diff --git a/.github/workflows/release-linux.yml b/.github/workflows/release-linux.yml
new file mode 100644
index 0000000..c441463
--- /dev/null
+++ b/.github/workflows/release-linux.yml
@@ -0,0 +1,62 @@
+name: Release Linux & macOS binaries
+
+# Triggered when a GitHub Release is published. Builds .so binaries for
+# Linux glibc (x86_64 + arm64) and macOS arm64 at PHP 8.4 + 8.5 (NTS
+# only, ZTS lives in the Windows lane), uploads each as a release asset
+# via php/pie-ext-binary-builder using PIE's pre-packaged-binary naming
+# convention. PIE's `pie install iliaal/php_clickhouse` then resolves
+# the prebuilt asset matching the host arch/os/libc/PHP-version instead
+# of building clickhouse-cpp from source (~3-5 min compile).
+#
+# Coverage matrix: 3 lanes x 2 PHP = 6 prebuilt assets per release.
+#
+# macOS Intel (macos-13) is intentionally absent: GitHub's Intel runner
+# pool is too contended to ship reliably, and macOS 26 drops Intel
+# support anyway. Intel-Mac users get source-build via PIE's
+# composer-default fallback.
+#
+# Linux musl is deferred: shivammathur/setup-php in Alpine containers
+# needs more setup; users on musl get source-build via PIE's
+# composer-default fallback meanwhile.
+#
+# This workflow + the existing release-windows.yml together produce
+# 6 Linux/macOS + 12 Windows = 18 prebuilt assets per release.
+
+on:
+  release:
+    types: [published]
+
+jobs:
+  build:
+    name: ${{ matrix.os }} php ${{ matrix.php }}
+    runs-on: ${{ matrix.os }}
+    permissions:
+      contents: write
+    strategy:
+      fail-fast: false
+      matrix:
+        # ubuntu-latest (24.04 x86_64), ubuntu-24.04-arm (arm64),
+        # macos-14 (Apple Silicon arm64).
+        # Pinned over `*-latest` to avoid silent runner shifts.
+        os:
+          - ubuntu-latest
+          - ubuntu-24.04-arm
+          - macos-14
+        php:
+          - '8.4'
+          - '8.5'
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Setup PHP
+        uses: shivammathur/setup-php@v2
+        with:
+          php-version: ${{ matrix.php }}
+          coverage: none
+
+      - name: Build and upload prebuilt binary
+        uses: php/pie-ext-binary-builder@0.0.3
+        with:
+          release-tag: ${{ github.event.release.tag_name }}
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          configure-flags: '--enable-clickhouse'
diff --git a/.github/workflows/release-windows.yml b/.github/workflows/release-windows.yml
new file mode 100644
index 0000000..60c5d1d
--- /dev/null
+++ b/.github/workflows/release-windows.yml
@@ -0,0 +1,55 @@
+name: Release Windows binaries
+
+# Triggered when a GitHub Release is published (manual or from a tag push
+# via /release-ext). Builds the extension on every supported PHP version
+# x architecture x thread-safety variant on Windows runners, then
+# uploads each .dll as a release asset following the PIE-compatible
+# naming convention (php_<ext>-<version>-<php_x.y>-<ts|nts>-vs17-<arch>.zip).
+#
+# Mirrors the Linux test matrix in build_matrix expansion so we ship
+# binaries for every PHP version we claim to support. Tests are off in
+# this lane -- the test workflow already validated the same arch/ts/php
+# combinations on master.
+
+on:
+  release:
+    types: [published]
+
+jobs:
+  matrix:
+    name: Resolve Windows release matrix
+    runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.matrix.outputs.matrix }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: matrix
+        uses: php/php-windows-builder/extension-matrix@v1
+        with:
+          php-version-list: '8.3, 8.4, 8.5'
+
+  build:
+    name: Build ${{ matrix.php-version }} ${{ matrix.ts }} ${{ matrix.arch }}
+    needs: matrix
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJson(needs.matrix.outputs.matrix) }}
+    steps:
+      - uses: actions/checkout@v6
+      - uses: php/php-windows-builder/extension@v1
+        with:
+          php-version: ${{ matrix.php-version }}
+          arch: ${{ matrix.arch }}
+          ts: ${{ matrix.ts }}
+          run-tests: 'false'
+
+  release:
+    name: Upload binaries to the release
+    runs-on: ubuntu-latest
+    needs: build
+    steps:
+      - uses: php/php-windows-builder/release@v1
+        with:
+          release: ${{ github.event.release.tag_name }}
+          token: ${{ secrets.GITHUB_TOKEN }}
diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
new file mode 100644
index 0000000..9aaf5cb
--- /dev/null
+++ b/.github/workflows/tests.yml
@@ -0,0 +1,334 @@
+name: Tests
+
+on:
+  push:
+    branches: [master]
+  pull_request:
+    branches: [master]
+
+jobs:
+  validate:
+    name: Validate composer.json
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - uses: shivammathur/setup-php@v2
+        with:
+          php-version: '8.4'
+          tools: composer
+          coverage: none
+      - run: composer validate --strict
+
+  pie-smoke:
+    name: PIE install smoke
+    runs-on: ubuntu-latest
+    needs: validate
+    steps:
+      - uses: actions/checkout@v6
+      - uses: shivammathur/setup-php@v2
+        with:
+          php-version: '8.4'
+          tools: phpize, php-config
+          coverage: none
+      - name: Install PIE build prerequisites
+        run: |
+          sudo apt-get update -qq
+          sudo apt-get install -y --no-install-recommends bison libtool-bin
+      - name: Download PIE
+        run: |
+          curl -fsSL -o /tmp/pie.phar https://github.com/php/pie/releases/latest/download/pie.phar
+          chmod +x /tmp/pie.phar
+      - name: PIE build smoke (base, no TLS)
+        run: |
+          php /tmp/pie.phar repository:add path .
+          php /tmp/pie.phar build iliaal/php_clickhouse:*@dev
+
+  linux:
+    name: PHP ${{ matrix.php }} - Linux
+    runs-on: ubuntu-latest
+    services:
+      clickhouse:
+        image: clickhouse/clickhouse-server:latest
+        ports:
+          - 9000:9000
+          - 8123:8123
+        env:
+          CLICKHOUSE_USER: test
+          CLICKHOUSE_PASSWORD: test
+        options: >-
+          --ulimit nofile=262144:262144
+          --health-cmd "wget -qO- http://localhost:8123/ping || exit 1"
+          --health-interval 5s
+          --health-timeout 3s
+          --health-retries 10
+
+    strategy:
+      fail-fast: false
+      matrix:
+        php: ['7.4', '8.1', '8.2', '8.3', '8.4', '8.5']
+
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Setup PHP
+        uses: shivammathur/setup-php@v2
+        with:
+          php-version: ${{ matrix.php }}
+          tools: phpize, php-config
+          extensions: none, dom, json, xml, simplexml, tokenizer
+          coverage: none
+
+      - name: Install OpenSSL dev (TLS build path)
+        run: sudo apt-get update -qq && sudo apt-get install -y libssl-dev
+
+      - name: Naming-consistency guard (no stray legacy references)
+        run: ./scripts/check-no-seasclick.sh
+
+      - name: Build extension (no TLS)
+        run: |
+          phpize
+          ./configure --enable-clickhouse
+          make -j$(nproc)
+
+      - name: Run test suite
+        env:
+          CLICKHOUSE_HOST: 127.0.0.1
+          CLICKHOUSE_PORT: 9000
+          CLICKHOUSE_USER: test
+          CLICKHOUSE_PASSWD: test
+          NO_INTERACTION: 1
+        run: |
+          # Stage the built .so into the system extension_dir so
+          # run-tests.php's --EXTENSIONS-- clickhouse directive resolves
+          # via the standard $ext_dir/clickhouse.so path on every PHP
+          # version. Loading via -d extension=$(pwd)/modules/clickhouse.so
+          # tripped PHP 7.4 dl.c's extension_dir prepend; overriding
+          # extension_dir breaks the system dom/xml/phar/etc that
+          # setup-php loads by basename.
+          EXT_DIR=$(php -r 'echo ini_get("extension_dir");')
+          sudo cp modules/clickhouse.so "$EXT_DIR/"
+          export TEST_PHP_EXECUTABLE=$(which php)
+          # run-tests.php exits non-zero on failures, zero on all-pass /
+          # all-skip. Let its exit code drive the step.
+          php run-tests.php --show-diff -g FAIL,BORK,LEAK,XLEAK tests/
+
+      - name: Rebuild with TLS (verifies SSL build path)
+        run: |
+          make clean
+          phpize --clean
+          phpize
+          ./configure --enable-clickhouse --enable-clickhouse-openssl
+          make -j$(nproc)
+
+  linux-zts:
+    name: PHP ${{ matrix.php }} ZTS - Linux
+    runs-on: ubuntu-latest
+    services:
+      clickhouse:
+        image: clickhouse/clickhouse-server:latest
+        ports:
+          - 9000:9000
+          - 8123:8123
+        env:
+          CLICKHOUSE_USER: test
+          CLICKHOUSE_PASSWORD: test
+        options: >-
+          --ulimit nofile=262144:262144
+          --health-cmd "wget -qO- http://localhost:8123/ping || exit 1"
+          --health-interval 5s
+          --health-timeout 3s
+          --health-retries 10
+
+    strategy:
+      fail-fast: false
+      matrix:
+        php: ['8.4']
+
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Install build deps
+        run: sudo apt-get update -qq && sudo apt-get install -y re2c bison autoconf libssl-dev
+
+      - name: Build PHP from source with --enable-zts
+        run: |
+          git clone --depth=1 --branch=PHP-${{ matrix.php }} https://github.com/php/php-src.git /tmp/php-src
+          cd /tmp/php-src
+          ./buildconf --force
+          ./configure \
+            --enable-zts \
+            --disable-all \
+            --enable-cli \
+            --prefix=/tmp/php-zts
+          make -j$(nproc)
+          make install
+
+      - name: Build extension under ZTS
+        run: |
+          export PATH="/tmp/php-zts/bin:$PATH"
+          phpize
+          ./configure --enable-clickhouse --with-php-config=/tmp/php-zts/bin/php-config
+          make -j$(nproc)
+
+      - name: Run test suite under ZTS
+        env:
+          CLICKHOUSE_HOST: 127.0.0.1
+          CLICKHOUSE_PORT: 9000
+          CLICKHOUSE_USER: test
+          CLICKHOUSE_PASSWD: test
+          NO_INTERACTION: 1
+        run: |
+          EXT_DIR=$(/tmp/php-zts/bin/php -r 'echo ini_get("extension_dir");')
+          mkdir -p "$EXT_DIR"
+          cp modules/clickhouse.so "$EXT_DIR/"
+          export TEST_PHP_EXECUTABLE=/tmp/php-zts/bin/php
+          /tmp/php-zts/bin/php run-tests.php --show-diff -g FAIL,BORK,LEAK,XLEAK tests/
+
+  # Windows extension builds need the official PHP SDK + matching MSVC
+  # toolset, not the autotools phpize/configure flow. The php-org
+  # composite action below bundles SDK download, VS detection,
+  # phpize / configure / nmake invocation, and a .dll smoke load.
+  #
+  # run-tests is enabled so the runner exercises extension load + class
+  # registration + skip handling. Server-required phpts skip up-front
+  # via clickhouse_skip_if_no_server() (which short-circuits on
+  # PHP_OS_FAMILY === Windows). The offline surface smoke (tests/051)
+  # actually runs and validates the public API and exception boundary.
+  windows-matrix:
+    name: Resolve Windows matrix
+    runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.matrix.outputs.matrix }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: matrix
+        uses: php/php-windows-builder/extension-matrix@v1
+        with:
+          php-version-list: '8.4'
+
+  windows:
+    name: Windows ${{ matrix.php-version }} ${{ matrix.ts }} ${{ matrix.arch }}
+    needs: windows-matrix
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJson(needs.windows-matrix.outputs.matrix) }}
+    steps:
+      - uses: actions/checkout@v6
+      - name: Build extension
+        uses: php/php-windows-builder/extension@v1
+        with:
+          php-version: ${{ matrix.php-version }}
+          arch: ${{ matrix.arch }}
+          ts: ${{ matrix.ts }}
+          run-tests: 'true'
+
+  asan:
+    name: PHP ${{ matrix.php }} - ASAN
+    runs-on: ubuntu-latest
+    services:
+      clickhouse:
+        image: clickhouse/clickhouse-server:latest
+        ports:
+          - 9000:9000
+          - 8123:8123
+        env:
+          CLICKHOUSE_USER: test
+          CLICKHOUSE_PASSWORD: test
+        options: >-
+          --ulimit nofile=262144:262144
+          --health-cmd "wget -qO- http://localhost:8123/ping || exit 1"
+          --health-interval 5s
+          --health-timeout 3s
+          --health-retries 10
+
+    strategy:
+      fail-fast: false
+      matrix:
+        php: ['8.4']
+
+    env:
+      # detect_leaks=0: vendored clickhouse-cpp creates a NonSecureSocketFactory
+      #   per Client that's only freed at process exit; suppress until upstream
+      #   releases a fix.
+      # detect_odr_violation=0: clickhouse-cpp templates produce duplicate
+      #   symbols across translation units, harmless but noisy.
+      # exitcode=139: keep ASan abort on the conventional SIGSEGV exit code so
+      #   run-tests.php classifies failures as crashes, not silent skips.
+      ASAN_OPTIONS: exitcode=139:detect_odr_violation=0:detect_leaks=0
+      UBSAN_OPTIONS: print_stacktrace=1
+      USE_ZEND_ALLOC: 0
+
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Install dependencies
+        run: sudo apt-get update -qq && sudo apt-get install -y llvm clang re2c bison autoconf libssl-dev
+
+      - name: Resolve clang runtime dir
+        id: clang_rt
+        run: echo "dir=$(clang -print-runtime-dir)" >> "$GITHUB_OUTPUT"
+
+      - name: Build PHP with ASAN
+        env:
+          LD_LIBRARY_PATH: ${{ steps.clang_rt.outputs.dir }}
+        run: |
+          git clone --depth=1 --branch=PHP-${{ matrix.php }} https://github.com/php/php-src.git /tmp/php-src
+          cd /tmp/php-src
+          ./buildconf --force
+          # -shared-libasan: link ASan as a shared library so the dynamic
+          # linker can resolve interceptors (notably __cxa_throw) across the
+          # dlopen'd extension boundary. With static asan, the extension's
+          # interceptor for __cxa_throw never finds the real libstdc++
+          # symbol and ASan aborts the moment the wrapper throws.
+          CC=clang CXX=clang++ \
+          CFLAGS="-fsanitize=address,undefined -fno-sanitize=function,vptr -shared-libasan -DZEND_TRACK_ARENA_ALLOC" \
+          LDFLAGS="-fsanitize=address,undefined -shared-libasan" \
+          ./configure \
+            --enable-debug \
+            --disable-zts \
+            --disable-all \
+            --enable-cli \
+            --prefix=/tmp/php-asan
+          make -j$(nproc)
+          make install
+
+      - name: Build extension with ASAN
+        env:
+          LD_LIBRARY_PATH: ${{ steps.clang_rt.outputs.dir }}
+        run: |
+          export PATH="/tmp/php-asan/bin:$PATH"
+          phpize
+          # Match PHP's exact -fsanitize / -fno-sanitize / -shared-libasan
+          # set so the runtime ABI lines up. Disable `function` (PHP's own
+          # build does) and `vptr` (the vptr check needs
+          # __ubsan_vptr_type_cache, which the dlopen'd extension can't
+          # resolve against the host PHP's runtime). Also disable
+          # `pointer-overflow`: vendored LZ4 evaluates `dictionary + dictSize`
+          # which is `NULL + 0` when no compression dictionary is set. This
+          # is technically UB per C, harmless in practice, and a known
+          # false positive that every LZ4 + UBSan integration suppresses.
+          CC=clang CXX=clang++ \
+          CFLAGS="-fsanitize=address,undefined -fno-sanitize=function,vptr,pointer-overflow -shared-libasan -DZEND_TRACK_ARENA_ALLOC" \
+          CXXFLAGS="-fsanitize=address,undefined -fno-sanitize=function,vptr,pointer-overflow -shared-libasan -DZEND_TRACK_ARENA_ALLOC" \
+          LDFLAGS="-fsanitize=address,undefined -shared-libasan" \
+          ./configure --enable-clickhouse --with-php-config=/tmp/php-asan/bin/php-config
+          make -j$(nproc)
+
+      - name: Run tests under ASAN
+        env:
+          LD_LIBRARY_PATH: ${{ steps.clang_rt.outputs.dir }}
+          CLICKHOUSE_HOST: 127.0.0.1
+          CLICKHOUSE_PORT: 9000
+          CLICKHOUSE_USER: test
+          CLICKHOUSE_PASSWD: test
+          NO_INTERACTION: 1
+        run: |
+          # PHP and the extension are linked against the shared libasan/ubsan
+          # runtimes, so the dynamic linker needs to find them. LD_LIBRARY_PATH
+          # points at clang's runtime dir; no LD_PRELOAD is needed.
+          EXT_DIR=$(/tmp/php-asan/bin/php -r 'echo ini_get("extension_dir");')
+          mkdir -p "$EXT_DIR"
+          cp modules/clickhouse.so "$EXT_DIR/"
+          export TEST_PHP_EXECUTABLE=/tmp/php-asan/bin/php
+          /tmp/php-asan/bin/php run-tests.php --show-diff --asan -g FAIL,BORK,LEAK,XLEAK tests/
diff --git a/.gitignore b/.gitignore
index 6e642d6..68fc939 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,6 @@
+# in-progress security findings (pre-disclosure); never commit
+docs/security/
+
 # Prerequisites
 *.d
 
@@ -31,3 +34,40 @@
 *.out
 *.app
 vendor
+
+# phpize / autotools build artifacts
+.deps/
+.libs/
+*.dep
+*.loT
+Makefile
+Makefile.fragments
+Makefile.objects
+acinclude.m4
+autom4te.cache/
+build/
+config.h
+config.h.in
+config.log
+config.nice
+config.status
+configure
+configure.ac
+configure.in
+install-sh
+libtool
+ltmain.sh
+missing
+mkinstalldirs
+run-tests.php
+modules/
+# Windows build script (kept tracked)
+
+# local AI tooling: per-developer scaffolding, not checked in
+.claude/
+.codex/
+AGENTS.md
+CLAUDE.md
+
+# editor backups
+*~
diff --git a/.release-config b/.release-config
new file mode 100644
index 0000000..db03f59
--- /dev/null
+++ b/.release-config
@@ -0,0 +1,32 @@
+# Read by /release-ext. Schema documented in ~/.claude/commands/release-ext.md.
+
+extension: clickhouse
+version_macro: PHP_CLICKHOUSE_VERSION
+version_file: php_clickhouse.h
+repo: iliaal/php_clickhouse
+build_matrix:
+  - PHP-8.3
+  - PHP-8.4
+  - PHP-8.5
+configure_base: "--enable-clickhouse"
+
+pre_test:
+  - command: "docker ps | grep clickhouse-test"
+    on_fail: "Start the test container per README before releasing"
+
+test_env:
+  CLICKHOUSE_HOST: "127.0.0.1"
+  CLICKHOUSE_PORT: "9000"
+  CLICKHOUSE_USER: "test"
+  CLICKHOUSE_PASSWD: "test"
+
+extra_build_variants:
+  - name: "TLS variant"
+    configure: "--enable-clickhouse --enable-clickhouse-openssl"
+    php_version: PHP-8.4
+
+# Toggle: ship pre-built Linux/macOS binaries via .github/workflows/release-linux.yml
+# (uses php/pie-ext-binary-builder). Linux glibc x86_64+arm64 and macOS x86_64+arm64,
+# NTS only, PHP 8.4+8.5. musl deferred. Windows binaries handled separately by
+# release-windows.yml.
+linux_prebuilt: true
diff --git a/.travis.yml b/.travis.yml
deleted file mode 100644
index 0fe131a..0000000
--- a/.travis.yml
+++ /dev/null
@@ -1,51 +0,0 @@
-language: php
-
-compiler:
-  - gcc
-  - clang
-
-os:
-  - linux
-
-dist: trusty
-
-env:
-  - REPORT_EXIT_STATUS=1 NO_INTERACTION=1
-
-services:
-  - docker
-
-#hosts
-addons:
-  hosts:
-    - clickhouse
-
-php:
-  - 5.4
-  - 5.5
-  - 5.6
-  - 7.0
-  - 7.1
-  - 7.2
-  - 7.3
-  - 7.4
-
-notifications:
-  email: whj199649@gmail.com
-
-#clickhouse
-before_install:
-    - docker run -d -p 9000:9000 --name some-clickhouse-server --ulimit nofile=262144:262144 yandex/clickhouse-server
-    - docker ps -a
-    - g++ -v
-    - gcc -v
-    - clang -v
-    - clang++ -v
-
-#Compile
-before_script:
-    - phpize && ./configure && make clean && make
-
-# test
-script:
-    - ./travis/run-tests.sh
diff --git a/.upstream/clickhouse-cpp.yml b/.upstream/clickhouse-cpp.yml
new file mode 100644
index 0000000..8cac713
--- /dev/null
+++ b/.upstream/clickhouse-cpp.yml
@@ -0,0 +1,8 @@
+# /release-ext library-freshness gate. Schema: ~/.claude/commands/release-ext.md
+# detected from lib/clickhouse-cpp/clickhouse/version.h
+
+pinned: 2.6.1
+vendored_at: lib/clickhouse-cpp
+resolver: github
+repo: ClickHouse/clickhouse-cpp
+note: "2.6.1 + LOCAL_PATCHES.md (empty string_view memcpy UB); upstream PR ClickHouse/clickhouse-cpp#489"
diff --git a/CHANGELOG.md b/CHANGELOG.md
new file mode 100644
index 0000000..e3c4f8f
--- /dev/null
+++ b/CHANGELOG.md
@@ -0,0 +1,703 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.8.5] - 2026-05-11
+
+### Added
+
+- `insertFromStream($table, $columns, $stream, $format, $batch_rows, $query_id, $settings)`
+  stream-parses a TSV or CSV file (or any readable PHP stream resource)
+  and INSERTs rows in batches of `$batch_rows` (default 10000) without
+  materializing the full file in PHP memory. The parser is C++ and
+  handles TSV escapes (`\\`, `\t`, `\n`, `\r`, `\0`) and RFC 4180 CSV
+  quoting (embedded `"`, `,`, CRLF, doubled `""`). Literal `\N` is the
+  NULL marker in both formats; empty CSV cells stay as empty strings.
+  `*WithNames` variants discard the first input row. Returns rows
+  inserted. Same recovery semantics as `insert()` / `writeStart()`: a
+  mid-stream throw resets the native client so the handle stays
+  usable. Smoke test: 100k single-block-shaped rows insert in ~140ms
+  on a local Memory-engine table.
+- 4 new PHPTs (103–106) cover the TSV happy path with header skipping
+  and escape decoding, CSV quoting / embedded commas / `""` / CRLF
+  rows / multi-line cells, NULL parsing in both formats plus batch
+  boundary correctness with `batch_rows=10`, and the rejection
+  surface (bad format, non-stream zval, zero batch, empty columns,
+  row-too-wide, unterminated quoted CSV cell, handle remains usable
+  after each rejection).
+- `selectToStream($sql, $params, $stream, $format, $query_id, $settings)`
+  writes SELECT rows directly to a PHP stream resource (`fopen` /
+  `php://memory` / `php://stdout` / any writable stream) in
+  `TabSeparated`, `TabSeparatedWithNames`, `CSV`, or `CSVWithNames`
+  format. Returns rows written. Cells are formatted block-by-block in
+  C++ from native column data and flushed without per-row PHP array
+  assembly or userland callback overhead, so large exports run
+  meaningfully faster than `selectStream` + manual `fwrite`. Dates
+  always emit as ISO strings; Decimal / Int128 / UInt128 as decimal
+  strings; NULL renders as `\N` in TSV and as an empty cell in CSV.
+  Nullable and LowCardinality wrappers around supported scalars are
+  fine; Array, Tuple, Map, and geometry columns are rejected with a
+  `ClickHouseException` — text formats can't unambiguously represent
+  them. Mid-stream errors `ResetConnection()` the handle to keep it
+  usable, mirroring the existing recovery in `insert()` / `writeStart()`.
+- 5 new PHPTs (098–102) cover TSV happy path with escape characters,
+  CSV RFC-4180 quoting (embedded `"`, `,`, `\n`, `\r`), `WithNames`
+  header emission (including the empty-result-still-emits-header
+  case), Nullable rendering, and the rejection surface (bad format,
+  non-stream argument, unsupported column types).
+- `selectWithExternalData($sql, $externals, ...)` sends one or more
+  named in-memory tables alongside a SELECT (ClickHouse "external
+  data" feature). Lets the query body reference the externals by name
+  (`WHERE id IN ext_ids`) instead of materializing a huge literal `IN
+  (...)` list. Each external is `['name' => ..., 'columns' => ['col'
+  => 'Type', ...], 'rows' => [[...], ...]]`; multiple externals per
+  call supported. Per-call `params`, `fetch_mode`, `query_id`, and
+  `settings` flow through identically to `select()`. Vendored
+  `clickhouse-cpp` patched with a `SelectWithExternalData(const
+  Query&, const ExternalTables&)` overload so settings + query_id
+  propagate (see `lib/clickhouse-cpp/LOCAL_PATCHES.md`).
+- 3 new PHPTs (095–097) cover the basic IN-clause path, multiple
+  externals with mixed column types (`String`, `Date`, `Nullable`,
+  multi-column lookup pairs), and the malformed-input rejection
+  surface (empty externals, missing/wrong keys, bad identifier,
+  unsupported type, row width mismatch, recovery after rejection).
+
+### Changed
+
+- Parser strictness for `insertFromStream` is RFC 4180-strict in CSV
+  (any byte other than `,` / CR / LF / `""`-escape after a closing
+  quote throws; a quoted-empty cell at EOF without a trailing newline
+  still flushes the row) and ClickHouse-TSV-strict (`\N` is the
+  whole-cell NULL marker — trailing bytes after it throw, `\\N`
+  decodes to the literal two-character string and not NULL). NULL
+  into a non-Nullable target column is rejected up front against the
+  server schema (Nullable / LowCardinality(Nullable(...)) accepted,
+  everything else throws with the column name). Empty external-table
+  rows in `selectWithExternalData` are rejected with a clear message
+  (the native protocol uses an empty block as the end-of-stream
+  marker, so the clickhouse-cpp client silently skips zero-row named
+  blocks and the server then sees no such table). Stream read errors
+  do not commit partial data — `php_stream_read` returning `n < 0`,
+  or `n == 0` while `!php_stream_eof()`, throws and routes through
+  the existing `ResetConnection()` recovery so any rows enqueued
+  before the failure discard before commit. TSV escapes split across
+  a `php_stream_read` chunk boundary (a `\` at byte 65 535) now
+  decode correctly via a `pending_backslash` parser-state field
+  (same mechanism `prev_was_cr` uses for CRLF straddling a chunk).
+- Internal: parser containers (`cell_buf`, `row_cells`) pre-sized at
+  setup so the first row pays the same allocation cost as subsequent
+  rows; `appendCellForStream` skips the `zend_string` round-trip
+  when the cell is already `IS_STRING` (String / FixedString / Date*
+  / DateTime* / Decimal* / Int128 / UInt128 / UUID / IPv4 / IPv6 —
+  produced as strings by `convertToZval` under
+  `SC_FETCH_DATE_AS_STRINGS`). `StreamOutFormat` and
+  `InsertStreamFormat` collapsed into one `StreamFormat` enum with a
+  single parse helper, removing the synchronization hazard between
+  the input and output sides of the format alias table.
+
+## [0.8.1] - 2026-05-01
+
+Hardening release covering nine rounds of reviewer-driven fixes
+across the insert, write, and type-conversion surfaces. Streaming
+and direct insert paths now recover the native client across every
+server-side rejection point (BeginInsert, SendInsertBlock,
+EndInsert) so a thrown insert no longer wedges the handle with
+"cannot execute query while inserting". Type conversion gained
+strict full-consumption parsers for the Map, narrow-int, UInt64,
+Int128/UInt128, and date/time surfaces; placeholders, hex literals,
+enums, and Nullable inserts validate up front instead of silently
+coercing. Insert and write batch transposition no longer
+materializes a full PHP column-major matrix — peak intermediate PHP
+memory drops from N_rows × N_cols to one column at a time.
+
+### Added
+
+- UInt64 inserts accept decimal and hex strings above
+  `ZEND_LONG_MAX` on both the scalar and `Map(*, UInt64)` paths
+  via a shared `strict_zval_u64` parser. Reads continue to surface
+  upper-half values as decimal strings.
+- 23 new PHPTs (072–094) pinning the new safety surfaces:
+  insert/write shape validation, UInt64 string round-trip,
+  server-side `EndInsert`/`BeginInsert` recovery, destructor
+  rollback, partial-commit abort, enum range/Nullable rejection,
+  single-token placeholder enforcement, Map narrow-int bounds,
+  DateTime64 fractional validation, `insertAssoc` key-set drift.
+
+### Changed
+
+- `insert()` and `write()` build native ClickHouse columns one at
+  a time directly from the user's row-major input. The previous
+  path materialized a full column-major PHP zval matrix before
+  building.
+- `insertAssoc()` no longer makes a positional PHP copy of input
+  rows; the column gatherer reads each column directly from the
+  original associative rows. Key validation now uses
+  `zend_hash_exists` against the first row's HashTable instead of
+  allocating an `std::string` for every row key.
+- `require.php` floor raised to `>=7.4` to match the CI matrix
+  (PHP 7.1–7.3 were never tested).
+
+### Fixed
+
+- `BeginInsert`, `SendInsertBlock`, and `EndInsert` exceptions
+  reset the native connection before rethrowing, so the same
+  handle stays usable. Previously a server-side rejection
+  (missing table, bad column, CHECK constraint, schema drift)
+  left the vendored client's `inserting_` flag set; subsequent
+  `select`/`execute` on the same handle threw "cannot execute
+  query while inserting" until manual `resetConnection()`.
+- Destructor cleanup mirrors the dirty/clean recovery split: an
+  in-flight streaming insert with sent blocks is dropped via
+  `ResetConnection` on `unset()` rather than committed via
+  `EndInsert`. Clean sessions still `EndInsert`. Avoids partial
+  commit on script bailout.
+- `write()` rejects rows narrower or wider than the `writeStart`
+  column count instead of silently sending a truncated block. The
+  previous path took the first row's element count as
+  authoritative, so a row like `[1]` against
+  `writeStart(t, ['a','b'])` landed `1` into column `a` with `b`
+  defaulted server-side.
+- `insert()` rejects rows with extra positional or named cells
+  instead of silently dropping them. A row like `[1, 99]` against
+  a single-column table previously landed as `1` with `99` lost.
+- A failed later `write()` no longer commits previously sent
+  blocks. The catch path tracks whether any block has been sent
+  in the current `writeStart()` session and chooses
+  `ResetConnection` (discard) over `EndInsert` (commit) on a
+  dirty session.
+- `insertAssoc()` rejects integer-keyed later rows and any
+  key-set drift from the first row. The first row defines the
+  column set; every later row must have the same string-key set.
+- `Enum8`/`Enum16` inserts reject undeclared integers, NULL on
+  non-Nullable columns, and unknown string names.
+- PHP 8.5 `ReflectionProperty::setAccessible()` deprecation
+  guarded by version. `ZEND_ACC_NOT_SERIALIZABLE` guarded for
+  PHP < 8.1.
+
+### Security & Hardening
+
+- Strict full-consumption integer and double parsers across the
+  Map, narrow-int, Int128/UInt128, geo, DateTime64, and Time64
+  insert paths. Non-numeric strings, fractional doubles, non-finite
+  floats, and out-of-range values throw instead of silently
+  coercing to 0/0.0 inside the column.
+- Single-token placeholder validator: `{name}` placeholders accept
+  exactly one identifier (column or table name) and reject
+  comma-separated lists. Comma-list callers must use array form.
+- Hex literal sanity: `0x...` strings reject embedded NUL bytes
+  and partial parses against unsigned integer columns.
+- Same-client reentry guard: a userland progress/profile callback
+  that fires another query on the same handle throws cleanly
+  instead of crashing the worker on the next `ReceiveData`.
+- Recursive type-conversion depth cap (32) keeps deeply nested
+  structures (`Array(Array(...))`, `Map(K, Tuple(...))`) from
+  blowing the stack.
+- Public-API consistency: `setSettings()` validates keys,
+  `setVerbose(null)` disables verbose mode cleanly, and
+  `setProgressCallback(null)` clears the registered callback.
+
+## [0.8.0] - 2026-04-30
+
+Architecture refactor that moves per-Client state from file-scope
+`std::map` banks onto the `zend_object` itself. The change unblocks
+ZTS support (no more global state to thread-isolate), plugs a
+pre-existing leak on bailout (`free_obj` fires; the old userspace-only
+`__destruct` did not), and fixes a refcount bug on the progress
+callback. ZTS Linux builds are now first-class; a Windows
+`config.w32` ships and is exercised in CI as a build-only smoke
+test. Adds streaming via `ClickHouseRowIterator` plus a true
+per-row callback path, four new Client knobs / introspection
+methods, seven DDL helpers, transparent `LowCardinality(Nullable(T))`
+and geo-type round-trips, and `query_id` echo through
+`getStatistics()`.
+
+### Added
+
+- Streaming row iterator: `selectStream(string $sql, ...)` returns a
+  `ClickHouseRowIterator` (`Iterator` + `Countable`) so large result
+  sets walk lazily without materializing as a single PHP array. The
+  iterator survives `unset($client)` since blocks own their column
+  data via `shared_ptr`.
+- True per-row callback path: `selectStreamCallback(string $sql,
+  callable $cb, ...)` invokes the callback once per row as blocks
+  arrive, never accumulating the full result. Use this for unbounded
+  streams.
+- Client knobs and introspection: `ping_before_query` constructor
+  config key (round-trip ping before each query), `resetConnection()`
+  method, `getServerInfo()` (name, version_major/minor/patch,
+  revision, timezone, display_name), `getCurrentEndpoint()`
+  (host/port of the active endpoint when an endpoints[] pool is in
+  use), `setProfileCallback(?callable $cb)` for `Profile` packets
+  (rows, blocks, bytes, rows_before_limit, applied_limit).
+- `query_id` echoed through `getStatistics()` so callers can correlate
+  a recorded statistics snapshot to a server-side query in
+  `system.query_log`.
+- DDL helpers: `isExists(string $database, string $table): bool`,
+  `showDatabases(): array`, `showProcesslist(): array`,
+  `getServerVersion(): string`, `tableSize(string $table): array`,
+  `truncateTable(string $table): bool`,
+  `dropPartition(string $table, string $partition): bool`. All
+  identifier arguments validated; `dropPartition` SQL-escapes the
+  partition value.
+- Type coverage: `LowCardinality(Nullable(String))` and
+  `LowCardinality(Nullable(FixedString))` round-trip on both read
+  and write paths. Geo types Point, Ring, Polygon, MultiPolygon
+  round-trip via `ColumnGeo` (Point as `[Float64, Float64]`, others
+  as nested arrays). `SimpleAggregateFunction(f, T)` reads
+  transparently as `T`.
+- Map matrix expansion. The insert path now accepts any
+  `Map(K, V)` over scalar K and V (String, all signed/unsigned
+  integer widths, Float32/64, UUID) plus `LowCardinality(String)`
+  keys and values. Read path mirrors the same matrix except for
+  `LowCardinality` keys (vendor gap). Previously only five
+  hardcoded combinations worked.
+- `ClickHouseRowIterator` class registered alongside `ClickHouse`.
+- smi2/phpClickHouse-style ergonomics:
+  - `setSettings()` returns `$this` instead of `bool` so callers
+    can chain. Truthy semantics preserved.
+  - `setSetting(string $key, mixed $value): static` for single-key
+    chainable sugar on top of the array form.
+  - `setDatabase(string $database): static` issues `USE` on the
+    server and updates the cached default used by helpers like
+    `databaseSize()` and `showTables()`. Validates the identifier.
+  - `ClickHouseException::getServerCode()`, `getServerName()`,
+    `getQueryId()` getter aliases for the existing public
+    `server_code`/`server_name`/`query_id` properties. Same data,
+    smi2-compatible call shape.
+- `selectStatement(string $sql, ...): ClickHouseStatement`
+  result-wrapper variant of `select()`. The new
+  `ClickHouseStatement` class implements Iterator, Countable,
+  ArrayAccess, and JsonSerializable over the materialized rows,
+  plus `fetchOne()` / `fetchKeyPair()` / `fetchColumn()` /
+  `toArray()` / `statistics()`. Carries a per-call stats snapshot
+  so it survives the client running other queries afterwards.
+  Read-only: `offsetSet`/`offsetUnset` throw. Plain `select()` is
+  unchanged and remains the faster path when you just need the
+  array.
+- `setVerbose(bool|callable $sink): static` for protocol-level
+  lifecycle tracing. Pass `true` to log JSON lines on STDERR,
+  `false` to disable, or a callable invoked with
+  `($eventName, $context)` per event. Events: `select_start`,
+  `data_block`, `select_finish`, `execute_start`,
+  `execute_finish`, `server_exception`. Existing
+  progress/profile callbacks are unaffected. No-op when off,
+  so the hot path stays cheap on production deployments.
+
+### Changed
+
+- Per-Client state (Client*, insert Block, ClientStats, settings,
+  progress/profile callbacks, log_enabled, query_log) lives on the
+  `zend_object` itself via custom `create_object`/`free_obj`
+  handlers. Replaces the seven file-scope `std::map<int, ...>` banks
+  keyed on `Z_OBJ_HANDLE`.
+- ZTS gate at MINIT removed. The extension loads under `--enable-zts`
+  builds; per-object state means no thread-shared mutable state to
+  protect.
+- `config.w32` rewritten from a 9-line warning stub to a full Windows
+  build script that mirrors `config.m4`'s source list, includes, and
+  flags. Optional `--enable-clickhouse-openssl` plumbing is mirrored
+  via `CHECK_LIB("libssl.lib", ...)`. CI exercises Windows as a build
+  + extension-load smoke (no live server tests on Windows).
+- CI matrix gains a `linux-zts` job (PHP 8.4 ZTS built from source)
+  and a `windows` job (build-only).
+
+### Security & Hardening
+
+- Multi-round security and correctness sweep across the input boundary,
+  Map / Int128 / hex parsers, recursion paths, and locale-sensitive
+  serialization. Wrong-type input now surfaces as `TypeError` instead
+  of corrupting memory; adversarial server schemas can no longer
+  stack-overflow the worker; narrow-int and 128-bit columns reject
+  out-of-range values up front; embedded NUL bytes no longer slip past
+  parsers; and Float / typed-parameter formatting is locale-independent.
+
+### Fixed
+
+- IPv4 / IPv6 read paths no longer crash. Vendored clickhouse-cpp
+  v2.6.1 made `ColumnIPv4`/`ColumnIPv6` siblings of (not subclasses
+  of) `ColumnUInt32`/`ColumnFixedString`, so the prior
+  `As<ColumnUInt32>()` / `As<ColumnFixedString>()` calls returned
+  null and segfaulted on dereference. Use `ColumnIPv*::AsString(row)`
+  for canonical dotted-quad / `::1` form.
+- Progress callback zval refcount: `setProgressCallback` now uses
+  `ZVAL_COPY` instead of a struct copy, so the callable doesn't get
+  freed out from under us when the caller goes out of scope.
+- Connection / insert-block leaks on bailout: cleanup runs in the new
+  `free_obj` handler, which fires unconditionally. Previously the
+  userspace `__destruct` didn't run on fatal errors, leaking the
+  underlying `Client*` and any half-open insert stream.
+- `selectStreamCallback` row-callback exceptions abort the stream and
+  surface immediately instead of sitting buffered in `EG(exception)`
+  while the stream consumes remaining rows and `recordQuerySuccess`
+  runs. Mirrors the progress / profile / verbose callback pattern.
+- Boolean config keys (`compression`, `tcp_nodelay`, `tcp_keepalive`,
+  `ssl`, etc.) read the coerced value correctly. The prior path read
+  the lval slot after `convert_to_boolean` flipped the type tag,
+  picking up stale union storage (a freed string pointer for
+  `IS_STRING` input, dval bits for `IS_DOUBLE`); `compression => "true"`
+  silently became compression mode 0.
+- Negative seconds-based timeouts and TCP-keepalive counters reject
+  up front instead of wrapping to giant unsigned through
+  `SetSendRetries` / `SetTcpKeepAlive*`.
+- `setSettings()` rejects integer keys and empty-string keys
+  consistently with `setSetting()`. The bulk form used to silently
+  drop integer entries and store unusable empty keys.
+- `setVerbose(null)` accepts null as a synonym for `false`, matching
+  the other `?callable` setters.
+- Streaming-callback OnData, `fetchKeyPair`, and `ClickHouseStatement`
+  iterator paths plug zend_string refcount and HashTable leaks on
+  `convertToZval` throw and on object-key coercion.
+
+### Changed
+
+- `composer.json` declares `ext-json`; `config.m4` and `config.w32`
+  declare `PHP_ADD_EXTENSION_DEP(clickhouse, json)`. Loading against
+  a json-less PHP fails at install / dlopen with a clear message
+  instead of `undefined symbol: php_json_serializable_ce`.
+- `composer.json` `require.php` floor bumped to `>=7.4`. PHP 7.1–7.3
+  were never CI-tested; the prior floor was aspirational.
+- Linux glibc `x86_64` + `arm64` and macOS `x86_64` + `arm64` pre-built
+  binaries now ship via `php/pie-ext-binary-builder` on tag push
+  (`release-linux.yml`). NTS only, PHP 8.4 + 8.5. Windows binaries
+  continue via `release-windows.yml`.
+- CI gains gating ASAN (PHP 8.4), PIE install smoke (PHP 8.4 base
+  build, no TLS), `composer.json` validation, and parallel Windows
+  builds across `nts`/`ts` × `x86`/`x64`.
+
+### Known limitations
+
+- `SELECT ... WITH TOTALS` and `SETTINGS extremes=1` still throw
+  `unimplemented 7` from the cpp layer. clickhouse-cpp v2.6.1 does
+  not dispatch the Totals/Extremes packet types
+  ([upstream issue #297](https://github.com/ClickHouse/clickhouse-cpp/issues/297));
+  `getTotals()` / `getExtremes()` are deferred to a future release.
+- `Map(LowCardinality(K), V)` read paths are not yet decoded by
+  the vendored library (writes succeed). `showProcesslist()`
+  selects a fixed projection of standard columns to avoid the
+  unsupported map columns (`ProfileEvents`, `Settings`, `used_*`).
+
+## [0.7.0] - 2026-04-26
+
+Feature release closing the ergonomics gap with smi2/phpClickHouse.
+Adds per-query and client-wide settings, server-side typed parameters,
+a progress callback, a statistics getter, structured exception fields,
+millisecond timeout precision, an associative-row insert helper, and a
+small set of SQL helper methods. All additive; no BC breaks.
+
+Method signatures, return types, and class properties are now declared
+via a stub-driven arginfo workflow and visible to Reflection / IDEs /
+static analyzers.
+
+### Added
+
+- `setSettings(array $settings)` for client-wide ClickHouse settings
+  (e.g. `max_execution_time`, `max_memory_usage`, `async_insert`).
+  Per-call settings take a 5th array argument on `select()`,
+  `insert()`, `execute()`, `writeStart()`. Per-call overrides global.
+- Server-side typed parameters via the `{name:Type}` placeholder
+  syntax. Routed through `Query::SetParam` so the server quotes and
+  parses according to the declared `Type`. Plain `{name}`
+  placeholders keep their existing client-side identifier-substitution
+  behavior. Arrays format as ClickHouse array literals so
+  `Array(UInt32)`, `Array(String)`, etc. round-trip cleanly.
+- `setProgressCallback(?callable $cb)` invokes the callable for every
+  `Progress` packet during a query, receiving an associative array of
+  `rows`, `bytes`, `total_rows`, `written_rows`, `written_bytes`.
+- `getStatistics()` returns `rows_read`, `bytes_read`, `total_rows`,
+  `written_rows`, `written_bytes`, `blocks`, `rows_before_limit`,
+  `applied_limit`, `elapsed_ms` from the last completed query. Reset
+  at the start of each `select` / `execute` / `insert` / `writeStart`.
+- Structured `ClickHouseException` fields: `server_code` (server
+  error code, e.g. 159 for TIMEOUT_EXCEEDED), `server_name` (e.g.
+  `DB::Exception`), and `query_id`. Populated on server errors and on
+  any throw that has a query-id context; unset on pure client errors.
+- `insertAssoc(string $table, array $rows, string $query_id = "",
+  array $settings = [])` derives the column list from the keys of
+  the first row and forwards to `insert()`.
+- SQL helper methods: `databaseSize(?string $database)`,
+  `tablesSize(?string $database)`, `partitions(string $table)`,
+  `showTables(?string $database, ?string $like)`,
+  `showCreateTable(string $table)`, `getServerUptime()`. Each
+  validates identifiers against the existing safe-character set.
+- Config keys `connect_timeout_ms`, `receive_timeout_ms`, and
+  `send_timeout_ms` for sub-second timeout precision. Override the
+  existing seconds-based keys when present.
+- `enableLogQueries(bool $enabled = true)` toggles a per-client query
+  log accumulator; `getLogQueries()` returns the entries and clears
+  the buffer. Each entry carries `sql`, `query_id`, `elapsed_ms`,
+  `rows_read`, `bytes_read`, `error_code`, `error_message`. Errors
+  are recorded with the ClickHouse server code (or `-1` for
+  client/network failures).
+
+### Changed
+
+- `select()`, `insert()`, `execute()`, and `writeStart()` now build a
+  full `clickhouse::Query` object internally so settings, server-side
+  params, progress, and profile callbacks can attach. Behavior of
+  existing call sites is unchanged.
+- Vendored `clickhouse-cpp` patched to expose
+  `Client::BeginInsert(const Query&)` so the streaming insert path
+  honors per-query settings and progress callbacks. Documented in
+  `lib/clickhouse-cpp/LOCAL_PATCHES.md`.
+- Migrated to stub-driven arginfo (`clickhouse.stub.php` + generated
+  `clickhouse_arginfo.h`). Method parameter and return types are now
+  declared and visible to Reflection / IDEs / static analyzers;
+  previously they were untyped at the engine boundary. Behavior is
+  unchanged for correctly-typed callers; wrong-type callers now hit
+  ZPP at the boundary instead of a custom thrown exception inside the
+  method body.
+- ClickHouse and ClickHouseException properties are now declared with
+  types via the stub.
+- Compat shims in `php7_wrapper.h` keep the generated arginfo header
+  compiling unchanged across the entire build matrix (PHP 7.4 through
+  8.5). On pre-8.0 builds the polyfills drop type information rather
+  than emulate it, so reflection signatures revert to untyped on those
+  versions; runtime behavior is unchanged.
+
+### For contributors
+
+- New phpt tests `034`–`041` cover structured exception fields,
+  settings precedence, typed-param round-trip, progress callback
+  firing, fast connect-timeout, `insertAssoc`, each SQL helper, and
+  the query log accumulator.
+
+## [0.6.0] - 2026-04-25
+
+Hardening release on top of 0.5.0. Closes a SQL-injection class
+through the `{placeholder}` substitution, fixes a handful of
+lifecycle crashes around `__construct` failure and orphan
+`writeStart`, and resolves several smaller correctness bugs in the
+data path. Two upstream clickhouse-cpp v2.6.1 bugs are patched
+locally and queued for upstream.
+
+### Security
+
+- `select()` and `execute()` placeholder substitution validates each
+  value against an identifier-and-numerics character set; quotes,
+  semicolons, backslashes, and other SQL meta-characters are
+  rejected before the SQL is built. Closes an injection class that
+  was reachable any time a caller fed user input through `$params`.
+- `insert()` and `writeStart()` validate table and column
+  identifiers against ClickHouse identifier syntax. Empty names,
+  leading digits, and shell-meta characters throw before the INSERT
+  is built.
+- New `ssl_min_protocol_version` config knob (default `tls1.2`) so
+  a server speaking only deprecated TLS versions fails closed
+  without explicit config.
+- `ClickHouseException` messages strip the SQL fragment that
+  clickhouse-cpp appends to its errors, so a literal placed in a
+  placeholder can't leak back to userland through `e.what()`.
+- The `passwd` config key is consumed into `ClientOptions` and
+  discarded; it's no longer stored as a PHP-object property, so
+  `var_dump`, `serialize`, and reflection don't expose it.
+- ZTS PHP builds refuse to load with an `E_CORE_ERROR`. Process-
+  global Client state would race on shared handle space.
+
+### Added
+
+- 9 new phpt tests (025-033) covering `ping()`, BC aliases, TLS
+  round-trip, `writeStart` query_id propagation, default-database
+  config, error paths, placeholder rejection, identifier
+  rejection, and object-lifecycle paths.
+
+### Fixed
+
+- `__construct` no longer `RETURN_TRUE`s on connection failure; the
+  half-constructed object never reaches userland.
+- Every `clientMap.at(key)` site routed through a `getClient()`
+  helper that throws `ClickHouseException` on miss. `ping()`
+  previously did the lookup outside its `try` block, so
+  `std::out_of_range` escaped the PHP boundary unhandled.
+- `__destruct` silently no-ops when no client was registered, and
+  calls `EndInsert()` if a `writeStart()` was left dangling so the
+  server doesn't see a half-open insert.
+- `write()` and `writeEnd()` reject calls without a matching
+  `writeStart()`. `writeEnd()` erases the in-progress flag only
+  after `EndInsert()` returns.
+- `Date` insert: dropped the `tm_gmtoff` shift so raw epoch ints
+  round-trip TZ-independently.
+- `Nullable(Enum8)` / `Nullable(Enum16)`: NULL rows no longer crash
+  inside `ColumnEnum::Append`.
+- `Decimal` reads apply column scale before formatting. A value
+  inserted as `12345.6789012345` now reads back as
+  `12345.6789012345`, not the unscaled storage integer.
+- `Int128` / `UInt128` string parse rejects malformed input and
+  detects overflow during accumulation.
+- `to_time_t()` uses `timegm()` instead of `mktime()` so Date and
+  DateTime string round-trips don't drift by the runner's TZ.
+- `FETCH_ONE` select returns the first row of the result, not the
+  first row of the last block.
+- `Tuple` insert iterates by tuple arity and validates per-row
+  arity, fixing an out-of-bounds read when row count differed
+  from arity.
+- Signed integer reads (Int8..Int64) cast through `zend_long`;
+  negative values no longer surface as huge unsigned numbers.
+- `tcp_keepalive_cnt`, `max_compression_chunk_size`, and endpoint
+  port bounds-check before truncating.
+- Unknown `compression` strings throw instead of silently
+  disabling compression.
+- HashTable leaks in `insert()` / `write()` error paths.
+- Six tests previously TODO-skipped now run.
+
+### Changed
+
+- LICENSE replaced with the canonical PHP-3.01 text (the file
+  previously held Apache 2.0, contradicting every source-file
+  header, `composer.json`, and the README's license section).
+- README documents `Tuple` insert as supported (was listed as
+  read-only).
+- README benchmark section dropped the unmaintained
+  `lizhichao/one-ck` comparison column.
+
+### For contributors
+
+- Vendored `clickhouse-cpp` v2.6.1 patched for two upstream bugs:
+  `Client::Impl::BeginInsert` was dropping `query_id` from the wire
+  packet, and `ColumnStringBlock::AppendUnsafe` called `memcpy`
+  with a NULL source on empty `string_view`. Both documented in
+  `lib/clickhouse-cpp/LOCAL_PATCHES.md`. The empty-string fix has
+  an upstream PR at clickhouse-cpp#489.
+- `CONTRIBUTING.md` spells out the LOCAL_PATCHES.md re-apply step
+  on lib bumps.
+- ASan CI job is gating, not informational. Switched to
+  `-shared-libasan` so `__cxa_throw` interception works across the
+  PHP / extension dlopen boundary.
+- Compile warnings cleared on PHP 7.4-8.5
+  (`-Wunused-but-set-variable`, `-Wswitch`,
+  `-Wmaybe-uninitialized`).
+- typesToPhp.cpp: integer arms in `insertColumn` and
+  `convertToZval`, the five `Map(K, V)` insert permutations, and
+  the `DateTime` / `Date` / `Date32` read paths collapsed into
+  templated helpers.
+- Dead-code sweep: `FAST_ZPP` dead arms (never defined), unused
+  macros in `php7_wrapper.h`, TSRM scaffolding in
+  `php_clickhouse.h`, commented-out `clickhouse_version`, unused
+  `<iostream>` includes, 18 unreachable `break;` after `return;`.
+  Roughly 750 lines net removed.
+- Migrated remaining `SC_HASHTABLE_FOREACH_START2` call sites to
+  `ZEND_HASH_FOREACH_*` directly. The old macro silently dropped
+  integer keys.
+- Renamed `clientInsertBlack` to `clientInsertBlock` (Block typo).
+
+## [0.5.0] - 2026-04-25
+
+This release renames the extension from `SeasClick` to `php_clickhouse`,
+upgrades the vendored client library to ClickHouse/clickhouse-cpp v2.6.1,
+and adds significant new functionality. The original SeasClick project
+(SeasX/SeasClick on GitHub) appears unmaintained — its last accepted
+external PR is from 2020. php_clickhouse is a soft fork that goes its
+own way.
+
+### Added
+
+- New PHP class names: `ClickHouse` and `ClickHouseException`. The old
+  `SeasClick` and `SeasClickException` names are kept as class aliases
+  for this release cycle and removed in 1.0.
+- New column types backed by clickhouse-cpp v2.6.1:
+  - `Date32`
+  - `Time` and `Time64(N)` (requires ClickHouse 25.x or later on the
+    server)
+  - `DateTime64(N[, timezone])`
+  - `Int128`, `UInt128`, `Decimal128(P, S)` — round-trip as decimal
+    strings since PHP integers are 64-bit
+  - `LowCardinality(String)` and `LowCardinality(FixedString(N))`
+  - `Map(K, V)` for `(String, String)`, `(String, Int64)`,
+    `(String, UInt64)`, `(String, Float64)`, and `(Int64, String)`
+- New connection / client options on the `__construct` config array:
+  - `compression` now accepts `"lz4"`, `"zstd"`, `"none"` in addition
+    to `true` / `false`. ZSTD is roughly 1.5-2x denser than LZ4 on
+    typical row payloads.
+  - `tcp_nodelay`, `tcp_keepalive`, `tcp_keepalive_idle`,
+    `tcp_keepalive_intvl`, `tcp_keepalive_cnt`
+  - `send_timeout` (companion to `receive_timeout` and
+    `connect_timeout`)
+  - `endpoints` — list of `[{host, port}, ...]` for round-robin
+    failover. The lib walks the list in order on connect failure.
+  - `max_compression_chunk_size`
+- `query_id` accepted as a final optional argument on `select()`,
+  `insert()`, `writeStart()`, and `execute()`. The id propagates to
+  ClickHouse's `system.query_log` and is useful for `KILL QUERY` and
+  trace correlation.
+- TLS support: build with `--enable-clickhouse-openssl`. New config
+  keys: `ssl`, `ssl_skip_verify`, `ssl_ca_files` (string or array),
+  `ssl_ca_directory`, `ssl_use_default_ca`. Building without the flag
+  and passing `ssl: true` raises `ClickHouseException` rather than
+  silently downgrading to plaintext.
+- Test suite: phpt coverage for Date32, Int128/UInt128/Decimal128,
+  LowCardinality, Map (all five K/V combinations), DateTime64,
+  Time/Time64, query_id propagation, ZSTD compression, multi-endpoint
+  failover. Driven by `tests/_clickhouse.inc` which reads
+  `CLICKHOUSE_HOST` / `CLICKHOUSE_PORT` / `CLICKHOUSE_USER` /
+  `CLICKHOUSE_PASSWD` from the environment and skips cleanly when no
+  server is reachable.
+- README documents installation via [PIE](https://github.com/php/pie),
+  a Docker recipe for a local ClickHouse server, the full configuration
+  surface, and a benchmark table comparing against
+  [smi2/phpClickHouse](https://github.com/smi2/phpClickHouse).
+
+### Changed
+
+- Vendored ClickHouse client library bumped from artpaul-fork v1.x to
+  the official ClickHouse/clickhouse-cpp v2.6.1.
+- Build now requires C++17 (clickhouse-cpp v2 hard requirement).
+- Optional dependencies are vendored under `lib/clickhouse-cpp/contrib/`:
+  cityhash, lz4, abseil int128, zstd. No external `pkg-config` lookups.
+- Insert API rewired from the v1.x `InsertQuery` / `InsertData` /
+  `InsertDataEnd` triplet to v2.x `BeginInsert` / `SendInsertBlock` /
+  `EndInsert`. PHP-side `writeStart()` / `write()` / `writeEnd()`
+  signatures unchanged.
+- `ClientOptions` field renames carried through to the PHP-side
+  config: `socket_receive_timeout` / `socket_connect_timeout` are now
+  `connection_recv_timeout` / `connection_connect_timeout` internally.
+  The old PHP-level keys `receive_timeout` / `connect_timeout` are
+  preserved.
+- PHP version floor: 7.1+ (drops 5.x and 7.0). PHP 7.1, 7.2, 7.4, 8.3,
+  8.4, 8.5 all build and pass tests.
+- Benchmarks moved from `tests/bench_mark/` to top-level `bench/`.
+- Tests `006`-`008` now read connection settings from the environment
+  via the shared `_clickhouse.inc` helper, drop their tables at the
+  top, and (`008`) order results deterministically. The old
+  `host=clickhouse` literal is still the default if no env is set.
+
+### Fixed
+
+- `Decimal` / `Decimal32` / `Decimal64` column reads previously
+  downcast to `ColumnFloat32` / `ColumnFloat64` and SEGV'd on any
+  pure `SELECT toDecimal128(42, 5)`. All four `Decimal*` variants now
+  go through `ColumnDecimal` and round-trip as scaled-integer strings.
+- `ZEND_ACC_DTOR` had no equivalent in PHP 8 (the bit value got
+  reassigned to `ZEND_ACC_VARIADIC`); the old fallback was crashing
+  `zend_register_functions` on extension load on every PHP 8.x build.
+- `<cstdint>` now explicitly included in vendored `types.h` and
+  `numeric.h`. Older GCC pulled it in transitively from `<atomic>` /
+  `<vector>`; GCC 15 stopped doing that and refused to compile
+  `uint8_t` / `int16_t` / `uint64_t` references. Backports the same
+  fix as upstream's [PR #18](https://github.com/SeasX/SeasClick/pull/18).
+- Two arg-info / zend_parse_parameters mismatches that PHP 8.3
+  flagged as fatal: `select()` declared 3 required args while ZPP
+  accepted 1, and `execute()` declared 2 while ZPP accepted 1. Both
+  now declare 1 required argument.
+
+### Removed
+
+- PHP 5 support: branches, `TSRMLS_*` macros, `MAKE_STD_ZVAL`-style
+  scaffolding deleted. Net -155 lines from the wrapper layer.
+- `package.xml`: PECL is closed; PIE pulls metadata from `composer.json`.
+- `EXPERIMENTAL`: PECL stability marker, empty file.
+- `travis/`: Travis CI is no longer used. CI lives on GitHub Actions.
+- `config.w32.without` Windows stub: replaced with a `config.w32` that
+  emits a clear "unsupported" warning. Full Windows build of the
+  vendored zstd + absl + lz4 + cityhash is a separate project.
+
+[Unreleased]: https://github.com/iliaal/php_clickhouse/compare/0.8.5...HEAD
+[0.8.5]: https://github.com/iliaal/php_clickhouse/releases/tag/0.8.5
+[0.8.1]: https://github.com/iliaal/php_clickhouse/releases/tag/0.8.1
+[0.8.0]: https://github.com/iliaal/php_clickhouse/releases/tag/0.8.0
+[0.7.0]: https://github.com/iliaal/php_clickhouse/releases/tag/0.7.0
+[0.6.0]: https://github.com/iliaal/php_clickhouse/releases/tag/0.6.0
+[0.5.0]: https://github.com/iliaal/php_clickhouse/releases/tag/0.5.0
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
new file mode 100644
index 0000000..af41431
--- /dev/null
+++ b/CONTRIBUTING.md
@@ -0,0 +1,83 @@
+# Contributing to php_clickhouse
+
+## Requirements
+
+- PHP 7.1 or later (8.x recommended; debug builds are useful for
+  development: `--enable-debug`)
+- C++17-capable compiler: GCC 8+, Clang 7+
+- `phpize` and `php-config` (from `php-dev` / `php8.x-dev`)
+- GNU Make
+
+The vendored `lib/clickhouse-cpp/` includes its own copies of LZ4,
+ZSTD, abseil's int128, and CityHash. No external libraries are
+required to build the default extension. Building with TLS support
+needs OpenSSL development headers (`libssl-dev` on Debian/Ubuntu).
+
+## Bug reports
+
+Use the [GitHub issue tracker](https://github.com/iliaal/php_clickhouse/issues).
+Include:
+
+- PHP version (`php -v`)
+- php_clickhouse version (`php -r 'echo phpversion("clickhouse");'`)
+- ClickHouse server version
+- Operating system and compiler version
+- Minimal reproducing code
+- Expected vs actual behavior
+- Any error messages, exceptions, or crash output
+
+Before filing, try to reproduce against the latest `master` branch.
+
+For security issues, do **not** file a public issue. See
+[SECURITY.md](SECURITY.md).
+
+## Pull requests
+
+1. Fork and clone the repo
+2. Create a topic branch off `master`
+3. Build clean (`./configure --enable-clickhouse && make`) and run
+   the test suite. Tests need a reachable ClickHouse server; the
+   README has a one-liner Docker recipe.
+4. Add `.phpt` coverage for any new behavior. Tests skip cleanly if
+   no server is reachable, so adding new ones doesn't punish people
+   without a local CH.
+5. Open a PR. Include the rationale, the test you added, and any
+   benchmark deltas if your change is in a hot path.
+
+## Code style
+
+- Two-space indent for PHP, four-space for C/C++.
+- File headers carry the standard PHP-3.01 license block plus an
+  `| Author: ` line. New files credit yourself there too.
+- Prefer adding to `clickhouse.cpp` and `typesToPhp.cpp` over creating
+  new translation units unless there's a structural reason. The build
+  rebuilds everything in one shot anyway and the symbol surface is
+  clearer when concentrated.
+- C++ exceptions caught at the PHP boundary should go through
+  `sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce, ...)`
+  so users see `ClickHouseException` and not generic `Exception`.
+
+## Vendored library updates
+
+`lib/clickhouse-cpp/` tracks a specific tag of
+[ClickHouse/clickhouse-cpp](https://github.com/ClickHouse/clickhouse-cpp).
+When bumping it:
+
+1. Drop the new release into `lib/clickhouse-cpp/clickhouse/` and
+   `lib/clickhouse-cpp/contrib/` (keep cityhash, lz4, zstd, absl;
+   skip gtest, ut, bench, tests).
+2. Re-apply (or drop, if upstreamed) every patch listed in
+   `lib/clickhouse-cpp/LOCAL_PATCHES.md`. Bumping without this step
+   silently regresses the fixes the test suite depends on.
+3. Update the source list in `config.m4`. The list there is
+   alphabetical by directory; keep that order.
+4. Run the full test suite against ClickHouse `latest` (the test
+   server in CI is `clickhouse/clickhouse-server:latest`).
+5. Note any breaking changes in `CHANGELOG.md` under the unreleased
+   section.
+
+## Releases
+
+Maintainers run `/release` (see `.claude/commands/release.md` for the
+playbook). The summary: bump version in `php_clickhouse.h`,
+date-stamp the top section of `CHANGELOG.md`, tag, push.
diff --git a/CREDITS b/CREDITS
index d2bc1f7..f4bd044 100644
--- a/CREDITS
+++ b/CREDITS
@@ -1,3 +1,7 @@
-SeasClick
+php_clickhouse
+Ilia Alshanetsky <ilia@ilia.ws>
 
-SeasX Group
\ No newline at end of file
+Original SeasClick author: SeasX Group <ahhhh.wang@gmail.com>
+
+Vendored ClickHouse C++ client (lib/clickhouse-cpp/): ClickHouse, Inc.
+and contributors, Apache License 2.0.
diff --git a/EXPERIMENTAL b/EXPERIMENTAL
deleted file mode 100644
index e69de29..0000000
diff --git a/LICENSE b/LICENSE
index 261eeb9..0611d38 100644
--- a/LICENSE
+++ b/LICENSE
@@ -1,201 +1,67 @@
-                                 Apache License
-                           Version 2.0, January 2004
-                        http://www.apache.org/licenses/
-
-   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
-
-   1. Definitions.
-
-      "License" shall mean the terms and conditions for use, reproduction,
-      and distribution as defined by Sections 1 through 9 of this document.
-
-      "Licensor" shall mean the copyright owner or entity authorized by
-      the copyright owner that is granting the License.
-
-      "Legal Entity" shall mean the union of the acting entity and all
-      other entities that control, are controlled by, or are under common
-      control with that entity. For the purposes of this definition,
-      "control" means (i) the power, direct or indirect, to cause the
-      direction or management of such entity, whether by contract or
-      otherwise, or (ii) ownership of fifty percent (50%) or more of the
-      outstanding shares, or (iii) beneficial ownership of such entity.
-
-      "You" (or "Your") shall mean an individual or Legal Entity
-      exercising permissions granted by this License.
-
-      "Source" form shall mean the preferred form for making modifications,
-      including but not limited to software source code, documentation
-      source, and configuration files.
-
-      "Object" form shall mean any form resulting from mechanical
-      transformation or translation of a Source form, including but
-      not limited to compiled object code, generated documentation,
-      and conversions to other media types.
-
-      "Work" shall mean the work of authorship, whether in Source or
-      Object form, made available under the License, as indicated by a
-      copyright notice that is included in or attached to the work
-      (an example is provided in the Appendix below).
-
-      "Derivative Works" shall mean any work, whether in Source or Object
-      form, that is based on (or derived from) the Work and for which the
-      editorial revisions, annotations, elaborations, or other modifications
-      represent, as a whole, an original work of authorship. For the purposes
-      of this License, Derivative Works shall not include works that remain
-      separable from, or merely link (or bind by name) to the interfaces of,
-      the Work and Derivative Works thereof.
-
-      "Contribution" shall mean any work of authorship, including
-      the original version of the Work and any modifications or additions
-      to that Work or Derivative Works thereof, that is intentionally
-      submitted to Licensor for inclusion in the Work by the copyright owner
-      or by an individual or Legal Entity authorized to submit on behalf of
-      the copyright owner. For the purposes of this definition, "submitted"
-      means any form of electronic, verbal, or written communication sent
-      to the Licensor or its representatives, including but not limited to
-      communication on electronic mailing lists, source code control systems,
-      and issue tracking systems that are managed by, or on behalf of, the
-      Licensor for the purpose of discussing and improving the Work, but
-      excluding communication that is conspicuously marked or otherwise
-      designated in writing by the copyright owner as "Not a Contribution."
-
-      "Contributor" shall mean Licensor and any individual or Legal Entity
-      on behalf of whom a Contribution has been received by Licensor and
-      subsequently incorporated within the Work.
-
-   2. Grant of Copyright License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      copyright license to reproduce, prepare Derivative Works of,
-      publicly display, publicly perform, sublicense, and distribute the
-      Work and such Derivative Works in Source or Object form.
-
-   3. Grant of Patent License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      (except as stated in this section) patent license to make, have made,
-      use, offer to sell, sell, import, and otherwise transfer the Work,
-      where such license applies only to those patent claims licensable
-      by such Contributor that are necessarily infringed by their
-      Contribution(s) alone or by combination of their Contribution(s)
-      with the Work to which such Contribution(s) was submitted. If You
-      institute patent litigation against any entity (including a
-      cross-claim or counterclaim in a lawsuit) alleging that the Work
-      or a Contribution incorporated within the Work constitutes direct
-      or contributory patent infringement, then any patent licenses
-      granted to You under this License for that Work shall terminate
-      as of the date such litigation is filed.
-
-   4. Redistribution. You may reproduce and distribute copies of the
-      Work or Derivative Works thereof in any medium, with or without
-      modifications, and in Source or Object form, provided that You
-      meet the following conditions:
-
-      (a) You must give any other recipients of the Work or
-          Derivative Works a copy of this License; and
-
-      (b) You must cause any modified files to carry prominent notices
-          stating that You changed the files; and
-
-      (c) You must retain, in the Source form of any Derivative Works
-          that You distribute, all copyright, patent, trademark, and
-          attribution notices from the Source form of the Work,
-          excluding those notices that do not pertain to any part of
-          the Derivative Works; and
-
-      (d) If the Work includes a "NOTICE" text file as part of its
-          distribution, then any Derivative Works that You distribute must
-          include a readable copy of the attribution notices contained
-          within such NOTICE file, excluding those notices that do not
-          pertain to any part of the Derivative Works, in at least one
-          of the following places: within a NOTICE text file distributed
-          as part of the Derivative Works; within the Source form or
-          documentation, if provided along with the Derivative Works; or,
-          within a display generated by the Derivative Works, if and
-          wherever such third-party notices normally appear. The contents
-          of the NOTICE file are for informational purposes only and
-          do not modify the License. You may add Your own attribution
-          notices within Derivative Works that You distribute, alongside
-          or as an addendum to the NOTICE text from the Work, provided
-          that such additional attribution notices cannot be construed
-          as modifying the License.
-
-      You may add Your own copyright statement to Your modifications and
-      may provide additional or different license terms and conditions
-      for use, reproduction, or distribution of Your modifications, or
-      for any such Derivative Works as a whole, provided Your use,
-      reproduction, and distribution of the Work otherwise complies with
-      the conditions stated in this License.
-
-   5. Submission of Contributions. Unless You explicitly state otherwise,
-      any Contribution intentionally submitted for inclusion in the Work
-      by You to the Licensor shall be under the terms and conditions of
-      this License, without any additional terms or conditions.
-      Notwithstanding the above, nothing herein shall supersede or modify
-      the terms of any separate license agreement you may have executed
-      with Licensor regarding such Contributions.
-
-   6. Trademarks. This License does not grant permission to use the trade
-      names, trademarks, service marks, or product names of the Licensor,
-      except as required for reasonable and customary use in describing the
-      origin of the Work and reproducing the content of the NOTICE file.
-
-   7. Disclaimer of Warranty. Unless required by applicable law or
-      agreed to in writing, Licensor provides the Work (and each
-      Contributor provides its Contributions) on an "AS IS" BASIS,
-      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
-      implied, including, without limitation, any warranties or conditions
-      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
-      PARTICULAR PURPOSE. You are solely responsible for determining the
-      appropriateness of using or redistributing the Work and assume any
-      risks associated with Your exercise of permissions under this License.
-
-   8. Limitation of Liability. In no event and under no legal theory,
-      whether in tort (including negligence), contract, or otherwise,
-      unless required by applicable law (such as deliberate and grossly
-      negligent acts) or agreed to in writing, shall any Contributor be
-      liable to You for damages, including any direct, indirect, special,
-      incidental, or consequential damages of any character arising as a
-      result of this License or out of the use or inability to use the
-      Work (including but not limited to damages for loss of goodwill,
-      work stoppage, computer failure or malfunction, or any and all
-      other commercial damages or losses), even if such Contributor
-      has been advised of the possibility of such damages.
-
-   9. Accepting Warranty or Additional Liability. While redistributing
-      the Work or Derivative Works thereof, You may choose to offer,
-      and charge a fee for, acceptance of support, warranty, indemnity,
-      or other liability obligations and/or rights consistent with this
-      License. However, in accepting such obligations, You may act only
-      on Your own behalf and on Your sole responsibility, not on behalf
-      of any other Contributor, and only if You agree to indemnify,
-      defend, and hold each Contributor harmless for any liability
-      incurred by, or claims asserted against, such Contributor by reason
-      of your accepting any such warranty or additional liability.
-
-   END OF TERMS AND CONDITIONS
-
-   APPENDIX: How to apply the Apache License to your work.
-
-      To apply the Apache License to your work, attach the following
-      boilerplate notice, with the fields enclosed by brackets "[]"
-      replaced with your own identifying information. (Don't include
-      the brackets!)  The text should be enclosed in the appropriate
-      comment syntax for the file format. We also recommend that a
-      file or class name and description of purpose be included on the
-      same "printed page" as the copyright notice for easier
-      identification within third-party archives.
-
-   Copyright [yyyy] [name of copyright owner]
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
+--------------------------------------------------------------------
+                  The PHP License, version 3.01
+Copyright (c) 1999 - 2026 The PHP Group. All rights reserved.
+--------------------------------------------------------------------
+
+Redistribution and use in source and binary forms, with or without
+modification, is permitted provided that the following conditions
+are met:
+
+  1. Redistributions of source code must retain the above copyright
+     notice, this list of conditions and the following disclaimer.
+
+  2. Redistributions in binary form must reproduce the above copyright
+     notice, this list of conditions and the following disclaimer in
+     the documentation and/or other materials provided with the
+     distribution.
+
+  3. The name "PHP" must not be used to endorse or promote products
+     derived from this software without prior written permission. For
+     written permission, please contact group@php.net.
+
+  4. Products derived from this software may not be called "PHP", nor
+     may "PHP" appear in their name, without prior written permission
+     from group@php.net.  You may indicate that your software works in
+     conjunction with PHP by saying "Foo for PHP" instead of calling
+     it "PHP Foo" or "phpfoo"
+
+  5. The PHP Group may publish revised and/or new versions of the
+     license from time to time. Each version will be given a
+     distinguishing version number.
+     Once covered code has been published under a particular version
+     of the license, you may always continue to use it under the terms
+     of that version. You may also choose to use such covered code
+     under the terms of any subsequent version of the license
+     published by the PHP Group. No one other than the PHP Group has
+     the right to modify the terms applicable to covered code created
+     under this License.
+
+  6. Redistributions of any form whatsoever must retain the following
+     acknowledgment:
+     "This product includes PHP software, freely available from
+     <http://www.php.net/software/>".
+
+THIS SOFTWARE IS PROVIDED BY THE PHP DEVELOPMENT TEAM ``AS IS'' AND
+ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
+PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE PHP
+DEVELOPMENT TEAM OR ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+OF THE POSSIBILITY OF SUCH DAMAGE.
+
+--------------------------------------------------------------------
+
+For more information on the PHP License, see
+<https://opensource.org/license/php-3-01> and
+<https://www.php.net/license/3_01.txt>.
+
+The vendored library at lib/clickhouse-cpp/ is
+ClickHouse/clickhouse-cpp, distributed under the Apache License 2.0.
+See lib/clickhouse-cpp/LICENSE for that license text and
+lib/clickhouse-cpp/contrib/ for the licenses covering the bundled
+LZ4, ZSTD, abseil, and CityHash sources.
diff --git a/README.md b/README.md
index 436ee47..635dbe1 100644
--- a/README.md
+++ b/README.md
@@ -1,90 +1,457 @@
-SeasClick
-=====
-[![Build Status](https://travis-ci.org/SeasX/SeasClick.svg?branch=master)](https://travis-ci.org/SeasX/SeasClick)
-
-PHP client for [Yandex ClickHouse](https://clickhouse.yandex/)，Based on [ClickHouse C++ client](https://github.com/aiwhj/clickhouse-cpp)
-
-## ClickHouse
-* [What is ClickHouse](https://clickhouse.yandex/docs/en/)
-* [ClickHouse Performance](https://clickhouse.yandex/docs/en/introduction/performance/)
-* [Performance comparison with MySQL](https://clickhouse.yandex/benchmark.html#[%22100000000%22,[%22ClickHouse%22,%22MySQL%22],[%220%22,%221%22]])
-
-## Supported data types
-
-* Array(T)
-    **Multidimensional arrays are not supported at this time**
-* Date
-* DateTime
-* Enum8, Enum16
-* FixedString(N)
-* Float32, Float64
-* Nullable(T)
-* String
-* UInt8, UInt16, UInt32, UInt64, Int8, Int16, Int32, Int64
-
-## Supported PHP version
-PHP 5.4+
-
-## Performance
-![image](https://github.com/SeasX/SeasClick/raw/master/tests/bench_mark/bench_mark.png)
-
-This performance test [demo](https://github.com/SeasX/SeasClick/blob/master/tests/bench_mark/bench_mark.php) is compared to [phpclickhouse](https://github.com/smi2/phpClickHouse)
-
-## Install
-```ssh
-git clone https://github.com/SeasX/SeasClick.git
-cd SeasClick
+# php_clickhouse
+
+[![Tests](https://github.com/iliaal/php_clickhouse/actions/workflows/tests.yml/badge.svg)](https://github.com/iliaal/php_clickhouse/actions/workflows/tests.yml)
+[![Version](https://img.shields.io/github/v/release/iliaal/php_clickhouse)](https://github.com/iliaal/php_clickhouse/releases)
+[![License: PHP-3.01](https://img.shields.io/badge/License-PHP--3.01-green.svg)](http://www.php.net/license/3_01.txt)
+[![Follow @iliaa](https://img.shields.io/badge/Follow-@iliaa-000000?style=flat&logo=x&logoColor=white)](https://x.com/intent/follow?screen_name=iliaa)
+
+![php_clickhouse: native binary protocol vs HTTP](images/php_clickhouse-hero.jpg)
+
+Native PHP extension for [ClickHouse](https://clickhouse.com/), built on the official [ClickHouse/clickhouse-cpp](https://github.com/ClickHouse/clickhouse-cpp) v2.6.1 client. Speaks the native binary TCP protocol with LZ4 / ZSTD compression and optional TLS, picking up where [SeasX/SeasClick](https://github.com/SeasX/SeasClick) left off in 2020. 30-40% faster than HTTP-based clients on heavy workloads, with modern types (Date32, Time64, Decimal128, LowCardinality, Map), multi-endpoint failover, and structured exceptions.
+
+## Why this fork?
+
+[SeasX/SeasClick](https://github.com/SeasX/SeasClick) was the canonical native PHP ClickHouse extension and stopped accepting PRs in 2020. Several follow-up PRs there have been pending for years. This fork:
+
+- Renames the extension to `php_clickhouse` (module `clickhouse`, classes `ClickHouse` / `ClickHouseException`)
+- Upgrades the vendored client from artpaul-fork v1.x to the official ClickHouse/clickhouse-cpp v2.6.1
+- Adds Date32 / Time / Time64 / DateTime64 / Int128 / UInt128 / Decimal128 / LowCardinality / Map column types, multi-endpoint failover, ZSTD compression, query_id propagation, and TLS
+- Ships an updated test suite, CI, PIE-based packaging, and benchmarks
+
+The original `SeasClick` and `SeasClickException` class names continue to work as deprecated aliases.
+
+Method signatures, return types, and class properties are declared with PHP types via a stub-driven arginfo workflow (`clickhouse.stub.php` → `clickhouse_arginfo.h`). Reflection, IDE completion, and static analyzers (PHPStan, Psalm) see the typed surface without manual stubs.
+
+## 🚀 Install
+
+Via [PIE](https://github.com/php/pie) (the PHP Foundation's PECL successor):
+
+```sh
+pie install iliaal/php_clickhouse
+```
+
+With TLS support:
+
+```sh
+pie install iliaal/php_clickhouse --enable-clickhouse-openssl
+```
+
+> Bare `php:X.Y-cli` Docker images lack `/usr/bin/unzip`, which composer needs to extract PIE's prebuilt `.so` zip. Run `apt-get install -y unzip` before `pie install`, otherwise composer falls back to PHP's ZipArchive and PIE fails with `ExtensionBinaryNotFound`. Host installs that already have `unzip` are fine.
+
+Building from source:
+
+```sh
+git clone https://github.com/iliaal/php_clickhouse.git
+cd php_clickhouse
 phpize
-./configure
-make && make install
+./configure                              # default build
+./configure --enable-clickhouse-openssl  # with TLS, requires OpenSSL development headers
+                                         #   (libssl-dev on Debian/Ubuntu, openssl-devel
+                                         #    on RHEL/Fedora, openssl-dev on Alpine)
+make && sudo make install
+```
+
+Add `extension=clickhouse.so` to your `php.ini`. The build needs a C++17-capable compiler (GCC 8+, Clang 7+, MSVC 2019+); LZ4, ZSTD, abseil-int128, and CityHash are vendored under `lib/clickhouse-cpp/contrib/`.
+
+### Platforms
+
+| Platform | Status | Notes |
+|----------|--------|-------|
+| Linux NTS | first-class | PHP 7.4 through 8.5, CI matrix |
+| Linux ZTS | first-class | PHP 8.4 ZTS in CI |
+| Windows (NTS, TS) | supported | PHP 8.4 x86 / x64 in CI; pre-built `.dll` released via PIE |
+| macOS | unverified | should build (POSIX path); no CI lane |
+
+Per-Client state lives on the `zend_object` itself (custom `create_object` / `free_obj` handlers), so ZTS works without locking. There is no module-global state to thread-isolate.
+
+### Test server
+
+For development and integration tests, the simplest path is the official ClickHouse server image:
+
+```sh
+docker run -d --name clickhouse-test \
+    --ulimit nofile=262144:262144 \
+    -p 9000:9000 -p 8123:8123 -p 9440:9440 \
+    -e CLICKHOUSE_USER=test \
+    -e CLICKHOUSE_PASSWORD=test \
+    clickhouse/clickhouse-server:latest
 ```
 
-## Example
+Stop and clean up: `docker rm -f clickhouse-test`.
+
+## 🛠️ Quick example
 
 ```php
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => 9000,
-    "compression" => true
-];
+$ch = new ClickHouse([
+    "host"        => "127.0.0.1",
+    "port"        => 9000,
+    "database"    => "test",
+    "user"        => "default",
+    "passwd"      => "",
+    "compression" => "lz4",   // or "zstd" / true / false
+]);
 
-clientTest($config);
+$ch->execute("CREATE TABLE IF NOT EXISTS events (
+    id UInt32, ts DateTime64(3), tag LowCardinality(String)
+) ENGINE = Memory");
 
-function clientTest($config)
-{
-    $deleteTable = true;
-    $client = new SeasClick($config);
+$ch->insert("events", ["id", "ts", "tag"], [
+    [1, time(), "alpha"],
+    [2, time(), "beta"],
+]);
 
-    $client->execute("CREATE DATABASE IF NOT EXISTS test");
+// Filter by an in-memory list of IDs without bloating the SQL — the
+// server reads `ext_ids` as a named temp table for this query only.
+$hits = $ch->selectWithExternalData(
+    "SELECT id, tag FROM events WHERE id IN ext_ids",
+    [["name" => "ext_ids",
+      "columns" => ["id" => "UInt32"],
+      "rows" => [[1], [42], [1337]]]]
+);
 
-    testArray($client, $deleteTable);
+foreach ($ch->select("SELECT id, ts, tag FROM events ORDER BY id",
+                     [], ClickHouse::DATE_AS_STRINGS) as $row) {
+    print_r($row);
 }
+```
+
+## 📦 Supported data types
+
+* `Array(T)` including nested server-side placeholder values such as `Array(Array(UInt32))`
+* `Date`, `Date32`, `DateTime`, `DateTime64(N[, timezone])`
+* `Time`, `Time64(N)` (server side requires ClickHouse 25.x or later)
+* `Decimal`, `Decimal32`, `Decimal64`, `Decimal128(P, S)` (read/write as scaled-integer strings)
+* `Enum8`, `Enum16`
+* `FixedString(N)`
+* `Float32`, `Float64`
+* `Int8` … `Int64`, `UInt8` … `UInt64`
+* `Int128`, `UInt128` (round-trip as decimal strings; PHP integers are 64-bit)
+* `IPv4`, `IPv6`
+* `LowCardinality(String)`, `LowCardinality(FixedString(N))`, `LowCardinality(Nullable(String))`, `LowCardinality(Nullable(FixedString(N)))`
+* `Map(K, V)` over scalar K and V (`String`, `Int8` through `Int64`, `UInt8` through `UInt64`, `Float32`, `Float64`, `UUID`) plus `LowCardinality(String)` keys and values. `Map(LowCardinality(K), V)` reads are not yet decoded by the vendored client; writes succeed and the data is queryable server side.
+* `Point`, `Ring`, `Polygon`, `MultiPolygon` (geo)
+* `Nullable(T)`
+* `String`
+* `Tuple` (read-only)
+* `UUID`
+
+## Configuration reference
+
+All keys go in the array passed to `new ClickHouse([...])`.
+
+### Connection
+
+| Key | Type | Default | Description |
+|---|---|---|---|
+| `host` | string | `127.0.0.1` | Server host |
+| `port` | int | `9000` | Native TCP port (or `9440` for TLS) |
+| `database` | string | `default` | Default database |
+| `user` | string | `default` | Username |
+| `passwd` | string | (empty) | Password |
+| `endpoints` | array | (none) | List of `[{host, port}, ...]` for round-robin failover. Tried in order on connect failure. |
+
+### Compression
+
+| Key | Type | Default | Description |
+|---|---|---|---|
+| `compression` | bool / string | `false` | `false`/`"none"` = uncompressed; `true`/`"lz4"` = LZ4 (fast); `"zstd"` = ZSTD (denser) |
+| `max_compression_chunk_size` | int | `65535` | Block size used by the compressor |
+
+### Timeouts and retry
+
+| Key | Type | Default | Description |
+|---|---|---|---|
+| `connect_timeout` | int (sec) | `5` | TCP connect deadline |
+| `connect_timeout_ms` | int (ms) | (none) | Sub-second connect deadline; overrides the seconds key when set |
+| `receive_timeout` | int (sec) | `0` | Read deadline (0 = no timeout) |
+| `receive_timeout_ms` | int (ms) | (none) | Sub-second read deadline; overrides the seconds key when set |
+| `send_timeout` | int (sec) | `0` | Write deadline |
+| `send_timeout_ms` | int (ms) | (none) | Sub-second write deadline; overrides the seconds key when set |
+| `retry_count` | int | `1` | Send retries on transient failure |
+| `retry_timeout` | int (sec) | `5` | Sleep between retries |
+| `tcp_nodelay` | bool | `true` | TCP_NODELAY |
+| `tcp_keepalive` | bool | `false` | TCP keepalive |
+| `tcp_keepalive_idle` | int (sec) | `60` | Idle time before first keepalive probe |
+| `tcp_keepalive_intvl` | int (sec) | `5` | Interval between probes |
+| `tcp_keepalive_cnt` | int | `3` | Failed probes before declaring dead |
+
+### TLS (build with `--enable-clickhouse-openssl`)
+
+| Key | Type | Default | Description |
+|---|---|---|---|
+| `ssl` | bool | `false` | Enable TLS |
+| `ssl_min_protocol_version` | string | `tls1.2` | Minimum protocol; one of `tls1.0`, `tls1.1`, `tls1.2`, `tls1.3` |
+| `ssl_skip_verify` | bool | `false` | Skip cert validation; dev only |
+| `ssl_use_default_ca` | bool | `true` | Trust the system CA bundle |
+| `ssl_ca_files` | string \| array | (none) | PEM CA file path(s) |
+| `ssl_ca_directory` | string | (none) | OpenSSL hashed-cert directory |
+
+Building without `--enable-clickhouse-openssl` and passing `ssl => true` raises `ClickHouseException` at construct time.
+
+## Methods
+
+```php
+$ch = new ClickHouse(array $config);
+
+// Schema / DDL
+$ch->execute(string $sql,
+             array $params = [],
+             string $query_id = "",
+             array $settings = []);
+
+// Read
+$rows = $ch->select(string $sql,
+                    array $params = [],
+                    int $fetch_mode = 0,
+                    string $query_id = "",
+                    array $settings = []);
+
+// Read with external in-memory tables sent alongside the query.
+// Each entry: ['name' => 'ext_x', 'columns' => ['c' => 'Type', ...],
+//              'rows' => [[...], [...], ...]]. The query body references
+// the external table by name (e.g. `WHERE id IN ext_ids`). Keeps the
+// SQL small when filtering by big lists; multiple externals per call.
+$rows = $ch->selectWithExternalData(string $sql,
+                                    array  $externals,
+                                    array  $params = [],
+                                    int    $fetch_mode = 0,
+                                    string $query_id = "",
+                                    array  $settings = []);
+
+// Bulk insert (entire dataset in one call)
+$ch->insert(string $table, array $columns, array $values,
+            string $query_id = "",
+            array $settings = []);
+
+// Same as insert(), but rows are associative arrays and the column list
+// is derived from the first row's keys.
+$ch->insertAssoc(string $table, array $rows,
+                 string $query_id = "",
+                 array $settings = []);
+
+// Stream-parse a TSV / CSV file (or any PHP stream resource) and insert
+// the rows in batches of $batch_rows. Bytes are parsed in C++; only
+// $batch_rows worth of per-column zvals exist at any time, so inputs
+// larger than memory work fine. Formats match selectToStream()'s set
+// (TabSeparated, TabSeparatedWithNames, CSV, CSVWithNames). Literal `\N`
+// in either format is the NULL marker; empty CSV cells become empty
+// strings. Returns rows inserted.
+$f = fopen("/tmp/events.csv", "rb");
+$n = $ch->insertFromStream(string $table, array $columns, mixed $stream,
+                           string $format = "TabSeparated",
+                           int    $batch_rows = 10000,
+                           string $query_id = "",
+                           array  $settings = []);
+fclose($f);
+
+// Streaming insert (open block, append, close)
+$ch->writeStart(string $table, array $columns,
+                string $query_id = "",
+                array $settings = []);
+$ch->write(array $values);
+$ch->write(array $more_values);
+$ch->writeEnd();
+
+$ch->ping();          // returns true on success, throws on failure
+
+// Streaming reads (no full-result PHP array)
+$iter = $ch->selectStream(string $sql, array $params = [],
+                          string $query_id = "", array $settings = []);
+foreach ($iter as $row) { /* ... */ }   // ClickHouseRowIterator: Iterator+Countable
+
+$ch->selectStreamCallback(string $sql, callable $cb,
+                          array $params = [], string $query_id = "",
+                          array $settings = []);  // true per-row stream
+
+// Write rows straight to a PHP stream resource as TSV / CSV — bypasses
+// per-row PHP array assembly and userland callbacks; cells are
+// formatted block-by-block in C++ and flushed to the stream. Returns
+// rows written. Formats: TabSeparated (alias TSV), TabSeparatedWithNames
+// (alias TSVWithNames), CSV, CSVWithNames. Dates emit as ISO strings;
+// Decimal / Int128 / UInt128 as decimal strings. NULL = `\N` (TSV) /
+// empty (CSV). Array / Tuple / Map columns are rejected.
+$f = fopen("/tmp/events.tsv", "wb");
+$n = $ch->selectToStream(string $sql, array $params, mixed $stream,
+                         string $format = "TabSeparated",
+                         string $query_id = "", array $settings = []);
+fclose($f);
 
-function testArray($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.array_test (string_c String, array_c Array(Int8), arraynull_c Array(Nullable(String))) ENGINE = Memory");
-
-    $client->insert("test.array_test", [
-        'string_c', 'array_c', 'arraynull_c'
-    ], [
-        ['string_c1', [1, 2, 3], ['string']],
-        ['string_c2', [4, 5, 6], [null]]
-    ]);
-
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'string_c, array_c, arraynull_c',
-        'table' => 'test.array_test'
-    ]);
-    var_dump($result);
-
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.array_test'
-        ]);
-    }
+// smi2-style result wrapper: returns a ClickHouseStatement (Iterator,
+// Countable, ArrayAccess, JsonSerializable) carrying a per-call stats
+// snapshot. Use when you want fetchOne / fetchKeyPair / fetchColumn /
+// statistics() on the result, or want to keep stats around after
+// running other queries on the client. Plain $ch->select() is faster
+// when you just need the array.
+$stmt = $ch->selectStatement(string $sql, array $params = [],
+                             string $query_id = "", array $settings = []);
+foreach ($stmt as $row) { /* ... */ }
+$stmt[0]; count($stmt); json_encode($stmt);
+$stmt->fetchOne(); $stmt->fetchKeyPair(); $stmt->fetchColumn();
+$stmt->toArray(); $stmt->statistics();
+
+// Settings, observability, helpers
+$ch->setSettings(array $settings);     // client-wide; per-call overrides; chainable
+$ch->setSetting(string $key, mixed $value);  // single-key sugar, chainable
+$ch->setDatabase(string $database);    // USE on the server, updates default; chainable
+$ch->setProgressCallback(?callable $cb);
+$ch->setProfileCallback(?callable $cb);
+$ch->setVerbose(true);                 // JSON lifecycle lines on STDERR
+$ch->setVerbose(fn($e, $ctx) => ...);  // or custom sink: select_start /
+                                       // data_block / select_finish /
+                                       // execute_start / execute_finish /
+                                       // server_exception
+$ch->setVerbose(false);                // disable
+$stats = $ch->getStatistics();         // last query: rows, bytes, elapsed_ms, query_id
+
+$ch->resetConnection();
+$info = $ch->getServerInfo();          // name, version_*, revision, timezone, display_name
+$ep   = $ch->getCurrentEndpoint();     // {host, port} of the active endpoint, or null
+
+$ch->enableLogQueries(bool $enabled = true);
+$log = $ch->getLogQueries();           // returns and clears the buffer
+
+// DDL / introspection helpers
+$ch->isExists(string $database, string $table);
+$ch->showDatabases();
+$ch->showProcesslist();
+$ch->getServerVersion();
+$ch->databaseSize(?string $database = null);     // {bytes_on_disk, rows}
+$ch->tablesSize(?string $database = null);
+$ch->tableSize(string $table);                   // {rows, bytes_on_disk, partitions, modification_time}
+$ch->partitions(string $table);
+$ch->showTables(?string $database = null, ?string $like = null);
+$ch->showCreateTable(string $table);
+$ch->getServerUptime();                // seconds
+$ch->truncateTable(string $table);
+$ch->dropPartition(string $table, string $partition);
+```
+
+`fetch_mode` is a bitmask of `ClickHouse::FETCH_ONE`, `ClickHouse::FETCH_KEY_PAIR`, `ClickHouse::FETCH_COLUMN`, and `ClickHouse::DATE_AS_STRINGS`.
+
+### Placeholders
+
+Two placeholder syntaxes are supported in `select` / `execute`:
+
+- `{name}` is client-side identifier substitution. Two value shapes:
+  - **Scalar** (string / int / float): coerces to a single token, validated as either an identifier (`[A-Za-z_][A-Za-z0-9_]*`, optionally db-qualified by one dot like `db.tbl`) or a numeric literal. Whitespace, commas, quotes, semicolons, backslashes, and other SQL meta-characters are rejected.
+  - **Array**: each element is validated as a single scalar token, then joined with `", "` for the SQL replacement. Use this for legitimate column lists; an element with internal whitespace or commas is still rejected. A scalar value containing commas is rejected — that ambiguity (single identifier vs. list) is the point of the array-shape API.
+- `{name:Type}` is a server-side typed parameter. The SQL text is passed through unchanged; the value is bound via `Query::SetParam` and the server quotes and parses it according to `Type`. Pass PHP arrays for `Array(T)` types; `null` becomes a server `NULL`.
+
+```php
+// Single-identifier substitution.
+$ch->select("SELECT * FROM {tbl}", ["tbl" => "users"]);
+
+// Column-list substitution via array value.
+$ch->select("SELECT {cols} FROM users",
+            ["cols" => ["id", "name", "email"]]);
+
+// Server-side typed parameters, no client-side quoting needed.
+$ch->select("SELECT * FROM users WHERE id IN ({ids:Array(UInt32)})",
+            ["ids" => [1, 2, 3]]);
+```
+
+### Settings
+
+`setSettings()` applies client-wide. The 5th argument on `select` / `insert` / `execute` / `writeStart` overrides per call. Both accept plain `string => string` pairs; PHP scalars are stringified for you.
+
+```php
+$ch->setSettings(["max_execution_time" => "30"]);
+
+// Per-call override.
+$ch->select("SELECT * FROM big_table",
+            [], 0, "",
+            ["max_execution_time" => "5",
+             "max_memory_usage"   => "1000000000"]);
+```
+
+### Statistics and progress
+
+```php
+$ch->setProgressCallback(function (array $p) {
+    fprintf(STDERR, "rows=%d bytes=%d\n", $p["rows"], $p["bytes"]);
+});
+
+$ch->select("SELECT count() FROM big_table");
+
+$stats = $ch->getStatistics();
+// rows_read, bytes_read, total_rows, written_rows, written_bytes,
+// blocks, rows_before_limit, applied_limit, elapsed_ms
+```
+
+### Query log
+
+`enableLogQueries(true)` turns on a per-client buffer that records each completed `select` / `insert` / `execute` / streaming `writeEnd`. Each entry is `{sql, query_id, elapsed_ms, rows_read, bytes_read, error_code, error_message}`. `sql` is capped before retention. `error_code` is `0` on success, the server error code on a `ServerException`, or `-1` on client/network failure. `getLogQueries()` returns the buffer and clears it.
+
+```php
+$ch->enableLogQueries(true);
+$ch->select("SELECT count() FROM users");
+$ch->insert("logins", ["user_id", "ts"], $batch);
+
+foreach ($ch->getLogQueries() as $q) {
+    fprintf(STDERR, "[%.1fms] %s\n", $q["elapsed_ms"], $q["sql"]);
 }
 ```
-#### [More examples](https://github.com/SeasX/SeasClick/blob/master/tests/test.php)
 
-## Support
-SeasX Group
+### Structured exceptions
+
+`ClickHouseException` carries three extra public properties:
+
+- `server_code`: ClickHouse error code (e.g. 159 = `TIMEOUT_EXCEEDED`). `0` for client-side errors.
+- `server_name`: server-reported exception name (e.g. `DB::Exception`). `null` for client-side errors.
+- `query_id`: the query id associated with the failed call, when one was supplied. `null` otherwise.
+
+## 📊 Benchmarks
+
+PHP 8.4.22 / ClickHouse 26.3.9.8 / localhost loopback / `Memory` table (no disk).
+
+Compared against [smi2/phpClickHouse](https://github.com/smi2/phpClickHouse), the most popular pure-PHP HTTP client.
+
+Each cell is total wall-clock seconds for `selectCount` queries plus a single bulk insert of `dataCount` rows.
+
+| dataCount × selectCount × limit | phpClickHouse (HTTP) | php_clickhouse (uncompressed) | php_clickhouse (LZ4) | php_clickhouse (ZSTD) |
+|---|---:|---:|---:|---:|
+| 10000 × 1 × 5000   | 0.112 | 0.085 | 0.074 | 0.023 |
+| 10000 × 1 × 5000   | 0.104 | 0.030 | 0.024 | 0.081 |
+| 10000 × 100 × 5000  | 0.298 | 0.263 | 0.209 | 0.218 |
+| 10000 × 100 × 10000 | 0.303 | 0.210 | 0.265 | 0.215 |
+| 1000 × 200 × 500   | 0.558 | 0.416 | 0.415 | 0.413 |
+| 1000 × 200 × 1000  | 0.611 | 0.408 | 0.410 | 0.395 |
+| 1000 × 500 × 500   | 1.428 | 1.063 | 0.976 | 0.982 |
+| 1000 × 500 × 1000  | 1.383 | 0.959 | 1.025 | 1.030 |
+| 1000 × 800 × 500   | 2.477 | 1.533 | 1.569 | 1.543 |
+| 1000 × 800 × 1000  | 2.498 | 1.588 | 1.563 | 1.519 |
+
+At high select counts the native binary protocol runs 30-40% faster than the HTTP client. On small bursts (`dataCount=10000, selectCount=1`), php_clickhouse with ZSTD or LZ4 is fastest. To reproduce, see [`bench/`](bench/).
+
+## 🔗 PHP Performance Toolkit
+
+Companion native PHP extensions for high-throughput PHP workloads:
+
+- **[php_excel](https://github.com/iliaal/php_excel)**: native Excel I/O. 7-10× faster than PhpSpreadsheet, full XLS/XLSX with formulas, conditional formatting, and rich text. Powered by LibXL.
+- **[mdparser](https://github.com/iliaal/mdparser)**: native CommonMark + GFM markdown parser. 15-30× faster than pure-PHP libraries (Parsedown, cebe, michelf). Powered by cmark-gfm.
+- **[fastchart](https://github.com/iliaal/fastchart)**: native chart-rendering extension. 26 chart types behind one fluent OO API, SVG-canonical with PNG/JPG/WebP output (no libgd dependency).
+
+## 📚 Read more
+
+Full background, fork rationale, and benchmark methodology in the launch post: [php_clickhouse: A Native ClickHouse Client for PHP, Picking Up Where SeasClick Left Off](https://ilia.ws/blog/php-clickhouse-a-native-clickhouse-client-for-php-picking-up-where-seasclick-left-off).
+
+## License
+
+The PHP-side wrapper is licensed under [PHP-3.01](LICENSE).
+
+The vendored client library at `lib/clickhouse-cpp/` is [ClickHouse/clickhouse-cpp](https://github.com/ClickHouse/clickhouse-cpp), licensed under the [Apache License 2.0](lib/clickhouse-cpp/LICENSE).
+
+The vendored compression libraries (`lib/clickhouse-cpp/contrib/lz4/`, `contrib/zstd/`, `contrib/cityhash/`) carry BSD-style licenses; abseil int128 (`contrib/absl/`) is Apache 2.0. See each subdirectory for the exact text.
+
+## Credits
+
+`php_clickhouse` started as a fork of [SeasX/SeasClick](https://github.com/SeasX/SeasClick) by SeasX Group (`ahhhh.wang@gmail.com`). The original PR-4 work to add fetch modes landed in 2019 and the upstream maintainer hasn't accepted external PRs since. Independent re-vendoring, port to clickhouse-cpp v2.6.1, new types, TLS, and packaging are by Ilia Alshanetsky <ilia@ilia.ws>.
+
+## Contributing
+
+See [CONTRIBUTING.md](CONTRIBUTING.md). Security issues: [SECURITY.md](SECURITY.md).
+
+---
+
+[Follow @iliaa on X](https://x.com/iliaa) • [Blog](https://ilia.ws) • If this sped up your stack, ⭐ star it!
diff --git a/SECURITY.md b/SECURITY.md
new file mode 100644
index 0000000..d05013e
--- /dev/null
+++ b/SECURITY.md
@@ -0,0 +1,45 @@
+# Security policy
+
+## Supported versions
+
+| Version | Supported          |
+|---------|--------------------|
+| 0.7.x   | :white_check_mark: |
+
+Once 1.0 ships, the two most recent minor versions will receive
+security fixes.
+
+## Reporting a vulnerability
+
+**Do not file a public GitHub issue for security vulnerabilities.**
+
+Use GitHub's private security advisory feature at
+<https://github.com/iliaal/php_clickhouse/security/advisories/new>
+or email Ilia Alshanetsky <ilia@ilia.ws> directly.
+
+Please include:
+
+- Affected php_clickhouse version (`php -r 'echo phpversion("clickhouse");'`)
+- Affected ClickHouse server version
+- A minimal reproducing case (PHP code + the SQL or input that triggers it)
+- Impact: crash / RCE / info disclosure / DoS / etc.
+- Whether you've coordinated disclosure with anyone else
+
+Acknowledgement within 7 days, fix or status update within 30. Once a
+fix is released the advisory becomes public.
+
+## Scope
+
+php_clickhouse is a thin binding to the official ClickHouse/clickhouse-cpp
+client library. Vulnerabilities in the vendored client (LZ4 / ZSTD /
+cityhash / absl int128 / clickhouse-cpp itself) are reported to their
+respective upstream projects; we'll re-vendor as soon as a fix is tagged.
+
+User input passes from PHP into the wire protocol via the typed column
+machinery in `typesToPhp.cpp`. Bugs in that layer (out-of-bounds reads,
+use-after-free, integer truncation that leads to incorrect SQL) are
+in scope.
+
+The TLS/SSL build (`--enable-clickhouse-openssl`) links against the
+system OpenSSL. Misuse of `ssl_skip_verify => true` is documented as a
+dev-only knob; it disables certificate validation entirely.
diff --git a/SeasClick.cpp b/SeasClick.cpp
deleted file mode 100644
index 93347fa..0000000
--- a/SeasClick.cpp
+++ /dev/null
@@ -1,803 +0,0 @@
-/*
-  +----------------------------------------------------------------------+
-  | SeasClick                                                            |
-  +----------------------------------------------------------------------+
-  | Copyright (c) 1997-2018 The PHP Group                                |
-  +----------------------------------------------------------------------+
-  | This source file is subject to version 3.01 of the PHP license,      |
-  | that is bundled with this package in the file LICENSE, and is        |
-  | available through the world-wide-web at the following url:           |
-  | http://www.php.net/license/3_01.txt                                  |
-  | If you did not receive a copy of the PHP license and are unable to   |
-  | obtain it through the world-wide-web, please send a note to          |
-  | license@php.net so we can mail you a copy immediately.               |
-  +----------------------------------------------------------------------+
-  | Author:  SeasX Group <ahhhh.wang@gmail.com>                          |
-  +----------------------------------------------------------------------+
-*/
-#ifdef HAVE_CONFIG_H
-#include "config.h"
-#endif
-
-extern "C" {
-#include "php.h"
-#include "php_ini.h"
-#include "ext/standard/info.h"
-#include "Zend/zend_exceptions.h"
-#include "php7_wrapper.h"
-}
-
-#include "php_SeasClick.h"
-
-#include "lib/clickhouse-cpp/clickhouse/client.h"
-#include "lib/clickhouse-cpp/clickhouse/error_codes.h"
-#include "lib/clickhouse-cpp/clickhouse/types/type_parser.h"
-#include "typesToPhp.hpp"
-#include <iostream>
-#include <map>
-#include <sstream>
-
-using namespace clickhouse;
-using namespace std;
-
-zend_class_entry *SeasClick_ce;
-map<int, Client*> clientMap;
-map<int, Block> clientInsertBlack;
-
-#ifdef COMPILE_DL_SEASCLICK
-extern "C" {
-    ZEND_GET_MODULE(SeasClick)
-}
-#endif
-
-// PHP_FUNCTION(SeasClick_version)
-// {
-//     SC_RETURN_STRINGL(PHP_SEASCLICK_VERSION, strlen(PHP_SEASCLICK_VERSION));
-// }
-
-static PHP_METHOD(SEASCLICK_RES_NAME, __construct);
-static PHP_METHOD(SEASCLICK_RES_NAME, __destruct);
-static PHP_METHOD(SEASCLICK_RES_NAME, select);
-static PHP_METHOD(SEASCLICK_RES_NAME, insert);
-static PHP_METHOD(SEASCLICK_RES_NAME, writeStart);
-static PHP_METHOD(SEASCLICK_RES_NAME, write);
-static PHP_METHOD(SEASCLICK_RES_NAME, writeEnd);
-static PHP_METHOD(SEASCLICK_RES_NAME, execute);
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_construct, 0, 0, 1)
-ZEND_ARG_INFO(0, connectParames)
-ZEND_END_ARG_INFO()
-
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_destruct, 0, 0, 0)
-ZEND_END_ARG_INFO()
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_select, 0, 0, 2)
-ZEND_ARG_INFO(0, sql)
-ZEND_ARG_INFO(0, params)
-ZEND_END_ARG_INFO()
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_insert, 0, 0, 3)
-ZEND_ARG_INFO(0, table)
-ZEND_ARG_INFO(0, columns)
-ZEND_ARG_INFO(0, values)
-ZEND_END_ARG_INFO()
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_writeStart, 0, 0, 2)
-ZEND_ARG_INFO(0, table)
-ZEND_ARG_INFO(0, columns)
-ZEND_END_ARG_INFO()
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_write, 0, 0, 1)
-ZEND_ARG_INFO(0, values)
-ZEND_END_ARG_INFO()
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_writeEnd, 0, 0, 0)
-ZEND_END_ARG_INFO()
-
-ZEND_BEGIN_ARG_INFO_EX(SeasCilck_execute, 0, 0, 2)
-ZEND_ARG_INFO(0, sql)
-ZEND_ARG_INFO(0, params)
-ZEND_END_ARG_INFO()
-
-/* {{{ SeasClick_functions[] */
-const zend_function_entry SeasClick_functions[] =
-{
-    //PHP_FE(SeasClick_version,	NULL)
-    PHP_FE_END
-};
-/* }}} */
-
-const zend_function_entry SeasClick_methods[] =
-{
-    PHP_ME(SEASCLICK_RES_NAME, __construct,   SeasCilck_construct, ZEND_ACC_PUBLIC | ZEND_ACC_CTOR)
-    PHP_ME(SEASCLICK_RES_NAME, __destruct,    SeasCilck_destruct, ZEND_ACC_PUBLIC)
-    PHP_ME(SEASCLICK_RES_NAME, select,        SeasCilck_select, ZEND_ACC_PUBLIC)
-    PHP_ME(SEASCLICK_RES_NAME, insert,        SeasCilck_insert, ZEND_ACC_PUBLIC)
-    PHP_ME(SEASCLICK_RES_NAME, writeStart,    SeasCilck_writeStart, ZEND_ACC_PUBLIC)
-    PHP_ME(SEASCLICK_RES_NAME, write,         SeasCilck_write, ZEND_ACC_PUBLIC)
-    PHP_ME(SEASCLICK_RES_NAME, writeEnd,      SeasCilck_writeEnd, ZEND_ACC_PUBLIC)
-    PHP_ME(SEASCLICK_RES_NAME, execute,       SeasCilck_execute, ZEND_ACC_PUBLIC)
-    PHP_FE_END
-};
-
-/* {{{ PHP_MINIT_FUNCTION
- */
-PHP_MINIT_FUNCTION(SeasClick)
-{
-    zend_class_entry SeasClick;
-    INIT_CLASS_ENTRY(SeasClick, SEASCLICK_RES_NAME, SeasClick_methods);
-#if PHP_VERSION_ID >= 70000
-    SeasClick_ce = zend_register_internal_class_ex(&SeasClick, NULL);
-#else
-    SeasClick_ce = zend_register_internal_class_ex(&SeasClick, NULL, NULL TSRMLS_CC);
-#endif
-
-#if PHP_VERSION_ID <= 70000
-    zend_declare_property_stringl(SeasClick_ce, "host", strlen("host"), "127.0.0.1", sizeof("127.0.0.1") - 1, ZEND_ACC_PROTECTED TSRMLS_CC);
-    zend_declare_property_long(SeasClick_ce, "port", strlen("port"), 9000, ZEND_ACC_PROTECTED TSRMLS_CC);
-    zend_declare_property_stringl(SeasClick_ce, "database", strlen("database"), "default", sizeof("default") - 1, ZEND_ACC_PROTECTED TSRMLS_CC);
-    zend_declare_property_null(SeasClick_ce, "user", strlen("user"), ZEND_ACC_PROTECTED TSRMLS_CC);
-    zend_declare_property_null(SeasClick_ce, "passwd", strlen("passwd"), ZEND_ACC_PROTECTED TSRMLS_CC);
-    zend_declare_property_bool(SeasClick_ce, "compression", strlen("compression"), false, ZEND_ACC_PROTECTED TSRMLS_CC);
-#else
-     zend_declare_property_stringl(SeasClick_ce, "host", strlen("host"), "127.0.0.1", sizeof("127.0.0.1") - 1, ZEND_ACC_PROTECTED );
-    zend_declare_property_long(SeasClick_ce, "port", strlen("port"), 9000, ZEND_ACC_PROTECTED );
-    zend_declare_property_stringl(SeasClick_ce, "database", strlen("database"), "default", sizeof("default") - 1, ZEND_ACC_PROTECTED );
-    zend_declare_property_null(SeasClick_ce, "user", strlen("user"), ZEND_ACC_PROTECTED );
-    zend_declare_property_null(SeasClick_ce, "passwd", strlen("passwd"), ZEND_ACC_PROTECTED );
-    zend_declare_property_bool(SeasClick_ce, "compression", strlen("compression"), false, ZEND_ACC_PROTECTED );
-#endif
-
-   
-
-    SeasClick_ce->ce_flags |= ZEND_ACC_FINAL;
-    return SUCCESS;
-}
-/* }}} */
-
-/* {{{ PHP_MSHUTDOWN_FUNCTION
- */
-PHP_MSHUTDOWN_FUNCTION(SeasClick)
-{
-    return SUCCESS;
-}
-/* }}} */
-
-/* {{{ PHP_RINIT_FUNCTION
- */
-PHP_RINIT_FUNCTION(SeasClick)
-{
-    return SUCCESS;
-}
-/* }}} */
-
-/* {{{ PHP_RSHUTDOWN_FUNCTION
- */
-PHP_RSHUTDOWN_FUNCTION(SeasClick)
-{
-    return SUCCESS;
-}
-/* }}} */
-
-/* {{{ PHP_MINFO_FUNCTION
- */
-PHP_MINFO_FUNCTION(SeasClick)
-{
-    php_info_print_table_start();
-    php_info_print_table_header(2, "SeasClick support", "enabled");
-    php_info_print_table_row(2, "Version", PHP_SEASCLICK_VERSION);
-    php_info_print_table_row(2, "Author", "SeasX Group[email: ahhhh.wang@gmail.com]");
-    php_info_print_table_end();
-
-    DISPLAY_INI_ENTRIES();
-}
-/* }}} */
-
-/* {{{ SeasClick_module_entry
- */
-zend_module_entry SeasClick_module_entry =
-{
-    STANDARD_MODULE_HEADER,
-    SEASCLICK_RES_NAME,
-    SeasClick_functions,
-    PHP_MINIT(SeasClick),
-    PHP_MSHUTDOWN(SeasClick),
-    PHP_RINIT(SeasClick),
-    PHP_RSHUTDOWN(SeasClick),
-    PHP_MINFO(SeasClick),
-    PHP_SEASCLICK_VERSION,
-    STANDARD_MODULE_PROPERTIES
-};
-/* }}} */
-
-/* {{{ proto object __construct(array connectParames)
- */
-PHP_METHOD(SEASCLICK_RES_NAME, __construct)
-{
-    zval *connectParames;
-
-#ifndef FAST_ZPP
-    if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "z", &connectParames) == FAILURE)
-    {
-        return;
-    }
-#else
-#undef IS_UNDEF
-#define IS_UNDEF Z_EXPECTED_LONG
-    ZEND_PARSE_PARAMETERS_START(1, 1)
-    Z_PARAM_ARRAY(connectParames)
-    ZEND_PARSE_PARAMETERS_END();
-#undef IS_UNDEF
-#define IS_UNDEF 0
-#endif
-
-    HashTable *_ht = Z_ARRVAL_P(connectParames);
-    zval *value;
-
-    zval *this_obj;
-    this_obj = getThis();
-    if (php_array_get_value(_ht, "host", value))
-    {
-        convert_to_string(value);
-        sc_zend_update_property_string(SeasClick_ce, this_obj, "host", sizeof("host") - 1, Z_STRVAL_P(value));
-    }
-
-    if (php_array_get_value(_ht, "port", value))
-    {
-        convert_to_long(value);
-        sc_zend_update_property_long(SeasClick_ce, this_obj, "port", sizeof("port") - 1, Z_LVAL_P(value));
-    }
-
-    if (php_array_get_value(_ht, "compression", value))
-    {
-        convert_to_boolean(value);
-        sc_zend_update_property_long(SeasClick_ce, this_obj, "compression", sizeof("compression") - 1, Z_LVAL_P(value));
-    }
-
-    zval *host = sc_zend_read_property(SeasClick_ce, this_obj, "host", sizeof("host") - 1, 0);
-    zval *port = sc_zend_read_property(SeasClick_ce, this_obj, "port", sizeof("port") - 1, 0);
-    zval *compression = sc_zend_read_property(SeasClick_ce, this_obj, "compression", sizeof("compression") - 1, 0);
-
-    ClientOptions Options = ClientOptions()
-                            .SetHost(Z_STRVAL_P(host))
-                            .SetPort(Z_LVAL_P(port))
-                            .SetPingBeforeQuery(false);
-    if (Z_TYPE_P(compression) == IS_TRUE)
-    {
-        Options = Options.SetCompressionMethod(CompressionMethod::LZ4);
-    }
-
-    if (php_array_get_value(_ht, "database", value))
-    {
-        convert_to_string(value);
-        sc_zend_update_property_string(SeasClick_ce, this_obj, "database", sizeof("database") - 1, Z_STRVAL_P(value));
-        Options = Options.SetDefaultDatabase(Z_STRVAL_P(value));
-    }
-
-    if (php_array_get_value(_ht, "user", value))
-    {
-        convert_to_string(value);
-        sc_zend_update_property_string(SeasClick_ce, this_obj, "user", sizeof("user") - 1, Z_STRVAL_P(value));
-        Options = Options.SetUser(Z_STRVAL_P(value));
-    }
-
-    if (php_array_get_value(_ht, "passwd", value))
-    {
-        convert_to_string(value);
-        sc_zend_update_property_string(SeasClick_ce, this_obj, "passwd", sizeof("passwd") - 1, Z_STRVAL_P(value));
-        Options = Options.SetPassword(Z_STRVAL_P(value));
-    }
-
-    try
-    {
-        Client *client = new Client(Options);
-        int key = Z_OBJ_HANDLE(*this_obj);
-
-        clientMap.insert(std::pair<int, Client*>(key, client));
-
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-
-    RETURN_TRUE;
-}
-/* }}} */
-
-void getInsertSql(string *sql, char *table_name, zval *columns)
-{
-    zval *pzval;
-    char *key;
-    uint32_t keylen;
-    int keytype;
-
-    std::stringstream fields_section;
-
-    HashTable *columns_ht = Z_ARRVAL_P(columns);
-    size_t count = zend_hash_num_elements(columns_ht);
-    size_t index = 0;
-
-    SC_HASHTABLE_FOREACH_START2(columns_ht, key, keylen, keytype, pzval)
-    {
-        convert_to_string(pzval);
-        if (index >= (count - 1))
-        {
-            fields_section << (string)Z_STRVAL_P(pzval);
-        }
-        else
-        {
-            fields_section << (string)Z_STRVAL_P(pzval) << ",";
-        }
-        index++;
-    }
-    SC_HASHTABLE_FOREACH_END();
-    *sql = "INSERT INTO " + (string)table_name + " ( " + fields_section.str() + " ) VALUES";
-}
-
-/* {{{ proto array select(string sql, array params)
- */
-PHP_METHOD(SEASCLICK_RES_NAME, select)
-{
-    char *sql = NULL;
-    size_t l_sql = 0;
-    zval* params = NULL;
-
-#ifndef FAST_ZPP
-    if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "s|z", &sql, &l_sql, &params) == FAILURE)
-    {
-        return;
-    }
-#else
-#undef IS_UNDEF
-#define IS_UNDEF Z_EXPECTED_LONG
-    ZEND_PARSE_PARAMETERS_START(1, 2)
-    Z_PARAM_STRING(sql, l_sql)
-    Z_PARAM_OPTIONAL
-    Z_PARAM_ARRAY(params)
-    ZEND_PARSE_PARAMETERS_END();
-#undef IS_UNDEF
-#define IS_UNDEF 0
-#endif
-    try
-    {
-        int key = Z_OBJ_HANDLE(*getThis());
-        Client *client = clientMap.at(key);
-
-        if (clientInsertBlack.count(key))
-        {
-            throw std::runtime_error("The insert operation is now in progress");
-        }
-
-        string sql_s = (string)sql;
-        if (ZEND_NUM_ARGS() > 1 && params != NULL)
-        {
-            if (Z_TYPE_P(params) != IS_ARRAY)
-            {
-                throw std::runtime_error("The second argument to the select function must be an array");
-            }
-
-            HashTable *params_ht = Z_ARRVAL_P(params);
-            zval *pzval;
-            char *str_key;
-            uint32_t str_keylen;
-            int keytype;
-
-            SC_HASHTABLE_FOREACH_START2(params_ht, str_key, str_keylen, keytype, pzval)
-            {
-                convert_to_string(pzval);
-                sql_s.replace(sql_s.find("{" + (string)str_key + "}"), str_keylen + 2, (string)Z_STRVAL_P(pzval));
-            }
-            SC_HASHTABLE_FOREACH_END();
-        }
-
-        array_init(return_value);
-
-        client->Select(sql_s, [return_value](const Block& block)
-        {
-            zval *return_tmp;
-            for (size_t row = 0; row < block.GetRowCount(); ++row)
-            {
-                SC_MAKE_STD_ZVAL(return_tmp);
-                array_init(return_tmp);
-                for (size_t column = 0; column < block.GetColumnCount(); ++column)
-                {
-                    string column_name = block.GetColumnName(column);
-                    convertToZval(return_tmp, block[column], row, column_name, 0);
-                }
-                add_next_index_zval(return_value, return_tmp);
-            }
-        }
-                      );
-
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-}
-/* }}} */
-
-/* {{{ proto array insert(string table, array columns, array values)
- */
-PHP_METHOD(SEASCLICK_RES_NAME, insert)
-{
-    char *table = NULL;
-    size_t l_table = 0;
-    zval *columns;
-    zval *values;
-
-    string sql;
-
-#ifndef FAST_ZPP
-    if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "szz", &table, &l_table, &columns, &values) == FAILURE)
-    {
-        return;
-    }
-#else
-#undef IS_UNDEF
-#define IS_UNDEF Z_EXPECTED_LONG
-    ZEND_PARSE_PARAMETERS_START(3, 3)
-    Z_PARAM_STRING(table, l_table)
-    Z_PARAM_ARRAY(columns)
-    Z_PARAM_ARRAY(values)
-    ZEND_PARSE_PARAMETERS_END();
-#undef IS_UNDEF
-#define IS_UNDEF 0
-#endif
-
-    try
-    {
-        int key = Z_OBJ_HANDLE(*getThis());
-        Client *client = clientMap.at(key);
-
-        if (clientInsertBlack.count(key))
-        {
-            throw std::runtime_error("The insert operation is now in progress");
-        }
-
-        HashTable *columns_ht = Z_ARRVAL_P(columns);
-        HashTable *values_ht = Z_ARRVAL_P(values);
-        size_t columns_count = zend_hash_num_elements(columns_ht);
-
-        zval *return_should;
-        SC_MAKE_STD_ZVAL(return_should);
-        array_init(return_should);
-
-        zval *fzval;
-        zval *pzval;
-        char *str_key;
-        uint32_t str_keylen;
-        int keytype;
-
-        zval *return_tmp;
-        for(size_t i = 0; i < columns_count; i++)
-        {
-            SC_MAKE_STD_ZVAL(return_tmp);
-            array_init(return_tmp);
-
-            SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, pzval)
-            {
-                if (Z_TYPE_P(pzval) != IS_ARRAY)
-                {
-                    throw std::runtime_error("The insert function needs to pass in a two-dimensional array");
-                }
-                fzval = sc_zend_hash_index_find(Z_ARRVAL_P(pzval), i);
-                if (NULL == fzval)
-                {
-                    throw std::runtime_error("The number of parameters inserted per line is inconsistent");
-                }
-                sc_zval_add_ref(fzval);
-                add_next_index_zval(return_tmp, fzval);
-            }
-            SC_HASHTABLE_FOREACH_END();
-
-            add_next_index_zval(return_should, return_tmp);
-        }
-
-        getInsertSql(&sql, table, columns);
-        Block blockQuery;
-
-        client->InsertQuery(sql, [&blockQuery](const Block& block)
-        {
-            blockQuery = block;
-        }
-                           );
-
-        Block blockInsert;
-        size_t index = 0;
-
-        SC_HASHTABLE_FOREACH_START2(Z_ARRVAL_P(return_should), str_key, str_keylen, keytype, pzval)
-        {
-            zvalToBlock(blockInsert, blockQuery, index, pzval);
-            index++;
-        }
-        SC_HASHTABLE_FOREACH_END();
-
-        client->InsertData(blockInsert);
-        client->InsertDataEnd();
-        sc_zval_ptr_dtor(&return_should);
-
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-    RETURN_TRUE;
-}
-/* }}} */
-
-/* {{{ proto array insert(string table, array columns, array values)
- */
-PHP_METHOD(SEASCLICK_RES_NAME, writeStart)
-{
-    char *table = NULL;
-    size_t l_table = 0;
-    zval *columns;
-
-    string sql;
-
-#ifndef FAST_ZPP
-    if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "sz", &table, &l_table, &columns) == FAILURE)
-    {
-        return;
-    }
-#else
-#undef IS_UNDEF
-#define IS_UNDEF Z_EXPECTED_LONG
-    ZEND_PARSE_PARAMETERS_START(2, 2)
-    Z_PARAM_STRING(table, l_table)
-    Z_PARAM_ARRAY(columns)
-    ZEND_PARSE_PARAMETERS_END();
-#undef IS_UNDEF
-#define IS_UNDEF 0
-#endif
-
-    try
-    {
-        int key = Z_OBJ_HANDLE(*getThis());
-        Client *client = clientMap.at(key);
-
-        if (clientInsertBlack.count(key))
-        {
-            throw std::runtime_error("The insert operation is now in progress");
-        }
-
-        getInsertSql(&sql, table, columns);
-        Block blockQuery;
-
-        client->InsertQuery(sql, [&blockQuery](const Block& block)
-        {
-            blockQuery = block;
-        }
-                           );
-        
-        clientInsertBlack.insert(std::pair<int, Block>(key, blockQuery));
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-    RETURN_TRUE;
-}
-/* }}} */
-
-/* {{{ proto array insert(string table, array columns, array values)
- */
-PHP_METHOD(SEASCLICK_RES_NAME, write)
-{
-    zval *values;
-
-#ifndef FAST_ZPP
-    if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "z", &values) == FAILURE)
-    {
-        return;
-    }
-#else
-#undef IS_UNDEF
-#define IS_UNDEF Z_EXPECTED_LONG
-    ZEND_PARSE_PARAMETERS_START(1, 1)
-    Z_PARAM_ARRAY(values)
-    ZEND_PARSE_PARAMETERS_END();
-#undef IS_UNDEF
-#define IS_UNDEF 0
-#endif
-
-    try
-    {
-#if PHP_VERSION_ID < 70000
-        zval **first_data;
-#else
-        zval *first_data;
-#endif
-        HashTable *values_ht = Z_ARRVAL_P(values);
-        sc_zend_hash_get_current_data(values_ht, (void**) &first_data);
-        if (NULL == first_data)
-        {
-            throw std::runtime_error("The conut of data inserted is empty");
-        }
-#if PHP_VERSION_ID < 70000
-        size_t columns_count = zend_hash_num_elements(Z_ARRVAL_P(*first_data));
-#else
-        size_t columns_count = zend_hash_num_elements(Z_ARRVAL_P(first_data));
-#endif
-        zval *return_should;
-        SC_MAKE_STD_ZVAL(return_should);
-        array_init(return_should);
-
-        zval *fzval;
-        zval *pzval;
-        char *str_key;
-        uint32_t str_keylen;
-        int keytype;
-
-        zval *return_tmp;
-        for(size_t i = 0; i < columns_count; i++)
-        {
-            SC_MAKE_STD_ZVAL(return_tmp);
-            array_init(return_tmp);
-
-            SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, pzval)
-            {
-                if (Z_TYPE_P(pzval) != IS_ARRAY)
-                {
-                    throw std::runtime_error("The insert function needs to pass in a two-dimensional array");
-                }
-                fzval = sc_zend_hash_index_find(Z_ARRVAL_P(pzval), i);
-                if (NULL == fzval)
-                {
-                    throw std::runtime_error("The number of parameters inserted per line is inconsistent");
-                }
-                sc_zval_add_ref(fzval);
-                add_next_index_zval(return_tmp, fzval);
-            }
-            SC_HASHTABLE_FOREACH_END();
-
-            add_next_index_zval(return_should, return_tmp);
-        }
-
-
-        int key = Z_OBJ_HANDLE(*getThis());
-        Client *client = clientMap.at(key);
-
-        Block blockQuery = clientInsertBlack.at(key);
-
-        Block blockInsert;
-        size_t index = 0;
-
-        SC_HASHTABLE_FOREACH_START2(Z_ARRVAL_P(return_should), str_key, str_keylen, keytype, pzval)
-        {
-            zvalToBlock(blockInsert, blockQuery, index, pzval);
-            index++;
-        }
-        SC_HASHTABLE_FOREACH_END();
-
-        client->InsertData(blockInsert);
-        sc_zval_ptr_dtor(&return_should);
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-    RETURN_TRUE;
-}
-/* }}} */
-
-/* {{{ proto array insert(string table, array columns, array values)
- */
-PHP_METHOD(SEASCLICK_RES_NAME, writeEnd)
-{
-    try
-    {
-        int key = Z_OBJ_HANDLE(*getThis());
-        Client *client = clientMap.at(key);
-        clientInsertBlack.erase(key);
-
-        client->InsertDataEnd();
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-    RETURN_TRUE;
-}
-/* }}} */
-
-/* {{{ proto bool execute(string sql, array params)
- */
-PHP_METHOD(SEASCLICK_RES_NAME, execute)
-{
-    char *sql = NULL;
-    size_t l_sql = 0;
-    zval* params = NULL;
-
-#ifndef FAST_ZPP
-    if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "s|z", &sql, &l_sql, &params) == FAILURE)
-    {
-        return;
-    }
-#else
-#undef IS_UNDEF
-#define IS_UNDEF Z_EXPECTED_LONG
-    ZEND_PARSE_PARAMETERS_START(1, 2)
-    Z_PARAM_STRING(sql, l_sql)
-    Z_PARAM_OPTIONAL
-    Z_PARAM_ARRAY(params)
-    ZEND_PARSE_PARAMETERS_END();
-#undef IS_UNDEF
-#define IS_UNDEF 0
-#endif
-
-    try
-    {
-        int key = Z_OBJ_HANDLE(*getThis());
-        Client *client = clientMap.at(key);
-
-        if (clientInsertBlack.count(key))
-        {
-            throw std::runtime_error("The insert operation is now in progress");
-        }
-
-        string sql_s = (string)sql;
-        if (ZEND_NUM_ARGS() > 1 && params != NULL)
-        {
-            if (Z_TYPE_P(params) != IS_ARRAY)
-            {
-                throw std::runtime_error("The second argument to the select function must be an array");
-            }
-
-            HashTable *params_ht = Z_ARRVAL_P(params);
-            zval *pzval;
-            char *str_key;
-            uint32_t str_keylen;
-            int keytype;
-
-            SC_HASHTABLE_FOREACH_START2(params_ht, str_key, str_keylen, keytype, pzval)
-            {
-                convert_to_string(pzval);
-                sql_s.replace(sql_s.find("{" + (string)str_key + "}"), str_keylen + 2, (string)Z_STRVAL_P(pzval));
-            }
-            SC_HASHTABLE_FOREACH_END();
-        }
-
-        client->Execute(sql_s);
-
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-    RETURN_TRUE;
-}
-/* }}} */
-
-/* {{{ proto array __destruct()
- */
-PHP_METHOD(SEASCLICK_RES_NAME, __destruct)
-{
-    try
-    {
-        int key = Z_OBJ_HANDLE(*getThis());
-        Client *client = clientMap.at(key);
-        delete client;
-        clientMap.erase(key);
-        clientInsertBlack.erase(key);
-
-    }
-    catch (const std::exception& e)
-    {
-        sc_zend_throw_exception_tsrmls_cc(NULL, e.what(), 0);
-    }
-    RETURN_TRUE;
-}
-/* }}} */
-
-/*
- * Local variables:
- * tab-width: 4
- * c-basic-offset: 4
- * End:
- * vim600: noet sw=4 ts=4 fdm=marker
- * vim<600: noet sw=4 ts=4
- */
diff --git a/TODO.md b/TODO.md
new file mode 100644
index 0000000..b8d4a65
--- /dev/null
+++ b/TODO.md
@@ -0,0 +1,71 @@
+# TODO
+
+Post-0.8.0 work. Gap analysis against [smi2/phpClickHouse](https://github.com/smi2/phpClickHouse), the canonical pure-PHP HTTP client. Items here are functional gaps (real missing capability) or porting friction for users coming from smi2.
+
+HTTP-only features (sessions, curl options, file-based `WHERE IN`/write-to-file, HTTP auth methods, `X-ClickHouse-Summary` header) are excluded since they have no native-binary analogue.
+
+## Real functional gaps
+
+### Async query execution
+
+`selectAsync()` / `executeAsync()` style. Queue several queries, then block until all complete. clickhouse-cpp exposes the primitives (`Client::ExecuteAsync` and friends); we don't bind them.
+
+Scope: architectural. Needs a `ClickHouseFuture` (or similar) result handle, a way to drive completion (poll vs blocking wait-all), and per-future statistics capture. Most natural fit is a 0.9.0 release with its own design pass.
+
+### Cluster / replica awareness
+
+Multi-node cluster support: replica health checks, automatic node selection, table-on-replica discovery, shard/replica counting. smi2's `Cluster` class is the reference shape.
+
+Scope: large. Likely a separate `ClickHouseCluster` class on top of N `ClickHouse` instances, with health probes via the existing `ping()`. 0.9.0+ scope.
+
+### Totals / Extremes capture
+
+`SELECT ... WITH TOTALS` and `SETTINGS extremes=1` currently throw `unimplemented 7` / `unimplemented 8` from the vendored library. clickhouse-cpp v2.6.1 doesn't dispatch the Totals / Extremes packet types ([upstream issue #297](https://github.com/ClickHouse/clickhouse-cpp/issues/297)).
+
+Scope: vendor blocker. Two paths:
+
+1. Local patch: extend `BlockInfo` and add the two cases to the packet dispatch (~30-50 LOC). Ships in `lib/clickhouse-cpp/LOCAL_PATCHES.md`.
+2. Wait for upstream.
+
+Once decoded, surface via `getTotals()` / `getExtremes()` methods on `ClickHouse`.
+
+### Protocol-level verbose tracing (landed in 0.8.0)
+
+`setVerbose(bool|callable $sink): static`. `true` writes JSON lines to STDERR; `false` disables; a callable is invoked with `($event, $context)` per event. Events emitted at lifecycle points (`select_start`, `data_block`, `select_finish`, `execute_start`, `execute_finish`) plus `server_exception` from the protocol's exception packet path. Existing progress/profile callbacks are unaffected.
+
+True byte-level wire tracing (raw packet hex dumps) was not implemented. The event-level surface above is what the native protocol's hookable callbacks expose, and matches the operational debugging value of smi2's HTTP request/response logging without the dependency on the underlying transport.
+
+## Cosmetic / porting friction (smi2 -> php_clickhouse)
+
+Landed in 0.8.0:
+- `setSettings()` returns `$this` (was `bool`).
+- `setSetting(string $key, mixed $value): static` for chainable single-key writes.
+- `setDatabase(string $database): static` issues `USE` on the server.
+- `ClickHouseException::getServerCode()` / `getServerName()` / `getQueryId()` getter aliases.
+- `selectStatement(): ClickHouseStatement` opt-in result wrapper. New `ClickHouseStatement` class implements Iterator, Countable, ArrayAccess, JsonSerializable plus `fetchOne` / `fetchKeyPair` / `fetchColumn` / `toArray` / `statistics`. `select()` is unchanged.
+
+Remaining items are lower-leverage and stay deferred.
+
+### Full chainable settings builder (deferred)
+
+smi2: `$client->settings()->max_execution_time(30)->max_memory_usage(...)`. We have `setSetting($key, $value)` for chainable single-key writes and `setSettings($array)` for bulk replacement, both returning `$this`. The remaining gap is the magic-method-per-key API.
+
+A full `ClickHouseSettings` builder class with one method per setting key is mostly cosmetic; the existing API already chains. ClickHouse has 200+ settings, half unstable, so generating one method per key would couple us to vendor-side churn for marginal benefit. Defer unless users specifically ask for it.
+
+### Bindings / placeholder syntax compatibility (deferred)
+
+smi2: `:param` style with `bindParams([':a' => 1])`. php_clickhouse: `{name}` (client-side identifier substitution) and `{name:Type}` (server-side typed parameter).
+
+The native-protocol typed form is strictly better. Adding `:param` as a third alias would mean more documentation surface and more cases where a query author has to remember which form does what. We should be steering smi2 users toward `{name:Type}`, not making the unsafe-substitution path feel familiar. Skip on principle.
+
+## Differentiators to keep highlighting
+
+(Not gaps; reminders for README/marketing.)
+
+- Multi-endpoint failover at the native protocol level.
+- Native LZ4 / ZSTD (smi2 only has HTTP gzip).
+- Native TLS via `--enable-clickhouse-openssl`.
+- Sub-millisecond timeout precision.
+- TCP socket knobs (`tcp_nodelay`, full `tcp_keepalive_*`).
+- True native decoding of `Map`, `LowCardinality`, `Tuple` (HTTP returns strings or JSON).
+- Custom `zend_object` model: ZTS first-class, no module-global state, `free_obj` reaps on bailout.
diff --git a/bench/README.md b/bench/README.md
new file mode 100644
index 0000000..8c05dc8
--- /dev/null
+++ b/bench/README.md
@@ -0,0 +1,31 @@
+# php_clickhouse benchmarks
+
+Compares `php_clickhouse` (LZ4, ZSTD, uncompressed) against
+[smi2/phpClickHouse](https://github.com/smi2/phpClickHouse), a
+pure-PHP HTTP client.
+
+## Run
+
+```sh
+cd bench
+composer update --no-dev          # installs phpClickHouse
+CLICKHOUSE_HOST=127.0.0.1 \
+CLICKHOUSE_PORT=9000 \
+CLICKHOUSE_HTTP_PORT=8123 \
+CLICKHOUSE_USER=test \
+CLICKHOUSE_PASSWD=test \
+php -d extension=../modules/clickhouse.so bench_mark.php
+```
+
+The composer dependencies need `ext-curl`, `ext-mbstring`, `ext-phar`,
+`ext-tokenizer`. Use a stock distro PHP for the benchmark run if your
+dev PHP is built with `--disable-all`.
+
+`tests/` runs the functional suite; this directory is performance-only
+and deliberately separate.
+
+## Results
+
+Latest run lives in the top-level [README.md](../README.md) under
+"Benchmarks". To update it after a code change, re-run with the same
+`dataCount × selectCount × limit` matrix and replace the table.
diff --git a/bench/bench_mark.php b/bench/bench_mark.php
new file mode 100644
index 0000000..f8843e5
--- /dev/null
+++ b/bench/bench_mark.php
@@ -0,0 +1,194 @@
+<?php
+/**
+ * php_clickhouse benchmark.
+ *
+ * Compares this extension (with and without LZ4 compression) against
+ * smi2/phpClickHouse, a pure-PHP HTTP client
+ * (https://github.com/smi2/phpClickHouse).
+ *
+ * Originally written by ahhhh.wang@gmail.com for SeasClick. Updated for
+ * php_clickhouse (PHP 8.4 + clickhouse-cpp v2.6.1).
+ *
+ * Connection settings come from the same env vars the phpt suite uses:
+ *   CLICKHOUSE_HOST (default: clickhouse), CLICKHOUSE_PORT (default 9000),
+ *   CLICKHOUSE_HTTP_PORT (default 8123), CLICKHOUSE_USER, CLICKHOUSE_PASSWD.
+ */
+
+require_once __DIR__ . '/vendor/autoload.php';
+
+$HOST   = getenv('CLICKHOUSE_HOST') ?: 'clickhouse';
+$TCP    = (int)(getenv('CLICKHOUSE_PORT') ?: '9000');
+$HTTP   = (int)(getenv('CLICKHOUSE_HTTP_PORT') ?: '8123');
+$USER   = getenv('CLICKHOUSE_USER') ?: 'default';
+$PASSWD = getenv('CLICKHOUSE_PASSWD') ?: '';
+
+// $dataCount, $selectCount, $limit
+$testDataSet = [
+    [10000,   1,  5000],
+    [10000,   1,  5000],
+    [10000, 100,  5000],
+    [10000, 100, 10000],
+    [ 1000, 200,   500],
+    [ 1000, 200,  1000],
+    [ 1000, 500,   500],
+    [ 1000, 500,  1000],
+    [ 1000, 800,   500],
+    [ 1000, 800,  1000],
+];
+
+$total = 0;
+
+foreach ($testDataSet as $value) {
+    [$dataCount, $selectCount, $limit] = $value;
+    $insertData = initData($dataCount);
+
+    echo "\n##### dataCount: {$dataCount}, selectCount: {$selectCount}, limit: {$limit} #####\n";
+
+    $t0 = $t = start_test();
+    testPhpClickhouse($insertData, $selectCount, $limit, $HOST, $HTTP, $USER, $PASSWD);
+    $t = end_test($t, 'phpClickHouse (HTTP)');
+
+    testClickhouseNoCompression($insertData, $selectCount, $limit, $HOST, $TCP, $USER, $PASSWD);
+    $t = end_test($t, 'php_clickhouse (uncompressed)');
+
+    testClickhouseLz4($insertData, $selectCount, $limit, $HOST, $TCP, $USER, $PASSWD);
+    $t = end_test($t, 'php_clickhouse (LZ4)');
+
+    testClickhouseZstd($insertData, $selectCount, $limit, $HOST, $TCP, $USER, $PASSWD);
+    $t = end_test($t, 'php_clickhouse (ZSTD)');
+
+    total($t0);
+}
+
+function start_test() { return microtime(true); }
+
+function end_test($start, $name) {
+    global $total;
+    $end = microtime(true);
+    $total += $end - $start;
+    $num = number_format($end - $start, 3);
+    $pad = str_repeat(' ', max(1, 60 - strlen($name) - strlen($num)));
+    echo $name . $pad . $num . "\n";
+    return microtime(true);
+}
+
+function total() {
+    global $total;
+    echo str_repeat('-', 32) . "\n";
+    $num = number_format($total, 3);
+    $pad = str_repeat(' ', max(1, 32 - strlen('Total') - strlen($num)));
+    echo 'Total' . $pad . $num . "\n";
+    $total = 0;
+}
+
+function makeClickhouse($host, $port, $user, $pass, $compression) {
+    $cfg = [
+        'host'        => $host,
+        'port'        => $port,
+        'compression' => $compression,
+    ];
+    if ($user !== '')  $cfg['user']   = $user;
+    if ($pass !== '')  $cfg['passwd'] = $pass;
+    return new ClickHouse($cfg);
+}
+
+function setupTable($db) {
+    $db->execute('CREATE DATABASE IF NOT EXISTS test');
+    $db->execute('DROP TABLE IF EXISTS test.summing_url_views');
+    $db->execute('
+        CREATE TABLE test.summing_url_views (
+            event_date Date DEFAULT toDate(event_time),
+            event_time DateTime,
+            site_id    Int32,
+            site_key   String,
+            views      Int32,
+            v_00       Int32,
+            v_55       Int32
+        )
+        ENGINE = SummingMergeTree
+        ORDER BY (site_id, site_key, event_time, event_date)
+        PARTITION BY event_date
+        SETTINGS index_granularity = 8192
+    ');
+}
+
+function testClickhouseNoCompression($insertData, $num, $limit, $host, $port, $user, $pass) {
+    $db = makeClickhouse($host, $port, $user, $pass, false);
+    setupTable($db);
+    $db->insert('test.summing_url_views',
+        ['event_time', 'site_key', 'site_id', 'views', 'v_00', 'v_55'],
+        $insertData);
+    for ($a = 0; $a < $num; $a++) {
+        $db->select('SELECT * FROM test.summing_url_views LIMIT ' . $limit);
+    }
+    $db->execute('DROP TABLE test.summing_url_views');
+}
+
+function testClickhouseLz4($insertData, $num, $limit, $host, $port, $user, $pass) {
+    $db = makeClickhouse($host, $port, $user, $pass, 'lz4');
+    setupTable($db);
+    $db->insert('test.summing_url_views',
+        ['event_time', 'site_key', 'site_id', 'views', 'v_00', 'v_55'],
+        $insertData);
+    for ($a = 0; $a < $num; $a++) {
+        $db->select('SELECT * FROM test.summing_url_views LIMIT ' . $limit);
+    }
+    $db->execute('DROP TABLE test.summing_url_views');
+}
+
+function testClickhouseZstd($insertData, $num, $limit, $host, $port, $user, $pass) {
+    $db = makeClickhouse($host, $port, $user, $pass, 'zstd');
+    setupTable($db);
+    $db->insert('test.summing_url_views',
+        ['event_time', 'site_key', 'site_id', 'views', 'v_00', 'v_55'],
+        $insertData);
+    for ($a = 0; $a < $num; $a++) {
+        $db->select('SELECT * FROM test.summing_url_views LIMIT ' . $limit);
+    }
+    $db->execute('DROP TABLE test.summing_url_views');
+}
+
+function testPhpClickhouse($insertData, $num, $limit, $host, $port, $user, $pass) {
+    $db = new ClickHouseDB\Client([
+        'host'     => $host,
+        'port'     => (string)$port,
+        'username' => $user ?: 'default',
+        'password' => $pass,
+    ]);
+    $db->write('CREATE DATABASE IF NOT EXISTS test');
+    $db->database('test');
+    $db->setTimeout(10);
+    $db->setConnectTimeOut(5);
+
+    $db->write('DROP TABLE IF EXISTS summing_url_views');
+    $db->write('
+        CREATE TABLE summing_url_views (
+            event_date Date DEFAULT toDate(event_time),
+            event_time DateTime,
+            site_id    Int32,
+            site_key   String,
+            views      Int32,
+            v_00       Int32,
+            v_55       Int32
+        )
+        ENGINE = SummingMergeTree
+        ORDER BY (site_id, site_key, event_time, event_date)
+        PARTITION BY event_date
+        SETTINGS index_granularity = 8192
+    ');
+    $db->insert('summing_url_views', $insertData,
+        ['event_time', 'site_key', 'site_id', 'views', 'v_00', 'v_55']);
+
+    for ($a = 0; $a < $num; $a++) {
+        $db->select('SELECT * FROM summing_url_views LIMIT ' . $limit)->rows();
+    }
+    $db->write('DROP TABLE summing_url_views');
+}
+
+function initData($num) {
+    $rows = [];
+    while ($num--) {
+        $rows[] = [time(), 'HASH2', 2345, 12, 9, 3];
+    }
+    return $rows;
+}
diff --git a/tests/bench_mark/bench_mark.png b/bench/bench_mark.png
similarity index 100%
rename from tests/bench_mark/bench_mark.png
rename to bench/bench_mark.png
diff --git a/bench/composer.json b/bench/composer.json
new file mode 100644
index 0000000..e07f9a4
--- /dev/null
+++ b/bench/composer.json
@@ -0,0 +1,5 @@
+{
+    "require": {
+        "smi2/phpclickhouse": "^1.1"
+    }
+}
diff --git a/bench/composer.lock b/bench/composer.lock
new file mode 100644
index 0000000..092837d
--- /dev/null
+++ b/bench/composer.lock
@@ -0,0 +1,78 @@
+{
+    "_readme": [
+        "This file locks the dependencies of your project to a known state",
+        "Read more about it at https://getcomposer.org/doc/01-basic-usage.md#installing-dependencies",
+        "This file is @generated automatically"
+    ],
+    "content-hash": "d3221306b62716ca07412ebf79d997fd",
+    "packages": [
+        {
+            "name": "smi2/phpclickhouse",
+            "version": "1.26.423",
+            "source": {
+                "type": "git",
+                "url": "https://github.com/smi2/phpClickHouse.git",
+                "reference": "ab164a69d5755ae91c77fdd196c5de4ffb1893a2"
+            },
+            "dist": {
+                "type": "zip",
+                "url": "https://api.github.com/repos/smi2/phpClickHouse/zipball/ab164a69d5755ae91c77fdd196c5de4ffb1893a2",
+                "reference": "ab164a69d5755ae91c77fdd196c5de4ffb1893a2",
+                "shasum": ""
+            },
+            "require": {
+                "ext-curl": "*",
+                "ext-json": "*",
+                "php": "^8.0"
+            },
+            "require-dev": {
+                "doctrine/coding-standard": "^8.2",
+                "phpstan/phpstan": "^2.0",
+                "phpunit/phpunit": "^9.5",
+                "sebastian/comparator": "^4.0"
+            },
+            "type": "library",
+            "autoload": {
+                "psr-4": {
+                    "ClickHouseDB\\": "src/"
+                }
+            },
+            "notification-url": "https://packagist.org/downloads/",
+            "license": [
+                "MIT"
+            ],
+            "authors": [
+                {
+                    "name": "Igor Strykhar",
+                    "email": "isublimity@gmail.com",
+                    "homepage": "https://github.com/isublimity"
+                }
+            ],
+            "description": "PHP ClickHouse Client",
+            "homepage": "https://github.com/smi2/phpClickHouse",
+            "keywords": [
+                "clickhouse",
+                "client",
+                "curl",
+                "driver",
+                "http",
+                "http client",
+                "php"
+            ],
+            "support": {
+                "issues": "https://github.com/smi2/phpClickHouse/issues",
+                "source": "https://github.com/smi2/phpClickHouse/tree/1.26.423"
+            },
+            "time": "2026-04-23T18:02:52+00:00"
+        }
+    ],
+    "packages-dev": [],
+    "aliases": [],
+    "minimum-stability": "stable",
+    "stability-flags": {},
+    "prefer-stable": false,
+    "prefer-lowest": false,
+    "platform": {},
+    "platform-dev": {},
+    "plugin-api-version": "2.9.0"
+}
diff --git a/clickhouse.cpp b/clickhouse.cpp
new file mode 100644
index 0000000..b8c73ba
--- /dev/null
+++ b/clickhouse.cpp
@@ -0,0 +1,5357 @@
+/*
+  +----------------------------------------------------------------------+
+  | php_clickhouse                                                       |
+  +----------------------------------------------------------------------+
+  | Copyright (c) 1997-2026 The PHP Group                                |
+  +----------------------------------------------------------------------+
+  | This source file is subject to version 3.01 of the PHP license,      |
+  | that is bundled with this package in the file LICENSE, and is        |
+  | available through the world-wide-web at the following url:           |
+  | http://www.php.net/license/3_01.txt                                  |
+  | If you did not receive a copy of the PHP license and are unable to   |
+  | obtain it through the world-wide-web, please send a note to          |
+  | license@php.net so we can mail you a copy immediately.               |
+  +----------------------------------------------------------------------+
+  | Author: Ilia Alshanetsky <ilia@ilia.ws>                              |
+  | Original SeasClick author: SeasX Group <ahhhh.wang@gmail.com>        |
+  +----------------------------------------------------------------------+
+*/
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+extern "C" {
+#include "php.h"
+#include "php_ini.h"
+#include "ext/standard/info.h"
+#include "Zend/zend_exceptions.h"
+#include "Zend/zend_interfaces.h"
+#include "Zend/zend_smart_str.h"
+#include "ext/json/php_json.h"
+#include "main/snprintf.h"  // php_gcvt: locale-independent double formatter
+#include "php7_wrapper.h"
+}
+
+#include "php_clickhouse.h"
+
+#include "lib/clickhouse-cpp/clickhouse/client.h"
+#include "lib/clickhouse-cpp/clickhouse/error_codes.h"
+#include "lib/clickhouse-cpp/clickhouse/exceptions.h"
+#include "lib/clickhouse-cpp/clickhouse/types/type_parser.h"
+#include "lib/clickhouse-cpp/clickhouse/columns/factory.h"
+#include "typesToPhp.hpp"
+#include <chrono>
+#include <deque>
+#include <optional>
+#include <unordered_map>
+
+using namespace clickhouse;
+using namespace std;
+
+zend_class_entry *clickhouse_ce, *clickhouse_exception_ce, *clickhouse_iter_ce, *clickhouse_statement_ce;
+
+struct ClientStats {
+    uint64_t rows_read = 0;
+    uint64_t bytes_read = 0;
+    uint64_t total_rows = 0;
+    uint64_t written_rows = 0;
+    uint64_t written_bytes = 0;
+    uint64_t blocks = 0;
+    uint64_t rows_before_limit = 0;
+    bool applied_limit = false;
+    double elapsed_ms = 0.0;
+    std::string last_query_id;
+};
+struct QueryLog {
+    std::string sql;
+    std::string query_id;
+    double elapsed_ms = 0.0;
+    uint64_t rows_read = 0;
+    uint64_t bytes_read = 0;
+    int error_code = 0;            // 0 = success; ServerException code on server failure; -1 on client/network failure
+    std::string error_message;
+};
+
+/*
+ * Per-Client state lives on the zend_object itself. Replaces the old
+ * file-scope std::map<int, ...> bank keyed on Z_OBJ_HANDLE. Two
+ * concrete benefits over the old layout: free_obj fires even on
+ * bailout (so a fatal error mid-query no longer leaks the Client*),
+ * and ZTS works naturally because there is no global state to
+ * thread-isolate.
+ *
+ * The std member must be last; create_object placement-news the
+ * non-POD members and free_obj placement-destructs them.
+ */
+struct clickhouse_object {
+    Client *client;
+    Block insert_block;
+    bool has_insert_block;
+    /* True once write() has handed any block to the native client during
+     * the current writeStart()/writeEnd() session. The streaming-insert
+     * recovery path keys off this: pre-send conversion failure on a
+     * still-clean session can EndInsert() to close an empty insert,
+     * but once even one block has been sent the recovery must
+     * ResetConnection() to discard partial data. EndInsert() in that
+     * case would commit the already-sent blocks and turn a thrown
+     * write() into a silent partial-commit. */
+    bool insert_blocks_sent;
+    /* Guards same-client reentry. clickhouse-cpp's Client uses a single
+     * socket and a single per-call packet loop; a userland callback that
+     * fires another query / insert on the same client mid-stream sends
+     * packets into a wire still owned by the outer call and crashes the
+     * worker on the next ReceiveData. Set true at the start of each
+     * client-touching method, cleared on exit. Reentry throws cleanly. */
+    bool query_active;
+    ClientStats stats;
+    std::unordered_map<std::string, std::string> settings;
+    zval progress_callback;        // IS_UNDEF when unset
+    zval profile_callback;         // IS_UNDEF when unset
+    zval verbose_callback;         // IS_UNDEF when off or stderr-mode
+    bool verbose_to_stderr;        // true when setVerbose(true) was used
+    bool log_enabled;
+    std::string insert_sql;
+    std::string insert_query_id;
+    std::chrono::steady_clock::time_point insert_started_at;
+    /* deque, not vector, so the cap-overflow path's pop-front is O(1).
+     * vector::erase(begin()) on a 1024-entry log shifted ~16 KB on
+     * every query past the cap. */
+    std::deque<QueryLog> query_log;
+    zend_object std;
+};
+
+static inline clickhouse_object *clickhouse_from_obj(zend_object *obj)
+{
+    return (clickhouse_object *)((char *)obj - offsetof(clickhouse_object, std));
+}
+
+#define Z_CLICKHOUSE_P(zv) clickhouse_from_obj(Z_OBJ_P(zv))
+
+static zend_object_handlers clickhouse_object_handlers;
+
+static zend_object *clickhouse_create_object(zend_class_entry *ce)
+{
+    clickhouse_object *obj = (clickhouse_object *)zend_object_alloc(sizeof(clickhouse_object), ce);
+
+    obj->client = nullptr;
+    obj->has_insert_block = false;
+    obj->insert_blocks_sent = false;
+    obj->query_active = false;
+    obj->log_enabled = false;
+    obj->verbose_to_stderr = false;
+    new (&obj->insert_block) Block();
+    new (&obj->stats) ClientStats();
+    new (&obj->settings) std::unordered_map<std::string, std::string>();
+    new (&obj->insert_sql) std::string();
+    new (&obj->insert_query_id) std::string();
+    new (&obj->insert_started_at) std::chrono::steady_clock::time_point();
+    new (&obj->query_log) std::deque<QueryLog>();
+    ZVAL_UNDEF(&obj->progress_callback);
+    ZVAL_UNDEF(&obj->profile_callback);
+    ZVAL_UNDEF(&obj->verbose_callback);
+
+    zend_object_std_init(&obj->std, ce);
+    object_properties_init(&obj->std, ce);
+    obj->std.handlers = &clickhouse_object_handlers;
+    return &obj->std;
+}
+
+static void clickhouse_free_obj(zend_object *object)
+{
+    clickhouse_object *obj = clickhouse_from_obj(object);
+
+    /* Mirror write()'s catch-side recovery policy for an orphaned
+     * streaming insert. EndInsert() commits whatever blocks have
+     * already been sent, so on a session that received write() rows
+     * but never writeEnd() — script bailout, exception unwinding,
+     * unset() before completion — calling EndInsert() here turns an
+     * incomplete user intent into an implicit partial commit. Use
+     * ResetConnection() in that case so the server discards the
+     * in-flight insert. A clean session (BeginInsert ran, no blocks
+     * sent yet) is harmless to close with EndInsert. Swallow errors:
+     * free_obj must not throw. */
+    if (obj->client && obj->has_insert_block) {
+        if (obj->insert_blocks_sent) {
+            try { obj->client->ResetConnection(); } catch (...) {}
+        } else {
+            try { obj->client->EndInsert(); } catch (...) {}
+        }
+    }
+
+    if (obj->client) {
+        delete obj->client;
+        obj->client = nullptr;
+    }
+
+    if (Z_TYPE(obj->progress_callback) != IS_UNDEF) {
+        zval_ptr_dtor(&obj->progress_callback);
+        ZVAL_UNDEF(&obj->progress_callback);
+    }
+    if (Z_TYPE(obj->profile_callback) != IS_UNDEF) {
+        zval_ptr_dtor(&obj->profile_callback);
+        ZVAL_UNDEF(&obj->profile_callback);
+    }
+    if (Z_TYPE(obj->verbose_callback) != IS_UNDEF) {
+        zval_ptr_dtor(&obj->verbose_callback);
+        ZVAL_UNDEF(&obj->verbose_callback);
+    }
+
+    obj->insert_block.~Block();
+    obj->stats.~ClientStats();
+    obj->settings.~unordered_map<std::string, std::string>();
+    obj->insert_sql.~basic_string();
+    obj->insert_query_id.~basic_string();
+    obj->insert_started_at.~time_point();
+    obj->query_log.~deque<QueryLog>();
+
+    zend_object_std_dtor(&obj->std);
+}
+
+/*
+ * Streaming row iterator state. blocks accumulate via OnData during
+ * selectStream(); the foreach loop walks them lazily without
+ * materializing the full result set as a single PHP array.
+ */
+struct clickhouse_iter_object {
+    std::vector<Block> blocks;
+    /* Column names cached once on the first OnData callback. Result-set
+     * schemas are stable across all blocks in a single query, so caching
+     * once in the iterator avoids ~10M std::string heap allocs from
+     * GetColumnName(col) on every current() call for million-row scans. */
+    std::vector<std::string> column_names;
+    size_t block_idx;
+    size_t row_idx;
+    uint64_t cumulative_row_idx;
+    uint64_t total_rows;
+    int fetch_mode;
+    zend_object std;
+};
+
+static inline clickhouse_iter_object *clickhouse_iter_from_obj(zend_object *obj)
+{
+    return (clickhouse_iter_object *)((char *)obj - offsetof(clickhouse_iter_object, std));
+}
+
+#define Z_CLICKHOUSE_ITER_P(zv) clickhouse_iter_from_obj(Z_OBJ_P(zv))
+
+static zend_object_handlers clickhouse_iter_object_handlers;
+
+static zend_object *clickhouse_iter_create_object(zend_class_entry *ce)
+{
+    clickhouse_iter_object *iter = (clickhouse_iter_object *)zend_object_alloc(sizeof(clickhouse_iter_object), ce);
+
+    iter->block_idx = 0;
+    iter->row_idx = 0;
+    iter->cumulative_row_idx = 0;
+    iter->total_rows = 0;
+    iter->fetch_mode = 0;
+    new (&iter->blocks) std::vector<Block>();
+    new (&iter->column_names) std::vector<std::string>();
+
+    zend_object_std_init(&iter->std, ce);
+    object_properties_init(&iter->std, ce);
+    iter->std.handlers = &clickhouse_iter_object_handlers;
+    return &iter->std;
+}
+
+static void clickhouse_iter_free_obj(zend_object *object)
+{
+    clickhouse_iter_object *iter = clickhouse_iter_from_obj(object);
+    iter->blocks.~vector<Block>();
+    iter->column_names.~vector<std::string>();
+    zend_object_std_dtor(&iter->std);
+}
+
+/*
+ * Materialized result wrapper, returned by selectStatement(). The rows
+ * zval is a PHP array built once at construction time; iteration uses
+ * the HashTable's internal pointer, so a single foreach is the supported
+ * mode (nested foreach on the same Statement would fight over one
+ * cursor). The statistics zval is a per-call snapshot of obj->stats at
+ * the moment selectStatement returned, so callers can stash a Statement
+ * across other queries without losing its stats.
+ */
+struct clickhouse_statement_object {
+    zval rows;
+    zval positional_rows;
+    zval statistics;
+    zend_object std;
+};
+
+static inline clickhouse_statement_object *clickhouse_statement_from_obj(zend_object *obj)
+{
+    return (clickhouse_statement_object *)((char *)obj - offsetof(clickhouse_statement_object, std));
+}
+
+#define Z_CLICKHOUSE_STATEMENT_P(zv) clickhouse_statement_from_obj(Z_OBJ_P(zv))
+
+static zend_object_handlers clickhouse_statement_object_handlers;
+
+static zend_object *clickhouse_statement_create_object(zend_class_entry *ce)
+{
+    clickhouse_statement_object *stmt = (clickhouse_statement_object *)zend_object_alloc(sizeof(clickhouse_statement_object), ce);
+    ZVAL_UNDEF(&stmt->rows);
+    ZVAL_UNDEF(&stmt->positional_rows);
+    ZVAL_UNDEF(&stmt->statistics);
+    zend_object_std_init(&stmt->std, ce);
+    object_properties_init(&stmt->std, ce);
+    stmt->std.handlers = &clickhouse_statement_object_handlers;
+    return &stmt->std;
+}
+
+static void clickhouse_statement_free_obj(zend_object *object)
+{
+    clickhouse_statement_object *stmt = clickhouse_statement_from_obj(object);
+    zval_ptr_dtor(&stmt->rows);
+    zval_ptr_dtor(&stmt->positional_rows);
+    zval_ptr_dtor(&stmt->statistics);
+    zend_object_std_dtor(&stmt->std);
+}
+
+static std::string sanitizeError(const char *what);
+static void throwClickHouseError(const std::exception &e, const std::string &query_id = std::string());
+static std::string currentDatabase(zval *this_obj);
+
+#ifdef COMPILE_DL_CLICKHOUSE
+extern "C" {
+#ifdef ZTS
+    ZEND_TSRMLS_CACHE_DEFINE()
+#endif
+    ZEND_GET_MODULE(clickhouse)
+}
+#endif
+
+static PHP_METHOD(ClickHouse, __construct);
+static PHP_METHOD(ClickHouse, __destruct);
+static PHP_METHOD(ClickHouse, select);
+static PHP_METHOD(ClickHouse, selectWithExternalData);
+static PHP_METHOD(ClickHouse, selectToStream);
+static PHP_METHOD(ClickHouse, insert);
+static PHP_METHOD(ClickHouse, insertAssoc);
+static PHP_METHOD(ClickHouse, insertFromStream);
+static PHP_METHOD(ClickHouse, writeStart);
+static PHP_METHOD(ClickHouse, write);
+static PHP_METHOD(ClickHouse, writeEnd);
+static PHP_METHOD(ClickHouse, execute);
+static PHP_METHOD(ClickHouse, ping);
+static PHP_METHOD(ClickHouse, setSettings);
+static PHP_METHOD(ClickHouse, setSetting);
+static PHP_METHOD(ClickHouse, setDatabase);
+static PHP_METHOD(ClickHouse, setProgressCallback);
+static PHP_METHOD(ClickHouse, setProfileCallback);
+static PHP_METHOD(ClickHouse, setVerbose);
+static PHP_METHOD(ClickHouse, resetConnection);
+static PHP_METHOD(ClickHouse, getServerInfo);
+static PHP_METHOD(ClickHouse, getCurrentEndpoint);
+static PHP_METHOD(ClickHouse, getStatistics);
+static PHP_METHOD(ClickHouse, databaseSize);
+static PHP_METHOD(ClickHouse, tablesSize);
+static PHP_METHOD(ClickHouse, partitions);
+static PHP_METHOD(ClickHouse, showTables);
+static PHP_METHOD(ClickHouse, showCreateTable);
+static PHP_METHOD(ClickHouse, getServerUptime);
+static PHP_METHOD(ClickHouse, enableLogQueries);
+static PHP_METHOD(ClickHouse, getLogQueries);
+static PHP_METHOD(ClickHouse, selectStream);
+static PHP_METHOD(ClickHouse, selectStatement);
+static PHP_METHOD(ClickHouse, selectStreamCallback);
+static PHP_METHOD(ClickHouse, isExists);
+static PHP_METHOD(ClickHouse, showDatabases);
+static PHP_METHOD(ClickHouse, showProcesslist);
+static PHP_METHOD(ClickHouse, getServerVersion);
+static PHP_METHOD(ClickHouse, tableSize);
+static PHP_METHOD(ClickHouse, truncateTable);
+static PHP_METHOD(ClickHouse, dropPartition);
+
+static PHP_METHOD(ClickHouseRowIterator, rewind);
+static PHP_METHOD(ClickHouseRowIterator, valid);
+static PHP_METHOD(ClickHouseRowIterator, current);
+static PHP_METHOD(ClickHouseRowIterator, key);
+static PHP_METHOD(ClickHouseRowIterator, next);
+static PHP_METHOD(ClickHouseRowIterator, count);
+
+static PHP_METHOD(ClickHouseException, getServerCode);
+static PHP_METHOD(ClickHouseException, getServerName);
+static PHP_METHOD(ClickHouseException, getQueryId);
+
+static PHP_METHOD(ClickHouseStatement, __construct);
+static PHP_METHOD(ClickHouseStatement, count);
+static PHP_METHOD(ClickHouseStatement, rewind);
+static PHP_METHOD(ClickHouseStatement, valid);
+static PHP_METHOD(ClickHouseStatement, current);
+static PHP_METHOD(ClickHouseStatement, key);
+static PHP_METHOD(ClickHouseStatement, next);
+static PHP_METHOD(ClickHouseStatement, offsetExists);
+static PHP_METHOD(ClickHouseStatement, offsetGet);
+static PHP_METHOD(ClickHouseStatement, offsetSet);
+static PHP_METHOD(ClickHouseStatement, offsetUnset);
+static PHP_METHOD(ClickHouseStatement, jsonSerialize);
+static PHP_METHOD(ClickHouseStatement, toArray);
+static PHP_METHOD(ClickHouseStatement, statistics);
+static PHP_METHOD(ClickHouseStatement, fetchOne);
+static PHP_METHOD(ClickHouseStatement, fetchKeyPair);
+static PHP_METHOD(ClickHouseStatement, fetchColumn);
+
+#include "clickhouse_arginfo.h"
+
+#if PHP_VERSION_ID < 80100
+static int clickhouse_serialize_deny(zval *object, unsigned char **buffer, size_t *buf_len, zend_serialize_data *data)
+{
+    (void)buffer;
+    (void)buf_len;
+    (void)data;
+    zend_throw_exception_ex(NULL, 0, "Serialization of '%s' is not allowed", ZSTR_VAL(Z_OBJCE_P(object)->name));
+    return FAILURE;
+}
+
+static int clickhouse_unserialize_deny(zval *object, zend_class_entry *ce, const unsigned char *buf, size_t buf_len, zend_unserialize_data *data)
+{
+    (void)object;
+    (void)buf;
+    (void)buf_len;
+    (void)data;
+    zend_throw_exception_ex(NULL, 0, "Unserialization of '%s' is not allowed", ZSTR_VAL(ce->name));
+    return FAILURE;
+}
+#endif
+
+static void clickhouse_mark_not_serializable(zend_class_entry *ce)
+{
+#if PHP_VERSION_ID >= 80100
+    ce->ce_flags |= ZEND_ACC_NOT_SERIALIZABLE;
+#else
+    ce->serialize = clickhouse_serialize_deny;
+    ce->unserialize = clickhouse_unserialize_deny;
+#endif
+}
+
+/* {{{ clickhouse_functions[] */
+const zend_function_entry clickhouse_functions[] =
+{
+    PHP_FE_END
+};
+/* }}} */
+
+/* {{{ PHP_MINIT_FUNCTION
+ */
+PHP_MINIT_FUNCTION(clickhouse)
+{
+#if defined(COMPILE_DL_CLICKHOUSE) && defined(ZTS)
+    ZEND_TSRMLS_CACHE_UPDATE();
+#endif
+
+    clickhouse_ce = register_class_ClickHouse();
+    clickhouse_mark_not_serializable(clickhouse_ce);
+    clickhouse_ce->create_object = clickhouse_create_object;
+#if PHP_VERSION_ID >= 80400
+    clickhouse_ce->default_object_handlers = &clickhouse_object_handlers;
+#endif
+
+    memcpy(&clickhouse_object_handlers, zend_get_std_object_handlers(), sizeof(zend_object_handlers));
+    clickhouse_object_handlers.offset = offsetof(clickhouse_object, std);
+    clickhouse_object_handlers.free_obj = clickhouse_free_obj;
+
+    clickhouse_exception_ce = register_class_ClickHouseException(zend_ce_exception);
+
+    clickhouse_iter_ce = register_class_ClickHouseRowIterator(zend_ce_iterator, zend_ce_countable);
+    clickhouse_mark_not_serializable(clickhouse_iter_ce);
+    clickhouse_iter_ce->create_object = clickhouse_iter_create_object;
+#if PHP_VERSION_ID >= 80400
+    clickhouse_iter_ce->default_object_handlers = &clickhouse_iter_object_handlers;
+#endif
+
+    memcpy(&clickhouse_iter_object_handlers, zend_get_std_object_handlers(), sizeof(zend_object_handlers));
+    clickhouse_iter_object_handlers.offset = offsetof(clickhouse_iter_object, std);
+    clickhouse_iter_object_handlers.free_obj = clickhouse_iter_free_obj;
+
+    clickhouse_statement_ce = register_class_ClickHouseStatement(zend_ce_iterator, zend_ce_countable, zend_ce_arrayaccess, php_json_serializable_ce);
+    clickhouse_mark_not_serializable(clickhouse_statement_ce);
+    clickhouse_statement_ce->create_object = clickhouse_statement_create_object;
+#if PHP_VERSION_ID >= 80400
+    clickhouse_statement_ce->default_object_handlers = &clickhouse_statement_object_handlers;
+#endif
+
+    memcpy(&clickhouse_statement_object_handlers, zend_get_std_object_handlers(), sizeof(zend_object_handlers));
+    clickhouse_statement_object_handlers.offset = offsetof(clickhouse_statement_object, std);
+    clickhouse_statement_object_handlers.free_obj = clickhouse_statement_free_obj;
+    clickhouse_statement_object_handlers.clone_obj = NULL;
+
+    /* Back-compat aliases for the original SeasClick name. Deprecated;
+     * removed in the next major release. */
+    zend_register_class_alias(CLICKHOUSE_RES_NAME_LEGACY, clickhouse_ce);
+    zend_register_class_alias(CLICKHOUSE_EXCEPTION_NAME_LEGACY, clickhouse_exception_ce);
+
+    return SUCCESS;
+}
+/* }}} */
+
+/* {{{ PHP_MINFO_FUNCTION
+ */
+PHP_MINFO_FUNCTION(clickhouse)
+{
+    php_info_print_table_start();
+    php_info_print_table_header(2, "ClickHouse support", "enabled");
+    php_info_print_table_row(2, "Version", PHP_CLICKHOUSE_VERSION);
+    php_info_print_table_row(2, "Author", "SeasX Group[email: ahhhh.wang@gmail.com], Ilia Alshanetsky");
+    php_info_print_table_end();
+
+    DISPLAY_INI_ENTRIES();
+}
+/* }}} */
+
+/* {{{ clickhouse_module_entry
+ */
+zend_module_entry clickhouse_module_entry =
+{
+    STANDARD_MODULE_HEADER,
+    CLICKHOUSE_RES_NAME,
+    clickhouse_functions,
+    PHP_MINIT(clickhouse),
+    NULL,
+    NULL,
+    NULL,
+    PHP_MINFO(clickhouse),
+    PHP_CLICKHOUSE_VERSION,
+    STANDARD_MODULE_PROPERTIES
+};
+/* }}} */
+
+/* {{{ proto object __construct(array connectParams)
+ */
+PHP_METHOD(ClickHouse, __construct)
+{
+    zval *connectParams;
+
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ARRAY(connectParams)
+    ZEND_PARSE_PARAMETERS_END();
+
+    HashTable *_ht = Z_ARRVAL_P(connectParams);
+    zval *value;
+
+    zval *this_obj;
+    this_obj = getThis();
+    bool host_configured = false;
+    bool port_configured = false;
+    if (php_array_get_value(_ht, "host", value))
+    {
+        host_configured = true;
+        ZStrGuard sg(value);
+        sc_zend_update_property_stringl(clickhouse_ce, this_obj, "host", sizeof("host") - 1,
+                                        sg.val(), sg.len());
+    }
+
+    if (php_array_get_value(_ht, "port", value))
+    {
+        port_configured = true;
+        zend_long _p = zval_get_long(value);
+        if (_p < 1 || _p > 65535) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "port out of 1..65535 range", 0);
+            return;
+        }
+        sc_zend_update_property_long(clickhouse_ce, this_obj, "port", sizeof("port") - 1, _p);
+    }
+
+    if (php_array_get_value(_ht, "compression", value))
+    {
+        long cv = 0;
+        if (Z_TYPE_P(value) == IS_STRING) {
+            const char *s = Z_STRVAL_P(value);
+            if (strcasecmp(s, "lz4") == 0)        cv = 1;
+            else if (strcasecmp(s, "zstd") == 0)  cv = 2;
+            else if (strcasecmp(s, "none") == 0)  cv = 0;
+            else {
+                sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                    "Unknown compression name; expected 'lz4', 'zstd', 'none', true, or false", 0);
+                return;
+            }
+        } else {
+            cv = zend_is_true(value) ? 1 : 0;
+        }
+        sc_zend_update_property_long(clickhouse_ce, this_obj, "compression", sizeof("compression") - 1, cv);
+    }
+
+    /* The seconds-based timeouts and retry counters all flow into
+     * std::chrono::seconds(...) or unsigned setters in clickhouse-cpp.
+     * A negative value silently wraps to a giant unsigned through
+     * SetSendRetries / SetTcpKeepAlive*; reject up front. */
+    /* php_array_get_value is a string-literal-only macro (it uses
+     * sizeof(str)-1 for the key length) so it can't be passed a const
+     * char* runtime key. The lambda goes through sc_zend_hash_find. */
+    auto load_nonneg_long = [&](const char *key, zend_long &out) -> bool {
+        zval *v = sc_zend_hash_find(_ht, (char*)key, strlen(key));
+        if (!v || ZVAL_IS_NULL(v)) return false;
+        zend_long n = zval_get_long(v);
+        if (n < 0) {
+            std::string msg = std::string(key) + " must be >= 0";
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce, msg.c_str(), 0);
+            return false; // caller checks EG(exception)
+        }
+        out = n;
+        return true;
+    };
+
+    {
+        zend_long n;
+        if (load_nonneg_long("retry_timeout", n)) {
+            sc_zend_update_property_long(clickhouse_ce, this_obj, "retry_timeout", sizeof("retry_timeout") - 1, n);
+        } else if (EG(exception)) { return; }
+    }
+    {
+        zend_long n;
+        if (load_nonneg_long("retry_count", n)) {
+            sc_zend_update_property_long(clickhouse_ce, this_obj, "retry_count", sizeof("retry_count") - 1, n);
+        } else if (EG(exception)) { return; }
+    }
+    {
+        zend_long n;
+        if (load_nonneg_long("connect_timeout", n)) {
+            sc_zend_update_property_long(clickhouse_ce, this_obj, "connect_timeout", sizeof("connect_timeout") - 1, n);
+        } else if (EG(exception)) { return; }
+    }
+    {
+        zend_long n;
+        if (load_nonneg_long("receive_timeout", n)) {
+            sc_zend_update_property_long(clickhouse_ce, this_obj, "receive_timeout", sizeof("receive_timeout") - 1, n);
+        } else if (EG(exception)) { return; }
+    }
+
+    zval *host = sc_zend_read_property(clickhouse_ce, this_obj, "host", sizeof("host") - 1, 0);
+    zval *port = sc_zend_read_property(clickhouse_ce, this_obj, "port", sizeof("port") - 1, 0);
+    zval *compression = sc_zend_read_property(clickhouse_ce, this_obj, "compression", sizeof("compression") - 1, 0);
+    zval *retry_timeout = sc_zend_read_property(clickhouse_ce, this_obj, "retry_timeout", sizeof("retry_timeout") - 1, 0);
+    zval *retry_count = sc_zend_read_property(clickhouse_ce, this_obj, "retry_count", sizeof("retry_count") - 1, 0);
+    zval *receive_timeout = sc_zend_read_property(clickhouse_ce, this_obj, "receive_timeout", sizeof("receive_timeout") - 1, 0);
+    zval *connect_timeout = sc_zend_read_property(clickhouse_ce, this_obj, "connect_timeout", sizeof("connect_timeout") - 1, 0);
+
+    ClientOptions Options = ClientOptions()
+                            .SetHost(std::string(Z_STRVAL_P(host), Z_STRLEN_P(host)))
+                            .SetPort((uint16_t)Z_LVAL_P(port))
+                            .SetSendRetries(Z_LVAL_P(retry_count))
+                            .SetRetryTimeout(std::chrono::seconds(Z_LVAL_P(retry_timeout)))
+                            .SetConnectionRecvTimeout(std::chrono::seconds(Z_LVAL_P(receive_timeout)))
+                            .SetConnectionConnectTimeout(std::chrono::seconds(Z_LVAL_P(connect_timeout)))
+                            .SetPingBeforeQuery(false);
+    long cv = Z_LVAL_P(compression);
+    if (cv == 1) Options = Options.SetCompressionMethod(CompressionMethod::LZ4);
+    else if (cv == 2) Options = Options.SetCompressionMethod(CompressionMethod::ZSTD);
+
+    if (php_array_get_value(_ht, "send_timeout", value)) {
+        zend_long n = zval_get_long(value);
+        if (n < 0) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "send_timeout must be >= 0", 0);
+            return;
+        }
+        Options = Options.SetConnectionSendTimeout(std::chrono::seconds(n));
+    }
+    /* Millisecond variants override the seconds-based keys. Useful when
+     * sub-second precision matters (CI test guards, low-latency hops). */
+    auto apply_timeout_ms = [&](const char *key,
+                                 ClientOptions& (ClientOptions::*setter)(const std::chrono::milliseconds&)) -> bool {
+        zval *v = sc_zend_hash_find(_ht, (char*)key, strlen(key));
+        if (!v || ZVAL_IS_NULL(v)) return true;
+        zend_long n = zval_get_long(v);
+        if (n < 0) {
+            std::string msg = std::string(key) + " must be >= 0";
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce, msg.c_str(), 0);
+            return false;
+        }
+        Options = (Options.*setter)(std::chrono::milliseconds(n));
+        return true;
+    };
+    if (!apply_timeout_ms("connect_timeout_ms", &ClientOptions::SetConnectionConnectTimeout)) return;
+    if (!apply_timeout_ms("receive_timeout_ms", &ClientOptions::SetConnectionRecvTimeout))    return;
+    if (!apply_timeout_ms("send_timeout_ms",    &ClientOptions::SetConnectionSendTimeout))   return;
+    if (php_array_get_value(_ht, "tcp_nodelay", value)) {
+        Options = Options.TcpNoDelay(zend_is_true(value));
+    }
+    if (php_array_get_value(_ht, "tcp_keepalive", value)) {
+        Options = Options.TcpKeepAlive(zend_is_true(value));
+    }
+    if (php_array_get_value(_ht, "ping_before_query", value)) {
+        Options = Options.SetPingBeforeQuery(zend_is_true(value));
+    }
+    if (php_array_get_value(_ht, "tcp_keepalive_idle", value)) {
+        zend_long n = zval_get_long(value);
+        if (n < 0) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "tcp_keepalive_idle must be >= 0", 0);
+            return;
+        }
+        Options = Options.SetTcpKeepAliveIdle(std::chrono::seconds(n));
+    }
+    if (php_array_get_value(_ht, "tcp_keepalive_intvl", value)) {
+        zend_long n = zval_get_long(value);
+        if (n < 0) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "tcp_keepalive_intvl must be >= 0", 0);
+            return;
+        }
+        Options = Options.SetTcpKeepAliveInterval(std::chrono::seconds(n));
+    }
+    if (php_array_get_value(_ht, "tcp_keepalive_cnt", value)) {
+        zend_long n = zval_get_long(value);
+        if (n < 0 || n > UINT_MAX) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "tcp_keepalive_cnt out of range", 0);
+            return;
+        }
+        Options = Options.SetTcpKeepAliveCount((unsigned int)n);
+    }
+    if (php_array_get_value(_ht, "max_compression_chunk_size", value)) {
+        zend_long n = zval_get_long(value);
+        if (n < 0 || n > UINT_MAX) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "max_compression_chunk_size out of range", 0);
+            return;
+        }
+        Options = Options.SetMaxCompressionChunkSize((unsigned int)n);
+    }
+#ifdef WITH_OPENSSL
+    bool want_ssl = false;
+    if (php_array_get_value(_ht, "ssl", value)) {
+        want_ssl = zend_is_true(value);
+    }
+    if (want_ssl) {
+        ClientOptions::SSLOptions ssl_opts;
+        // Default to TLS 1.2 minimum so a server speaking only 1.0 / 1.1
+        // is rejected without the caller having to remember to set this.
+        // Caller can override via ssl_min_protocol_version.
+        ssl_opts.SetMinProtocolVersion(0x0303);
+        if (php_array_get_value(_ht, "ssl_min_protocol_version", value)) {
+            static const struct { const char *name; int version; } tls_versions[] = {
+                {"tls1.0", 0x0301}, {"tls1.1", 0x0302},
+                {"tls1.2", 0x0303}, {"tls1.3", 0x0304},
+            };
+            int ver = 0;
+            {
+                ZStrGuard sg(value);
+                for (const auto &tv : tls_versions) {
+                    if (strcasecmp(sg.val(), tv.name) == 0) {
+                        ver = tv.version;
+                        break;
+                    }
+                }
+            }
+            if (ver == 0) {
+                sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                    "ssl_min_protocol_version must be one of tls1.0, tls1.1, tls1.2, tls1.3", 0);
+                return;
+            }
+            ssl_opts.SetMinProtocolVersion(ver);
+        }
+        if (php_array_get_value(_ht, "ssl_skip_verify", value)) {
+            ssl_opts.SetSkipVerification(zend_is_true(value));
+        }
+        if (php_array_get_value(_ht, "ssl_use_default_ca", value)) {
+            ssl_opts.SetUseDefaultCALocations(zend_is_true(value));
+        }
+        if (php_array_get_value(_ht, "ssl_ca_directory", value)) {
+            ZStrGuard sg(value);
+            ssl_opts.SetPathToCADirectory(std::string(sg.val(), sg.len()));
+        }
+        if (php_array_get_value(_ht, "ssl_ca_files", value)) {
+            std::vector<std::string> files;
+            if (Z_TYPE_P(value) == IS_STRING) {
+                files.emplace_back(Z_STRVAL_P(value), Z_STRLEN_P(value));
+            } else if (Z_TYPE_P(value) == IS_ARRAY) {
+                HashTable *fh = Z_ARRVAL_P(value);
+                zval *fv;
+                ZEND_HASH_FOREACH_VAL(fh, fv) {
+                    ZStrGuard sg(fv);
+                    files.emplace_back(sg.val(), sg.len());
+                } ZEND_HASH_FOREACH_END();
+            }
+            ssl_opts.SetPathToCAFiles(files);
+        }
+        Options = Options.SetSSLOptions(ssl_opts);
+    }
+#else
+    if (php_array_get_value(_ht, "ssl", value)) {
+        if (zend_is_true(value)) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "php_clickhouse was built without TLS support. Reconfigure with --enable-clickhouse-openssl",
+                0);
+            return;
+        }
+    }
+#endif
+
+    if (php_array_get_value(_ht, "endpoints", value) && Z_TYPE_P(value) == IS_ARRAY) {
+        std::vector<Endpoint> eps;
+        HashTable *eps_ht = Z_ARRVAL_P(value);
+        zval *ep_zv;
+        ZEND_HASH_FOREACH_VAL(eps_ht, ep_zv) {
+            if (Z_TYPE_P(ep_zv) != IS_ARRAY) continue;
+            HashTable *eh = Z_ARRVAL_P(ep_zv);
+            zval *hz = sc_zend_hash_find(eh, (char*)"host", 4);
+            zval *pz = sc_zend_hash_find(eh, (char*)"port", 4);
+            if (!hz) continue;
+            Endpoint e;
+            {
+                ZStrGuard host_sg(hz);
+                e.host = std::string(host_sg.val(), host_sg.len());
+            }
+            if (pz) {
+                zend_long p = zval_get_long(pz);
+                if (p < 1 || p > 65535) {
+                    sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                        "Endpoint port out of 1..65535 range", 0);
+                    return;
+                }
+                e.port = (uint16_t)p;
+            }
+            eps.push_back(std::move(e));
+        } ZEND_HASH_FOREACH_END();
+        if (!eps.empty()) {
+            if (!host_configured && !port_configured) {
+                Options = Options.SetHost(std::string());
+            }
+            Options = Options.SetEndpoints(eps);
+        }
+    }
+
+    if (php_array_get_value(_ht, "database", value))
+    {
+        ZStrGuard sg(value);
+        sc_zend_update_property_stringl(clickhouse_ce, this_obj, "database", sizeof("database") - 1,
+                                        sg.val(), sg.len());
+        Options = Options.SetDefaultDatabase(std::string(sg.val(), sg.len()));
+    }
+
+    if (php_array_get_value(_ht, "user", value))
+    {
+        ZStrGuard sg(value);
+        sc_zend_update_property_stringl(clickhouse_ce, this_obj, "user", sizeof("user") - 1,
+                                        sg.val(), sg.len());
+        Options = Options.SetUser(std::string(sg.val(), sg.len()));
+    }
+
+    if (php_array_get_value(_ht, "passwd", value))
+    {
+        ZStrGuard sg(value);
+        Options = Options.SetPassword(std::string(sg.val(), sg.len()));
+    }
+
+    try
+    {
+        clickhouse_object *obj = Z_CLICKHOUSE_P(this_obj);
+        if (obj->client) {
+            throw std::runtime_error("ClickHouse object is already constructed");
+        }
+        obj->client = new Client(Options);
+    }
+    catch (const std::exception& e)
+    {
+        throwClickHouseError(e, std::string());
+        return;
+    }
+
+    RETURN_TRUE;
+}
+/* }}} */
+
+/*
+ * Permit identifier chars only: ASCII letter/digit/underscore, plus a
+ * single dot for the optional database prefix on a table name. Length
+ * must be > 0 and the first character of each segment must be a letter
+ * or underscore. Rejects anything else, including the empty string and
+ * any quoting characters that could break out of the INSERT statement.
+ */
+static void validateIdentifier(const char *s, size_t len, const char *what, bool allow_dot)
+{
+    if (len == 0) {
+        throw std::runtime_error(std::string(what) + " must not be empty");
+    }
+    bool seg_start = true;
+    bool dot_seen = false;
+    for (size_t i = 0; i < len; ++i) {
+        unsigned char c = (unsigned char)s[i];
+        if (seg_start) {
+            if (!((c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || c == '_')) {
+                throw std::runtime_error(
+                    std::string(what) + " must start with a letter or underscore");
+            }
+            seg_start = false;
+            continue;
+        }
+        if ((c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') ||
+            (c >= '0' && c <= '9') || c == '_') {
+            continue;
+        }
+        if (allow_dot && c == '.' && !dot_seen) {
+            dot_seen = true;
+            seg_start = true;
+            continue;
+        }
+        throw std::runtime_error(
+            std::string(what) + " contains an invalid character");
+    }
+    if (seg_start) {
+        throw std::runtime_error(std::string(what) + " has an empty segment");
+    }
+}
+
+// Convert an optional zend_string * (often the query_id PHP_METHOD
+// arg) into a std::string, treating NULL and empty as "no qid".
+static inline std::string makeQid(zend_string *s)
+{
+    return (s && ZSTR_LEN(s) > 0) ? std::string(ZSTR_VAL(s), ZSTR_LEN(s)) : std::string();
+}
+
+static std::string sqlStringLiteral(const std::string &s)
+{
+    std::string out;
+    out.reserve(s.size() + 2);
+    out.push_back('\'');
+    for (char c : s) {
+        switch (c) {
+            case '\'':
+            case '\\':
+                out.push_back('\\');
+                out.push_back(c);
+                break;
+            case '\0':
+                out.append("\\0", 2);
+                break;
+            default:
+                out.push_back(c);
+                break;
+        }
+    }
+    out.push_back('\'');
+    return out;
+}
+
+static std::string sqlQuotedIdentifier(const std::string &s)
+{
+    std::string out;
+    out.reserve(s.size() + 2);
+    out.push_back('`');
+    for (char c : s) {
+        if (c == '`' || c == '\\') {
+            out.push_back('\\');
+        }
+        out.push_back(c);
+    }
+    out.push_back('`');
+    return out;
+}
+
+// Max bytes of an exception message that crosses into userland. Bigger
+// than 1024 because real ClickHouse errors with stack hints can run
+// long, smaller than the few KB that would let a verbose dump leak.
+#define CLICKHOUSE_ERROR_MAX_LEN 4096
+
+/*
+ * Strip the embedded SQL fragment from a clickhouse-cpp error message
+ * before it crosses into userland. Upstream typically appends the full
+ * failing query after a "While executing" / "in query" prefix, which
+ * leaks any literal a caller placed in a placeholder (passwords with
+ * digits-only values still pass our placeholder validator). Cap length
+ * at CLICKHOUSE_ERROR_MAX_LEN as a final defense.
+ */
+static std::string sanitizeError(const char *what)
+{
+    std::string msg(what ? what : "");
+    static const char *sql_markers[] = {
+        "While executing",
+        "in query: ",
+        "While processing",
+    };
+    for (const char *marker : sql_markers) {
+        std::string::size_type pos = msg.find(marker);
+        if (pos != std::string::npos) {
+            msg.erase(pos);
+            // Drop trailing whitespace/punct left from the cut.
+            while (!msg.empty() && (msg.back() == ' ' || msg.back() == ',' ||
+                                     msg.back() == ':' || msg.back() == '.')) {
+                msg.pop_back();
+            }
+            break;
+        }
+    }
+    if (msg.size() > CLICKHOUSE_ERROR_MAX_LEN) {
+        /* Reserve space for the suffix so the final string respects
+         * the documented cap rather than exceeding it by 15 bytes. */
+        static const char suffix[] = "... (truncated)";
+        constexpr size_t suffix_len = sizeof(suffix) - 1;
+        msg.resize(CLICKHOUSE_ERROR_MAX_LEN - suffix_len);
+        msg.append(suffix, suffix_len);
+    }
+    return msg;
+}
+
+/*
+ * Resolve the Client* on the given object, or throw if __construct
+ * never finished installing one (failed connect, or a method called on
+ * a half-built object).
+ */
+static Client* getClient(clickhouse_object *obj)
+{
+    if (!obj->client) {
+        throw std::runtime_error("ClickHouse client is not initialized");
+    }
+    return obj->client;
+}
+
+static zend_string *zvalGetStringOrThrow(zval *v)
+{
+    zend_string *s = zval_get_string(v);
+    if (!s || EG(exception)) {
+        if (s) {
+            zend_string_release(s);
+        }
+        throw std::runtime_error("PHP string conversion failed");
+    }
+    return s;
+}
+
+static void setElapsedSince(clickhouse_object *obj,
+                            std::chrono::steady_clock::time_point started_at)
+{
+    auto now = std::chrono::steady_clock::now();
+    obj->stats.elapsed_ms =
+        std::chrono::duration<double, std::milli>(now - started_at).count();
+}
+
+static void clearStreamingInsertState(clickhouse_object *obj)
+{
+    obj->insert_block = Block();
+    obj->has_insert_block = false;
+    obj->insert_blocks_sent = false;
+    obj->insert_sql.clear();
+    obj->insert_query_id.clear();
+    obj->insert_started_at = std::chrono::steady_clock::time_point();
+}
+
+static void resetConnectionReapplyDatabase(zval *this_obj, clickhouse_object *obj,
+                                           bool clear_insert_state)
+{
+    Client *client = getClient(obj);
+    client->ResetConnection();
+    if (clear_insert_state && obj->has_insert_block) {
+        clearStreamingInsertState(obj);
+    }
+    std::string dbname = currentDatabase(this_obj);
+    if (!dbname.empty() && dbname != "default") {
+        client->Execute(Query("USE " + sqlQuotedIdentifier(dbname)));
+    }
+}
+
+static void tryResetConnectionReapplyDatabase(zval *this_obj, clickhouse_object *obj,
+                                              bool clear_insert_state)
+{
+    try {
+        resetConnectionReapplyDatabase(this_obj, obj, clear_insert_state);
+    } catch (...) {}
+}
+
+static void validateMetadataFilterName(const std::string &s, const char *what)
+{
+    if (s.empty()) {
+        throw std::runtime_error(std::string(what) + " must not be empty");
+    }
+}
+
+/*
+ * Central thrower. Replaces every catch-block sc_zend_throw call so the
+ * server fields land on the exception in one place. ServerException is
+ * the only branch that knows the server's error code and name; every
+ * other exception (network, validation, ours) leaves the structured
+ * fields at their MINIT defaults.
+ */
+static void throwClickHouseError(const std::exception &e, const std::string &query_id)
+{
+    /* Preserve a PHP exception that was already raised (e.g. from inside a
+     * user-supplied progress/profile/verbose callback that we re-raised as a
+     * sentinel C++ throw). Overwriting it would drop the user's stack and
+     * leave them with our generic "callback aborted" wrapper. */
+    if (EG(exception)) {
+        return;
+    }
+    std::string msg = sanitizeError(e.what());
+    zval ex;
+    object_init_ex(&ex, clickhouse_exception_ce);
+
+    if (auto se = dynamic_cast<const clickhouse::ServerException*>(&e)) {
+        const clickhouse::Exception &exc = se->GetException();
+        sc_zend_update_property_long(clickhouse_exception_ce, &ex, "server_code", sizeof("server_code") - 1, (zend_long)exc.code);
+        if (!exc.name.empty()) {
+            sc_zend_update_property_stringl(clickhouse_exception_ce, &ex, "server_name", sizeof("server_name") - 1, exc.name.c_str(), exc.name.size());
+        }
+    }
+    if (!query_id.empty()) {
+        sc_zend_update_property_stringl(clickhouse_exception_ce, &ex, "query_id", sizeof("query_id") - 1, query_id.c_str(), query_id.size());
+    }
+    sc_zend_update_property_stringl(clickhouse_exception_ce, &ex, "message", sizeof("message") - 1, msg.c_str(), msg.size());
+    zend_throw_exception_object(&ex);
+}
+
+/*
+ * Coerce a PHP zval into the string format ClickHouse expects for
+ * server-side parameter values. Matches the textual format the server
+ * parses for {name:Type} placeholders. Strings/dates/scalars pass
+ * through verbatim (the wire layer adds the surrounding quotes); arrays
+ * are formatted as ClickHouse array literals so Array(T) parses cleanly.
+ */
+static std::string formatParamValue(zval *v, const std::string &type, bool inside_array);
+
+static std::string formatScalarParam(zval *v)
+{
+    switch (Z_TYPE_P(v)) {
+        case IS_NULL:
+            return std::string();
+        case IS_TRUE:
+            return std::string("true");
+        case IS_FALSE:
+            return std::string("false");
+        case IS_LONG: {
+            char buf[32];
+            int n = snprintf(buf, sizeof(buf), ZEND_LONG_FMT, Z_LVAL_P(v));
+            return std::string(buf, (n > 0 && (size_t)n < sizeof(buf)) ? (size_t)n : 0);
+        }
+        case IS_DOUBLE: {
+            /* snprintf("%g") honors LC_NUMERIC, so a PHP user calling
+             * setlocale(LC_NUMERIC, 'de_DE') would emit "1,5" on the
+             * wire and the ClickHouse server would reject the typed
+             * parameter or setting value. php_gcvt takes the decimal
+             * separator and exponent char explicitly and is locale-
+             * independent, which is what we need at the SQL boundary.
+             * 17 significant digits is the IEEE 754 round-trip bound
+             * for double, matching the prior %.17g behavior. Buffer
+             * needs ~25 bytes worst case; 64 leaves comfortable margin. */
+            char buf[64];
+            php_gcvt(Z_DVAL_P(v), 17, '.', 'e', buf);
+            return std::string(buf);
+        }
+        default: {
+            zend_string *coerced = zvalGetStringOrThrow(v);
+            std::string out(ZSTR_VAL(coerced), ZSTR_LEN(coerced));
+            zend_string_release(coerced);
+            return out;
+        }
+    }
+}
+
+static bool wrappedTypeInner(const std::string &t, const char *prefix, size_t prefix_len, std::string &inner)
+{
+    if (t.size() > prefix_len + 1 &&
+        t.compare(0, prefix_len, prefix) == 0 &&
+        t.back() == ')') {
+        inner = t.substr(prefix_len, t.size() - prefix_len - 1);
+        return true;
+    }
+    return false;
+}
+
+static bool arrayInnerType(const std::string &type, std::string &inner)
+{
+    std::string t = type;
+    for (;;) {
+        if (wrappedTypeInner(t, "LowCardinality(", 15, inner) ||
+            wrappedTypeInner(t, "Nullable(", 9, inner)) {
+            t = inner;
+            continue;
+        }
+        return wrappedTypeInner(t, "Array(", 6, inner);
+    }
+}
+
+static bool typeNeedsQuoting(const std::string &t)
+{
+    if (t.compare(0, 9, "Nullable(") == 0 && t.back() == ')') {
+        return typeNeedsQuoting(t.substr(9, t.size() - 10));
+    }
+    if (t.compare(0, 15, "LowCardinality(") == 0 && t.back() == ')') {
+        return typeNeedsQuoting(t.substr(15, t.size() - 16));
+    }
+
+    /* Inner type for an Array(T) typed param. Numeric and bool parse
+     * raw; everything else needs single-quotes around each element.
+     * Lengths are baked in so we don't strlen each compile-time literal
+     * on every dispatch. */
+    struct BareType { const char *prefix; size_t len; };
+    static constexpr BareType bare[] = {
+        {"Int",     3},
+        {"UInt",    4},
+        {"Float",   5},
+        {"Decimal", 7},
+        {"Bool",    4},
+    };
+    for (const auto &b : bare) {
+        if (t.compare(0, b.len, b.prefix) == 0) return false;
+    }
+    return true;
+}
+
+static bool typeAllowsNull(const std::string &t)
+{
+    if (t.compare(0, 9, "Nullable(") == 0 && t.back() == ')') {
+        return true;
+    }
+    if (t.compare(0, 15, "LowCardinality(") == 0 && t.back() == ')') {
+        return typeAllowsNull(t.substr(15, t.size() - 16));
+    }
+    return false;
+}
+
+static std::string formatParamValue(zval *v, const std::string &type, bool inside_array)
+{
+    if (Z_TYPE_P(v) == IS_NULL) {
+        return std::string();
+    }
+
+    if (Z_TYPE_P(v) == IS_ARRAY) {
+        std::string inner;
+        arrayInnerType(type, inner);
+        bool quote = inner.empty() ? true : typeNeedsQuoting(inner);
+        bool allow_null = inner.empty() ? false : typeAllowsNull(inner);
+        std::string nested_inner;
+        bool nested_array = !inner.empty() && arrayInnerType(inner, nested_inner);
+        std::string out = "[";
+        bool first = true;
+        zval *iv;
+        ZEND_HASH_FOREACH_VAL(Z_ARRVAL_P(v), iv) {
+            if (!first) out += ",";
+            first = false;
+            if (Z_TYPE_P(iv) == IS_NULL) {
+                if (!allow_null) {
+                    throw std::runtime_error(
+                        "NULL element in non-Nullable Array typed parameter");
+                }
+                out += "NULL";
+                continue;
+            }
+            if (nested_array) {
+                if (Z_TYPE_P(iv) != IS_ARRAY) {
+                    throw std::runtime_error(
+                        "Array typed parameter element must be an array");
+                }
+                out += formatParamValue(iv, inner, true);
+                continue;
+            }
+            std::string sv = formatScalarParam(iv);
+            if (quote) {
+                std::string esc;
+                esc.reserve(sv.size() + 2);
+                esc += "'";
+                for (char c : sv) {
+                    if (c == '\'' || c == '\\') esc += '\\';
+                    esc += c;
+                }
+                esc += "'";
+                out += esc;
+            } else {
+                out += sv;
+            }
+        } ZEND_HASH_FOREACH_END();
+        out += "]";
+        return out;
+    }
+
+    /* Scalar inside an Array goes through the same path; the caller is
+     * responsible for the surrounding quotes. The wire layer quotes the
+     * outer value for non-array typed params. */
+    (void)inside_array;
+    return formatScalarParam(v);
+}
+
+/*
+ * Apply the global setSettings map merged with a per-call settings array
+ * onto a Query object via Query::SetSetting. Per-call settings override
+ * global. Empty per-call array means "use global only".
+ */
+static void applyMergedSettings(Query &q, clickhouse_object *obj, zval *per_call)
+{
+    bool have_per_call = (per_call != NULL && Z_TYPE_P(per_call) == IS_ARRAY
+                          && zend_hash_num_elements(Z_ARRVAL_P(per_call)) > 0);
+
+    /* Common case: no per-call overrides. Iterate the global map directly
+     * instead of paying for a full unordered_map copy on every query. */
+    if (!have_per_call) {
+        for (const auto &kv : obj->settings) {
+            QuerySettingsField f;
+            f.value = kv.second;
+            f.flags = 0;
+            q.SetSetting(kv.first, f);
+        }
+        return;
+    }
+
+    std::unordered_map<std::string, std::string> merged = obj->settings;
+    HashTable *ht = Z_ARRVAL_P(per_call);
+    zval *vz;
+    zend_string *zk;
+    zend_ulong nk;
+    ZEND_HASH_FOREACH_KEY_VAL(ht, nk, zk, vz) {
+        (void)nk;
+        if (!zk) continue;
+        std::string sval = formatScalarParam(vz);
+        merged[std::string(ZSTR_VAL(zk), ZSTR_LEN(zk))] = sval;
+    } ZEND_HASH_FOREACH_END();
+    for (const auto &kv : merged) {
+        QuerySettingsField f;
+        f.value = kv.second;
+        f.flags = 0;
+        q.SetSetting(kv.first, f);
+    }
+}
+
+/*
+ * Wire OnProgress and OnProfile to (a) populate the per-object stats
+ * struct and (b) forward to the user's PHP progress callback if one is
+ * registered. Stats reset happens at query start in the caller.
+ */
+static void attachProgressAndProfile(Query &q, clickhouse_object *obj)
+{
+    q.OnProgress([obj](const Progress &p) {
+        ClientStats &st = obj->stats;
+        st.rows_read += p.rows;
+        st.bytes_read += p.bytes;
+        if (p.total_rows > st.total_rows) st.total_rows = p.total_rows;
+        st.written_rows += p.written_rows;
+        st.written_bytes += p.written_bytes;
+
+        if (Z_TYPE(obj->progress_callback) != IS_UNDEF) {
+            zval args[1], retval;
+            ZVAL_NULL(&retval);
+            array_init(&args[0]);
+            add_assoc_long(&args[0], "rows", (zend_long)p.rows);
+            add_assoc_long(&args[0], "bytes", (zend_long)p.bytes);
+            add_assoc_long(&args[0], "total_rows", (zend_long)p.total_rows);
+            add_assoc_long(&args[0], "written_rows", (zend_long)p.written_rows);
+            add_assoc_long(&args[0], "written_bytes", (zend_long)p.written_bytes);
+            call_user_function(NULL, NULL, &obj->progress_callback, &retval, 1, args);
+            zval_ptr_dtor(&args[0]);
+            zval_ptr_dtor(&retval);
+            /* If the user callback raised, propagate to the packet loop
+             * so subsequent OnData/OnProgress callbacks don't run and the
+             * outer try-catch in do_select_into surfaces the user's
+             * exception (preserved by throwClickHouseError). */
+            if (EG(exception)) {
+                throw std::runtime_error("progress callback aborted query");
+            }
+        }
+    });
+    q.OnProfile([obj](const Profile &pr) {
+        ClientStats &st = obj->stats;
+        st.blocks = pr.blocks;
+        if (pr.calculated_rows_before_limit) {
+            st.rows_before_limit = pr.rows_before_limit;
+            st.applied_limit = pr.applied_limit;
+        }
+        /* Profile.bytes is bytes processed server-side; merge as a
+         * floor so we never report less than what Progress saw. */
+        if (pr.bytes > st.bytes_read) st.bytes_read = pr.bytes;
+        if (pr.rows > st.rows_read) st.rows_read = pr.rows;
+
+        if (Z_TYPE(obj->profile_callback) != IS_UNDEF) {
+            zval args[1], retval;
+            ZVAL_NULL(&retval);
+            array_init(&args[0]);
+            add_assoc_long(&args[0], "rows", (zend_long)pr.rows);
+            add_assoc_long(&args[0], "blocks", (zend_long)pr.blocks);
+            add_assoc_long(&args[0], "bytes", (zend_long)pr.bytes);
+            add_assoc_long(&args[0], "rows_before_limit", (zend_long)pr.rows_before_limit);
+            add_assoc_bool(&args[0], "applied_limit", pr.applied_limit ? 1 : 0);
+            add_assoc_bool(&args[0], "calculated_rows_before_limit", pr.calculated_rows_before_limit ? 1 : 0);
+            call_user_function(NULL, NULL, &obj->profile_callback, &retval, 1, args);
+            zval_ptr_dtor(&args[0]);
+            zval_ptr_dtor(&retval);
+            if (EG(exception)) {
+                throw std::runtime_error("profile callback aborted query");
+            }
+        }
+    });
+}
+
+/*
+ * Verbose tracing: when enabled (via setVerbose(true|callable)), emit
+ * lifecycle events as either JSON lines on stderr or as calls to a
+ * user sink. ctx is consumed (zval_ptr_dtor'd) by the helper, so call
+ * sites can build a fresh array per event without worrying about
+ * cleanup. ctx may be NULL for events with no payload.
+ */
+static inline bool verbose_active(const clickhouse_object *obj)
+{
+    return obj->verbose_to_stderr || Z_TYPE(obj->verbose_callback) != IS_UNDEF;
+}
+
+static void emitVerbose(clickhouse_object *obj, const char *event, zval *ctx)
+{
+    if (!verbose_active(obj)) {
+        if (ctx) zval_ptr_dtor(ctx);
+        return;
+    }
+    zval payload;
+    if (ctx) {
+        ZVAL_COPY(&payload, ctx);
+    } else {
+        array_init(&payload);
+    }
+    if (obj->verbose_to_stderr) {
+        smart_str buf = {0};
+        php_json_encode(&buf, &payload, 0);
+        smart_str_0(&buf);
+        /* php_json_encode can raise (non-UTF8 string in payload, etc.).
+         * Don't let that exception bleed onto the next user-facing call;
+         * trace output is best-effort. */
+        if (EG(exception)) {
+            zend_clear_exception();
+        }
+        const char *body = buf.s ? ZSTR_VAL(buf.s) : "{}";
+        size_t body_len = buf.s ? ZSTR_LEN(buf.s) : 2;
+        fprintf(stderr, "[clickhouse] %s %.*s\n", event, (int)body_len, body);
+        smart_str_free(&buf);
+    } else if (Z_TYPE(obj->verbose_callback) != IS_UNDEF) {
+        zval args[2], retval;
+        ZVAL_NULL(&retval);
+        ZVAL_STRING(&args[0], event);
+        ZVAL_COPY(&args[1], &payload);
+        call_user_function(NULL, NULL, &obj->verbose_callback, &retval, 2, args);
+        zval_ptr_dtor(&args[0]);
+        zval_ptr_dtor(&args[1]);
+        zval_ptr_dtor(&retval);
+        /* If the user callback raised, propagate to the packet loop so
+         * subsequent OnData / event emissions don't run, mirroring the
+         * progress / profile callback re-raise pattern. Without this,
+         * the user's PHP exception buffers in EG(exception) and bleeds
+         * onto the next unrelated query. Cleanup of payload / ctx
+         * happens after this block; do that first then throw. */
+        if (EG(exception)) {
+            zval_ptr_dtor(&payload);
+            if (ctx) zval_ptr_dtor(ctx);
+            throw std::runtime_error("verbose callback aborted query");
+        }
+    }
+    zval_ptr_dtor(&payload);
+    if (ctx) zval_ptr_dtor(ctx);
+}
+
+/*
+ * Attach the OnException hook so a server_exception verbose event fires
+ * with the server's code/name/message. The data_block events come from
+ * the OnData closures in do_select_into / selectStreamCallback so they
+ * have access to the per-call block_index counter. Idempotent if
+ * verbose is off so the hot path stays cheap.
+ */
+static void attachVerbose(Query &q, clickhouse_object *obj)
+{
+    if (!verbose_active(obj)) return;
+
+    q.OnException([obj](const Exception &e) {
+        zval ctx;
+        array_init(&ctx);
+        add_assoc_long(&ctx, "code", (zend_long)e.code);
+        add_assoc_string(&ctx, "name", e.name.c_str());
+        add_assoc_string(&ctx, "message", e.display_text.c_str());
+        emitVerbose(obj, "server_exception", &ctx);
+    });
+}
+
+static void resetStats(clickhouse_object *obj)
+{
+    obj->stats = ClientStats();
+}
+
+/*
+ * RAII guard that asserts no other operation is currently active on
+ * this client. clickhouse-cpp's Client owns a single TCP socket and a
+ * single per-call packet loop; a userland callback (row, progress,
+ * profile, verbose) that fires another query / insert / ping on the
+ * SAME ClickHouse instance pushes packets onto a wire still owned by
+ * the outer call. The next ReceiveData walks invalidated state and
+ * SEGVs. Throw cleanly instead. A separate ClickHouse instance is
+ * fine — the guard is per-object.
+ */
+struct QueryActiveGuard {
+    clickhouse_object *obj;
+    bool armed;
+    explicit QueryActiveGuard(clickhouse_object *o) : obj(o), armed(false) {
+        if (obj->query_active) {
+            throw std::runtime_error(
+                "Reentrant operation: another query is already in progress "
+                "on this ClickHouse instance. Use a separate ClickHouse "
+                "instance from inside row / progress / profile / verbose "
+                "callbacks.");
+        }
+        obj->query_active = true;
+        armed = true;
+    }
+    ~QueryActiveGuard() { if (armed) obj->query_active = false; }
+    QueryActiveGuard(const QueryActiveGuard&) = delete;
+    QueryActiveGuard& operator=(const QueryActiveGuard&) = delete;
+};
+
+/*
+ * Append a completed-query record to the per-client log if logging is
+ * enabled. Pulls elapsed_ms / rows_read / bytes_read from the just-
+ * populated stats. No-op when logging is off so the hot path stays
+ * cheap on production deployments.
+ */
+/* Cap on retained QueryLog entries before getLogQueries() is called to
+ * drain. A long-running PHP-FPM worker with logging on otherwise grows
+ * the vector unboundedly (each entry holds two arbitrary-length strings).
+ * When the cap is reached we drop the oldest in-place. */
+#define CLICKHOUSE_QUERY_LOG_MAX 1024
+#define CLICKHOUSE_QUERY_LOG_STRING_MAX_BYTES 8192
+
+static std::string queryLogString(const std::string &value)
+{
+    if (value.size() <= CLICKHOUSE_QUERY_LOG_STRING_MAX_BYTES) {
+        return value;
+    }
+    static const char suffix[] = "... (truncated)";
+    constexpr size_t suffix_len = sizeof(suffix) - 1;
+    constexpr size_t prefix_len = CLICKHOUSE_QUERY_LOG_STRING_MAX_BYTES - suffix_len;
+    std::string out(value.data(), prefix_len);
+    out.append(suffix, suffix_len);
+    return out;
+}
+
+static void appendQueryLogCapped(clickhouse_object *obj, QueryLog &&ql)
+{
+    if (obj->query_log.size() >= CLICKHOUSE_QUERY_LOG_MAX) {
+        obj->query_log.pop_front();
+    }
+    obj->query_log.push_back(std::move(ql));
+}
+
+/* Build the QueryLog row shared by success and error paths. Caller fills
+ * error_code / error_message for the error variant. */
+static QueryLog buildQueryLog(const clickhouse_object *obj,
+                              const std::string &sql, const std::string &qid)
+{
+    QueryLog ql;
+    ql.sql = queryLogString(sql);
+    ql.query_id = queryLogString(qid);
+    ql.elapsed_ms = obj->stats.elapsed_ms;
+    ql.rows_read = obj->stats.rows_read;
+    ql.bytes_read = obj->stats.bytes_read;
+    return ql;
+}
+
+static void recordQuerySuccess(clickhouse_object *obj, const std::string &sql, const std::string &qid)
+{
+    if (!obj->log_enabled) return;
+    /* Never let a nested allocation failure here escape the wrapper. The caller
+     * may already be inside a catch-block (recording the previous error); a
+     * second uncaught exception would call std::terminate. */
+    try {
+        appendQueryLogCapped(obj, buildQueryLog(obj, sql, qid));
+    } catch (...) { /* swallow; logging is best-effort */ }
+}
+
+static void recordQueryError(clickhouse_object *obj, const std::string &sql, const std::string &qid, const std::exception &e)
+{
+    if (!obj->log_enabled) return;
+    try {
+        QueryLog ql = buildQueryLog(obj, sql, qid);
+        if (auto se = dynamic_cast<const clickhouse::ServerException*>(&e)) {
+            ql.error_code = se->GetException().code;
+        } else {
+            ql.error_code = -1;
+        }
+        ql.error_message = sanitizeError(e.what());
+        appendQueryLogCapped(obj, std::move(ql));
+    } catch (...) { /* swallow; logging must not throw from inside a catch */ }
+}
+
+/*
+ * Build the static INSERT INTO ... ( cols ) VALUES prefix from a PHP
+ * column-name list. Returns by value (NRVO). Each column name is
+ * validated through validateIdentifier; the table name is validated
+ * with allow_dot=true so "db.tbl" works.
+ *
+ * The prior version used std::stringstream and an out-param + raw
+ * char* table_name which forced a strlen() and lost the length the
+ * caller already had.
+ */
+static std::string getInsertSql(std::string_view table_name, const zval *columns)
+{
+    validateIdentifier(table_name.data(), table_name.size(), "table name", true);
+
+    HashTable *columns_ht = Z_ARRVAL_P(const_cast<zval*>(columns));
+    size_t count = zend_hash_num_elements(columns_ht);
+    if (count == 0) {
+        throw std::runtime_error("Column list must not be empty");
+    }
+
+    /* Pre-size: "INSERT INTO " + table + " ( " + cols + " ) VALUES",
+     * estimating ~16 chars per column on average. */
+    std::string out;
+    out.reserve(table_name.size() + 16 * count + 32);
+    out.append("INSERT INTO ");
+    out.append(table_name.data(), table_name.size());
+    out.append(" ( ");
+
+    bool first = true;
+    zval *pzval;
+    ZEND_HASH_FOREACH_VAL(columns_ht, pzval)
+    {
+        ZStrGuard sg(pzval);
+        validateIdentifier(sg.val(), sg.len(), "column name", false);
+        if (!first) out.append(",");
+        out.append(sg.val(), sg.len());
+        first = false;
+    }
+    ZEND_HASH_FOREACH_END();
+
+    out.append(" ) VALUES");
+    return out;
+}
+
+/*
+ * Substitute placeholders in `sql` with values from `params_ht`.
+ *
+ * Two syntaxes are supported and routed differently:
+ *
+ *   {name}        client-side identifier substitution. Two value
+ *                 shapes:
+ *
+ *                   - Scalar (string / int / float / bool): the value
+ *                     coerces to one token, validated as either a
+ *                     single identifier (`[A-Za-z_][A-Za-z0-9_]*`,
+ *                     optionally db-qualified by exactly one dot) or
+ *                     a numeric literal (optional sign, digits,
+ *                     optional fractional part, optional exponent).
+ *                     Whitespace, commas, and any other punctuation
+ *                     are rejected — the prior whitelist allowed
+ *                     comma-lists, which let `{tbl}` with value
+ *                     "a, b" turn `FROM {tbl}` into a cross join.
+ *
+ *                   - Array: each element is validated as a single
+ *                     scalar token by the same rule, then the
+ *                     elements are joined with ", " for the SQL
+ *                     replacement. Use this for legitimate column
+ *                     lists; an element with internal whitespace or
+ *                     commas is still rejected.
+ *
+ *                 Used for table and column names. Callers that need
+ *                 expression fragments should pre-validate upstream.
+ *
+ *   {name:Type}   server-side parameter (ClickHouse native). The SQL
+ *                 text is left untouched (the server parses {name:Type}
+ *                 itself); the value is collected into `out_params` so
+ *                 the caller can pass it to Query::SetParam. The wire
+ *                 layer single-quotes and the server parses according
+ *                 to Type. PHP arrays format as ClickHouse array
+ *                 literals so Array(T) parses cleanly.
+ *
+ * If a parameter is provided that doesn't appear in the SQL (in either
+ * form), the call throws. Multiple occurrences of the same `{name}`
+ * placeholder are all replaced.
+ *
+ * `out_params` collects (name, optional<value>) pairs. nullopt is
+ * routed to Query::SetParam as the server-side NULL sentinel.
+ */
+struct TypedParam {
+    std::string name;
+    std::optional<std::string> value;  // nullopt → server NULL
+};
+
+/*
+ * Validate a single client-side `{name}` placeholder token. Accepts a
+ * numeric literal (optional sign, digits, optional fractional part,
+ * optional exponent) or an identifier (`[A-Za-z_][A-Za-z0-9_]*`,
+ * optionally db-qualified by exactly one dot). Whitespace, commas, and
+ * other punctuation are rejected — list semantics live in the
+ * array-valued placeholder branch below. Returns the validated string
+ * by value or throws std::runtime_error with a contextual message.
+ */
+static std::string validatePlaceholderToken(const char *val, size_t vlen,
+                                            const std::string &name)
+{
+    auto throwInvalid = [&](const char *why) {
+        throw std::runtime_error(
+            "Placeholder value for {" + name + "} is invalid: " + why);
+    };
+    if (vlen == 0) throwInvalid("empty");
+
+    auto isAlpha = [](unsigned char c) {
+        return (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || c == '_';
+    };
+    auto isDigit = [](unsigned char c) { return c >= '0' && c <= '9'; };
+
+    size_t i = 0;
+    unsigned char c0 = (unsigned char)val[0];
+    if (isDigit(c0) || c0 == '+' || c0 == '-') {
+        if (c0 == '+' || c0 == '-') {
+            i++;
+            if (i >= vlen || !isDigit((unsigned char)val[i])) {
+                throwInvalid("sign without digits");
+            }
+        }
+        bool seen_digit = false;
+        while (i < vlen && isDigit((unsigned char)val[i])) { i++; seen_digit = true; }
+        if (i < vlen && val[i] == '.') {
+            i++;
+            while (i < vlen && isDigit((unsigned char)val[i])) { i++; seen_digit = true; }
+        }
+        if (i < vlen && (val[i] == 'e' || val[i] == 'E')) {
+            i++;
+            if (i < vlen && (val[i] == '+' || val[i] == '-')) i++;
+            if (i >= vlen || !isDigit((unsigned char)val[i])) {
+                throwInvalid("malformed exponent");
+            }
+            while (i < vlen && isDigit((unsigned char)val[i])) i++;
+        }
+        if (!seen_digit) throwInvalid("numeric token without digits");
+    } else if (isAlpha(c0)) {
+        i++;
+        bool dot_seen = false;
+        while (i < vlen) {
+            unsigned char c = (unsigned char)val[i];
+            if (isAlpha(c) || isDigit(c)) { i++; continue; }
+            if (c == '.' && !dot_seen) {
+                dot_seen = true; i++;
+                if (i >= vlen) throwInvalid("trailing dot");
+                if (!isAlpha((unsigned char)val[i])) {
+                    throwInvalid("segment after dot must start with a letter or underscore");
+                }
+                i++;
+                continue;
+            }
+            break;
+        }
+    } else {
+        throwInvalid("must start with a digit, sign, letter, or underscore");
+    }
+    if (i != vlen) {
+        throwInvalid("only one identifier or numeric literal per token; "
+                     "use array-valued placeholders for column lists");
+    }
+    return std::string(val, vlen);
+}
+
+static void applyPlaceholders(string &sql, HashTable *params_ht, std::vector<TypedParam> &out_params)
+{
+    zval *pzval;
+    zend_string *zk;
+    zend_ulong nk;
+
+    ZEND_HASH_FOREACH_KEY_VAL(params_ht, nk, zk, pzval) {
+        (void)nk;
+        if (!zk) {
+            throw std::runtime_error("Placeholder array keys must be strings");
+        }
+        std::string name(ZSTR_VAL(zk), ZSTR_LEN(zk));
+
+        /* Detect the {name:Type} server-side form. We scan the SQL for
+         * the prefix `{name:` and capture the matching Type up to the
+         * closing `}`. If found, this parameter is server-side. */
+        std::string typed_prefix = "{" + name + ":";
+        size_t tpos = sql.find(typed_prefix);
+        if (tpos != std::string::npos) {
+            size_t close = sql.find('}', tpos + typed_prefix.size());
+            if (close == std::string::npos) {
+                throw std::runtime_error(
+                    "Unterminated typed placeholder for {" + name + "}");
+            }
+            std::string type = sql.substr(tpos + typed_prefix.size(),
+                                          close - (tpos + typed_prefix.size()));
+            TypedParam tp;
+            tp.name = name;
+            if (Z_TYPE_P(pzval) != IS_NULL) {
+                tp.value = formatParamValue(pzval, type, false);
+            }
+            out_params.push_back(std::move(tp));
+            continue;
+        }
+
+        /* Fall through: client-side {name} identifier substitution.
+         *
+         * Two value shapes are supported:
+         *   - String value: a single identifier (optionally db-qualified
+         *     by one dot) or a numeric literal. Whitespace, commas, and
+         *     punctuation are rejected. The single-token contract is
+         *     what the API documentation promises as a safe identifier
+         *     substitution.
+         *   - Array value: each element is validated as a single token
+         *     and the elements are joined with ", " for the SQL
+         *     replacement. Use this for legitimate column lists; a
+         *     scalar string with commas like "a, b" would otherwise
+         *     bypass the single-identifier guarantee (`FROM {tbl}` with
+         *     "a, b" turned into a cross join in scan.md's repro). */
+        std::string repl;
+        if (Z_TYPE_P(pzval) == IS_ARRAY) {
+            HashTable *aht = Z_ARRVAL_P(pzval);
+            if (zend_hash_num_elements(aht) == 0) {
+                throw std::runtime_error(
+                    "Placeholder value for {" + name + "} is invalid: empty array");
+            }
+            zval *iv;
+            bool first = true;
+            ZEND_HASH_FOREACH_VAL(aht, iv) {
+                zend_string *coerced = zvalGetStringOrThrow(iv);
+                std::string tok;
+                try {
+                    tok = validatePlaceholderToken(
+                        ZSTR_VAL(coerced), ZSTR_LEN(coerced), name);
+                } catch (...) {
+                    zend_string_release(coerced);
+                    throw;
+                }
+                zend_string_release(coerced);
+                if (!first) repl += ", ";
+                first = false;
+                repl += tok;
+            } ZEND_HASH_FOREACH_END();
+        } else {
+            zend_string *coerced = zvalGetStringOrThrow(pzval);
+            try {
+                repl = validatePlaceholderToken(
+                    ZSTR_VAL(coerced), ZSTR_LEN(coerced), name);
+            } catch (...) {
+                zend_string_release(coerced);
+                throw;
+            }
+            zend_string_release(coerced);
+        }
+        std::string needle = "{" + name + "}";
+        size_t pos = sql.find(needle);
+        if (pos == std::string::npos) {
+            throw std::runtime_error(
+                "Placeholder {" + name + "} does not appear in the SQL");
+        }
+        while (pos != std::string::npos) {
+            sql.replace(pos, needle.size(), repl);
+            pos = sql.find(needle, pos + repl.size());
+        }
+    } ZEND_HASH_FOREACH_END();
+}
+
+static void attachTypedParams(Query &q, const std::vector<TypedParam> &params)
+{
+    for (const auto &p : params) {
+        q.SetParam(p.name, p.value);
+    }
+}
+
+/* {{{ proto bool ping()
+ */
+PHP_METHOD(ClickHouse, ping)
+{
+    try {
+        clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+        client->Ping();
+    } catch (const std::exception& e) {
+        throwClickHouseError(e);
+        return;
+    }
+    RETURN_TRUE;
+}
+
+/*
+ * Internal: run a SELECT and write rows into `out`, which the caller
+ * must have zero-initialized (we either array_init it or, for
+ * SC_FETCH_ONE, write a scalar zval directly). On error, throws via
+ * throwClickHouseError() and leaves `out` undefined; callers should
+ * check EG(exception) on return.
+ */
+static void do_select_into(zval *out, zval *this_obj,
+                           const char *sql, size_t l_sql,
+                           zval *params, zend_long fetch_mode,
+                           const std::string &qid, zval *settings,
+                           const ExternalTables *external_tables,
+                           zval *positional_out = NULL)
+{
+    clickhouse_object *obj = Z_CLICKHOUSE_P(this_obj);
+    std::string log_sql(sql, l_sql);
+    try
+    {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block)
+        {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        std::string sql_s = log_sql;
+        std::vector<TypedParam> typed_params;
+
+        if (params != NULL && Z_TYPE_P(params) == IS_ARRAY)
+        {
+            applyPlaceholders(sql_s, Z_ARRVAL_P(params), typed_params);
+            log_sql = sql_s;
+        } else if (params != NULL && Z_TYPE_P(params) != IS_ARRAY) {
+            throw std::runtime_error("The second argument to the select function must be an array");
+        }
+
+        Query query = qid.empty() ? Query(sql_s) : Query(sql_s, qid);
+        attachTypedParams(query, typed_params);
+        applyMergedSettings(query, obj, settings);
+        attachProgressAndProfile(query, obj);
+        attachVerbose(query, obj);
+
+        if (verbose_active(obj)) {
+            zval ctx;
+            array_init(&ctx);
+            add_assoc_stringl(&ctx, "sql", (char*)sql_s.data(), sql_s.size());
+            add_assoc_stringl(&ctx, "query_id", (char*)qid.data(), qid.size());
+            add_assoc_long(&ctx, "settings_count", (zend_long)obj->settings.size());
+            add_assoc_long(&ctx, "fetch_mode", (zend_long)fetch_mode);
+            emitVerbose(obj, "select_start", &ctx);
+        }
+
+        if (!(fetch_mode & SC_FETCH_ONE)) {
+            array_init(out);
+            if (positional_out) {
+                array_init(positional_out);
+            }
+        }
+
+        size_t verbose_block_idx = 0;
+        bool fetched_one = false;
+        query.OnData([out, positional_out, fetch_mode, &fetched_one, obj, &verbose_block_idx](const Block &block) {
+            if (verbose_active(obj)) {
+                zval ctx;
+                array_init(&ctx);
+                add_assoc_long(&ctx, "rows", (zend_long)block.GetRowCount());
+                add_assoc_long(&ctx, "columns", (zend_long)block.GetColumnCount());
+                add_assoc_long(&ctx, "block_index", (zend_long)verbose_block_idx++);
+                emitVerbose(obj, "data_block", &ctx);
+            }
+            if (fetch_mode & SC_FETCH_ONE) {
+                if (!fetched_one && block.GetRowCount() > 0 && block.GetColumnCount() > 0) {
+                    convertToZval(out, block[0], 0, "", 0, fetch_mode);
+                    fetched_one = true;
+                }
+                return;
+            }
+
+            /* Hoist column names out of the row loop; clickhouse-cpp
+             * returns a fresh std::string per call, and the names are
+             * stable across all rows of a block. */
+            const size_t col_count = block.GetColumnCount();
+            std::vector<std::string> col_names;
+            col_names.reserve(col_count);
+            for (size_t c = 0; c < col_count; ++c) {
+                col_names.emplace_back(block.GetColumnName(c));
+            }
+
+            for (size_t row = 0; row < block.GetRowCount(); ++row)
+            {
+                if (fetch_mode & SC_FETCH_KEY_PAIR) {
+                    if (col_count < 2) {
+                        throw std::runtime_error("Key pair mode requires at least 2 columns to be present");
+                    }
+                    zval kp_col1, kp_col2;
+                    ZVAL_UNDEF(&kp_col1);
+                    ZVAL_UNDEF(&kp_col2);
+                    try {
+                        convertToZval(&kp_col1, block[0], row, "", 0, fetch_mode|SC_FETCH_ONE);
+                        convertToZval(&kp_col2, block[1], row, "", 0, fetch_mode|SC_FETCH_ONE);
+                    } catch (...) {
+                        if (Z_TYPE(kp_col1) != IS_UNDEF) zval_ptr_dtor(&kp_col1);
+                        if (Z_TYPE(kp_col2) != IS_UNDEF) zval_ptr_dtor(&kp_col2);
+                        throw;
+                    }
+
+                    if (Z_TYPE(kp_col1) == IS_LONG) {
+                         sc_zend_hash_index_update(Z_ARRVAL_P(out), Z_LVAL(kp_col1), &kp_col2);
+                    } else {
+                        /* zval_get_string is the modern non-mutating
+                         * coercion (matches the post-CR-312 fetchKeyPair
+                         * path). The prior convert_to_string mutated
+                         * kp_col1 in place; both worked but the new form
+                         * keeps the two key-pair paths in lockstep. */
+                        zend_string *coerced = NULL;
+                        try {
+                            coerced = zvalGetStringOrThrow(&kp_col1);
+                            zend_symtable_update(Z_ARRVAL_P(out), coerced, &kp_col2);
+                            zend_string_release(coerced);
+                        } catch (...) {
+                            if (coerced) zend_string_release(coerced);
+                            zval_ptr_dtor(&kp_col1);
+                            zval_ptr_dtor(&kp_col2);
+                            throw;
+                        }
+                    }
+                    zval_ptr_dtor(&kp_col1);
+                    continue;
+                }
+
+                zval row_tmp;
+                ZVAL_UNDEF(&row_tmp);
+                if (!(fetch_mode & SC_FETCH_COLUMN)) {
+                    array_init(&row_tmp);
+                }
+
+                try {
+                    if (positional_out && !(fetch_mode & SC_FETCH_COLUMN)) {
+                        zval pos_tmp;
+                        array_init(&pos_tmp);
+                        try {
+                            for (size_t column = 0; column < col_count; ++column)
+                            {
+                                zval cell;
+                                ZVAL_UNDEF(&cell);
+                                convertToZval(&cell, block[column], row, "", 0, fetch_mode|SC_FETCH_ONE);
+
+                                zval assoc_cell;
+                                ZVAL_COPY(&assoc_cell, &cell);
+                                sc_add_assoc_zval_ex(&row_tmp,
+                                    col_names[column].c_str(),
+                                    col_names[column].length(),
+                                    &assoc_cell);
+                                add_next_index_zval(&pos_tmp, &cell);
+                            }
+                            add_next_index_zval(positional_out, &pos_tmp);
+                        } catch (...) {
+                            zval_ptr_dtor(&pos_tmp);
+                            throw;
+                        }
+                    } else {
+                        for (size_t column = 0; column < col_count; ++column)
+                        {
+                            if (fetch_mode & SC_FETCH_COLUMN) {
+                                convertToZval(&row_tmp, block[0], row, "", 0, fetch_mode|SC_FETCH_ONE);
+                                break;
+                            } else {
+                                convertToZval(&row_tmp, block[column], row, col_names[column], 0, fetch_mode);
+                            }
+                        }
+                    }
+                } catch (...) {
+                    if (Z_TYPE(row_tmp) != IS_UNDEF) zval_ptr_dtor(&row_tmp);
+                    throw;
+                }
+                add_next_index_zval(out, &row_tmp);
+            }
+        });
+
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            if (external_tables && !external_tables->empty()) {
+                client->SelectWithExternalData(query, *external_tables);
+            } else {
+                client->Select(query);
+            }
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            /* OnData lambda throws (KEY_PAIR shape mismatch, etc.) bubble
+             * out mid-stream and leave the native client with unread
+             * blocks on the wire. Subsequent queries on the same handle
+             * would then receive the residual data and appear corrupted.
+             * Reset to recover; the original throw still propagates. */
+            tryResetConnectionReapplyDatabase(this_obj, obj, false);
+            throw;
+        }
+        auto t1 = std::chrono::steady_clock::now();
+        obj->stats.elapsed_ms =
+            std::chrono::duration<double, std::milli>(t1 - t0).count();
+        recordQuerySuccess(obj, sql_s, qid);
+
+        if (verbose_active(obj)) {
+            zval ctx;
+            array_init(&ctx);
+            add_assoc_double(&ctx, "elapsed_ms", obj->stats.elapsed_ms);
+            add_assoc_long(&ctx, "rows_read", (zend_long)obj->stats.rows_read);
+            add_assoc_long(&ctx, "bytes_read", (zend_long)obj->stats.bytes_read);
+            add_assoc_long(&ctx, "blocks", (zend_long)verbose_block_idx);
+            emitVerbose(obj, "select_finish", &ctx);
+        }
+    }
+    catch (const std::exception& e)
+    {
+        recordQueryError(obj, log_sql, qid, e);
+        throwClickHouseError(e, qid);
+    }
+}
+
+/*
+ * Internal: snapshot a ClientStats into a fresh PHP assoc array. Used
+ * by both ClickHouse::getStatistics() and ClickHouseStatement (where
+ * the snapshot lives on the Statement object so it survives the
+ * Client running other queries afterwards).
+ */
+static void buildStatsArray(zval *out, const ClientStats &st)
+{
+    array_init(out);
+    add_assoc_long(out, "rows_read", (zend_long)st.rows_read);
+    add_assoc_long(out, "bytes_read", (zend_long)st.bytes_read);
+    add_assoc_long(out, "total_rows", (zend_long)st.total_rows);
+    add_assoc_long(out, "written_rows", (zend_long)st.written_rows);
+    add_assoc_long(out, "written_bytes", (zend_long)st.written_bytes);
+    add_assoc_long(out, "blocks", (zend_long)st.blocks);
+    add_assoc_long(out, "rows_before_limit", (zend_long)st.rows_before_limit);
+    add_assoc_bool(out, "applied_limit", st.applied_limit ? 1 : 0);
+    add_assoc_double(out, "elapsed_ms", st.elapsed_ms);
+    add_assoc_stringl(out, "query_id", st.last_query_id.data(), st.last_query_id.size());
+}
+
+/* {{{ proto array select(string sql, array params, int mode, string query_id, array settings)
+ */
+PHP_METHOD(ClickHouse, select)
+{
+    zend_string *sql = NULL;
+    zval* params = NULL;
+    zend_long fetch_mode = 0;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(1, 5)
+        Z_PARAM_STR(sql)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_ARRAY(params)
+        Z_PARAM_LONG(fetch_mode)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string qid = makeQid(query_id);
+    do_select_into(return_value, getThis(), ZSTR_VAL(sql), ZSTR_LEN(sql), params, fetch_mode, qid, settings, NULL);
+}
+/* }}} */
+
+/* Forward decls: buildExternalTableBlock reuses the row-shape validator
+ * and per-column packed-array builder that insert() also uses, but they
+ * are defined further down in the file. */
+static void validateRowShapes(HashTable *values_ht, size_t columns_count);
+static void buildSingleColumnZval(HashTable *values_ht, size_t column_index,
+                                  const std::vector<zend_string*> *column_names,
+                                  zval *out);
+
+/*
+ * Build a clickhouse-cpp Block from a single external-table entry of the
+ * shape ['name' => ..., 'columns' => ['col' => 'Type', ...], 'rows' => [...]].
+ * Validates structure and types; throws std::runtime_error on any
+ * malformed shape so the caller's outer try/catch can route it to
+ * ClickHouseException. The block is appended-into via the same
+ * insertColumn() path that insert() uses, so type coverage is identical.
+ */
+static Block buildExternalTableBlock(zval *entry, std::string &name_out)
+{
+    if (Z_TYPE_P(entry) != IS_ARRAY) {
+        throw std::runtime_error("externals must be a list of arrays");
+    }
+    HashTable *ht = Z_ARRVAL_P(entry);
+
+    zval *name_zv    = sc_zend_hash_find(ht, "name", sizeof("name") - 1);
+    zval *columns_zv = sc_zend_hash_find(ht, "columns", sizeof("columns") - 1);
+    zval *rows_zv    = sc_zend_hash_find(ht, "rows", sizeof("rows") - 1);
+
+    if (!name_zv || Z_TYPE_P(name_zv) != IS_STRING) {
+        throw std::runtime_error("external table requires string 'name'");
+    }
+    if (!columns_zv || Z_TYPE_P(columns_zv) != IS_ARRAY) {
+        throw std::runtime_error("external table requires array 'columns'");
+    }
+    if (!rows_zv || Z_TYPE_P(rows_zv) != IS_ARRAY) {
+        throw std::runtime_error("external table requires array 'rows'");
+    }
+
+    validateIdentifier(Z_STRVAL_P(name_zv), Z_STRLEN_P(name_zv),
+                       "external table name", /*allow_dot=*/false);
+    name_out.assign(Z_STRVAL_P(name_zv), Z_STRLEN_P(name_zv));
+
+    HashTable *columns_ht = Z_ARRVAL_P(columns_zv);
+    HashTable *rows_ht    = Z_ARRVAL_P(rows_zv);
+    size_t columns_count  = zend_hash_num_elements(columns_ht);
+    if (columns_count == 0) {
+        throw std::runtime_error("external table '" + name_out + "' has no columns");
+    }
+
+    std::vector<std::string> col_names;
+    std::vector<TypeRef>     col_types;
+    std::vector<zend_string*> col_names_zs;
+    col_names.reserve(columns_count);
+    col_types.reserve(columns_count);
+    col_names_zs.reserve(columns_count);
+    {
+        zend_string *k;
+        zend_ulong   nk;
+        zval        *type_zv;
+        ZEND_HASH_FOREACH_KEY_VAL(columns_ht, nk, k, type_zv) {
+            (void)nk;
+            if (!k) {
+                throw std::runtime_error("external table '" + name_out +
+                    "' columns must be an associative array of name => type");
+            }
+            if (Z_TYPE_P(type_zv) != IS_STRING) {
+                throw std::runtime_error("external table '" + name_out +
+                    "' column '" + std::string(ZSTR_VAL(k), ZSTR_LEN(k)) +
+                    "' type must be a string");
+            }
+            validateIdentifier(ZSTR_VAL(k), ZSTR_LEN(k),
+                               "external column name", /*allow_dot=*/false);
+            std::string type_name(Z_STRVAL_P(type_zv), Z_STRLEN_P(type_zv));
+            ColumnRef templ = CreateColumnByType(type_name);
+            if (!templ) {
+                throw std::runtime_error("external table '" + name_out +
+                    "' column '" + std::string(ZSTR_VAL(k), ZSTR_LEN(k)) +
+                    "' has unsupported type '" + type_name + "'");
+            }
+            col_names.emplace_back(ZSTR_VAL(k), ZSTR_LEN(k));
+            col_types.push_back(templ->Type());
+            col_names_zs.push_back(k);
+        } ZEND_HASH_FOREACH_END();
+    }
+
+    validateRowShapes(rows_ht, columns_count);
+
+    /* The native protocol uses a zero-row block as the "end of stream"
+     * marker for the external-data section, so clickhouse-cpp skips
+     * empty named tables on the wire (lib/clickhouse-cpp/clickhouse/
+     * client.cpp:354). The server then never sees the table identifier
+     * and the query fails with "Unknown expression or table expression
+     * identifier ext_X". Reject upfront with a message that points to
+     * the userland workaround (skip the query when the filter set is
+     * empty). */
+    if (zend_hash_num_elements(rows_ht) == 0) {
+        throw std::runtime_error(
+            "external table '" + name_out + "' has no rows; the native protocol "
+            "cannot carry an empty named external table — guard the call in "
+            "userland and skip the query when the filter set is empty");
+    }
+
+    Block block;
+    for (size_t c = 0; c < columns_count; ++c) {
+        zval inner;
+        buildSingleColumnZval(rows_ht, c, &col_names_zs, &inner);
+        ColumnRef column;
+        try {
+            column = insertColumn(col_types[c], &inner);
+        } catch (...) {
+            zval_ptr_dtor(&inner);
+            throw;
+        }
+        zval_ptr_dtor(&inner);
+        block.AppendColumn(col_names[c], column);
+    }
+    return block;
+}
+
+/* {{{ proto mixed selectWithExternalData(string sql, array externals, array params, int mode, string query_id, array settings)
+ *
+ * SELECT with one or more named in-memory tables sent alongside the
+ * query (ClickHouse "external data" feature). Use this to keep the SQL
+ * body small when filtering by a big list — e.g.
+ * `SELECT id, name FROM users WHERE id IN ext_ids` with `ext_ids`
+ * supplied as 50k rows in memory.
+ *
+ * Each entry of `externals`:
+ *   ['name'    => 'ext_ids',
+ *    'columns' => ['id' => 'UInt64'],
+ *    'rows'    => [[1], [2], ...]]
+ *
+ * Multiple externals per call supported; names must appear literally
+ * in the query body. Empty `externals` is rejected — call select()
+ * directly when no external data is needed.
+ */
+PHP_METHOD(ClickHouse, selectWithExternalData)
+{
+    zend_string *sql = NULL;
+    zval *externals = NULL;
+    zval *params = NULL;
+    zend_long fetch_mode = 0;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(2, 6)
+        Z_PARAM_STR(sql)
+        Z_PARAM_ARRAY(externals)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_ARRAY(params)
+        Z_PARAM_LONG(fetch_mode)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+
+    HashTable *externals_ht = Z_ARRVAL_P(externals);
+    size_t externals_count = zend_hash_num_elements(externals_ht);
+    if (externals_count == 0) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "selectWithExternalData requires at least one external table; "
+            "use select() when no external data is needed", 0);
+        return;
+    }
+
+    /* Build the backing storage for ExternalTable entries. ExternalTable
+     * holds a string_view into the name and a const reference into the
+     * Block; both must outlive the SelectWithExternalData call. Reserve
+     * so push_back never relocates. */
+    std::vector<std::string> ext_names;
+    std::vector<Block>       ext_blocks;
+    ExternalTables           ext_tables;
+    ext_names.reserve(externals_count);
+    ext_blocks.reserve(externals_count);
+    ext_tables.reserve(externals_count);
+
+    try {
+        zval *entry;
+        ZEND_HASH_FOREACH_VAL(externals_ht, entry) {
+            std::string nm;
+            Block blk = buildExternalTableBlock(entry, nm);
+            ext_names.push_back(std::move(nm));
+            ext_blocks.push_back(std::move(blk));
+            ext_tables.push_back(ExternalTable{
+                std::string_view(ext_names.back()),
+                ext_blocks.back()
+            });
+        } ZEND_HASH_FOREACH_END();
+    } catch (const std::exception &e) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            e.what(), 0);
+        return;
+    }
+
+    std::string qid = makeQid(query_id);
+    do_select_into(return_value, getThis(), ZSTR_VAL(sql), ZSTR_LEN(sql),
+                   params, fetch_mode, qid, settings, &ext_tables);
+}
+/* }}} */
+
+/* Wire format shared by selectToStream() (output) and insertFromStream()
+ * (input). The same four formats apply to both directions; aliases
+ * TSV / TSVWithNames match ClickHouse's own short names. */
+enum class StreamFormat { TSV, TSVWithNames, CSV, CSVWithNames };
+
+static bool parseStreamFormat(const char *s, size_t l, StreamFormat &out)
+{
+    auto eq = [&](const char *lit) {
+        size_t n = strlen(lit);
+        return l == n && memcmp(s, lit, n) == 0;
+    };
+    if (eq("TabSeparated") || eq("TSV"))                    { out = StreamFormat::TSV;          return true; }
+    if (eq("TabSeparatedWithNames") || eq("TSVWithNames"))  { out = StreamFormat::TSVWithNames; return true; }
+    if (eq("CSV"))                                          { out = StreamFormat::CSV;          return true; }
+    if (eq("CSVWithNames"))                                 { out = StreamFormat::CSVWithNames; return true; }
+    return false;
+}
+
+static inline bool streamFormatIsCSV(StreamFormat f)
+{
+    return f == StreamFormat::CSV || f == StreamFormat::CSVWithNames;
+}
+
+static inline bool streamFormatHasHeader(StreamFormat f)
+{
+    return f == StreamFormat::TSVWithNames || f == StreamFormat::CSVWithNames;
+}
+
+/* Walk Nullable / LowCardinality wrappers and reject the composite
+ * column types that text formats can't unambiguously serialize. Returns
+ * true when the column is OK; on false, *reason_out names the offending
+ * type for the error message. */
+static bool isStreamableColumnType(const TypeRef &t, std::string &reason_out)
+{
+    switch (t->GetCode()) {
+        case Type::Code::Nullable:
+            return isStreamableColumnType(t->As<NullableType>()->GetNestedType(), reason_out);
+        case Type::Code::LowCardinality:
+            return isStreamableColumnType(t->As<LowCardinalityType>()->GetNestedType(), reason_out);
+        case Type::Code::Array:
+        case Type::Code::Tuple:
+        case Type::Code::Map:
+        case Type::Code::Point:
+        case Type::Code::Ring:
+        case Type::Code::Polygon:
+        case Type::Code::MultiPolygon:
+            reason_out = t->GetName();
+            return false;
+        default:
+            return true;
+    }
+}
+
+static void tsvAppendEscaped(smart_str *buf, const char *s, size_t len)
+{
+    for (size_t i = 0; i < len; ++i) {
+        char c = s[i];
+        switch (c) {
+            case '\\': smart_str_appendl(buf, "\\\\", 2); break;
+            case '\t': smart_str_appendl(buf, "\\t", 2);  break;
+            case '\n': smart_str_appendl(buf, "\\n", 2);  break;
+            case '\r': smart_str_appendl(buf, "\\r", 2);  break;
+            case '\0': smart_str_appendl(buf, "\\0", 2);  break;
+            default:   smart_str_appendc(buf, c);
+        }
+    }
+}
+
+/* RFC 4180-style CSV escape. Cell wrapped in double-quotes if it
+ * contains ", ',', '\r', or '\n'; embedded '"' is doubled. The literal
+ * non-NULL string "\N" is quoted so insertFromStream() does not read it
+ * back as the NULL marker. */
+static void csvAppendEscaped(smart_str *buf, const char *s, size_t len)
+{
+    bool needs_quoting = (len == 2 && s[0] == '\\' && s[1] == 'N');
+    for (size_t i = 0; i < len; ++i) {
+        char c = s[i];
+        if (c == '"' || c == ',' || c == '\n' || c == '\r') {
+            needs_quoting = true;
+            break;
+        }
+    }
+    if (!needs_quoting) {
+        smart_str_appendl(buf, s, len);
+        return;
+    }
+    smart_str_appendc(buf, '"');
+    for (size_t i = 0; i < len; ++i) {
+        char c = s[i];
+        if (c == '"') {
+            smart_str_appendl(buf, "\"\"", 2);
+        } else {
+            smart_str_appendc(buf, c);
+        }
+    }
+    smart_str_appendc(buf, '"');
+}
+
+/* Append one cell (already-formatted text or IS_NULL) to the per-block
+ * buffer. NULL renders as `\N` in both TSV and CSV; a non-NULL CSV string
+ * equal to "\N" is quoted by csvAppendEscaped(). */
+static void appendCellForStream(smart_str *buf, zval *cell, StreamFormat fmt)
+{
+    if (Z_TYPE_P(cell) == IS_NULL) {
+        smart_str_appendl(buf, "\\N", 2);
+        return;
+    }
+    /* Fast path for cells that are already IS_STRING — convertToZval
+     * with SC_FETCH_DATE_AS_STRINGS produces strings directly for
+     * String/FixedString, all Date* and DateTime* variants, Decimal*,
+     * Int128/UInt128, UUID, IPv4, IPv6. Avoids one zend_string heap
+     * round-trip per cell. Other types (numeric scalars) still go
+     * through zval_get_string for the snprintf-driven conversion. */
+    const char *p;
+    size_t l;
+    zend_string *zs = NULL;
+    if (Z_TYPE_P(cell) == IS_STRING) {
+        p = Z_STRVAL_P(cell);
+        l = Z_STRLEN_P(cell);
+    } else {
+        zs = zval_get_string(cell);
+        p = ZSTR_VAL(zs);
+        l = ZSTR_LEN(zs);
+    }
+    if (streamFormatIsCSV(fmt)) {
+        csvAppendEscaped(buf, p, l);
+    } else {
+        tsvAppendEscaped(buf, p, l);
+    }
+    if (zs) zend_string_release(zs);
+}
+
+/* Flush the per-block buffer to the PHP stream and reset it. Throws on
+ * short-write so the caller's catch path can map to ClickHouseException. */
+static void flushStreamBuf(smart_str *buf, php_stream *stream)
+{
+    if (!buf->s || ZSTR_LEN(buf->s) == 0) return;
+    size_t n = ZSTR_LEN(buf->s);
+    ssize_t w = php_stream_write(stream, ZSTR_VAL(buf->s), n);
+    if (w < 0 || (size_t)w != n) {
+        throw std::runtime_error("selectToStream: short write to PHP stream");
+    }
+    smart_str_free(buf);
+}
+
+/*
+ * Internal: run a SELECT and write rows directly to a PHP stream in
+ * TSV / CSV (with optional header). No PHP row-array assembly; the
+ * type-aware text comes from convertToZval into a temporary scalar
+ * zval, then zval_get_string + format-specific escape, written
+ * block-by-block to keep syscall count low.
+ */
+static zend_long do_select_to_stream(zval *this_obj,
+                                     const char *sql, size_t l_sql,
+                                     zval *params, php_stream *stream,
+                                     StreamFormat fmt,
+                                     const std::string &qid, zval *settings)
+{
+    clickhouse_object *obj = Z_CLICKHOUSE_P(this_obj);
+    zend_long total_rows = 0;
+    std::string log_sql(sql, l_sql);
+    try
+    {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block) {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        std::string sql_s = log_sql;
+        std::vector<TypedParam> typed_params;
+        if (params != NULL && Z_TYPE_P(params) == IS_ARRAY) {
+            applyPlaceholders(sql_s, Z_ARRVAL_P(params), typed_params);
+            log_sql = sql_s;
+        } else if (params != NULL && Z_TYPE_P(params) != IS_ARRAY) {
+            throw std::runtime_error("The second argument to selectToStream must be an array");
+        }
+
+        Query query = qid.empty() ? Query(sql_s) : Query(sql_s, qid);
+        attachTypedParams(query, typed_params);
+        applyMergedSettings(query, obj, settings);
+        attachProgressAndProfile(query, obj);
+        attachVerbose(query, obj);
+
+        bool header_written = false;
+        smart_str buf = {0};
+        const long fetch_mode = SC_FETCH_DATE_AS_STRINGS | SC_FETCH_ONE;
+        const char *row_term  = streamFormatIsCSV(fmt) ? "\r\n" : "\n";
+        const size_t row_term_len = streamFormatIsCSV(fmt) ? 2 : 1;
+        const char  cell_sep  = streamFormatIsCSV(fmt) ? ','  : '\t';
+
+        query.OnData([&](const Block &block) {
+            const size_t col_count = block.GetColumnCount();
+            const size_t row_count = block.GetRowCount();
+            if (col_count == 0) return;
+
+            if (!header_written) {
+                /* Validate types once, against the first non-empty block.
+                 * The server guarantees identical schema across all blocks
+                 * of a single query, so checking once is sufficient. */
+                for (size_t c = 0; c < col_count; ++c) {
+                    std::string reason;
+                    if (!isStreamableColumnType(block[c]->Type(), reason)) {
+                        throw std::runtime_error(
+                            "selectToStream: column '" + std::string(block.GetColumnName(c)) +
+                            "' has unsupported type '" + reason +
+                            "'; TSV/CSV cannot represent it");
+                    }
+                }
+                if (streamFormatHasHeader(fmt)) {
+                    for (size_t c = 0; c < col_count; ++c) {
+                        if (c > 0) smart_str_appendc(&buf, cell_sep);
+                        std::string nm = block.GetColumnName(c);
+                        if (streamFormatIsCSV(fmt)) {
+                            csvAppendEscaped(&buf, nm.data(), nm.size());
+                        } else {
+                            tsvAppendEscaped(&buf, nm.data(), nm.size());
+                        }
+                    }
+                    smart_str_appendl(&buf, row_term, row_term_len);
+                }
+                header_written = true;
+            }
+
+            for (size_t r = 0; r < row_count; ++r) {
+                for (size_t c = 0; c < col_count; ++c) {
+                    if (c > 0) smart_str_appendc(&buf, cell_sep);
+                    zval cell;
+                    ZVAL_UNDEF(&cell);
+                    try {
+                        convertToZval(&cell, block[c], r, "", 0, fetch_mode);
+                    } catch (...) {
+                        if (Z_TYPE(cell) != IS_UNDEF) zval_ptr_dtor(&cell);
+                        throw;
+                    }
+                    appendCellForStream(&buf, &cell, fmt);
+                    zval_ptr_dtor(&cell);
+                }
+                smart_str_appendl(&buf, row_term, row_term_len);
+                ++total_rows;
+            }
+
+            flushStreamBuf(&buf, stream);
+        });
+
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            client->Select(query);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            /* OnData lambda throws (unsupported column type, short stream
+             * write) bubble out mid-stream and leave the native client
+             * with unread blocks on the wire. Subsequent queries on the
+             * same handle would then receive the residual data and
+             * appear corrupted. Reset to recover; the original throw
+             * still propagates. */
+            smart_str_free(&buf);
+            tryResetConnectionReapplyDatabase(this_obj, obj, false);
+            throw;
+        }
+        flushStreamBuf(&buf, stream);
+        smart_str_free(&buf);
+        auto t1 = std::chrono::steady_clock::now();
+        obj->stats.elapsed_ms =
+            std::chrono::duration<double, std::milli>(t1 - t0).count();
+        recordQuerySuccess(obj, sql_s, qid);
+    }
+    catch (const std::exception &e)
+    {
+        recordQueryError(obj, log_sql, qid, e);
+        throwClickHouseError(e, qid);
+        return 0;
+    }
+    return total_rows;
+}
+
+/* {{{ proto int selectToStream(string sql, array params, mixed stream, string format = "TabSeparated", string query_id = "", array settings = [])
+ *
+ * Run a SELECT and write rows directly to a PHP stream resource in
+ * TSV / CSV format (with optional column-name header). Returns the
+ * number of rows written. Skips per-row PHP array assembly entirely —
+ * cells are formatted from native column data and flushed block by
+ * block to the stream. Use for large exports where selectStream() +
+ * userland fwrite() is too slow.
+ *
+ * Supported formats: TabSeparated (alias TSV), TabSeparatedWithNames
+ * (alias TSVWithNames), CSV, CSVWithNames. Other values are rejected.
+ *
+ * Dates always emit as YYYY-MM-DD / YYYY-MM-DD HH:MM:SS[.fff] strings;
+ * Decimal / Int128 / UInt128 as decimal strings. Array / Tuple / Map /
+ * geometry columns are rejected — text formats can't unambiguously
+ * serialize them. Nullable and LowCardinality wrappers around supported
+ * scalars are fine.
+ */
+PHP_METHOD(ClickHouse, selectToStream)
+{
+    zend_string *sql = NULL;
+    zval *params = NULL;
+    zval *stream_zv = NULL;
+    zend_string *format_s = NULL;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(3, 6)
+        Z_PARAM_STR(sql)
+        Z_PARAM_ARRAY(params)
+        Z_PARAM_ZVAL(stream_zv)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR(format_s)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+
+    StreamFormat fmt = StreamFormat::TSV;
+    if (format_s) {
+        if (!parseStreamFormat(ZSTR_VAL(format_s), ZSTR_LEN(format_s), fmt)) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "selectToStream: unknown format; expected TabSeparated, "
+                "TabSeparatedWithNames, CSV, or CSVWithNames", 0);
+            return;
+        }
+    }
+
+    php_stream *stream = NULL;
+    php_stream_from_zval_no_verify(stream, stream_zv);
+    if (!stream) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "selectToStream: argument 3 must be an open stream resource", 0);
+        return;
+    }
+
+    std::string qid = makeQid(query_id);
+    zend_long n = do_select_to_stream(getThis(), ZSTR_VAL(sql), ZSTR_LEN(sql),
+                                      params, stream, fmt, qid, settings);
+    if (EG(exception)) return;
+    RETURN_LONG(n);
+}
+/* }}} */
+
+/* {{{ proto ClickHouseStatement selectStatement(string sql, array params, string query_id, array settings)
+ *
+ * smi2/phpClickHouse-style result wrapper. Runs the SELECT and returns
+ * a ClickHouseStatement that implements Iterator + Countable +
+ * ArrayAccess + JsonSerializable over the materialized rows, plus
+ * fetchOne / fetchKeyPair / fetchColumn / statistics / toArray. The
+ * Statement carries a per-call stats snapshot so it survives the
+ * Client running other queries afterwards.
+ */
+PHP_METHOD(ClickHouse, selectStatement)
+{
+    zend_string *sql = NULL;
+    zval *params = NULL;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(1, 4)
+        Z_PARAM_STR(sql)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_ARRAY(params)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string qid = makeQid(query_id);
+
+    object_init_ex(return_value, clickhouse_statement_ce);
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(return_value);
+
+    do_select_into(&stmt->rows, getThis(), ZSTR_VAL(sql), ZSTR_LEN(sql), params, 0, qid, settings, NULL, &stmt->positional_rows);
+    if (EG(exception)) {
+        zval_ptr_dtor(return_value);
+        ZVAL_UNDEF(return_value);
+        return;
+    }
+
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    buildStatsArray(&stmt->statistics, obj->stats);
+}
+/* }}} */
+
+/*
+ * Pre-flight row shape check shared by insert() and write(). Each row
+ * must be an array and have at most columns_count cells; extras would
+ * be silently dropped by the per-column gather below, which was the
+ * data-integrity hole behind a row like `[1, 99]` against a
+ * single-column table landing as `1` with `99` dropped. Missing cells
+ * still surface inside buildSingleColumnZval via the per-row lookup.
+ */
+static void validateRowShapes(HashTable *values_ht, size_t columns_count)
+{
+    zval *pzval;
+    ZEND_HASH_FOREACH_VAL(values_ht, pzval) {
+        if (Z_TYPE_P(pzval) != IS_ARRAY) {
+            throw std::runtime_error(
+                "The insert function needs to pass in a two-dimensional array");
+        }
+        size_t row_count = zend_hash_num_elements(Z_ARRVAL_P(pzval));
+        if (row_count > columns_count) {
+            throw std::runtime_error(
+                "row has " + std::to_string(row_count) +
+                " cells but only " + std::to_string(columns_count) +
+                " columns were declared; extra cells would be silently dropped");
+        }
+    } ZEND_HASH_FOREACH_END();
+}
+
+/*
+ * Build a per-column PHP zval array of refcount-bumped cells for one
+ * column index. column_names != NULL: missing positional entries are
+ * looked up by name (insert() accepts {col=>val} rows). NULL:
+ * positional only.
+ *
+ * Sets *out to an IS_ARRAY zval the caller owns. On throw, *out is
+ * reset to IS_UNDEF and any partial state is freed.
+ *
+ * Streaming column-by-column instead of materializing the full
+ * column-major matrix keeps peak PHP memory at one column at a time
+ * (plus the original row-major input the caller still holds) instead
+ * of N_rows * N_cols. The native ClickHouse column built from this
+ * intermediate is also independent per call, so the caller can drop
+ * the inner zval before moving to the next column.
+ */
+static void buildSingleColumnZval(HashTable *values_ht, size_t column_index,
+                                  const std::vector<zend_string*> *column_names,
+                                  zval *out)
+{
+    array_init(out);
+    try {
+        zval *pzval, *fzval;
+        ZEND_HASH_FOREACH_VAL(values_ht, pzval) {
+            fzval = sc_zend_hash_index_find(Z_ARRVAL_P(pzval), column_index);
+            if (!fzval && column_names) {
+                zend_string *col = (*column_names)[column_index];
+                fzval = sc_zend_hash_find(Z_ARRVAL_P(pzval),
+                                          ZSTR_VAL(col), ZSTR_LEN(col));
+            }
+            if (!fzval) {
+                throw std::runtime_error(
+                    "The number of parameters inserted per line is inconsistent");
+            }
+            sc_zval_add_ref(fzval);
+            add_next_index_zval(out, fzval);
+        } ZEND_HASH_FOREACH_END();
+    } catch (...) {
+        zval_ptr_dtor(out);
+        ZVAL_UNDEF(out);
+        throw;
+    }
+}
+
+/* {{{ proto array insert(string table, array columns, array values, string query_id, array settings)
+ */
+/*
+ * Internal: run an INSERT against `table` with the column-name list
+ * `columns` and the row-major matrix `values`. On error throws a PHP
+ * exception via throwClickHouseError; callers should check EG(exception)
+ * on return. Used by both ClickHouse::insert and ClickHouse::insertAssoc
+ * (insertAssoc transposes the assoc-array input first, then calls here
+ * directly instead of going through call_user_function on "insert").
+ */
+static void do_insert_into(zval *this_obj, zend_string *table,
+                           zval *columns, zval *values,
+                           const std::string &qid, zval *settings)
+{
+    string sql;
+
+    clickhouse_object *obj = Z_CLICKHOUSE_P(this_obj);
+    try
+    {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block)
+        {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        HashTable *columns_ht = Z_ARRVAL_P(columns);
+        HashTable *values_ht = Z_ARRVAL_P(values);
+        size_t columns_count = zend_hash_num_elements(columns_ht);
+
+        /* Materialize the column-name list once, validating each entry is
+         * a string. Indexing columns_ht by 0..N-1 directly assumes a packed
+         * list; sparse or associative inputs (e.g. ['a' => 'col1']) used
+         * to NULL-deref `key` a few lines down. Iterating in HashTable order
+         * is the safe substitute. */
+        std::vector<zend_string*> column_names;
+        column_names.reserve(columns_count);
+        {
+            zval *cz;
+            ZEND_HASH_FOREACH_VAL(columns_ht, cz) {
+                if (Z_TYPE_P(cz) != IS_STRING) {
+                    throw std::runtime_error(
+                        "The columns array must be a list of column-name strings");
+                }
+                column_names.push_back(Z_STR_P(cz));
+            } ZEND_HASH_FOREACH_END();
+        }
+
+        validateRowShapes(values_ht, columns_count);
+
+        sql = getInsertSql(std::string_view(ZSTR_VAL(table), ZSTR_LEN(table)), columns);
+
+        Query insertQuery = qid.empty() ? Query(sql) : Query(sql, qid);
+        applyMergedSettings(insertQuery, obj, settings);
+        attachProgressAndProfile(insertQuery, obj);
+        attachVerbose(insertQuery, obj);
+        /* BeginInsert can throw on server-side schema errors (missing
+         * table, bad column names, permissions). The vendored client
+         * sets its inserting_ flag before sending the query and before
+         * receiving the server's schema block, so a throw past that
+         * point leaves the native client wedged: every subsequent
+         * select/execute on the same handle throws "cannot execute
+         * query while inserting" until userland resetConnection()s by
+         * hand. Recover the handle here. */
+        Block blockQuery;
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            blockQuery = client->BeginInsert(insertQuery);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            tryResetConnectionReapplyDatabase(this_obj, obj, false);
+            throw;
+        }
+        bool insert_open = true;
+        bool block_sent = false;
+
+        try {
+            Block blockInsert;
+
+            /* Build one column at a time and feed it directly into the
+             * native block. The previous path materialized a full
+             * column-major PHP zval matrix first, which doubled peak
+             * PHP memory for the duration of the call. Streaming
+             * column-by-column keeps peak intermediate memory at one
+             * column. */
+            for (size_t index = 0; index < columns_count; ++index) {
+                zval inner;
+                buildSingleColumnZval(values_ht, index, &column_names, &inner);
+                try {
+                    zvalToBlock(blockInsert, blockQuery, index, &inner);
+                } catch (...) {
+                    zval_ptr_dtor(&inner);
+                    throw;
+                }
+                zval_ptr_dtor(&inner);
+            }
+
+            /* Mark the wire dirty before SendInsertBlock so a throw
+             * during the call (transport error mid-frame, server
+             * pushback packet, etc.) routes the catch path through
+             * ResetConnection instead of EndInsert. EndInsert on a
+             * partially transmitted block can either commit the
+             * transmitted prefix or leave the native client wedged
+             * in inserting state. Same ordering streaming write()
+             * uses for insert_blocks_sent. */
+            block_sent = true;
+            client->SendInsertBlock(blockInsert);
+            client->EndInsert();
+            insert_open = false;
+            setElapsedSince(obj, t0);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            /* BeginInsert opened the wire. Three cases:
+             *   - Pre-send failure (zvalToBlock conversion throw): the
+             *     wire is healthy and no rows crossed it; EndInsert()
+             *     closes the empty insert and clears the vendored
+             *     client's inserting_ flag.
+             *   - Send/end failure (SendInsertBlock or EndInsert
+             *     itself threw): rows have been transmitted (or
+             *     partially so) and the server may have rejected them
+             *     via constraint, schema, or transport error. The
+             *     wire's native inserting_ state is dirty; EndInsert
+             *     would commit any successfully transmitted rows and
+             *     leave the client unable to run subsequent queries
+             *     ("cannot execute query while inserting" until manual
+             *     resetConnection). ResetConnection() drops the socket
+             *     and reconnects — discards any in-flight insert and
+             *     leaves the handle reusable.
+             * A throw past SendInsertBlock means the data left this
+             * process; the server alone determined whether to commit.
+             * Reset to recover the handle either way. */
+            if (insert_open) {
+                if (block_sent) {
+                    tryResetConnectionReapplyDatabase(this_obj, obj, false);
+                } else {
+                    try { client->EndInsert(); } catch (...) {}
+                }
+            }
+            throw;
+        }
+        recordQuerySuccess(obj, sql, qid);
+    }
+    catch (const std::exception& e)
+    {
+        recordQueryError(obj, sql, qid, e);
+        throwClickHouseError(e, qid);
+    }
+}
+
+PHP_METHOD(ClickHouse, insert)
+{
+    zend_string *table = NULL;
+    zval *columns;
+    zval *values;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(3, 5)
+        Z_PARAM_STR(table)
+        Z_PARAM_ARRAY(columns)
+        Z_PARAM_ARRAY(values)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+
+    do_insert_into(getThis(), table, columns, values, makeQid(query_id), settings);
+    if (EG(exception)) {
+        return;
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* === insertFromStream() ===================================================
+ *
+ * Streaming TSV / CSV ingester. Reads a PHP stream, parses rows in C++,
+ * batches them into native ClickHouse Blocks of N rows, and sends each
+ * block as it fills. No row-major PHP zval matrix is materialized for
+ * the whole file — only one batch of per-column packed arrays at a time.
+ */
+
+/* True when the column type can accept a PHP NULL — i.e. Nullable(X)
+ * or LowCardinality(Nullable(X)). Anything else would have
+ * insertColumn() silently coerce NULL to "" or 0; we want to reject
+ * the row at parse time to honor the documented contract that `\N` is
+ * only valid against Nullable columns. */
+static bool acceptsNullCell(const TypeRef &t)
+{
+    switch (t->GetCode()) {
+        case Type::Code::Nullable:
+            return true;
+        case Type::Code::LowCardinality:
+            return acceptsNullCell(t->As<LowCardinalityType>()->GetNestedType());
+        default:
+            return false;
+    }
+}
+
+/* Push the just-parsed cell into the current row. TSV uses the
+ * parser's cell_is_null flag (set when `\N` is decoded at cell start),
+ * so `\\N` — which decodes to the literal byte sequence `\N` — round-
+ * trips as the two-character IS_STRING value, not NULL. CSV has no
+ * escape protocol, so a literal unquoted `\N` cell is the only way to
+ * write NULL by our convention; we keep the bytes-based fallback for
+ * that path. The cell buffer is cleared on return. */
+static void pushCell(std::string &cell_buf, bool cell_is_quoted,
+                     bool cell_is_null, StreamFormat fmt,
+                     std::vector<zval> &row_cells)
+{
+    zval z;
+    if (cell_is_null) {
+        ZVAL_NULL(&z);
+    } else if (streamFormatIsCSV(fmt) && !cell_is_quoted &&
+               cell_buf.size() == 2 && cell_buf[0] == '\\' &&
+               cell_buf[1] == 'N') {
+        ZVAL_NULL(&z);
+    } else {
+        ZVAL_STRINGL(&z, cell_buf.data(), cell_buf.size());
+    }
+    row_cells.push_back(z);
+    cell_buf.clear();
+}
+
+struct InsertStreamParser {
+    StreamFormat fmt;
+    size_t expected_cols;
+
+    enum class State { CellStart, InCell, InQuoted, QuotePending } state = State::CellStart;
+    std::string cell_buf;
+    bool cell_is_quoted = false;     // CSV: did this cell start with `"`?
+    std::vector<zval> row_cells;
+    bool first_row_skipped = false;
+    bool prev_was_cr = false;
+    size_t pending_empty_rows = 0;
+    /* TSV-only: a `\` at the very end of one feed() chunk has to wait
+     * for the next chunk to know what it escapes. Same role prev_was_cr
+     * plays for CRLF straddling a chunk boundary. */
+    bool pending_backslash = false;
+    /* TSV-only: set when the cell opens with the literal `\N` NULL
+     * marker. ClickHouse's TSV format treats `\N` as the entire cell
+     * content for NULL — any trailing bytes before the cell separator
+     * are a parse error. Without this strictness, `\Nx` silently became
+     * the literal three-character string `\Nx` for String columns. */
+    bool cell_is_null = false;
+
+    /* Owned by the parser between calls; transferred to the caller via
+     * finishRow() and consumed there. */
+
+    void finishCell() {
+        pushCell(cell_buf, cell_is_quoted, cell_is_null, fmt, row_cells);
+        cell_is_quoted = false;
+        cell_is_null = false;
+    }
+
+    /* Hand the just-completed row to the row handler. The handler takes
+     * ownership of the zvals (we move out of row_cells). On throw, the
+     * remaining zvals in row_cells get dtor'd by the parser destructor. */
+    template<typename RowHandler>
+    void finishRow(RowHandler &on_row) {
+        if (streamFormatHasHeader(fmt) && !first_row_skipped) {
+            /* Discard the header row entirely. We don't verify it
+             * against $columns; the user is responsible for matching
+             * order. */
+            for (zval &z : row_cells) zval_ptr_dtor(&z);
+            row_cells.clear();
+            first_row_skipped = true;
+            return;
+        }
+        if (row_cells.size() != expected_cols) {
+            size_t got = row_cells.size();
+            for (zval &z : row_cells) zval_ptr_dtor(&z);
+            row_cells.clear();
+            throw std::runtime_error(
+                "insertFromStream: row has " + std::to_string(got) +
+                " cells but " + std::to_string(expected_cols) +
+                " columns were declared");
+        }
+        on_row(row_cells);
+        row_cells.clear();
+    }
+
+    ~InsertStreamParser() {
+        for (zval &z : row_cells) zval_ptr_dtor(&z);
+    }
+
+    /* Feed `len` bytes and emit each completed row through on_row. */
+    template<typename RowHandler>
+    void feed(const char *data, size_t len, RowHandler &on_row) {
+        const bool csv = streamFormatIsCSV(fmt);
+        const char cell_sep = csv ? ',' : '\t';
+
+        /* Decode the byte that follows a `\` in a TSV escape. Returns
+         * true when the byte was the escape's second character (caller
+         * should advance past it). Returns false when the byte should
+         * be re-processed as ordinary cell content (the `\` is pushed
+         * literally and the byte falls through to the rest of the
+         * byte-processing loop). Shared between the inline path (both
+         * bytes in this chunk) and the cross-chunk drain at the top of
+         * the loop, so a `\` straddling a feed() boundary decodes the
+         * same way it would inside a single chunk. */
+        auto decode_escape_byte = [&](char n) -> bool {
+            switch (n) {
+                case '\\': cell_buf.push_back('\\'); return true;
+                case 't':  cell_buf.push_back('\t'); return true;
+                case 'n':  cell_buf.push_back('\n'); return true;
+                case 'r':  cell_buf.push_back('\r'); return true;
+                case '0':  cell_buf.push_back('\0'); return true;
+                case 'N':
+                    if (cell_buf.empty()) {
+                        cell_buf.push_back('\\');
+                        cell_buf.push_back('N');
+                        cell_is_null = true;
+                        return true;
+                    }
+                    cell_buf.push_back('\\');
+                    return false;
+                default:
+                    cell_buf.push_back('\\');
+                    return false;
+            }
+        };
+
+        auto flush_pending_empty_rows = [&]() {
+            while (pending_empty_rows > 0) {
+                finishCell();
+                finishRow(on_row);
+                --pending_empty_rows;
+            }
+        };
+
+        for (size_t i = 0; i < len; ++i) {
+            char c = data[i];
+
+            /* Drain a `\` that hung at the end of the previous chunk:
+             * the byte now in hand is the escape's second character. */
+            if (pending_backslash) {
+                pending_backslash = false;
+                if (state == State::CellStart) state = State::InCell;
+                if (decode_escape_byte(c)) continue;
+                /* else `\` got pushed; fall through and process c as
+                 * ordinary content (cell_sep, row term, or push). */
+            }
+
+            /* CSV-only quoted-cell state machine. TSV ignores '"' entirely. */
+            if (csv && state == State::InQuoted) {
+                if (c == '"') {
+                    state = State::QuotePending;
+                } else {
+                    cell_buf.push_back(c);
+                }
+                continue;
+            }
+            if (csv && state == State::QuotePending) {
+                if (c == '"') {
+                    /* Doubled quote -> literal " inside cell, stay quoted. */
+                    cell_buf.push_back('"');
+                    state = State::InQuoted;
+                    continue;
+                }
+                /* RFC 4180: the only valid bytes after a closing quote
+                 * are the cell separator, a row terminator, or EOF.
+                 * Permissive parsers that accept `"ab"c` as `abc` silently
+                 * hide upstream export bugs; ClickHouse's own CSV reader
+                 * rejects this. */
+                if (c != cell_sep && c != '\n' && c != '\r') {
+                    throw std::runtime_error(
+                        "insertFromStream: malformed CSV - byte after closing "
+                        "quote must be ',', newline, or end of input");
+                }
+                state = State::InCell;
+                /* fall through so cell_sep / row terminator handlers below
+                 * see c and finalize the cell / row. */
+            }
+
+            if (c == cell_sep) {
+                finishCell();
+                state = State::CellStart;
+                continue;
+            }
+            if (c == '\n' || c == '\r') {
+                if (c == '\r') {
+                    prev_was_cr = true;
+                } else {
+                    /* '\n' after '\r' is the CRLF tail; swallow it. */
+                    if (prev_was_cr) { prev_was_cr = false; continue; }
+                }
+                /* End of row. Tolerate a trailing blank line at EOF. */
+                if (state == State::CellStart && row_cells.empty() && cell_buf.empty()) {
+                    ++pending_empty_rows;
+                    state = State::CellStart;
+                    continue;
+                }
+                finishCell();
+                finishRow(on_row);
+                state = State::CellStart;
+                continue;
+            }
+            prev_was_cr = false;
+
+            if (pending_empty_rows > 0) {
+                flush_pending_empty_rows();
+            }
+
+            /* `\N` at cell start is the whole-cell NULL marker; bytes
+             * other than the cell separator or row terminator after it
+             * are a parse error. (Pre-fix the cell silently became the
+             * literal string "\Nx..." for String columns.) */
+            if (cell_is_null) {
+                throw std::runtime_error(
+                    "insertFromStream: TSV `\\N` is the whole-cell NULL "
+                    "marker and cannot be followed by other data");
+            }
+
+            if (state == State::CellStart) {
+                if (csv && c == '"') {
+                    state = State::InQuoted;
+                    cell_is_quoted = true;
+                    continue;
+                }
+                state = State::InCell;
+                /* fallthrough */
+            }
+
+            /* TSV escapes inside an unquoted cell. CSV unquoted cells
+             * have no escape syntax. */
+            if (!csv && c == '\\') {
+                if (i + 1 >= len) {
+                    /* `\` is the last byte of this chunk; stash the
+                     * pending state and let the next feed() decode the
+                     * escape from its first byte. Without this, the `\`
+                     * would be pushed as a literal and the next chunk's
+                     * first byte would be treated as ordinary content,
+                     * silently corrupting any escape whose two bytes
+                     * straddle a 64 KiB read boundary. */
+                    pending_backslash = true;
+                    break;
+                }
+                if (decode_escape_byte(data[i + 1])) {
+                    ++i;
+                    continue;
+                }
+                /* `\` was pushed; the second byte falls through to
+                 * normal cell-content handling on the next iteration. */
+                continue;
+            }
+
+            cell_buf.push_back(c);
+        }
+    }
+
+    /* Flush any trailing partial row at EOF. A non-empty cell or
+     * non-empty row_cells means the last line had no newline; treat it
+     * as a complete row. */
+    template<typename RowHandler>
+    void finish(RowHandler &on_row) {
+        if (pending_backslash) {
+            /* TSV file ends mid-escape. Treat the dangling `\` as a
+             * literal trailing character of the current cell; matches
+             * how a `\X` at end of a single chunk would have been
+             * handled if the next byte were not an escape letter. */
+            cell_buf.push_back('\\');
+            pending_backslash = false;
+            if (state == State::CellStart) state = State::InCell;
+        }
+        if (state == State::QuotePending) {
+            /* Closing quote on the final line. */
+            state = State::InCell;
+        }
+        if (state == State::InQuoted) {
+            throw std::runtime_error("insertFromStream: unterminated quoted CSV cell at EOF");
+        }
+        pending_empty_rows = 0;
+        /* cell_is_quoted catches the "" no-trailing-newline case: the
+         * cell was started by an opening quote and immediately closed,
+         * leaving cell_buf and row_cells both empty; without this flag
+         * the row would be silently dropped at EOF. */
+        if (!cell_buf.empty() || !row_cells.empty() || cell_is_quoted) {
+            finishCell();
+            finishRow(on_row);
+        }
+    }
+};
+
+/* {{{ proto int insertFromStream(string table, array columns, mixed stream, string format = "TabSeparated", int batch_rows = 10000, string query_id = "", array settings = [])
+ *
+ * Stream-parse a TSV / CSV file (or any PHP stream resource) and INSERT
+ * the rows into `table` in batches of `batch_rows`. Bytes are parsed in
+ * C++; only `batch_rows` worth of per-column zvals exist at a time, so
+ * the function works on inputs larger than memory.
+ *
+ * Formats: TabSeparated (alias TSV), TabSeparatedWithNames (alias
+ * TSVWithNames), CSV, CSVWithNames. For `*WithNames`, the first row of
+ * the input is discarded (the user is responsible for matching column
+ * order to `$columns`).
+ *
+ * NULL: a literal cell `\N` (in either format) becomes a PHP null and
+ * is rejected unless the target column is Nullable. Empty CSV cells
+ * become empty strings, not NULL.
+ *
+ * Type coercion is delegated to the existing insertColumn() path; all
+ * cells arrive there as IS_STRING (PHP coerces numeric strings to int
+ * and float for the appropriate column types). `Time` columns reject
+ * string input — TSV / CSV imports are not supported for Time today.
+ *
+ * Returns the total number of rows inserted.
+ */
+PHP_METHOD(ClickHouse, insertFromStream)
+{
+    zend_string *table = NULL;
+    zval *columns = NULL;
+    zval *stream_zv = NULL;
+    zend_string *format_s = NULL;
+    zend_long batch_rows = 10000;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(3, 7)
+        Z_PARAM_STR(table)
+        Z_PARAM_ARRAY(columns)
+        Z_PARAM_ZVAL(stream_zv)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR(format_s)
+        Z_PARAM_LONG(batch_rows)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+
+    StreamFormat fmt = StreamFormat::TSV;
+    if (format_s) {
+        if (!parseStreamFormat(ZSTR_VAL(format_s), ZSTR_LEN(format_s), fmt)) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "insertFromStream: unknown format; expected TabSeparated, "
+                "TabSeparatedWithNames, CSV, or CSVWithNames", 0);
+            return;
+        }
+    }
+
+    if (batch_rows < 1) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "insertFromStream: batch_rows must be >= 1", 0);
+        return;
+    }
+
+    php_stream *stream = NULL;
+    php_stream_from_zval_no_verify(stream, stream_zv);
+    if (!stream) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "insertFromStream: argument 3 must be an open stream resource", 0);
+        return;
+    }
+
+    HashTable *columns_ht = Z_ARRVAL_P(columns);
+    size_t columns_count = zend_hash_num_elements(columns_ht);
+    if (columns_count == 0) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "insertFromStream: columns list cannot be empty", 0);
+        return;
+    }
+
+    std::string qid = makeQid(query_id);
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    std::string sql;
+    zend_long total_rows = 0;
+
+    try {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block) {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        /* Validate column-name list shape (same rules as insert()). */
+        std::vector<zend_string*> column_names;
+        column_names.reserve(columns_count);
+        {
+            zval *cz;
+            ZEND_HASH_FOREACH_VAL(columns_ht, cz) {
+                if (Z_TYPE_P(cz) != IS_STRING) {
+                    throw std::runtime_error(
+                        "insertFromStream: columns must be a list of column-name strings");
+                }
+                column_names.push_back(Z_STR_P(cz));
+            } ZEND_HASH_FOREACH_END();
+        }
+
+        sql = getInsertSql(std::string_view(ZSTR_VAL(table), ZSTR_LEN(table)), columns);
+
+        Query insertQuery = qid.empty() ? Query(sql) : Query(sql, qid);
+        applyMergedSettings(insertQuery, obj, settings);
+        attachProgressAndProfile(insertQuery, obj);
+        attachVerbose(insertQuery, obj);
+
+        Block blockQuery;
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            blockQuery = client->BeginInsert(insertQuery);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            tryResetConnectionReapplyDatabase(getThis(), obj, false);
+            throw;
+        }
+
+        bool insert_open  = true;
+        bool block_dirty  = false;
+
+        /* Per-column packed-zval accumulators. One IS_ARRAY zval per
+         * column; each insert from the row handler appends one cell. */
+        std::vector<zval> col_zvals(columns_count);
+        for (size_t c = 0; c < columns_count; ++c) array_init(&col_zvals[c]);
+
+        auto cleanup_col_zvals = [&]() {
+            for (size_t c = 0; c < columns_count; ++c) {
+                if (Z_TYPE(col_zvals[c]) != IS_UNDEF) zval_ptr_dtor(&col_zvals[c]);
+            }
+        };
+
+        size_t pending_rows = 0;
+
+        auto flush_batch = [&]() {
+            if (pending_rows == 0) return;
+            Block blockInsert;
+            for (size_t c = 0; c < columns_count; ++c) {
+                zvalToBlock(blockInsert, blockQuery, c, &col_zvals[c]);
+                zval_ptr_dtor(&col_zvals[c]);
+                array_init(&col_zvals[c]);
+            }
+            block_dirty = true;
+            client->SendInsertBlock(blockInsert);
+            block_dirty = false;
+            pending_rows = 0;
+        };
+
+        auto on_row = [&](std::vector<zval> &row) {
+            /* row.size() == columns_count is guaranteed by InsertStreamParser. */
+            for (size_t c = 0; c < columns_count; ++c) {
+                /* `\N` (TSV) or literal-bytes `\N` (CSV) become IS_NULL
+                 * at the parser. The server schema tells us whether the
+                 * destination column can accept that NULL. Without this
+                 * check, insertColumn() silently coerces NULL to "" for
+                 * String / FixedString (zval_get_string on IS_NULL) and
+                 * to 0 for numeric types — contradicting the documented
+                 * "\N is rejected unless target column is Nullable"
+                 * contract and corrupting imports. */
+                if (Z_TYPE(row[c]) == IS_NULL &&
+                    !acceptsNullCell(blockQuery[c]->Type())) {
+                    throw std::runtime_error(
+                        "insertFromStream: NULL (`\\N`) value for column '" +
+                        std::string(blockQuery.GetColumnName(c)) +
+                        "' which is not Nullable");
+                }
+                /* Move cell zval into the column accumulator. The cell
+                 * already owns its IS_STRING buffer; ownership transfers
+                 * with add_next_index_zval. */
+                add_next_index_zval(&col_zvals[c], &row[c]);
+                ZVAL_UNDEF(&row[c]);
+            }
+            ++pending_rows;
+            ++total_rows;
+            if (pending_rows >= (size_t)batch_rows) {
+                flush_batch();
+            }
+        };
+
+        try {
+            InsertStreamParser parser;
+            parser.fmt = fmt;
+            parser.expected_cols = columns_count;
+            /* Pre-size the parser's per-row containers so the first row
+             * doesn't pay vector / std::string reallocations. cell_buf's
+             * SSO is only ~15 bytes on common libstdc++ builds; a 256-
+             * byte hint covers most TSV/CSV cells (URLs, names, UUIDs,
+             * timestamps) without growth. */
+            parser.row_cells.reserve(columns_count);
+            parser.cell_buf.reserve(256);
+
+            constexpr size_t CHUNK = 64 * 1024;
+            char buf[CHUNK];
+            while (true) {
+                ssize_t n = php_stream_read(stream, buf, CHUNK);
+                if (n < 0) {
+                    /* Wrapper signaled an error. Routing through the
+                     * catch path means the existing recovery resets the
+                     * connection if any rows have been queued, so the
+                     * partially-parsed data does not commit. The
+                     * pre-fix `if (n <= 0) break` treated this case as
+                     * clean EOF and silently inserted whatever had been
+                     * parsed up to the failure. */
+                    throw std::runtime_error(
+                        "insertFromStream: stream read returned an error");
+                }
+                if (n == 0) {
+                    if (php_stream_eof(stream)) break;
+                    /* Some wrappers return 0 without setting the EOF
+                     * flag (transient unavailability, badly-written
+                     * userland wrapper). Don't conflate with EOF. */
+                    throw std::runtime_error(
+                        "insertFromStream: stream returned 0 bytes without reaching EOF");
+                }
+                parser.feed(buf, (size_t)n, on_row);
+            }
+            parser.finish(on_row);
+
+            flush_batch();
+            client->EndInsert();
+            insert_open = false;
+            setElapsedSince(obj, t0);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            cleanup_col_zvals();
+            if (insert_open) {
+                if (block_dirty || total_rows > 0) {
+                    tryResetConnectionReapplyDatabase(getThis(), obj, false);
+                } else {
+                    try { client->EndInsert(); } catch (...) {}
+                }
+            }
+            throw;
+        }
+
+        cleanup_col_zvals();
+        recordQuerySuccess(obj, sql, qid);
+    }
+    catch (const std::exception &e) {
+        recordQueryError(obj, sql, qid, e);
+        throwClickHouseError(e, qid);
+        return;
+    }
+
+    RETURN_LONG(total_rows);
+}
+/* }}} */
+
+/* {{{ proto bool writeStart(string table, array columns, string query_id, array settings)
+ */
+PHP_METHOD(ClickHouse, writeStart)
+{
+    zend_string *table = NULL;
+    zval *columns;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    string sql;
+
+    ZEND_PARSE_PARAMETERS_START(2, 4)
+        Z_PARAM_STR(table)
+        Z_PARAM_ARRAY(columns)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string qid = makeQid(query_id);
+
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    try
+    {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block)
+        {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        sql = getInsertSql(std::string_view(ZSTR_VAL(table), ZSTR_LEN(table)), columns);
+
+        Query insertQuery = qid.empty() ? Query(sql) : Query(sql, qid);
+        applyMergedSettings(insertQuery, obj, settings);
+        attachProgressAndProfile(insertQuery, obj);
+        attachVerbose(insertQuery, obj);
+        /* Same recovery pattern as direct insert(): BeginInsert can
+         * throw on server-side schema errors after the vendored
+         * client has set its inserting_ flag. Reset the connection
+         * so the next call on this handle isn't met with "cannot
+         * execute query while inserting". */
+        Block blockQuery;
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            blockQuery = client->BeginInsert(insertQuery);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            tryResetConnectionReapplyDatabase(getThis(), obj, false);
+            throw;
+        }
+
+        obj->insert_block = blockQuery;
+        obj->has_insert_block = true;
+        obj->insert_blocks_sent = false;
+        obj->insert_sql = sql;
+        obj->insert_query_id = qid;
+        obj->insert_started_at = t0;
+    }
+    catch (const std::exception& e)
+    {
+        recordQueryError(obj, sql, qid, e);
+        throwClickHouseError(e, qid);
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto array insert(string table, array columns, array values)
+ */
+PHP_METHOD(ClickHouse, write)
+{
+    zval *values;
+
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ARRAY(values)
+    ZEND_PARSE_PARAMETERS_END();
+
+    try
+    {
+        /* Acquire the reentry guard before any heavy lifting. Fail-fast
+         * on reentry rather than spending CPU and memory on a per-column
+         * build that will be discarded. */
+        clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        if (!obj->has_insert_block) {
+            throw std::runtime_error("write() called without a matching writeStart()");
+        }
+
+        HashTable *values_ht = Z_ARRVAL_P(values);
+        if (zend_hash_num_elements(values_ht) == 0) {
+            throw std::runtime_error("Empty rows array passed to write()");
+        }
+
+        /* The native block was prepared by BeginInsert() with one column
+         * per declared writeStart() column. Use that count as the
+         * authoritative row width — not the first row's element count.
+         * Otherwise a row like [1] against `writeStart(t, ['a','b'])`
+         * silently sends a one-column block; the server fills the
+         * missing column with its default and the streamed insert
+         * returns success on truncated rows. */
+        size_t columns_count = obj->insert_block.GetColumnCount();
+        if (columns_count == 0) {
+            throw std::runtime_error(
+                "writeStart() block has no columns; cannot stream rows");
+        }
+
+        validateRowShapes(values_ht, columns_count);
+        Block &blockQuery = obj->insert_block;
+
+        Block blockInsert;
+        for (size_t index = 0; index < columns_count; ++index) {
+            zval inner;
+            buildSingleColumnZval(values_ht, index, NULL, &inner);
+            try {
+                zvalToBlock(blockInsert, blockQuery, index, &inner);
+            } catch (...) {
+                zval_ptr_dtor(&inner);
+                throw;
+            }
+            zval_ptr_dtor(&inner);
+        }
+
+        /* Mark the session dirty before SendInsertBlock so any failure
+         * mid-send routes the catch path to ResetConnection() instead
+         * of EndInsert(). EndInsert on a wire that already saw partial
+         * data would commit it. */
+        obj->insert_blocks_sent = true;
+        client->SendInsertBlock(blockInsert);
+    }
+    catch (const std::exception& e)
+    {
+        /* Three failure phases now, each needing a different recovery:
+         *   - Pre-send on a clean session (no prior write() succeeded):
+         *     the wire is healthy and no data has crossed it; close the
+         *     empty insert with EndInsert() so the vendored Client
+         *     clears its `inserting_` flag. Without this, the next
+         *     select/execute throws "cannot execute query while
+         *     inserting" and the caller has to resetConnection() by
+         *     hand.
+         *   - Pre-send on a dirty session (a prior write() already sent
+         *     blocks): EndInsert() would commit those earlier blocks
+         *     and turn this thrown write() into a silent partial commit.
+         *     Drop the connection instead so the server discards the
+         *     in-flight insert.
+         *   - Mid-send (SendInsertBlock itself threw): the wire is
+         *     dirty regardless of prior writes; reset for the same
+         *     reason. insert_blocks_sent is set true before
+         *     SendInsertBlock so this case lands here too.
+         * Swallow secondary failures from the recovery call — we are
+         * already throwing the original error to PHP. */
+        clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+        std::string sql = obj->insert_sql;
+        std::string qid = obj->insert_query_id;
+        if (obj->has_insert_block) {
+            setElapsedSince(obj, obj->insert_started_at);
+        }
+        if (obj->client && obj->has_insert_block) {
+            if (obj->insert_blocks_sent) {
+                tryResetConnectionReapplyDatabase(getThis(), obj, false);
+            } else {
+                try { obj->client->EndInsert(); } catch (...) {}
+            }
+        }
+        if (!sql.empty() || !qid.empty()) {
+            recordQueryError(obj, sql, qid, e);
+        }
+        clearStreamingInsertState(obj);
+        throwClickHouseError(e, qid);
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto array insert(string table, array columns, array values)
+ */
+PHP_METHOD(ClickHouse, writeEnd)
+{
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    std::string sql;
+    std::string qid;
+    bool end_insert_started = false;
+    try
+    {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+        if (!obj->has_insert_block) {
+            throw std::runtime_error("writeEnd() called without a matching writeStart()");
+        }
+
+        sql = obj->insert_sql;
+        qid = obj->insert_query_id;
+        auto started_at = obj->insert_started_at;
+        end_insert_started = true;
+        client->EndInsert();
+        setElapsedSince(obj, started_at);
+        recordQuerySuccess(obj, sql, qid);
+        clearStreamingInsertState(obj);
+    }
+    catch (const std::exception& e)
+    {
+        /* Reset the native connection on EndInsert failure. The server
+         * may have rejected the insert (constraint violation, schema
+         * drift, etc.) or the wire is otherwise dirty; in either case
+         * the vendored client's inserting_ flag is still set, so the
+         * next select/execute on this same handle would throw "cannot
+         * execute query while inserting" and the caller would have to
+         * resetConnection() by hand. Drop the socket and reconnect so
+         * the handle stays usable. Swallow secondary failures from the
+         * reset — we are already throwing the original error. */
+        if (obj->has_insert_block) {
+            setElapsedSince(obj, obj->insert_started_at);
+        }
+        if (end_insert_started && obj->client) {
+            tryResetConnectionReapplyDatabase(getThis(), obj, true);
+        }
+        if (!sql.empty() || !qid.empty()) {
+            recordQueryError(obj, sql, qid, e);
+        }
+        if (end_insert_started && obj->has_insert_block) {
+            clearStreamingInsertState(obj);
+        }
+        throwClickHouseError(e, qid);
+        return;
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto bool execute(string sql, array params, string query_id, array settings)
+ */
+/*
+ * Internal: run a non-result-bearing statement (DDL, INSERT...SELECT,
+ * SET, etc.). On error throws a PHP exception via throwClickHouseError;
+ * callers should check EG(exception) on return. Used by ClickHouse::execute
+ * and the SQL-helper one-liners (truncateTable, dropPartition, ...) so
+ * those don't have to round-trip through call_user_function on "execute".
+ */
+static void do_execute_into(zval *this_obj,
+                            const char *sql, size_t l_sql,
+                            zval *params, const std::string &qid, zval *settings)
+{
+    clickhouse_object *obj = Z_CLICKHOUSE_P(this_obj);
+    std::string log_sql(sql, l_sql);
+    try
+    {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block)
+        {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        string sql_s = log_sql;
+        std::vector<TypedParam> typed_params;
+
+        if (params != NULL && Z_TYPE_P(params) == IS_ARRAY)
+        {
+            applyPlaceholders(sql_s, Z_ARRVAL_P(params), typed_params);
+            log_sql = sql_s;
+        } else if (params != NULL && Z_TYPE_P(params) != IS_ARRAY) {
+            throw std::runtime_error("The second argument to execute must be an array");
+        }
+
+        Query query = qid.empty() ? Query(sql_s) : Query(sql_s, qid);
+        attachTypedParams(query, typed_params);
+        applyMergedSettings(query, obj, settings);
+        attachProgressAndProfile(query, obj);
+        attachVerbose(query, obj);
+
+        if (verbose_active(obj)) {
+            zval ctx;
+            array_init(&ctx);
+            add_assoc_stringl(&ctx, "sql", (char*)sql_s.data(), sql_s.size());
+            add_assoc_stringl(&ctx, "query_id", (char*)qid.data(), qid.size());
+            add_assoc_long(&ctx, "settings_count", (zend_long)obj->settings.size());
+            emitVerbose(obj, "execute_start", &ctx);
+        }
+
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            client->Execute(query);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            tryResetConnectionReapplyDatabase(this_obj, obj, false);
+            throw;
+        }
+        auto t1 = std::chrono::steady_clock::now();
+        obj->stats.elapsed_ms =
+            std::chrono::duration<double, std::milli>(t1 - t0).count();
+        recordQuerySuccess(obj, sql_s, qid);
+
+        if (verbose_active(obj)) {
+            zval ctx;
+            array_init(&ctx);
+            add_assoc_double(&ctx, "elapsed_ms", obj->stats.elapsed_ms);
+            emitVerbose(obj, "execute_finish", &ctx);
+        }
+    }
+    catch (const std::exception& e)
+    {
+        recordQueryError(obj, log_sql, qid, e);
+        throwClickHouseError(e, qid);
+    }
+}
+
+PHP_METHOD(ClickHouse, execute)
+{
+    zend_string *sql = NULL;
+    zval* params = NULL;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(1, 4)
+        Z_PARAM_STR(sql)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_ARRAY(params)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+
+    do_execute_into(getThis(), ZSTR_VAL(sql), ZSTR_LEN(sql), params,
+                    makeQid(query_id), settings);
+    if (EG(exception)) {
+        return;
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto static setSettings(array settings)
+ *
+ * Replace the client-wide settings map. Per-call settings supplied to
+ * select/insert/execute/writeStart override these. Pass an empty array
+ * to clear. Returns $this so callers can chain.
+ */
+PHP_METHOD(ClickHouse, setSettings)
+{
+    zval *arr;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ARRAY(arr)
+    ZEND_PARSE_PARAMETERS_END();
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    /* Build into a temporary first so a malformed key doesn't leave the
+     * caller's settings half-replaced. The validator matches setSetting's
+     * (numeric keys rejected, empty strings rejected) so a setting array
+     * accepted by one entry point is also accepted by the other. */
+    std::unordered_map<std::string, std::string> m;
+    HashTable *ht = Z_ARRVAL_P(arr);
+    zval *vz;
+    zend_string *zk;
+    zend_ulong nk;
+    try {
+        ZEND_HASH_FOREACH_KEY_VAL(ht, nk, zk, vz) {
+            (void)nk;
+            if (!zk) {
+                sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                    "setting keys must be strings", 0);
+                return;
+            }
+            if (ZSTR_LEN(zk) == 0) {
+                sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                    "setting key must not be empty", 0);
+                return;
+            }
+            m[std::string(ZSTR_VAL(zk), ZSTR_LEN(zk))] = formatScalarParam(vz);
+        } ZEND_HASH_FOREACH_END();
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    obj->settings = std::move(m);
+    RETURN_ZVAL(getThis(), 1, 0);
+}
+/* }}} */
+
+/* {{{ proto static setSetting(string key, mixed value)
+ *
+ * Set a single client-wide setting. Equivalent to calling setSettings()
+ * with a one-key array merged onto the existing map. Returns $this.
+ */
+PHP_METHOD(ClickHouse, setSetting)
+{
+    zend_string *key = NULL;
+    zval *value = NULL;
+    ZEND_PARSE_PARAMETERS_START(2, 2)
+        Z_PARAM_STR(key)
+        Z_PARAM_ZVAL(value)
+    ZEND_PARSE_PARAMETERS_END();
+    if (ZSTR_LEN(key) == 0) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "setting key must not be empty", 0);
+        return;
+    }
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    try {
+        obj->settings[std::string(ZSTR_VAL(key), ZSTR_LEN(key))] = formatScalarParam(value);
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    RETURN_ZVAL(getThis(), 1, 0);
+}
+/* }}} */
+
+/* {{{ proto static setDatabase(string database)
+ *
+ * Switch the active database for subsequent queries. Issues USE on the
+ * server, then updates the cached `database` property used by helpers
+ * that take a default database (databaseSize, tablesSize, etc.).
+ * Returns $this.
+ */
+PHP_METHOD(ClickHouse, setDatabase)
+{
+    zend_string *db = NULL;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_STR(db)
+    ZEND_PARSE_PARAMETERS_END();
+    if (ZSTR_LEN(db) == 0) {
+        throwClickHouseError(std::runtime_error("database name must not be empty"));
+        return;
+    }
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    if (!obj->client) {
+        throwClickHouseError(std::runtime_error("Client is not connected"));
+        return;
+    }
+    try {
+        QueryActiveGuard guard(obj);
+        obj->client->Execute(Query("USE " + sqlQuotedIdentifier(std::string(ZSTR_VAL(db), ZSTR_LEN(db)))));
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    sc_zend_update_property_stringl(clickhouse_ce, getThis(), "database", sizeof("database") - 1, ZSTR_VAL(db), ZSTR_LEN(db));
+    RETURN_ZVAL(getThis(), 1, 0);
+}
+/* }}} */
+
+/* {{{ proto bool setProgressCallback(?callable callback)
+ *
+ * Register a callable invoked for each Progress packet the server
+ * sends during select/execute. Pass null to remove. Callback receives
+ * a single associative array: rows, bytes, total_rows, written_rows,
+ * written_bytes.
+ */
+/*
+ * Register-or-clear shared body for the three callback setters. Returns
+ * false on the not-callable error path (with EG(exception) set), true
+ * on success or null-clear.
+ */
+static bool setCallbackField(zval *target, zval *cb, const char *err_name)
+{
+    bool clear_only = Z_TYPE_P(cb) == IS_NULL;
+    if (!clear_only && !zend_is_callable(cb, 0, NULL)) {
+        std::string msg = std::string(err_name) + " expects a callable or null";
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce, msg.c_str(), 0);
+        return false;
+    }
+    if (Z_TYPE(*target) != IS_UNDEF) {
+        zval_ptr_dtor(target);
+        ZVAL_UNDEF(target);
+    }
+    if (clear_only) {
+        return true;
+    }
+    ZVAL_COPY(target, cb);
+    return true;
+}
+
+PHP_METHOD(ClickHouse, setProgressCallback)
+{
+    zval *cb;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ZVAL(cb)
+    ZEND_PARSE_PARAMETERS_END();
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    if (!setCallbackField(&obj->progress_callback, cb, "setProgressCallback")) {
+        return;
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto bool setProfileCallback(?callable callback)
+ *
+ * Register a callable invoked for each Profile packet the server sends
+ * (typically once at end of select / execute). Pass null to remove.
+ * Callback receives a single associative array: rows, blocks, bytes,
+ * rows_before_limit, applied_limit, calculated_rows_before_limit.
+ */
+PHP_METHOD(ClickHouse, setProfileCallback)
+{
+    zval *cb;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ZVAL(cb)
+    ZEND_PARSE_PARAMETERS_END();
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    if (!setCallbackField(&obj->profile_callback, cb, "setProfileCallback")) {
+        return;
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto static setVerbose(bool|callable sink)
+ *
+ * Enable protocol-level lifecycle tracing. Pass true to log JSON
+ * lines on STDERR, false to disable, or a callable invoked with
+ * (string $event, array $context) per event. Events: select_start,
+ * data_block, select_finish, execute_start, execute_finish,
+ * server_exception (plus the existing progress / profile callbacks
+ * are unaffected). Returns $this so callers can chain.
+ */
+PHP_METHOD(ClickHouse, setVerbose)
+{
+    zval *sink;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ZVAL(sink)
+    ZEND_PARSE_PARAMETERS_END();
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+
+    bool new_verbose_to_stderr = false;
+    bool new_callback = false;
+
+    if (Z_TYPE_P(sink) == IS_TRUE) {
+        new_verbose_to_stderr = true;
+    } else if (Z_TYPE_P(sink) == IS_FALSE || Z_TYPE_P(sink) == IS_NULL) {
+    } else if (zend_is_callable(sink, 0, NULL)) {
+        new_callback = true;
+    } else {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "setVerbose expects bool, null, or callable", 0);
+        return;
+    }
+
+    if (Z_TYPE(obj->verbose_callback) != IS_UNDEF) {
+        zval_ptr_dtor(&obj->verbose_callback);
+        ZVAL_UNDEF(&obj->verbose_callback);
+    }
+    obj->verbose_to_stderr = new_verbose_to_stderr;
+
+    if (new_callback) {
+        ZVAL_COPY(&obj->verbose_callback, sink);
+    }
+    RETURN_ZVAL(getThis(), 1, 0);
+}
+/* }}} */
+
+/* {{{ proto bool resetConnection()
+ *
+ * Force-close and re-open the underlying TCP connection. Useful for
+ * long-lived workers that want a clean socket after an idle period or
+ * after a server-side restart.
+ */
+PHP_METHOD(ClickHouse, resetConnection)
+{
+    try {
+        clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+        QueryActiveGuard guard(obj);
+        resetConnectionReapplyDatabase(getThis(), obj, true);
+    } catch (const std::exception& e) {
+        throwClickHouseError(e);
+        return;
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto array getServerInfo()
+ *
+ * Return the server identification banner from the most recent connect:
+ * name, display_name, version_major, version_minor, version_patch,
+ * revision, timezone.
+ */
+PHP_METHOD(ClickHouse, getServerInfo)
+{
+    try {
+        clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+        Client *client = getClient(obj);
+        const ServerInfo &si = client->GetServerInfo();
+        array_init(return_value);
+        add_assoc_stringl(return_value, "name", (char*)si.name.c_str(), si.name.size());
+        add_assoc_stringl(return_value, "display_name", (char*)si.display_name.c_str(), si.display_name.size());
+        add_assoc_long(return_value, "version_major", (zend_long)si.version_major);
+        add_assoc_long(return_value, "version_minor", (zend_long)si.version_minor);
+        add_assoc_long(return_value, "version_patch", (zend_long)si.version_patch);
+        add_assoc_long(return_value, "revision", (zend_long)si.revision);
+        add_assoc_stringl(return_value, "timezone", (char*)si.timezone.c_str(), si.timezone.size());
+    } catch (const std::exception& e) {
+        throwClickHouseError(e);
+        return;
+    }
+}
+/* }}} */
+
+/* {{{ proto ?array getCurrentEndpoint()
+ *
+ * Return the active endpoint as ["host" => ..., "port" => ...]. The
+ * single host/port config is modeled internally as a 1-item endpoints
+ * list, so this is non-null in normal operation. Returns null only if
+ * the underlying client has no endpoint resolved yet.
+ */
+PHP_METHOD(ClickHouse, getCurrentEndpoint)
+{
+    try {
+        clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+        Client *client = getClient(obj);
+        const std::optional<Endpoint> &ep = client->GetCurrentEndpoint();
+        if (!ep.has_value()) {
+            RETURN_NULL();
+        }
+        array_init(return_value);
+        add_assoc_stringl(return_value, "host", (char*)ep->host.c_str(), ep->host.size());
+        add_assoc_long(return_value, "port", (zend_long)ep->port);
+    } catch (const std::exception& e) {
+        throwClickHouseError(e);
+        return;
+    }
+}
+/* }}} */
+
+/* {{{ proto array getStatistics()
+ *
+ * Return the rows / bytes / time recorded for the last completed
+ * select / execute / insert. Reset on every query.
+ */
+PHP_METHOD(ClickHouse, getStatistics)
+{
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    buildStatsArray(return_value, obj->stats);
+}
+/* }}} */
+
+/* {{{ proto bool insertAssoc(string table, array rows, string query_id, array settings)
+ *
+ * Convenience wrapper over insert(): derives the column list from the
+ * keys of the first row, then forwards to insert(). All rows must share
+ * the same key set (positional alignment is by first-row key order).
+ */
+PHP_METHOD(ClickHouse, insertAssoc)
+{
+    zend_string *table = NULL;
+    zval *rows;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(2, 4)
+        Z_PARAM_STR(table)
+        Z_PARAM_ARRAY(rows)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string qid = makeQid(query_id);
+
+    try {
+        HashTable *rows_ht = Z_ARRVAL_P(rows);
+        if (zend_hash_num_elements(rows_ht) == 0) {
+            throw std::runtime_error("insertAssoc: rows is empty");
+        }
+        zval *first = NULL;
+        {
+            zval *fz;
+            ZEND_HASH_FOREACH_VAL(rows_ht, fz) {
+                first = fz;
+                break;
+            } ZEND_HASH_FOREACH_END();
+        }
+        if (!first || Z_TYPE_P(first) != IS_ARRAY) {
+            throw std::runtime_error("insertAssoc: each row must be an associative array");
+        }
+
+        zval columns_zv;
+        array_init(&columns_zv);
+
+        /* First row defines the column set. Validate its keys are all
+         * strings while building the columns list. The first row's
+         * HashTable then doubles as the expected-key oracle for every
+         * later row — zend_hash_exists takes a zend_string and
+         * compares hashes without allocating, which avoids the
+         * std::string copy that an std::unordered_set lookup needed. */
+        HashTable *first_ht = Z_ARRVAL_P(first);
+        size_t expected_count = zend_hash_num_elements(first_ht);
+        {
+            zval *fv;
+            zend_string *fk;
+            zend_ulong fnk;
+            ZEND_HASH_FOREACH_KEY_VAL(first_ht, fnk, fk, fv) {
+                (void)fv;
+                (void)fnk;
+                if (!fk) {
+                    zval_ptr_dtor(&columns_zv);
+                    throw std::runtime_error("insertAssoc: each row must have string keys (column names)");
+                }
+                add_next_index_stringl(&columns_zv, ZSTR_VAL(fk), ZSTR_LEN(fk));
+            } ZEND_HASH_FOREACH_END();
+        }
+
+        /* Validate every row against the first row's key set without
+         * materializing a positional copy. do_insert_into →
+         * buildSingleColumnZval handles the value lookup later, but
+         * its gatherer tries integer-index lookup before name lookup;
+         * a later row like `[0 => 99, "b" => 4]` would silently land
+         * 99 into the first column instead of throwing. Reject any
+         * row that has integer keys or whose string-key set drifts
+         * from the first row's set. */
+        zval *row_zv;
+        ZEND_HASH_FOREACH_VAL(rows_ht, row_zv) {
+            if (Z_TYPE_P(row_zv) != IS_ARRAY) {
+                zval_ptr_dtor(&columns_zv);
+                throw std::runtime_error("insertAssoc: each row must be an associative array");
+            }
+            HashTable *row_ht = Z_ARRVAL_P(row_zv);
+            if (zend_hash_num_elements(row_ht) != expected_count) {
+                zval_ptr_dtor(&columns_zv);
+                throw std::runtime_error(
+                    "insertAssoc: row key count differs from first row");
+            }
+            zend_string *rk;
+            zend_ulong rnk;
+            zval *rv;
+            ZEND_HASH_FOREACH_KEY_VAL(row_ht, rnk, rk, rv) {
+                (void)rv;
+                (void)rnk;
+                if (!rk) {
+                    zval_ptr_dtor(&columns_zv);
+                    throw std::runtime_error(
+                        "insertAssoc: each row must have string keys (column names)");
+                }
+                if (!zend_hash_exists(first_ht, rk)) {
+                    zval_ptr_dtor(&columns_zv);
+                    throw std::runtime_error(
+                        std::string("insertAssoc: unexpected key '") +
+                        std::string(ZSTR_VAL(rk), ZSTR_LEN(rk)) +
+                        "' (not in first row's column set)");
+                }
+            } ZEND_HASH_FOREACH_END();
+        } ZEND_HASH_FOREACH_END();
+
+        /* Dispatch directly into the shared insert helper. The previous
+         * version went through call_user_function("insert", ...) which
+         * added a full PHP method-dispatch frame on every assoc insert
+         * and exposed the helper to user-defined subclass overrides
+         * of insert(). */
+        do_insert_into(getThis(), table, &columns_zv, rows, qid, settings);
+        zval_ptr_dtor(&columns_zv);
+        if (EG(exception)) {
+            return;
+        }
+    } catch (const std::exception &e) {
+        throwClickHouseError(e, qid);
+        return;
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/*
+ * SQL-helper one-liners. Each builds a small SELECT and reuses the
+ * select() machinery directly through do_select_into / do_execute_into
+ * so settings, progress, stats, and the verbose trace surface apply
+ * exactly the same as on the user-visible select() / execute(). The
+ * old call_user_function indirection added a full PHP method-dispatch
+ * frame per helper call (and exposed the helpers to user-defined
+ * subclass overrides of select/execute, which wasn't intended).
+ */
+static void runHelperSelect(zval *return_value, zval *this_obj, const std::string &sql, zend_long fetch_mode)
+{
+    do_select_into(return_value, this_obj, sql.c_str(), sql.size(),
+                   /*params=*/NULL, fetch_mode, /*qid=*/std::string(),
+                   /*settings=*/NULL, /*external_tables=*/NULL);
+}
+
+static bool runHelperExec(zval *this_obj, const std::string &sql)
+{
+    do_execute_into(this_obj, sql.c_str(), sql.size(),
+                    /*params=*/NULL, /*qid=*/std::string(), /*settings=*/NULL);
+    return !EG(exception);
+}
+
+static std::string currentDatabase(zval *this_obj)
+{
+    zval *db = sc_zend_read_property(clickhouse_ce, this_obj, "database", sizeof("database") - 1, 0);
+    if (db && Z_TYPE_P(db) == IS_STRING) {
+        return std::string(Z_STRVAL_P(db), Z_STRLEN_P(db));
+    }
+    return std::string("default");
+}
+
+/*
+ * Return the first row of a helper result as an assoc array, or
+ * an empty array if there were no rows.
+ */
+static void runHelperSelectFirstRow(zval *return_value, zval *this_obj, const std::string &sql)
+{
+    zval rows;
+    runHelperSelect(&rows, this_obj, sql, 0);
+    if (EG(exception)) {
+        return;
+    }
+    if (Z_TYPE(rows) == IS_ARRAY && zend_hash_num_elements(Z_ARRVAL(rows)) > 0) {
+        zval *first = NULL;
+        zval *fz;
+        ZEND_HASH_FOREACH_VAL(Z_ARRVAL(rows), fz) {
+            first = fz;
+            break;
+        } ZEND_HASH_FOREACH_END();
+        if (first && Z_TYPE_P(first) == IS_ARRAY) {
+            ZVAL_COPY(return_value, first);
+            zval_ptr_dtor(&rows);
+            return;
+        }
+    }
+    array_init(return_value);
+    zval_ptr_dtor(&rows);
+}
+
+/* {{{ proto array databaseSize(?string database)
+ */
+PHP_METHOD(ClickHouse, databaseSize)
+{
+    zend_string *db = NULL;
+    ZEND_PARSE_PARAMETERS_START(0, 1)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR_OR_NULL(db)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string dbname = (db && ZSTR_LEN(db) > 0) ? std::string(ZSTR_VAL(db), ZSTR_LEN(db)) : currentDatabase(getThis());
+    try {
+        validateMetadataFilterName(dbname, "database name");
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    std::string sql =
+        "SELECT sum(bytes_on_disk) AS bytes_on_disk, sum(rows) AS rows "
+        "FROM system.parts WHERE active AND database = " + sqlStringLiteral(dbname);
+    runHelperSelectFirstRow(return_value, getThis(), sql);
+}
+/* }}} */
+
+/* {{{ proto array tablesSize(?string database)
+ */
+PHP_METHOD(ClickHouse, tablesSize)
+{
+    zend_string *db = NULL;
+    ZEND_PARSE_PARAMETERS_START(0, 1)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR_OR_NULL(db)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string dbname = (db && ZSTR_LEN(db) > 0) ? std::string(ZSTR_VAL(db), ZSTR_LEN(db)) : currentDatabase(getThis());
+    try {
+        validateMetadataFilterName(dbname, "database name");
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    std::string sql =
+        "SELECT table, sum(bytes_on_disk) AS bytes_on_disk, sum(rows) AS rows, "
+        "max(modification_time) AS modification_time "
+        "FROM system.parts WHERE active AND database = " + sqlStringLiteral(dbname) + " "
+        "GROUP BY table ORDER BY table";
+    runHelperSelect(return_value, getThis(), sql, 0);
+}
+/* }}} */
+
+/* {{{ proto array partitions(string table)
+ */
+PHP_METHOD(ClickHouse, partitions)
+{
+    zend_string *table = NULL;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_STR(table)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string tname(ZSTR_VAL(table), ZSTR_LEN(table));
+    std::string dbname = currentDatabase(getThis());
+    /* Allow `db.table` in the argument; split on the dot. */
+    auto dot = tname.find('.');
+    if (dot != std::string::npos) {
+        dbname = tname.substr(0, dot);
+        tname = tname.substr(dot + 1);
+    }
+    try {
+        validateMetadataFilterName(dbname, "database name");
+        validateMetadataFilterName(tname, "table name");
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    std::string sql =
+        "SELECT partition, count() AS parts, sum(rows) AS rows, "
+        "sum(bytes_on_disk) AS bytes_on_disk, "
+        "min(min_time) AS min_time, max(max_time) AS max_time "
+        "FROM system.parts WHERE active AND database = " + sqlStringLiteral(dbname) + " "
+        "AND table = " + sqlStringLiteral(tname) + " "
+        "GROUP BY partition ORDER BY partition";
+    runHelperSelect(return_value, getThis(), sql, 0);
+}
+/* }}} */
+
+/* {{{ proto array showTables(?string database, ?string like)
+ */
+PHP_METHOD(ClickHouse, showTables)
+{
+    zend_string *db = NULL, *like = NULL;
+    ZEND_PARSE_PARAMETERS_START(0, 2)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_STR_OR_NULL(db)
+        Z_PARAM_STR_OR_NULL(like)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string dbname = (db && ZSTR_LEN(db) > 0) ? std::string(ZSTR_VAL(db), ZSTR_LEN(db)) : currentDatabase(getThis());
+    std::string sql = "SELECT name FROM system.tables WHERE database = " + sqlStringLiteral(dbname);
+    if (like && ZSTR_LEN(like) > 0) {
+        sql += " AND name LIKE " +
+            sqlStringLiteral(std::string(ZSTR_VAL(like), ZSTR_LEN(like)));
+    }
+    sql += " ORDER BY name";
+    runHelperSelect(return_value, getThis(), sql, SC_FETCH_COLUMN);
+}
+/* }}} */
+
+/* {{{ proto string showCreateTable(string table)
+ */
+PHP_METHOD(ClickHouse, showCreateTable)
+{
+    zend_string *table = NULL;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_STR(table)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string tname(ZSTR_VAL(table), ZSTR_LEN(table));
+    try {
+        validateIdentifier(tname.c_str(), tname.size(), "table name", true);
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    std::string sql = "SHOW CREATE TABLE " + tname;
+    runHelperSelect(return_value, getThis(), sql, SC_FETCH_ONE | SC_FETCH_COLUMN);
+}
+/* }}} */
+
+/* {{{ proto int getServerUptime()
+ */
+PHP_METHOD(ClickHouse, getServerUptime)
+{
+    if (zend_parse_parameters_none() == FAILURE) {
+        return;
+    }
+    runHelperSelect(return_value, getThis(),
+        "SELECT uptime() AS uptime",
+        SC_FETCH_ONE | SC_FETCH_COLUMN);
+}
+/* }}} */
+
+/* {{{ proto bool enableLogQueries(bool enabled = true)
+ *
+ * Toggle the query log accumulator. While enabled, each completed
+ * select / insert / execute / writeStart appends an entry. Toggling
+ * off does NOT clear; getLogQueries() returns and clears.
+ */
+PHP_METHOD(ClickHouse, enableLogQueries)
+{
+    zend_bool enabled = 1;
+    ZEND_PARSE_PARAMETERS_START(0, 1)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_BOOL(enabled)
+    ZEND_PARSE_PARAMETERS_END();
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    obj->log_enabled = (enabled != 0);
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* {{{ proto array getLogQueries()
+ *
+ * Return all accumulated query log entries and clear the buffer. Each
+ * entry is an associative array: sql, query_id, elapsed_ms, rows_read,
+ * bytes_read, error_code (0 = success, server code on server error,
+ * -1 on client/network error), error_message.
+ */
+PHP_METHOD(ClickHouse, getLogQueries)
+{
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    array_init(return_value);
+    for (const auto &ql : obj->query_log) {
+        zval entry;
+        array_init(&entry);
+        add_assoc_stringl(&entry, "sql", (char*)ql.sql.c_str(), ql.sql.size());
+        add_assoc_stringl(&entry, "query_id", (char*)ql.query_id.c_str(), ql.query_id.size());
+        add_assoc_double(&entry, "elapsed_ms", ql.elapsed_ms);
+        add_assoc_long(&entry, "rows_read", (zend_long)ql.rows_read);
+        add_assoc_long(&entry, "bytes_read", (zend_long)ql.bytes_read);
+        add_assoc_long(&entry, "error_code", (zend_long)ql.error_code);
+        add_assoc_stringl(&entry, "error_message",
+            (char*)ql.error_message.c_str(), ql.error_message.size());
+        add_next_index_zval(return_value, &entry);
+    }
+    obj->query_log.clear();
+}
+/* }}} */
+
+/* {{{ proto ClickHouseRowIterator selectStream(string sql, array params, string query_id, array settings)
+ *
+ * Run a SELECT and return a ClickHouseRowIterator over the rows
+ * without materializing the full result as a single PHP array. The
+ * implementation buffers all blocks before returning (so the network
+ * round-trip is finished by the time the iterator is handed back),
+ * then walks them lazily during iteration. Use this when the row
+ * shape is fine but the row count is large enough that a full PHP
+ * array would balloon zval overhead.
+ *
+ * For true unbounded streaming where rows must be consumed as they
+ * arrive, use selectStreamCallback() instead.
+ */
+PHP_METHOD(ClickHouse, selectStream)
+{
+    zend_string *sql = NULL;
+    zval *params = NULL;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(1, 4)
+        Z_PARAM_STR(sql)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_ARRAY(params)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string qid = makeQid(query_id);
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+
+    object_init_ex(return_value, clickhouse_iter_ce);
+    clickhouse_iter_object *iter = Z_CLICKHOUSE_ITER_P(return_value);
+    std::string log_sql(ZSTR_VAL(sql), ZSTR_LEN(sql));
+
+    try {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block) {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        std::string sql_s = log_sql;
+        std::vector<TypedParam> typed_params;
+        if (params != NULL && Z_TYPE_P(params) == IS_ARRAY) {
+            applyPlaceholders(sql_s, Z_ARRVAL_P(params), typed_params);
+            log_sql = sql_s;
+        }
+
+        Query query = qid.empty() ? Query(sql_s) : Query(sql_s, qid);
+        attachTypedParams(query, typed_params);
+        applyMergedSettings(query, obj, settings);
+        attachProgressAndProfile(query, obj);
+        attachVerbose(query, obj);
+
+        query.OnData([iter](const Block &block) {
+            if (block.GetRowCount() == 0 || block.GetColumnCount() == 0) return;
+            iter->total_rows += block.GetRowCount();
+            /* Cache column names on the first non-empty block. The schema
+             * is identical across all blocks in a single result, so we
+             * pay one std::string copy per column, once, instead of one
+             * per (row, column) on every current() call. */
+            if (iter->column_names.empty()) {
+                const size_t nc = block.GetColumnCount();
+                iter->column_names.reserve(nc);
+                for (size_t c = 0; c < nc; ++c) {
+                    iter->column_names.emplace_back(block.GetColumnName(c));
+                }
+            }
+            iter->blocks.push_back(block);
+        });
+
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            client->Select(query);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            tryResetConnectionReapplyDatabase(getThis(), obj, false);
+            throw;
+        }
+        auto t1 = std::chrono::steady_clock::now();
+        obj->stats.elapsed_ms =
+            std::chrono::duration<double, std::milli>(t1 - t0).count();
+        recordQuerySuccess(obj, sql_s, qid);
+    }
+    catch (const std::exception &e) {
+        recordQueryError(obj, log_sql, qid, e);
+        zval_ptr_dtor(return_value);
+        ZVAL_NULL(return_value);
+        throwClickHouseError(e, qid);
+    }
+}
+/* }}} */
+
+/* {{{ proto bool selectStreamCallback(string sql, callable cb, array params, string query_id, array settings)
+ *
+ * True per-row streaming: invoke the user callback once per row as
+ * blocks arrive from the server, never accumulating the full result
+ * in memory. The callback receives a single argument, the row as an
+ * associative array keyed by column name. Returns true on success.
+ */
+PHP_METHOD(ClickHouse, selectStreamCallback)
+{
+    zend_string *sql = NULL;
+    zval *cb = NULL;
+    zval *params = NULL;
+    zend_string *query_id = NULL;
+    zval *settings = NULL;
+
+    ZEND_PARSE_PARAMETERS_START(2, 5)
+        Z_PARAM_STR(sql)
+        Z_PARAM_ZVAL(cb)
+        Z_PARAM_OPTIONAL
+        Z_PARAM_ARRAY(params)
+        Z_PARAM_STR(query_id)
+        Z_PARAM_ARRAY(settings)
+    ZEND_PARSE_PARAMETERS_END();
+
+    if (!zend_is_callable(cb, 0, NULL)) {
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+            "Argument 2 passed to selectStreamCallback must be callable", 0);
+        return;
+    }
+
+    std::string qid = makeQid(query_id);
+    clickhouse_object *obj = Z_CLICKHOUSE_P(getThis());
+    std::string log_sql(ZSTR_VAL(sql), ZSTR_LEN(sql));
+
+    try {
+        Client *client = getClient(obj);
+        QueryActiveGuard guard(obj);
+
+        resetStats(obj);
+        obj->stats.last_query_id = qid;
+
+        if (obj->has_insert_block) {
+            throw std::runtime_error("The insert operation is now in progress");
+        }
+
+        std::string sql_s = log_sql;
+        std::vector<TypedParam> typed_params;
+        if (params != NULL && Z_TYPE_P(params) == IS_ARRAY) {
+            applyPlaceholders(sql_s, Z_ARRVAL_P(params), typed_params);
+            log_sql = sql_s;
+        }
+
+        Query query = qid.empty() ? Query(sql_s) : Query(sql_s, qid);
+        attachTypedParams(query, typed_params);
+        applyMergedSettings(query, obj, settings);
+        attachProgressAndProfile(query, obj);
+        attachVerbose(query, obj);
+
+        query.OnData([cb](const Block &block) {
+            if (block.GetRowCount() == 0 || block.GetColumnCount() == 0) return;
+            /* Hoist column names out of the row loop. clickhouse-cpp returns
+             * a fresh std::string per GetColumnName call but the names are
+             * stable across all rows of a block. */
+            const size_t col_count = block.GetColumnCount();
+            std::vector<std::string> col_names;
+            col_names.reserve(col_count);
+            for (size_t c = 0; c < col_count; ++c) {
+                col_names.emplace_back(block.GetColumnName(c));
+            }
+            for (size_t row = 0; row < block.GetRowCount(); ++row) {
+                zval row_zv;
+                array_init(&row_zv);
+                /* Mirrors do_select_into's exception-safety pattern: a
+                 * convertToZval throw mid-row would otherwise leak the
+                 * partially-built row_zv HashTable. */
+                try {
+                    for (size_t col = 0; col < col_count; ++col) {
+                        convertToZval(&row_zv, block[col], row, col_names[col], 0, 0);
+                    }
+                } catch (...) {
+                    zval_ptr_dtor(&row_zv);
+                    throw;
+                }
+                zval args[1], retval;
+                ZVAL_NULL(&retval);
+                ZVAL_COPY_VALUE(&args[0], &row_zv);
+                call_user_function(NULL, NULL, cb, &retval, 1, args);
+                zval_ptr_dtor(&args[0]);
+                zval_ptr_dtor(&retval);
+                /* Match the progress / profile / verbose pattern: a throwing
+                 * row callback aborts the packet loop instead of silently
+                 * over-processing remaining rows and recording the query
+                 * as successful. The user's PHP exception is preserved in
+                 * EG(exception); the surrounding catch translates the C++
+                 * throw to recordQueryError + throwClickHouseError. */
+                if (EG(exception)) {
+                    throw std::runtime_error("row callback aborted query");
+                }
+            }
+        });
+
+        auto t0 = std::chrono::steady_clock::now();
+        try {
+            client->Select(query);
+        } catch (...) {
+            setElapsedSince(obj, t0);
+            tryResetConnectionReapplyDatabase(getThis(), obj, false);
+            throw;
+        }
+        auto t1 = std::chrono::steady_clock::now();
+        obj->stats.elapsed_ms =
+            std::chrono::duration<double, std::milli>(t1 - t0).count();
+        recordQuerySuccess(obj, sql_s, qid);
+    }
+    catch (const std::exception &e) {
+        recordQueryError(obj, log_sql, qid, e);
+        throwClickHouseError(e, qid);
+    }
+    RETURN_TRUE;
+}
+/* }}} */
+
+/* ClickHouseRowIterator Iterator interface ------------------------- */
+
+PHP_METHOD(ClickHouseRowIterator, rewind)
+{
+    clickhouse_iter_object *iter = Z_CLICKHOUSE_ITER_P(getThis());
+    iter->block_idx = 0;
+    iter->row_idx = 0;
+    iter->cumulative_row_idx = 0;
+}
+
+PHP_METHOD(ClickHouseRowIterator, valid)
+{
+    clickhouse_iter_object *iter = Z_CLICKHOUSE_ITER_P(getThis());
+    if (iter->block_idx >= iter->blocks.size()) {
+        RETURN_FALSE;
+    }
+    if (iter->row_idx >= iter->blocks[iter->block_idx].GetRowCount()) {
+        RETURN_FALSE;
+    }
+    RETURN_TRUE;
+}
+
+PHP_METHOD(ClickHouseRowIterator, current)
+{
+    clickhouse_iter_object *iter = Z_CLICKHOUSE_ITER_P(getThis());
+    if (iter->block_idx >= iter->blocks.size()) {
+        RETURN_NULL();
+    }
+    const Block &block = iter->blocks[iter->block_idx];
+    if (iter->row_idx >= block.GetRowCount()) {
+        RETURN_NULL();
+    }
+    array_init(return_value);
+    /* convertToZval throws on unsupported / malformed server-side types.
+     * The Zend dispatcher is C; let the exception cross it would be UB. */
+    static const std::string empty_name;
+    try {
+        const size_t col_count = block.GetColumnCount();
+        for (size_t col = 0; col < col_count; ++col) {
+            /* Use the iter-cached names rather than block.GetColumnName(col)
+             * which would heap-alloc a fresh std::string per cell. */
+            const std::string &name = (col < iter->column_names.size())
+                ? iter->column_names[col]
+                : empty_name;
+            convertToZval(return_value, block[col], iter->row_idx, name, 0, 0);
+        }
+    } catch (const std::exception &e) {
+        zval_ptr_dtor(return_value);
+        sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce, e.what(), 0);
+        RETURN_NULL();
+    }
+}
+
+PHP_METHOD(ClickHouseRowIterator, key)
+{
+    clickhouse_iter_object *iter = Z_CLICKHOUSE_ITER_P(getThis());
+    RETURN_LONG((zend_long)iter->cumulative_row_idx);
+}
+
+PHP_METHOD(ClickHouseRowIterator, next)
+{
+    clickhouse_iter_object *iter = Z_CLICKHOUSE_ITER_P(getThis());
+    if (iter->block_idx >= iter->blocks.size()) return;
+    iter->row_idx++;
+    iter->cumulative_row_idx++;
+    while (iter->block_idx < iter->blocks.size() &&
+           iter->row_idx >= iter->blocks[iter->block_idx].GetRowCount()) {
+        iter->block_idx++;
+        iter->row_idx = 0;
+    }
+}
+
+PHP_METHOD(ClickHouseRowIterator, count)
+{
+    clickhouse_iter_object *iter = Z_CLICKHOUSE_ITER_P(getThis());
+    RETURN_LONG((zend_long)iter->total_rows);
+}
+
+/* {{{ proto bool isExists(string database, string table)
+ *
+ * Returns true when the (database, table) pair exists in
+ * system.tables (which also covers views and dictionaries on modern
+ * ClickHouse). Both arguments are validated as identifiers.
+ */
+PHP_METHOD(ClickHouse, isExists)
+{
+    zend_string *db = NULL, *table = NULL;
+    ZEND_PARSE_PARAMETERS_START(2, 2)
+        Z_PARAM_STR(db)
+        Z_PARAM_STR(table)
+    ZEND_PARSE_PARAMETERS_END();
+    try {
+        validateIdentifier(ZSTR_VAL(db), ZSTR_LEN(db), "database name", false);
+        validateIdentifier(ZSTR_VAL(table), ZSTR_LEN(table), "table name", false);
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    std::string sql =
+        "SELECT count() AS c FROM system.tables WHERE database = '" +
+        std::string(ZSTR_VAL(db), ZSTR_LEN(db)) + "' AND name = '" + std::string(ZSTR_VAL(table), ZSTR_LEN(table)) + "'";
+    zval row;
+    runHelperSelectFirstRow(&row, getThis(), sql);
+    if (EG(exception)) return;
+    bool exists = false;
+    if (Z_TYPE(row) == IS_ARRAY) {
+        zval *cnt = zend_hash_str_find(Z_ARRVAL(row), "c", sizeof("c") - 1);
+        if (cnt) {
+            exists = (zval_get_long(cnt) > 0);
+        }
+    }
+    zval_ptr_dtor(&row);
+    RETURN_BOOL(exists);
+}
+/* }}} */
+
+/* {{{ proto array showDatabases()
+ */
+PHP_METHOD(ClickHouse, showDatabases)
+{
+    if (zend_parse_parameters_none() == FAILURE) {
+        return;
+    }
+    runHelperSelect(return_value, getThis(),
+        "SELECT name FROM system.databases ORDER BY name", SC_FETCH_COLUMN);
+}
+/* }}} */
+
+/* {{{ proto array showProcesslist()
+ *
+ * Projects a fixed set of common columns from system.processes
+ * instead of `SELECT *`, because the wider table includes
+ * Map(LowCardinality(String), ...) columns (ProfileEvents, Settings,
+ * used_*) that our Map read path doesn't yet decode.
+ */
+PHP_METHOD(ClickHouse, showProcesslist)
+{
+    if (zend_parse_parameters_none() == FAILURE) {
+        return;
+    }
+    runHelperSelect(return_value, getThis(),
+        "SELECT query_id, user, address, port, initial_user, initial_query_id, "
+        "initial_address, interface, os_user, client_hostname, client_name, "
+        "client_revision, client_version_major, client_version_minor, "
+        "client_version_patch, http_method, http_user_agent, http_referer, "
+        "forwarded_for, query, elapsed, read_rows, read_bytes, total_rows_approx, "
+        "memory_usage, peak_memory_usage "
+        "FROM system.processes", 0);
+}
+/* }}} */
+
+/* {{{ proto string getServerVersion()
+ */
+PHP_METHOD(ClickHouse, getServerVersion)
+{
+    if (zend_parse_parameters_none() == FAILURE) {
+        return;
+    }
+    zval row;
+    runHelperSelectFirstRow(&row, getThis(), "SELECT version() AS v");
+    if (EG(exception)) return;
+    if (Z_TYPE(row) == IS_ARRAY) {
+        zval *v = zend_hash_str_find(Z_ARRVAL(row), "v", sizeof("v") - 1);
+        if (v && Z_TYPE_P(v) == IS_STRING) {
+            ZVAL_STR_COPY(return_value, Z_STR_P(v));
+            zval_ptr_dtor(&row);
+            return;
+        }
+    }
+    zval_ptr_dtor(&row);
+    RETURN_EMPTY_STRING();
+}
+/* }}} */
+
+/* {{{ proto array tableSize(string table)
+ *
+ * Aggregate row/byte/partition count from system.parts for a single
+ * table. Accepts `db.table` form. Returns the assoc row or an empty
+ * array when the table has no active parts.
+ */
+PHP_METHOD(ClickHouse, tableSize)
+{
+    zend_string *table = NULL;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_STR(table)
+    ZEND_PARSE_PARAMETERS_END();
+    std::string tname(ZSTR_VAL(table), ZSTR_LEN(table));
+    std::string dbname = currentDatabase(getThis());
+    auto dot = tname.find('.');
+    if (dot != std::string::npos) {
+        dbname = tname.substr(0, dot);
+        tname = tname.substr(dot + 1);
+    }
+    try {
+        validateMetadataFilterName(dbname, "database name");
+        validateMetadataFilterName(tname, "table name");
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    std::string sql =
+        "SELECT sum(rows) AS rows, sum(bytes_on_disk) AS bytes_on_disk, "
+        "uniqExact(partition) AS partitions, max(modification_time) AS modification_time "
+        "FROM system.parts WHERE active AND database = " + sqlStringLiteral(dbname) +
+        " AND table = " + sqlStringLiteral(tname) +
+        " GROUP BY database, table";
+    runHelperSelectFirstRow(return_value, getThis(), sql);
+}
+/* }}} */
+
+/* {{{ proto bool truncateTable(string table)
+ */
+PHP_METHOD(ClickHouse, truncateTable)
+{
+    zend_string *table = NULL;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_STR(table)
+    ZEND_PARSE_PARAMETERS_END();
+    try {
+        validateIdentifier(ZSTR_VAL(table), ZSTR_LEN(table), "table name", true);
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    std::string sql = "TRUNCATE TABLE " + std::string(ZSTR_VAL(table), ZSTR_LEN(table));
+    RETURN_BOOL(runHelperExec(getThis(), sql));
+}
+/* }}} */
+
+/* {{{ proto bool dropPartition(string table, string partition)
+ *
+ * Drop a partition by string value. The partition argument is always
+ * single-quote-escaped and emitted as a SQL string literal, so dates
+ * ('2024-01-01') and named partitions are safe by default. For
+ * integer partitions or partition IDs, fall back to execute() with a
+ * hand-built ALTER TABLE statement.
+ */
+PHP_METHOD(ClickHouse, dropPartition)
+{
+    zend_string *table = NULL, *part = NULL;
+    ZEND_PARSE_PARAMETERS_START(2, 2)
+        Z_PARAM_STR(table)
+        Z_PARAM_STR(part)
+    ZEND_PARSE_PARAMETERS_END();
+    try {
+        validateIdentifier(ZSTR_VAL(table), ZSTR_LEN(table), "table name", true);
+    } catch (const std::exception &e) {
+        throwClickHouseError(e);
+        return;
+    }
+    /* Guard against control characters that could break the literal. */
+    for (size_t i = 0; i < ZSTR_LEN(part); ++i) {
+        unsigned char c = (unsigned char)ZSTR_VAL(part)[i];
+        if (c < 0x20) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "dropPartition: partition value contains a control character", 0);
+            return;
+        }
+    }
+    /* ClickHouse accepts both SQL-standard '' and C-style \' as escapes inside
+     * single-quoted literals; backslash must be escaped too or '\'' would be
+     * parsed as an escaped quote followed by a closing quote. */
+    std::string escaped;
+    escaped.reserve(ZSTR_LEN(part) + 8);
+    for (size_t i = 0; i < ZSTR_LEN(part); ++i) {
+        char c = ZSTR_VAL(part)[i];
+        if (c == '\'')      escaped += "\\'";
+        else if (c == '\\') escaped += "\\\\";
+        else                escaped += c;
+    }
+    std::string sql = "ALTER TABLE " + std::string(ZSTR_VAL(table), ZSTR_LEN(table)) +
+        " DROP PARTITION '" + escaped + "'";
+    RETURN_BOOL(runHelperExec(getThis(), sql));
+}
+/* }}} */
+
+/* {{{ proto void __destruct()
+ *
+ * No-op. All cleanup (Client teardown, in-progress insert close,
+ * progress callback release, settings/log destruction) lives in the
+ * free_obj handler so it fires even on bailout.
+ */
+PHP_METHOD(ClickHouse, __destruct)
+{
+}
+/* }}} */
+
+/* {{{ proto int ClickHouseException::getServerCode()
+ * proto ?string ClickHouseException::getServerName()
+ * proto ?string ClickHouseException::getQueryId()
+ *
+ * smi2/phpClickHouse-style getter aliases for the public
+ * `server_code`/`server_name`/`query_id` properties. Same data, easier
+ * to call from code that was written against the smi2 client.
+ */
+PHP_METHOD(ClickHouseException, getServerCode)
+{
+    if (zend_parse_parameters_none() == FAILURE) {
+        return;
+    }
+    zval *p = sc_zend_read_property(clickhouse_exception_ce, getThis(), "server_code", sizeof("server_code") - 1, 0);
+    if (p && Z_TYPE_P(p) == IS_LONG) {
+        RETURN_LONG(Z_LVAL_P(p));
+    }
+    RETURN_LONG(0);
+}
+
+PHP_METHOD(ClickHouseException, getServerName)
+{
+    if (zend_parse_parameters_none() == FAILURE) {
+        return;
+    }
+    zval *p = sc_zend_read_property(clickhouse_exception_ce, getThis(), "server_name", sizeof("server_name") - 1, 0);
+    if (p && Z_TYPE_P(p) == IS_STRING) {
+        RETURN_STRINGL(Z_STRVAL_P(p), Z_STRLEN_P(p));
+    }
+    RETURN_NULL();
+}
+
+PHP_METHOD(ClickHouseException, getQueryId)
+{
+    if (zend_parse_parameters_none() == FAILURE) {
+        return;
+    }
+    zval *p = sc_zend_read_property(clickhouse_exception_ce, getThis(), "query_id", sizeof("query_id") - 1, 0);
+    if (p && Z_TYPE_P(p) == IS_STRING) {
+        RETURN_STRINGL(Z_STRVAL_P(p), Z_STRLEN_P(p));
+    }
+    RETURN_NULL();
+}
+/* }}} */
+
+/* {{{ ClickHouseStatement methods
+ *
+ * Materialized result wrapper. Constructed only by
+ * ClickHouse::selectStatement(); the public constructor throws.
+ */
+PHP_METHOD(ClickHouseStatement, __construct)
+{
+    sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+        "ClickHouseStatement is constructed by ClickHouse::selectStatement(); the constructor is private",
+        0);
+}
+
+PHP_METHOD(ClickHouseStatement, count)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        RETURN_LONG(0);
+    }
+    RETURN_LONG((zend_long)zend_hash_num_elements(Z_ARRVAL(stmt->rows)));
+}
+
+PHP_METHOD(ClickHouseStatement, rewind)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) == IS_ARRAY) {
+        zend_hash_internal_pointer_reset(Z_ARRVAL(stmt->rows));
+    }
+}
+
+PHP_METHOD(ClickHouseStatement, valid)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        RETURN_FALSE;
+    }
+    RETURN_BOOL(zend_hash_get_current_data(Z_ARRVAL(stmt->rows)) != NULL);
+}
+
+PHP_METHOD(ClickHouseStatement, current)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        RETURN_NULL();
+    }
+    zval *cur = zend_hash_get_current_data(Z_ARRVAL(stmt->rows));
+    if (!cur) {
+        RETURN_NULL();
+    }
+    ZVAL_COPY(return_value, cur);
+}
+
+PHP_METHOD(ClickHouseStatement, key)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        RETURN_NULL();
+    }
+    zend_string *str_key;
+    zend_ulong num_key;
+    int t = zend_hash_get_current_key(Z_ARRVAL(stmt->rows), &str_key, &num_key);
+    if (t == HASH_KEY_IS_STRING) {
+        RETURN_STR_COPY(str_key);
+    } else if (t == HASH_KEY_IS_LONG) {
+        RETURN_LONG((zend_long)num_key);
+    }
+    RETURN_NULL();
+}
+
+PHP_METHOD(ClickHouseStatement, next)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) == IS_ARRAY) {
+        zend_hash_move_forward(Z_ARRVAL(stmt->rows));
+    }
+}
+
+PHP_METHOD(ClickHouseStatement, offsetExists)
+{
+    zval *offset;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ZVAL(offset)
+    ZEND_PARSE_PARAMETERS_END();
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        RETURN_FALSE;
+    }
+    HashTable *ht = Z_ARRVAL(stmt->rows);
+    if (Z_TYPE_P(offset) == IS_LONG) {
+        RETURN_BOOL(zend_hash_index_exists(ht, Z_LVAL_P(offset)));
+    }
+    if (Z_TYPE_P(offset) == IS_STRING) {
+        RETURN_BOOL(zend_symtable_exists(ht, Z_STR_P(offset)));
+    }
+    RETURN_FALSE;
+}
+
+PHP_METHOD(ClickHouseStatement, offsetGet)
+{
+    zval *offset;
+    ZEND_PARSE_PARAMETERS_START(1, 1)
+        Z_PARAM_ZVAL(offset)
+    ZEND_PARSE_PARAMETERS_END();
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        RETURN_NULL();
+    }
+    HashTable *ht = Z_ARRVAL(stmt->rows);
+    zval *v = NULL;
+    if (Z_TYPE_P(offset) == IS_LONG) {
+        v = zend_hash_index_find(ht, Z_LVAL_P(offset));
+    } else if (Z_TYPE_P(offset) == IS_STRING) {
+        v = zend_symtable_find(ht, Z_STR_P(offset));
+    }
+    if (!v) {
+        RETURN_NULL();
+    }
+    ZVAL_COPY(return_value, v);
+}
+
+PHP_METHOD(ClickHouseStatement, offsetSet)
+{
+    sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+        "ClickHouseStatement is read-only; offsetSet is not supported", 0);
+}
+
+PHP_METHOD(ClickHouseStatement, offsetUnset)
+{
+    sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+        "ClickHouseStatement is read-only; offsetUnset is not supported", 0);
+}
+
+/* jsonSerialize and toArray return the same value: a copy of the
+ * stored rows array, or an empty array when the statement has none.
+ * Shared body keeps the two PHP-visible methods in lockstep. */
+static void statement_emit_rows(zval *return_value, zval *this_obj)
+{
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(this_obj);
+    if (Z_TYPE(stmt->rows) == IS_ARRAY) {
+        ZVAL_COPY(return_value, &stmt->rows);
+    } else {
+        array_init(return_value);
+    }
+}
+
+PHP_METHOD(ClickHouseStatement, jsonSerialize)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    statement_emit_rows(return_value, getThis());
+}
+
+PHP_METHOD(ClickHouseStatement, toArray)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    statement_emit_rows(return_value, getThis());
+}
+
+PHP_METHOD(ClickHouseStatement, statistics)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->statistics) == IS_ARRAY) {
+        ZVAL_COPY(return_value, &stmt->statistics);
+    } else {
+        array_init(return_value);
+    }
+}
+
+PHP_METHOD(ClickHouseStatement, fetchOne)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    if (Z_TYPE(stmt->rows) != IS_ARRAY || zend_hash_num_elements(Z_ARRVAL(stmt->rows)) == 0) {
+        RETURN_NULL();
+    }
+    HashPosition pos;
+    zend_hash_internal_pointer_reset_ex(Z_ARRVAL(stmt->rows), &pos);
+    zval *first = zend_hash_get_current_data_ex(Z_ARRVAL(stmt->rows), &pos);
+    if (!first) {
+        RETURN_NULL();
+    }
+    /* If the row is itself an assoc array with a single column, return
+     * the scalar value (smi2 fetchOne semantics). Otherwise return the
+     * full row. */
+    if (Z_TYPE_P(first) == IS_ARRAY && zend_hash_num_elements(Z_ARRVAL_P(first)) == 1) {
+        zend_hash_internal_pointer_reset_ex(Z_ARRVAL_P(first), &pos);
+        zval *only = zend_hash_get_current_data_ex(Z_ARRVAL_P(first), &pos);
+        if (only) {
+            ZVAL_COPY(return_value, only);
+            return;
+        }
+    }
+    ZVAL_COPY(return_value, first);
+}
+
+PHP_METHOD(ClickHouseStatement, fetchKeyPair)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    array_init(return_value);
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        return;
+    }
+    zval *source_rows = (Z_TYPE(stmt->positional_rows) == IS_ARRAY)
+        ? &stmt->positional_rows
+        : &stmt->rows;
+    zval *row;
+    ZEND_HASH_FOREACH_VAL(Z_ARRVAL_P(source_rows), row) {
+        if (Z_TYPE_P(row) != IS_ARRAY || zend_hash_num_elements(Z_ARRVAL_P(row)) < 2) {
+            sc_zend_throw_exception_tsrmls_cc(clickhouse_exception_ce,
+                "fetchKeyPair requires each row to have at least 2 columns", 0);
+            zend_array_destroy(Z_ARR_P(return_value));
+            ZVAL_UNDEF(return_value);
+            return;
+        }
+        HashPosition pos;
+        zend_hash_internal_pointer_reset_ex(Z_ARRVAL_P(row), &pos);
+        zval *kv = zend_hash_get_current_data_ex(Z_ARRVAL_P(row), &pos);
+        zend_hash_move_forward_ex(Z_ARRVAL_P(row), &pos);
+        zval *vv = zend_hash_get_current_data_ex(Z_ARRVAL_P(row), &pos);
+        if (!kv || !vv) continue;
+
+        zval val_copy;
+        ZVAL_COPY(&val_copy, vv);
+        if (Z_TYPE_P(kv) == IS_LONG) {
+            zend_hash_index_update(Z_ARRVAL_P(return_value), Z_LVAL_P(kv), &val_copy);
+        } else {
+            /* zval_get_string is non-mutating and handles object __toString
+             * without leaking the pre-bumped val_copy if conversion raises.
+             * The prior pattern (ZVAL_COPY + convert_to_string) buffered
+             * EG(exception) and continued the loop, accumulating refcount
+             * leaks across rows. */
+            zend_string *coerced = zval_get_string(kv);
+            if (EG(exception)) {
+                zval_ptr_dtor(&val_copy);
+                if (coerced) zend_string_release(coerced);
+                return;
+            }
+            zend_symtable_update(Z_ARRVAL_P(return_value), coerced, &val_copy);
+            zend_string_release(coerced);
+        }
+    } ZEND_HASH_FOREACH_END();
+}
+
+PHP_METHOD(ClickHouseStatement, fetchColumn)
+{
+    if (zend_parse_parameters_none() == FAILURE) return;
+    clickhouse_statement_object *stmt = Z_CLICKHOUSE_STATEMENT_P(getThis());
+    array_init(return_value);
+    if (Z_TYPE(stmt->rows) != IS_ARRAY) {
+        return;
+    }
+    zval *source_rows = (Z_TYPE(stmt->positional_rows) == IS_ARRAY)
+        ? &stmt->positional_rows
+        : &stmt->rows;
+    zval *row;
+    ZEND_HASH_FOREACH_VAL(Z_ARRVAL_P(source_rows), row) {
+        if (Z_TYPE_P(row) != IS_ARRAY) {
+            /* Already a flat list of scalars (e.g. FETCH_COLUMN result fed in). */
+            zval c;
+            ZVAL_COPY(&c, row);
+            add_next_index_zval(return_value, &c);
+            continue;
+        }
+        HashPosition pos;
+        zend_hash_internal_pointer_reset_ex(Z_ARRVAL_P(row), &pos);
+        zval *first = zend_hash_get_current_data_ex(Z_ARRVAL_P(row), &pos);
+        if (first) {
+            zval c;
+            ZVAL_COPY(&c, first);
+            add_next_index_zval(return_value, &c);
+        }
+    } ZEND_HASH_FOREACH_END();
+}
+/* }}} */
+
+/*
+ * Local variables:
+ * tab-width: 4
+ * c-basic-offset: 4
+ * End:
+ * vim600: noet sw=4 ts=4 fdm=marker
+ * vim<600: noet sw=4 ts=4
+ */
diff --git a/clickhouse.stub.php b/clickhouse.stub.php
new file mode 100644
index 0000000..faf68b6
--- /dev/null
+++ b/clickhouse.stub.php
@@ -0,0 +1,238 @@
+<?php
+
+/** @generate-class-entries */
+
+final class ClickHouse
+{
+    public const int FETCH_ONE = 1;
+    public const int FETCH_KEY_PAIR = 2;
+    public const int DATE_AS_STRINGS = 4;
+    public const int FETCH_COLUMN = 8;
+
+    protected string $host = "127.0.0.1";
+    protected int $port = 9000;
+    protected string $database = "default";
+    protected ?string $user = null;
+    // No `passwd` property is declared. The secret stays out of
+    // get_object_vars, var_dump, serialize, and reflection by simply
+    // not being stored on the object.
+    // 0=none, 1=lz4, 2=zstd. Was `bool` but that coerced 2 → true → 1
+    // on read-back, silently downgrading "zstd" callers to LZ4.
+    protected int $compression = 0;
+    protected int $retry_timeout = 5;
+    protected int $retry_count = 1;
+    protected int $receive_timeout = 0;
+    protected int $connect_timeout = 5;
+
+    public function __construct(array $connectParams) {}
+
+    public function __destruct() {}
+
+    public function select(
+        string $sql,
+        array $params = [],
+        int $fetch_mode = 0,
+        string $query_id = "",
+        array $settings = []
+    ): mixed {}
+
+    public function selectWithExternalData(
+        string $sql,
+        array $externals,
+        array $params = [],
+        int $fetch_mode = 0,
+        string $query_id = "",
+        array $settings = []
+    ): mixed {}
+
+    public function selectToStream(
+        string $sql,
+        array $params,
+        mixed $stream,
+        string $format = "TabSeparated",
+        string $query_id = "",
+        array $settings = []
+    ): int {}
+
+    public function insert(
+        string $table,
+        array $columns,
+        array $values,
+        string $query_id = "",
+        array $settings = []
+    ): bool {}
+
+    public function insertAssoc(
+        string $table,
+        array $rows,
+        string $query_id = "",
+        array $settings = []
+    ): bool {}
+
+    public function insertFromStream(
+        string $table,
+        array $columns,
+        mixed $stream,
+        string $format = "TabSeparated",
+        int $batch_rows = 10000,
+        string $query_id = "",
+        array $settings = []
+    ): int {}
+
+    public function writeStart(
+        string $table,
+        array $columns,
+        string $query_id = "",
+        array $settings = []
+    ): bool {}
+
+    public function write(array $values): bool {}
+
+    public function writeEnd(): bool {}
+
+    public function execute(
+        string $sql,
+        array $params = [],
+        string $query_id = "",
+        array $settings = []
+    ): bool {}
+
+    public function ping(): bool {}
+
+    public function setSettings(array $settings): static {}
+
+    public function setSetting(string $key, mixed $value): static {}
+
+    public function setDatabase(string $database): static {}
+
+    public function setProgressCallback(?callable $callback): bool {}
+
+    public function setProfileCallback(?callable $callback): bool {}
+
+    public function setVerbose(mixed $sink): static {}
+
+    public function resetConnection(): bool {}
+
+    public function getServerInfo(): array {}
+
+    public function getCurrentEndpoint(): ?array {}
+
+    public function getStatistics(): array {}
+
+    public function databaseSize(?string $database = null): array {}
+
+    public function tablesSize(?string $database = null): array {}
+
+    public function partitions(string $table): array {}
+
+    public function showTables(?string $database = null, ?string $like = null): array {}
+
+    public function showCreateTable(string $table): string {}
+
+    public function getServerUptime(): int {}
+
+    public function enableLogQueries(bool $enabled = true): bool {}
+
+    public function getLogQueries(): array {}
+
+    public function selectStream(
+        string $sql,
+        array $params = [],
+        string $query_id = "",
+        array $settings = []
+    ): ClickHouseRowIterator {}
+
+    public function selectStatement(
+        string $sql,
+        array $params = [],
+        string $query_id = "",
+        array $settings = []
+    ): ClickHouseStatement {}
+
+    public function selectStreamCallback(
+        string $sql,
+        callable $callback,
+        array $params = [],
+        string $query_id = "",
+        array $settings = []
+    ): bool {}
+
+    public function isExists(string $database, string $table): bool {}
+
+    public function showDatabases(): array {}
+
+    public function showProcesslist(): array {}
+
+    public function getServerVersion(): string {}
+
+    public function tableSize(string $table): array {}
+
+    public function truncateTable(string $table): bool {}
+
+    public function dropPartition(string $table, string $partition): bool {}
+}
+
+final class ClickHouseRowIterator implements Iterator, Countable
+{
+    public function rewind(): void {}
+
+    public function valid(): bool {}
+
+    public function current(): ?array {}
+
+    public function key(): int {}
+
+    public function next(): void {}
+
+    public function count(): int {}
+}
+
+final class ClickHouseStatement implements Iterator, Countable, ArrayAccess, JsonSerializable
+{
+    private function __construct() {}
+
+    public function count(): int {}
+
+    public function rewind(): void {}
+
+    public function valid(): bool {}
+
+    public function current(): mixed {}
+
+    public function key(): mixed {}
+
+    public function next(): void {}
+
+    public function offsetExists(mixed $offset): bool {}
+
+    public function offsetGet(mixed $offset): mixed {}
+
+    public function offsetSet(mixed $offset, mixed $value): void {}
+
+    public function offsetUnset(mixed $offset): void {}
+
+    public function jsonSerialize(): array {}
+
+    public function toArray(): array {}
+
+    public function statistics(): array {}
+
+    public function fetchOne(): mixed {}
+
+    public function fetchKeyPair(): array {}
+
+    public function fetchColumn(): array {}
+}
+
+class ClickHouseException extends Exception
+{
+    public int $server_code = 0;
+    public ?string $server_name = null;
+    public ?string $query_id = null;
+
+    public function getServerCode(): int {}
+
+    public function getServerName(): ?string {}
+
+    public function getQueryId(): ?string {}
+}
diff --git a/clickhouse_arginfo.h b/clickhouse_arginfo.h
new file mode 100644
index 0000000..9762cee
--- /dev/null
+++ b/clickhouse_arginfo.h
@@ -0,0 +1,544 @@
+/* This is a generated file, edit the .stub.php file instead.
+ * Stub hash: 1abb4cb3aed42538a5407daf3cd854c5631e8c08 */
+
+ZEND_BEGIN_ARG_INFO_EX(arginfo_class_ClickHouse___construct, 0, 0, 1)
+	ZEND_ARG_TYPE_INFO(0, connectParams, IS_ARRAY, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_INFO_EX(arginfo_class_ClickHouse___destruct, 0, 0, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_select, 0, 1, IS_MIXED, 0)
+	ZEND_ARG_TYPE_INFO(0, sql, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, params, IS_ARRAY, 0, "[]")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, fetch_mode, IS_LONG, 0, "0")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_selectWithExternalData, 0, 2, IS_MIXED, 0)
+	ZEND_ARG_TYPE_INFO(0, sql, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, externals, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, params, IS_ARRAY, 0, "[]")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, fetch_mode, IS_LONG, 0, "0")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_selectToStream, 0, 3, IS_LONG, 0)
+	ZEND_ARG_TYPE_INFO(0, sql, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, params, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO(0, stream, IS_MIXED, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, format, IS_STRING, 0, "\"TabSeparated\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_insert, 0, 3, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, columns, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO(0, values, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_insertAssoc, 0, 2, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, rows, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_insertFromStream, 0, 3, IS_LONG, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, columns, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO(0, stream, IS_MIXED, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, format, IS_STRING, 0, "\"TabSeparated\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, batch_rows, IS_LONG, 0, "10000")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_writeStart, 0, 2, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, columns, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_write, 0, 1, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, values, IS_ARRAY, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_writeEnd, 0, 0, _IS_BOOL, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_execute, 0, 1, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, sql, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, params, IS_ARRAY, 0, "[]")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_ping arginfo_class_ClickHouse_writeEnd
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_setSettings, 0, 1, IS_STATIC, 0)
+	ZEND_ARG_TYPE_INFO(0, settings, IS_ARRAY, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_setSetting, 0, 2, IS_STATIC, 0)
+	ZEND_ARG_TYPE_INFO(0, key, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, value, IS_MIXED, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_setDatabase, 0, 1, IS_STATIC, 0)
+	ZEND_ARG_TYPE_INFO(0, database, IS_STRING, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_setProgressCallback, 0, 1, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, callback, IS_CALLABLE, 1)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_setProfileCallback arginfo_class_ClickHouse_setProgressCallback
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_setVerbose, 0, 1, IS_STATIC, 0)
+	ZEND_ARG_TYPE_INFO(0, sink, IS_MIXED, 0)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_resetConnection arginfo_class_ClickHouse_writeEnd
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_getServerInfo, 0, 0, IS_ARRAY, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_getCurrentEndpoint, 0, 0, IS_ARRAY, 1)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_getStatistics arginfo_class_ClickHouse_getServerInfo
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_databaseSize, 0, 0, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, database, IS_STRING, 1, "null")
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_tablesSize arginfo_class_ClickHouse_databaseSize
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_partitions, 0, 1, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_showTables, 0, 0, IS_ARRAY, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, database, IS_STRING, 1, "null")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, like, IS_STRING, 1, "null")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_showCreateTable, 0, 1, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_getServerUptime, 0, 0, IS_LONG, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_enableLogQueries, 0, 0, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, enabled, _IS_BOOL, 0, "true")
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_getLogQueries arginfo_class_ClickHouse_getServerInfo
+
+ZEND_BEGIN_ARG_WITH_RETURN_OBJ_INFO_EX(arginfo_class_ClickHouse_selectStream, 0, 1, ClickHouseRowIterator, 0)
+	ZEND_ARG_TYPE_INFO(0, sql, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, params, IS_ARRAY, 0, "[]")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_OBJ_INFO_EX(arginfo_class_ClickHouse_selectStatement, 0, 1, ClickHouseStatement, 0)
+	ZEND_ARG_TYPE_INFO(0, sql, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, params, IS_ARRAY, 0, "[]")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_selectStreamCallback, 0, 2, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, sql, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, callback, IS_CALLABLE, 0)
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, params, IS_ARRAY, 0, "[]")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, query_id, IS_STRING, 0, "\"\"")
+	ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(0, settings, IS_ARRAY, 0, "[]")
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_isExists, 0, 2, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, database, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_showDatabases arginfo_class_ClickHouse_getServerInfo
+
+#define arginfo_class_ClickHouse_showProcesslist arginfo_class_ClickHouse_getServerInfo
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_getServerVersion, 0, 0, IS_STRING, 0)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouse_tableSize arginfo_class_ClickHouse_partitions
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_truncateTable, 0, 1, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouse_dropPartition, 0, 2, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, table, IS_STRING, 0)
+	ZEND_ARG_TYPE_INFO(0, partition, IS_STRING, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouseRowIterator_rewind, 0, 0, IS_VOID, 0)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouseRowIterator_valid arginfo_class_ClickHouse_writeEnd
+
+#define arginfo_class_ClickHouseRowIterator_current arginfo_class_ClickHouse_getCurrentEndpoint
+
+#define arginfo_class_ClickHouseRowIterator_key arginfo_class_ClickHouse_getServerUptime
+
+#define arginfo_class_ClickHouseRowIterator_next arginfo_class_ClickHouseRowIterator_rewind
+
+#define arginfo_class_ClickHouseRowIterator_count arginfo_class_ClickHouse_getServerUptime
+
+#define arginfo_class_ClickHouseStatement___construct arginfo_class_ClickHouse___destruct
+
+#define arginfo_class_ClickHouseStatement_count arginfo_class_ClickHouse_getServerUptime
+
+#define arginfo_class_ClickHouseStatement_rewind arginfo_class_ClickHouseRowIterator_rewind
+
+#define arginfo_class_ClickHouseStatement_valid arginfo_class_ClickHouse_writeEnd
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouseStatement_current, 0, 0, IS_MIXED, 0)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouseStatement_key arginfo_class_ClickHouseStatement_current
+
+#define arginfo_class_ClickHouseStatement_next arginfo_class_ClickHouseRowIterator_rewind
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouseStatement_offsetExists, 0, 1, _IS_BOOL, 0)
+	ZEND_ARG_TYPE_INFO(0, offset, IS_MIXED, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouseStatement_offsetGet, 0, 1, IS_MIXED, 0)
+	ZEND_ARG_TYPE_INFO(0, offset, IS_MIXED, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouseStatement_offsetSet, 0, 2, IS_VOID, 0)
+	ZEND_ARG_TYPE_INFO(0, offset, IS_MIXED, 0)
+	ZEND_ARG_TYPE_INFO(0, value, IS_MIXED, 0)
+ZEND_END_ARG_INFO()
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouseStatement_offsetUnset, 0, 1, IS_VOID, 0)
+	ZEND_ARG_TYPE_INFO(0, offset, IS_MIXED, 0)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouseStatement_jsonSerialize arginfo_class_ClickHouse_getServerInfo
+
+#define arginfo_class_ClickHouseStatement_toArray arginfo_class_ClickHouse_getServerInfo
+
+#define arginfo_class_ClickHouseStatement_statistics arginfo_class_ClickHouse_getServerInfo
+
+#define arginfo_class_ClickHouseStatement_fetchOne arginfo_class_ClickHouseStatement_current
+
+#define arginfo_class_ClickHouseStatement_fetchKeyPair arginfo_class_ClickHouse_getServerInfo
+
+#define arginfo_class_ClickHouseStatement_fetchColumn arginfo_class_ClickHouse_getServerInfo
+
+#define arginfo_class_ClickHouseException_getServerCode arginfo_class_ClickHouse_getServerUptime
+
+ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_class_ClickHouseException_getServerName, 0, 0, IS_STRING, 1)
+ZEND_END_ARG_INFO()
+
+#define arginfo_class_ClickHouseException_getQueryId arginfo_class_ClickHouseException_getServerName
+
+ZEND_METHOD(ClickHouse, __construct);
+ZEND_METHOD(ClickHouse, __destruct);
+ZEND_METHOD(ClickHouse, select);
+ZEND_METHOD(ClickHouse, selectWithExternalData);
+ZEND_METHOD(ClickHouse, selectToStream);
+ZEND_METHOD(ClickHouse, insert);
+ZEND_METHOD(ClickHouse, insertAssoc);
+ZEND_METHOD(ClickHouse, insertFromStream);
+ZEND_METHOD(ClickHouse, writeStart);
+ZEND_METHOD(ClickHouse, write);
+ZEND_METHOD(ClickHouse, writeEnd);
+ZEND_METHOD(ClickHouse, execute);
+ZEND_METHOD(ClickHouse, ping);
+ZEND_METHOD(ClickHouse, setSettings);
+ZEND_METHOD(ClickHouse, setSetting);
+ZEND_METHOD(ClickHouse, setDatabase);
+ZEND_METHOD(ClickHouse, setProgressCallback);
+ZEND_METHOD(ClickHouse, setProfileCallback);
+ZEND_METHOD(ClickHouse, setVerbose);
+ZEND_METHOD(ClickHouse, resetConnection);
+ZEND_METHOD(ClickHouse, getServerInfo);
+ZEND_METHOD(ClickHouse, getCurrentEndpoint);
+ZEND_METHOD(ClickHouse, getStatistics);
+ZEND_METHOD(ClickHouse, databaseSize);
+ZEND_METHOD(ClickHouse, tablesSize);
+ZEND_METHOD(ClickHouse, partitions);
+ZEND_METHOD(ClickHouse, showTables);
+ZEND_METHOD(ClickHouse, showCreateTable);
+ZEND_METHOD(ClickHouse, getServerUptime);
+ZEND_METHOD(ClickHouse, enableLogQueries);
+ZEND_METHOD(ClickHouse, getLogQueries);
+ZEND_METHOD(ClickHouse, selectStream);
+ZEND_METHOD(ClickHouse, selectStatement);
+ZEND_METHOD(ClickHouse, selectStreamCallback);
+ZEND_METHOD(ClickHouse, isExists);
+ZEND_METHOD(ClickHouse, showDatabases);
+ZEND_METHOD(ClickHouse, showProcesslist);
+ZEND_METHOD(ClickHouse, getServerVersion);
+ZEND_METHOD(ClickHouse, tableSize);
+ZEND_METHOD(ClickHouse, truncateTable);
+ZEND_METHOD(ClickHouse, dropPartition);
+ZEND_METHOD(ClickHouseRowIterator, rewind);
+ZEND_METHOD(ClickHouseRowIterator, valid);
+ZEND_METHOD(ClickHouseRowIterator, current);
+ZEND_METHOD(ClickHouseRowIterator, key);
+ZEND_METHOD(ClickHouseRowIterator, next);
+ZEND_METHOD(ClickHouseRowIterator, count);
+ZEND_METHOD(ClickHouseStatement, __construct);
+ZEND_METHOD(ClickHouseStatement, count);
+ZEND_METHOD(ClickHouseStatement, rewind);
+ZEND_METHOD(ClickHouseStatement, valid);
+ZEND_METHOD(ClickHouseStatement, current);
+ZEND_METHOD(ClickHouseStatement, key);
+ZEND_METHOD(ClickHouseStatement, next);
+ZEND_METHOD(ClickHouseStatement, offsetExists);
+ZEND_METHOD(ClickHouseStatement, offsetGet);
+ZEND_METHOD(ClickHouseStatement, offsetSet);
+ZEND_METHOD(ClickHouseStatement, offsetUnset);
+ZEND_METHOD(ClickHouseStatement, jsonSerialize);
+ZEND_METHOD(ClickHouseStatement, toArray);
+ZEND_METHOD(ClickHouseStatement, statistics);
+ZEND_METHOD(ClickHouseStatement, fetchOne);
+ZEND_METHOD(ClickHouseStatement, fetchKeyPair);
+ZEND_METHOD(ClickHouseStatement, fetchColumn);
+ZEND_METHOD(ClickHouseException, getServerCode);
+ZEND_METHOD(ClickHouseException, getServerName);
+ZEND_METHOD(ClickHouseException, getQueryId);
+
+static const zend_function_entry class_ClickHouse_methods[] = {
+	ZEND_ME(ClickHouse, __construct, arginfo_class_ClickHouse___construct, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, __destruct, arginfo_class_ClickHouse___destruct, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, select, arginfo_class_ClickHouse_select, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, selectWithExternalData, arginfo_class_ClickHouse_selectWithExternalData, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, selectToStream, arginfo_class_ClickHouse_selectToStream, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, insert, arginfo_class_ClickHouse_insert, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, insertAssoc, arginfo_class_ClickHouse_insertAssoc, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, insertFromStream, arginfo_class_ClickHouse_insertFromStream, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, writeStart, arginfo_class_ClickHouse_writeStart, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, write, arginfo_class_ClickHouse_write, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, writeEnd, arginfo_class_ClickHouse_writeEnd, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, execute, arginfo_class_ClickHouse_execute, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, ping, arginfo_class_ClickHouse_ping, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, setSettings, arginfo_class_ClickHouse_setSettings, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, setSetting, arginfo_class_ClickHouse_setSetting, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, setDatabase, arginfo_class_ClickHouse_setDatabase, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, setProgressCallback, arginfo_class_ClickHouse_setProgressCallback, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, setProfileCallback, arginfo_class_ClickHouse_setProfileCallback, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, setVerbose, arginfo_class_ClickHouse_setVerbose, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, resetConnection, arginfo_class_ClickHouse_resetConnection, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, getServerInfo, arginfo_class_ClickHouse_getServerInfo, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, getCurrentEndpoint, arginfo_class_ClickHouse_getCurrentEndpoint, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, getStatistics, arginfo_class_ClickHouse_getStatistics, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, databaseSize, arginfo_class_ClickHouse_databaseSize, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, tablesSize, arginfo_class_ClickHouse_tablesSize, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, partitions, arginfo_class_ClickHouse_partitions, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, showTables, arginfo_class_ClickHouse_showTables, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, showCreateTable, arginfo_class_ClickHouse_showCreateTable, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, getServerUptime, arginfo_class_ClickHouse_getServerUptime, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, enableLogQueries, arginfo_class_ClickHouse_enableLogQueries, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, getLogQueries, arginfo_class_ClickHouse_getLogQueries, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, selectStream, arginfo_class_ClickHouse_selectStream, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, selectStatement, arginfo_class_ClickHouse_selectStatement, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, selectStreamCallback, arginfo_class_ClickHouse_selectStreamCallback, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, isExists, arginfo_class_ClickHouse_isExists, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, showDatabases, arginfo_class_ClickHouse_showDatabases, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, showProcesslist, arginfo_class_ClickHouse_showProcesslist, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, getServerVersion, arginfo_class_ClickHouse_getServerVersion, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, tableSize, arginfo_class_ClickHouse_tableSize, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, truncateTable, arginfo_class_ClickHouse_truncateTable, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouse, dropPartition, arginfo_class_ClickHouse_dropPartition, ZEND_ACC_PUBLIC)
+	ZEND_FE_END
+};
+
+static const zend_function_entry class_ClickHouseRowIterator_methods[] = {
+	ZEND_ME(ClickHouseRowIterator, rewind, arginfo_class_ClickHouseRowIterator_rewind, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseRowIterator, valid, arginfo_class_ClickHouseRowIterator_valid, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseRowIterator, current, arginfo_class_ClickHouseRowIterator_current, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseRowIterator, key, arginfo_class_ClickHouseRowIterator_key, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseRowIterator, next, arginfo_class_ClickHouseRowIterator_next, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseRowIterator, count, arginfo_class_ClickHouseRowIterator_count, ZEND_ACC_PUBLIC)
+	ZEND_FE_END
+};
+
+static const zend_function_entry class_ClickHouseStatement_methods[] = {
+	ZEND_ME(ClickHouseStatement, __construct, arginfo_class_ClickHouseStatement___construct, ZEND_ACC_PRIVATE)
+	ZEND_ME(ClickHouseStatement, count, arginfo_class_ClickHouseStatement_count, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, rewind, arginfo_class_ClickHouseStatement_rewind, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, valid, arginfo_class_ClickHouseStatement_valid, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, current, arginfo_class_ClickHouseStatement_current, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, key, arginfo_class_ClickHouseStatement_key, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, next, arginfo_class_ClickHouseStatement_next, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, offsetExists, arginfo_class_ClickHouseStatement_offsetExists, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, offsetGet, arginfo_class_ClickHouseStatement_offsetGet, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, offsetSet, arginfo_class_ClickHouseStatement_offsetSet, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, offsetUnset, arginfo_class_ClickHouseStatement_offsetUnset, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, jsonSerialize, arginfo_class_ClickHouseStatement_jsonSerialize, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, toArray, arginfo_class_ClickHouseStatement_toArray, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, statistics, arginfo_class_ClickHouseStatement_statistics, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, fetchOne, arginfo_class_ClickHouseStatement_fetchOne, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, fetchKeyPair, arginfo_class_ClickHouseStatement_fetchKeyPair, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseStatement, fetchColumn, arginfo_class_ClickHouseStatement_fetchColumn, ZEND_ACC_PUBLIC)
+	ZEND_FE_END
+};
+
+static const zend_function_entry class_ClickHouseException_methods[] = {
+	ZEND_ME(ClickHouseException, getServerCode, arginfo_class_ClickHouseException_getServerCode, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseException, getServerName, arginfo_class_ClickHouseException_getServerName, ZEND_ACC_PUBLIC)
+	ZEND_ME(ClickHouseException, getQueryId, arginfo_class_ClickHouseException_getQueryId, ZEND_ACC_PUBLIC)
+	ZEND_FE_END
+};
+
+static zend_class_entry *register_class_ClickHouse(void)
+{
+	zend_class_entry ce, *class_entry;
+
+	INIT_CLASS_ENTRY(ce, "ClickHouse", class_ClickHouse_methods);
+	class_entry = zend_register_internal_class_with_flags(&ce, NULL, ZEND_ACC_FINAL);
+
+	zval const_FETCH_ONE_value;
+	ZVAL_LONG(&const_FETCH_ONE_value, 1);
+	zend_string *const_FETCH_ONE_name = zend_string_init_interned("FETCH_ONE", sizeof("FETCH_ONE") - 1, 1);
+	zend_declare_typed_class_constant(class_entry, const_FETCH_ONE_name, &const_FETCH_ONE_value, ZEND_ACC_PUBLIC, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(const_FETCH_ONE_name);
+
+	zval const_FETCH_KEY_PAIR_value;
+	ZVAL_LONG(&const_FETCH_KEY_PAIR_value, 2);
+	zend_string *const_FETCH_KEY_PAIR_name = zend_string_init_interned("FETCH_KEY_PAIR", sizeof("FETCH_KEY_PAIR") - 1, 1);
+	zend_declare_typed_class_constant(class_entry, const_FETCH_KEY_PAIR_name, &const_FETCH_KEY_PAIR_value, ZEND_ACC_PUBLIC, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(const_FETCH_KEY_PAIR_name);
+
+	zval const_DATE_AS_STRINGS_value;
+	ZVAL_LONG(&const_DATE_AS_STRINGS_value, 4);
+	zend_string *const_DATE_AS_STRINGS_name = zend_string_init_interned("DATE_AS_STRINGS", sizeof("DATE_AS_STRINGS") - 1, 1);
+	zend_declare_typed_class_constant(class_entry, const_DATE_AS_STRINGS_name, &const_DATE_AS_STRINGS_value, ZEND_ACC_PUBLIC, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(const_DATE_AS_STRINGS_name);
+
+	zval const_FETCH_COLUMN_value;
+	ZVAL_LONG(&const_FETCH_COLUMN_value, 8);
+	zend_string *const_FETCH_COLUMN_name = zend_string_init_interned("FETCH_COLUMN", sizeof("FETCH_COLUMN") - 1, 1);
+	zend_declare_typed_class_constant(class_entry, const_FETCH_COLUMN_name, &const_FETCH_COLUMN_value, ZEND_ACC_PUBLIC, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(const_FETCH_COLUMN_name);
+
+	zval property_host_default_value;
+	zend_string *property_host_default_value_str = zend_string_init("127.0.0.1", strlen("127.0.0.1"), 1);
+	ZVAL_STR(&property_host_default_value, property_host_default_value_str);
+	zend_string *property_host_name = zend_string_init("host", sizeof("host") - 1, 1);
+	zend_declare_typed_property(class_entry, property_host_name, &property_host_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_STRING));
+	zend_string_release(property_host_name);
+
+	zval property_port_default_value;
+	ZVAL_LONG(&property_port_default_value, 9000);
+	zend_string *property_port_name = zend_string_init("port", sizeof("port") - 1, 1);
+	zend_declare_typed_property(class_entry, property_port_name, &property_port_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(property_port_name);
+
+	zval property_database_default_value;
+	zend_string *property_database_default_value_str = zend_string_init("default", strlen("default"), 1);
+	ZVAL_STR(&property_database_default_value, property_database_default_value_str);
+	zend_string *property_database_name = zend_string_init("database", sizeof("database") - 1, 1);
+	zend_declare_typed_property(class_entry, property_database_name, &property_database_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_STRING));
+	zend_string_release(property_database_name);
+
+	zval property_user_default_value;
+	ZVAL_NULL(&property_user_default_value);
+	zend_string *property_user_name = zend_string_init("user", sizeof("user") - 1, 1);
+	zend_declare_typed_property(class_entry, property_user_name, &property_user_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_STRING|MAY_BE_NULL));
+	zend_string_release(property_user_name);
+
+	zval property_compression_default_value;
+	ZVAL_LONG(&property_compression_default_value, 0);
+	zend_string *property_compression_name = zend_string_init("compression", sizeof("compression") - 1, 1);
+	zend_declare_typed_property(class_entry, property_compression_name, &property_compression_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(property_compression_name);
+
+	zval property_retry_timeout_default_value;
+	ZVAL_LONG(&property_retry_timeout_default_value, 5);
+	zend_string *property_retry_timeout_name = zend_string_init("retry_timeout", sizeof("retry_timeout") - 1, 1);
+	zend_declare_typed_property(class_entry, property_retry_timeout_name, &property_retry_timeout_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(property_retry_timeout_name);
+
+	zval property_retry_count_default_value;
+	ZVAL_LONG(&property_retry_count_default_value, 1);
+	zend_string *property_retry_count_name = zend_string_init("retry_count", sizeof("retry_count") - 1, 1);
+	zend_declare_typed_property(class_entry, property_retry_count_name, &property_retry_count_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(property_retry_count_name);
+
+	zval property_receive_timeout_default_value;
+	ZVAL_LONG(&property_receive_timeout_default_value, 0);
+	zend_string *property_receive_timeout_name = zend_string_init("receive_timeout", sizeof("receive_timeout") - 1, 1);
+	zend_declare_typed_property(class_entry, property_receive_timeout_name, &property_receive_timeout_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(property_receive_timeout_name);
+
+	zval property_connect_timeout_default_value;
+	ZVAL_LONG(&property_connect_timeout_default_value, 5);
+	zend_string *property_connect_timeout_name = zend_string_init("connect_timeout", sizeof("connect_timeout") - 1, 1);
+	zend_declare_typed_property(class_entry, property_connect_timeout_name, &property_connect_timeout_default_value, ZEND_ACC_PROTECTED, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(property_connect_timeout_name);
+
+	return class_entry;
+}
+
+static zend_class_entry *register_class_ClickHouseRowIterator(zend_class_entry *class_entry_Iterator, zend_class_entry *class_entry_Countable)
+{
+	zend_class_entry ce, *class_entry;
+
+	INIT_CLASS_ENTRY(ce, "ClickHouseRowIterator", class_ClickHouseRowIterator_methods);
+	class_entry = zend_register_internal_class_with_flags(&ce, NULL, ZEND_ACC_FINAL);
+	zend_class_implements(class_entry, 2, class_entry_Iterator, class_entry_Countable);
+
+	return class_entry;
+}
+
+static zend_class_entry *register_class_ClickHouseStatement(zend_class_entry *class_entry_Iterator, zend_class_entry *class_entry_Countable, zend_class_entry *class_entry_ArrayAccess, zend_class_entry *class_entry_JsonSerializable)
+{
+	zend_class_entry ce, *class_entry;
+
+	INIT_CLASS_ENTRY(ce, "ClickHouseStatement", class_ClickHouseStatement_methods);
+	class_entry = zend_register_internal_class_with_flags(&ce, NULL, ZEND_ACC_FINAL);
+	zend_class_implements(class_entry, 4, class_entry_Iterator, class_entry_Countable, class_entry_ArrayAccess, class_entry_JsonSerializable);
+
+	return class_entry;
+}
+
+static zend_class_entry *register_class_ClickHouseException(zend_class_entry *class_entry_Exception)
+{
+	zend_class_entry ce, *class_entry;
+
+	INIT_CLASS_ENTRY(ce, "ClickHouseException", class_ClickHouseException_methods);
+	class_entry = zend_register_internal_class_with_flags(&ce, class_entry_Exception, 0);
+
+	zval property_server_code_default_value;
+	ZVAL_LONG(&property_server_code_default_value, 0);
+	zend_string *property_server_code_name = zend_string_init("server_code", sizeof("server_code") - 1, 1);
+	zend_declare_typed_property(class_entry, property_server_code_name, &property_server_code_default_value, ZEND_ACC_PUBLIC, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_LONG));
+	zend_string_release(property_server_code_name);
+
+	zval property_server_name_default_value;
+	ZVAL_NULL(&property_server_name_default_value);
+	zend_string *property_server_name_name = zend_string_init("server_name", sizeof("server_name") - 1, 1);
+	zend_declare_typed_property(class_entry, property_server_name_name, &property_server_name_default_value, ZEND_ACC_PUBLIC, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_STRING|MAY_BE_NULL));
+	zend_string_release(property_server_name_name);
+
+	zval property_query_id_default_value;
+	ZVAL_NULL(&property_query_id_default_value);
+	zend_string *property_query_id_name = zend_string_init("query_id", sizeof("query_id") - 1, 1);
+	zend_declare_typed_property(class_entry, property_query_id_name, &property_query_id_default_value, ZEND_ACC_PUBLIC, NULL, (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_STRING|MAY_BE_NULL));
+	zend_string_release(property_query_id_name);
+
+	return class_entry;
+}
diff --git a/composer.json b/composer.json
new file mode 100644
index 0000000..89f6b66
--- /dev/null
+++ b/composer.json
@@ -0,0 +1,41 @@
+{
+    "name": "iliaal/php_clickhouse",
+    "type": "php-ext",
+    "description": "Native PHP extension for ClickHouse using the official ClickHouse/clickhouse-cpp client. Connects over the native TCP protocol with LZ4 / ZSTD compression and optional TLS.",
+    "keywords": ["clickhouse", "php-extension", "pie", "olap", "database"],
+    "license": "PHP-3.01",
+    "homepage": "https://github.com/iliaal/php_clickhouse",
+    "authors": [
+        {
+            "name": "Ilia Alshanetsky",
+            "email": "ilia@ilia.ws"
+        },
+        {
+            "name": "SeasX Group (original SeasClick)",
+            "email": "ahhhh.wang@gmail.com",
+            "role": "Original author of SeasClick which php_clickhouse forks from"
+        }
+    ],
+    "require": {
+        "php": ">=7.4",
+        "ext-json": "*"
+    },
+    "php-ext": {
+        "extension-name": "clickhouse",
+        "configure-options": [
+            {
+                "name": "enable-clickhouse",
+                "description": "Enable ClickHouse client support",
+                "needs-value": false
+            },
+            {
+                "name": "enable-clickhouse-openssl",
+                "description": "Enable TLS/SSL support (requires OpenSSL development headers)",
+                "needs-value": false
+            }
+        ],
+        "support-zts": false,
+        "support-nts": true,
+        "download-url-method": ["pre-packaged-binary", "composer-default"]
+    }
+}
diff --git a/config.m4 b/config.m4
index 485b54b..a448224 100644
--- a/config.m4
+++ b/config.m4
@@ -1,107 +1,122 @@
 dnl $Id$
-dnl config.m4 for extension SeasClick
+dnl config.m4 for extension clickhouse
 
-dnl Comments in this file start with the string 'dnl'.
-dnl Remove where necessary. This file will not work
-dnl without editing.
+PHP_ARG_ENABLE(clickhouse, whether to enable ClickHouse support,
+[  --enable-clickhouse           Enable ClickHouse client support])
 
-dnl If your extension references something external, use with:
+PHP_ARG_ENABLE(clickhouse-openssl, whether to enable OpenSSL/TLS support in php_clickhouse,
+[  --enable-clickhouse-openssl Enable TLS/SSL support (requires OpenSSL)], no, no)
 
-dnl PHP_ARG_WITH(SeasClick, for SeasClick support,
-dnl Make sure that the comment is aligned:
-dnl [  --with-SeasClick             Include SeasClick support])
-
-dnl Otherwise use enable:
-
-PHP_ARG_ENABLE(SeasClick, whether to enable SeasClick support,
-Make sure that the comment is aligned:
-[  --enable-SeasClick           Enable SeasClick support])
-
-if test "$PHP_SEASCLICK" != "no"; then
-  dnl Write more examples of tests here...
-
-  dnl # --with-SeasClick -> check with-path
-  dnl SEARCH_PATH="/usr/local /usr"     # you might want to change this
-  dnl SEARCH_FOR="/include/SeasClick.h"  # you most likely want to change this
-  dnl if test -r $PHP_SEASCLICK/$SEARCH_FOR; then # path given as parameter
-  dnl   SEASCLICK_DIR=$PHP_SEASCLICK
-  dnl else # search default path list
-  dnl   AC_MSG_CHECKING([for SeasClick files in default path])
-  dnl   for i in $SEARCH_PATH ; do
-  dnl     if test -r $i/$SEARCH_FOR; then
-  dnl       SEASCLICK_DIR=$i
-  dnl       AC_MSG_RESULT(found in $i)
-  dnl     fi
-  dnl   done
-  dnl fi
-  dnl
-  dnl if test -z "$SEASCLICK_DIR"; then
-  dnl   AC_MSG_RESULT([not found])
-  dnl   AC_MSG_ERROR([Please reinstall the SeasClick distribution])
-  dnl fi
-
-  dnl # --with-SeasClick -> add include path
-  dnl PHP_ADD_INCLUDE($SEASCLICK_DIR/include)
+if test "$PHP_CLICKHOUSE" != "no"; then
+  PHP_REQUIRE_CXX()
+  PHP_SUBST(CLICKHOUSE_SHARED_LIBADD)
+  PHP_ADD_LIBRARY(stdc++, 1, CLICKHOUSE_SHARED_LIBADD)
+  CXXFLAGS="$CXXFLAGS -Wall -Wno-unused-function -Wno-deprecated -Wno-deprecated-declarations -std=c++17 -DZSTD_DISABLE_ASM"
+  CFLAGS="$CFLAGS -DZSTD_DISABLE_ASM"
 
-  dnl # --with-SeasClick -> check for lib and symbol presence
-  dnl LIBNAME=SeasClick # you may want to change this
-  dnl LIBSYMBOL=SeasClick # you most likely want to change this 
+  CLICKHOUSE_SSL_SRC=""
+  if test "$PHP_CLICKHOUSE_OPENSSL" != "no"; then
+    AC_MSG_CHECKING([for OpenSSL libraries])
+    PHP_ADD_LIBRARY(ssl, 1, CLICKHOUSE_SHARED_LIBADD)
+    PHP_ADD_LIBRARY(crypto, 1, CLICKHOUSE_SHARED_LIBADD)
+    AC_DEFINE([WITH_OPENSSL], [1], [Build php_clickhouse with TLS/SSL support])
+    CXXFLAGS="$CXXFLAGS -DWITH_OPENSSL"
+    CLICKHOUSE_SSL_SRC="lib/clickhouse-cpp/clickhouse/base/sslsocket.cpp"
+    AC_MSG_RESULT([enabled])
+  fi
 
-  dnl PHP_CHECK_LIBRARY($LIBNAME,$LIBSYMBOL,
-  dnl [
-  dnl   PHP_ADD_LIBRARY_WITH_PATH($LIBNAME, $SEASCLICK_DIR/$PHP_LIBDIR, SEASCLICK_SHARED_LIBADD)
-  dnl   AC_DEFINE(HAVE_SEASCLICKLIB,1,[ ])
-  dnl ],[
-  dnl   AC_MSG_ERROR([wrong SeasClick lib version or lib not found])
-  dnl ],[
-  dnl   -L$SEASCLICK_DIR/$PHP_LIBDIR -lm
-  dnl ])
-  dnl
-  dnl PHP_SUBST(SEASCLICK_SHARED_LIBADD)
-  PHP_REQUIRE_CXX()
-  PHP_SUBST(SEASCLICK_SHARED_LIBADD)
-  PHP_ADD_LIBRARY(stdc++, 1, SEASCLICK_SHARED_LIBADD)
-  CXXFLAGS="$CXXFLAGS -Wall -Wno-unused-function -Wno-deprecated -Wno-deprecated-declarations -std=c++11"
-  SeasClick_source_file="SeasClick.cpp \
+  clickhouse_source_file="clickhouse.cpp \
         typesToPhp.cpp \
-        lib/clickhouse-cpp/clickhouse/base/coded.cpp \
+        $CLICKHOUSE_SSL_SRC \
         lib/clickhouse-cpp/clickhouse/base/compressed.cpp \
+        lib/clickhouse-cpp/clickhouse/base/endpoints_iterator.cpp \
         lib/clickhouse-cpp/clickhouse/base/input.cpp \
         lib/clickhouse-cpp/clickhouse/base/output.cpp \
         lib/clickhouse-cpp/clickhouse/base/platform.cpp \
         lib/clickhouse-cpp/clickhouse/base/socket.cpp \
+        lib/clickhouse-cpp/clickhouse/base/wire_format.cpp \
         lib/clickhouse-cpp/clickhouse/columns/array.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/column.cpp \
         lib/clickhouse-cpp/clickhouse/columns/date.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/decimal.cpp \
         lib/clickhouse-cpp/clickhouse/columns/enum.cpp \
         lib/clickhouse-cpp/clickhouse/columns/factory.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/geo.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/ip4.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/ip6.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/itemview.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/lowcardinality.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/map.cpp \
         lib/clickhouse-cpp/clickhouse/columns/nullable.cpp \
         lib/clickhouse-cpp/clickhouse/columns/numeric.cpp \
         lib/clickhouse-cpp/clickhouse/columns/string.cpp \
+        lib/clickhouse-cpp/clickhouse/columns/time.cpp \
         lib/clickhouse-cpp/clickhouse/columns/tuple.cpp \
         lib/clickhouse-cpp/clickhouse/columns/uuid.cpp \
         lib/clickhouse-cpp/clickhouse/types/type_parser.cpp \
         lib/clickhouse-cpp/clickhouse/types/types.cpp \
-        lib/clickhouse-cpp/contrib/cityhash/city.cc \
-        lib/clickhouse-cpp/contrib/lz4/lz4.c \
-        lib/clickhouse-cpp/contrib/lz4/lz4hc.c \
-        lib/clickhouse-cpp/contrib/gtest/gtest-all.cc \
         lib/clickhouse-cpp/clickhouse/block.cpp \
         lib/clickhouse-cpp/clickhouse/client.cpp \
-        lib/clickhouse-cpp/clickhouse/query.cpp"
-  SeasClick_header_file="lib/clickhouse-cpp/contrib"
+        lib/clickhouse-cpp/clickhouse/query.cpp \
+        lib/clickhouse-cpp/contrib/cityhash/cityhash/city.cc \
+        lib/clickhouse-cpp/contrib/lz4/lz4/lz4.c \
+        lib/clickhouse-cpp/contrib/lz4/lz4/lz4hc.c \
+        lib/clickhouse-cpp/contrib/absl/absl/numeric/int128.cc \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/debug.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/entropy_common.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/error_private.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/fse_decompress.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/pool.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/threading.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/xxhash.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_common.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/fse_compress.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/hist.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/huf_compress.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_literals.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_sequences.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_superblock.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_double_fast.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_fast.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_lazy.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_opt.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstdmt_compress.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/decompress/huf_decompress.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_ddict.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress.c \
+        lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_block.c"
+
+  PHP_NEW_EXTENSION(clickhouse, $clickhouse_source_file, $ext_shared,,-DZEND_ENABLE_STATIC_TSRMLS_CACHE=1)
 
-  THIS_DIR=`dirname $0`
-  PHP_ADD_INCLUDE($THIS_DIR/lib/clickhouse-cpp/contrib)
-  
-  PHP_NEW_EXTENSION(SeasClick, $SeasClick_source_file, $ext_shared,,-DZEND_ENABLE_STATIC_TSRMLS_CACHE=1)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib/absl)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib/cityhash)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib/cityhash/cityhash)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib/lz4)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib/lz4/lz4)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib/zstd/zstd)
+  PHP_ADD_INCLUDE($ext_srcdir/lib/clickhouse-cpp/contrib/zstd/zstd/common)
+
+  PHP_ADD_EXTENSION_DEP(clickhouse, json)
   PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp)
-  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib)
-  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/cityhash)
-  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/gtest)
-  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/lz4)
   PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/clickhouse)
   PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/clickhouse/base)
   PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/clickhouse/types)
   PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/clickhouse/columns)
-  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/cityhash)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/cityhash/cityhash)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/lz4)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/lz4/lz4)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/absl)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/absl/absl)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/absl/absl/numeric)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/zstd)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/zstd/zstd)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/zstd/zstd/common)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/zstd/zstd/compress)
+  PHP_ADD_BUILD_DIR($ext_builddir/lib/clickhouse-cpp/contrib/zstd/zstd/decompress)
 fi
diff --git a/config.w32 b/config.w32
new file mode 100644
index 0000000..4c32610
--- /dev/null
+++ b/config.w32
@@ -0,0 +1,63 @@
+// vim:ft=javascript
+
+ARG_ENABLE("clickhouse", "Enable ClickHouse client support", "no");
+ARG_ENABLE("clickhouse-openssl", "Enable TLS/SSL support in php_clickhouse (requires OpenSSL)", "no");
+
+if (PHP_CLICKHOUSE != "no") {
+    var ext_dir = configure_module_dirname;
+    var ch_root = ext_dir + "\\lib\\clickhouse-cpp";
+
+    // C++17 + zstd ASM workaround + ws2_32 for sockets. /EHsc enables C++
+    // exception unwinding (clickhouse-cpp throws across the wire layer).
+    // /DNOMINMAX suppresses the windows.h max/min macros that otherwise
+    // collide with std::numeric_limits<T>::max() inside clickhouse-cpp
+    // (array.h template specializations). /bigobj covers the object-
+    // section count blow-up from the heavily templated columns code.
+    ADD_FLAG("CFLAGS_CLICKHOUSE",   "/std:c++17 /EHsc /bigobj /DNOMINMAX /DZSTD_DISABLE_ASM /D_CRT_SECURE_NO_WARNINGS /D_WINSOCK_DEPRECATED_NO_WARNINGS");
+    ADD_FLAG("CXXFLAGS_CLICKHOUSE", "/std:c++17 /EHsc /bigobj /DNOMINMAX /DZSTD_DISABLE_ASM /D_CRT_SECURE_NO_WARNINGS /D_WINSOCK_DEPRECATED_NO_WARNINGS");
+    ADD_FLAG("LDFLAGS_CLICKHOUSE",  "ws2_32.lib");
+
+    ADD_FLAG("CFLAGS_CLICKHOUSE",
+        "/I" + ch_root +
+        " /I" + ch_root + "\\contrib" +
+        " /I" + ch_root + "\\contrib\\absl" +
+        " /I" + ch_root + "\\contrib\\cityhash" +
+        " /I" + ch_root + "\\contrib\\cityhash\\cityhash" +
+        " /I" + ch_root + "\\contrib\\lz4" +
+        " /I" + ch_root + "\\contrib\\lz4\\lz4" +
+        " /I" + ch_root + "\\contrib\\zstd\\zstd" +
+        " /I" + ch_root + "\\contrib\\zstd\\zstd\\common");
+
+    var clickhouse_ssl_src = "";
+    if (PHP_CLICKHOUSE_OPENSSL != "no") {
+        if (CHECK_LIB("libssl.lib", "clickhouse") &&
+            CHECK_LIB("libcrypto.lib", "clickhouse") &&
+            CHECK_HEADER_ADD_INCLUDE("openssl/ssl.h", "CFLAGS_CLICKHOUSE")) {
+            ADD_FLAG("CXXFLAGS_CLICKHOUSE", "/DWITH_OPENSSL");
+            ADD_FLAG("CFLAGS_CLICKHOUSE",   "/DWITH_OPENSSL");
+            AC_DEFINE("WITH_OPENSSL", 1, "Build php_clickhouse with TLS/SSL support");
+            clickhouse_ssl_src = "sslsocket.cpp";
+        } else {
+            WARNING("clickhouse-openssl was requested but libssl/libcrypto were not found; building without TLS support.");
+        }
+    }
+
+    EXTENSION("clickhouse", "clickhouse.cpp typesToPhp.cpp", null, "/DZEND_ENABLE_STATIC_TSRMLS_CACHE=1");
+    ADD_EXTENSION_DEP("clickhouse", "json");
+
+    ADD_SOURCES(ch_root + "\\clickhouse",            "block.cpp client.cpp query.cpp",                                                                                                                                                                                                                              "clickhouse");
+    ADD_SOURCES(ch_root + "\\clickhouse\\base",      "compressed.cpp endpoints_iterator.cpp input.cpp output.cpp platform.cpp socket.cpp wire_format.cpp " + clickhouse_ssl_src,                                                                                                                                    "clickhouse");
+    ADD_SOURCES(ch_root + "\\clickhouse\\columns",   "array.cpp column.cpp date.cpp decimal.cpp enum.cpp factory.cpp geo.cpp ip4.cpp ip6.cpp itemview.cpp lowcardinality.cpp map.cpp nullable.cpp numeric.cpp string.cpp time.cpp tuple.cpp uuid.cpp",                                                              "clickhouse");
+    ADD_SOURCES(ch_root + "\\clickhouse\\types",     "type_parser.cpp types.cpp",                                                                                                                                                                                                                                   "clickhouse");
+
+    // Vendored dependencies (pure C/C++; built into the same module).
+    ADD_SOURCES(ch_root + "\\contrib\\cityhash\\cityhash", "city.cc", "clickhouse");
+    ADD_SOURCES(ch_root + "\\contrib\\lz4\\lz4",           "lz4.c lz4hc.c", "clickhouse");
+    ADD_SOURCES(ch_root + "\\contrib\\absl\\absl\\numeric", "int128.cc", "clickhouse");
+    ADD_SOURCES(ch_root + "\\contrib\\zstd\\zstd\\common",
+        "debug.c entropy_common.c error_private.c fse_decompress.c pool.c threading.c xxhash.c zstd_common.c", "clickhouse");
+    ADD_SOURCES(ch_root + "\\contrib\\zstd\\zstd\\compress",
+        "fse_compress.c hist.c huf_compress.c zstd_compress.c zstd_compress_literals.c zstd_compress_sequences.c zstd_compress_superblock.c zstd_double_fast.c zstd_fast.c zstd_lazy.c zstd_ldm.c zstd_opt.c zstdmt_compress.c", "clickhouse");
+    ADD_SOURCES(ch_root + "\\contrib\\zstd\\zstd\\decompress",
+        "huf_decompress.c zstd_ddict.c zstd_decompress.c zstd_decompress_block.c", "clickhouse");
+}
diff --git a/config.w32.without b/config.w32.without
deleted file mode 100644
index 398f697..0000000
--- a/config.w32.without
+++ /dev/null
@@ -1,13 +0,0 @@
-// $Id$
-// vim:ft=javascript
-
-// If your extension references something external, use ARG_WITH
-// ARG_WITH("SeasClick", "for SeasClick support", "no");
-
-// Otherwise, use ARG_ENABLE
-// ARG_ENABLE("SeasClick", "enable SeasClick support", "no");
-
-if (PHP_SEASCLICK != "no") {
-	EXTENSION("SeasClick", "SeasClick.c", PHP_EXTNAME_SHARED, "/DZEND_ENABLE_STATIC_TSRMLS_CACHE=1");
-}
-
diff --git a/images/php_clickhouse-hero.jpg b/images/php_clickhouse-hero.jpg
new file mode 100755
index 0000000..7fa42cf
Binary files /dev/null and b/images/php_clickhouse-hero.jpg differ
diff --git a/lib/clickhouse-cpp/.buckconfig b/lib/clickhouse-cpp/.buckconfig
deleted file mode 100644
index c885451..0000000
--- a/lib/clickhouse-cpp/.buckconfig
+++ /dev/null
@@ -1,2 +0,0 @@
-[cxx]
-  gtest_dep = //contrib/gtest:gtest
diff --git a/lib/clickhouse-cpp/.gitattributes b/lib/clickhouse-cpp/.gitattributes
deleted file mode 100644
index 1ff0c42..0000000
--- a/lib/clickhouse-cpp/.gitattributes
+++ /dev/null
@@ -1,63 +0,0 @@
-###############################################################################
-# Set default behavior to automatically normalize line endings.
-###############################################################################
-* text=auto
-
-###############################################################################
-# Set default behavior for command prompt diff.
-#
-# This is need for earlier builds of msysgit that does not have it on by
-# default for csharp files.
-# Note: This is only used by command line
-###############################################################################
-#*.cs     diff=csharp
-
-###############################################################################
-# Set the merge driver for project and solution files
-#
-# Merging from the command prompt will add diff markers to the files if there
-# are conflicts (Merging from VS is not affected by the settings below, in VS
-# the diff markers are never inserted). Diff markers may cause the following 
-# file extensions to fail to load in VS. An alternative would be to treat
-# these files as binary and thus will always conflict and require user
-# intervention with every merge. To do so, just uncomment the entries below
-###############################################################################
-#*.sln       merge=binary
-#*.csproj    merge=binary
-#*.vbproj    merge=binary
-#*.vcxproj   merge=binary
-#*.vcproj    merge=binary
-#*.dbproj    merge=binary
-#*.fsproj    merge=binary
-#*.lsproj    merge=binary
-#*.wixproj   merge=binary
-#*.modelproj merge=binary
-#*.sqlproj   merge=binary
-#*.wwaproj   merge=binary
-
-###############################################################################
-# behavior for image files
-#
-# image files are treated as binary by default.
-###############################################################################
-#*.jpg   binary
-#*.png   binary
-#*.gif   binary
-
-###############################################################################
-# diff behavior for common document formats
-# 
-# Convert binary document formats to text before diffing them. This feature
-# is only available from the command line. Turn it on by uncommenting the 
-# entries below.
-###############################################################################
-#*.doc   diff=astextplain
-#*.DOC   diff=astextplain
-#*.docx  diff=astextplain
-#*.DOCX  diff=astextplain
-#*.dot   diff=astextplain
-#*.DOT   diff=astextplain
-#*.pdf   diff=astextplain
-#*.PDF   diff=astextplain
-#*.rtf   diff=astextplain
-#*.RTF   diff=astextplain
diff --git a/lib/clickhouse-cpp/.gitignore b/lib/clickhouse-cpp/.gitignore
deleted file mode 100644
index 7b251ba..0000000
--- a/lib/clickhouse-cpp/.gitignore
+++ /dev/null
@@ -1,271 +0,0 @@
-## Ignore Visual Studio temporary files, build results, and
-## files generated by popular Visual Studio add-ons.
-
-# User-specific files
-*.suo
-*.user
-*.userosscache
-*.sln.docstates
-
-# User-specific files (MonoDevelop/Xamarin Studio)
-*.userprefs
-# Temporary solution for VS 2015
-vs/
-
-# Build results
-[Dd]ebug/
-[Dd]ebugPublic/
-[Rr]elease/
-[Rr]eleases/
-x64/
-x86/
-bld/
-build/
-[Bb]in/
-[Oo]bj/
-[Ll]og/
-
-# Visual Studio 2015 cache/options directory
-.vs/
-# Uncomment if you have tasks that create the project's static files in wwwroot
-#wwwroot/
-
-# MSTest test Results
-[Tt]est[Rr]esult*/
-[Bb]uild[Ll]og.*
-
-# NUNIT
-*.VisualState.xml
-TestResult.xml
-
-# Build Results of an ATL Project
-[Dd]ebugPS/
-[Rr]eleasePS/
-dlldata.c
-
-# DNX
-project.lock.json
-project.fragment.lock.json
-artifacts/
-
-*_i.c
-*_p.c
-*_i.h
-*.ilk
-*.meta
-*.obj
-*.pch
-*.pdb
-*.pgc
-*.pgd
-*.rsp
-*.sbr
-*.tlb
-*.tli
-*.tlh
-*.tmp
-*.tmp_proj
-*.log
-*.vspscc
-*.vssscc
-.builds
-*.pidb
-*.svclog
-*.scc
-
-# Chutzpah Test files
-_Chutzpah*
-
-# Visual C++ cache files
-ipch/
-*.aps
-*.ncb
-*.opendb
-*.opensdf
-*.sdf
-*.cachefile
-*.VC.db
-*.VC.VC.opendb
-
-# Visual Studio profiler
-*.psess
-*.vsp
-*.vspx
-*.sap
-
-# TFS 2012 Local Workspace
-$tf/
-
-# Guidance Automation Toolkit
-*.gpState
-
-# ReSharper is a .NET coding add-in
-_ReSharper*/
-*.[Rr]e[Ss]harper
-*.DotSettings.user
-
-# JustCode is a .NET coding add-in
-.JustCode
-
-# TeamCity is a build add-in
-_TeamCity*
-
-# DotCover is a Code Coverage Tool
-*.dotCover
-
-# NCrunch
-_NCrunch_*
-.*crunch*.local.xml
-nCrunchTemp_*
-
-# MightyMoose
-*.mm.*
-AutoTest.Net/
-
-# Web workbench (sass)
-.sass-cache/
-
-# Installshield output folder
-[Ee]xpress/
-
-# DocProject is a documentation generator add-in
-DocProject/buildhelp/
-DocProject/Help/*.HxT
-DocProject/Help/*.HxC
-DocProject/Help/*.hhc
-DocProject/Help/*.hhk
-DocProject/Help/*.hhp
-DocProject/Help/Html2
-DocProject/Help/html
-
-# Click-Once directory
-publish/
-
-# Publish Web Output
-*.[Pp]ublish.xml
-*.azurePubxml
-# TODO: Comment the next line if you want to checkin your web deploy settings
-# but database connection strings (with potential passwords) will be unencrypted
-#*.pubxml
-*.publishproj
-
-# Microsoft Azure Web App publish settings. Comment the next line if you want to
-# checkin your Azure Web App publish settings, but sensitive information contained
-# in these scripts will be unencrypted
-PublishScripts/
-
-# NuGet Packages
-*.nupkg
-# The packages folder can be ignored because of Package Restore
-**/packages/*
-# except build/, which is used as an MSBuild target.
-!**/packages/build/
-# Uncomment if necessary however generally it will be regenerated when needed
-#!**/packages/repositories.config
-# NuGet v3's project.json files produces more ignoreable files
-*.nuget.props
-*.nuget.targets
-
-# Microsoft Azure Build Output
-csx/
-*.build.csdef
-
-# Microsoft Azure Emulator
-ecf/
-rcf/
-
-# Windows Store app package directories and files
-AppPackages/
-BundleArtifacts/
-Package.StoreAssociation.xml
-_pkginfo.txt
-
-# Visual Studio cache files
-# files ending in .cache can be ignored
-*.[Cc]ache
-# but keep track of directories ending in .cache
-!*.[Cc]ache/
-
-# Others
-ClientBin/
-~$*
-*~
-*.dbmdl
-*.dbproj.schemaview
-*.jfm
-*.pfx
-*.publishsettings
-node_modules/
-orleans.codegen.cs
-
-# Since there are multiple workflows, uncomment next line to ignore bower_components
-# (https://github.com/github/gitignore/pull/1529#issuecomment-104372622)
-#bower_components/
-
-# RIA/Silverlight projects
-Generated_Code/
-
-# Backup & report files from converting an old project file
-# to a newer Visual Studio version. Backup files are not needed,
-# because we have git ;-)
-_UpgradeReport_Files/
-Backup*/
-UpgradeLog*.XML
-UpgradeLog*.htm
-
-# SQL Server files
-*.mdf
-*.ldf
-
-# Business Intelligence projects
-*.rdl.data
-*.bim.layout
-*.bim_*.settings
-
-# Microsoft Fakes
-FakesAssemblies/
-
-# GhostDoc plugin setting file
-*.GhostDoc.xml
-
-# Node.js Tools for Visual Studio
-.ntvs_analysis.dat
-
-# Visual Studio 6 build log
-*.plg
-
-# Visual Studio 6 workspace options file
-*.opt
-
-# Visual Studio LightSwitch build output
-**/*.HTMLClient/GeneratedArtifacts
-**/*.DesktopClient/GeneratedArtifacts
-**/*.DesktopClient/ModelManifest.xml
-**/*.Server/GeneratedArtifacts
-**/*.Server/ModelManifest.xml
-_Pvt_Extensions
-
-# Paket dependency manager
-.paket/paket.exe
-paket-files/
-
-# FAKE - F# Make
-.fake/
-
-# JetBrains Rider
-.idea/
-*.sln.iml
-
-# CodeRush
-.cr/
-
-# Python Tools for Visual Studio (PTVS)
-__pycache__/
-*.pyc
-
-# Buck
-/buck-out/
-/.buckd/
-/buckaroo/
-.buckconfig.local
-BUCKAROO_DEPS
diff --git a/lib/clickhouse-cpp/.travis.yml b/lib/clickhouse-cpp/.travis.yml
deleted file mode 100644
index 1e26bee..0000000
--- a/lib/clickhouse-cpp/.travis.yml
+++ /dev/null
@@ -1,32 +0,0 @@
-# Enable C++ support
-language: cpp
-
-matrix:
-  include:
-    - os: linux
-      dist: trusty
-      sudo: required
-      compiler: gcc
-
-    - os: linux
-      dist: trusty
-      sudo: required
-      compiler: clang
-
-    - os: osx
-      osx_image: xcode8.2
-      compiler: clang
-
-before_install:
-  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then echo 'deb http://repo.yandex.ru/clickhouse/deb/stable main/' | sudo tee /etc/apt/sources.list.d/clickhouse.list ; fi
-  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then sudo apt-key adv --keyserver keyserver.ubuntu.com --recv E0C56BD4 ; fi
-  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then sudo apt-get update -q && sudo apt-get install -q -y clickhouse-server-common ; fi
-  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then sudo service clickhouse-server start ; fi
-
-# Build steps
-script:
-  - mkdir build
-  - cd build
-  - cmake .. && make
-  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then ./ut/clickhouse-cpp-ut ; fi
-  - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then ./ut/clickhouse-cpp-ut --gtest_filter='-Client/*' ; fi
diff --git a/lib/clickhouse-cpp/BUCK b/lib/clickhouse-cpp/BUCK
deleted file mode 100644
index 0c86b12..0000000
--- a/lib/clickhouse-cpp/BUCK
+++ /dev/null
@@ -1,20 +0,0 @@
-cxx_library(
-  name = 'clickhouse-cpp',
-  header_namespace = 'clickhouse',
-  exported_headers = subdir_glob([
-    ('clickhouse', '**/*.h'),
-  ]),
-  srcs = glob([
-    'clickhouse/**/*.cpp',
-  ]),
-  compiler_flags = [
-    '-std=c++11',
-  ],
-  visibility = [
-    'PUBLIC',
-  ],
-  deps = [
-    '//contrib/cityhash:cityhash',
-    '//contrib/lz4:lz4',
-  ]
-)
diff --git a/lib/clickhouse-cpp/CMakeLists.txt b/lib/clickhouse-cpp/CMakeLists.txt
deleted file mode 100644
index fa3856f..0000000
--- a/lib/clickhouse-cpp/CMakeLists.txt
+++ /dev/null
@@ -1,35 +0,0 @@
-CMAKE_MINIMUM_REQUIRED(VERSION 3.0.2)
-
-INCLUDE (cmake/cpp11.cmake)
-INCLUDE (cmake/subdirs.cmake)
-
-OPTION(BUILD_BENCHMARK "Build benchmark" OFF)
-
-PROJECT (CLICKHOUSE-CLIENT)
-
-    USE_CXX11()
-
-    IF (UNIX)
-        IF (APPLE)
-            SET (CMAKE_CXX_FLAGS "-O2 -Wall -Wextra -Werror")
-        ELSE ()
-            SET (CMAKE_CXX_FLAGS "-O2 -pthread -Wall -Wextra -Werror")
-        ENDIF ()
-        SET (CMAKE_EXE_LINKER_FLAGS, "-lpthread")
-    ENDIF ()
-
-    INCLUDE_DIRECTORIES(.)
-    INCLUDE_DIRECTORIES(contrib)
-
-    SUBDIRS (
-        clickhouse
-        contrib/cityhash
-        contrib/gtest
-        contrib/lz4
-        tests/simple
-        ut
-    )
-
-    IF (BUILD_BENCHMARK)
-        SUBDIRS(bench)
-    ENDIF (BUILD_BENCHMARK)
diff --git a/lib/clickhouse-cpp/LICENSE b/lib/clickhouse-cpp/LICENSE
index 59dec86..9269404 100644
--- a/lib/clickhouse-cpp/LICENSE
+++ b/lib/clickhouse-cpp/LICENSE
@@ -1,13 +1,206 @@
+Copyright 2018-2023 ClickHouse, Inc.
 Copyright 2017 Pavel Artemkin
 
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
 
-    http://www.apache.org/licenses/LICENSE-2.0
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
 
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2018-2023 ClickHouse, Inc.
+   Copyright 2017 Pavel Artemkin
+   
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
diff --git a/lib/clickhouse-cpp/LOCAL_PATCHES.md b/lib/clickhouse-cpp/LOCAL_PATCHES.md
new file mode 100644
index 0000000..c3722f2
--- /dev/null
+++ b/lib/clickhouse-cpp/LOCAL_PATCHES.md
@@ -0,0 +1,99 @@
+# Local patches against vendored clickhouse-cpp v2.6.1
+
+Anything listed here must be re-applied (or upstreamed and dropped)
+when the vendored library is bumped.
+
+## clickhouse/client.cpp: `Client::Impl::BeginInsert` drops `query_id`
+
+`Client::Impl::BeginInsert(Query query)` constructs the wire-level
+`SendQuery` from `query.GetText()` rather than from the full `Query`
+object. The implicit `Query(const std::string&)` constructor uses
+`Query::default_query_id` (empty), so any `query_id` passed into
+`Client::BeginInsert(query, query_id)` is silently discarded and the
+INSERT lands in `system.query_log` with an auto-generated id.
+
+The one-shot `Client::Impl::Insert` path is unaffected because it
+calls `SendQuery(query)` directly with the `Query` object.
+
+Patch: change `SendQuery(query.GetText())` to `SendQuery(query)` in
+`Client::Impl::BeginInsert`.
+
+This is exercised by `tests/028.phpt` (writeStart query_id propagation).
+
+## clickhouse/columns/string.cpp: `memcpy(NULL, 0)` UB on empty string_view
+
+`StringBlock::AppendUnsafe` calls `memcpy(pos, str.data(), str.size())`
+unconditionally. When `str` was constructed from an empty
+`std::string`, `str.data()` is allowed to be `NULL`, and libc's memcpy
+declares argument 2 with `__attribute__((nonnull))` regardless of the
+size. UBSan flags every empty append as undefined behavior:
+
+```
+runtime error: null pointer passed as argument 2,
+  which is declared to never be null
+```
+
+Every libc no-ops `memcpy(_, NULL, 0)` in practice, so the bug is
+benign on real workloads, but the false-positive UBSan trip noised the
+extension's ASan job and obscured real findings.
+
+Patch: guard the `memcpy` with `if (str.size() > 0)`. This is
+exercised by `tests/018.phpt` (LowCardinality(String) with empty
+values).
+
+## contrib/cityhash/cityhash/city.cc: unconditional `#include "config.h"`
+
+The vendored cityhash drop ships an unguarded `#include "config.h"`
+at the top of `city.cc`. autoconf-style: the file is generated by
+the upstream cityhash configure step and defines `WORDS_BIGENDIAN`
+and `HAVE_BUILTIN_EXPECT` (both with safe fall-throughs to LE / no
+__builtin_expect when undefined).
+
+clickhouse-cpp does not generate one for vendored use, and PHP's
+build system does not put a compatible `config.h` on the include
+path on Windows (autotools does on Linux only by accident, where
+the include resolves but the file's contents end up irrelevant).
+The Windows CI build failed with:
+
+```
+contrib/cityhash/cityhash/city.cc(30): fatal error C1083:
+  Cannot open include file: 'config.h': No such file or directory
+```
+
+Patch: gate the include with `HAVE_CONFIG_H` per the standard
+autoconf convention. cityhash's downstream defaults are safe; the
+file falls through to `__BYTE_ORDER == __LITTLE_ENDIAN` detection
+or unconditional little-endian on MSVC.
+
+## clickhouse/client.{h,cpp}: no public `BeginInsert(const Query&)` overload
+
+`Client::Impl::BeginInsert` already takes a `Query`, but the public
+surface only exposes `BeginInsert(string)` and
+`BeginInsert(string, string)`. That means callers cannot attach
+per-query settings, server-side params, or progress callbacks to an
+INSERT, even though the impl supports it.
+
+Patch: declare and forward a third overload, `Block
+BeginInsert(const Query&)`, so the binding can pass a fully
+configured `Query` (settings + params + callbacks + query_id) into
+the streaming insert path. Used by `php_clickhouse` to honor
+`setSettings()` / per-call settings on `insert()` and `writeStart()`.
+
+## clickhouse/client.{h,cpp}: no public `SelectWithExternalData(const Query&, ...)` overload
+
+`Client::Impl::SelectWithExternalData` already takes a `Query`, but
+the public surface only exposes the `std::string` overloads
+(`SelectWithExternalData(string, externals, cb)` and
+`SelectWithExternalData(string, query_id, externals, cb)`). That
+means callers cannot attach per-query settings, server-side params,
+or progress / verbose callbacks to a SELECT-with-external-data, even
+though the impl supports it.
+
+Patch: declare and forward a thin overload,
+`void SelectWithExternalData(const Query& query, const ExternalTables& external_tables);`,
+so the binding can pass a fully configured `Query` (settings + params
++ OnData + query_id + OnProgress) into the external-data SELECT path.
+Used by `php_clickhouse::selectWithExternalData()` to honor
+`setSettings()` / per-call settings and to thread `query_id` through
+`system.query_log`. Exercised by `tests/095.phpt`, `tests/096.phpt`,
+`tests/097.phpt`.
diff --git a/lib/clickhouse-cpp/README.md b/lib/clickhouse-cpp/README.md
index 1a840db..bd48575 100644
--- a/lib/clickhouse-cpp/README.md
+++ b/lib/clickhouse-cpp/README.md
@@ -1,74 +1,260 @@
-ClickHouse C++ client [![Build Status](https://travis-ci.org/artpaul/clickhouse-cpp.svg?branch=master)](https://travis-ci.org/artpaul/clickhouse-cpp)
+ClickHouse C++ client [![Linux](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/linux.yml/badge.svg)](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/linux.yml) [![macOS](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/macos.yml/badge.svg)](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/macos.yml) [![Windows MSVC](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/windows_msvc.yml/badge.svg)](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/windows_msvc.yml) [![Windows mingw](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/windows_mingw.yml/badge.svg)](https://github.com/ClickHouse/clickhouse-cpp/actions/workflows/windows_mingw.yml)
 =====
 
-C++ client for [Yandex ClickHouse](https://clickhouse.yandex/)
-
-## This repositorie change
-* [Add InsertQuery and InsertData methods](https://github.com/aiwhj/clickhouse-cpp/commit/bab28bcb5a509d80b8e2e0c7e89512446283dde5)
+C++ client for [ClickHouse](https://clickhouse.com/).
 
 ## Supported data types
 
 * Array(T)
 * Date
-* DateTime
+* DateTime, DateTime64
+* DateTime([timezone]), DateTime64(N, [timezone])
+* Decimal32, Decimal64, Decimal128
 * Enum8, Enum16
 * FixedString(N)
 * Float32, Float64
+* IPv4, IPv6
 * Nullable(T)
 * String
+* LowCardinality(String) or LowCardinality(FixedString(N))
 * Tuple
 * UInt8, UInt16, UInt32, UInt64, Int8, Int16, Int32, Int64
+* UInt128, Int128
+* UUID
+* Map
+* Point, Ring, Polygon, MultiPolygon
+
+## Dependencies
+In the most basic case one needs only:
+- a C++-17-complaint compiler,
+- `cmake` (3.12 or newer), and
+- `ninja`
+
+Optional dependencies:
+- openssl
+- liblz4
+- libabsl
+- libzstd
 
 ## Building
 
 ```sh
 $ mkdir build .
 $ cd build
-$ cmake ..
+$ cmake .. [-DBUILD_TESTS=ON]
 $ make
 ```
 
-## Example
+Please refer to the workflows for the reference on dependencies/build options
+- https://github.com/ClickHouse/clickhouse-cpp/blob/master/.github/workflows/linux.yml
+- https://github.com/ClickHouse/clickhouse-cpp/blob/master/.github/workflows/windows_msvc.yml
+- https://github.com/ClickHouse/clickhouse-cpp/blob/master/.github/workflows/windows_mingw.yml
+- https://github.com/ClickHouse/clickhouse-cpp/blob/master/.github/workflows/macos.yml
+
+
+## Example application build with clickhouse-cpp
+
+There are various ways to integrate clickhouse-cpp with the build system of an application. Below example uses the simple approach based on
+submodules presented in https://www.youtube.com/watch?v=ED-WUk440qc .
+
+- `mkdir clickhouse-app && cd clickhouse-app && git init`
+- `git submodule add https://github.com/ClickHouse/clickhouse-cpp.git contribs/clickhouse-cpp`
+- `touch app.cpp`, then copy the following C++ code into that file
 
 ```cpp
+#include <iostream>
 #include <clickhouse/client.h>
 
 using namespace clickhouse;
 
-/// Initialize client connection.
-Client client(ClientOptions().SetHost("localhost"));
+int main()
+{
+    /// Initialize client connection.
+    Client client(ClientOptions().SetHost("localhost"));
 
-/// Create a table.
-client.Execute("CREATE TABLE IF NOT EXISTS test.numbers (id UInt64, name String) ENGINE = Memory");
+    /// Create a table.
+    client.Execute("CREATE TABLE IF NOT EXISTS default.numbers (id UInt64, name String) ENGINE = Memory");
 
-/// Insert some values.
-{
-    Block block;
+    /// Insert some values.
+    {
+        Block block;
 
-    auto id = std::make_shared<ColumnUInt64>();
-    id->Append(1);
-    id->Append(7);
+        auto id = std::make_shared<ColumnUInt64>();
+        id->Append(1);
+        id->Append(7);
 
-    auto name = std::make_shared<ColumnString>();
-    name->Append("one");
-    name->Append("seven");
+        auto name = std::make_shared<ColumnString>();
+        name->Append("one");
+        name->Append("seven");
 
-    block.AppendColumn("id"  , id);
-    block.AppendColumn("name", name);
+        block.AppendColumn("id"  , id);
+        block.AppendColumn("name", name);
 
-    client.Insert("test.numbers", block);
-}
+        client.Insert("default.numbers", block);
+    }
 
-/// Select values inserted in the previous step.
-client.Select("SELECT id, name FROM test.numbers", [] (const Block& block)
-    {
-        for (size_t i = 0; i < block.GetRowCount(); ++i) {
-            std::cout << block[0]->As<ColumnUInt64>()->At(i) << " "
-                      << block[1]->As<ColumnString>()->At(i) << "\n";
+    /// Select values inserted in the previous step.
+    client.Select("SELECT id, name FROM default.numbers", [] (const Block& block)
+        {
+            for (size_t i = 0; i < block.GetRowCount(); ++i) {
+                std::cout << block[0]->As<ColumnUInt64>()->At(i) << " "
+                          << block[1]->As<ColumnString>()->At(i) << "\n";
+            }
         }
+    );
+
+    /// Select values inserted in the previous step using external data feature
+    /// See https://clickhouse.com/docs/engines/table-engines/special/external-data
+    {
+        Block block1, block2;
+        auto id = std::make_shared<ColumnUInt64>();
+        id->Append(1);
+        block1.AppendColumn("id"  , id);
+
+        auto name = std::make_shared<ColumnString>();
+        name->Append("seven");
+        block2.AppendColumn("name", name);
+
+        const std::string _1 = "_1";
+        const std::string _2 = "_2";
+
+        const ExternalTables external = {{_1, block1}, {_2, block2}};
+        client.SelectWithExternalData("SELECT id, name FROM default.numbers where id in (_1) or name in (_2)",
+                                      external, [] (const Block& block)
+            {
+                for (size_t i = 0; i < block.GetRowCount(); ++i) {
+                    std::cout << block[0]->As<ColumnUInt64>()->At(i) << " "
+                              << block[1]->As<ColumnString>()->At(i) << "\n";
+                }
+            }
+        );
     }
-);
 
-/// Delete table.
-client.Execute("DROP TABLE test.numbers");
+    /// Delete table.
+    client.Execute("DROP TABLE default.numbers");
+
+    return 0;
+}
+```
+
+- `touch CMakeLists.txt`, then copy the following CMake code into that file
+
+```cmake
+cmake_minimum_required(VERSION 3.12)
+project(application-example)
+
+set(CMAKE_CXX_STANDARD 17)
+
+add_subdirectory(contribs/clickhouse-cpp)
+
+add_executable(${PROJECT_NAME} "app.cpp")
+
+target_include_directories(${PROJECT_NAME} PRIVATE contribs/clickhouse-cpp/ contribs/clickhouse-cpp/contrib/absl)
+
+target_link_libraries(${PROJECT_NAME} PRIVATE clickhouse-cpp-lib)
 ```
+
+- run `rm -rf build && cmake -B build -S . && cmake --build build -j32` to remove remainders of the previous builds, run CMake and build the
+  application. The generated binary is located in location `build/application-example`.
+
+## Batch Insertion
+
+In addition to the `Insert` method, which inserts all the data in a block in a
+single call, you can use the `BeginInsert` / `InsertData` / `EndInsert`
+pattern to insert batches of data. This can be useful for managing larger data
+sets without inflating memory with the entire set.
+
+To use it pass `BeginInsert` an `INSERT` statement ending in `VALUES` but with
+no actual values. Use the resulting `Block` to append batches of data, sending
+each to the sever with `InsertData`. Finally, call `EndInsert` (or let the
+client go out of scope) to signal the server that insertion is complete.
+Example:
+
+```cpp
+// Start the insertion.
+auto block = client->BeginInsert("INSERT INTO foo (id, name) VALUES");
+
+// Grab the columns from the block.
+auto col1 = block[0]->As<ColumnUInt64>();
+auto col2 = block[1]->As<ColumnString>();
+
+// Add a couple of records to the block.
+col1.Append(1);
+col1.Append(2);
+col2.Append("holden");
+col2.Append("naomi");
+
+// Send those records.
+block.RefreshRowCount();
+client->InsertData(block);
+block.Clear();
+
+// Add another record.
+col1.Append(3);
+col2.Append("amos");
+
+// Send it and finish.
+block.RefreshRowCount();
+client->EndInsert(block);
+```
+
+## Thread-safety
+⚠ Please note that `Client` instance is NOT thread-safe. I.e. you must create a separate `Client` for each thread or utilize some synchronization techniques. ⚠
+
+## Retries
+If you wish to implement some retry logic atop of `clickhouse::Client` there are few simple rules to make you life easier:
+- If previous attempt threw an exception, then make sure to call `clickhouse::Client::ResetConnection()` before the next try.
+- For `clickhouse::Client::Insert()` you can reuse a block from previous try, no need to rebuild it from scratch.
+
+See https://github.com/ClickHouse/clickhouse-cpp/issues/184 for details.
+
+## Asynchronous inserts
+See https://clickhouse.com/docs/en/cloud/bestpractices/asynchronous-inserts for details.
+
+⚠ The asynchronous setting is different according to the clickhouse-server version. The under example with clickhouse-server version 24.8.4.13. ⚠
+
+> Our strong recommendation is to use async_insert=1,wait_for_async_insert=1 if using asynchronous inserts. Using wait_for_async_insert=0 is very risky because your INSERT client may not be aware if there are errors, and also can cause potential overload if your client continues to write quickly in a situation where the ClickHouse server needs to slow down the writes and create some backpressure in order to ensure reliability of the service.
+
+- Only use the SDK, do not need to change the clickhouse-server config. Asynchronous inserts only work if the data is sent as SQL text format. Here is the example.
+```cpp
+// You can specify the asynchronous insert settings by using the SETTINGS clause of insert queries
+clickhouse::Query query("INSERT INTO default.test SETTINGS async_insert=1,wait_for_async_insert=1,async_insert_busy_timeout_ms=5000,async_insert_use_adaptive_busy_timeout=0,async_insert_max_data_size=104857600 VALUES(10,10)");
+client.Execute(query);
+
+// Or by SetSetting
+clickhouse::Query query("INSERT INTO default.test VALUES(10,10)");
+query.SetSetting("async_insert", clickhouse::QuerySettingsField{ "1", 1 });
+query.SetSetting("wait_for_async_insert", clickhouse::QuerySettingsField{ "1", 1 }); // strong recommendation
+query.SetSetting("async_insert_busy_timeout_ms", clickhouse::QuerySettingsField{ "5000", 1 });
+query.SetSetting("async_insert_max_data_size", clickhouse::QuerySettingsField{ "104857600", 1 });
+query.SetSetting("async_insert_use_adaptive_busy_timeout", clickhouse::QuerySettingsField{ "0", 1 });
+client.Execute(query);
+
+// Not available case. The Insert interface actually use the native data format
+clickhouse::Block block;
+client.Insert("default.test", block);
+```
+- Change the clickhouse-server users.xml, enable asynchronous inserts (available for the native data format). Here is the example.
+```xml
+<profiles>
+        <!-- Default settings. -->
+        <default>
+            <async_insert>1</async_insert>
+            <wait_for_async_insert>1</wait_for_async_insert>
+            <async_insert_use_adaptive_busy_timeout>0</async_insert_use_adaptive_busy_timeout>
+            <async_insert_busy_timeout_ms>5000</async_insert_busy_timeout_ms>
+            <async_insert_max_data_size>104857600</async_insert_max_data_size>
+        </default>
+
+        <!-- Profile that allows only read queries. -->
+        <readonly>
+            <readonly>1</readonly>
+        </readonly>
+    </profiles>
+```
+- Enabling asynchronous inserts at the user level. Ensure your login account has the privileges about ALTER USER. Then you can use insert_account for asynchronous inserts.
+```sql
+ALTER USER insert_account SETTINGS async_insert=1,wait_for_async_insert=1,async_insert_use_adaptive_busy_timeout=0,async_insert_busy_timeout_ms=5000,async_insert_max_data_size=104857600
+```
+
+
diff --git a/lib/clickhouse-cpp/bench/CMakeLists.txt b/lib/clickhouse-cpp/bench/CMakeLists.txt
deleted file mode 100644
index ac99470..0000000
--- a/lib/clickhouse-cpp/bench/CMakeLists.txt
+++ /dev/null
@@ -1,8 +0,0 @@
-ADD_EXECUTABLE (bench
-    bench.cpp
-)
-
-TARGET_LINK_LIBRARIES (bench
-    clickhouse-cpp-lib
-    benchmark
-)
diff --git a/lib/clickhouse-cpp/bench/bench.cpp b/lib/clickhouse-cpp/bench/bench.cpp
deleted file mode 100644
index 9ca8ff7..0000000
--- a/lib/clickhouse-cpp/bench/bench.cpp
+++ /dev/null
@@ -1,34 +0,0 @@
-#include <benchmark/benchmark.h>
-
-#include <clickhouse/client.h>
-
-namespace clickhouse {
-
-Client g_client(ClientOptions()
-        .SetHost("localhost")
-        .SetPingBeforeQuery(false));
-
-static void SelectNumber(benchmark::State& state) {
-    while (state.KeepRunning()) {
-        g_client.Select("SELECT number, number, number FROM system.numbers LIMIT 1000",
-            [](const Block& block) { block.GetRowCount(); }
-        );
-    }
-}
-BENCHMARK(SelectNumber);
-
-static void SelectNumberMoreColumns(benchmark::State& state) {
-    // Mainly test performance on type name parsing.
-    while (state.KeepRunning()) {
-        g_client.Select("SELECT "
-                "number, number, number, number, number, number, number, number, number, number "
-                "FROM system.numbers LIMIT 100",
-            [](const Block& block) { block.GetRowCount(); }
-        );
-    }
-}
-BENCHMARK(SelectNumberMoreColumns);
-
-}
-
-BENCHMARK_MAIN();
diff --git a/lib/clickhouse-cpp/clickhouse/CMakeLists.txt b/lib/clickhouse-cpp/clickhouse/CMakeLists.txt
index d7a1de7..0ab1a48 100644
--- a/lib/clickhouse-cpp/clickhouse/CMakeLists.txt
+++ b/lib/clickhouse-cpp/clickhouse/CMakeLists.txt
@@ -1,33 +1,246 @@
-ADD_LIBRARY (clickhouse-cpp-lib
-    base/coded.cpp
+SET ( clickhouse-cpp-lib-src
     base/compressed.cpp
     base/input.cpp
     base/output.cpp
     base/platform.cpp
     base/socket.cpp
+    base/wire_format.cpp
+    base/endpoints_iterator.cpp
 
     columns/array.cpp
+    columns/column.cpp
     columns/date.cpp
+    columns/decimal.cpp
     columns/enum.cpp
     columns/factory.cpp
+    columns/geo.cpp
+    columns/ip4.cpp
+    columns/ip6.cpp
+    columns/lowcardinality.cpp
     columns/nullable.cpp
     columns/numeric.cpp
+    columns/map.cpp
     columns/string.cpp
     columns/tuple.cpp
+    columns/time.cpp
     columns/uuid.cpp
 
+    columns/itemview.cpp
+
     types/type_parser.cpp
     types/types.cpp
 
     block.cpp
     client.cpp
     query.cpp
+
+    # Headers
+    base/buffer.h
+    base/compressed.h
+    base/endpoints_iterator.h
+    base/input.h
+    base/open_telemetry.h
+    base/output.h
+    base/platform.h
+    base/projected_iterator.h
+    base/singleton.h
+    base/socket.h
+    base/sslsocket.h
+    base/string_utils.h
+    base/string_view.h
+    base/uuid.h
+    base/wire_format.h
+
+    columns/array.h
+    columns/column.h
+    columns/date.h
+    columns/decimal.h
+    columns/enum.h
+    columns/factory.h
+    columns/geo.h
+    columns/ip4.h
+    columns/ip6.h
+    columns/itemview.h
+    columns/lowcardinality.h
+    columns/lowcardinalityadaptor.h
+    columns/map.h
+    columns/nothing.h
+    columns/nullable.h
+    columns/numeric.h
+    columns/string.h
+    columns/time.h
+    columns/tuple.h
+    columns/utils.h
+    columns/uuid.h
+
+    types/type_parser.h
+    types/types.h
+
+    block.h
+    client.h
+    error_codes.h
+    exceptions.h
+    protocol.h
+    query.h
+    server_exception.h
 )
 
-SET_TARGET_PROPERTIES(clickhouse-cpp-lib
-    PROPERTIES LINKER_LANGUAGE CXX)
+if (MSVC)
+    add_definitions(-D_CRT_SECURE_NO_WARNINGS)
+    add_compile_options(/W4)
+    # remove in 3.0
+    add_compile_options(/wd4996)
+else()
+    set(cxx_extra_wall "-Wempty-body -Wconversion -Wreturn-type -Wparentheses -Wuninitialized -Wunreachable-code -Wunused-function -Wunused-value -Wunused-variable")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${cxx_extra_wall}")
+
+    if(CMAKE_CXX_COMPILER_ID MATCHES "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang")
+        # a little abnormal when clang check conversion
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${cxx_extra_wall} -Wno-conversion")
+    endif()
+endif()
 
+IF (WITH_OPENSSL)
+    LIST(APPEND clickhouse-cpp-lib-src base/sslsocket.cpp)
+ENDIF ()
+
+ADD_LIBRARY (clickhouse-cpp-lib ${clickhouse-cpp-lib-src}
+    version.h)
+SET_TARGET_PROPERTIES (clickhouse-cpp-lib
+    PROPERTIES
+    LINKER_LANGUAGE CXX
+    VERSION ${CLICKHOUSE_CPP_VERSION}
+)
 TARGET_LINK_LIBRARIES (clickhouse-cpp-lib
-    cityhash-lib
-    lz4-lib
+    absl::int128
+    cityhash::cityhash
+    lz4::lz4
+    zstd::zstd
+)
+TARGET_INCLUDE_DIRECTORIES (clickhouse-cpp-lib
+    PUBLIC ${PROJECT_SOURCE_DIR}
+)
+
+IF (NOT BUILD_SHARED_LIBS)
+    ADD_LIBRARY (clickhouse-cpp-lib-static ALIAS clickhouse-cpp-lib)
+ELSE ()
+    SET_TARGET_PROPERTIES (clickhouse-cpp-lib
+        PROPERTIES
+        SO_VERSION ${CLICKHOUSE_CPP_VERSION}
+        SO_VERSION ${CLICKHOUSE_CPP_VERSION_MAJOR}
+    )
+ENDIF ()
+
+
+IF (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" AND NOT DISABLE_CLANG_LIBC_WORKAROUND)
+    INCLUDE (CheckCXXSourceCompiles)
+
+    CHECK_CXX_SOURCE_COMPILES("#include <bits/c++config.h>\nint main() { return __GLIBCXX__ != 0; }"
+                              BUILDING_WITH_LIB_STDCXX)
+
+    IF (BUILDING_WITH_LIB_STDCXX)
+        # there is a problem with __builtin_mul_overflow call at link time
+        # the error looks like: ... undefined reference to `__muloti4' ...
+        # caused by clang bug https://bugs.llvm.org/show_bug.cgi?id=16404
+        # explicit linking to compiler-rt allows to workaround the problem
+        SET (CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} --rtlib=compiler-rt")
+
+        # some workaround for linking issues on linux:
+        # /usr/bin/ld: CMakeFiles/simple-test.dir/main.cpp.o: undefined reference to symbol '_Unwind_Resume@@GCC_3.0'
+        # /usr/bin/ld: /lib/x86_64-linux-gnu/libgcc_s.so.1: error adding symbols: DSO missing from command line
+        # FIXME: that workaround breaks clang build on mingw
+        TARGET_LINK_LIBRARIES (clickhouse-cpp-lib gcc_s)
+    ENDIF ()
+ENDIF ()
+
+
+INSTALL (TARGETS clickhouse-cpp-lib
+    ARCHIVE DESTINATION lib
+    LIBRARY DESTINATION lib
 )
+IF (NOT BUILD_SHARED_LIBS)
+    IF (NOT WITH_SYSTEM_CITYHASH)
+        INSTALL (TARGETS cityhash
+            ARCHIVE DESTINATION lib
+            LIBRARY DESTINATION lib
+        )
+    ENDIF()
+    IF (NOT WITH_SYSTEM_LZ4)
+        INSTALL (TARGETS lz4
+            ARCHIVE DESTINATION lib
+            LIBRARY DESTINATION lib
+        )
+    ENDIF()
+    IF (NOT WITH_SYSTEM_ZSTD)
+        INSTALL (TARGETS zstdstatic
+            ARCHIVE DESTINATION lib
+            LIBRARY DESTINATION lib
+        )
+    ENDIF()
+    IF (NOT WITH_SYSTEM_ABSEIL)
+        INSTALL (TARGETS absl_int128
+            ARCHIVE DESTINATION lib
+            LIBRARY DESTINATION lib
+        )
+    ENDIF()
+ENDIF()
+
+# general
+INSTALL(FILES block.h DESTINATION include/clickhouse/)
+INSTALL(FILES client.h DESTINATION include/clickhouse/)
+INSTALL(FILES error_codes.h DESTINATION include/clickhouse/)
+INSTALL(FILES exceptions.h DESTINATION include/clickhouse/)
+INSTALL(FILES server_exception.h DESTINATION include/clickhouse/)
+INSTALL(FILES protocol.h DESTINATION include/clickhouse/)
+INSTALL(FILES query.h DESTINATION include/clickhouse/)
+INSTALL(FILES version.h DESTINATION include/clickhouse/)
+
+# base
+INSTALL(FILES base/buffer.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/compressed.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/input.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/open_telemetry.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/output.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/platform.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/projected_iterator.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/singleton.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/socket.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/string_utils.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/string_view.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/uuid.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/wire_format.h DESTINATION include/clickhouse/base/)
+INSTALL(FILES base/endpoints_iterator.h DESTINATION include/clickhouse/base/)
+
+# columns
+INSTALL(FILES columns/array.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/column.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/date.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/decimal.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/enum.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/factory.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/geo.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/ip4.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/ip6.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/itemview.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/lowcardinality.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/nothing.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/nullable.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/numeric.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/map.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/string.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/time.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/tuple.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/utils.h DESTINATION include/clickhouse/columns/)
+INSTALL(FILES columns/uuid.h DESTINATION include/clickhouse/columns/)
+
+# types
+INSTALL(FILES types/type_parser.h DESTINATION include/clickhouse/types/)
+INSTALL(FILES types/types.h DESTINATION include/clickhouse/types/)
+
+IF (WITH_OPENSSL)
+    TARGET_LINK_LIBRARIES (clickhouse-cpp-lib OpenSSL::SSL)
+ENDIF ()
+
+IF (WIN32 OR MINGW)
+    TARGET_LINK_LIBRARIES (clickhouse-cpp-lib wsock32 ws2_32)
+ENDIF ()
diff --git a/lib/clickhouse-cpp/clickhouse/base/coded.cpp b/lib/clickhouse-cpp/clickhouse/base/coded.cpp
deleted file mode 100644
index 9e638ca..0000000
--- a/lib/clickhouse-cpp/clickhouse/base/coded.cpp
+++ /dev/null
@@ -1,100 +0,0 @@
-#include "coded.h"
-
-#include <memory.h>
-
-namespace clickhouse {
-
-static const int MAX_VARINT_BYTES = 10;
-
-CodedInputStream::CodedInputStream(ZeroCopyInput* input)
-    : input_(input)
-{
-}
-
-bool CodedInputStream::ReadRaw(void* buffer, size_t size) {
-    uint8_t* p = static_cast<uint8_t*>(buffer);
-
-    while (size > 0) {
-        const void* ptr;
-        size_t len = input_->Next(&ptr, size);
-
-        memcpy(p, ptr, len);
-
-        p += len;
-        size -= len;
-    }
-
-    return true;
-}
-
-bool CodedInputStream::Skip(size_t count) {
-    while (count > 0) {
-        const void* ptr;
-        size_t len = input_->Next(&ptr, count);
-
-        if (len == 0) {
-            return false;
-        }
-
-        count -= len;
-    }
-
-    return true;
-}
-
-bool CodedInputStream::ReadVarint64(uint64_t* value) {
-    *value = 0;
-
-    for (size_t i = 0; i < 9; ++i) {
-        uint8_t byte;
-
-        if (!input_->ReadByte(&byte)) {
-            return false;
-        } else {
-            *value |= (byte & 0x7F) << (7 * i);
-
-            if (!(byte & 0x80)) {
-                return true;
-            }
-        }
-    }
-
-    // TODO skip invalid
-    return false;
-}
-
-
-CodedOutputStream::CodedOutputStream(ZeroCopyOutput* output)
-    : output_(output)
-{
-}
-
-void CodedOutputStream::Flush() {
-    output_->Flush();
-}
-
-void CodedOutputStream::WriteRaw(const void* buffer, int size) {
-    output_->Write(buffer, size);
-}
-
-void CodedOutputStream::WriteVarint64(uint64_t value) {
-    uint8_t bytes[MAX_VARINT_BYTES];
-    int size = 0;
-
-    for (size_t i = 0; i < 9; ++i) {
-        uint8_t byte = value & 0x7F;
-        if (value > 0x7F)
-            byte |= 0x80;
-
-        bytes[size++] = byte;
-
-        value >>= 7;
-        if (!value) {
-            break;
-        }
-    }
-
-    WriteRaw(bytes, size);
-}
-
-}
diff --git a/lib/clickhouse-cpp/clickhouse/base/coded.h b/lib/clickhouse-cpp/clickhouse/base/coded.h
deleted file mode 100644
index a171ac5..0000000
--- a/lib/clickhouse-cpp/clickhouse/base/coded.h
+++ /dev/null
@@ -1,65 +0,0 @@
-#pragma once
-
-#include "input.h"
-#include "output.h"
-
-#include <string>
-
-namespace clickhouse {
-
-/**
- * Class which reads and decodes binary data which is composed of varint-
- * encoded integers and fixed-width pieces.
- */
-class CodedInputStream {
-public:
-    /// Create a CodedInputStream that reads from the given ZeroCopyInput.
-    explicit CodedInputStream(ZeroCopyInput* input);
-
-    // Read an unsigned integer with Varint encoding, truncating to 32 bits.
-    // Reading a 32-bit value is equivalent to reading a 64-bit one and casting
-    // it to uint32, but may be more efficient.
-    bool ReadVarint32(uint32_t* value);
-
-    // Read an unsigned integer with Varint encoding.
-    bool ReadVarint64(uint64_t* value);
-
-    // Read raw bytes, copying them into the given buffer.
-    bool ReadRaw(void* buffer, size_t size);
-
-    // Like ReadRaw, but reads into a string.
-    //
-    // Implementation Note:  ReadString() grows the string gradually as it
-    // reads in the data, rather than allocating the entire requested size
-    // upfront.  This prevents denial-of-service attacks in which a client
-    // could claim that a string is going to be MAX_INT bytes long in order to
-    // crash the server because it can't allocate this much space at once.
-    bool ReadString(std::string* buffer, int size);
-
-    // Skips a number of bytes.  Returns false if an underlying read error
-    // occurs.
-    bool Skip(size_t count);
-
-private:
-    ZeroCopyInput* input_;
-};
-
-
-class CodedOutputStream {
-public:
-    /// Create a CodedInputStream that writes to the given ZeroCopyOutput.
-    explicit CodedOutputStream(ZeroCopyOutput* output);
-
-    void Flush();
-
-    // Write raw bytes, copying them from the given buffer.
-    void WriteRaw(const void* buffer, int size);
-
-    /// Write an unsigned integer with Varint encoding.
-    void WriteVarint64(const uint64_t value);
-
-private:
-    ZeroCopyOutput* output_;
-};
-
-}
diff --git a/lib/clickhouse-cpp/clickhouse/base/compressed.cpp b/lib/clickhouse-cpp/clickhouse/base/compressed.cpp
index 33bc111..f30b5f8 100644
--- a/lib/clickhouse-cpp/clickhouse/base/compressed.cpp
+++ b/lib/clickhouse-cpp/clickhouse/base/compressed.cpp
@@ -1,24 +1,45 @@
 #include "compressed.h"
 #include "wire_format.h"
+#include "output.h"
+#include "clickhouse/exceptions.h"
 
-#include <cityhash/city.h>
-#include <lz4/lz4.h>
-
+#include <city.h>
+#include <lz4.h>
+#include <exception>
+#include <zstd.h>
+#include <stdexcept>
 #include <system_error>
 
-#define DBMS_MAX_COMPRESSED_SIZE    0x40000000ULL   // 1GB
+namespace {
+constexpr size_t HEADER_SIZE = 9;
+
+// see DB::CompressionMethodByte from src/Compression/CompressionInfo.h of ClickHouse project
+enum class CompressionMethodByte : uint8_t {
+    NONE = 0x02,
+    LZ4  = 0x82,
+    ZSTD = 0x90,
+};
+
+// Documentation says that compression is faster when output buffer is larger than LZ4_compressBound/ZSTD_compressBound estimation.
+constexpr size_t EXTRA_COMPRESS_BUFFER_SIZE = 4096;
+constexpr size_t DBMS_MAX_COMPRESSED_SIZE = 0x40000000ULL;   // 1GB
+}
 
 namespace clickhouse {
 
-CompressedInput::CompressedInput(CodedInputStream* input)
+CompressedInput::CompressedInput(InputStream* input)
     : input_(input)
 {
 }
 
 CompressedInput::~CompressedInput() {
     if (!mem_.Exhausted()) {
+#if __cplusplus < 201703L
         if (!std::uncaught_exception()) {
-            throw std::runtime_error("some data was not readed");
+#else
+        if (!std::uncaught_exceptions()) {
+#endif
+            throw CompressionError("some data was not read");
         }
     }
 }
@@ -39,56 +60,199 @@ bool CompressedInput::Decompress() {
     uint32_t original = 0;
     uint8_t method = 0;
 
-    if (!WireFormat::ReadFixed(input_, &hash)) {
+    if (!WireFormat::ReadFixed(*input_, &hash)) {
         return false;
     }
-    if (!WireFormat::ReadFixed(input_, &method)) {
+    if (!WireFormat::ReadFixed(*input_, &method)) {
         return false;
     }
 
-    if (method != 0x82) {
-        throw std::runtime_error("unsupported compression method " +
-                                 std::to_string(int(method)));
-    } else {
-        if (!WireFormat::ReadFixed(input_, &compressed)) {
-            return false;
-        }
-        if (!WireFormat::ReadFixed(input_, &original)) {
-            return false;
-        }
+    if (method != static_cast<uint8_t>(CompressionMethodByte::LZ4) && method != static_cast<uint8_t>(CompressionMethodByte::ZSTD)) {
+        throw CompressionError("unsupported compression method " + std::to_string((method)));
+    }
 
-        if (compressed > DBMS_MAX_COMPRESSED_SIZE) {
-            throw std::runtime_error("compressed data too big");
-        }
+    if (!WireFormat::ReadFixed(*input_, &compressed)) {
+        return false;
+    }
+    if (!WireFormat::ReadFixed(*input_, &original)) {
+        return false;
+    }
 
-        Buffer tmp(compressed);
+    if (compressed > DBMS_MAX_COMPRESSED_SIZE) {
+        throw CompressionError("compressed data too big");
+    }
 
-        // Заполнить заголовок сжатых данных.
-        {
-            BufferOutput out(&tmp);
-            out.Write(&method,     sizeof(method));
-            out.Write(&compressed, sizeof(compressed));
-            out.Write(&original,   sizeof(original));
+    Buffer tmp(compressed);
+
+    // Data header
+    {
+        BufferOutput out(&tmp);
+        out.Write(&method, sizeof(method));
+        out.Write(&compressed, sizeof(compressed));
+        out.Write(&original, sizeof(original));
+        out.Flush();
+    }
+
+    if (!WireFormat::ReadBytes(*input_, tmp.data() + HEADER_SIZE, compressed - HEADER_SIZE)) {
+        return false;
+    } else {
+        if (hash != CityHash128((const char*)tmp.data(), compressed)) {
+            throw CompressionError("data was corrupted");
         }
+    }
 
-        if (!WireFormat::ReadBytes(input_, tmp.data() + 9, compressed - 9)) {
-            return false;
+    data_ = Buffer(original);
+
+    switch (method) {
+    case static_cast<uint8_t>(CompressionMethodByte::LZ4): {
+        if (LZ4_decompress_safe((const char*)tmp.data() + HEADER_SIZE, (char*)data_.data(), static_cast<int>(compressed - HEADER_SIZE), original) < 0) {
+            throw CompressionError("can't decompress LZ4-encoded data");
         } else {
-            if (hash != CityHash128((const char*)tmp.data(), compressed)) {
-                throw std::runtime_error("data was corrupted");
-            }
+            mem_.Reset(data_.data(), original);
         }
+        return true;
+    }
 
-        data_ = Buffer(original);
+    case static_cast<uint8_t>(CompressionMethodByte::ZSTD): {
+        size_t res = ZSTD_decompress((char*)data_.data(), original, (const char*)tmp.data() + HEADER_SIZE, static_cast<int>(compressed - HEADER_SIZE));
 
-        if (LZ4_decompress_fast((const char*)tmp.data() + 9, (char*)data_.data(), original) < 0) {
-            throw std::runtime_error("can't decompress data");
+        if (ZSTD_isError(res)) {
+            throw CompressionError("can't decompress ZSTD-encoded data, ZSTD error: " + std::string(ZSTD_getErrorName(res)));
         } else {
             mem_.Reset(data_.data(), original);
         }
+        return true;
+    }
+
+    case static_cast<uint8_t>(CompressionMethodByte::NONE): {
+        throw CompressionError("compression method not defined" + std::to_string((method)));
+    }
+    default: {
+        throw CompressionError("Unknown or unsupported compression method " + std::to_string((method)));
+    }
     }
 
     return true;
 }
 
+
+CompressedOutput::CompressedOutput(OutputStream * destination, size_t max_compressed_chunk_size, CompressionMethod method)
+    : destination_(destination)
+    , max_compressed_chunk_size_(max_compressed_chunk_size)
+    , method_(method)
+{
+    PreallocateCompressBuffer(max_compressed_chunk_size);
+}
+
+CompressedOutput::~CompressedOutput() { }
+
+size_t CompressedOutput::DoWrite(const void* data, size_t len) {
+    const size_t original_len = len;
+    // what if len > max_compressed_chunk_size_ ?
+    const size_t max_chunk_size = max_compressed_chunk_size_ > 0 ? max_compressed_chunk_size_ : len;
+    if (max_chunk_size > max_compressed_chunk_size_) {
+        PreallocateCompressBuffer(len);
+    }
+
+    while (len > 0) {
+        auto to_compress = std::min(len, max_chunk_size);
+        Compress(data, to_compress);
+
+        len -= to_compress;
+        data = reinterpret_cast<const char*>(data) + to_compress;
+    }
+
+    return original_len - len;
+}
+
+void CompressedOutput::DoFlush() {
+    destination_->Flush();
+}
+
+void CompressedOutput::Compress(const void * data, size_t len) {
+    switch (method_) {  
+    case clickhouse::CompressionMethod::LZ4: {
+        const auto compressed_size = LZ4_compress_default(
+                (const char*)data,
+                (char*)compressed_buffer_.data() + HEADER_SIZE,
+                static_cast<int>(len),
+                static_cast<int>(compressed_buffer_.size() - HEADER_SIZE));
+        if (compressed_size <= 0)
+            throw CompressionError("Failed to compress chunk of " + std::to_string(len) + " bytes, "
+                    "LZ4 error: " + std::to_string(compressed_size));
+
+        {
+            auto header = compressed_buffer_.data();
+            WriteUnaligned(header, CompressionMethodByte::LZ4);
+            // Compressed data size with header
+            WriteUnaligned(header + 1, static_cast<uint32_t>(compressed_size + HEADER_SIZE));
+            // Original data size
+            WriteUnaligned(header + 5, static_cast<uint32_t>(len));
+        }
+
+        WireFormat::WriteFixed(*destination_, CityHash128((const char*)compressed_buffer_.data(), compressed_size + HEADER_SIZE));
+        WireFormat::WriteBytes(*destination_, compressed_buffer_.data(), compressed_size + HEADER_SIZE);
+        break;
+    }
+
+    case clickhouse::CompressionMethod::ZSTD: {
+        const size_t compressed_size = ZSTD_compress(
+                (char*)compressed_buffer_.data() + HEADER_SIZE,
+                static_cast<int>(compressed_buffer_.size() - HEADER_SIZE),
+                (const char*)data,
+                static_cast<int>(len),
+                ZSTD_fast);
+        if (ZSTD_isError(compressed_size))
+            throw CompressionError("Failed to compress chunk of " + std::to_string(len) + " bytes, "
+                    "ZSTD error: " + std::string(ZSTD_getErrorName(compressed_size)));
+
+        {
+            auto header = compressed_buffer_.data();
+            WriteUnaligned(header, CompressionMethodByte::ZSTD);
+            // Compressed data size with header
+            WriteUnaligned(header + 1, static_cast<uint32_t>(compressed_size + HEADER_SIZE));
+            // Original data size
+            WriteUnaligned(header + 5, static_cast<uint32_t>(len));
+        }
+
+        WireFormat::WriteFixed(*destination_, CityHash128((const char*)compressed_buffer_.data(), compressed_size + HEADER_SIZE));
+        WireFormat::WriteBytes(*destination_, compressed_buffer_.data(), compressed_size + HEADER_SIZE);
+        break;
+    }
+
+    case clickhouse::CompressionMethod::None: {
+        throw CompressionError("no compression defined");
+    }
+    }
+
+    destination_->Flush();
+}
+
+void CompressedOutput::PreallocateCompressBuffer(size_t input_size) {
+    switch (method_) {  
+    case clickhouse::CompressionMethod::LZ4: {
+        const auto estimated_compressed_buffer_size = LZ4_compressBound(static_cast<int>(input_size));
+        if (estimated_compressed_buffer_size <= 0)
+            throw CompressionError("Failed to estimate compressed buffer size, LZ4 error: " + std::to_string(estimated_compressed_buffer_size));
+
+        compressed_buffer_.resize(estimated_compressed_buffer_size + HEADER_SIZE + EXTRA_COMPRESS_BUFFER_SIZE);
+        break;
+    }
+
+    case clickhouse::CompressionMethod::ZSTD: {
+        const size_t estimated_compressed_buffer_size = ZSTD_compressBound(static_cast<int>(input_size));
+        if (ZSTD_isError(estimated_compressed_buffer_size))
+            throw CompressionError("Failed to estimate compressed buffer size, ZSTD error: " + std::string(ZSTD_getErrorName(estimated_compressed_buffer_size)));
+
+        compressed_buffer_.resize(estimated_compressed_buffer_size + HEADER_SIZE + EXTRA_COMPRESS_BUFFER_SIZE);
+        break;
+    }
+
+    case clickhouse::CompressionMethod::None: {
+        /// do nothing
+        break;
+    }
+    }
+}
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/base/compressed.h b/lib/clickhouse-cpp/clickhouse/base/compressed.h
index 8c1b461..f1e6a6d 100644
--- a/lib/clickhouse-cpp/clickhouse/base/compressed.h
+++ b/lib/clickhouse-cpp/clickhouse/base/compressed.h
@@ -1,13 +1,17 @@
 #pragma once
 
-#include "coded.h"
+#include "input.h"
+#include "output.h"
+#include "buffer.h"
+
+#include "clickhouse/client.h"
 
 namespace clickhouse {
 
 class CompressedInput : public ZeroCopyInput {
 public:
-     CompressedInput(CodedInputStream* input);
-    ~CompressedInput();
+    explicit CompressedInput(InputStream* input);
+    ~CompressedInput() override;
 
 protected:
     size_t DoNext(const void** ptr, size_t len) override;
@@ -15,10 +19,30 @@ class CompressedInput : public ZeroCopyInput {
     bool Decompress();
 
 private:
-    CodedInputStream* const input_;
+    InputStream* const input_;
 
     Buffer data_;
     ArrayInput mem_;
 };
 
+class CompressedOutput : public OutputStream {
+public:
+    explicit CompressedOutput(OutputStream* destination, size_t max_compressed_chunk_size = 0, CompressionMethod method = CompressionMethod::LZ4);
+    ~CompressedOutput() override;
+
+protected:
+    size_t DoWrite(const void* data, size_t len) override;
+    void DoFlush() override;
+
+private:
+    void Compress(const void * data, size_t len);
+    void PreallocateCompressBuffer(size_t input_size);
+
+private:
+    OutputStream * destination_;
+    const size_t max_compressed_chunk_size_;
+    Buffer compressed_buffer_;
+    CompressionMethod method_;
+};
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/base/endpoints_iterator.cpp b/lib/clickhouse-cpp/clickhouse/base/endpoints_iterator.cpp
new file mode 100644
index 0000000..30d3593
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/endpoints_iterator.cpp
@@ -0,0 +1,20 @@
+#include "endpoints_iterator.h"
+#include <clickhouse/client.h>
+
+namespace clickhouse {
+
+RoundRobinEndpointsIterator::RoundRobinEndpointsIterator(const std::vector<Endpoint>& _endpoints)
+   :  endpoints (_endpoints)
+   , current_index (endpoints.size() - 1ull)
+{
+}
+
+Endpoint RoundRobinEndpointsIterator::Next()
+{
+   current_index = (current_index + 1ull) % endpoints.size();
+   return endpoints[current_index];
+}
+
+RoundRobinEndpointsIterator::~RoundRobinEndpointsIterator() = default;
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/base/endpoints_iterator.h b/lib/clickhouse-cpp/clickhouse/base/endpoints_iterator.h
new file mode 100644
index 0000000..ba6a850
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/endpoints_iterator.h
@@ -0,0 +1,34 @@
+#pragma once
+
+#include "clickhouse/client.h"
+#include <vector>
+
+namespace clickhouse {
+
+struct ClientOptions;
+
+/**
+ * Base class for iterating through endpoints.
+*/
+class EndpointsIteratorBase
+{
+ public:
+   virtual ~EndpointsIteratorBase() = default;
+
+   virtual Endpoint Next() = 0;
+};
+
+class RoundRobinEndpointsIterator : public EndpointsIteratorBase
+{
+ public:
+    explicit RoundRobinEndpointsIterator(const std::vector<Endpoint>& opts);
+    Endpoint Next() override;
+
+    ~RoundRobinEndpointsIterator() override;
+
+ private:
+    const std::vector<Endpoint>& endpoints;
+    size_t current_index;
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/base/input.cpp b/lib/clickhouse-cpp/clickhouse/base/input.cpp
index f7b7ff6..e704fe5 100644
--- a/lib/clickhouse-cpp/clickhouse/base/input.cpp
+++ b/lib/clickhouse-cpp/clickhouse/base/input.cpp
@@ -5,6 +5,21 @@
 
 namespace clickhouse {
 
+bool ZeroCopyInput::Skip(size_t bytes) {
+    while (bytes > 0) {
+        const void* ptr;
+        size_t len = Next(&ptr, bytes);
+
+        if (len == 0) {
+            return false;
+        }
+
+        bytes -= len;
+    }
+
+    return true;
+}
+
 size_t ZeroCopyInput::DoRead(void* buf, size_t len) {
     const void* ptr;
     size_t result = DoNext(&ptr, len);
@@ -41,8 +56,8 @@ size_t ArrayInput::DoNext(const void** ptr, size_t len) {
 }
 
 
-BufferedInput::BufferedInput(InputStream* slave, size_t buflen)
-    : slave_(slave)
+BufferedInput::BufferedInput(std::unique_ptr<InputStream> source, size_t buflen)
+    : source_(std::move(source))
     , array_input_(nullptr, 0)
     , buffer_(buflen)
 {
@@ -57,7 +72,7 @@ void BufferedInput::Reset() {
 size_t BufferedInput::DoNext(const void** ptr, size_t len)  {
     if (array_input_.Exhausted()) {
         array_input_.Reset(
-            buffer_.data(), slave_->Read(buffer_.data(), buffer_.size())
+            buffer_.data(), source_->Read(buffer_.data(), buffer_.size())
         );
     }
 
@@ -67,11 +82,11 @@ size_t BufferedInput::DoNext(const void** ptr, size_t len)  {
 size_t BufferedInput::DoRead(void* buf, size_t len) {
     if (array_input_.Exhausted()) {
         if (len > buffer_.size() / 2) {
-            return slave_->Read(buf, len);
+            return source_->Read(buf, len);
         }
 
         array_input_.Reset(
-            buffer_.data(), slave_->Read(buffer_.data(), buffer_.size())
+            buffer_.data(), source_->Read(buffer_.data(), buffer_.size())
         );
     }
 
diff --git a/lib/clickhouse-cpp/clickhouse/base/input.h b/lib/clickhouse-cpp/clickhouse/base/input.h
index 052fab8..a8885b3 100644
--- a/lib/clickhouse-cpp/clickhouse/base/input.h
+++ b/lib/clickhouse-cpp/clickhouse/base/input.h
@@ -3,6 +3,7 @@
 #include <cstddef>
 #include <cstdint>
 #include <vector>
+#include <memory>
 
 namespace clickhouse {
 
@@ -21,6 +22,9 @@ class InputStream {
         return DoRead(buf, len);
     }
 
+    // Skips a number of bytes.  Returns false if an underlying read error occurs.
+    virtual bool Skip(size_t bytes) = 0;
+
 protected:
     virtual size_t DoRead(void* buf, size_t len) = 0;
 };
@@ -32,6 +36,8 @@ class ZeroCopyInput : public InputStream {
         return DoNext(buf, len);
     }
 
+    bool Skip(size_t bytes) override;
+
 protected:
     virtual size_t DoNext(const void** ptr, size_t len) = 0;
 
@@ -79,7 +85,7 @@ class ArrayInput : public ZeroCopyInput {
 
 class BufferedInput : public ZeroCopyInput {
 public:
-     BufferedInput(InputStream* slave, size_t buflen = 8192);
+    BufferedInput(std::unique_ptr<InputStream> source, size_t buflen = 8192);
     ~BufferedInput() override;
 
     void Reset();
@@ -89,7 +95,7 @@ class BufferedInput : public ZeroCopyInput {
     size_t DoNext(const void** ptr, size_t len) override;
 
 private:
-    InputStream* const slave_;
+    std::unique_ptr<InputStream> const source_;
     ArrayInput array_input_;
     std::vector<uint8_t> buffer_;
 };
diff --git a/lib/clickhouse-cpp/clickhouse/base/open_telemetry.h b/lib/clickhouse-cpp/clickhouse/base/open_telemetry.h
new file mode 100644
index 0000000..34f3311
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/open_telemetry.h
@@ -0,0 +1,23 @@
+#pragma once
+
+#include "uuid.h"
+
+#include <string>
+
+namespace clickhouse::open_telemetry {
+
+/// See https://www.w3.org/TR/trace-context/ for trace_flags definition
+enum TraceFlags : uint8_t {
+    TRACE_FLAG_NONE    = 0,
+    TRACE_FLAG_SAMPLED = 1,
+};
+
+/// The runtime info we need to create new OpenTelemetry spans.
+struct TracingContext {
+    UUID trace_id{};
+    uint64_t span_id = 0;
+    std::string tracestate;
+    uint8_t trace_flags = TRACE_FLAG_NONE;
+};
+
+}  // namespace clickhouse::open_telemetry
diff --git a/lib/clickhouse-cpp/clickhouse/base/output.cpp b/lib/clickhouse-cpp/clickhouse/base/output.cpp
index ee8e147..1b98272 100644
--- a/lib/clickhouse-cpp/clickhouse/base/output.cpp
+++ b/lib/clickhouse-cpp/clickhouse/base/output.cpp
@@ -6,7 +6,8 @@
 
 namespace clickhouse {
 
-void ZeroCopyOutput::DoWrite(const void* data, size_t len) {
+size_t ZeroCopyOutput::DoWrite(const void* data, size_t len) {
+    const size_t original_len = len;
     while (len > 0) {
         void* ptr;
         size_t result = DoNext(&ptr, len);
@@ -19,12 +20,15 @@ void ZeroCopyOutput::DoWrite(const void* data, size_t len) {
             break;
         }
     }
+
+    return original_len - len;
 }
 
 
 ArrayOutput::ArrayOutput(void* buf, size_t len)
     : buf_(static_cast<uint8_t*>(buf))
     , end_(buf_ + len)
+    , buffer_size_(len)
 {
 }
 
@@ -62,16 +66,14 @@ size_t BufferOutput::DoNext(void** data, size_t len) {
 }
 
 
-BufferedOutput::BufferedOutput(OutputStream* slave, size_t buflen)
-    : slave_(slave)
+BufferedOutput::BufferedOutput(std::unique_ptr<OutputStream> destination, size_t buflen)
+    : destination_(std::move(destination))
     , buffer_(buflen)
     , array_output_(buffer_.data(), buflen)
 {
 }
 
-BufferedOutput::~BufferedOutput() {
-    Flush();
-}
+BufferedOutput::~BufferedOutput() { }
 
 void BufferedOutput::Reset() {
     array_output_.Reset(buffer_.data(), buffer_.size());
@@ -79,8 +81,15 @@ void BufferedOutput::Reset() {
 
 void BufferedOutput::DoFlush() {
     if (array_output_.Data() != buffer_.data()) {
-        slave_->Write(buffer_.data(), array_output_.Data() - buffer_.data());
-        slave_->Flush();
+        size_t len = array_output_.Data() - buffer_.data();
+        const uint8_t* buf = buffer_.data();
+        while (len > 0) {
+            const size_t written = destination_->Write(buf, len);
+            buf += written;
+            len -= written;
+        }
+
+        destination_->Flush();
 
         array_output_.Reset(buffer_.data(), buffer_.size());
     }
@@ -95,17 +104,16 @@ size_t BufferedOutput::DoNext(void** data, size_t len) {
 
 }
 
-void BufferedOutput::DoWrite(const void* data, size_t len) {
+size_t BufferedOutput::DoWrite(const void* data, size_t len) {
     if (array_output_.Avail() < len) {
         Flush();
 
         if (len > buffer_.size() / 2) {
-            slave_->Write(data, len);
-            return;
+            return destination_->Write(data, len);
         }
     }
 
-    array_output_.Write(data, len);
+    return array_output_.Write(data, len);
 }
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/base/output.h b/lib/clickhouse-cpp/clickhouse/base/output.h
index e53aadf..bb804ce 100644
--- a/lib/clickhouse-cpp/clickhouse/base/output.h
+++ b/lib/clickhouse-cpp/clickhouse/base/output.h
@@ -6,6 +6,7 @@
 #include <cstdint>
 #include <vector>
 #include <memory.h>
+#include <memory>
 
 namespace clickhouse {
 
@@ -18,14 +19,14 @@ class OutputStream {
         DoFlush();
     }
 
-    inline void Write(const void* data, size_t len) {
-        DoWrite(data, len);
+    inline size_t Write(const void* data, size_t len) {
+        return DoWrite(data, len);
     }
 
 protected:
     virtual void DoFlush() { }
 
-    virtual void DoWrite(const void* data, size_t len) = 0;
+    virtual size_t DoWrite(const void* data, size_t len) = 0;
 };
 
 
@@ -41,7 +42,7 @@ class ZeroCopyOutput : public OutputStream {
     // be written to the output.
     virtual size_t DoNext(void** data, size_t len) = 0;
 
-    void DoWrite(const void* data, size_t len) override;
+    size_t DoWrite(const void* data, size_t len) override;
 };
 
 
@@ -72,6 +73,12 @@ class ArrayOutput : public ZeroCopyOutput {
     inline void Reset(void* buf, size_t len) noexcept {
         buf_ = static_cast<uint8_t*>(buf);
         end_ = buf_ + len;
+        buffer_size_ = len;
+    }
+
+    /// Number of bytes written to the buffer.
+    inline size_t Size() const noexcept {
+        return buffer_size_ - Avail();
     }
 
 protected:
@@ -80,16 +87,19 @@ class ArrayOutput : public ZeroCopyOutput {
 private:
     uint8_t* buf_;
     uint8_t* end_;
+    size_t buffer_size_;
 };
 
 
 /**
- * A ZeroCopyOutput stream backed by an vector of bytes.
+ * A ZeroCopyOutput stream backed by a vector.
+ *
+ * Doesn't Flush() in destructor, client must ensure to do it manually at some point.
  */
 class BufferOutput : public ZeroCopyOutput {
 public:
      BufferOutput(Buffer* buf);
-    ~BufferOutput();
+    ~BufferOutput() override;
 
 protected:
     size_t DoNext(void** data, size_t len) override;
@@ -99,10 +109,16 @@ class BufferOutput : public ZeroCopyOutput {
     size_t pos_;
 };
 
-
+/** BufferedOutput writes data to internal buffer first.
+ *
+ *  Any data goes to underlying stream only if internal buffer is full
+ *  or when client invokes Flush() on this.
+ *
+ * Doesn't Flush() in destructor, client must ensure to do it manually at some point.
+ */
 class BufferedOutput : public ZeroCopyOutput {
 public:
-     BufferedOutput(OutputStream* slave, size_t buflen = 8192);
+    explicit BufferedOutput(std::unique_ptr<OutputStream> destination, size_t buflen = 8192);
     ~BufferedOutput() override;
 
     void Reset();
@@ -110,10 +126,10 @@ class BufferedOutput : public ZeroCopyOutput {
 protected:
     void DoFlush() override;
     size_t DoNext(void** data, size_t len) override;
-    void DoWrite(const void* data, size_t len) override;
+    size_t DoWrite(const void* data, size_t len) override;
 
 private:
-    OutputStream* const slave_;
+    std::unique_ptr<OutputStream> const destination_;
     Buffer buffer_;
     ArrayOutput array_output_;
 };
diff --git a/lib/clickhouse-cpp/clickhouse/base/platform.h b/lib/clickhouse-cpp/clickhouse/base/platform.h
index e8bf4d1..f6d896b 100644
--- a/lib/clickhouse-cpp/clickhouse/base/platform.h
+++ b/lib/clickhouse-cpp/clickhouse/base/platform.h
@@ -13,9 +13,14 @@
 
 #if defined(_win32_) || defined(_win64_)
 #   define _win_
+#   if !defined(_WIN32_WINNT) || (_WIN32_WINNT < 0x0600)
+#      undef _WIN32_WINNT
+#      define _WIN32_WINNT 0x0600 // The WSAPoll function is defined on Windows Vista and later.
+#   endif
+#   define WIN32_LEAN_AND_MEAN 1  // don't include too much header automatically
 #endif
 
-#if defined(_linux_)
+#if defined(_linux_) || defined (_darwin_)
 #   define _unix_
 #endif
 
diff --git a/lib/clickhouse-cpp/clickhouse/base/projected_iterator.h b/lib/clickhouse-cpp/clickhouse/base/projected_iterator.h
new file mode 100644
index 0000000..ca5aecf
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/projected_iterator.h
@@ -0,0 +1,55 @@
+#pragma once
+
+#include <iterator>
+#include <type_traits>
+#include <utility>
+
+namespace clickhouse {
+
+template <typename UnaryFunction, typename Iterator, typename Reference = decltype(std::declval<UnaryFunction>()(std::declval<Iterator>())),
+          typename Value = std::decay_t<Reference>>
+class ProjectedIterator {
+public:
+    using value_type        = Value;
+    using reference         = Reference;
+    using pointer           = Reference;
+    using difference_type   = typename std::iterator_traits<Iterator>::difference_type;
+    using iterator_category = typename std::iterator_traits<Iterator>::iterator_category;
+
+    ProjectedIterator() = default;
+
+    inline ProjectedIterator(Iterator const& iterator, UnaryFunction functor)
+        : iterator_(iterator)
+        , functor_(std::move(functor)) {
+    }
+
+    inline UnaryFunction functor() const { return functor; }
+
+    inline Iterator const& base() const { return iterator_; }
+
+    inline reference operator*() const { return functor_(iterator_); }
+
+    inline ProjectedIterator& operator++() {
+        ++iterator_;
+        return *this;
+    }
+
+    inline ProjectedIterator& operator--() {
+        --iterator_;
+        return *this;
+    }
+
+    inline bool operator==(const ProjectedIterator& other) const {
+        return this->iterator_ == other.iterator_;
+    }
+
+    inline bool operator!=(const ProjectedIterator& other) const {
+        return !(*this == other);
+    }
+
+private:
+    Iterator iterator_;
+    UnaryFunction functor_;
+};
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/base/socket.cpp b/lib/clickhouse-cpp/clickhouse/base/socket.cpp
index 927f193..3bb1aa5 100644
--- a/lib/clickhouse-cpp/clickhouse/base/socket.cpp
+++ b/lib/clickhouse-cpp/clickhouse/base/socket.cpp
@@ -1,48 +1,270 @@
 #include "socket.h"
 #include "singleton.h"
+#include "../client.h"
 
 #include <assert.h>
 #include <stdexcept>
 #include <system_error>
 #include <unordered_set>
 #include <memory.h>
+#include <thread>
 
 #if !defined(_win_)
 #   include <errno.h>
+#   include <fcntl.h>
 #   include <netdb.h>
+#   include <netinet/tcp.h>
 #   include <signal.h>
 #   include <unistd.h>
 #endif
 
 namespace clickhouse {
+
+#if defined(_win_)
+char const* windowsErrorCategory::name() const noexcept {
+    return "WindowsSocketError";
+}
+
+std::string windowsErrorCategory::message(int c) const {
+    char error[UINT8_MAX];
+    auto len = FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM, nullptr, static_cast<DWORD>(c), 0, error, sizeof(error), nullptr);
+    if (len == 0) {
+        return "unknown";
+    }
+    while (len && (error[len - 1] == '\r' || error[len - 1] == '\n')) {
+        --len;
+    }
+    return std::string(error, len);
+}
+
+windowsErrorCategory const& windowsErrorCategory::category() {
+    static windowsErrorCategory c;
+    return c;
+}
+#endif
+
+#if defined(_unix_)
+char const* getaddrinfoErrorCategory::name() const noexcept {
+    return "getaddrinfoError";
+}
+
+std::string getaddrinfoErrorCategory::message(int c) const {
+    return gai_strerror(c);
+}
+
+getaddrinfoErrorCategory const& getaddrinfoErrorCategory::category() {
+    static getaddrinfoErrorCategory c;
+    return c;
+}
+#endif
+
 namespace {
 
-    class LocalNames : public std::unordered_set<std::string> {
-    public:
-        LocalNames() {
-            emplace("localhost");
-            emplace("localhost.localdomain");
-            emplace("localhost6");
-            emplace("localhost6.localdomain6");
-            emplace("::1");
-            emplace("127.0.0.1");
+class LocalNames : public std::unordered_set<std::string> {
+public:
+    LocalNames() {
+        emplace("localhost");
+        emplace("localhost.localdomain");
+        emplace("localhost6");
+        emplace("localhost6.localdomain6");
+        emplace("::1");
+        emplace("127.0.0.1");
+    }
+
+    inline bool IsLocalName(const std::string& name) const noexcept {
+        return find(name) != end();
+    }
+};
+
+inline int getSocketErrorCode() {
+#if defined(_win_)
+    return WSAGetLastError();
+#else
+    return errno;
+#endif
+}
+
+const std::error_category& getErrorCategory() noexcept {
+#if defined(_win_)
+    return windowsErrorCategory::category();
+#else
+    return std::system_category();
+#endif
+}
+
+void SetNonBlock(SOCKET fd, bool value) {
+#if defined(_unix_) || defined(__CYGWIN__)
+    int flags;
+    int ret;
+    #if defined(O_NONBLOCK)
+        if ((flags = fcntl(fd, F_GETFL, 0)) == -1)
+            flags = 0;
+        if (value) {
+            flags |= O_NONBLOCK;
+        } else {
+            flags &= ~O_NONBLOCK;
         }
+        ret = fcntl(fd, F_SETFL, flags);
+    #else
+        flags = value;
+        return ioctl(fd, FIOBIO, &flags);
+    #endif
+    if (ret == -1) {
+        throw std::system_error(getSocketErrorCode(), getErrorCategory(), "fail to set nonblocking mode");
+    }
+#elif defined(_win_)
+    unsigned long inbuf = value;
+    unsigned long outbuf = 0;
+    DWORD written = 0;
+
+    if (!inbuf) {
+        WSAEventSelect(fd, nullptr, 0);
+    }
+
+    if (WSAIoctl(fd, FIONBIO, &inbuf, sizeof(inbuf), &outbuf, sizeof(outbuf), &written, 0, 0) == SOCKET_ERROR) {
+        throw std::system_error(getSocketErrorCode(), getErrorCategory(), "fail to set nonblocking mode");
+    }
+#endif
+}
+
+void SetTimeout(SOCKET fd, const SocketTimeoutParams& timeout_params) {
+#if defined(_unix_)
+    timeval recv_timeout{ static_cast<time_t>(timeout_params.recv_timeout.count() / 1000), static_cast<suseconds_t>(timeout_params.recv_timeout.count() % 1000 * 1000) };
+    auto recv_ret = setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, &recv_timeout, sizeof(recv_timeout));
 
-        inline bool IsLocalName(const std::string& name) const noexcept {
-            return find(name) != end();
+    timeval send_timeout{ static_cast<time_t>(timeout_params.send_timeout.count() / 1000), static_cast<suseconds_t>(timeout_params.send_timeout.count() % 1000 * 1000) };
+    auto send_ret = setsockopt(fd, SOL_SOCKET, SO_SNDTIMEO, &send_timeout, sizeof(send_timeout));
+
+    if (recv_ret == -1 || send_ret == -1) {
+        throw std::system_error(getSocketErrorCode(), getErrorCategory(), "fail to set socket timeout");
+    }
+#else
+    DWORD recv_timeout = static_cast<DWORD>(timeout_params.recv_timeout.count());
+    auto recv_ret = setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, (const char*)&recv_timeout, sizeof(DWORD));
+   
+    DWORD send_timeout = static_cast<DWORD>(timeout_params.send_timeout.count());
+    auto send_ret = setsockopt(fd, SOL_SOCKET, SO_SNDTIMEO, (const char*)&send_timeout, sizeof(DWORD));
+    
+    if (recv_ret == SOCKET_ERROR || send_ret == SOCKET_ERROR) {
+        throw std::system_error(getSocketErrorCode(), getErrorCategory(), "fail to set socket timeout");
+    }
+#endif
+};
+
+ssize_t Poll(struct pollfd* fds, int nfds, int timeout) noexcept {
+#if defined(_win_)
+    return WSAPoll(fds, nfds, timeout);
+#else
+    return poll(fds, nfds, timeout);
+#endif
+}
+
+#ifndef INVALID_SOCKET
+const SOCKET INVALID_SOCKET = -1;
+#endif
+
+void CloseSocket(SOCKET socket) {
+    if (socket == INVALID_SOCKET)
+        return;
+
+#if defined(_win_)
+    closesocket(socket);
+#else
+    close(socket);
+#endif
+}
+
+struct SocketRAIIWrapper {
+    SOCKET socket = INVALID_SOCKET;
+
+    ~SocketRAIIWrapper() {
+        CloseSocket(socket);
+    }
+
+    SOCKET operator*() const {
+        return socket;
+    }
+
+    SOCKET release() {
+        auto result = socket;
+        socket = INVALID_SOCKET;
+
+        return result;
+    }
+};
+
+SOCKET SocketConnect(const NetworkAddress& addr, const SocketTimeoutParams& timeout_params) {
+    int last_err = 0;
+    for (auto res = addr.Info(); res != nullptr; res = res->ai_next) {
+        SocketRAIIWrapper s{socket(res->ai_family, res->ai_socktype, res->ai_protocol)};
+
+        if (*s == INVALID_SOCKET) {
+            continue;
         }
-    };
+
+        SetNonBlock(*s, true);
+        SetTimeout(*s, timeout_params);
+
+        if (connect(*s, res->ai_addr, (int)res->ai_addrlen) != 0) {
+            int err = getSocketErrorCode();
+            if (
+                err == EINPROGRESS || err == EAGAIN || err == EWOULDBLOCK
+#if defined(_win_)
+                || err == WSAEWOULDBLOCK || err == WSAEINPROGRESS
+#endif
+            ) {
+                pollfd fd;
+                fd.fd = *s;
+                fd.events = POLLOUT;
+                fd.revents = 0;
+                ssize_t rval = Poll(&fd, 1, static_cast<int>(timeout_params.connect_timeout.count()));
+
+                if (rval == -1) {
+                    throw std::system_error(getSocketErrorCode(), getErrorCategory(), "fail to connect");
+                }
+                if (rval == 0) {
+#if defined(_win_)
+                    last_err = WSAETIMEDOUT;
+#else
+                    last_err = ETIMEDOUT;
+#endif
+                }
+                if (rval > 0) {
+                    socklen_t len = sizeof(err);
+                    getsockopt(*s, SOL_SOCKET, SO_ERROR, (char*)&err, &len);
+
+                    if (!err) {
+                        SetNonBlock(*s, false);
+                        return s.release();
+                    }
+                   last_err = err;
+                }
+            }
+        } else {
+            SetNonBlock(*s, false);
+            return s.release();
+        }
+    }
+    if (last_err > 0) {
+        throw std::system_error(last_err, getErrorCategory(), "fail to connect");
+    }
+    throw std::system_error(getSocketErrorCode(), getErrorCategory(), "fail to connect");
 }
 
+} // namespace
+
 NetworkAddress::NetworkAddress(const std::string& host, const std::string& port)
-    : info_(nullptr)
+    : host_(host)
+    , info_(nullptr)
 {
     struct addrinfo hints;
     memset(&hints, 0, sizeof(hints));
 
     hints.ai_family = PF_UNSPEC;
     hints.ai_socktype = SOCK_STREAM;
-
+    // using AI_ADDRCONFIG on windows will cause getaddrinfo to return WSAHOST_NOT_FOUND
+    // for more information, see https://github.com/ClickHouse/clickhouse-cpp/issues/195
+#if defined(_unix_) 
     if (!Singleton<LocalNames>()->IsLocalName(host)) {
         // https://linux.die.net/man/3/getaddrinfo
         // If hints.ai_flags includes the AI_ADDRCONFIG flag,
@@ -54,11 +276,18 @@ NetworkAddress::NetworkAddress(const std::string& host, const std::string& port)
         // as valid as a configured address.
         hints.ai_flags |= AI_ADDRCONFIG;
     }
+#endif
 
     const int error = getaddrinfo(host.c_str(), port.c_str(), &hints, &info_);
 
+#if defined(_unix_)
+    if (error && error != EAI_SYSTEM) {
+        throw std::system_error(error, getaddrinfoErrorCategory::category());
+    }
+#endif
+
     if (error) {
-        throw std::system_error(errno, std::system_category());
+        throw std::system_error(getSocketErrorCode(), getErrorCategory());
     }
 }
 
@@ -72,55 +301,119 @@ const struct addrinfo* NetworkAddress::Info() const {
     return info_;
 }
 
-
-SocketHolder::SocketHolder()
-    : handle_(-1)
-{
+const std::string & NetworkAddress::Host() const {
+    return host_;
 }
 
-SocketHolder::SocketHolder(SOCKET s)
-    : handle_(s)
-{
+
+SocketBase::~SocketBase() = default;
+
+
+SocketFactory::~SocketFactory() = default;
+
+void SocketFactory::sleepFor(const std::chrono::milliseconds& duration) {
+    std::this_thread::sleep_for(duration);
 }
 
-SocketHolder::SocketHolder(SocketHolder&& other)
+
+Socket::Socket(const NetworkAddress& addr, const SocketTimeoutParams& timeout_params)
+    : handle_(SocketConnect(addr, timeout_params))
+{}
+
+Socket::Socket(const NetworkAddress & addr)
+    : handle_(SocketConnect(addr, SocketTimeoutParams{}))
+{}
+
+Socket::Socket(Socket&& other) noexcept
     : handle_(other.handle_)
 {
-    other.handle_ = -1;
+    other.handle_ = INVALID_SOCKET;
 }
 
-SocketHolder::~SocketHolder() {
+Socket& Socket::operator=(Socket&& other) noexcept {
+    if (this != &other) {
+        Close();
+
+        handle_ = other.handle_;
+        other.handle_ = INVALID_SOCKET;
+    }
+
+    return *this;
+}
+
+Socket::~Socket() {
     Close();
 }
 
-void SocketHolder::Close() noexcept {
-    if (handle_ != -1) {
-#if defined(_win_)
-        closesocket(handle_);
+void Socket::Close() {
+    CloseSocket(handle_);
+    handle_ = INVALID_SOCKET;
+}
+
+void Socket::SetTcpKeepAlive(int idle, int intvl, int cnt) noexcept {
+    int val = 1;
+
+#if defined(_unix_)
+    setsockopt(handle_, SOL_SOCKET, SO_KEEPALIVE, &val, sizeof(val));
+#   if defined(_linux_)
+        setsockopt(handle_, IPPROTO_TCP, TCP_KEEPIDLE, &idle, sizeof(idle));
+#   elif defined(_darwin_)
+        setsockopt(handle_, IPPROTO_TCP, TCP_KEEPALIVE, &idle, sizeof(idle));
+#   else
+#       error "platform is not supported"
+#   endif
+    setsockopt(handle_, IPPROTO_TCP, TCP_KEEPINTVL, &intvl, sizeof(intvl));
+    setsockopt(handle_, IPPROTO_TCP, TCP_KEEPCNT, &cnt, sizeof(cnt));
 #else
-        close(handle_);
+    setsockopt(handle_, SOL_SOCKET, SO_KEEPALIVE, (const char*)&val, sizeof(val));
+    std::ignore = idle = intvl = cnt;
 #endif
-        handle_ = -1;
-    }
 }
 
-bool SocketHolder::Closed() const noexcept {
-    return handle_ == -1;
+void Socket::SetTcpNoDelay(bool nodelay) noexcept {
+    int val = nodelay;
+#if defined(_unix_)
+    setsockopt(handle_, IPPROTO_TCP, TCP_NODELAY, &val, sizeof(val));
+#else
+    setsockopt(handle_, IPPROTO_TCP, TCP_NODELAY, (const char*)&val, sizeof(val));
+#endif
 }
 
-SocketHolder& SocketHolder::operator = (SocketHolder&& other) noexcept {
-    if (this != &other) {
-        Close();
+std::unique_ptr<InputStream> Socket::makeInputStream() const {
+    return std::make_unique<SocketInput>(handle_);
+}
 
-        handle_ = other.handle_;
-        other.handle_ = -1;
-    }
+std::unique_ptr<OutputStream> Socket::makeOutputStream() const {
+    return std::make_unique<SocketOutput>(handle_);
+}
 
-    return *this;
+
+NonSecureSocketFactory::~NonSecureSocketFactory()  {}
+
+std::unique_ptr<SocketBase> NonSecureSocketFactory::connect(const ClientOptions &opts, const Endpoint& endpoint) {
+
+    const auto address = NetworkAddress(endpoint.host, std::to_string(endpoint.port));
+    auto socket = doConnect(address, opts);
+    setSocketOptions(*socket, opts);
+
+    return socket;
 }
 
-SocketHolder::operator SOCKET () const noexcept {
-    return handle_;
+std::unique_ptr<Socket> NonSecureSocketFactory::doConnect(const NetworkAddress& address, const ClientOptions& opts) {
+    SocketTimeoutParams timeout_params { opts.connection_connect_timeout, opts.connection_recv_timeout, opts.connection_send_timeout };
+    return std::make_unique<Socket>(address, timeout_params);
+}
+
+void NonSecureSocketFactory::setSocketOptions(Socket &socket, const ClientOptions &opts) {
+    if (opts.tcp_keepalive) {
+        socket.SetTcpKeepAlive(
+                static_cast<int>(opts.tcp_keepalive_idle.count()),
+                static_cast<int>(opts.tcp_keepalive_intvl.count()),
+                static_cast<int>(opts.tcp_keepalive_cnt));
+    }
+    if (opts.tcp_nodelay) {
+        socket.SetTcpNoDelay(opts.tcp_nodelay);
+    }
 }
 
 
@@ -139,14 +432,14 @@ size_t SocketInput::DoRead(void* buf, size_t len) {
     }
 
     if (ret == 0) {
-        throw std::system_error(
-            errno, std::system_category(), "closed"
-        );
+        throw std::system_error(getSocketErrorCode(), getErrorCategory(), "closed");
     }
 
-    throw std::system_error(
-        errno, std::system_category(), "can't receive string data"
-    );
+    throw std::system_error(getSocketErrorCode(), getErrorCategory(), "can't receive string data");
+}
+
+bool SocketInput::Skip(size_t /*bytes*/) {
+    return false;
 }
 
 
@@ -157,18 +450,19 @@ SocketOutput::SocketOutput(SOCKET s)
 
 SocketOutput::~SocketOutput() = default;
 
-void SocketOutput::DoWrite(const void* data, size_t len) {
+size_t SocketOutput::DoWrite(const void* data, size_t len) {
 #if defined (_linux_)
     static const int flags = MSG_NOSIGNAL;
 #else
     static const int flags = 0;
 #endif
 
-    if (::send(s_, (const char*)data, len, flags) != (int)len) {
-        throw std::system_error(
-            errno, std::system_category(), "fail to send data"
-        );
+    const ssize_t ret = ::send(s_, (const char*)data, (int)len, flags);
+    if (ret < 0) {
+        throw std::system_error(getSocketErrorCode(), getErrorCategory(), "fail to send " + std::to_string(len) + " bytes of data");
     }
+
+    return (size_t)ret;
 }
 
 
@@ -192,53 +486,4 @@ NetrworkInitializer::NetrworkInitializer() {
     (void)Singleton<NetrworkInitializerImpl>();
 }
 
-
-SOCKET SocketConnect(const NetworkAddress& addr) {
-    for (auto res = addr.Info(); res != nullptr; res = res->ai_next) {
-        SOCKET s(socket(res->ai_family, res->ai_socktype, res->ai_protocol));
-
-        if (s == -1) {
-            continue;
-        }
-
-        if (connect(s, res->ai_addr, (int)res->ai_addrlen)) {
-            if (errno == EINPROGRESS ||
-                errno == EAGAIN ||
-                errno == EWOULDBLOCK)
-            {
-                pollfd fd;
-                fd.fd = s;
-                fd.events = POLLOUT;
-                int rval = Poll(&fd, 1, 1000);
-
-                if (rval > 0) {
-                    int opt;
-                    socklen_t len = sizeof(opt);
-                    getsockopt(s, SOL_SOCKET, SO_ERROR, (char*)&opt, &len);
-
-                    return opt;
-                } else {
-                    continue;
-                }
-            }
-        } else {
-            return s;
-        }
-    }
-
-    throw std::system_error(
-        errno, std::system_category(), "fail to connect"
-    );
-}
-
-
-ssize_t Poll(struct pollfd* fds, int nfds, int timeout) noexcept {
-#if defined(_win_)
-    int rval = WSAPoll(fds, nfds, timeout);
-#else
-    return poll(fds, nfds, timeout);
-#endif
-    return -1;
-}
-
 }
diff --git a/lib/clickhouse-cpp/clickhouse/base/socket.h b/lib/clickhouse-cpp/clickhouse/base/socket.h
index f2fc32c..9bd9ca3 100644
--- a/lib/clickhouse-cpp/clickhouse/base/socket.h
+++ b/lib/clickhouse-cpp/clickhouse/base/socket.h
@@ -1,18 +1,19 @@
 #pragma once
 
+#include "platform.h"
 #include "input.h"
 #include "output.h"
-#include "platform.h"
+#include "endpoints_iterator.h"
 
 #include <cstddef>
 #include <string>
+#include <chrono>
 
 #if defined(_win_)
-#   pragma comment(lib, "Ws2_32.lib")
-
 #   include <winsock2.h>
 #   include <ws2tcpip.h>
 #else
+#   include <arpa/inet.h>
 #   include <sys/types.h>
 #   include <sys/socket.h>
 #   include <poll.h>
@@ -22,11 +23,16 @@
 #   endif
 #endif
 
+#include <memory>
+#include <system_error>
+
 struct addrinfo;
 
 namespace clickhouse {
 
-/**
+struct ClientOptions;
+
+/** Address of a host to establish connection to.
  *
  */
 class NetworkAddress {
@@ -36,45 +42,116 @@ class NetworkAddress {
     ~NetworkAddress();
 
     const struct addrinfo* Info() const;
+    const std::string & Host() const;
 
 private:
+    const std::string host_;
     struct addrinfo* info_;
 };
 
+#if defined(_win_)
 
-class SocketHolder {
+class windowsErrorCategory : public std::error_category {
 public:
-    SocketHolder();
-    SocketHolder(SOCKET s);
-    SocketHolder(SocketHolder&& other);
+    char const* name() const noexcept override final;
+    std::string message(int c) const override final;
 
-    ~SocketHolder();
+    static windowsErrorCategory const& category();
+};
 
-    void Close() noexcept;
+#endif
 
-    bool Closed() const noexcept;
+#if defined(_unix_)
 
-    SocketHolder& operator = (SocketHolder&& other) noexcept;
+class getaddrinfoErrorCategory : public std::error_category {
+public:
+    char const* name() const noexcept override final;
+    std::string message(int c) const override final;
 
-    operator SOCKET () const noexcept;
+    static getaddrinfoErrorCategory const& category();
+};
 
-private:
-    SocketHolder(const SocketHolder&) = delete;
-    SocketHolder& operator = (const SocketHolder&) = delete;
+#endif
+
+
+class SocketBase {
+public:
+    virtual ~SocketBase();
+
+    virtual std::unique_ptr<InputStream> makeInputStream() const = 0;
+    virtual std::unique_ptr<OutputStream> makeOutputStream() const = 0;
+};
+
+
+class SocketFactory {
+public:
+    virtual ~SocketFactory();
+
+    // TODO: move connection-related options to ConnectionOptions structure.
+
+    virtual std::unique_ptr<SocketBase> connect(const ClientOptions& opts, const Endpoint& endpoint) = 0;
+
+    virtual void sleepFor(const std::chrono::milliseconds& duration);
+};
+
+
+struct SocketTimeoutParams {
+    std::chrono::milliseconds connect_timeout{ 5000 };
+    std::chrono::milliseconds recv_timeout{ 0 };
+    std::chrono::milliseconds send_timeout{ 0 };
+};
+
+class Socket : public SocketBase {
+public:
+    Socket(const NetworkAddress& addr, const SocketTimeoutParams& timeout_params);
+    Socket(const NetworkAddress& addr);
+    Socket(Socket&& other) noexcept;
+    Socket& operator=(Socket&& other) noexcept;
+
+    ~Socket() override;
+
+    /// @params idle the time (in seconds) the connection needs to remain
+    ///         idle before TCP starts sending keepalive probes.
+    /// @params intvl the time (in seconds) between individual keepalive probes.
+    /// @params cnt the maximum number of keepalive probes TCP should send
+    ///         before dropping the connection.
+    void SetTcpKeepAlive(int idle, int intvl, int cnt) noexcept;
+
+    /// @params nodelay whether to enable TCP_NODELAY
+    void SetTcpNoDelay(bool nodelay) noexcept;
+
+    std::unique_ptr<InputStream> makeInputStream() const override;
+    std::unique_ptr<OutputStream> makeOutputStream() const override;
+
+protected:
+    Socket(const Socket&) = delete;
+    Socket& operator = (const Socket&) = delete;
+    void Close();
 
     SOCKET handle_;
 };
 
 
-/**
- *
- */
+class NonSecureSocketFactory : public SocketFactory {
+public:
+    ~NonSecureSocketFactory() override;
+
+    std::unique_ptr<SocketBase> connect(const ClientOptions& opts, const Endpoint& endpoint) override;
+
+protected:
+    virtual std::unique_ptr<Socket> doConnect(const NetworkAddress& address, const ClientOptions& opts);
+
+    void setSocketOptions(Socket& socket, const ClientOptions& opts);
+};
+
+
 class SocketInput : public InputStream {
 public:
     explicit SocketInput(SOCKET s);
     ~SocketInput();
 
 protected:
+    bool Skip(size_t bytes) override;
     size_t DoRead(void* buf, size_t len) override;
 
 private:
@@ -87,7 +164,7 @@ class SocketOutput : public OutputStream {
     ~SocketOutput();
 
 protected:
-    void DoWrite(const void* data, size_t len) override;
+    size_t DoWrite(const void* data, size_t len) override;
 
 private:
     SOCKET s_;
@@ -97,9 +174,4 @@ static struct NetrworkInitializer {
     NetrworkInitializer();
 } gNetrworkInitializer;
 
-///
-SOCKET SocketConnect(const NetworkAddress& addr);
-
-ssize_t Poll(struct pollfd* fds, int nfds, int timeout) noexcept;
-
 }
diff --git a/lib/clickhouse-cpp/clickhouse/base/sslsocket.cpp b/lib/clickhouse-cpp/clickhouse/base/sslsocket.cpp
new file mode 100644
index 0000000..4c5185d
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/sslsocket.cpp
@@ -0,0 +1,307 @@
+#include "sslsocket.h"
+#include "../client.h"
+#include "../exceptions.h"
+
+#include <stdexcept>
+
+#include <openssl/ssl.h>
+#include <openssl/x509v3.h>
+#include <openssl/err.h>
+#include <openssl/asn1.h>
+
+
+namespace {
+
+std::string getCertificateInfo(X509* cert)
+{
+    if (!cert)
+        return "No certificate";
+
+    std::unique_ptr<BIO, decltype(&BIO_free)> mem_bio(BIO_new(BIO_s_mem()), &BIO_free);
+    X509_print(mem_bio.get(), cert);
+
+    char * data = nullptr;
+    auto len = BIO_get_mem_data(mem_bio.get(), &data);
+    if (len < 0)
+        return "Can't get certificate info due to BIO error " + std::to_string(len);
+
+    return std::string(data, len);
+}
+
+void throwSSLError(SSL * ssl, int error, const char * /*location*/, const char * /*statement*/, const std::string prefix = "OpenSSL error: ") {
+    const auto detail_error = ERR_get_error();
+    auto reason = ERR_reason_error_string(detail_error);
+    reason = reason ? reason : "Unknown SSL error";
+
+    std::string reason_str = reason;
+    if (ssl) {
+        // Print certificate only if handshake isn't completed 
+        if (auto ssl_session = SSL_get_session(ssl); ssl_session && SSL_get_state(ssl) != TLS_ST_OK)
+            reason_str += "\nServer certificate: " + getCertificateInfo(SSL_SESSION_get0_peer(ssl_session));
+    }
+
+//    std::cerr << "!!! SSL error at " << location
+//              << "\n\tcaused by " << statement
+//              << "\n\t: "<< reason_str << "(" << error << ")"
+//              << "\n\t last err: " << ERR_peek_last_error()
+//              << std::endl;
+
+    throw clickhouse::OpenSSLError(prefix + std::to_string(error) + " : " + reason_str);
+}
+
+void configureSSL(const clickhouse::SSLParams::ConfigurationType & configuration, SSL * ssl, SSL_CTX * context = nullptr) {
+    std::unique_ptr<SSL_CONF_CTX, decltype(&SSL_CONF_CTX_free)> conf_ctx_holder(SSL_CONF_CTX_new(), SSL_CONF_CTX_free);
+    auto conf_ctx = conf_ctx_holder.get();
+
+    // To make both cmdline and flag file commands start with no prefix.
+    SSL_CONF_CTX_set1_prefix(conf_ctx, "");
+    // Allow all set of client commands, also turn on proper error reporting to reuse throwSSLError().
+    SSL_CONF_CTX_set_flags(conf_ctx, SSL_CONF_FLAG_CMDLINE | SSL_CONF_FLAG_FILE | SSL_CONF_FLAG_CLIENT | SSL_CONF_FLAG_SHOW_ERRORS | SSL_CONF_FLAG_CERTIFICATE );
+    if (ssl)
+        SSL_CONF_CTX_set_ssl(conf_ctx, ssl);
+    else if (context)
+        SSL_CONF_CTX_set_ssl_ctx(conf_ctx, context);
+
+    for (const auto & kv : configuration) {
+        const int err = SSL_CONF_cmd(conf_ctx, kv.first.c_str(), (kv.second ? kv.second->c_str() : nullptr));
+        // From the documentation:
+        //  2 - both key and value used
+        //  1 - only key used
+        //  0 - error during processing
+        // -2 - key not recodnized
+        // -3 - missing value
+        const bool value_present = !!kv.second;
+        if (err == 2 || (err == 1 && !value_present))
+            continue;
+        else if (err == 0)
+            throwSSLError(ssl, SSL_ERROR_NONE, nullptr, nullptr, "Failed to configure OpenSSL with command '" + kv.first + "' ");
+        else if (err == 1 && value_present)
+            throw clickhouse::OpenSSLError("Failed to configure OpenSSL: command '" + kv.first + "' needs no value");
+        else if (err == -2)
+            throw clickhouse::OpenSSLError("Failed to configure OpenSSL: unknown command '" + kv.first + "'");
+        else if (err == -3)
+            throw clickhouse::OpenSSLError("Failed to configure OpenSSL: command '" + kv.first + "' requires a value");
+        else
+            throw clickhouse::OpenSSLError("Failed to configure OpenSSL: command '" + kv.first + "' unknown error: " + std::to_string(err));
+    }
+}
+
+#define STRINGIFY_HELPER(x) #x
+#define STRINGIFY(x) STRINGIFY_HELPER(x)
+#define LOCATION __FILE__  ":" STRINGIFY(__LINE__)
+
+struct SSLInitializer {
+    SSLInitializer() {
+        SSL_library_init();
+        SSLeay_add_ssl_algorithms();
+        SSL_load_error_strings();
+    }
+};
+
+SSL_CTX * prepareSSLContext(const clickhouse::SSLParams & context_params) {
+    static const SSLInitializer ssl_initializer;
+
+    const SSL_METHOD *method = TLS_client_method();
+    std::unique_ptr<SSL_CTX, decltype(&SSL_CTX_free)> ctx(SSL_CTX_new(method), &SSL_CTX_free);
+
+    if (!ctx)
+        throw clickhouse::OpenSSLError("Failed to initialize SSL context");
+
+#define HANDLE_SSL_CTX_ERROR(statement) do { \
+    if (const auto ret_code = (statement); !ret_code) \
+        throwSSLError(nullptr, static_cast<int>(ERR_peek_error()), LOCATION, #statement); \
+} while(false);
+
+    if (context_params.use_default_ca_locations)
+        HANDLE_SSL_CTX_ERROR(SSL_CTX_set_default_verify_paths(ctx.get()));
+    if (!context_params.path_to_ca_directory.empty())
+        HANDLE_SSL_CTX_ERROR(
+            SSL_CTX_load_verify_locations(
+                ctx.get(),
+                nullptr,
+                context_params.path_to_ca_directory.c_str())
+        );
+
+    for (const auto & f : context_params.path_to_ca_files)
+        HANDLE_SSL_CTX_ERROR(SSL_CTX_load_verify_locations(ctx.get(), f.c_str(), nullptr));
+
+    if (context_params.context_options != -1)
+        SSL_CTX_set_options(ctx.get(), context_params.context_options);
+    if (context_params.min_protocol_version != -1)
+        HANDLE_SSL_CTX_ERROR(
+            SSL_CTX_set_min_proto_version(ctx.get(), context_params.min_protocol_version));
+    if (context_params.max_protocol_version != -1)
+        HANDLE_SSL_CTX_ERROR(
+            SSL_CTX_set_max_proto_version(ctx.get(), context_params.max_protocol_version));
+
+    return ctx.release();
+#undef HANDLE_SSL_CTX_ERROR
+}
+
+auto convertConfiguration(const decltype(clickhouse::ClientOptions::SSLOptions::configuration) & configuration)
+{
+    auto result = decltype(clickhouse::SSLParams::configuration){};
+    for (const auto & cv : configuration)
+        result.push_back({cv.command, cv.value});
+
+    return result;
+}
+
+clickhouse::SSLParams GetSSLParams(const clickhouse::ClientOptions& opts) {
+    const auto& ssl_options = *opts.ssl_options;
+    return clickhouse::SSLParams{
+            ssl_options.path_to_ca_files,
+            ssl_options.path_to_ca_directory,
+            ssl_options.use_default_ca_locations,
+            ssl_options.context_options,
+            ssl_options.min_protocol_version,
+            ssl_options.max_protocol_version,
+            ssl_options.use_sni,
+            ssl_options.skip_verification,
+            ssl_options.host_flags,
+            convertConfiguration(ssl_options.configuration)
+    };
+}
+
+}
+
+namespace clickhouse {
+
+SSLContext::SSLContext(SSL_CTX & context)
+    : context_(&context, &SSL_CTX_free)
+{
+    SSL_CTX_up_ref(context_.get());
+}
+
+SSLContext::SSLContext(const SSLParams & context_params)
+    : context_(prepareSSLContext(context_params), &SSL_CTX_free)
+{
+}
+
+SSL_CTX * SSLContext::getContext() {
+    return context_.get();
+}
+
+// Allows caller to use returned value of `statement` if there was no error, throws exception otherwise.
+#define HANDLE_SSL_ERROR(SSL_PTR, statement) [&] { \
+    if (const auto ret_code = (statement); ret_code <= 0) { \
+        throwSSLError(SSL_PTR, SSL_get_error(SSL_PTR, static_cast<int>(ret_code)), LOCATION, #statement); \
+        return static_cast<std::decay_t<decltype(ret_code)>>(0); \
+    } \
+    else \
+        return ret_code; \
+} ()
+
+/* // debug macro for tracing SSL state
+#define LOG_SSL_STATE() std::cerr << "!!!!" << LOCATION << " @" << __FUNCTION__ \
+    << "\t" << SSL_get_version(ssl_) << " state: "  << SSL_state_string_long(ssl_) \
+    << "\n\t handshake state: " << SSL_get_state(ssl_) \
+    << std::endl
+*/
+SSLSocket::SSLSocket(const NetworkAddress& addr, const SocketTimeoutParams& timeout_params,
+                     const SSLParams & ssl_params, SSLContext& context)
+    : Socket(addr, timeout_params)
+    , ssl_(SSL_new(context.getContext()), &SSL_free)
+{
+    auto ssl = ssl_.get();
+    if (!ssl)
+        throw clickhouse::OpenSSLError("Failed to create SSL instance");
+
+    std::unique_ptr<ASN1_OCTET_STRING, decltype(&ASN1_OCTET_STRING_free)> ip_addr(a2i_IPADDRESS(addr.Host().c_str()), &ASN1_OCTET_STRING_free);
+
+    HANDLE_SSL_ERROR(ssl, SSL_set_fd(ssl, static_cast<int>(handle_)));
+    if (ssl_params.use_SNI)
+        HANDLE_SSL_ERROR(ssl, SSL_set_tlsext_host_name(ssl, addr.Host().c_str()));
+
+    if (ssl_params.host_flags != -1)
+        SSL_set_hostflags(ssl, ssl_params.host_flags);
+    HANDLE_SSL_ERROR(ssl, SSL_set1_host(ssl, addr.Host().c_str()));
+
+    // DO NOT use SSL_set_verify(ssl, SSL_VERIFY_PEER, nullptr), since
+    // we check verification result later, and that provides better error message.
+
+    if (ssl_params.configuration.size() > 0)
+        configureSSL(ssl_params.configuration, ssl);
+
+    SSL_set_connect_state(ssl);
+    HANDLE_SSL_ERROR(ssl, SSL_connect(ssl));
+    HANDLE_SSL_ERROR(ssl, SSL_set_mode(ssl, SSL_MODE_AUTO_RETRY));
+
+    if (const auto verify_result = SSL_get_verify_result(ssl); !ssl_params.skip_verification && verify_result != X509_V_OK) {
+        auto error_message = X509_verify_cert_error_string(verify_result);
+        throw clickhouse::OpenSSLError("Failed to verify SSL connection, X509_v error: "
+                + std::to_string(verify_result)
+                + " " + error_message
+                + "\nServer certificate: " + getCertificateInfo(SSL_get_peer_certificate(ssl)));
+    }
+
+    // Host name verification is done by OpenSSL itself, however if we are connecting to an ip-address,
+    // no verification is made, so we have to do it manually.
+    // Just in case if this is ever required, leave it here commented out.
+//    if (ip_addr) {
+//        // if hostname is actually an IP address
+//        HANDLE_SSL_ERROR(ssl, X509_check_ip(
+//                SSL_get_peer_certificate(ssl),
+//                ASN1_STRING_get0_data(ip_addr.get()),
+//                ASN1_STRING_length(ip_addr.get()),
+//                0));
+//    }
+}
+
+void SSLSocket::validateParams(const SSLParams & ssl_params) {
+    // We need either SSL or SSL_CTX to properly validate configuration, so create a temporary one.
+    std::unique_ptr<SSL_CTX, decltype(&SSL_CTX_free)> ctx(SSL_CTX_new(TLS_client_method()), &SSL_CTX_free);
+    configureSSL(ssl_params.configuration, nullptr, ctx.get());
+}
+
+
+SSLSocketFactory::SSLSocketFactory(const ClientOptions& opts)
+    : NonSecureSocketFactory()
+    , ssl_params_(GetSSLParams(opts)) {
+    if (opts.ssl_options->ssl_context) {
+        ssl_context_ = std::make_unique<SSLContext>(*opts.ssl_options->ssl_context);
+    } else {
+        ssl_context_ = std::make_unique<SSLContext>(ssl_params_);
+    }
+}
+
+SSLSocketFactory::~SSLSocketFactory() = default;
+
+std::unique_ptr<Socket> SSLSocketFactory::doConnect(const NetworkAddress& address, const ClientOptions& opts) {
+    SocketTimeoutParams timeout_params { opts.connection_connect_timeout, opts.connection_recv_timeout, opts.connection_send_timeout };
+    return std::make_unique<SSLSocket>(address, timeout_params, ssl_params_, *ssl_context_);
+}
+
+std::unique_ptr<InputStream> SSLSocket::makeInputStream() const {
+    return std::make_unique<SSLSocketInput>(ssl_.get());
+}
+
+std::unique_ptr<OutputStream> SSLSocket::makeOutputStream() const {
+    return std::make_unique<SSLSocketOutput>(ssl_.get());
+}
+
+SSLSocketInput::SSLSocketInput(SSL *ssl)
+    : ssl_(ssl)
+{}
+
+size_t SSLSocketInput::DoRead(void* buf, size_t len) {
+    size_t actually_read;
+    HANDLE_SSL_ERROR(ssl_, SSL_read_ex(ssl_, buf, len, &actually_read));
+    return actually_read;
+}
+
+SSLSocketOutput::SSLSocketOutput(SSL *ssl)
+    : ssl_(ssl)
+{}
+
+size_t SSLSocketOutput::DoWrite(const void* data, size_t len) {
+    if (len > std::numeric_limits<int>::max())
+        // FIXME(vnemkov): We should do multiple `SSL_write`s in this case.
+        throw AssertionError("Failed to write too big chunk at once "
+                + std::to_string(len) + " > " + std::to_string(std::numeric_limits<int>::max()));
+    return static_cast<size_t>(HANDLE_SSL_ERROR(ssl_, SSL_write(ssl_, data, static_cast<int>(len))));
+}
+
+#undef HANDLE_SSL_ERROR
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/base/sslsocket.h b/lib/clickhouse-cpp/clickhouse/base/sslsocket.h
new file mode 100644
index 0000000..945de86
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/sslsocket.h
@@ -0,0 +1,111 @@
+#pragma once
+
+#include "socket.h"
+
+#include <memory>
+#include <optional>
+#include <vector>
+
+typedef struct ssl_ctx_st SSL_CTX;
+typedef struct ssl_st SSL;
+
+namespace clickhouse {
+
+struct SSLParams
+{
+    std::vector<std::string> path_to_ca_files;
+    std::string path_to_ca_directory;
+    bool use_default_ca_locations;
+    int context_options;
+    int min_protocol_version;
+    int max_protocol_version;
+    bool use_SNI;
+    bool skip_verification;
+    int host_flags;
+    using ConfigurationType = std::vector<std::pair<std::string, std::optional<std::string>>>;
+    ConfigurationType configuration;
+};
+
+class SSLContext
+{
+public:
+    explicit SSLContext(SSL_CTX & context);
+    explicit SSLContext(const SSLParams & context_params);
+    ~SSLContext() = default;
+
+    SSLContext(const SSLContext &) = delete;
+    SSLContext& operator=(const SSLContext &) = delete;
+    SSLContext(SSLContext &&) = delete;
+    SSLContext& operator=(SSLContext &) = delete;
+
+private:
+    friend class SSLSocket;
+    SSL_CTX * getContext();
+
+private:
+    std::unique_ptr<SSL_CTX, void (*)(SSL_CTX*)> context_;
+};
+
+class SSLSocket : public Socket {
+public:
+    explicit SSLSocket(const NetworkAddress& addr, const SocketTimeoutParams& timeout_params,
+                       const SSLParams& ssl_params, SSLContext& context);
+
+    SSLSocket(SSLSocket &&) = default;
+    ~SSLSocket() override = default;
+
+    SSLSocket(const SSLSocket & ) = delete;
+    SSLSocket& operator=(const SSLSocket & ) = delete;
+
+    std::unique_ptr<InputStream> makeInputStream() const override;
+    std::unique_ptr<OutputStream> makeOutputStream() const override;
+
+    static void validateParams(const SSLParams & ssl_params);
+private:
+    std::unique_ptr<SSL, void (*)(SSL *s)> ssl_;
+};
+
+class SSLSocketFactory : public NonSecureSocketFactory {
+public:
+    explicit SSLSocketFactory(const ClientOptions& opts);
+    ~SSLSocketFactory() override;
+
+protected:
+    std::unique_ptr<Socket> doConnect(const NetworkAddress& address, const ClientOptions& opts) override;
+
+private:
+    const SSLParams ssl_params_;
+    std::unique_ptr<SSLContext> ssl_context_;
+};
+
+class SSLSocketInput : public InputStream {
+public:
+    explicit SSLSocketInput(SSL *ssl);
+    ~SSLSocketInput() = default;
+
+    bool Skip(size_t /*bytes*/) override {
+        return false;
+    }
+
+protected:
+    size_t DoRead(void* buf, size_t len) override;
+
+private:
+    // Not owning
+    SSL *ssl_;
+};
+
+class SSLSocketOutput : public OutputStream {
+public:
+    explicit SSLSocketOutput(SSL *ssl);
+    ~SSLSocketOutput() = default;
+
+protected:
+    size_t DoWrite(const void* data, size_t len) override;
+
+private:
+    // Not owning
+    SSL *ssl_;
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/base/string_utils.h b/lib/clickhouse-cpp/clickhouse/base/string_utils.h
index f2e66ba..4d19485 100644
--- a/lib/clickhouse-cpp/clickhouse/base/string_utils.h
+++ b/lib/clickhouse-cpp/clickhouse/base/string_utils.h
@@ -8,6 +8,7 @@
 namespace clickhouse {
 
 template <typename T>
+[[deprecated("Not used by clickhosue-cpp itself, and will be removed in next major release (3.0) ")]]
 inline T FromString(const std::string& s) {
    std::istringstream iss(s);
    T result;
@@ -16,6 +17,7 @@ inline T FromString(const std::string& s) {
 }
 
 template <typename T>
+[[deprecated("Not used by clickhosue-cpp itself, and will be removed in next major release (3.0) ")]]
 inline T FromString(const StringView& s) {
    std::istringstream iss((std::string(s)));
    T result;
diff --git a/lib/clickhouse-cpp/clickhouse/base/string_view.h b/lib/clickhouse-cpp/clickhouse/base/string_view.h
index 7d35585..ad71907 100644
--- a/lib/clickhouse-cpp/clickhouse/base/string_view.h
+++ b/lib/clickhouse-cpp/clickhouse/base/string_view.h
@@ -11,7 +11,9 @@ template <
     typename TChar,
     typename TTraits = std::char_traits<TChar>
 >
-class StringViewImpl {
+class
+[[deprecated("Obsolete due to C++17's std::string_view. Will be removed in next major release (3.0) ")]]
+StringViewImpl {
 public:
     using size_type = size_t;
     using traits_type = TTraits;
@@ -75,6 +77,11 @@ class StringViewImpl {
         return size_;
     }
 
+    // to mimic std::string and std::string_view
+    inline size_type length() const noexcept {
+        return size();
+    }
+
 public:
     // Returns a substring [pos, pos + count).
     // If the requested substring extends past the end of the string,
diff --git a/lib/clickhouse-cpp/clickhouse/base/uuid.h b/lib/clickhouse-cpp/clickhouse/base/uuid.h
new file mode 100644
index 0000000..87db5ec
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/uuid.h
@@ -0,0 +1,10 @@
+#pragma once
+
+#include <cstdint>
+#include <utility>
+
+namespace clickhouse {
+
+using UUID = std::pair<uint64_t, uint64_t>;
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/base/wire_format.cpp b/lib/clickhouse-cpp/clickhouse/base/wire_format.cpp
new file mode 100644
index 0000000..55d4fb8
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/base/wire_format.cpp
@@ -0,0 +1,177 @@
+#include <assert.h>
+#include "wire_format.h"
+
+#include "input.h"
+#include "output.h"
+
+#include "../exceptions.h"
+
+#include <stdexcept>
+#include <algorithm>
+
+namespace {
+constexpr int MAX_VARINT_BYTES = 10;
+}
+
+namespace clickhouse {
+
+bool WireFormat::ReadAll(InputStream& input, void* buf, size_t len) {
+    uint8_t* p = static_cast<uint8_t*>(buf);
+
+    size_t read_previously = 1; // 1 to execute loop at least once
+    while (len > 0 && read_previously) {
+        read_previously = input.Read(p, len);
+
+        p += read_previously;
+        len -= read_previously;
+    }
+
+    return !len;
+}
+
+void WireFormat::WriteAll(OutputStream& output, const void* buf, size_t len) {
+    const size_t original_len = len;
+    const uint8_t* p = static_cast<const uint8_t*>(buf);
+
+    size_t written_previously = 1; // 1 to execute loop at least once
+    while (len > 0 && written_previously) {
+        written_previously = output.Write(p, len);
+
+        p += written_previously;
+        len -= written_previously;
+    }
+
+    if (len) {
+        throw ProtocolError("Failed to write " + std::to_string(original_len)
+                + " bytes, only written " + std::to_string(original_len - len));
+    }
+}
+
+bool WireFormat::ReadVarint64(InputStream& input, uint64_t* value) {
+    *value = 0;
+
+    for (size_t i = 0; i < MAX_VARINT_BYTES; ++i) {
+        uint8_t byte = 0;
+
+        if (!input.ReadByte(&byte)) {
+            return false;
+        } else {
+            *value |= uint64_t(byte & 0x7F) << (7 * i);
+
+            if (!(byte & 0x80)) {
+                return true;
+            }
+        }
+    }
+
+    // TODO skip invalid
+    return false;
+}
+
+void WireFormat::WriteVarint64(OutputStream& output, uint64_t value) {
+    uint8_t bytes[MAX_VARINT_BYTES];
+    int size = 0;
+
+    for (size_t i = 0; i < MAX_VARINT_BYTES; ++i) {
+        uint8_t byte = value & 0x7F;
+        if (value > 0x7F)
+            byte |= 0x80;
+
+        bytes[size++] = byte;
+
+        value >>= 7;
+        if (!value) {
+            break;
+        }
+    }
+
+    WriteAll(output, bytes, size);
+}
+
+bool WireFormat::SkipString(InputStream& input) {
+    uint64_t len = 0;
+
+    if (ReadVarint64(input, &len)) {
+        if (len > 0x00FFFFFFULL)
+            return false;
+
+        return input.Skip((size_t)len);
+    }
+
+    return false;
+}
+
+inline const char* find_quoted_chars(const char* start, const char* end)
+{
+    static constexpr char quoted_chars[] = {'\0', '\b', '\t', '\n', '\'', '\\'};
+    const auto first  = std::find_first_of(start, end, std::begin(quoted_chars), std::end(quoted_chars));
+
+    return (first == end) ? nullptr : first;
+}
+
+void WireFormat::WriteQuotedString(OutputStream& output, std::string_view value) {
+    auto size               = value.size();
+    const char* start       = value.data();
+    const char* end         = start + size;
+    const char* quoted_char = find_quoted_chars(start, end);
+    if (quoted_char == nullptr) {
+        WriteVarint64(output, size + 2);
+        WriteAll(output, "'", 1);
+        WriteAll(output, start, size);
+        WriteAll(output, "'", 1);
+        return;
+    }
+
+    // calculate quoted chars count
+    int quoted_count             = 1;
+    const char* next_quoted_char = quoted_char + 1;
+    while ((next_quoted_char = find_quoted_chars(next_quoted_char, end))) {
+        quoted_count++;
+        next_quoted_char++;
+    }
+    WriteVarint64(output, size + 2 + 3 * quoted_count);  // length
+
+    WriteAll(output, "'", 1);
+
+    do {
+        auto write_size = quoted_char - start;
+        WriteAll(output, start, write_size);
+        WriteAll(output, "\\", 1);
+        char c = quoted_char[0];
+        switch (c) {
+            case '\0':
+                WriteAll(output, "x00", 3);
+                break;
+            case '\b':
+                WriteAll(output, "x08", 3);
+                break;
+            case '\t':
+                WriteAll(output, R"(\\t)", 3);
+                break;
+            case '\n':
+                WriteAll(output, R"(\\n)", 3);
+                break;
+            case '\'':
+                WriteAll(output, "x27", 3);
+                break;
+            case '\\':
+                WriteAll(output, R"(\\\)", 3);
+                break;
+            default:
+                break;
+        }
+        start       = quoted_char + 1;
+        quoted_char = find_quoted_chars(start, end);
+    } while (quoted_char);
+
+    WriteAll(output, start, end - start);
+    WriteAll(output, "'", 1);
+}
+
+void WireFormat::WriteParamNullRepresentation(OutputStream& output) {
+    const std::string NULL_REPRESENTATION(R"('\\N')");
+    WriteVarint64(output, NULL_REPRESENTATION.size());
+    WriteAll(output, NULL_REPRESENTATION.data(), NULL_REPRESENTATION.size());
+}
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/base/wire_format.h b/lib/clickhouse-cpp/clickhouse/base/wire_format.h
index f383fcb..be58d7a 100644
--- a/lib/clickhouse-cpp/clickhouse/base/wire_format.h
+++ b/lib/clickhouse-cpp/clickhouse/base/wire_format.h
@@ -1,101 +1,79 @@
 #pragma once
 
-#include "coded.h"
-
 #include <string>
+#include <cstdint>
 
 namespace clickhouse {
 
+class InputStream;
+class OutputStream;
+
 class WireFormat {
 public:
     template <typename T>
-    static bool ReadFixed(CodedInputStream* input, T* value);
-
-    static bool ReadString(CodedInputStream* input, std::string* value);
-
-    static bool ReadBytes(CodedInputStream* input, void* buf, size_t len);
-
-    static bool ReadUInt64(CodedInputStream* input, uint64_t* value);
-
+    static bool ReadFixed(InputStream& input, T* value);
+    static bool ReadString(InputStream& input, std::string* value);
+    static bool SkipString(InputStream& input);
+    static bool ReadBytes(InputStream& input, void* buf, size_t len);
+    static bool ReadUInt64(InputStream& input, uint64_t* value);
+    static bool ReadVarint64(InputStream& output, uint64_t* value);
 
     template <typename T>
-    static void WriteFixed(CodedOutputStream* output, const T& value);
-
-    static void WriteBytes(CodedOutputStream* output, const void* buf, size_t len);
-
-    static void WriteString(CodedOutputStream* output, const std::string& value);
-
-    static void WriteUInt64(CodedOutputStream* output, const uint64_t value);
+    static void WriteFixed(OutputStream& output, const T& value);
+    static void WriteBytes(OutputStream& output, const void* buf, size_t len);
+    static void WriteString(OutputStream& output, std::string_view value);
+    static void WriteQuotedString(OutputStream& output, std::string_view value);
+    static void WriteParamNullRepresentation(OutputStream& output);
+    static void WriteUInt64(OutputStream& output, const uint64_t value);
+    static void WriteVarint64(OutputStream& output, uint64_t value);
+
+private:
+    static bool ReadAll(InputStream& input, void* buf, size_t len);
+    static void WriteAll(OutputStream& output, const void* buf, size_t len);
 };
 
 template <typename T>
-inline bool WireFormat::ReadFixed(
-    CodedInputStream* input,
-    T* value)
-{
-    return input->ReadRaw(value, sizeof(T));
+inline bool WireFormat::ReadFixed(InputStream& input, T* value) {
+    return ReadAll(input, value, sizeof(T));
 }
 
-inline bool WireFormat::ReadString(
-    CodedInputStream* input,
-    std::string* value)
-{
-    uint64_t len;
-
-    if (input->ReadVarint64(&len)) {
+inline bool WireFormat::ReadString(InputStream& input, std::string* value) {
+    uint64_t len = 0;
+    if (ReadVarint64(input, &len)) {
         if (len > 0x00FFFFFFULL) {
             return false;
         }
         value->resize((size_t)len);
-        return input->ReadRaw(&(*value)[0], (size_t)len);
+        return ReadAll(input, value->data(), (size_t)len);
     }
 
     return false;
 }
 
-inline bool WireFormat::ReadBytes(
-    CodedInputStream* input, void* buf, size_t len)
-{
-    return input->ReadRaw(buf, len);
+inline bool WireFormat::ReadBytes(InputStream& input, void* buf, size_t len) {
+    return ReadAll(input, buf, len);
 }
 
-inline bool WireFormat::ReadUInt64(
-    CodedInputStream* input,
-    uint64_t* value)
-{
-    return input->ReadVarint64(value);
+inline bool WireFormat::ReadUInt64(InputStream& input, uint64_t* value) {
+    return ReadVarint64(input, value);
 }
 
-
 template <typename T>
-inline void WireFormat::WriteFixed(
-    CodedOutputStream* output,
-    const T& value)
-{
-    output->WriteRaw(&value, sizeof(T));
+inline void WireFormat::WriteFixed(OutputStream& output, const T& value) {
+    WriteAll(output, &value, sizeof(T));
 }
 
-inline void WireFormat::WriteBytes(
-    CodedOutputStream* output,
-    const void* buf,
-    size_t len)
-{
-    output->WriteRaw(buf, len);
+inline void WireFormat::WriteBytes(OutputStream& output, const void* buf, size_t len) {
+    WriteAll(output, buf, len);
 }
 
-inline void WireFormat::WriteString(
-    CodedOutputStream* output,
-    const std::string& value)
-{
-    output->WriteVarint64(value.size());
-    output->WriteRaw(value.data(), value.size());
+inline void WireFormat::WriteString(OutputStream& output, std::string_view value) {
+    WriteVarint64(output, value.size());
+    WriteAll(output, value.data(), value.size());
 }
 
-inline void WireFormat::WriteUInt64(
-    CodedOutputStream* output,
-    const uint64_t value)
-{
-    output->WriteVarint64(value);
+inline void WireFormat::WriteUInt64(OutputStream& output, const uint64_t value) {
+    WriteVarint64(output, value);
 }
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/block.cpp b/lib/clickhouse-cpp/clickhouse/block.cpp
index 3291f98..519cfa4 100644
--- a/lib/clickhouse-cpp/clickhouse/block.cpp
+++ b/lib/clickhouse-cpp/clickhouse/block.cpp
@@ -1,5 +1,7 @@
 #include "block.h"
 
+#include "exceptions.h"
+
 #include <stdexcept>
 
 namespace clickhouse {
@@ -10,6 +12,11 @@ Block::Iterator::Iterator(const Block& block)
 {
 }
 
+Block::Iterator::Iterator(const Block& block, Block::Iterator::ConstructAtEndTag /*at_end*/)
+    : block_(block)
+    , idx_(block.GetColumnCount())
+{}
+
 const std::string& Block::Iterator::Name() const {
     return block_.columns_[idx_].name;
 }
@@ -22,8 +29,9 @@ ColumnRef Block::Iterator::Column() const {
     return block_.columns_[idx_].column;
 }
 
-void Block::Iterator::Next() {
+bool Block::Iterator::Next() {
     ++idx_;
+    return IsValid();
 }
 
 bool Block::Iterator::IsValid() const {
@@ -48,7 +56,7 @@ void Block::AppendColumn(const std::string& name, const ColumnRef& col) {
     if (columns_.empty()) {
         rows_ = col->Size();
     } else if (col->Size() != rows_) {
-        throw std::runtime_error("all columns in block must have same count of rows. Name: ["+name+"], rows: ["+std::to_string(rows_)+"], columns: [" + std::to_string(col->Size())+"]");
+        throw ValidationError("all columns in block must have same count of rows. Name: ["+name+"], rows: ["+std::to_string(rows_)+"], columns: [" + std::to_string(col->Size())+"]");
     }
 
     columns_.push_back(ColumnItem{name, col});
@@ -63,11 +71,50 @@ const BlockInfo& Block::Info() const {
     return info_;
 }
 
+/// Set block info
+void Block::SetInfo(BlockInfo info) {
+    info_ = std::move(info);
+}
+
 /// Count of rows in the block.
 size_t Block::GetRowCount() const {
     return rows_;
 }
 
+size_t Block::RefreshRowCount() {
+    size_t rows = 0UL;
+
+    for (size_t idx = 0UL; idx < columns_.size(); ++idx)
+    {
+       const std::string& name = columns_[idx].name;
+       const ColumnRef& col = columns_[idx].column;
+
+       if (idx == 0UL)
+           rows = col->Size();
+       else if (rows != col->Size())
+           throw ValidationError("all columns in block must have same count of rows. Name: ["+name+"], rows: ["+std::to_string(rows)+"], columns: [" + std::to_string(col->Size())+"]");
+    }
+
+    rows_ = rows;
+    return rows_;
+}
+
+void Block::Clear() {
+    for (auto & c : columns_) {
+        c.column->Clear();
+    }
+
+    RefreshRowCount();
+}
+
+void Block::Reserve(size_t new_cap) {
+    for (auto & c : columns_) {
+        c.column->Reserve(new_cap);
+    }
+}
+
+
+
 ColumnRef Block::operator [] (size_t idx) const {
     if (idx < columns_.size()) {
         return columns_[idx].column;
@@ -76,4 +123,12 @@ ColumnRef Block::operator [] (size_t idx) const {
     throw std::out_of_range("column index is out of range. Index: ["+std::to_string(idx)+"], columns: [" + std::to_string(columns_.size())+"]");
 }
 
+Block::Iterator Block::begin() const {
+    return Iterator(*this);
+}
+
+Block::Iterator Block::end() const {
+    return Iterator(*this, Iterator::ConstructAtEndTag{});
+}
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/block.h b/lib/clickhouse-cpp/clickhouse/block.h
index 4322396..a32cd02 100644
--- a/lib/clickhouse-cpp/clickhouse/block.h
+++ b/lib/clickhouse-cpp/clickhouse/block.h
@@ -25,13 +25,35 @@ class Block {
         /// Reference to column object.
         ColumnRef Column() const;
 
-        /// Move to next column.
-        void Next();
+        /// Move to next column, returns false if next call to IsValid() would return false;
+        bool Next();
 
         /// Is the iterator still valid.
         bool IsValid() const;
 
+        size_t ColumnIndex() const {
+            return idx_;
+        }
+
+        Iterator& operator*() { return *this; }
+        const Iterator& operator*() const { return *this; }
+
+        bool operator==(const Iterator & other) const {
+            return &block_ == &other.block_ && idx_ == other.idx_;
+        }
+        bool operator!=(const Iterator & other) const {
+            return !(*this == other);
+        }
+
+        Iterator& operator++() {
+            this->Next();
+            return *this;
+        }
+
     private:
+        friend class Block;
+        struct ConstructAtEndTag {};
+        Iterator(const Block& block, ConstructAtEndTag at_end);
         Iterator() = delete;
 
         const Block& block_;
@@ -51,16 +73,32 @@ class Block {
 
     const BlockInfo& Info() const;
 
+    /// Set block info
+    void SetInfo(BlockInfo info);
+
     /// Count of rows in the block.
     size_t GetRowCount() const;
 
+    size_t RefreshRowCount();
+
     const std::string& GetColumnName(size_t idx) const {
         return columns_.at(idx).name;
     }
 
+    /// Convenience method to wipe out all rows from all columns
+    void Clear();
+
+    /// Convenience method to do Reserve() on all columns
+    void Reserve(size_t new_cap);
+
     /// Reference to column by index in the block.
     ColumnRef operator [] (size_t idx) const;
 
+    Iterator begin() const;
+    Iterator end() const;
+    Iterator cbegin() const { return begin(); }
+    Iterator cend() const { return end(); }
+
 private:
     struct ColumnItem {
         std::string name;
diff --git a/lib/clickhouse-cpp/clickhouse/client.cpp b/lib/clickhouse-cpp/clickhouse/client.cpp
index 060a298..e8d0c02 100644
--- a/lib/clickhouse-cpp/clickhouse/client.cpp
+++ b/lib/clickhouse-cpp/clickhouse/client.cpp
@@ -1,28 +1,23 @@
 #include "client.h"
+#include "clickhouse/version.h"
 #include "protocol.h"
 
-#include "base/coded.h"
 #include "base/compressed.h"
 #include "base/socket.h"
 #include "base/wire_format.h"
 
 #include "columns/factory.h"
 
-#include <cityhash/city.h>
-#include <lz4/lz4.h>
-
 #include <assert.h>
-#include <atomic>
 #include <system_error>
-#include <thread>
 #include <vector>
 #include <sstream>
-#include <iostream>
 
-#define DBMS_NAME                                       "ClickHouse"
-#define DBMS_VERSION_MAJOR                              1
-#define DBMS_VERSION_MINOR                              1
-#define REVISION                                        54126
+#if defined(WITH_OPENSSL)
+#include "base/sslsocket.h"
+#endif
+
+#define CLIENT_NAME "clickhouse-cpp"
 
 #define DBMS_MIN_REVISION_WITH_TEMPORARY_TABLES         50264
 #define DBMS_MIN_REVISION_WITH_TOTAL_ROWS_IN_PROGRESS   51554
@@ -30,6 +25,26 @@
 #define DBMS_MIN_REVISION_WITH_CLIENT_INFO              54032
 #define DBMS_MIN_REVISION_WITH_SERVER_TIMEZONE          54058
 #define DBMS_MIN_REVISION_WITH_QUOTA_KEY_IN_CLIENT_INFO 54060
+//#define DBMS_MIN_REVISION_WITH_TABLES_STATUS            54226
+#define DBMS_MIN_REVISION_WITH_TIME_ZONE_PARAMETER_IN_DATETIME_DATA_TYPE 54337
+#define DBMS_MIN_REVISION_WITH_SERVER_DISPLAY_NAME      54372
+#define DBMS_MIN_REVISION_WITH_VERSION_PATCH            54401
+#define DBMS_MIN_REVISION_WITH_LOW_CARDINALITY_TYPE     54405
+#define DBMS_MIN_REVISION_WITH_COLUMN_DEFAULTS_METADATA 54410
+#define DBMS_MIN_REVISION_WITH_CLIENT_WRITE_INFO        54420
+#define DBMS_MIN_REVISION_WITH_SETTINGS_SERIALIZED_AS_STRINGS 54429
+#define DBMS_MIN_REVISION_WITH_INTERSERVER_SECRET       54441
+#define DBMS_MIN_REVISION_WITH_OPENTELEMETRY            54442
+#define DBMS_MIN_REVISION_WITH_DISTRIBUTED_DEPTH        54448
+#define DBMS_MIN_REVISION_WITH_INITIAL_QUERY_START_TIME 54449
+#define DBMS_MIN_REVISION_WITH_INCREMENTAL_PROFILE_EVENTS 54451
+#define DBMS_MIN_REVISION_WITH_PARALLEL_REPLICAS 54453
+#define DBMS_MIN_REVISION_WITH_CUSTOM_SERIALIZATION  54454 // Client can get some fields in JSon format
+#define DBMS_MIN_PROTOCOL_VERSION_WITH_ADDENDUM 54458 // send quota key after handshake
+#define DBMS_MIN_PROTOCOL_REVISION_WITH_QUOTA_KEY 54458 // the same
+#define DBMS_MIN_PROTOCOL_VERSION_WITH_PARAMETERS 54459
+
+#define DMBS_PROTOCOL_REVISION  DBMS_MIN_PROTOCOL_VERSION_WITH_PARAMETERS
 
 namespace clickhouse {
 
@@ -45,61 +60,139 @@ struct ClientInfo {
     std::string initial_address = "[::ffff:127.0.0.1]:0";
     uint64_t client_version_major = 0;
     uint64_t client_version_minor = 0;
+    uint64_t client_version_patch = 0;
     uint32_t client_revision = 0;
 };
 
-struct ServerInfo {
-    std::string name;
-    std::string timezone;
-    uint64_t    version_major;
-    uint64_t    version_minor;
-    uint64_t    revision;
-};
+std::ostream& operator<<(std::ostream& os, const Endpoint& endpoint) {
+    return os << endpoint.host << ":" << endpoint.port;
+}
 
 std::ostream& operator<<(std::ostream& os, const ClientOptions& opt) {
-    os << "Client(" << opt.user << '@' << opt.host << ":" << opt.port
+    os << "Client("
+       << " Endpoints : [";
+    size_t extra_endpoints = 0;
+
+    if (!opt.host.empty()) {
+        extra_endpoints = 1;
+        os << opt.user << '@' << Endpoint{opt.host, opt.port};
+
+        if (opt.endpoints.size())
+            os << ", ";
+    }
+
+    for (size_t i = 0; i < opt.endpoints.size(); i++) {
+        os << opt.user << '@' << opt.endpoints[i]
+           << ((i == opt.endpoints.size() - 1) ? "" : ", ");
+    }
+
+    os << "] (" << opt.endpoints.size() + extra_endpoints << " items )"
        << " ping_before_query:" << opt.ping_before_query
        << " send_retries:" << opt.send_retries
        << " retry_timeout:" << opt.retry_timeout.count()
        << " compression_method:"
-       << (opt.compression_method == CompressionMethod::LZ4 ? "LZ4" : "None")
-       << ")";
+       << (opt.compression_method == CompressionMethod::LZ4    ? "LZ4"
+           : opt.compression_method == CompressionMethod::ZSTD ? "ZSTD"
+                                                               : "None");
+#if defined(WITH_OPENSSL)
+    if (opt.ssl_options) {
+        const auto & ssl_options = *opt.ssl_options;
+        os << " SSL ("
+           << " ssl_context: " << (ssl_options.ssl_context ? "provided by user" : "created internally")
+           << " use_default_ca_locations: " << ssl_options.use_default_ca_locations
+           << " path_to_ca_files: " << ssl_options.path_to_ca_files.size() << " items"
+           << " path_to_ca_directory: " << ssl_options.path_to_ca_directory
+           << " min_protocol_version: " << ssl_options.min_protocol_version
+           << " max_protocol_version: " << ssl_options.max_protocol_version
+           << " context_options: " << ssl_options.context_options
+           << ")";
+    }
+#endif
+    os << ")";
     return os;
 }
 
+ClientOptions& ClientOptions::SetSSLOptions(ClientOptions::SSLOptions options)
+{
+#ifdef WITH_OPENSSL
+    ssl_options = options;
+    return *this;
+#else
+    (void)options;
+    throw OpenSSLError("Library was built with no SSL support");
+#endif
+}
+
+namespace {
+
+std::unique_ptr<SocketFactory> GetSocketFactory(const ClientOptions& opts) {
+    (void)opts;
+#if defined(WITH_OPENSSL)
+    if (opts.ssl_options)
+        return std::make_unique<SSLSocketFactory>(opts);
+    else
+#endif
+        return std::make_unique<NonSecureSocketFactory>();
+}
+
+std::unique_ptr<EndpointsIteratorBase> GetEndpointsIterator(const ClientOptions& opts) {
+    if (opts.endpoints.empty())
+    {
+        throw ValidationError("The list of endpoints is empty");
+    }
+
+    return std::make_unique<RoundRobinEndpointsIterator>(opts.endpoints);
+}
+
+}
+
 class Client::Impl {
 public:
      Impl(const ClientOptions& opts);
+     Impl(const ClientOptions& opts,
+          std::unique_ptr<SocketFactory> socket_factory);
     ~Impl();
 
     void ExecuteQuery(Query query);
 
+    void SelectWithExternalData(Query query, const ExternalTables& external_tables);
+
     void SendCancel();
 
-    void Insert(const std::string& table_name, const Block& block);
+    void Insert(const std::string& table_name, const std::string& query_id, const Block& block);
 
-    void InsertQuery(Query query);
+    Block BeginInsert(Query query);
 
-    void InsertData(const Block& block);
+    void SendInsertBlock(const Block& block);
 
-    void InsertDataEnd();
+    void EndInsert();
 
     void Ping();
 
     void ResetConnection();
 
+    void ResetConnectionEndpoint();
+
+    const ServerInfo& GetServerInfo() const;
+
+    const std::optional<Endpoint>& GetCurrentEndpoint() const;
+
 private:
     bool Handshake();
 
     bool ReceivePacket(uint64_t* server_packet = nullptr);
 
-    void SendQuery(const std::string& query);
+    void SendQuery(const Query& query, bool finalize = true);
+    void FinalizeQuery();
 
     void SendData(const Block& block);
 
+    void SendBlockData(const Block& block);
+    void SendExternalData(const ExternalTables& external_tables);
+
     bool SendHello();
 
-    bool ReadBlock(Block* block, CodedInputStream* input);
+    bool ReadBlock(InputStream& input, Block* block);
 
     bool ReceiveHello();
 
@@ -109,16 +202,23 @@ class Client::Impl {
     /// Reads exception packet form input stream.
     bool ReceiveException(bool rethrow = false);
 
-    void WriteBlock(const Block& block, CodedOutputStream* output);
+    void WriteBlock(const Block& block, OutputStream& output);
 
-private:
-    void Disconnect() {
-        socket_.Close();
+    void CreateConnection();
+
+    void InitializeStreams(std::unique_ptr<SocketBase>&& socket);
+
+    inline size_t GetConnectionAttempts() const
+    {
+        return options_.endpoints.size() * options_.send_retries;
     }
 
+private:
     /// In case of network errors tries to reconnect to server and
     /// call fuc several times.
-    void RetryGuard(std::function<void()> fuc);
+    void RetryGuard(std::function<void()> func);
+
+    void RetryConnectToTheEndpoint(std::function<void()>& func);
 
 private:
     class EnsureNull {
@@ -147,43 +247,41 @@ class Client::Impl {
     QueryEvents* events_;
     int compression_ = CompressionState::Disable;
 
-    SocketHolder socket_;
+    std::unique_ptr<SocketFactory> socket_factory_;
 
-    SocketInput socket_input_;
-    BufferedInput buffered_input_;
-    CodedInputStream input_;
+    std::unique_ptr<InputStream> input_;
+    std::unique_ptr<OutputStream> output_;
+    std::unique_ptr<SocketBase> socket_;
+    std::unique_ptr<EndpointsIteratorBase> endpoints_iterator;
 
-    SocketOutput socket_output_;
-    BufferedOutput buffered_output_;
-    CodedOutputStream output_;
+    std::optional<Endpoint> current_endpoint_;
 
     ServerInfo server_info_;
+
+    bool inserting_;
 };
 
+ClientOptions modifyClientOptions(ClientOptions opts)
+{
+    if (opts.host.empty())
+        return opts;
+
+    Endpoint default_endpoint({opts.host, opts.port});
+    opts.endpoints.emplace(opts.endpoints.begin(), default_endpoint);
+    return opts;
+}
 
 Client::Impl::Impl(const ClientOptions& opts)
-    : options_(opts)
+    : Impl(opts, GetSocketFactory(opts)) {}
+
+Client::Impl::Impl(const ClientOptions& opts,
+                   std::unique_ptr<SocketFactory> socket_factory)
+    : options_(modifyClientOptions(opts))
     , events_(nullptr)
-    , socket_(-1)
-    , socket_input_(socket_)
-    , buffered_input_(&socket_input_)
-    , input_(&buffered_input_)
-    , socket_output_(socket_)
-    , buffered_output_(&socket_output_)
-    , output_(&buffered_output_)
+    , socket_factory_(std::move(socket_factory))
+    , endpoints_iterator(GetEndpointsIterator(options_))
 {
-    for (int i = 0; ; ) {
-        try {
-            ResetConnection();
-            break;
-        } catch (const std::system_error&) {
-            if (++i > options_.send_retries) {
-                throw;
-            }
-
-            std::this_thread::sleep_for(options_.retry_timeout);
-        }
-    }
+    CreateConnection();
 
     if (options_.compression_method != CompressionMethod::None) {
         compression_ = CompressionState::Enable;
@@ -191,148 +289,271 @@ Client::Impl::Impl(const ClientOptions& opts)
 }
 
 Client::Impl::~Impl() {
-    Disconnect();
+    try {
+        EndInsert();
+    } catch (...) {
+    }
 }
 
 void Client::Impl::ExecuteQuery(Query query) {
+    if (inserting_) {
+        throw ValidationError("cannot execute query while inserting");
+    }
+
     EnsureNull en(static_cast<QueryEvents*>(&query), &events_);
 
     if (options_.ping_before_query) {
         RetryGuard([this]() { Ping(); });
     }
 
-    SendQuery(query.GetText());
+    SendQuery(query);
 
     while (ReceivePacket()) {
         ;
     }
 }
 
-void Client::Impl::Insert(const std::string& table_name, const Block& block) {
+
+void Client::Impl::SelectWithExternalData(Query query, const ExternalTables& external_tables) {
+    if (inserting_) {
+        throw ValidationError("cannot execute query while inserting");
+    }
+
+    if (server_info_.revision < DBMS_MIN_REVISION_WITH_TEMPORARY_TABLES) {
+       throw UnimplementedError("This version of ClickHouse server doesn't support temporary tables");
+    }
+
+    EnsureNull en(static_cast<QueryEvents*>(&query), &events_);
+
     if (options_.ping_before_query) {
         RetryGuard([this]() { Ping(); });
     }
 
-    std::vector<std::string> fields;
-    fields.reserve(block.GetColumnCount());
+    SendQuery(query, false);
+    SendExternalData(external_tables);
+    FinalizeQuery();
 
-    // Enumerate all fields
-    for (unsigned int i = 0; i < block.GetColumnCount(); i++) {
-        fields.push_back(block.GetColumnName(i));
+    while (ReceivePacket()) {
+        ;
     }
+}
 
-    std::stringstream fields_section;
+void Client::Impl::SendBlockData(const Block& block) {
+    if (compression_ == CompressionState::Enable) {
+        std::unique_ptr<OutputStream> compressed_output = std::make_unique<CompressedOutput>(output_.get(), options_.max_compression_chunk_size, options_.compression_method);
+        BufferedOutput buffered(std::move(compressed_output), options_.max_compression_chunk_size);
+    
+        WriteBlock(block, buffered);
+    } else {
+        WriteBlock(block, *output_);
+    }
+}
 
-    for (auto elem = fields.begin(); elem != fields.end(); ++elem) {
-        if (std::distance(elem, fields.end()) == 1) {
-            fields_section << *elem;
+void Client::Impl::SendExternalData(const ExternalTables& external_tables) {
+    for (const auto& table: external_tables) {
+        if (!table.data.GetRowCount()) {
+           // skip empty blocks to keep the connection in the consistent state as the current request would be marked as finished by such an empty block
+           continue;
+        }
+        WireFormat::WriteFixed<uint8_t>(*output_, ClientCodes::Data);
+        WireFormat::WriteString(*output_, table.name);
+        SendBlockData(table.data);
+    }
+}
+
+
+std::string NameToQueryString(const std::string &input)
+{
+    std::string output;
+    output.reserve(input.size() + 2);
+    output += '`';
+
+    for (const auto & c : input) {
+        if (c == '`') {
+            //escape ` with ``
+            output.append("``");
         } else {
-            fields_section << *elem << ",";
+            output.push_back(c);
         }
     }
 
-    SendQuery("INSERT INTO " + table_name + " ( " + fields_section.str() + " ) VALUES");
+    output += '`';
+    return output;
+}
 
-    uint64_t server_packet;
-    // Receive data packet.
-    while (true) {
-        bool ret = ReceivePacket(&server_packet);
+void Client::Impl::Insert(const std::string& table_name, const std::string& query_id, const Block& block) {
+    if (inserting_) {
+        throw ValidationError("cannot execute query while inserting, use SendInsertData instead");
+    }
+
+    if (options_.ping_before_query) {
+        RetryGuard([this]() { Ping(); });
+    }
+
+    inserting_ = true;
+
+    std::stringstream fields_section;
+    const auto num_columns = block.GetColumnCount();
 
-        if (!ret) {
-            throw std::runtime_error("fail to receive data packet");
+    for (unsigned int i = 0; i < num_columns; ++i) {
+        if (i == num_columns - 1) {
+            fields_section << NameToQueryString(block.GetColumnName(i));
+        } else {
+            fields_section << NameToQueryString(block.GetColumnName(i)) << ",";
         }
+    }
+
+    Query query("INSERT INTO " + table_name + " ( " + fields_section.str() + " ) VALUES", query_id);
+    SendQuery(query);
+
+    // Wait for a data packet and return
+    uint64_t server_packet = 0;
+    while (ReceivePacket(&server_packet)) {
         if (server_packet == ServerCodes::Data) {
-            break;
-        }
-        if (server_packet == ServerCodes::Progress) {
-            continue;
+            SendData(block);
+            EndInsert();
+            return;
         }
     }
 
-    // Send data.
-    SendData(block);
-    // Send empty block as marker of
-    // end of data.
-    SendData(Block());
+    throw ProtocolError("fail to receive data packet");
+}
 
-    // Wait for EOS.
-    while (ReceivePacket()) {
-        ;
+Block Client::Impl::BeginInsert(Query query) {
+    if (inserting_) {
+        throw ValidationError("cannot execute query while inserting");
     }
-}
 
-void Client::Impl::InsertQuery(Query query) {
     EnsureNull en(static_cast<QueryEvents*>(&query), &events_);
 
     if (options_.ping_before_query) {
         RetryGuard([this]() { Ping(); });
     }
 
-    SendQuery(query.GetText());
-    
-    uint64_t server_packet;
-    // Receive data packet.
-    while (true) {
-        bool ret = ReceivePacket(&server_packet);
+    inserting_ = true;
 
-        if (!ret) {
-            throw std::runtime_error("fail to receive data packet");
-        }
+    // Create a callback to extract the block with the proper query columns.
+    Block block;
+    query.OnData([&block](const Block& b) {
+        block = std::move(b);
+        return true;
+    });
+
+    SendQuery(query);
+
+    // Wait for a data packet and return
+    uint64_t server_packet = 0;
+    while (ReceivePacket(&server_packet)) {
         if (server_packet == ServerCodes::Data) {
-            break;
-        }
-        if (server_packet == ServerCodes::Progress) {
-            continue;
+            return block;
         }
     }
+
+    throw ProtocolError("fail to receive data packet");
 }
 
-void Client::Impl::InsertData(const Block& block) {
-    // Send data.
+void Client::Impl::SendInsertBlock(const Block& block) {
+    if (!inserting_) {
+        throw ValidationError("illegal call to InsertData without first calling BeginInsert");
+    }
+
     SendData(block);
 }
 
-void Client::Impl::InsertDataEnd() {
-    // Send empty block as marker of
-    // end of data.
+void Client::Impl::EndInsert() {
+    if (!inserting_) {
+        return;
+    }
+
+    // Send empty block as marker of end of data.
     SendData(Block());
 
     // Wait for EOS.
-    while (ReceivePacket()) {
+    uint64_t eos_packet{0};
+    while (ReceivePacket(&eos_packet)) {
         ;
     }
+
+    if (eos_packet != ServerCodes::EndOfStream && eos_packet != ServerCodes::Exception
+        && eos_packet != ServerCodes::Log && options_.rethrow_exceptions) {
+        throw ProtocolError(std::string{"unexpected packet from server while receiving end of query, expected (expected Exception, EndOfStream or Log, got: "}
+                            + (eos_packet ? std::to_string(eos_packet) : "nothing") + ")");
+    }
+    inserting_ = false;
 }
 
 void Client::Impl::Ping() {
-    WireFormat::WriteUInt64(&output_, ClientCodes::Ping);
-    output_.Flush();
+    if (inserting_) {
+        throw ValidationError("cannot execute query while inserting");
+    }
+
+    WireFormat::WriteUInt64(*output_, ClientCodes::Ping);
+    output_->Flush();
 
     uint64_t server_packet;
     const bool ret = ReceivePacket(&server_packet);
 
     if (!ret || server_packet != ServerCodes::Pong) {
-        throw std::runtime_error("fail to ping server");
+        throw ProtocolError("fail to ping server");
     }
 }
 
 void Client::Impl::ResetConnection() {
-    SocketHolder s(SocketConnect(NetworkAddress(options_.host, std::to_string(options_.port))));
+    InitializeStreams(socket_factory_->connect(options_, current_endpoint_.value()));
+    inserting_ = false;
 
-    if (s.Closed()) {
-        throw std::system_error(errno, std::system_category());
+    if (!Handshake()) {
+        throw ProtocolError("fail to connect to " + options_.host);
     }
+}
 
-    socket_ = std::move(s);
-    socket_input_ = SocketInput(socket_);
-    socket_output_ = SocketOutput(socket_);
-    buffered_input_.Reset();
-    buffered_output_.Reset();
+void Client::Impl::ResetConnectionEndpoint() {
+    current_endpoint_.reset();
+    for (size_t i = 0; i < options_.endpoints.size();)
+    {
+        try
+        {
+            current_endpoint_ = endpoints_iterator->Next();
+            ResetConnection();
+            return;
+        } catch (const std::system_error&) {
+            if (++i == options_.endpoints.size())
+            {
+                current_endpoint_.reset();
+                throw;
+            }
+        }
+    }
+}
 
-    if (!Handshake()) {
-        throw std::runtime_error("fail to connect to " + options_.host);
+void Client::Impl::CreateConnection() {
+    // make sure to try to connect to each endpoint at least once even if `options_.send_retries` is 0
+    const size_t max_attempts = (options_.send_retries ? options_.send_retries : 1);
+    for (size_t i = 0; i < max_attempts;)
+    {
+        try
+        {
+            // Try to connect to each endpoint before throwing exception.
+            ResetConnectionEndpoint();
+            return;
+        } catch (const std::system_error&) {
+            if (++i >= max_attempts)
+            {
+                throw;
+            }
+        }
     }
 }
 
+const ServerInfo& Client::Impl::GetServerInfo() const {
+    return server_info_;
+}
+
+
+const std::optional<Endpoint>& Client::Impl::GetCurrentEndpoint() const {
+    return current_endpoint_;
+}
+
 bool Client::Impl::Handshake() {
     if (!SendHello()) {
         return false;
@@ -340,13 +561,18 @@ bool Client::Impl::Handshake() {
     if (!ReceiveHello()) {
         return false;
     }
+
+    if (server_info_.revision >= DBMS_MIN_PROTOCOL_VERSION_WITH_ADDENDUM) {
+        WireFormat::WriteString(*output_, std::string());
+    }
+
     return true;
 }
 
 bool Client::Impl::ReceivePacket(uint64_t* server_packet) {
     uint64_t packet_type = 0;
 
-    if (!input_.ReadVarint64(&packet_type)) {
+    if (!WireFormat::ReadVarint64(*input_, &packet_type)) {
         return false;
     }
     if (server_packet) {
@@ -356,7 +582,7 @@ bool Client::Impl::ReceivePacket(uint64_t* server_packet) {
     switch (packet_type) {
     case ServerCodes::Data: {
         if (!ReceiveData()) {
-            throw std::runtime_error("can't read data packet from input stream");
+            throw ProtocolError("can't read data packet from input stream");
         }
         return true;
     }
@@ -369,22 +595,22 @@ bool Client::Impl::ReceivePacket(uint64_t* server_packet) {
     case ServerCodes::ProfileInfo: {
         Profile profile;
 
-        if (!WireFormat::ReadUInt64(&input_, &profile.rows)) {
+        if (!WireFormat::ReadUInt64(*input_, &profile.rows)) {
             return false;
         }
-        if (!WireFormat::ReadUInt64(&input_, &profile.blocks)) {
+        if (!WireFormat::ReadUInt64(*input_, &profile.blocks)) {
             return false;
         }
-        if (!WireFormat::ReadUInt64(&input_, &profile.bytes)) {
+        if (!WireFormat::ReadUInt64(*input_, &profile.bytes)) {
             return false;
         }
-        if (!WireFormat::ReadFixed(&input_, &profile.applied_limit)) {
+        if (!WireFormat::ReadFixed(*input_, &profile.applied_limit)) {
             return false;
         }
-        if (!WireFormat::ReadUInt64(&input_, &profile.rows_before_limit)) {
+        if (!WireFormat::ReadUInt64(*input_, &profile.rows_before_limit)) {
             return false;
         }
-        if (!WireFormat::ReadFixed(&input_, &profile.calculated_rows_before_limit)) {
+        if (!WireFormat::ReadFixed(*input_, &profile.calculated_rows_before_limit)) {
             return false;
         }
 
@@ -398,14 +624,23 @@ bool Client::Impl::ReceivePacket(uint64_t* server_packet) {
     case ServerCodes::Progress: {
         Progress info;
 
-        if (!WireFormat::ReadUInt64(&input_, &info.rows)) {
+        if (!WireFormat::ReadUInt64(*input_, &info.rows)) {
             return false;
         }
-        if (!WireFormat::ReadUInt64(&input_, &info.bytes)) {
+        if (!WireFormat::ReadUInt64(*input_, &info.bytes)) {
             return false;
         }
-        if (REVISION >= DBMS_MIN_REVISION_WITH_TOTAL_ROWS_IN_PROGRESS) {
-            if (!WireFormat::ReadUInt64(&input_, &info.total_rows)) {
+        if constexpr(DMBS_PROTOCOL_REVISION >= DBMS_MIN_REVISION_WITH_TOTAL_ROWS_IN_PROGRESS) {
+            if (!WireFormat::ReadUInt64(*input_, &info.total_rows)) {
+                return false;
+            }
+        }
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_CLIENT_WRITE_INFO)
+        {
+            if (!WireFormat::ReadUInt64(*input_, &info.written_rows)) {
+                return false;
+            }
+            if (!WireFormat::ReadUInt64(*input_, &info.written_bytes)) {
                 return false;
             }
         }
@@ -421,6 +656,10 @@ bool Client::Impl::ReceivePacket(uint64_t* server_packet) {
         return true;
     }
 
+    case ServerCodes::Hello: {
+        return true;
+    }
+
     case ServerCodes::EndOfStream: {
         if (events_) {
             events_->OnFinish();
@@ -428,17 +667,62 @@ bool Client::Impl::ReceivePacket(uint64_t* server_packet) {
         return false;
     }
 
+    case ServerCodes::Log: {
+        // log tag
+        if (!WireFormat::SkipString(*input_)) {
+            return false;
+        }
+        Block block;
+
+        // Use uncompressed stream since log blocks usually contain only one row
+        if (!ReadBlock(*input_, &block)) {
+            return false;
+        }
+
+        if (events_) {
+            events_->OnServerLog(block);
+        }
+        return true;
+    }
+
+    case ServerCodes::TableColumns: {
+        // external table name
+        if (!WireFormat::SkipString(*input_)) {
+            return false;
+        }
+
+        //  columns metadata
+        if (!WireFormat::SkipString(*input_)) {
+            return false;
+        }
+        return true;
+    }
+
+    case ServerCodes::ProfileEvents: {
+        if (!WireFormat::SkipString(*input_)) {
+            return false;
+        }
+
+        Block block;
+        if (!ReadBlock(*input_, &block)) {
+            return false;
+        }
+
+        if (events_) {
+            events_->OnProfileEvents(block);
+        }
+        return true;
+    }
+
     default:
-        throw std::runtime_error("unimplemented " + std::to_string((int)packet_type));
+        throw UnimplementedError("unimplemented " + std::to_string((int)packet_type));
         break;
     }
-
-    return false;
 }
 
-bool Client::Impl::ReadBlock(Block* block, CodedInputStream* input) {
+bool Client::Impl::ReadBlock(InputStream& input, Block* block) {
     // Additional information about block.
-    if (REVISION >= DBMS_MIN_REVISION_WITH_BLOCK_INFO) {
+    if (server_info_.revision >= DBMS_MIN_REVISION_WITH_BLOCK_INFO) {
         uint64_t num;
         BlockInfo info;
 
@@ -459,7 +743,7 @@ bool Client::Impl::ReadBlock(Block* block, CodedInputStream* input) {
             return false;
         }
 
-        // TODO use data
+        block->SetInfo(std::move(info));
     }
 
     uint64_t num_columns = 0;
@@ -472,25 +756,37 @@ bool Client::Impl::ReadBlock(Block* block, CodedInputStream* input) {
         return false;
     }
 
+    CreateColumnByTypeSettings create_column_settings;
+    create_column_settings.low_cardinality_as_wrapped_column = options_.backward_compatibility_lowcardinality_as_wrapped_column;
+
     for (size_t i = 0; i < num_columns; ++i) {
         std::string name;
         std::string type;
-
         if (!WireFormat::ReadString(input, &name)) {
             return false;
         }
         if (!WireFormat::ReadString(input, &type)) {
             return false;
         }
+    
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_CUSTOM_SERIALIZATION) {
+            uint8_t custom_format_len;
+            if (!WireFormat::ReadFixed(input, &custom_format_len)) {
+                return false;
+            }
+            if (custom_format_len > 0) {
+                throw UnimplementedError(std::string("unsupported custom serialization"));
+            }
+        }  
 
-        if (ColumnRef col = CreateColumnByType(type)) {
-            if (num_rows && !col->Load(input, num_rows)) {
-                throw std::runtime_error("can't load");
+        if (ColumnRef col = CreateColumnByType(type, create_column_settings)) {
+            if (num_rows && !col->Load(&input, num_rows)) {
+                throw ProtocolError("can't load column '" + name + "' of type " + type);
             }
 
             block->AppendColumn(name, col);
         } else {
-            throw std::runtime_error(std::string("unsupported column type: ") + type);
+            throw UnimplementedError(std::string("unsupported column type: ") + type);
         }
     }
 
@@ -500,23 +796,19 @@ bool Client::Impl::ReadBlock(Block* block, CodedInputStream* input) {
 bool Client::Impl::ReceiveData() {
     Block block;
 
-    if (REVISION >= DBMS_MIN_REVISION_WITH_TEMPORARY_TABLES) {
-        std::string table_name;
-
-        if (!WireFormat::ReadString(&input_, &table_name)) {
+    if (server_info_.revision >= DBMS_MIN_REVISION_WITH_TEMPORARY_TABLES) {
+        if (!WireFormat::SkipString(*input_)) {
             return false;
         }
     }
 
     if (compression_ == CompressionState::Enable) {
-        CompressedInput compressed(&input_);
-        CodedInputStream coded(&compressed);
-
-        if (!ReadBlock(&block, &coded)) {
+        CompressedInput compressed(input_.get());
+        if (!ReadBlock(compressed, &block)) {
             return false;
         }
     } else {
-        if (!ReadBlock(&block, &input_)) {
+        if (!ReadBlock(*input_, &block)) {
             return false;
         }
     }
@@ -532,26 +824,32 @@ bool Client::Impl::ReceiveData() {
 }
 
 bool Client::Impl::ReceiveException(bool rethrow) {
-    std::unique_ptr<Exception> e(new Exception);
+    std::shared_ptr<Exception> e(new Exception);
     Exception* current = e.get();
 
+    bool exception_received = true;
     do {
         bool has_nested = false;
 
-        if (!WireFormat::ReadFixed(&input_, &current->code)) {
-            return false;
+        if (!WireFormat::ReadFixed(*input_, &current->code)) {
+           exception_received = false;
+           break;
         }
-        if (!WireFormat::ReadString(&input_, &current->name)) {
-            return false;
+        if (!WireFormat::ReadString(*input_, &current->name)) {
+            exception_received = false;
+            break;
         }
-        if (!WireFormat::ReadString(&input_, &current->display_text)) {
-            return false;
+        if (!WireFormat::ReadString(*input_, &current->display_text)) {
+            exception_received = false;
+            break;
         }
-        if (!WireFormat::ReadString(&input_, &current->stack_trace)) {
-            return false;
+        if (!WireFormat::ReadString(*input_, &current->stack_trace)) {
+            exception_received = false;
+            break;
         }
-        if (!WireFormat::ReadFixed(&input_, &has_nested)) {
-            return false;
+        if (!WireFormat::ReadFixed(*input_, &has_nested)) {
+            exception_received = false;
+            break;
         }
 
         if (has_nested) {
@@ -567,73 +865,143 @@ bool Client::Impl::ReceiveException(bool rethrow) {
     }
 
     if (rethrow || options_.rethrow_exceptions) {
-        throw ServerException(std::move(e));
+        throw ServerError(e);
     }
 
-    return true;
+    return exception_received;
 }
 
 void Client::Impl::SendCancel() {
-    WireFormat::WriteUInt64(&output_, ClientCodes::Cancel);
-    output_.Flush();
+    WireFormat::WriteUInt64(*output_, ClientCodes::Cancel);
+    output_->Flush();
 }
 
-void Client::Impl::SendQuery(const std::string& query) {
-    WireFormat::WriteUInt64(&output_, ClientCodes::Query);
-    WireFormat::WriteString(&output_, std::string());
+void Client::Impl::SendQuery(const Query& query, bool finalize) {
+    WireFormat::WriteUInt64(*output_, ClientCodes::Query);
+    WireFormat::WriteString(*output_, query.GetQueryID());
 
     /// Client info.
     if (server_info_.revision >= DBMS_MIN_REVISION_WITH_CLIENT_INFO) {
         ClientInfo info;
 
         info.query_kind = 1;
-        info.client_name = "ClickHouse client";
-        info.client_version_major = DBMS_VERSION_MAJOR;
-        info.client_version_minor = DBMS_VERSION_MINOR;
-        info.client_revision = REVISION;
+        info.client_name          = CLIENT_NAME;
+        info.client_version_major = CLICKHOUSE_CPP_VERSION_MAJOR;
+        info.client_version_minor = CLICKHOUSE_CPP_VERSION_MINOR;
+        info.client_version_patch = CLICKHOUSE_CPP_VERSION_PATCH;
+        info.client_revision = DMBS_PROTOCOL_REVISION;
+
+
+        WireFormat::WriteFixed(*output_, info.query_kind);
+        WireFormat::WriteString(*output_, info.initial_user);
+        WireFormat::WriteString(*output_, info.initial_query_id);
+        WireFormat::WriteString(*output_, info.initial_address);
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_INITIAL_QUERY_START_TIME) {
+            WireFormat::WriteFixed<int64_t>(*output_, 0);
+        }
+        WireFormat::WriteFixed(*output_, info.iface_type);
 
+        WireFormat::WriteString(*output_, info.os_user);
+        WireFormat::WriteString(*output_, info.client_hostname);
+        WireFormat::WriteString(*output_, info.client_name);
+        WireFormat::WriteUInt64(*output_, info.client_version_major);
+        WireFormat::WriteUInt64(*output_, info.client_version_minor);
+        WireFormat::WriteUInt64(*output_, info.client_revision);
 
-        WireFormat::WriteFixed(&output_, info.query_kind);
-        WireFormat::WriteString(&output_, info.initial_user);
-        WireFormat::WriteString(&output_, info.initial_query_id);
-        WireFormat::WriteString(&output_, info.initial_address);
-        WireFormat::WriteFixed(&output_, info.iface_type);
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_QUOTA_KEY_IN_CLIENT_INFO)
+            WireFormat::WriteString(*output_, info.quota_key);
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_DISTRIBUTED_DEPTH)
+            WireFormat::WriteUInt64(*output_, 0u);
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_VERSION_PATCH) {
+            WireFormat::WriteUInt64(*output_, info.client_version_patch);
+        }
 
-        WireFormat::WriteString(&output_, info.os_user);
-        WireFormat::WriteString(&output_, info.client_hostname);
-        WireFormat::WriteString(&output_, info.client_name);
-        WireFormat::WriteUInt64(&output_, info.client_version_major);
-        WireFormat::WriteUInt64(&output_, info.client_version_minor);
-        WireFormat::WriteUInt64(&output_, info.client_revision);
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_OPENTELEMETRY) {
+            if (const auto& tracing_context = query.GetTracingContext()) {
+                // Have OpenTelemetry header.
+                WireFormat::WriteFixed(*output_, uint8_t(1));
+                // No point writing these numbers with variable length, because they
+                // are random and will probably require the full length anyway.
+                WireFormat::WriteFixed(*output_, tracing_context->trace_id);
+                WireFormat::WriteFixed(*output_, tracing_context->span_id);
+                WireFormat::WriteString(*output_, tracing_context->tracestate);
+                WireFormat::WriteFixed(*output_, tracing_context->trace_flags);
+            } else {
+                // Don't have OpenTelemetry header.
+                WireFormat::WriteFixed(*output_, uint8_t(0));
+            }
+        } else {
+            if (query.GetTracingContext()) {
+                // Current implementation works only for server version >= v20.11.2.1-stable
+                throw UnimplementedError(std::string("Can't send open telemetry tracing context to a server, server version is too old"));
+            }
+        }
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_PARALLEL_REPLICAS) {
+            // replica dont supported by client
+            WireFormat::WriteUInt64(*output_, 0);
+            WireFormat::WriteUInt64(*output_, 0);
+            WireFormat::WriteUInt64(*output_, 0);
+        }
+    }
 
-        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_QUOTA_KEY_IN_CLIENT_INFO)
-            WireFormat::WriteString(&output_, info.quota_key);
+    /// Per query settings
+    if (server_info_.revision >= DBMS_MIN_REVISION_WITH_SETTINGS_SERIALIZED_AS_STRINGS) {
+        for(const auto& [name, field] : query.GetQuerySettings()) {
+            WireFormat::WriteString(*output_, name);
+            WireFormat::WriteVarint64(*output_, field.flags);
+            WireFormat::WriteString(*output_, field.value);
+        }
+    }
+    else if (query.GetQuerySettings().size() > 0) {
+        // Current implementation works only for server version >= v20.1.2.4-stable, since we do not implement binary settings serialization.
+        throw UnimplementedError(std::string("Can't send query settings to a server, server version is too old"));
+    }
+    // Empty string signals end of serialized settings
+    WireFormat::WriteString(*output_, std::string());
+
+    if (server_info_.revision >= DBMS_MIN_REVISION_WITH_INTERSERVER_SECRET) {
+        WireFormat::WriteString(*output_, "");
     }
 
-    /// Per query settings.
-    //if (settings)
-    //    settings->serialize(*out);
-    //else
-    WireFormat::WriteString(&output_, std::string());
+    WireFormat::WriteUInt64(*output_, Stages::Complete);
+    WireFormat::WriteUInt64(*output_, compression_);
+    WireFormat::WriteString(*output_, query.GetText());
+
+    //Send params after query text
+    if (server_info_.revision >= DBMS_MIN_PROTOCOL_VERSION_WITH_PARAMETERS) {
+        for(const auto& [name, value] : query.GetParams()) {
+            // params is like query settings
+            WireFormat::WriteString(*output_, name);
+            const uint64_t Custom = 2;
+            WireFormat::WriteVarint64(*output_, Custom);
+            if (value)
+                WireFormat::WriteQuotedString(*output_, *value);
+            else
+                WireFormat::WriteParamNullRepresentation(*output_);
+        }
+        WireFormat::WriteString(*output_, std::string()); // empty string after last param
+    }
+ 
+    if (finalize) {
+        FinalizeQuery();
+    }
+}
 
-    WireFormat::WriteUInt64(&output_, Stages::Complete);
-    WireFormat::WriteUInt64(&output_, compression_);
-    WireFormat::WriteString(&output_, query);
+void Client::Impl::FinalizeQuery() {
     // Send empty block as marker of
     // end of data
     SendData(Block());
 
-    output_.Flush();
+    output_->Flush();
 }
 
-
-void Client::Impl::WriteBlock(const Block& block, CodedOutputStream* output) {
+void Client::Impl::WriteBlock(const Block& block, OutputStream& output) {
     // Additional information about block.
     if (server_info_.revision >= DBMS_MIN_REVISION_WITH_BLOCK_INFO) {
         WireFormat::WriteUInt64(output, 1);
-        WireFormat::WriteFixed (output, block.Info().is_overflows);
+        WireFormat::WriteFixed<uint8_t>(output, block.Info().is_overflows);
         WireFormat::WriteUInt64(output, 2);
-        WireFormat::WriteFixed (output, block.Info().bucket_num);
+        WireFormat::WriteFixed<int32_t>(output, block.Info().bucket_num);
         WireFormat::WriteUInt64(output, 0);
     }
 
@@ -644,73 +1012,52 @@ void Client::Impl::WriteBlock(const Block& block, CodedOutputStream* output) {
         WireFormat::WriteString(output, bi.Name());
         WireFormat::WriteString(output, bi.Type()->GetName());
 
-        bi.Column()->Save(output);
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_CUSTOM_SERIALIZATION) {
+            // TODO: custom serialization
+            WireFormat::WriteFixed<uint8_t>(output, 0);
+        }
+
+        // Empty columns are not serialized and occupy exactly 0 bytes.
+        // ref https://github.com/ClickHouse/ClickHouse/blob/39b37a3240f74f4871c8c1679910e065af6bea19/src/Formats/NativeWriter.cpp#L163
+        const bool containsData = block.GetRowCount() > 0;
+        if (containsData) {
+            bi.Column()->Save(&output);
+        }
     }
+    output.Flush();
 }
 
 void Client::Impl::SendData(const Block& block) {
-    WireFormat::WriteUInt64(&output_, ClientCodes::Data);
+    WireFormat::WriteUInt64(*output_, ClientCodes::Data);
 
     if (server_info_.revision >= DBMS_MIN_REVISION_WITH_TEMPORARY_TABLES) {
-        WireFormat::WriteString(&output_, std::string());
+        WireFormat::WriteString(*output_, std::string());
     }
+    SendBlockData(block);
 
-    if (compression_ == CompressionState::Enable) {
-        switch (options_.compression_method) {
-            case CompressionMethod::None: {
-                assert(false);
-                break;
-            }
+    output_->Flush();
+}
 
-            case CompressionMethod::LZ4: {
-                Buffer tmp;
-                // Serialize block's data
-                {
-                    BufferOutput out(&tmp);
-                    CodedOutputStream coded(&out);
-                    WriteBlock(block, &coded);
-                }
-                // Reserver space for data
-                Buffer buf;
-                buf.resize(9 + LZ4_compressBound(tmp.size()));
-
-                // Compress data
-                int size = LZ4_compress((const char*)tmp.data(), (char*)buf.data() + 9, tmp.size());
-                buf.resize(9 + size);
-
-                // Fill header
-                uint8_t* p = buf.data();
-                // Compression method
-                WriteUnaligned(p, (uint8_t)0x82); p += 1;
-                // Compressed data size with header
-                WriteUnaligned(p, (uint32_t)buf.size()); p += 4;
-                // Original data size
-                WriteUnaligned(p, (uint32_t)tmp.size());
-
-                WireFormat::WriteFixed(&output_, CityHash128(
-                                    (const char*)buf.data(), buf.size()));
-                WireFormat::WriteBytes(&output_, buf.data(), buf.size());
-                break;
-            }
-        }
-    } else {
-        WriteBlock(block, &output_);
-    }
+void Client::Impl::InitializeStreams(std::unique_ptr<SocketBase>&& socket) {
+    std::unique_ptr<OutputStream> output = std::make_unique<BufferedOutput>(socket->makeOutputStream());
+    std::unique_ptr<InputStream> input = std::make_unique<BufferedInput>(socket->makeInputStream());
 
-    output_.Flush();
+    std::swap(input, input_);
+    std::swap(output, output_);
+    std::swap(socket, socket_);
 }
 
 bool Client::Impl::SendHello() {
-    WireFormat::WriteUInt64(&output_, ClientCodes::Hello);
-    WireFormat::WriteString(&output_, std::string(DBMS_NAME) + " client");
-    WireFormat::WriteUInt64(&output_, DBMS_VERSION_MAJOR);
-    WireFormat::WriteUInt64(&output_, DBMS_VERSION_MINOR);
-    WireFormat::WriteUInt64(&output_, REVISION);
-    WireFormat::WriteString(&output_, options_.default_database);
-    WireFormat::WriteString(&output_, options_.user);
-    WireFormat::WriteString(&output_, options_.password);
+    WireFormat::WriteUInt64(*output_, ClientCodes::Hello);
+    WireFormat::WriteString(*output_, std::string(CLIENT_NAME));
+    WireFormat::WriteUInt64(*output_, CLICKHOUSE_CPP_VERSION_MAJOR);
+    WireFormat::WriteUInt64(*output_, CLICKHOUSE_CPP_VERSION_MINOR);
+    WireFormat::WriteUInt64(*output_, DMBS_PROTOCOL_REVISION);
+    WireFormat::WriteString(*output_, options_.default_database);
+    WireFormat::WriteString(*output_, options_.user);
+    WireFormat::WriteString(*output_, options_.password);
 
-    output_.Flush();
+    output_->Flush();
 
     return true;
 }
@@ -718,26 +1065,38 @@ bool Client::Impl::SendHello() {
 bool Client::Impl::ReceiveHello() {
     uint64_t packet_type = 0;
 
-    if (!input_.ReadVarint64(&packet_type)) {
+    if (!WireFormat::ReadVarint64(*input_, &packet_type)) {
         return false;
     }
 
     if (packet_type == ServerCodes::Hello) {
-        if (!WireFormat::ReadString(&input_, &server_info_.name)) {
+        if (!WireFormat::ReadString(*input_, &server_info_.name)) {
             return false;
         }
-        if (!WireFormat::ReadUInt64(&input_, &server_info_.version_major)) {
+        if (!WireFormat::ReadUInt64(*input_, &server_info_.version_major)) {
             return false;
         }
-        if (!WireFormat::ReadUInt64(&input_, &server_info_.version_minor)) {
+        if (!WireFormat::ReadUInt64(*input_, &server_info_.version_minor)) {
             return false;
         }
-        if (!WireFormat::ReadUInt64(&input_, &server_info_.revision)) {
+        if (!WireFormat::ReadUInt64(*input_, &server_info_.revision)) {
             return false;
         }
 
         if (server_info_.revision >= DBMS_MIN_REVISION_WITH_SERVER_TIMEZONE) {
-            if (!WireFormat::ReadString(&input_, &server_info_.timezone)) {
+            if (!WireFormat::ReadString(*input_, &server_info_.timezone)) {
+                return false;
+            }
+        }
+
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_SERVER_DISPLAY_NAME) {
+            if (!WireFormat::ReadString(*input_, &server_info_.display_name)) {
+                return false;
+            }
+        }
+
+        if (server_info_.revision >= DBMS_MIN_REVISION_WITH_VERSION_PATCH) {
+            if (!WireFormat::ReadUInt64(*input_, &server_info_.version_patch)) {
                 return false;
             }
         }
@@ -752,21 +1111,45 @@ bool Client::Impl::ReceiveHello() {
 }
 
 void Client::Impl::RetryGuard(std::function<void()> func) {
-    for (int i = 0; i <= options_.send_retries; ++i) {
-        try {
-            func();
-            return;
-        } catch (const std::system_error&) {
-            bool ok = true;
 
+    if (current_endpoint_)
+    {
+        for (unsigned int i = 0; ; ++i) {
             try {
-                std::this_thread::sleep_for(options_.retry_timeout);
-                ResetConnection();
-            } catch (...) {
-                ok = false;
-            }
+                func();
+                return;
+            } catch (const std::system_error&) {
+                bool ok = true;
+
+                try {
+                    socket_factory_->sleepFor(options_.retry_timeout);
+                    ResetConnection();
+                } catch (...) {
+                    ok = false;
+                }
 
-            if (!ok) {
+                if (!ok && i == options_.send_retries) {
+                    break;
+                }
+            }
+        }
+    }
+    // Connections with current_endpoint_ are broken.
+    // Trying to establish  with the another one from the list.
+    size_t connection_attempts_count = GetConnectionAttempts();
+    for (size_t i = 0; i < connection_attempts_count;)
+    {
+        try
+        {
+            socket_factory_->sleepFor(options_.retry_timeout);
+            current_endpoint_ = endpoints_iterator->Next();
+            ResetConnection();
+            func();
+            return;
+        } catch (const std::system_error&) {
+            if (++i == connection_attempts_count)
+            {
+                current_endpoint_.reset();
                 throw;
             }
         }
@@ -779,6 +1162,13 @@ Client::Client(const ClientOptions& opts)
 {
 }
 
+Client::Client(const ClientOptions& opts,
+               std::unique_ptr<SocketFactory> socket_factory)
+    : options_(opts)
+    , impl_(new Impl(opts, std::move(socket_factory)))
+{
+}
+
 Client::~Client()
 { }
 
@@ -787,31 +1177,71 @@ void Client::Execute(const Query& query) {
 }
 
 void Client::Select(const std::string& query, SelectCallback cb) {
-    Execute(Query(query).OnData(cb));
+    Execute(Query(query).OnData(std::move(cb)));
+}
+
+void Client::Select(const std::string& query, const std::string& query_id, SelectCallback cb) {
+    Execute(Query(query, query_id).OnData(std::move(cb)));
 }
 
 void Client::SelectCancelable(const std::string& query, SelectCancelableCallback cb) {
-    Execute(Query(query).OnDataCancelable(cb));
+    Execute(Query(query).OnDataCancelable(std::move(cb)));
+}
+
+void Client::SelectCancelable(const std::string& query, const std::string& query_id, SelectCancelableCallback cb) {
+    Execute(Query(query, query_id).OnDataCancelable(std::move(cb)));
 }
 
 void Client::Select(const Query& query) {
     Execute(query);
 }
 
+void Client::SelectWithExternalData(const std::string& query, const ExternalTables& external_tables, SelectCallback cb) {
+    impl_->SelectWithExternalData(Query(query).OnData(std::move(cb)), external_tables);
+}
+
+void Client::SelectWithExternalData(const std::string& query, const std::string& query_id, const ExternalTables& external_tables, SelectCallback cb) {
+    impl_->SelectWithExternalData(Query(query, query_id).OnData(std::move(cb)), external_tables);
+}
+
+void Client::SelectWithExternalDataCancelable(const std::string& query, const ExternalTables& external_tables, SelectCancelableCallback cb) {
+    impl_->SelectWithExternalData(Query(query).OnDataCancelable(std::move(cb)), external_tables);
+}
+
+void Client::SelectWithExternalDataCancelable(const std::string& query, const std::string& query_id, const ExternalTables& external_tables, SelectCancelableCallback cb) {
+    impl_->SelectWithExternalData(Query(query, query_id).OnDataCancelable(std::move(cb)), external_tables);
+}
+
+void Client::SelectWithExternalData(const Query& query, const ExternalTables& external_tables) {
+    impl_->SelectWithExternalData(query, external_tables);
+}
+
 void Client::Insert(const std::string& table_name, const Block& block) {
-    impl_->Insert(table_name, block);
+    impl_->Insert(table_name, Query::default_query_id, block);
+}
+
+void Client::Insert(const std::string& table_name, const std::string& query_id, const Block& block) {
+    impl_->Insert(table_name, query_id, block);
+}
+
+Block Client::BeginInsert(const std::string& query) {
+    return impl_->BeginInsert(Query(query));
+}
+
+Block Client::BeginInsert(const std::string& query, const std::string& query_id) {
+    return impl_->BeginInsert(Query(query, query_id));
 }
 
-void Client::InsertQuery(const std::string& query, SelectCallback cb) {
-    impl_->InsertQuery(Query(query).OnData(cb));
+Block Client::BeginInsert(const Query& query) {
+    return impl_->BeginInsert(query);
 }
 
-void Client::InsertData(const Block& block) {
-    impl_->InsertData(block);
+void Client::SendInsertBlock(const Block& block) {
+    impl_->SendInsertBlock(block);
 }
 
-void Client::InsertDataEnd() {
-    impl_->InsertDataEnd();
+void Client::EndInsert() {
+    impl_->EndInsert();
 }
 
 void Client::Ping() {
@@ -822,4 +1252,26 @@ void Client::ResetConnection() {
     impl_->ResetConnection();
 }
 
+void Client::ResetConnectionEndpoint() {
+    impl_->ResetConnectionEndpoint();
+}
+
+const std::optional<Endpoint>& Client::GetCurrentEndpoint() const {
+    return impl_->GetCurrentEndpoint();
+}
+
+const ServerInfo& Client::GetServerInfo() const {
+    return impl_->GetServerInfo();
+}
+
+Client::Version Client::GetVersion() {
+    return Version {
+        CLICKHOUSE_CPP_VERSION_MAJOR,
+        CLICKHOUSE_CPP_VERSION_MINOR,
+        CLICKHOUSE_CPP_VERSION_PATCH,
+        CLICKHOUSE_CPP_VERSION_BUILD,
+        ""
+    };
+}
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/client.h b/lib/clickhouse-cpp/clickhouse/client.h
index b7c25e6..301882b 100644
--- a/lib/clickhouse-cpp/clickhouse/client.h
+++ b/lib/clickhouse-cpp/clickhouse/client.h
@@ -5,38 +5,82 @@
 
 #include "columns/array.h"
 #include "columns/date.h"
+#include "columns/decimal.h"
 #include "columns/enum.h"
+#include "columns/geo.h"
+#include "columns/ip4.h"
+#include "columns/ip6.h"
+#include "columns/lowcardinality.h"
+#include "columns/nothing.h"
 #include "columns/nullable.h"
 #include "columns/numeric.h"
+#include "columns/map.h"
 #include "columns/string.h"
 #include "columns/tuple.h"
+#include "columns/time.h"
 #include "columns/uuid.h"
 
 #include <chrono>
+#include <cstdint>
 #include <memory>
 #include <ostream>
 #include <string>
+#include <optional>
+
+typedef struct ssl_ctx_st SSL_CTX;
 
 namespace clickhouse {
 
+struct ServerInfo {
+    std::string name;
+    std::string timezone;
+    std::string display_name;
+    uint64_t    version_major;
+    uint64_t    version_minor;
+    uint64_t    version_patch;
+    uint64_t    revision;
+};
+
 /// Methods of block compression.
-enum class CompressionMethod {
-    None    = -1,
-    LZ4     =  1,
+enum class CompressionMethod : int8_t {
+    None = -1,
+    LZ4  = 1,
+    ZSTD = 2,
+};
+
+struct Endpoint {
+    std::string host;
+    uint16_t port = 9000;
+    inline bool operator==(const Endpoint& right) const {
+        return host == right.host && port == right.port;
+    }
+};
+
+enum class EndpointsIterationAlgorithm {
+    RoundRobin = 0,
 };
 
 struct ClientOptions {
-#define DECLARE_FIELD(name, type, setter, default) \
-    type name = default; \
-    inline ClientOptions& setter(const type& value) { \
+    // Setter goes first, so it is possible to apply 'deprecated' annotation safely.
+#define DECLARE_FIELD(name, type, setter, default_value) \
+    inline auto & setter(const type& value) { \
         name = value; \
         return *this; \
-    }
+    } \
+    type name = default_value
 
     /// Hostname of the server.
     DECLARE_FIELD(host, std::string, SetHost, std::string());
     /// Service port.
-    DECLARE_FIELD(port, int, SetPort, 9000);
+    DECLARE_FIELD(port, uint16_t, SetPort, 9000);
+
+    /** Set endpoints (host+port), only one is used.
+     * Client tries to connect to those endpoints one by one, on the round-robin basis:
+     * first default enpoint (set via SetHost() + SetPort()), then each of endpoints, from begin() to end(),
+     * the first one to establish connection is used for the rest of the session.
+     * If port isn't specified, default(9000) value will be used.
+     */
+    DECLARE_FIELD(endpoints, std::vector<Endpoint>, SetEndpoints, {});
 
     /// Default database.
     DECLARE_FIELD(default_database, std::string, SetDefaultDatabase, "default");
@@ -53,17 +97,142 @@ struct ClientOptions {
     /// Ping server every time before execute any query.
     DECLARE_FIELD(ping_before_query, bool, SetPingBeforeQuery, false);
     /// Count of retry to send request to server.
-    DECLARE_FIELD(send_retries, int, SetSendRetries, 1);
+    DECLARE_FIELD(send_retries, unsigned int, SetSendRetries, 1);
     /// Amount of time to wait before next retry.
     DECLARE_FIELD(retry_timeout, std::chrono::seconds, SetRetryTimeout, std::chrono::seconds(5));
 
     /// Compression method.
     DECLARE_FIELD(compression_method, CompressionMethod, SetCompressionMethod, CompressionMethod::None);
 
+    /// TCP Keep alive options
+    DECLARE_FIELD(tcp_keepalive, bool, TcpKeepAlive, false);
+    DECLARE_FIELD(tcp_keepalive_idle, std::chrono::seconds, SetTcpKeepAliveIdle, std::chrono::seconds(60));
+    DECLARE_FIELD(tcp_keepalive_intvl, std::chrono::seconds, SetTcpKeepAliveInterval, std::chrono::seconds(5));
+    DECLARE_FIELD(tcp_keepalive_cnt, unsigned int, SetTcpKeepAliveCount, 3);
+
+    // TCP options
+    DECLARE_FIELD(tcp_nodelay, bool, TcpNoDelay, true);
+
+    /// Connection socket connect timeout. If the timeout is negative then the connect operation will never timeout.
+    DECLARE_FIELD(connection_connect_timeout, std::chrono::milliseconds, SetConnectionConnectTimeout, std::chrono::seconds(5));
+
+    /// Connection socket timeout. If the timeout is set to zero then the operation will never timeout.
+    DECLARE_FIELD(connection_recv_timeout, std::chrono::milliseconds, SetConnectionRecvTimeout, std::chrono::milliseconds(0));
+    DECLARE_FIELD(connection_send_timeout, std::chrono::milliseconds, SetConnectionSendTimeout, std::chrono::milliseconds(0));
+
+    /** It helps to ease migration of the old codebases, which can't afford to switch
+    * to using ColumnLowCardinalityT or ColumnLowCardinality directly,
+    * but still want to benefit from smaller on-wire LowCardinality bandwidth footprint.
+    *
+    * @see LowCardinalitySerializationAdaptor, CreateColumnByType
+    */
+    [[deprecated("Makes implementation of LC(X) harder and code uglier. Will be removed in next major release (3.0) ")]]
+    DECLARE_FIELD(backward_compatibility_lowcardinality_as_wrapped_column, bool, SetBakcwardCompatibilityFeatureLowCardinalityAsWrappedColumn, false);
+
+    /** Set max size data to compress if compression enabled.
+     *
+     *  Allows choosing tradeoff between RAM\CPU:
+     *  - Lower value reduces RAM usage, but slightly increases CPU usage.
+     *  - Higher value increases RAM usage but slightly decreases CPU usage.
+     */
+    DECLARE_FIELD(max_compression_chunk_size, unsigned int, SetMaxCompressionChunkSize, 65535);
+
+    struct SSLOptions {
+        /** There are two ways to configure an SSL connection:
+         *  - provide a pre-configured SSL_CTX, which is not modified and not owned by the Client.
+         *  - provide a set of options and allow the Client to create and configure SSL_CTX by itself.
+         */
+
+        /** Pre-configured SSL-context for SSL-connection.
+         *  If NOT null client DONES NOT take ownership of context and it must be valid for client lifetime.
+         *  If null client initlaizes OpenSSL and creates his own context, initializes it using
+         *  other options, like path_to_ca_files, path_to_ca_directory, use_default_ca_locations, etc.
+         *
+         *  Either way context is used to create an SSL-connection, which is then configured with
+         *  whatever was provided as `configuration`, `host_flags`, `skip_verification` and `use_sni`.
+         */
+        SSL_CTX * ssl_context = nullptr;
+        auto & SetExternalSSLContext(SSL_CTX * new_ssl_context) {
+            ssl_context = new_ssl_context;
+            return *this;
+        }
+
+        /** Means to validate the server-supplied certificate against trusted Certificate Authority (CA).
+         *  If no CAs are configured, the server's identity can't be validated, and the Client would err.
+         *  See https://www.openssl.org/docs/man1.1.1/man3/SSL_CTX_set_default_verify_paths.html
+        */
+        /// Load default CA certificates from default locations.
+        DECLARE_FIELD(use_default_ca_locations, bool, SetUseDefaultCALocations, true);
+        /// Path to the CA files to verify server certificate, may be empty.
+        DECLARE_FIELD(path_to_ca_files, std::vector<std::string>, SetPathToCAFiles, {});
+        /// Path to the directory with CA files used to validate server certificate, may be empty.
+        DECLARE_FIELD(path_to_ca_directory, std::string, SetPathToCADirectory, "");
+
+        /** Min and max protocol versions to use, set with SSL_CTX_set_min_proto_version and SSL_CTX_set_max_proto_version
+         *  for details see https://www.openssl.org/docs/man1.1.1/man3/SSL_CTX_set_min_proto_version.html
+         */
+        DECLARE_FIELD(min_protocol_version, int, SetMinProtocolVersion, DEFAULT_VALUE);
+        DECLARE_FIELD(max_protocol_version, int, SetMaxProtocolVersion, DEFAULT_VALUE);
+
+        /** Options to be set with SSL_CTX_set_options,
+         * for details see https://www.openssl.org/docs/man1.1.1/man3/SSL_CTX_set_options.html
+        */
+        DECLARE_FIELD(context_options, int, SetContextOptions, DEFAULT_VALUE);
+
+        /** Use SNI at ClientHello
+         */
+        DECLARE_FIELD(use_sni, bool, SetUseSNI, true);
+
+        /** Skip SSL session verification (server's certificate, etc).
+         *
+         *  WARNING: settig to true will bypass all SSL session checks, which
+         *  is dangerous, but can be used against self-signed certificates, e.g. for testing purposes.
+         */
+        DECLARE_FIELD(skip_verification, bool, SetSkipVerification, false);
+
+        /** Mode of verifying host ssl certificate against name of the host, set with SSL_set_hostflags.
+         *  For details see https://www.openssl.org/docs/man1.1.1/man3/SSL_set_hostflags.html
+         */
+        DECLARE_FIELD(host_flags, int, SetHostVerifyFlags, DEFAULT_VALUE);
+
+        struct CommandAndValue {
+            std::string command;
+            std::optional<std::string> value = std::nullopt;
+        };
+        /** Extra configuration options, set with SSL_CONF_cmd.
+         *  For deatils see https://www.openssl.org/docs/man1.1.1/man3/SSL_CONF_cmd.html
+         *
+         *  Takes multiple pairs of command-value strings, all commands are supported,
+         *  and prefix is empty.
+         *  i.e. pass `sigalgs` or `SignatureAlgorithms` instead of `-sigalgs`.
+         *
+         *  Rewrites any other options/flags if set in other ways.
+         */
+        DECLARE_FIELD(configuration, std::vector<CommandAndValue>, SetConfiguration, {});
+
+        static const int DEFAULT_VALUE = -1;
+    };
+
+    // By default SSL is turned off.
+    std::optional<SSLOptions> ssl_options = std::nullopt;
+
+    // Will throw an exception if client was built without SSL support.
+    ClientOptions& SetSSLOptions(SSLOptions options);
+
 #undef DECLARE_FIELD
 };
 
 std::ostream& operator<<(std::ostream& os, const ClientOptions& options);
+std::ostream& operator<<(std::ostream& os, const Endpoint& options);
+
+class SocketFactory;
+
+struct ExternalTable {
+  const std::string_view name;
+  const Block& data;
+};
+
+using ExternalTables = std::vector<ExternalTable>;
 
 /**
  *
@@ -71,6 +240,8 @@ std::ostream& operator<<(std::ostream& os, const ClientOptions& options);
 class Client {
 public:
      Client(const ClientOptions& opts);
+     Client(const ClientOptions& opts,
+            std::unique_ptr<SocketFactory> socket_factory);
     ~Client();
 
     /// Intends for execute arbitrary queries.
@@ -79,22 +250,45 @@ class Client {
     /// Intends for execute select queries.  Data will be returned with
     /// one or more call of \p cb.
     void Select(const std::string& query, SelectCallback cb);
+    void Select(const std::string& query, const std::string& query_id, SelectCallback cb);
 
     /// Executes a select query which can be canceled by returning false from
     /// the data handler function \p cb.
     void SelectCancelable(const std::string& query, SelectCancelableCallback cb);
+    void SelectCancelable(const std::string& query, const std::string& query_id, SelectCancelableCallback cb);
+
+    // The same as Select but with an external data
+    // required for the query, see https://clickhouse.com/docs/engines/table-engines/special/external-data
+    void SelectWithExternalData(const std::string& query, const ExternalTables& external_tables, SelectCallback cb);
+    void SelectWithExternalData(const std::string& query, const std::string& query_id, const ExternalTables& external_tables, SelectCallback cb);
+
+    // The same as SelectWithExternalData but can be canceled by returning false from
+    // the data handler function \p cb.
+    void SelectWithExternalDataCancelable(const std::string& query, const ExternalTables& external_tables, SelectCancelableCallback cb);
+    void SelectWithExternalDataCancelable(const std::string& query, const std::string& query_id, const ExternalTables& external_tables, SelectCancelableCallback cb);
+
+    /// Same as SelectWithExternalData but takes a fully-configured Query
+    /// (settings, params, callbacks, query_id, OnData) instead of a bare string.
+    void SelectWithExternalData(const Query& query, const ExternalTables& external_tables);
 
     /// Alias for Execute.
     void Select(const Query& query);
 
     /// Intends for insert block of data into a table \p table_name.
     void Insert(const std::string& table_name, const Block& block);
-    
-    void InsertQuery(const std::string& query, SelectCallback cb);
+    void Insert(const std::string& table_name, const std::string& query_id, const Block& block);
+
+    /// Start an \p INSERT statement, insert batches of data, then finish the insert.
+    Block BeginInsert(const std::string& query);
+    Block BeginInsert(const std::string& query, const std::string& query_id);
+    /// Start an \p INSERT statement with a fully-configured Query (settings, params, query_id).
+    Block BeginInsert(const Query& query);
 
-    void InsertData(const Block& block);
+    /// Insert data using a \p block returned by \p BeginInsert.
+    void SendInsertBlock(const Block& block);
 
-    void InsertDataEnd();
+    /// End an \p INSERT session started by \p BeginInsert.
+    void EndInsert();
 
     /// Ping server for aliveness.
     void Ping();
@@ -102,8 +296,28 @@ class Client {
     /// Reset connection with initial params.
     void ResetConnection();
 
+    const ServerInfo& GetServerInfo() const;
+
+    /// Get current connected endpoint.
+    /// In case when client is not connected to any endpoint, nullopt will returned.
+    const std::optional<Endpoint>& GetCurrentEndpoint() const;
+
+    // Try to connect to different endpoints one by one only one time. If it doesn't work, throw an exception.
+    void ResetConnectionEndpoint();
+
+    struct Version
+    {
+        uint16_t major;
+        uint16_t minor;
+        uint16_t patch;
+        uint16_t build;
+        const char * extra;
+    };
+
+    static Version GetVersion();
+
 private:
-    ClientOptions options_;
+    const ClientOptions options_;
 
     class Impl;
     std::unique_ptr<Impl> impl_;
diff --git a/lib/clickhouse-cpp/clickhouse/columns/array.cpp b/lib/clickhouse-cpp/clickhouse/columns/array.cpp
index ab7d145..94163c4 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/array.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/array.cpp
@@ -1,61 +1,130 @@
 #include "array.h"
+#include "numeric.h"
 
 #include <stdexcept>
 
 namespace clickhouse {
 
-ColumnArray::ColumnArray(ColumnRef data)
+namespace {
+std::shared_ptr<ColumnUInt64> make_single_offset(size_t value) {
+    auto res = std::make_shared<ColumnUInt64>();
+    if (value != 0) {
+        res->Append(value);
+    }
+    return res;
+}
+}  // namespace
+
+ColumnArray::ColumnArray(ColumnRef data) : ColumnArray(data, make_single_offset(data->Size())) {
+}
+
+ColumnArray::ColumnArray(ColumnRef data, std::shared_ptr<ColumnUInt64> offsets)
     : Column(Type::CreateArray(data->Type()))
     , data_(data)
-    , offsets_(std::make_shared<ColumnUInt64>())
+    , offsets_(offsets)
 {
-}
-
-void ColumnArray::AppendAsColumn(ColumnRef array) {
-    if (!data_->Type()->IsEqual(array->Type())) {
-        throw std::runtime_error(
-            "can't append column of type " + array->Type()->GetName() + " "
-            "to column type " + data_->Type()->GetName());
+    const auto rows            = offsets_->Size();
+    const auto expected_values = rows > 0 ? offsets_->At(rows - 1) : 0;
+    std::uint64_t prev         = 0;
+    // Ensure the offset array is internally consistent
+    for (const auto& o : offsets_->GetWritableData()) {
+        if (o < prev) {
+            throw ValidationError("offsets must be monotonically increasing");
+        }
+        prev = o;
     }
-
-    if (offsets_->Size() == 0) {
-        offsets_->Append(array->Size());
-    } else {
-        offsets_->Append((*offsets_)[offsets_->Size() - 1] + array->Size());
+    if (data_->Size() != expected_values) {
+        throw ValidationError("Mismatch between data and offsets: Expected " + std::to_string(expected_values) +
+                              " values in data, but got " + std::to_string(data_->Size()));
     }
+}
 
+ColumnArray::ColumnArray(ColumnArray&& other)
+    : Column(other.Type())
+    , data_(std::move(other.data_))
+    , offsets_(std::move(other.offsets_))
+{
+}
+
+void ColumnArray::AppendAsColumn(ColumnRef array) {
+    // appending data may throw (i.e. due to ype check failure), so do it first to avoid partly modified state.
     data_->Append(array);
+    AddOffset(array->Size());
 }
 
 ColumnRef ColumnArray::GetAsColumn(size_t n) const {
+    if (n >= Size())
+        throw ValidationError("Index is out ouf bounds: " + std::to_string(n));
+
     return data_->Slice(GetOffset(n), GetSize(n));
 }
 
+ColumnRef ColumnArray::Slice(size_t begin, size_t size) const {
+    if (size && begin + size > Size())
+        throw ValidationError("Slice indexes are out of bounds");
+
+    auto sliced_data = data_->Slice(GetOffset(begin), GetOffset(begin + size) - GetOffset(begin));
+    auto offsets     = std::make_shared<ColumnUInt64>();
+    auto offset      = uint64_t{0};
+    for (size_t i = 0; i < size; i++) {
+        offset += GetSize(begin + i);
+        offsets->Append(offset);
+    }
+
+    return std::make_shared<ColumnArray>(std::move(sliced_data), std::move(offsets));
+}
+
+ColumnRef ColumnArray::CloneEmpty() const {
+    return std::make_shared<ColumnArray>(data_->CloneEmpty());
+}
+
+void ColumnArray::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+    offsets_->Reserve(new_cap);
+}
+
 void ColumnArray::Append(ColumnRef column) {
     if (auto col = column->As<ColumnArray>()) {
-        if (!col->data_->Type()->IsEqual(data_->Type())) {
-            return;
-        }
-
         for (size_t i = 0; i < col->Size(); ++i) {
             AppendAsColumn(col->GetAsColumn(i));
         }
     }
 }
 
-bool ColumnArray::Load(CodedInputStream* input, size_t rows) {
-    if (!offsets_->Load(input, rows)) {
+bool ColumnArray::LoadPrefix(InputStream* input, size_t rows) {
+    if (!rows) {
+        return true;
+    }
+
+    return data_->LoadPrefix(input, rows);
+}
+
+bool ColumnArray::LoadBody(InputStream* input, size_t rows) {
+    if (!rows) {
+        return true;
+    }
+    if (!offsets_->LoadBody(input, rows)) {
         return false;
     }
-    if (!data_->Load(input, (*offsets_)[rows - 1])) {
+    const auto nested_rows = (*offsets_)[rows - 1];
+    if (nested_rows == 0) {
+        return true;
+    }
+    if (!data_->LoadBody(input, nested_rows)) {
         return false;
     }
     return true;
 }
 
-void ColumnArray::Save(CodedOutputStream* output) {
-    offsets_->Save(output);
-    data_->Save(output);
+void ColumnArray::SavePrefix(OutputStream* output) {
+    data_->SavePrefix(output);
+}
+
+void ColumnArray::SaveBody(OutputStream* output) {
+    offsets_->SaveBody(output);
+    if (data_->Size() > 0) {
+        data_->SaveBody(output);
+    }
 }
 
 void ColumnArray::Clear() {
@@ -67,12 +136,40 @@ size_t ColumnArray::Size() const {
     return offsets_->Size();
 }
 
+void ColumnArray::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnArray &>(other);
+    data_.swap(col.data_);
+    offsets_.swap(col.offsets_);
+}
+
+void ColumnArray::OffsetsIncrease(size_t n) {
+    offsets_->Append(n);
+}
+
 size_t ColumnArray::GetOffset(size_t n) const {
+
     return (n == 0) ? 0 : (*offsets_)[n - 1];
 }
 
+void ColumnArray::AddOffset(size_t n) {
+    if (offsets_->Size() == 0) {
+        offsets_->Append(n);
+    } else {
+        offsets_->Append((*offsets_)[offsets_->Size() - 1] + n);
+    }
+}
+
 size_t ColumnArray::GetSize(size_t n) const {
     return (n == 0) ? (*offsets_)[n] : ((*offsets_)[n] - (*offsets_)[n - 1]);
 }
 
+ColumnRef ColumnArray::GetData() {
+    return data_;
+}
+
+void ColumnArray::Reset() {
+    data_.reset();
+    offsets_.reset();
+}
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/array.h b/lib/clickhouse-cpp/clickhouse/columns/array.h
index c631bb2..3ad9c94 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/array.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/array.h
@@ -1,51 +1,321 @@
 #pragma once
 
+#include "column.h"
 #include "numeric.h"
+#include "utils.h"
+
+#include <memory>
 
 namespace clickhouse {
 
+template <typename NestedColumnType>
+class ColumnArrayT;
+
 /**
  * Represents column of Array(T).
  */
 class ColumnArray : public Column {
 public:
-    ColumnArray(ColumnRef data);
+    using ValueType = ColumnRef;
+
+    /** Create an array of given type.
+     *
+     *  `data` is used internally (and modified) by ColumnArray.
+     *  Users are strongly advised against supplying non-empty columns and/or modifying
+     *  contents of `data` afterwards.
+     */
+    explicit ColumnArray(ColumnRef data);
+
+    /** Create an array of given type, with actual values and offsets.
+     *
+     *  Both `data` and `offsets` are used (and modified) internally bye ColumnArray.
+     *  Users are strongly advised against modifying contents of `data` or `offsets` afterwards.
+     */
+    ColumnArray(ColumnRef data, std::shared_ptr<ColumnUInt64> offsets);
 
-    /// Converts input column to array and appends
-    /// as one row to the current column.
+    /// Converts input column to array and appends as one row to the current column.
     void AppendAsColumn(ColumnRef array);
 
-    /// Convets array at pos n to column.
+    /// Converts array at pos n to column.
     /// Type of element of result column same as type of array element.
     ColumnRef GetAsColumn(size_t n) const;
 
+    /// Shorthand to get a column casted to a proper type.
+    template <typename T>
+    auto GetAsColumnTyped(size_t n) const {
+        return GetAsColumn(n)->AsStrict<T>();
+    }
+
 public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
+    /// Loads column prefix from input stream.
+    bool LoadPrefix(InputStream* input, size_t rows) override;
+
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column prefix to output stream.
+    void SavePrefix(OutputStream* output) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
-    
+    void SaveBody(OutputStream* output) override;
+
     /// Clear column data .
     void Clear() override;
-    
+
     /// Returns count of rows in the column.
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t, size_t) override { return ColumnRef(); }
+    ColumnRef Slice(size_t, size_t) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column&) override;
 
-private:
-    size_t GetOffset(size_t n) const;
+    void OffsetsIncrease(size_t);
 
+protected:
+    template<typename T> friend class ColumnArrayT;
+
+    ColumnArray(ColumnArray&& array);
+
+    size_t GetOffset(size_t n) const;
     size_t GetSize(size_t n) const;
+    ColumnRef GetData();
+    void AddOffset(size_t n);
+    void Reset();
 
 private:
     ColumnRef data_;
     std::shared_ptr<ColumnUInt64> offsets_;
 };
 
+template <typename ColumnType>
+class ColumnArrayT : public ColumnArray {
+public:
+    class ArrayValueView;
+    using ValueType = ArrayValueView;
+    using NestedColumnType = ColumnType;
+
+    explicit ColumnArrayT(std::shared_ptr<NestedColumnType> data)
+        : ColumnArray(data)
+        , typed_nested_data_(data)
+    {}
+
+    ColumnArrayT(std::shared_ptr<NestedColumnType> data, std::shared_ptr<ColumnUInt64> offsets)
+        : ColumnArray(data, offsets)
+        , typed_nested_data_(data)
+    {}
+
+    template <typename ...Args>
+    explicit ColumnArrayT(Args &&... args)
+        : ColumnArrayT(std::make_shared<NestedColumnType>(std::forward<Args>(args)...))
+    {}
+
+    /** Create a ColumnArrayT from a ColumnArray, without copying data and offsets, but by 'stealing' those from `col`.
+     *
+     *  Ownership of column internals is transferred to returned object, original (argument) object
+     *  MUST NOT BE USED IN ANY WAY, it is only safe to dispose it.
+     *
+     *  Throws an exception if `col` is of wrong type, it is safe to use original col in this case.
+     *  This is a static method to make such conversion verbose.
+     */
+    static auto Wrap(ColumnArray&& col) {
+        auto nested_data = WrapColumn<NestedColumnType>(col.GetData());
+        return std::make_shared<ColumnArrayT<NestedColumnType>>(nested_data, col.offsets_);
+    }
+
+    static auto Wrap(Column&& col) {
+        return Wrap(std::move(dynamic_cast<ColumnArray&&>(col)));
+    }
+
+    // Helper to simplify integration with other APIs
+    static auto Wrap(ColumnRef&& col) {
+        return Wrap(std::move(*col->AsStrict<ColumnArray>()));
+    }
+
+    /// A single (row) value of the Array-column, i.e. readonly array of items.
+    class ArrayValueView {
+        const std::shared_ptr<NestedColumnType> typed_nested_data_;
+        const size_t offset_;
+        const size_t size_;
+
+    public:
+        using ValueType = std::decay_t<decltype(std::declval<NestedColumnType>().At(0))>;
+
+        ArrayValueView(std::shared_ptr<NestedColumnType> data, size_t offset = 0, size_t size = std::numeric_limits<size_t>::max())
+            : typed_nested_data_(data)
+            , offset_(offset)
+            , size_(std::min(typed_nested_data_->Size() - offset, size))
+        {}
+
+        inline auto operator[](size_t index) const {
+            return (*typed_nested_data_)[offset_ + index];
+        }
+
+        inline auto At(size_t index) const {
+            if (index >= size_)
+                throw ValidationError("ColumnArray value index out of bounds: "
+                        + std::to_string(index) + ", max is " + std::to_string(size_));
+            return typed_nested_data_->At(offset_ + index);
+        }
+
+        class Iterator {
+            const std::shared_ptr<NestedColumnType> typed_nested_data_;
+            const size_t offset_;
+            const size_t size_;
+            size_t index_;
+        public:
+            Iterator() = default;
+
+            Iterator(std::shared_ptr<NestedColumnType> typed_nested_data, size_t offset, size_t size, size_t index)
+                : typed_nested_data_(typed_nested_data)
+                , offset_(offset)
+                , size_(size)
+                , index_(index)
+            {}
+
+            using ValueType = typename ArrayValueView::ValueType;
+
+            inline auto operator*() const {
+                return typed_nested_data_->At(offset_ + index_);
+            }
+
+            inline Iterator& operator++() {
+                ++index_;
+                return *this;
+            }
+
+            inline bool operator==(const Iterator& other) const {
+                return this->typed_nested_data_ == other.typed_nested_data_
+                        && this->offset_ == other.offset_
+                        && this->size_ == other.size_
+                        && this->index_ == other.index_;
+            }
+
+            inline bool operator!=(const Iterator& other) const {
+                return !(*this == other);
+            }
+        };
+
+        // minimalistic stl-like container interface, hence the lowercase
+        inline Iterator begin() const {
+            return Iterator{typed_nested_data_, offset_, size_, 0};
+        }
+
+        inline Iterator cbegin() const {
+            return Iterator{typed_nested_data_, offset_, size_, 0};
+        }
+
+        inline Iterator end() const {
+            return Iterator{typed_nested_data_, offset_, size_, size_};
+        }
+
+        inline Iterator cend() const {
+            return Iterator{typed_nested_data_, offset_, size_, size_};
+        }
+
+        inline size_t size() const {
+            return size_;
+        }
+
+        // It is ugly to have both size() and Size(), but it is for compatitability with both STL and rest of the clickhouse-cpp.
+        inline size_t Size() const {
+            return size_;
+        }
+
+        inline bool operator==(const ArrayValueView& other) const {
+            if (size() != other.size()) {
+                return false;
+            }
+            for (size_t i = 0; i < size_; ++i) {
+                if ((*this)[i] != other[i]) {
+                    return false;
+                }
+            }
+            return true;
+        }
+
+        inline bool operator!=(const ArrayValueView& other) const {
+            return !(*this == other);
+        }
+    };
+
+    inline auto At(size_t index) const {
+        if (index >= Size())
+            throw ValidationError("ColumnArray row index out of bounds: "
+                    + std::to_string(index) + ", max is " + std::to_string(Size()));
+
+        return ArrayValueView{typed_nested_data_, GetOffset(index), GetSize(index)};
+    }
+
+    inline auto operator[](size_t index) const {
+        return ArrayValueView{typed_nested_data_, GetOffset(index), GetSize(index)};
+    }
+
+    using ColumnArray::Append;
+
+    template <typename Container>
+    inline void Append(Container&& container) {
+        using container_type = decltype(container);
+        if constexpr (std::is_lvalue_reference_v<container_type> || 
+            std::is_const_v<std::remove_reference_t<container_type>>) {
+            Append(std::begin(container), std::end(container));
+        }
+        else {
+            Append(std::make_move_iterator(std::begin(container)),
+                std::make_move_iterator(std::end(container)));
+        }
+    }
+
+    template <typename ValueType>
+    inline void Append(const std::initializer_list<ValueType>& container) {
+        Append(std::begin(container), std::end(container));
+    }
+
+    template <typename Begin, typename End>
+    inline void Append(Begin begin, End end) {
+        auto & nested_data = *typed_nested_data_;
+        size_t counter = 0;
+
+        while (begin != end) {
+            nested_data.Append(*begin);
+            ++begin;
+            ++counter;
+        }
+
+        // Even if there are 0 items, increase counter, creating empty array item.
+        AddOffset(counter);
+    }
+
+    ColumnRef Slice(size_t begin, size_t size) const override {
+        return Wrap(ColumnArray::Slice(begin, size));
+    }
+
+    ColumnRef CloneEmpty() const override {
+        return Wrap(ColumnArray::CloneEmpty());
+    }
+
+    void Swap(Column& other) override {
+        auto & col = dynamic_cast<ColumnArrayT<NestedColumnType> &>(other);
+        typed_nested_data_.swap(col.typed_nested_data_);
+        ColumnArray::Swap(other);
+    }
+
+private:
+    /// Helper to allow wrapping a "typeless" ColumnArray
+    ColumnArrayT(ColumnArray&& array, std::shared_ptr<NestedColumnType> nested_data)
+        : ColumnArray(std::move(array))
+        , typed_nested_data_(std::move(nested_data))
+    {}
+
+
+private:
+    std::shared_ptr<NestedColumnType> typed_nested_data_;
+};
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/column.cpp b/lib/clickhouse-cpp/clickhouse/columns/column.cpp
new file mode 100644
index 0000000..7f881d7
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/column.cpp
@@ -0,0 +1,24 @@
+#include "column.h"
+
+namespace clickhouse {
+
+bool Column::LoadPrefix(InputStream*, size_t) {
+    /// does nothing by default
+    return true;
+}
+
+bool Column::Load(InputStream* input, size_t rows) {
+    return LoadPrefix(input, rows) && LoadBody(input, rows);
+}
+
+void Column::SavePrefix(OutputStream*) {
+    /// does nothing by default
+}
+
+/// Saves column data to output stream.
+void Column::Save(OutputStream* output) {
+    SavePrefix(output);
+    SaveBody(output);
+}
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/column.h b/lib/clickhouse-cpp/clickhouse/columns/column.h
index efec2eb..475df89 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/column.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/column.h
@@ -1,51 +1,82 @@
 #pragma once
 
-#include "../base/input.h"
-#include "../base/coded.h"
 #include "../types/types.h"
+#include "../columns/itemview.h"
+#include "../exceptions.h"
+
+#include <memory>
+#include <stdexcept>
 
 namespace clickhouse {
 
+class InputStream;
+class OutputStream;
+
 using ColumnRef = std::shared_ptr<class Column>;
 
 /**
  * An abstract base of all columns classes.
  */
-class Column : public std::enable_shared_from_this<Column>
-{
+class Column : public std::enable_shared_from_this<Column> {
 public:
-    explicit inline Column(TypeRef type)
-        : type_(type)
-    {
-    }
+    explicit inline Column(TypeRef type) : type_(type) {}
 
-    virtual ~Column()
-    { }
+    virtual ~Column() {}
 
-    /// Downcast pointer to the specific culumn's subtype.
+    /// Downcast pointer to the specific column's subtype.
     template <typename T>
     inline std::shared_ptr<T> As() {
         return std::dynamic_pointer_cast<T>(shared_from_this());
     }
 
-    /// Downcast pointer to the specific culumn's subtype.
+    /// Downcast pointer to the specific column's subtype.
     template <typename T>
     inline std::shared_ptr<const T> As() const {
         return std::dynamic_pointer_cast<const T>(shared_from_this());
     }
 
+    /// Downcast pointer to the specific column's subtype.
+    template <typename T>
+    inline std::shared_ptr<T> AsStrict() {
+        auto result = std::dynamic_pointer_cast<T>(shared_from_this());
+        if (!result) {
+            throw ValidationError("Can't cast from " + type_->GetName());
+        }
+        return result;
+    }
+
     /// Get type object of the column.
     inline TypeRef Type() const { return type_; }
+    inline const class Type& GetType() const { return *type_; }
 
     /// Appends content of given column to the end of current one.
     virtual void Append(ColumnRef column) = 0;
 
+    /// Increase the capacity of the column for large block insertion.
+    virtual void Reserve(size_t new_cap) = 0;
+
+    /// Template method to load column data from input stream. It'll call LoadPrefix and LoadBody.
+    /// Should be called only once from the client. Derived classes should not call it.
+    bool Load(InputStream* input, size_t rows);
+
+    /// Loads column prefix from input stream.
+    virtual bool LoadPrefix(InputStream* input, size_t rows);
+
     /// Loads column data from input stream.
-    virtual bool Load(CodedInputStream* input, size_t rows) = 0;
+    virtual bool LoadBody(InputStream* input, size_t rows) = 0;
+
+    /// Saves column prefix to output stream. Column types with prefixes must implement it.
+    virtual void SavePrefix(OutputStream* output);
+
+    /// Saves column body to output stream.
+    virtual void SaveBody(OutputStream* output) = 0;
+
+    /// Template method to save to output stream. It'll call SavePrefix and SaveBody respectively
+    /// Should be called only once from the client. Derived classes should not call it.
+    /// Save is split in Prefix and Body because some data types require prefixes and specific serialization order.
+    /// For instance, Array(LowCardinality(X)) requires LowCardinality.key_version bytes to come before Array.offsets
+    void Save(OutputStream* output);
 
-    /// Saves column data to output stream.
-    virtual void Save(CodedOutputStream* output) = 0;
-    
     /// Clear column data .
     virtual void Clear() = 0;
 
@@ -53,10 +84,24 @@ class Column : public std::enable_shared_from_this<Column>
     virtual size_t Size() const = 0;
 
     /// Makes slice of the current column.
-    virtual ColumnRef Slice(size_t begin, size_t len) = 0;
+    virtual ColumnRef Slice(size_t begin, size_t len) const = 0;
+
+    virtual ColumnRef CloneEmpty() const = 0;
+
+    virtual void Swap(Column&) = 0;
+
+    /// Get a view on raw item data if it is supported by column, will throw an exception if index is out of range.
+    /// Please note that view is invalidated once column items are added or deleted, column is loaded from strean or destroyed.
+    virtual ItemView GetItem(size_t) const {
+        throw UnimplementedError("GetItem() is not supported for column of " + type_->GetName());
+    }
+
+    friend void swap(Column& left, Column& right) {
+        left.Swap(right);
+    }
 
 protected:
     TypeRef type_;
 };
 
-}
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/columns/date.cpp b/lib/clickhouse-cpp/clickhouse/columns/date.cpp
index eab41c8..ad14992 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/date.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/date.cpp
@@ -1,4 +1,5 @@
 #include "date.h"
+#include <cstdint>
 
 namespace clickhouse {
 
@@ -8,8 +9,15 @@ ColumnDate::ColumnDate()
 {
 }
 
+ColumnDate::ColumnDate(std::vector<uint16_t>&& data)
+    : Column(Type::CreateDate())
+    , data_(std::make_shared<ColumnUInt16>(std::move(data)))
+{
+}
+
 void ColumnDate::Append(const std::time_t& value) {
-    data_->Append(static_cast<uint16_t>(value / 86400));
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
+    data_->Append(static_cast<uint16_t>(value / std::time_t(86400)));
 }
 
 void ColumnDate::Clear() {
@@ -17,7 +25,16 @@ void ColumnDate::Clear() {
 }
 
 std::time_t ColumnDate::At(size_t n) const {
-    return data_->At(n) * 86400;
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
+    return static_cast<std::time_t>(data_->At(n)) * 86400;
+}
+
+void ColumnDate::AppendRaw(uint16_t value) {
+    data_->Append(value);
+}
+
+uint16_t ColumnDate::RawAt(size_t n) const {
+    return data_->At(n);
 }
 
 void ColumnDate::Append(ColumnRef column) {
@@ -26,19 +43,31 @@ void ColumnDate::Append(ColumnRef column) {
     }
 }
 
-bool ColumnDate::Load(CodedInputStream* input, size_t rows) {
-    return data_->Load(input, rows);
+std::vector<uint16_t>& ColumnDate::GetWritableData() {
+    return data_->GetWritableData();
 }
 
-void ColumnDate::Save(CodedOutputStream* output) {
-    data_->Save(output);
+void ColumnDate::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+size_t ColumnDate::Capacity() const {
+    return data_->Capacity();
+}
+
+bool ColumnDate::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnDate::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
 }
 
 size_t ColumnDate::Size() const {
     return data_->Size();
 }
 
-ColumnRef ColumnDate::Slice(size_t begin, size_t len) {
+ColumnRef ColumnDate::Slice(size_t begin, size_t len) const {
     auto col = data_->Slice(begin, len)->As<ColumnUInt16>();
     auto result = std::make_shared<ColumnDate>();
 
@@ -47,6 +76,105 @@ ColumnRef ColumnDate::Slice(size_t begin, size_t len) {
     return result;
 }
 
+ColumnRef ColumnDate::CloneEmpty() const {
+    return std::make_shared<ColumnDate>();
+}
+
+void ColumnDate::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnDate &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnDate::GetItem(size_t index) const {
+    return ItemView(Type::Date, data_->GetItem(index));
+}
+
+
+ColumnDate32::ColumnDate32()
+    : Column(Type::CreateDate32())
+    , data_(std::make_shared<ColumnInt32>())
+{
+}
+
+ColumnDate32::ColumnDate32(std::vector<int32_t>&& data)
+    : Column(Type::CreateDate32())
+    , data_(std::make_shared<ColumnInt32>(std::move(data)))
+{
+}
+
+void ColumnDate32::Append(const std::time_t& value) {
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
+    data_->Append(static_cast<int32_t>(value / std::time_t(86400)));
+}
+
+void ColumnDate32::Clear() {
+    data_->Clear();
+}
+
+std::time_t ColumnDate32::At(size_t n) const {
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
+    return static_cast<std::time_t>(data_->At(n)) * 86400;
+}
+
+void ColumnDate32::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnDate32>()) {
+        data_->Append(col->data_);
+    }
+}
+
+std::vector<int32_t>& ColumnDate32::GetWritableData() {
+    return data_->GetWritableData();
+}
+
+void ColumnDate32::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+size_t ColumnDate32::Capacity() const {
+    return data_->Capacity();
+}
+
+void ColumnDate32::AppendRaw(int32_t value) {
+    data_->Append(value);
+}
+
+int32_t ColumnDate32::RawAt(size_t n) const {
+    return data_->At(n);
+}
+
+bool ColumnDate32::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnDate32::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+size_t ColumnDate32::Size() const {
+    return data_->Size();
+}
+
+ColumnRef ColumnDate32::Slice(size_t begin, size_t len) const {
+    auto col = data_->Slice(begin, len)->As<ColumnInt32>();
+    auto result = std::make_shared<ColumnDate32>();
+
+    result->data_->Append(col);
+
+    return result;
+}
+
+ColumnRef ColumnDate32::CloneEmpty() const {
+    return std::make_shared<ColumnDate32>();
+}
+
+void ColumnDate32::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnDate32 &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnDate32::GetItem(size_t index) const {
+    return ItemView{Type()->GetCode(), data_->GetItem(index)};
+}
 
 ColumnDateTime::ColumnDateTime()
     : Column(Type::CreateDateTime())
@@ -54,6 +182,22 @@ ColumnDateTime::ColumnDateTime()
 {
 }
 
+ColumnDateTime::ColumnDateTime(std::string timezone)
+    : Column(Type::CreateDateTime(std::move(timezone)))
+    , data_(std::make_shared<ColumnUInt32>())
+{
+}
+
+ColumnDateTime::ColumnDateTime(std::vector<uint32_t>&& data)
+    : Column(Type::CreateDateTime())
+    , data_(std::make_shared<ColumnUInt32>(std::move(data))) {
+}
+
+ColumnDateTime::ColumnDateTime(std::string timezone, std::vector<uint32_t>&& data)
+    : Column(Type::CreateDateTime(std::move(timezone)))
+    , data_(std::make_shared<ColumnUInt32>(std::move(data))) {
+}
+
 void ColumnDateTime::Append(const std::time_t& value) {
     data_->Append(static_cast<uint32_t>(value));
 }
@@ -62,18 +206,42 @@ std::time_t ColumnDateTime::At(size_t n) const {
     return data_->At(n);
 }
 
+void ColumnDateTime::AppendRaw(uint32_t value) {
+    data_->Append(value);
+}
+
+uint32_t ColumnDateTime::RawAt(size_t n) const {
+	return data_->At(n);
+}
+
+std::string ColumnDateTime::Timezone() const {
+    return type_->As<DateTimeType>()->Timezone();
+}
+
 void ColumnDateTime::Append(ColumnRef column) {
     if (auto col = column->As<ColumnDateTime>()) {
         data_->Append(col->data_);
     }
 }
 
-bool ColumnDateTime::Load(CodedInputStream* input, size_t rows) {
-    return data_->Load(input, rows);
+std::vector<uint32_t>& ColumnDateTime::GetWritableData() {
+    return data_->GetWritableData();
+}
+
+void ColumnDateTime::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+size_t ColumnDateTime::Capacity() const {
+    return data_->Capacity();
+}
+
+bool ColumnDateTime::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
 }
 
-void ColumnDateTime::Save(CodedOutputStream* output) {
-    data_->Save(output);
+void ColumnDateTime::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
 }
 
 size_t ColumnDateTime::Size() const {
@@ -84,7 +252,7 @@ void ColumnDateTime::Clear() {
     data_->Clear();
 }
 
-ColumnRef ColumnDateTime::Slice(size_t begin, size_t len) {
+ColumnRef ColumnDateTime::Slice(size_t begin, size_t len) const {
     auto col = data_->Slice(begin, len)->As<ColumnUInt32>();
     auto result = std::make_shared<ColumnDateTime>();
 
@@ -93,5 +261,105 @@ ColumnRef ColumnDateTime::Slice(size_t begin, size_t len) {
     return result;
 }
 
+ColumnRef ColumnDateTime::CloneEmpty() const {
+    return std::make_shared<ColumnDateTime>();
+}
+
+void ColumnDateTime::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnDateTime &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnDateTime::GetItem(size_t index) const {
+    return ItemView(Type::DateTime, data_->GetItem(index));
+}
+
+ColumnDateTime64::ColumnDateTime64(size_t precision)
+    : ColumnDateTime64(Type::CreateDateTime64(precision), std::make_shared<ColumnDecimal>(18ul, precision))
+{}
+
+ColumnDateTime64::ColumnDateTime64(size_t precision, std::string timezone)
+    : ColumnDateTime64(Type::CreateDateTime64(precision, std::move(timezone)), std::make_shared<ColumnDecimal>(18ul, precision))
+{}
+
+ColumnDateTime64::ColumnDateTime64(TypeRef type, std::shared_ptr<ColumnDecimal> data)
+    : Column(type),
+      data_(data),
+      precision_(type->As<DateTime64Type>()->GetPrecision())
+{}
+
+void ColumnDateTime64::Append(const Int64& value) {
+    // TODO: we need a type, which safely represents datetime.
+    // The precision of Poco.DateTime is not big enough.
+    data_->Append(value);
+}
+
+//void ColumnDateTime64::Append(const std::string& value) {
+//    data_->Append(value);
+//}
+
+Int64 ColumnDateTime64::At(size_t n) const {
+    // make sure to use Absl's Int128 conversion
+    return static_cast<Int64>(data_->At(n));
+}
+
+std::string ColumnDateTime64::Timezone() const {
+    return type_->As<DateTime64Type>()->Timezone();
+}
+
+void ColumnDateTime64::Reserve(size_t new_cap)
+{
+    data_->Reserve(new_cap);
+}
+
+void ColumnDateTime64::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnDateTime64>()) {
+        data_->Append(col->data_);
+    }
+}
+
+bool ColumnDateTime64::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnDateTime64::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+void ColumnDateTime64::Clear() {
+    data_->Clear();
+}
+
+size_t ColumnDateTime64::Size() const {
+    return data_->Size();
+}
+
+ItemView ColumnDateTime64::GetItem(size_t index) const {
+    return ItemView(Type::DateTime64, data_->GetItem(index));
+}
+
+void ColumnDateTime64::Swap(Column& other) {
+    auto& col = dynamic_cast<ColumnDateTime64&>(other);
+    if (col.GetPrecision() != GetPrecision()) {
+        throw ValidationError("Can't swap DateTime64 columns when precisions are not the same: "
+                + std::to_string(GetPrecision()) + "(this) != " + std::to_string(col.GetPrecision()) + "(that)");
+    }
+
+    data_.swap(col.data_);
+}
+
+ColumnRef ColumnDateTime64::Slice(size_t begin, size_t len) const {
+    auto sliced_data = data_->Slice(begin, len)->As<ColumnDecimal>();
+
+    return ColumnRef{new ColumnDateTime64(type_, sliced_data)};
+}
+
+ColumnRef ColumnDateTime64::CloneEmpty() const {
+    return ColumnRef{new ColumnDateTime64(type_, data_->CloneEmpty()->As<ColumnDecimal>())};
+}
+
+size_t ColumnDateTime64::GetPrecision() const {
+    return precision_;
+}
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/date.h b/lib/clickhouse-cpp/clickhouse/columns/date.h
index 8df9ba1..324c67c 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/date.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/date.h
@@ -1,5 +1,6 @@
 #pragma once
 
+#include "decimal.h"
 #include "numeric.h"
 
 #include <ctime>
@@ -9,69 +10,236 @@ namespace clickhouse {
 /** */
 class ColumnDate : public Column {
 public:
+    using ValueType = std::time_t;
+
     ColumnDate();
+    explicit ColumnDate(std::vector<uint16_t>&& data);
 
     /// Appends one element to the end of column.
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
     void Append(const std::time_t& value);
 
     /// Returns element at given row number.
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
     std::time_t At(size_t n) const;
+    inline std::time_t operator [] (size_t n) const { return At(n); }
+
+    /// Do append data as is -- number of day in Unix epoch, no conversions performed.
+    void AppendRaw(uint16_t value);
+    uint16_t RawAt(size_t n) const;
 
-public:
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
+    /// Get Raw Vector Contents
+    std::vector<uint16_t>& GetWritableData();
+
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Returns the capacity of the column
+    size_t Capacity() const;
+
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
+    void SaveBody(OutputStream* output) override;
 
     /// Clear column data .
     void Clear() override;
-    
+
     /// Returns count of rows in the column.
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
 
 private:
     std::shared_ptr<ColumnUInt16> data_;
 };
 
-/** */
+
+class ColumnDate32 : public Column {
+public:
+    using ValueType = std::time_t;
+
+    ColumnDate32();
+    explicit ColumnDate32(std::vector<int32_t>&& data);
+
+    /// Appends one element to the end of column.
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
+    void Append(const std::time_t& value);
+
+    /// Returns element at given row number.
+    /// The implementation is fundamentally wrong, ignores timezones, leap years and daylight saving.
+    std::time_t At(size_t n) const;
+
+    inline std::time_t operator [] (size_t n) const { return At(n); }
+
+    /// Do append data as is -- number of day in Unix epoch (32bit signed), no conversions performed.
+    void AppendRaw(int32_t value);
+    int32_t RawAt(size_t n) const;
+
+    /// Get Raw Vector Contents
+    std::vector<int32_t>& GetWritableData();
+
+    /// Returns the capacity of the column
+    size_t Capacity() const;
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
+
+private:
+    std::shared_ptr<ColumnInt32> data_;
+};
+
+
+/** DateTime64 supports date-time values (number of seconds since UNIX epoch), from 1970 up to 2130. */
 class ColumnDateTime : public Column {
 public:
+    using ValueType = std::time_t;
+
     ColumnDateTime();
+    explicit ColumnDateTime(std::vector<uint32_t>&& data);
+
+    explicit ColumnDateTime(std::string timezone);
+    ColumnDateTime(std::string timezone, std::vector<uint32_t>&& data);
 
     /// Appends one element to the end of column.
     void Append(const std::time_t& value);
 
     /// Returns element at given row number.
     std::time_t At(size_t n) const;
+    inline std::time_t operator [] (size_t n) const { return At(n); }
+
+    /// Append raw as UNIX epoch seconds in uint32
+    void AppendRaw(uint32_t value);
+    uint32_t RawAt(size_t n) const;
+
+    /// Timezone associated with a data column.
+    std::string Timezone() const;
+
+    /// Get Raw Vector Contents
+    std::vector<uint32_t>& GetWritableData();
+
+    /// Returns the capacity of the column
+    size_t Capacity() const;
 
 public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
 
     /// Clear column data .
     void Clear() override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
+    void SaveBody(OutputStream* output) override;
 
     /// Returns count of rows in the column.
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
 
 private:
     std::shared_ptr<ColumnUInt32> data_;
 };
 
+
+/** DateTime64 supports date-time values of arbitrary sub-second precision, from 1900 up to 2300. */
+class ColumnDateTime64 : public Column {
+public:
+    using ValueType = Int64;
+
+    explicit ColumnDateTime64(size_t precision);
+    ColumnDateTime64(size_t precision, std::string timezone);
+
+    /// Appends one element to the end of column.
+    void Append(const Int64& value);
+    // It is a bit controversial: users might expect it to parse string of ISO8601 or some other human-friendly format,
+    // but current implementation parses it as fractional integer with decimal point, e.g. "123.456".
+//    void Append(const std::string& value);
+
+    /// Returns element at given row number.
+    Int64 At(size_t n) const;
+
+    inline Int64 operator[](size_t n) const { return At(n); }
+
+    /// Timezone associated with a data column.
+    std::string Timezone() const;
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
+
+    size_t GetPrecision() const;
+
+private:
+    ColumnDateTime64(TypeRef type, std::shared_ptr<ColumnDecimal> data);
+
+private:
+    std::shared_ptr<ColumnDecimal> data_;
+    const size_t precision_;
+};
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/decimal.cpp b/lib/clickhouse-cpp/clickhouse/columns/decimal.cpp
new file mode 100644
index 0000000..b2e8294
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/decimal.cpp
@@ -0,0 +1,255 @@
+#include "decimal.h"
+
+namespace
+{
+using namespace clickhouse;
+
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+template <typename T>
+inline bool addOverflow(const Int128 & l, const T & r, Int128 * result)
+{
+    __int128 res;
+    const auto ret_value = __builtin_add_overflow(static_cast<__int128>(l), static_cast<__int128>(r), &res);
+
+    *result = res;
+    return ret_value;
+}
+
+template <typename T>
+inline bool mulOverflow(const Int128 & l, const T & r, Int128 * result)
+{
+    __int128 res;
+    const auto ret_value = __builtin_mul_overflow(static_cast<__int128>(l), static_cast<__int128>(r), &res);
+
+    *result = res;
+    return ret_value;
+}
+
+#else
+template <typename T>
+inline bool getSignBit(const T & v)
+{
+    return v < static_cast<T>(0);
+}
+
+inline bool getSignBit(const Int128 & v)
+{
+//    static constexpr Int128 zero {};
+//    return v < zero;
+
+    // Sign of the whole absl::int128 value is determined by sign of higher 64 bits.
+    return absl::Int128High64(v) < 0;
+}
+
+inline bool addOverflow(const Int128 & l, const Int128 & r, Int128 * result)
+{
+    //    *result = l + r;
+    //    const auto result_sign = getSignBit(*result);
+    //    return l_sign == r_sign && l_sign != result_sign;
+
+    // Based on code from:
+    // https://wiki.sei.cmu.edu/confluence/display/c/INT32-C.+Ensure+that+operations+on+signed+integers+do+not+result+in+overflow#INT32C.Ensurethatoperationsonsignedintegersdonotresultinoverflow-CompliantSolution
+    const auto r_positive = !getSignBit(r);
+
+    if ((r_positive && (l > (std::numeric_limits<Int128>::max() - r))) ||
+        (!r_positive && (l < (std::numeric_limits<Int128>::min() - r)))) {
+        return true;
+    }
+    *result = l + r;
+
+    return false;
+}
+
+template <typename T>
+inline bool mulOverflow(const Int128 & l, const T & r, Int128 * result)
+{
+    // Based on code from:
+    // https://wiki.sei.cmu.edu/confluence/display/c/INT32-C.+Ensure+that+operations+on+signed+integers+do+not+result+in+overflow#INT32C.Ensurethatoperationsonsignedintegersdonotresultinoverflow-CompliantSolution.3
+    const auto l_positive = !getSignBit(l);
+    const auto r_positive = !getSignBit(r);
+
+    if (l_positive) {
+        if (r_positive) {
+            if (r != 0 && l > (std::numeric_limits<Int128>::max() / r)) {
+                return true;
+            }
+        } else {
+            if (l != 0 && r < (std::numeric_limits<Int128>::min() / l)) {
+                return true;
+            }
+        }
+    } else {
+        if (r_positive) {
+            if (r != 0 && l < (std::numeric_limits<Int128>::min() / r)) {
+                return true;
+            }
+        } else {
+            if (l != 0 && (r < (std::numeric_limits<Int128>::max() / l))) {
+                return true;
+            }
+        }
+    }
+
+    *result = l * r;
+    return false;
+}
+#endif
+
+}
+
+namespace clickhouse {
+
+ColumnDecimal::ColumnDecimal(size_t precision, size_t scale)
+    : Column(Type::CreateDecimal(precision, scale))
+{
+    if (precision <= 9) {
+        data_ = std::make_shared<ColumnInt32>();
+    } else if (precision <= 18) {
+        data_ = std::make_shared<ColumnInt64>();
+    } else {
+        data_ = std::make_shared<ColumnInt128>();
+    }
+    data_type_code_ = data_->Type()->GetCode();
+}
+
+ColumnDecimal::ColumnDecimal(TypeRef type, ColumnRef data)
+    : Column(type),
+      data_(data),
+      data_type_code_(data_->Type()->GetCode())
+{
+}
+
+void ColumnDecimal::Append(const Int128& value) {
+    switch (data_type_code_) {
+        case Type::Int32:
+            static_cast<ColumnInt32*>(data_.get())->Append(static_cast<int32_t>(value));
+            break;
+        case Type::Int64:
+            static_cast<ColumnInt64*>(data_.get())->Append(static_cast<int64_t>(value));
+            break;
+        default:
+            static_cast<ColumnInt128*>(data_.get())->Append(static_cast<Int128>(value));
+            break;
+    }
+}
+
+void ColumnDecimal::Append(const std::string& value) {
+    Int128 int_value = 0;
+    auto c = value.begin();
+    auto end = value.end();
+    bool sign = true;
+    bool has_dot = false;
+
+    size_t zeros = 0;
+
+    while (c != end) {
+        if (*c == '-') {
+            sign = false;
+            if (c != value.begin()) {
+                break;
+            }
+        } else if (*c == '.' && !has_dot) {
+            size_t distance = std::distance(c, end) - 1;
+            auto scale = type_->As<DecimalType>()->GetScale();
+
+            if (distance <= scale) {
+                zeros = scale - distance;
+            } else {
+                std::advance(end, scale - distance);
+            }
+
+            has_dot = true;
+        } else if (*c >= '0' && *c <= '9') {
+            if (mulOverflow(int_value, 10, &int_value) ||
+                addOverflow(int_value, *c - '0', &int_value)) {
+                throw AssertionError("value is too big for 128-bit integer");
+            }
+        } else {
+            throw ValidationError(std::string("unexpected symbol '") + (*c) + "' in decimal value");
+        }
+        ++c;
+    }
+
+    if (c != end) {
+        throw ValidationError("unexpected symbol '-' in decimal value");
+    }
+
+    while (zeros) {
+        if (mulOverflow(int_value, 10, &int_value)) {
+            throw AssertionError("value is too big for 128-bit integer");
+        }
+        --zeros;
+    }
+
+    Append(sign ? int_value : -int_value);
+}
+
+Int128 ColumnDecimal::At(size_t i) const {
+    switch (data_type_code_) {
+        case Type::Int32:
+            return static_cast<Int128>(static_cast<const ColumnInt32*>(data_.get())->At(i));
+        case Type::Int64:
+            return static_cast<Int128>(static_cast<const ColumnInt64*>(data_.get())->At(i));
+        case Type::Int128:
+            return static_cast<const ColumnInt128*>(data_.get())->At(i);
+        default:
+            throw ValidationError("Invalid data_ column type in ColumnDecimal");
+    }
+}
+
+void ColumnDecimal::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+void ColumnDecimal::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnDecimal>()) {
+        data_->Append(col->data_);
+    }
+}
+
+bool ColumnDecimal::LoadBody(InputStream * input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnDecimal::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+void ColumnDecimal::Clear() {
+    data_->Clear();
+}
+
+size_t ColumnDecimal::Size() const {
+    return data_->Size();
+}
+
+ColumnRef ColumnDecimal::Slice(size_t begin, size_t len) const {
+    // coundn't use std::make_shared since this c-tor is private
+    return ColumnRef{new ColumnDecimal(type_, data_->Slice(begin, len))};
+}
+
+ColumnRef ColumnDecimal::CloneEmpty() const {
+    // coundn't use std::make_shared since this c-tor is private
+    return ColumnRef{new ColumnDecimal(type_, data_->CloneEmpty())};
+}
+
+void ColumnDecimal::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnDecimal &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnDecimal::GetItem(size_t index) const {
+    return ItemView{GetType().GetCode(), data_->GetItem(index)};
+}
+
+size_t ColumnDecimal::GetScale() const
+{
+    return type_->As<DecimalType>()->GetScale();
+}
+
+size_t ColumnDecimal::GetPrecision() const
+{
+    return type_->As<DecimalType>()->GetPrecision();
+}
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/decimal.h b/lib/clickhouse-cpp/clickhouse/columns/decimal.h
new file mode 100644
index 0000000..28425ab
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/decimal.h
@@ -0,0 +1,50 @@
+#pragma once
+
+#include "column.h"
+#include "numeric.h"
+
+namespace clickhouse {
+
+/**
+ * Represents a column of decimal type.
+ */
+class ColumnDecimal : public Column {
+public:
+    using ValueType = Int128;
+
+    ColumnDecimal(size_t precision, size_t scale);
+
+    void Append(const Int128& value);
+    void Append(const std::string& value);
+
+    Int128 At(size_t i) const;
+    inline auto operator[](size_t i) const { return At(i); }
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+    void Append(ColumnRef column) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
+    void SaveBody(OutputStream* output) override;
+    void Clear() override;
+    size_t Size() const override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+    ItemView GetItem(size_t index) const override;
+
+    size_t GetScale() const;
+    size_t GetPrecision() const;
+
+private:
+    /// Depending on a precision it can be one of:
+    ///  - ColumnInt32
+    ///  - ColumnInt64
+    ///  - ColumnInt128
+    ColumnRef data_;
+    Type::Code data_type_code_;
+
+    explicit ColumnDecimal(TypeRef type, ColumnRef data);
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/enum.cpp b/lib/clickhouse-cpp/clickhouse/columns/enum.cpp
index e8ae38c..43fab89 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/enum.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/enum.cpp
@@ -1,6 +1,10 @@
 #include "enum.h"
 #include "utils.h"
 
+#include "../base/input.h"
+#include "../base/output.h"
+#include "../base/wire_format.h"
+
 namespace clickhouse {
 
 template <typename T>
@@ -16,17 +20,24 @@ ColumnEnum<T>::ColumnEnum(TypeRef type, const std::vector<T>& data)
 {
 }
 
+template <typename T>
+ColumnEnum<T>::ColumnEnum(TypeRef type, std::vector<T>&& data)
+    : Column(type)
+    , data_(std::move(data))
+{
+}
+
 template <typename T>
 void ColumnEnum<T>::Append(const T& value, bool checkValue) {
     if  (checkValue) {
-        // TODO type_->HasEnumValue(value), "Enum type doesn't have value " + std::to_string(value);
+        // TODO: type_->HasEnumValue(value), "Enum type doesn't have value " + std::to_string(value);
     }
     data_.push_back(value);
 }
 
 template <typename T>
 void ColumnEnum<T>::Append(const std::string& name) {
-    data_.push_back(EnumType(type_).GetEnumValue(name));
+    data_.push_back(static_cast<T>(type_->As<EnumType>()->GetEnumValue(name)));
 }
 
 template <typename T>
@@ -40,13 +51,8 @@ const T& ColumnEnum<T>::At(size_t n) const {
 }
 
 template <typename T>
-const std::string ColumnEnum<T>::NameAt(size_t n) const {
-    return EnumType(type_).GetEnumName(data_.at(n));
-}
-
-template <typename T>
-const T& ColumnEnum<T>::operator[] (size_t n) const {
-    return data_[n];
+std::string_view ColumnEnum<T>::NameAt(size_t n) const {
+    return type_->As<EnumType>()->GetEnumName(data_.at(n));
 }
 
 template <typename T>
@@ -59,7 +65,12 @@ void ColumnEnum<T>::SetAt(size_t n, const T& value, bool checkValue) {
 
 template <typename T>
 void ColumnEnum<T>::SetNameAt(size_t n, const std::string& name) {
-    data_.at(n) = EnumType(type_).GetEnumValue(name);
+    data_.at(n) = static_cast<T>(type_->As<EnumType>()->GetEnumValue(name));
+}
+
+template<typename T>
+void ColumnEnum<T>::Reserve(size_t new_cap) {
+    data_.reserve(new_cap);
 }
 
 template <typename T>
@@ -70,14 +81,14 @@ void ColumnEnum<T>::Append(ColumnRef column) {
 }
 
 template <typename T>
-bool ColumnEnum<T>::Load(CodedInputStream* input, size_t rows) {
+bool ColumnEnum<T>::LoadBody(InputStream* input, size_t rows) {
     data_.resize(rows);
-    return input->ReadRaw(data_.data(), data_.size() * sizeof(T));
+    return WireFormat::ReadBytes(*input, data_.data(), data_.size() * sizeof(T));
 }
 
 template <typename T>
-void ColumnEnum<T>::Save(CodedOutputStream* output) {
-    output->WriteRaw(data_.data(), data_.size() * sizeof(T));
+void ColumnEnum<T>::SaveBody(OutputStream* output) {
+    WireFormat::WriteBytes(*output, data_.data(), data_.size() * sizeof(T));
 }
 
 template <typename T>
@@ -86,10 +97,26 @@ size_t ColumnEnum<T>::Size() const {
 }
 
 template <typename T>
-ColumnRef ColumnEnum<T>::Slice(size_t begin, size_t len) {
+ColumnRef ColumnEnum<T>::Slice(size_t begin, size_t len) const {
     return std::make_shared<ColumnEnum<T>>(type_, SliceVector(data_, begin, len));
 }
 
+template <typename T>
+ColumnRef ColumnEnum<T>::CloneEmpty() const {
+    return std::make_shared<ColumnEnum<T>>(type_);
+}
+
+template <typename T>
+void ColumnEnum<T>::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnEnum<T> &>(other);
+    data_.swap(col.data_);
+}
+
+template <typename T>
+ItemView ColumnEnum<T>::GetItem(size_t index) const {
+    return ItemView{type_->GetCode(), data_[index]};
+}
+
 template class ColumnEnum<int8_t>;
 template class ColumnEnum<int16_t>;
 
diff --git a/lib/clickhouse-cpp/clickhouse/columns/enum.h b/lib/clickhouse-cpp/clickhouse/columns/enum.h
index 4877866..43900f6 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/enum.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/enum.h
@@ -8,8 +8,11 @@ namespace clickhouse {
 template <typename T>
 class ColumnEnum : public Column {
 public:
+    using ValueType = T;
+
     ColumnEnum(TypeRef type);
     ColumnEnum(TypeRef type, const std::vector<T>& data);
+    ColumnEnum(TypeRef type, std::vector<T>&& data);
 
     /// Appends one element to the end of column.
     void Append(const T& value, bool checkValue = false);
@@ -17,33 +20,40 @@ class ColumnEnum : public Column {
 
     /// Returns element at given row number.
     const T& At(size_t n) const;
-    const std::string NameAt(size_t n) const;
+    std::string_view NameAt(size_t n) const;
 
     /// Returns element at given row number.
-    const T& operator[] (size_t n) const;
+    inline const T& operator[] (size_t n) const { return At(n); }
 
     /// Set element at given row number.
     void SetAt(size_t n, const T& value, bool checkValue = false);
     void SetNameAt(size_t n, const std::string& name);
 
 public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
-    
-    /// Clear column data .
+    void SaveBody(OutputStream* output) override;
+
+    /// Clear column data.
     void Clear() override;
 
     /// Returns count of rows in the column.
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
 
 private:
     std::vector<T> data_;
diff --git a/lib/clickhouse-cpp/clickhouse/columns/factory.cpp b/lib/clickhouse-cpp/clickhouse/columns/factory.cpp
index 47dddf9..460d66f 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/factory.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/factory.cpp
@@ -2,20 +2,53 @@
 
 #include "array.h"
 #include "date.h"
+#include "decimal.h"
 #include "enum.h"
+#include "geo.h"
+#include "ip4.h"
+#include "ip6.h"
+#include "lowcardinality.h"
+#include "lowcardinalityadaptor.h"
+#include "map.h"
+#include "nothing.h"
 #include "nullable.h"
 #include "numeric.h"
 #include "string.h"
+#include "./time.h" // `./` avoids possible conflicts with standard C time.h
 #include "tuple.h"
 #include "uuid.h"
 
+
 #include "../types/type_parser.h"
 
+#include "../exceptions.h"
+
+#include <stdexcept>
+#include <string>
+
 namespace clickhouse {
 namespace {
 
+// Like Python's list's []:
+// * 0 - first element
+// * 1 - second element
+// * -1 - last element
+// * -2 - one before last, etc.
+const auto& GetASTChildElement(const TypeAst & ast, int position) {
+    if (static_cast<size_t>(abs(position)) >= ast.elements.size())
+        throw ValidationError("AST child element index out of bounds: " + std::to_string(position));
+
+    if (position < 0)
+        position = static_cast<int>(ast.elements.size() + position);
+
+    return ast.elements[static_cast<size_t>(position)];
+}
+
 static ColumnRef CreateTerminalColumn(const TypeAst& ast) {
     switch (ast.code) {
+    case Type::Void:
+        return std::make_shared<ColumnNothing>();
+
     case Type::UInt8:
         return std::make_shared<ColumnUInt8>();
     case Type::UInt16:
@@ -33,41 +66,92 @@ static ColumnRef CreateTerminalColumn(const TypeAst& ast) {
         return std::make_shared<ColumnInt32>();
     case Type::Int64:
         return std::make_shared<ColumnInt64>();
-
-    case Type::UUID:
-        return std::make_shared<ColumnUUID>();
+    case Type::Int128:
+        return std::make_shared<ColumnInt128>();
+    case Type::UInt128:
+        return std::make_shared<ColumnUInt128>();
 
     case Type::Float32:
         return std::make_shared<ColumnFloat32>();
     case Type::Float64:
         return std::make_shared<ColumnFloat64>();
 
+    case Type::Decimal:
+        return std::make_shared<ColumnDecimal>(GetASTChildElement(ast, 0).value, GetASTChildElement(ast, -1).value);
+    case Type::Decimal32:
+        return std::make_shared<ColumnDecimal>(9, GetASTChildElement(ast, 0).value);
+    case Type::Decimal64:
+        return std::make_shared<ColumnDecimal>(18, GetASTChildElement(ast, 0).value);
+    case Type::Decimal128:
+        return std::make_shared<ColumnDecimal>(38, GetASTChildElement(ast, 0).value);
+
     case Type::String:
         return std::make_shared<ColumnString>();
     case Type::FixedString:
-        return std::make_shared<ColumnFixedString>(ast.elements.front().value);
+        return std::make_shared<ColumnFixedString>(GetASTChildElement(ast, 0).value);
 
     case Type::DateTime:
-        return std::make_shared<ColumnDateTime>();
+        if (ast.elements.empty()) {
+            return std::make_shared<ColumnDateTime>();
+        } else {
+            return std::make_shared<ColumnDateTime>(GetASTChildElement(ast, 0).value_string);
+        }
+    case Type::DateTime64:
+        if (ast.elements.empty()) {
+            return nullptr;
+        }
+        if (ast.elements.size() == 1) {
+            return std::make_shared<ColumnDateTime64>(ast.elements[0].value);
+        } else {
+            return std::make_shared<ColumnDateTime64>(ast.elements[0].value, ast.elements[1].value_string);
+        }
     case Type::Date:
         return std::make_shared<ColumnDate>();
+    case Type::Date32:
+        return std::make_shared<ColumnDate32>();
+    case Type::IPv4:
+        return std::make_shared<ColumnIPv4>();
+    case Type::IPv6:
+        return std::make_shared<ColumnIPv6>();
+
+    case Type::UUID:
+        return std::make_shared<ColumnUUID>();
+
+    case Type::Point:
+        return std::make_shared<ColumnPoint>();
+
+    case Type::Ring:
+        return std::make_shared<ColumnRing>();
+
+    case Type::Polygon:
+        return std::make_shared<ColumnPolygon>();
+
+    case Type::MultiPolygon:
+        return std::make_shared<ColumnMultiPolygon>();
 
+    case Type::Time:
+        return std::make_shared<ColumnTime>();
+    case Type::Time64:
+        if (ast.elements.empty()) {
+            return nullptr;
+        }
+        return std::make_shared<ColumnTime64>(GetASTChildElement(ast, 0).value);
     default:
         return nullptr;
     }
 }
 
-static ColumnRef CreateColumnFromAst(const TypeAst& ast) {
+static ColumnRef CreateColumnFromAst(const TypeAst& ast, CreateColumnByTypeSettings settings) {
     switch (ast.meta) {
         case TypeAst::Array: {
             return std::make_shared<ColumnArray>(
-                CreateColumnFromAst(ast.elements.front())
+                CreateColumnFromAst(GetASTChildElement(ast, 0), settings)
             );
         }
 
         case TypeAst::Nullable: {
             return std::make_shared<ColumnNullable>(
-                CreateColumnFromAst(ast.elements.front()),
+                CreateColumnFromAst(GetASTChildElement(ast, 0), settings),
                 std::make_shared<ColumnUInt8>()
             );
         }
@@ -79,8 +163,9 @@ static ColumnRef CreateColumnFromAst(const TypeAst& ast) {
         case TypeAst::Tuple: {
             std::vector<ColumnRef> columns;
 
+            columns.reserve(ast.elements.size());
             for (const auto& elem : ast.elements) {
-                if (auto col = CreateColumnFromAst(elem)) {
+                if (auto col = CreateColumnFromAst(elem, settings)) {
                     columns.push_back(col);
                 } else {
                     return nullptr;
@@ -92,10 +177,17 @@ static ColumnRef CreateColumnFromAst(const TypeAst& ast) {
 
         case TypeAst::Enum: {
             std::vector<Type::EnumItem> enum_items;
+            //ast.elements.size() minimum is 1.
+            if ((ast.elements.size() % 2) != 0) {
+                throw ValidationError(ast.name + " content is not correct");
+            }
 
-            for (const auto& elem : ast.elements) {
-                enum_items.push_back(
-                    Type::EnumItem{elem.name, (int16_t)elem.value});
+            enum_items.reserve(ast.elements.size() / 2);
+            for (size_t i = 0; i < ast.elements.size(); i += 2) {
+                enum_items.push_back(Type::EnumItem{
+                    ast.elements[i].value_string,
+                    static_cast<int16_t>(ast.elements[i + 1].value)
+                });
             }
 
             if (ast.code == Type::Enum8) {
@@ -109,9 +201,69 @@ static ColumnRef CreateColumnFromAst(const TypeAst& ast) {
             }
             break;
         }
+        case TypeAst::LowCardinality: {
+            const auto nested = GetASTChildElement(ast, 0);
+            if (settings.low_cardinality_as_wrapped_column) {
+                switch (nested.code) {
+                    // TODO (nemkov): update this to maximize code reuse.
+                    case Type::String:
+                        return std::make_shared<LowCardinalitySerializationAdaptor<ColumnString>>();
+                    case Type::FixedString:
+                        return std::make_shared<LowCardinalitySerializationAdaptor<ColumnFixedString>>(GetASTChildElement(nested, 0).value);
+                    case Type::Nullable:
+                        throw UnimplementedError("LowCardinality(" + nested.name + ") is not supported with LowCardinalityAsWrappedColumn on");
+                    default:
+                        throw UnimplementedError("LowCardinality(" + nested.name + ") is not supported");
+                }
+            }
+            else {
+                switch (nested.code) {
+                    // TODO (nemkov): update this to maximize code reuse.
+                    case Type::String:
+                        return std::make_shared<ColumnLowCardinalityT<ColumnString>>();
+                    case Type::FixedString:
+                        return std::make_shared<ColumnLowCardinalityT<ColumnFixedString>>(GetASTChildElement(nested, 0).value);
+                    case Type::Nullable:
+                        return std::make_shared<ColumnLowCardinality>(
+                            std::make_shared<ColumnNullable>(
+                                CreateColumnFromAst(GetASTChildElement(nested, 0), settings),
+                                std::make_shared<ColumnUInt8>()
+                            )
+                        );
+                    default:
+                        throw UnimplementedError("LowCardinality(" + nested.name + ") is not supported");
+                }
+            }
+        }
+        case TypeAst::SimpleAggregateFunction: {
+            return CreateTerminalColumn(GetASTChildElement(ast, -1));
+        }
+
+        case TypeAst::Map: {
+            if (ast.elements.size() != 2) {
+                throw ValidationError(ast.name + " content is not correct");
+            }
+
+            std::vector<ColumnRef> columns;
+
+            columns.reserve(ast.elements.size());
+            for (const auto& elem : ast.elements) {
+                if (auto col = CreateColumnFromAst(elem, settings)) {
+                    columns.push_back(col);
+                } else {
+                    return nullptr;
+                }
+            }
+
+            return std::make_shared<ColumnMap>(
+                std::make_shared<ColumnArray>(
+                    std::make_shared<ColumnTuple>(columns)));
+        }
 
+        case TypeAst::Assign:
         case TypeAst::Null:
         case TypeAst::Number:
+        case TypeAst::String:
             break;
     }
 
@@ -121,10 +273,10 @@ static ColumnRef CreateColumnFromAst(const TypeAst& ast) {
 } // namespace
 
 
-ColumnRef CreateColumnByType(const std::string& type_name) {
+ColumnRef CreateColumnByType(const std::string& type_name, CreateColumnByTypeSettings settings) {
     auto ast = ParseTypeName(type_name);
     if (ast != nullptr) {
-        return CreateColumnFromAst(*ast);
+        return CreateColumnFromAst(*ast, settings);
     }
 
     return nullptr;
diff --git a/lib/clickhouse-cpp/clickhouse/columns/factory.h b/lib/clickhouse-cpp/clickhouse/columns/factory.h
index 05c7358..9d2bed1 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/factory.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/factory.h
@@ -4,6 +4,11 @@
 
 namespace clickhouse {
 
-ColumnRef CreateColumnByType(const std::string& type_name);
+struct CreateColumnByTypeSettings
+{
+    bool low_cardinality_as_wrapped_column = false;
+};
+
+ColumnRef CreateColumnByType(const std::string& type_name, CreateColumnByTypeSettings settings = {});
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/geo.cpp b/lib/clickhouse-cpp/clickhouse/columns/geo.cpp
new file mode 100644
index 0000000..fa98773
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/geo.cpp
@@ -0,0 +1,108 @@
+#include "geo.h"
+
+#include "utils.h"
+
+namespace {
+using namespace ::clickhouse;
+
+template <Type::Code type_code>
+TypeRef CreateGeoType() {
+    if constexpr (type_code == Type::Code::Point) {
+        return Type::CreatePoint();
+    } else if constexpr (type_code == Type::Code::Ring) {
+        return Type::CreateRing();
+    } else if constexpr (type_code == Type::Code::Polygon) {
+        return Type::CreatePolygon();
+    } else if constexpr (type_code == Type::Code::MultiPolygon) {
+        return Type::CreateMultiPolygon();
+    }
+}
+
+template <typename ColumnType>
+std::shared_ptr<ColumnType> CreateColumn() {
+    if constexpr (std::is_same_v<ColumnType, ColumnTupleT<ColumnFloat64, ColumnFloat64>>) {
+        return std::make_shared<ColumnTupleT<ColumnFloat64, ColumnFloat64>>(
+            std::make_tuple(std::make_shared<ColumnFloat64>(), std::make_shared<ColumnFloat64>()));
+    } else {
+        return std::make_shared<ColumnType>();
+    }
+}
+
+}  // namespace
+
+namespace clickhouse {
+
+template <typename NestedColumnType, Type::Code type_code>
+ColumnGeo<NestedColumnType, type_code>::ColumnGeo()
+    : Column(CreateGeoType<type_code>()),
+      data_(CreateColumn<NestedColumnType>()) {
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+ColumnGeo<NestedColumnType, type_code>::ColumnGeo(ColumnRef data)
+    : Column(CreateGeoType<type_code>())
+    , data_(WrapColumn<NestedColumnType>(std::move(data))) {
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+void ColumnGeo<NestedColumnType, type_code>::Clear() {
+    data_->Clear();
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+const typename ColumnGeo<NestedColumnType, type_code>::ValueType ColumnGeo<NestedColumnType, type_code>::At(size_t n) const {
+    return data_->At(n);
+}
+
+template<typename NestedColumnType, Type::Code type_code>
+void ColumnGeo<NestedColumnType, type_code>::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+void ColumnGeo<NestedColumnType, type_code>::Append(ColumnRef column) {
+    if (auto col = column->template As<ColumnGeo>()) {
+        data_->Append(col->data_->template As<Column>());
+    }
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+bool ColumnGeo<NestedColumnType, type_code>::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+void ColumnGeo<NestedColumnType, type_code>::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+size_t ColumnGeo<NestedColumnType, type_code>::Size() const {
+    return data_->Size();
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+ColumnRef ColumnGeo<NestedColumnType, type_code>::Slice(size_t begin, size_t len) const {
+    return std::make_shared<ColumnGeo>(data_->Slice(begin, len));
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+ColumnRef ColumnGeo<NestedColumnType, type_code>::CloneEmpty() const {
+    return std::make_shared<ColumnGeo>();
+}
+
+template <typename NestedColumnType, Type::Code type_code>
+void ColumnGeo<NestedColumnType, type_code>::Swap(Column& other) {
+    auto& col = dynamic_cast<ColumnGeo&>(other);
+    data_.swap(col.data_);
+}
+
+template class ColumnGeo<ColumnTupleT<ColumnFloat64, ColumnFloat64>, Type::Code::Point>;
+
+template class ColumnGeo<ColumnArrayT<ColumnPoint>, Type::Code::Ring>;
+
+template class ColumnGeo<ColumnArrayT<ColumnRing>, Type::Code::Polygon>;
+
+template class ColumnGeo<ColumnArrayT<ColumnPolygon>, Type::Code::MultiPolygon>;
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/columns/geo.h b/lib/clickhouse-cpp/clickhouse/columns/geo.h
new file mode 100644
index 0000000..1b12973
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/geo.h
@@ -0,0 +1,79 @@
+#pragma once
+
+#include "array.h"
+#include "column.h"
+#include "numeric.h"
+#include "tuple.h"
+
+namespace clickhouse {
+
+template <typename NestedColumnType, Type::Code type_code>
+class ColumnGeo : public Column {
+public:
+    using ValueType = typename NestedColumnType::ValueType;
+
+    ColumnGeo();
+
+    explicit ColumnGeo(ColumnRef data);
+
+    /// Appends one element to the end of column.
+    template <typename T = ValueType>
+    void Append(const T& value) {
+        data_->Append(value);
+    }
+
+    /// Returns element at given row number.
+    const ValueType At(size_t n) const;
+
+    /// Returns element at given row number.
+    inline const ValueType operator[](size_t n) const { return At(n); }
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+private:
+    std::shared_ptr<NestedColumnType> data_;
+};
+
+// /**
+//  * Represents a Point column.
+//  */
+using ColumnPoint = ColumnGeo<ColumnTupleT<ColumnFloat64, ColumnFloat64>, Type::Code::Point>;
+
+/**
+ * Represents a Ring column.
+ */
+using ColumnRing = ColumnGeo<ColumnArrayT<ColumnPoint>, Type::Code::Ring>;
+
+/**
+ * Represents a Polygon column.
+ */
+using ColumnPolygon = ColumnGeo<ColumnArrayT<ColumnRing>, Type::Code::Polygon>;
+
+/**
+ * Represents a MultiPolygon column.
+ */
+using ColumnMultiPolygon = ColumnGeo<ColumnArrayT<ColumnPolygon>, Type::Code::MultiPolygon>;
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/columns/ip4.cpp b/lib/clickhouse-cpp/clickhouse/columns/ip4.cpp
new file mode 100644
index 0000000..78ec5bb
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/ip4.cpp
@@ -0,0 +1,117 @@
+#include "ip4.h"
+
+#include "../base/socket.h" // for platform-specific IPv4-related functions
+#include <stdexcept>
+
+namespace clickhouse {
+
+ColumnIPv4::ColumnIPv4()
+    : Column(Type::CreateIPv4())
+    , data_(std::make_shared<ColumnUInt32>())
+{
+}
+
+ColumnIPv4::ColumnIPv4(ColumnRef data)
+    : Column(Type::CreateIPv4())
+    , data_(data ? data->As<ColumnUInt32>() : nullptr)
+{
+    if (!data_)
+        throw ValidationError("Expecting ColumnUInt32, got " + (data ? data->GetType().GetName() : "null"));
+}
+
+ColumnIPv4::ColumnIPv4(std::vector<uint32_t>&& data)
+    : Column(Type::CreateIPv4())
+{
+    for (auto& addr : data) {
+        addr = htonl(addr);
+    }
+    data_ = std::make_shared<ColumnUInt32>(std::move(data));
+}
+
+void ColumnIPv4::Append(const std::string& str) {
+    uint32_t address;
+    if (inet_pton(AF_INET, str.c_str(), &address) != 1)
+        throw ValidationError("invalid IPv4 format, ip: " + str);
+    data_->Append(htonl(address));
+}
+
+void ColumnIPv4::Append(uint32_t ip) {
+    data_->Append(htonl(ip));
+}
+
+void ColumnIPv4::Append(in_addr ip) {
+    data_->Append(htonl(ip.s_addr));
+}
+
+void ColumnIPv4::Clear() {
+    data_->Clear();
+}
+
+in_addr ColumnIPv4::At(size_t n) const {
+    in_addr addr;
+    addr.s_addr = ntohl(data_->At(n));
+    return addr;
+}
+
+in_addr ColumnIPv4::operator [] (size_t n) const {
+    in_addr addr;
+    addr.s_addr = ntohl(data_->operator[](n));
+    return addr;
+}
+
+std::string ColumnIPv4::AsString(size_t n) const {
+    const auto& addr = this->At(n);
+    auto tmp_addr = addr;
+
+    char buf[INET_ADDRSTRLEN];
+    const char* ip_str = inet_ntop(AF_INET, &tmp_addr, buf, INET_ADDRSTRLEN);
+
+    if (ip_str == nullptr) {
+        throw std::system_error(
+                std::error_code(errno, std::generic_category()),
+                "Invalid IPv4 data");
+    }
+
+    return ip_str;
+}
+
+void ColumnIPv4::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+void ColumnIPv4::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnIPv4>()) {
+        data_->Append(col->data_);
+    }
+}
+
+bool ColumnIPv4::LoadBody(InputStream * input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnIPv4::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+size_t ColumnIPv4::Size() const {
+    return data_->Size();
+}
+
+ColumnRef ColumnIPv4::Slice(size_t begin, size_t len) const {
+    return std::make_shared<ColumnIPv4>(data_->Slice(begin, len));
+}
+
+ColumnRef ColumnIPv4::CloneEmpty() const {
+    return std::make_shared<ColumnIPv4>(data_->CloneEmpty());
+}
+
+void ColumnIPv4::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnIPv4 &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnIPv4::GetItem(size_t index) const {
+    return ItemView(Type::IPv4, data_->GetItem(index));
+}
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/ip4.h b/lib/clickhouse-cpp/clickhouse/columns/ip4.h
new file mode 100644
index 0000000..2253e30
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/ip4.h
@@ -0,0 +1,71 @@
+#pragma once
+
+#include "numeric.h"
+
+struct in_addr;
+
+namespace clickhouse {
+
+class ColumnIPv4 : public Column {
+public:
+    using DataType = in_addr;
+    using ValueType = in_addr;
+
+    ColumnIPv4();
+    /** Takes ownership of the data, expects ColumnUInt32.
+     * Modifying memory pointed by `data` from outside is UB.
+     *
+     * TODO: deprecate and remove as it is too dangerous and error-prone.
+     */
+    explicit ColumnIPv4(ColumnRef data);
+
+    explicit ColumnIPv4(std::vector<uint32_t>&& data);
+
+    /// Appends one element to the column.
+    void Append(const std::string& ip);
+
+    /// @params ip numeric value with host byte order.
+    void Append(uint32_t ip);
+
+    ///
+    void Append(in_addr ip);
+
+    /// Returns element at given row number.
+    in_addr At(size_t n) const;
+
+    /// Returns element at given row number.
+    in_addr operator [] (size_t n) const;
+
+    std::string AsString(size_t n) const;
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
+
+private:
+    std::shared_ptr<ColumnUInt32> data_;
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/ip6.cpp b/lib/clickhouse-cpp/clickhouse/columns/ip6.cpp
new file mode 100644
index 0000000..55ba2d4
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/ip6.cpp
@@ -0,0 +1,108 @@
+#include "ip6.h"
+#include "../base/socket.h" // for IPv6 platform-specific stuff
+#include "../exceptions.h"
+
+#include <stdexcept>
+
+namespace clickhouse {
+
+static_assert(sizeof(struct in6_addr) == 16, "sizeof in6_addr should be 16 bytes");
+
+ColumnIPv6::ColumnIPv6()
+    : Column(Type::CreateIPv6())
+    , data_(std::make_shared<ColumnFixedString>(16))
+{
+}
+
+ColumnIPv6::ColumnIPv6(ColumnRef data)
+    : Column(Type::CreateIPv6())
+    , data_(data ? data->As<ColumnFixedString>() : nullptr)
+{
+    if (!data_ || data_->FixedSize() != sizeof(in6_addr))
+        throw ValidationError("Expecting ColumnFixedString(16), got " + (data ? data->GetType().GetName() : "null"));
+}
+
+void ColumnIPv6::Append(const std::string_view& str) {
+    unsigned char buf[16];
+    if (inet_pton(AF_INET6, str.data(), buf) != 1) {
+        throw ValidationError("invalid IPv6 format, ip: " + std::string(str));
+    }
+    data_->Append(std::string_view((const char*)buf, 16));
+}
+
+void ColumnIPv6::Append(const in6_addr* addr) {
+    data_->Append(std::string_view((const char*)addr->s6_addr, 16));
+}
+
+void ColumnIPv6::Append(const in6_addr& addr) {
+    Append(&addr);
+}
+
+void ColumnIPv6::Clear() {
+    data_->Clear();
+}
+
+std::string ColumnIPv6::AsString (size_t n) const {
+    const auto& addr = this->At(n);
+    auto tmp_addr = addr;
+
+    char buf[INET6_ADDRSTRLEN];
+    const char* ip_str = inet_ntop(AF_INET6, &tmp_addr, buf, INET6_ADDRSTRLEN);
+
+    if (ip_str == nullptr) {
+        throw std::system_error(
+                std::error_code(errno, std::generic_category()),
+                "Invalid IPv6 data");
+    }
+
+    return ip_str;
+}
+
+in6_addr ColumnIPv6::At(size_t n) const {
+    return *reinterpret_cast<const in6_addr*>(data_->At(n).data());
+}
+
+in6_addr ColumnIPv6::operator [] (size_t n) const {
+    return *reinterpret_cast<const in6_addr*>(data_->At(n).data());
+}
+
+void ColumnIPv6::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+void ColumnIPv6::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnIPv6>()) {
+        data_->Append(col->data_);
+    }
+}
+
+bool ColumnIPv6::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnIPv6::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+size_t ColumnIPv6::Size() const {
+    return data_->Size();
+}
+
+ColumnRef ColumnIPv6::Slice(size_t begin, size_t len) const {
+    return std::make_shared<ColumnIPv6>(data_->Slice(begin, len));
+}
+
+ColumnRef ColumnIPv6::CloneEmpty() const {
+    return std::make_shared<ColumnIPv6>(data_->CloneEmpty());
+}
+
+void ColumnIPv6::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnIPv6 &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnIPv6::GetItem(size_t index) const {
+    return ItemView{Type::IPv6, data_->GetItem(index)};
+}
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/ip6.h b/lib/clickhouse-cpp/clickhouse/columns/ip6.h
new file mode 100644
index 0000000..41af0d5
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/ip6.h
@@ -0,0 +1,66 @@
+#pragma once
+
+#include "string.h"
+#include <memory>
+
+struct in6_addr;
+
+namespace clickhouse {
+
+class ColumnIPv6 : public Column {
+public:
+    using DataType = in6_addr;
+    using ValueType = in6_addr;
+
+    ColumnIPv6();
+    /** Takes ownership of the data, expects ColumnFixedString.
+     * Modifying memory pointed by `data` from outside is UB.
+     *
+     * TODO: deprecate and remove as it is too dangerous and error-prone.
+     */
+    explicit ColumnIPv6(ColumnRef data);
+
+    /// Appends one element to the column.
+    void Append(const std::string_view& str);
+
+    void Append(const in6_addr* addr);
+    void Append(const in6_addr& addr);
+
+    /// Returns element at given row number.
+    in6_addr At(size_t n) const;
+
+    /// Returns element at given row number.
+    in6_addr operator [] (size_t n) const;
+
+    std::string AsString(size_t n) const;
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+    ItemView GetItem(size_t index) const override;
+
+private:
+    std::shared_ptr<ColumnFixedString> data_;
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/itemview.cpp b/lib/clickhouse-cpp/clickhouse/columns/itemview.cpp
new file mode 100644
index 0000000..0116070
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/itemview.cpp
@@ -0,0 +1,101 @@
+#include "../columns/itemview.h"
+
+#include <algorithm>
+#include <sstream>
+
+namespace {
+
+template <typename Container>
+std::string ContainerToString(Container container, const char * separator = ", ") {
+    std::stringstream sstr;
+    const auto end = std::end(container);
+    for (auto i = std::begin(container); i != end; /*intentionally no ++i*/) {
+        const auto & elem = *i;
+        sstr << elem;
+
+        if (++i != end) {
+            sstr << separator;
+        }
+    }
+
+    return sstr.str();
+}
+
+}
+
+namespace clickhouse {
+
+void ItemView::ValidateData(Type::Code type, DataType data) {
+
+    auto AssertSize = [type, &data](std::initializer_list<int> allowed_sizes) -> void {
+        const auto end = std::end(allowed_sizes);
+        if (std::find(std::begin(allowed_sizes), end, static_cast<int>(data.size())) == end) {
+            throw AssertionError(std::string("ItemView value size mismatch for ")
+                    + Type::TypeName(type)
+                    + " expected: " + ContainerToString(allowed_sizes, " or ")
+                    + ", got: " + std::to_string(data.size()));
+        }
+    };
+
+    switch (type) {
+        case Type::Code::Void:
+            return AssertSize({0});
+
+        case Type::Code::Int8:
+        case Type::Code::UInt8:
+        case Type::Code::Enum8:
+            return AssertSize({1});
+
+        case Type::Code::Int16:
+        case Type::Code::UInt16:
+        case Type::Code::Date:
+        case Type::Code::Enum16:
+            return AssertSize({2});
+
+        case Type::Code::Int32:
+        case Type::Code::UInt32:
+        case Type::Code::Float32:
+        case Type::Code::DateTime:
+        case Type::Code::Date32:
+        case Type::Code::IPv4:
+        case Type::Code::Decimal32:
+        case Type::Code::Time:
+            return AssertSize({4});
+
+        case Type::Code::Int64:
+        case Type::Code::UInt64:
+        case Type::Code::Float64:
+        case Type::Code::DateTime64:
+        case Type::Code::Decimal64:
+        case Type::Code::Time64:
+            return AssertSize({8});
+
+        case Type::Code::String:
+        case Type::Code::FixedString:
+            // value can be of any size
+            return;
+
+        case Type::Code::Array:
+        case Type::Code::Nullable:
+        case Type::Code::Tuple:
+        case Type::Code::LowCardinality:
+        case Type::Code::Map:
+            throw AssertionError("Unsupported type in ItemView: " + std::string(Type::TypeName(type)));
+
+        case Type::Code::IPv6:
+        case Type::Code::UUID:
+        case Type::Code::UInt128:
+        case Type::Code::Int128:
+        case Type::Code::Decimal128:
+            return AssertSize({16});
+
+        case Type::Code::Decimal:
+            // Could be either Decimal32, Decimal64 or Decimal128
+            return AssertSize({4, 8, 16});
+
+        default:
+            throw UnimplementedError("Unknown type code:" + std::to_string(static_cast<int>(type)));
+    }
+}
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/itemview.h b/lib/clickhouse-cpp/clickhouse/columns/itemview.h
new file mode 100644
index 0000000..199994b
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/itemview.h
@@ -0,0 +1,86 @@
+#pragma once
+
+#include "../types/types.h"
+#include "../exceptions.h"
+
+#include <string_view>
+#include <stdexcept>
+#include <type_traits>
+
+namespace clickhouse {
+
+/** ItemView is a view on a data stored in Column, safe-ish interface for reading values from Column.
+ *
+ * Data is not owned (hence the name View) and will be invalidated on column update, load
+ * or destruction (basically on calling any non-const method of Column).
+ * `type` reflects what is stored in `data` and can be almost any value-type
+ * (except Nullable, Array, Tuple, LowCardinality).
+ *
+ */
+struct ItemView {
+    using DataType = std::string_view;
+
+    const Type::Code type;
+    const DataType data;
+
+private:
+    template <typename T>
+    inline auto ConvertToStorageValue(const T& t) {
+        if constexpr (std::is_same_v<std::string_view, T> || std::is_same_v<std::string, T>) {
+            return std::string_view{t};
+        } else if constexpr (std::is_fundamental_v<T> || std::is_same_v<Int128, std::decay_t<T>> || std::is_same_v<UInt128, std::decay_t<T>>) {
+            return std::string_view{reinterpret_cast<const char*>(&t), sizeof(T)};
+        } else {
+            static_assert(!std::is_same_v<T, T>, "Unknown type, which can't be stored in ItemView");
+            return;
+        }
+    }
+
+public:
+    ItemView(Type::Code type, DataType data)
+        : type(type),
+          data(data)
+    {
+        ValidateData(type, data);
+    }
+
+    ItemView(Type::Code type, ItemView other)
+        : type(type),
+          data(other.data)
+    {
+        ValidateData(type, data);
+    }
+
+    explicit ItemView()
+        : ItemView(Type::Void, std::string_view{})
+    {}
+
+    template <typename T>
+    explicit ItemView(Type::Code type, const T & value)
+        : ItemView(type, ConvertToStorageValue(value))
+    {}
+
+    template <typename T>
+    auto get() const {
+        using ValueType = std::remove_cv_t<std::decay_t<T>>;
+        if constexpr (std::is_same_v<std::string_view, ValueType> || std::is_same_v<std::string, ValueType>) {
+            return data;
+        } else if constexpr (std::is_fundamental_v<ValueType> || std::is_same_v<Int128, ValueType> || std::is_same_v<UInt128, ValueType>) {
+            if (sizeof(ValueType) == data.size()) {
+                return *reinterpret_cast<const T*>(data.data());
+            } else {
+                throw AssertionError("Incompatitable value type and size. Requested size: "
+                        + std::to_string(sizeof(ValueType)) + " stored size: " + std::to_string(data.size()));
+            }
+        }
+    }
+
+    inline std::string_view AsBinaryData() const {
+        return data;
+    }
+
+    // Validate that value matches type, will throw an exception if validation fails.
+    static void ValidateData(Type::Code type, DataType data);
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/lowcardinality.cpp b/lib/clickhouse-cpp/clickhouse/columns/lowcardinality.cpp
new file mode 100644
index 0000000..9b2675d
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/lowcardinality.cpp
@@ -0,0 +1,527 @@
+#include "lowcardinality.h"
+
+#include "string.h"
+#include "nullable.h"
+#include "../base/wire_format.h"
+
+#include <city.h>
+
+#include <functional>
+#include <string_view>
+#include <type_traits>
+#include <cmath>
+
+#include <cassert>
+
+namespace {
+using namespace clickhouse;
+
+enum KeySerializationVersion {
+    SharedDictionariesWithAdditionalKeys = 1,
+};
+
+enum IndexType {
+    UInt8 = 0,
+    UInt16,
+    UInt32,
+    UInt64,
+};
+
+constexpr uint64_t IndexTypeMask = 0b11111111;
+
+enum IndexFlag {
+    /// Need to read dictionary if it wasn't.
+    NeedGlobalDictionaryBit = 1u << 8u,
+    /// Need to read additional keys. Additional keys are stored before indexes as value N and N keys after them.
+    HasAdditionalKeysBit = 1u << 9u,
+    /// Need to update dictionary. It means that previous granule has different dictionary.
+    NeedUpdateDictionary = 1u << 10u
+};
+
+ColumnRef createIndexColumn(IndexType type) {
+    switch (type) {
+        case IndexType::UInt8:
+            return std::make_shared<ColumnUInt8>();
+        case IndexType::UInt16:
+            return std::make_shared<ColumnUInt16>();
+        case IndexType::UInt32:
+            return std::make_shared<ColumnUInt32>();
+        case IndexType::UInt64:
+            return std::make_shared<ColumnUInt64>();
+    }
+
+    throw ValidationError("Invalid LowCardinality index type value: " + std::to_string(static_cast<uint64_t>(type)));
+}
+
+IndexType indexTypeFromIndexColumn(const Column & index_column) {
+    switch (index_column.Type()->GetCode()) {
+        case Type::UInt8:
+            return IndexType::UInt8;
+        case Type::UInt16:
+            return IndexType::UInt16;
+        case Type::UInt32:
+            return IndexType::UInt32;
+        case Type::UInt64:
+            return IndexType::UInt64;
+        default:
+            throw ValidationError("Invalid index column type for LowCardinality column:" + index_column.Type()->GetName());
+    }
+}
+
+template <typename ResultColumnType, typename ColumnType>
+inline const ResultColumnType & column_down_cast(const ColumnType & c) {
+    return dynamic_cast<const ResultColumnType &>(c);
+}
+
+template <typename ResultColumnType, typename ColumnType>
+inline ResultColumnType & column_down_cast(ColumnType & c) {
+    return dynamic_cast<ResultColumnType &>(c);
+}
+
+// std::visit-ish function to avoid including <variant> header, which is not present in older version of XCode.
+template <typename Vizitor, typename ColumnType>
+inline auto VisitIndexColumn(Vizitor && vizitor, ColumnType && col) {
+    switch (col.Type()->GetCode()) {
+        case Type::UInt8:
+            return vizitor(column_down_cast<ColumnUInt8>(col));
+        case Type::UInt16:
+            return vizitor(column_down_cast<ColumnUInt16>(col));
+        case Type::UInt32:
+            return vizitor(column_down_cast<ColumnUInt32>(col));
+        case Type::UInt64:
+            return vizitor(column_down_cast<ColumnUInt64>(col));
+        default:
+            throw ValidationError("Invalid index column type " + col.GetType().GetName());
+    }
+}
+
+// A special NULL-item, which is expected at pos(0) in dictionary,
+// note that we distinguish empty string from NULL-value.
+inline auto GetNullItemForDictionary(const ColumnRef dictionary) {
+    if (auto n = dictionary->As<ColumnNullable>()) {
+        return ItemView {};
+    } else {
+        return ItemView{dictionary->Type()->GetCode(), std::string_view{}};
+    }
+}
+
+// A special default item, which is expected at pos(0) in dictionary,
+// note that we distinguish empty string from NULL-value.
+inline ItemView GetDefaultItemForDictionary(const ColumnRef dictionary) {
+    if (auto n = dictionary->As<ColumnNullable>()) {
+        return GetDefaultItemForDictionary(n->Nested());
+    } else {
+        return ItemView{dictionary->Type()->GetCode(), std::string_view{}};
+    }
+}
+
+void AppendToDictionary(Column& dictionary, const ItemView & item);
+
+inline void AppendNullableToDictionary(ColumnNullable& nullable, const ItemView & item) {
+    auto nested = nullable.Nested();
+
+    const bool isNullValue = item.type == Type::Void;
+
+    if (isNullValue) {
+        AppendToDictionary(*nested, GetNullItemForDictionary(nested));
+    } else {
+        const auto nestedType = nested->GetType().GetCode();
+        if (nestedType != item.type) {
+            throw ValidationError("Invalid value. Type expected: " + nested->GetType().GetName());
+        }
+
+        AppendToDictionary(*nested, item);
+    }
+
+    nullable.Append(isNullValue);
+}
+
+inline void AppendToDictionary(Column& dictionary, const ItemView & item) {
+    switch (dictionary.GetType().GetCode()) {
+        case Type::FixedString:
+            column_down_cast<ColumnFixedString>(dictionary).Append(item.get<std::string_view>());
+            return;
+        case Type::String:
+            column_down_cast<ColumnString>(dictionary).Append(item.get<std::string_view>());
+            return;
+        case Type::Nullable:
+            AppendNullableToDictionary(column_down_cast<ColumnNullable>(dictionary), item);
+            return;
+        default:
+            throw ValidationError("Unexpected dictionary column type: " + dictionary.GetType().GetName());
+    }
+}
+
+}
+
+namespace clickhouse {
+ColumnLowCardinality::ColumnLowCardinality(ColumnRef dictionary_column)
+    : Column(Type::CreateLowCardinality(dictionary_column->Type())),
+      dictionary_column_(dictionary_column->CloneEmpty()), // safe way to get an column of the same type.
+      index_column_(std::make_shared<ColumnUInt32>()),
+      index_type_code_(Type::UInt32)
+{
+    Setup(dictionary_column);
+}
+
+ColumnLowCardinality::ColumnLowCardinality(std::shared_ptr<ColumnNullable> dictionary_column)
+    : Column(Type::CreateLowCardinality(dictionary_column->Type())),
+      dictionary_column_(dictionary_column->CloneEmpty()), // safe way to get an column of the same type.
+      index_column_(std::make_shared<ColumnUInt32>()),
+      index_type_code_(Type::UInt32)
+{
+    AppendNullItem();
+    Setup(dictionary_column);
+}
+
+ColumnLowCardinality::~ColumnLowCardinality()
+{}
+
+void ColumnLowCardinality::Reserve(size_t new_cap) {
+    // Assumption is that dictionary must be smaller than index.
+    // NOTE(vnemkov): Formula below (`ceil(sqrt(x))`) is a gut-feeling-good-enough estimation,
+    // feel free to replace/adjust if you have better one suported by actual data.
+    dictionary_column_->Reserve(static_cast<size_t>(ceil(sqrt(static_cast<double>(new_cap)))));    
+    index_column_->Reserve(new_cap + 2); // + 1 for null item (at pos 0), + 1 for default item (at pos 1)
+}
+
+void ColumnLowCardinality::Setup(ColumnRef dictionary_column) {
+    AppendDefaultItem();
+
+    if (dictionary_column->Size() != 0) {
+        // Add values, updating index_column_ and unique_items_map_.
+
+        // TODO: it would be possible to eliminate copying
+        // by adding InsertUnsafe(pos, ItemView) method to a Column
+        // (to insert null-item at pos 0),
+        // but that is too much work for now.
+        for (size_t i = 0; i < dictionary_column->Size(); ++i) {
+            AppendUnsafe(dictionary_column->GetItem(i));
+        }
+    }
+}
+
+std::uint64_t ColumnLowCardinality::getDictionaryIndex(std::uint64_t item_index) const {
+    switch (index_type_code_) {
+        case Type::UInt8:
+            return static_cast<const ColumnUInt8&>(*index_column_)[item_index];
+        case Type::UInt16:
+            return static_cast<const ColumnUInt16&>(*index_column_)[item_index];
+        case Type::UInt32:
+            return static_cast<const ColumnUInt32&>(*index_column_)[item_index];
+        case Type::UInt64:
+            return static_cast<const ColumnUInt64&>(*index_column_)[item_index];
+        default:
+            throw ValidationError("Invalid index column type");
+    }
+}
+
+void ColumnLowCardinality::appendIndex(std::uint64_t item_index) {
+    // TODO (nemkov): handle case when index should go from UInt8 to UInt16, etc.
+    switch (index_type_code_) {
+        case Type::UInt8:
+            static_cast<ColumnUInt8&>(*index_column_).Append(static_cast<uint8_t>(item_index));
+            break;
+        case Type::UInt16:
+            static_cast<ColumnUInt16&>(*index_column_).Append(static_cast<uint16_t>(item_index));
+            break;
+        case Type::UInt32:
+            static_cast<ColumnUInt32&>(*index_column_).Append(static_cast<uint32_t>(item_index));
+            break;
+        case Type::UInt64:
+            static_cast<ColumnUInt64&>(*index_column_).Append(static_cast<uint64_t>(item_index));
+            break;
+        default:
+            throw ValidationError("Invalid index column type");
+    }
+}
+
+void ColumnLowCardinality::removeLastIndex() {
+    switch (index_type_code_) {
+        case Type::UInt8: {
+            auto& col = static_cast<ColumnUInt8&>(*index_column_);
+            col.Erase(col.Size() - 1);
+            break;
+        }
+        case Type::UInt16: {
+            auto& col = static_cast<ColumnUInt16&>(*index_column_);
+            col.Erase(col.Size() - 1);
+            break;
+        }
+        case Type::UInt32: {
+            auto& col = static_cast<ColumnUInt32&>(*index_column_);
+            col.Erase(col.Size() - 1);
+            break;
+        }
+        case Type::UInt64: {
+            auto& col = static_cast<ColumnUInt64&>(*index_column_);
+            col.Erase(col.Size() - 1);
+            break;
+        }
+        default:
+            throw ValidationError("Invalid index column type");
+    }
+}
+
+details::LowCardinalityHashKey ColumnLowCardinality::computeHashKey(const ItemView & item) {
+    static const auto hasher = std::hash<ItemView::DataType>{};
+    if (item.type == Type::Void) {
+        // to distinguish NULL of ColumnNullable and empty string.
+        return {0u, 0u};
+    }
+
+    const auto hash1 = hasher(item.data);
+    const auto hash2 = CityHash64(item.data.data(), item.data.size());
+
+    return details::LowCardinalityHashKey{hash1, hash2};
+}
+
+ColumnRef ColumnLowCardinality::GetDictionary() {
+    return dictionary_column_;
+}
+
+void ColumnLowCardinality::Append(ColumnRef col) {
+    // Append values from col only if it is either
+    // - exactly same type as `this`: LowCardinality wrapping same dictionary type
+    // - same type as dictionary column
+
+    auto c = col->As<ColumnLowCardinality>();
+    // If not LowCardinality of same dictionary type
+    if (!c || !dictionary_column_->Type()->IsEqual(c->dictionary_column_->Type())) {
+        // If not column of the same type as dictionary type
+        if (!dictionary_column_->Type()->IsEqual(col->GetType())) {
+            return;
+        }
+    }
+
+    for (size_t i = 0; i < col->Size(); ++i) {
+        AppendUnsafe(col->GetItem(i));
+    }
+}
+
+namespace {
+
+auto Load(ColumnRef new_dictionary_column, InputStream& input, size_t rows) {
+    // This code tries to follow original implementation of ClickHouse's LowCardinality serialization with
+    // NativeBlockOutputStream::writeData() for DataTypeLowCardinality
+    // (see corresponding serializeBinaryBulkStateSuffix, serializeBinaryBulkStatePrefix, and serializeBinaryBulkWithMultipleStreams),
+    // but with certain simplifications: no shared dictionaries, no on-the-fly dictionary updates.
+    //
+    // As for now those features are not used in client-server protocol and minimal implementation suffices,
+    // however some day they may.
+
+    uint64_t index_serialization_type;
+    if (!WireFormat::ReadFixed(input, &index_serialization_type))
+        throw ProtocolError("Failed to read index serializaton type.");
+
+    auto new_index_column = createIndexColumn(static_cast<IndexType>(index_serialization_type & IndexTypeMask));
+    if (index_serialization_type & IndexFlag::NeedGlobalDictionaryBit)
+        throw UnimplementedError("Global dictionary is not supported.");
+
+    if ((index_serialization_type & IndexFlag::HasAdditionalKeysBit) == 0)
+        throw ValidationError("HasAdditionalKeysBit is missing.");
+
+    uint64_t number_of_keys;
+    if (!WireFormat::ReadFixed(input, &number_of_keys))
+        throw ProtocolError("Failed to read number of rows in dictionary column.");
+
+    auto dataColumn = new_dictionary_column;
+    if (auto nullable = new_dictionary_column->As<ColumnNullable>()) {
+        dataColumn = nullable->Nested();
+    }
+
+    if (!dataColumn->LoadBody(&input, number_of_keys))
+        throw ProtocolError("Failed to read values of dictionary column.");
+
+    uint64_t number_of_rows;
+    if (!WireFormat::ReadFixed(input, &number_of_rows))
+        throw ProtocolError("Failed to read number of rows in index column.");
+
+    if (number_of_rows != rows)
+        throw AssertionError("LowCardinality column must be read in full.");
+
+    new_index_column->LoadBody(&input, number_of_rows);
+
+    if (auto nullable = new_dictionary_column->As<ColumnNullable>()) {
+        nullable->Append(true);
+        for(std::size_t i = 1; i < dataColumn->Size(); i++) {
+            nullable->Append(false);
+        }
+    }
+
+    ColumnLowCardinality::UniqueItems new_unique_items_map;
+    for (size_t i = 0; i < dataColumn->Size(); ++i) {
+        const auto key = ColumnLowCardinality::computeHashKey(new_dictionary_column->GetItem(i));
+        new_unique_items_map.emplace(key, i);
+    }
+
+    // suffix
+    // NOP
+
+    return std::make_tuple(new_dictionary_column, new_index_column, new_unique_items_map);
+}
+
+}
+
+bool ColumnLowCardinality::LoadPrefix(InputStream* input, size_t) {
+    uint64_t key_version;
+
+    if (!WireFormat::ReadFixed(*input, &key_version))
+        throw ProtocolError("Failed to read key serialization version.");
+
+    if (key_version != KeySerializationVersion::SharedDictionariesWithAdditionalKeys)
+        throw ProtocolError("Invalid key serialization version value.");
+
+    return true;
+}
+
+bool ColumnLowCardinality::LoadBody(InputStream* input, size_t rows) {
+    try {
+        auto [new_dictionary, new_index, new_unique_items_map] = ::Load(dictionary_column_->CloneEmpty(), *input, rows);
+
+        dictionary_column_->Swap(*new_dictionary);
+        index_column_.swap(new_index);
+        unique_items_map_.swap(new_unique_items_map);
+        index_type_code_ = index_column_->Type()->GetCode();
+
+        return true;
+    } catch (...) {
+        return false;
+    }
+}
+
+void ColumnLowCardinality::SavePrefix(OutputStream* output) {
+    const auto version = static_cast<uint64_t>(KeySerializationVersion::SharedDictionariesWithAdditionalKeys);
+    WireFormat::WriteFixed(*output, version);
+}
+
+void ColumnLowCardinality::SaveBody(OutputStream* output) {
+    const uint64_t index_serialization_type = indexTypeFromIndexColumn(*index_column_) | IndexFlag::HasAdditionalKeysBit;
+    WireFormat::WriteFixed(*output, index_serialization_type);
+
+    const uint64_t number_of_keys = dictionary_column_->Size();
+    WireFormat::WriteFixed(*output, number_of_keys);
+
+    if (auto columnNullable = dictionary_column_->As<ColumnNullable>()) {
+        columnNullable->Nested()->SaveBody(output);
+    } else {
+        dictionary_column_->SaveBody(output);
+    }
+
+    const uint64_t number_of_rows = index_column_->Size();
+    WireFormat::WriteFixed(*output, number_of_rows);
+
+    index_column_->SaveBody(output);
+}
+
+void ColumnLowCardinality::Clear() {
+    index_column_->Clear();
+    dictionary_column_->Clear();
+    unique_items_map_.clear();
+
+    if (auto columnNullable = dictionary_column_->As<ColumnNullable>()) {
+        AppendNullItem();
+    }
+    AppendDefaultItem();
+}
+
+size_t ColumnLowCardinality::Size() const {
+    return index_column_->Size();
+}
+
+ColumnRef ColumnLowCardinality::Slice(size_t begin, size_t len) const {
+    begin = std::min(begin, Size());
+    len = std::min(len, Size() - begin);
+
+    auto result = std::make_shared<ColumnLowCardinality>(dictionary_column_->CloneEmpty());
+
+    for (size_t i = begin; i < begin + len; ++i)
+        result->AppendUnsafe(this->GetItem(i));
+
+    return result;
+}
+
+ColumnRef ColumnLowCardinality::CloneEmpty() const {
+    return std::make_shared<ColumnLowCardinality>(dictionary_column_->CloneEmpty());
+}
+
+void ColumnLowCardinality::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnLowCardinality &>(other);
+    if (!dictionary_column_->Type()->IsEqual(col.dictionary_column_->Type()))
+        throw ValidationError("Can't swap() LowCardinality columns of different types.");
+
+    // It is important here not to swap pointers to dictionary object,
+    // but swap contents of dictionaries, so the object inside shared_ptr stays the same
+    // (needed for ColumnLowCardinalityT)
+    dictionary_column_->Swap(*col.dictionary_column_);
+
+    index_column_.swap(col.index_column_);
+    unique_items_map_.swap(col.unique_items_map_);
+    std::swap(index_type_code_, col.index_type_code_);
+}
+
+ItemView ColumnLowCardinality::GetItem(size_t index) const {
+    const auto dictionaryIndex = getDictionaryIndex(index);
+
+    if (auto nullable = dictionary_column_->As<ColumnNullable>()) {
+        const auto isNull = dictionaryIndex == 0u;
+
+        if (isNull) {
+            return GetNullItemForDictionary(nullable);
+        }
+    }
+
+    return dictionary_column_->GetItem(dictionaryIndex);
+}
+
+// No checks regarding value type or validity of value is made.
+void ColumnLowCardinality::AppendUnsafe(const ItemView & value) {
+    const auto key = computeHashKey(value);
+    const auto initial_index_size = index_column_->Size();
+    // If the value is unique, then we are going to append it to a dictionary, hence new index is Size().
+    auto [iterator, is_new_item] = unique_items_map_.try_emplace(key, dictionary_column_->Size());
+    try {
+        // Order is important, adding to dictionary last, since it is much (MUCH!!!!) harder
+        // to remove item from dictionary column than from index column
+        // (also, there is currently no API to do that).
+        // Hence in catch-block we assume that dictionary wasn't modified on exception
+        // and there is nothing to rollback.
+
+        appendIndex(iterator->second);
+        if (is_new_item) {
+            AppendToDictionary(*dictionary_column_, value);
+        }
+    }
+    catch (...) {
+        if (index_column_->Size() != initial_index_size)
+            removeLastIndex();
+        if (is_new_item)
+            unique_items_map_.erase(iterator);
+
+        throw;
+    }
+}
+
+void ColumnLowCardinality::AppendNullItem()
+{
+    const auto null_item = GetNullItemForDictionary(dictionary_column_);
+    AppendToDictionary(*dictionary_column_, null_item);
+    unique_items_map_.emplace(computeHashKey(null_item), 0);
+}
+
+void ColumnLowCardinality::AppendDefaultItem()
+{
+    const auto defaultItem = GetDefaultItemForDictionary(dictionary_column_);
+    unique_items_map_.emplace(computeHashKey(defaultItem), dictionary_column_->Size());
+    AppendToDictionary(*dictionary_column_, defaultItem);
+}
+
+size_t ColumnLowCardinality::GetDictionarySize() const {
+    return dictionary_column_->Size();
+}
+
+TypeRef ColumnLowCardinality::GetNestedType() const {
+    return dictionary_column_->Type();
+}
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/lowcardinality.h b/lib/clickhouse-cpp/clickhouse/columns/lowcardinality.h
new file mode 100644
index 0000000..33c339e
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/lowcardinality.h
@@ -0,0 +1,221 @@
+#pragma once
+
+#include "column.h"
+#include "numeric.h"
+#include "nullable.h"
+
+#include <functional>
+#include <string>
+#include <unordered_map>
+#include <utility>
+
+namespace clickhouse {
+
+template <typename NestedColumnType>
+class ColumnLowCardinalityT;
+
+namespace details {
+
+/** LowCardinalityHashKey used as key in unique items hashmap to abstract away key value
+ * (type of which depends on dictionary column) and to reduce likelehood of collisions.
+ *
+ * In order to dramatically reduce collision rate, we use 2 different hashes from 2 different hash functions.
+ * First hash is used in hashtable (to calculate item position).
+ * Second one is used as part of key value and accessed via `operator==()` upon collision resolution/detection.
+ */
+using LowCardinalityHashKey = std::pair<std::uint64_t, std::uint64_t>;
+
+struct LowCardinalityHashKeyHash {
+    inline std::size_t operator()(const LowCardinalityHashKey &hash_key) const noexcept {
+        return hash_key.first;
+    }
+};
+
+}
+
+/*
+ * LC column contains an "invisible" default item at the beginning of the collection. [default, ...]
+ * If the nested type is Nullable, it contains a null-item at the beginning and a default item at the second position. [null, default, ...]
+ * Null map is not serialized in LC columns. Instead, nulls are tracked by having an index of 0.
+ * */
+class ColumnLowCardinality : public Column {
+public:
+    using UniqueItems = std::unordered_map<details::LowCardinalityHashKey, size_t /*dictionary index*/, details::LowCardinalityHashKeyHash>;
+
+    template <typename T>
+    friend class ColumnLowCardinalityT;
+
+private:
+    // IMPLEMENTATION NOTE: ColumnLowCardinalityT takes reference to underlying dictionary column object,
+    // so make sure to NOT change address of the dictionary object (with reset(), swap()) or with anything else.
+    ColumnRef dictionary_column_;
+    ColumnRef index_column_;
+    UniqueItems unique_items_map_;
+
+public:
+    ColumnLowCardinality(ColumnLowCardinality&& col) = default;
+    // c-tor makes a deep copy of the dictionary_column.
+    explicit ColumnLowCardinality(ColumnRef dictionary_column);
+    explicit ColumnLowCardinality(std::shared_ptr<ColumnNullable> dictionary_column);
+
+    template <typename T>
+    explicit ColumnLowCardinality(std::shared_ptr<ColumnNullableT<T>> dictionary_column)
+        : ColumnLowCardinality(dictionary_column->template As<ColumnNullable>())
+    {}
+
+    ~ColumnLowCardinality();
+
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Appends another LowCardinality column to the end of this one, updating dictionary.
+    void Append(ColumnRef /*column*/) override;
+
+    bool LoadPrefix(InputStream* input, size_t rows) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column prefix to output stream.
+    void SavePrefix(OutputStream* output) override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Clear column data.
+    void Clear() override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of current column, with compacted dictionary
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+    ItemView GetItem(size_t index) const override;
+
+    size_t GetDictionarySize() const;
+    TypeRef GetNestedType() const;
+
+protected:
+    std::uint64_t getDictionaryIndex(std::uint64_t item_index) const;
+    void appendIndex(std::uint64_t item_index);
+    void removeLastIndex();
+    ColumnRef GetDictionary();
+
+    void AppendUnsafe(const ItemView &);
+
+private:
+    void Setup(ColumnRef dictionary_column);
+    void AppendNullItem();
+    void AppendDefaultItem();
+
+    Type::Code index_type_code_;
+
+public:
+    static details::LowCardinalityHashKey computeHashKey(const ItemView &);
+};
+
+/** Type-aware wrapper that provides simple convenience interface for accessing/appending individual items.
+ */
+template <typename DictionaryColumnType>
+class ColumnLowCardinalityT : public ColumnLowCardinality {
+
+    DictionaryColumnType& typed_dictionary_;
+    const Type::Code type_;
+
+public:
+    using WrappedColumnType = DictionaryColumnType;
+    // Type this column takes as argument of Append and returns with At() and operator[]
+    using ValueType = typename DictionaryColumnType::ValueType;
+
+    explicit ColumnLowCardinalityT(ColumnLowCardinality&& col)
+        : ColumnLowCardinality(std::move(col))
+        ,  typed_dictionary_(dynamic_cast<DictionaryColumnType &>(*GetDictionary()))
+        ,  type_(GetTypeCode(typed_dictionary_))
+    {
+    }
+
+    template <typename ...Args>
+    explicit ColumnLowCardinalityT(Args &&... args)
+        : ColumnLowCardinalityT(std::make_shared<DictionaryColumnType>(std::forward<Args>(args)...))
+    {}
+
+    // Create LC<T> column from existing T-column, making a deep copy of all contents.
+    explicit ColumnLowCardinalityT(std::shared_ptr<DictionaryColumnType> dictionary_col)
+        : ColumnLowCardinality(dictionary_col)
+        , typed_dictionary_(dynamic_cast<DictionaryColumnType &>(*GetDictionary()))
+        , type_(GetTypeCode(typed_dictionary_))
+    {}
+
+    /// Extended interface to simplify reading/adding individual items.
+
+    /// Returns element at given row number.
+    inline ValueType At(size_t n) const {
+        return typed_dictionary_.At(getDictionaryIndex(n));
+    }
+
+    /// Returns element at given row number.
+    inline ValueType operator [] (size_t n) const {
+        return typed_dictionary_[getDictionaryIndex(n)];
+    }
+
+    // so the non-virtual Append below doesn't shadow Append() from base class when compiled with older compilers.
+    using ColumnLowCardinality::Append;
+
+    inline void Append(const ValueType & value) {
+        if constexpr (IsNullable<WrappedColumnType>) {
+            if (value.has_value()) {
+                AppendUnsafe(ItemView{type_, *value});
+            } else {
+                AppendUnsafe(ItemView{});
+            }
+        } else {
+            AppendUnsafe(ItemView{type_, value});
+        }
+    }
+
+    template <typename T>
+    inline void AppendMany(const T& container) {
+        for (const auto & item : container) {
+            Append(item);
+        }
+    }
+
+    /** Create a ColumnLowCardinalityT from a ColumnLowCardinality, without copying data and offsets, but by
+     * 'stealing' those from `col`.
+     *
+     *  Ownership of column internals is transferred to returned object, original (argument) object
+     *  MUST NOT BE USED IN ANY WAY, it is only safe to dispose it.
+     *
+     *  Throws an exception if `col` is of wrong type, it is safe to use original col in this case.
+     *  This is a static method to make such conversion verbose.
+     */
+    static auto Wrap(ColumnLowCardinality&& col) {
+        return std::make_shared<ColumnLowCardinalityT<WrappedColumnType>>(std::move(col));
+    }
+
+    static auto Wrap(Column&& col) { return Wrap(std::move(dynamic_cast<ColumnLowCardinality&&>(col))); }
+
+    // Helper to simplify integration with other APIs
+    static auto Wrap(ColumnRef&& col) { return Wrap(std::move(*col->AsStrict<ColumnLowCardinality>())); }
+
+    ColumnRef Slice(size_t begin, size_t size) const override {
+        return Wrap(ColumnLowCardinality::Slice(begin, size));
+    }
+
+    ColumnRef CloneEmpty() const override { return Wrap(ColumnLowCardinality::CloneEmpty()); }
+
+private:
+
+    template <typename T>
+    static auto GetTypeCode(T& column) {
+        if constexpr (IsNullable<T>) {
+            return GetTypeCode(*column.Nested()->template AsStrict<typename T::NestedColumnType>());
+        } else {
+            return column.Type()->GetCode();
+        }
+    }
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/lowcardinalityadaptor.h b/lib/clickhouse-cpp/clickhouse/columns/lowcardinalityadaptor.h
new file mode 100644
index 0000000..bcde1a9
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/lowcardinalityadaptor.h
@@ -0,0 +1,63 @@
+#pragma once
+
+#include "column.h"
+#include "lowcardinality.h"
+
+#include <cassert>
+
+namespace clickhouse {
+
+class OutputStream;
+class CodedInputStream;
+
+/** Adapts any ColumnType to be serialized\deserialized as LowCardinality,
+ *  and to be castable to ColumnType via ColumnPtr->As<ColumnType>().
+ *
+ * It helps to ease migration of the old codebases, which can't afford to switch
+ * to using ColumnLowCardinalityT or ColumnLowCardinality directly,
+ * but still want to benefit from smaller on-wire LowCardinality bandwidth footprint.
+ *
+ * Not intended to be used by users directly.
+ *
+ * @see ClientOptions, CreateColumnByType
+ */
+template <typename AdaptedColumnType>
+class
+[[deprecated("Makes implementation of LC(X) harder and code uglier. Will be removed in next major release (3.0) ")]]
+LowCardinalitySerializationAdaptor : public AdaptedColumnType
+{
+public:
+    using AdaptedColumnType::AdaptedColumnType;
+
+    bool LoadPrefix(InputStream* input, size_t rows) override {
+        auto new_data_column = this->Slice(0, 0)->template As<AdaptedColumnType>();
+        ColumnLowCardinalityT<AdaptedColumnType> low_cardinality_col(new_data_column);
+
+        return low_cardinality_col.LoadPrefix(input, rows);
+    }
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override {
+        auto new_data_column = this->CloneEmpty()->template As<AdaptedColumnType>();
+
+        ColumnLowCardinalityT<AdaptedColumnType> low_cardinality_col(new_data_column);
+        if (!low_cardinality_col.LoadBody(input, rows))
+            return false;
+
+        // It safe to reuse `flat_data_column` later since ColumnLowCardinalityT makes a deep copy, but still check just in case.
+        assert(new_data_column->Size() == 0);
+
+        for (size_t i = 0; i < low_cardinality_col.Size(); ++i)
+            new_data_column->Append(low_cardinality_col[i]);
+
+        this->Swap(*new_data_column);
+        return true;
+    }
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override {
+        ColumnLowCardinalityT<AdaptedColumnType>(this->template As<AdaptedColumnType>()).SaveBody(output);
+    }
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/map.cpp b/lib/clickhouse-cpp/clickhouse/columns/map.cpp
new file mode 100644
index 0000000..839b066
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/map.cpp
@@ -0,0 +1,87 @@
+#include "map.h"
+
+#include <stdexcept>
+
+#include "../exceptions.h"
+#include "utils.h"
+
+namespace {
+
+using namespace clickhouse;
+
+TypeRef GetMapType(const Type& data_type) {
+    auto array = data_type.As<ArrayType>();
+    if (!array) {
+        throw ValidationError("Wrong type  " + data_type.GetName() + " of data for map");
+    }
+    auto tuple = array->GetItemType()->As<TupleType>();
+    if (!tuple) {
+        throw ValidationError("Wrong type  " + data_type.GetName() + " of data for map");
+    }
+    auto types = tuple->GetTupleType();
+    if (types.size() != 2) {
+        throw ValidationError("Wrong type  " + data_type.GetName() + " of data for map");
+    }
+    return Type::CreateMap(types[0], types[1]);
+}
+
+}  // namespace
+
+namespace clickhouse {
+
+ColumnMap::ColumnMap(ColumnRef data)
+    : Column(GetMapType(data->GetType())), data_(data->As<ColumnArray>()) {
+}
+
+void ColumnMap::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+void ColumnMap::Clear() {
+    data_->Clear();
+}
+
+void ColumnMap::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnMap>()) {
+        data_->Append(col->data_);
+    }
+}
+
+bool ColumnMap::LoadPrefix(InputStream* input, size_t rows) {
+    return data_->LoadPrefix(input, rows);
+}
+
+bool ColumnMap::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnMap::SavePrefix(OutputStream* output) {
+    data_->SavePrefix(output);
+}
+
+void ColumnMap::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+size_t ColumnMap::Size() const {
+    return data_->Size();
+}
+
+ColumnRef ColumnMap::Slice(size_t begin, size_t len) const {
+    return std::make_shared<ColumnMap>(data_->Slice(begin, len));
+}
+
+ColumnRef ColumnMap::CloneEmpty() const {
+    return std::make_shared<ColumnMap>(data_->CloneEmpty());
+}
+
+void ColumnMap::Swap(Column& other) {
+    auto& col = dynamic_cast<ColumnMap&>(other);
+    data_.swap(col.data_);
+}
+
+ColumnRef ColumnMap::GetAsColumn(size_t n) const {
+    return data_->GetAsColumn(n);
+}
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/columns/map.h b/lib/clickhouse-cpp/clickhouse/columns/map.h
new file mode 100644
index 0000000..4d64480
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/map.h
@@ -0,0 +1,257 @@
+#pragma once
+
+#include "../base/projected_iterator.h"
+#include "array.h"
+#include "column.h"
+#include "tuple.h"
+
+#include <functional>
+#include <map>
+
+namespace clickhouse {
+
+template <typename KeyColumnType, typename ValueColumnType>
+class ColumnMapT;
+
+/**
+ * Represents column of Map(K, V).
+ */
+class ColumnMap : public Column {
+public:
+    /** Create a map of given type, with actual values and offsets.
+     *
+     *  Both `data` and `offsets` are used (and modified) internally bye ColumnArray.
+     *  Users are strongly advised against modifying contents of `data` or `offsets` afterwards.
+     */
+    explicit ColumnMap(ColumnRef data);
+
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column prefix from input stream.
+    bool LoadPrefix(InputStream* input, size_t rows) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column prefix to output stream.
+    void SavePrefix(OutputStream* output) override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t, size_t) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column&) override;
+
+    /// Converts map at pos n to column.
+    /// Type of row is tuple {key, value}.
+    ColumnRef GetAsColumn(size_t n) const;
+
+protected:
+    template <typename K, typename V>
+    friend class ColumnMapT;
+
+    ColumnMap(ColumnMap&& map);
+
+private:
+    std::shared_ptr<ColumnArray> data_;
+};
+
+template <typename K, typename V>
+class ColumnMapT : public ColumnMap {
+public:
+    using KeyColumnType   = K;
+    using ValueColumnType = V;
+    using Key             = std::decay_t<decltype(std::declval<KeyColumnType>().At(0))>;
+    using Value           = std::decay_t<decltype(std::declval<ValueColumnType>().At(0))>;
+    using TupleColumnType = ColumnTupleT<KeyColumnType, ValueColumnType>;
+    using ArrayColumnType = ColumnArrayT<TupleColumnType>;
+
+    ColumnMapT(ColumnRef data)
+        : ColumnMap(data), typed_data_(data->AsStrict<ColumnArrayT<TupleColumnType>>()) {}
+
+    ColumnMapT(std::shared_ptr<KeyColumnType> keys, std::shared_ptr<ValueColumnType> values)
+        : ColumnMap(std::make_shared<ArrayColumnType>(std::make_shared<TupleColumnType>(
+              std::make_tuple(std::move(keys), std::move(values))))),
+          typed_data_(data_->template As<ArrayColumnType>()) {}
+
+    ColumnRef Slice(size_t begin, size_t len) const override {
+        return std::make_shared<ColumnMapT<K, V>>(typed_data_->Slice(begin, len));
+    }
+
+    ColumnRef CloneEmpty() const override {
+        return std::make_shared<ColumnMapT<K, V>>(typed_data_->CloneEmpty());
+    }
+
+    void Swap(Column& other) override {
+        auto& col = dynamic_cast<ColumnMapT<K, V>&>(other);
+        col.typed_data_.swap(typed_data_);
+        ColumnMap::Swap(other);
+    }
+
+    /// A single (row) value of the Map-column i.e. read-only map.
+    /// It has a linear time complexity to access items
+    /// Because data base type has same structure
+    /// "This lookup works now with a linear complexity."
+    /// https://clickhouse.com/docs/en/sql-reference/data-types/map
+    /// Convert it to a suitable container required to access more than one element
+
+    class MapValueView {
+        const typename ArrayColumnType::ArrayValueView data_;
+
+    public:
+        using ValueType = std::pair<Key, Value>;
+
+        MapValueView(typename ArrayColumnType::ArrayValueView data) : data_(std::move(data)) {}
+
+        inline auto operator[](const Key& key) const { return (*Find(key)).second; }
+
+        inline auto At(const Key& key) const {
+            auto it = Find(key);
+            if (it == end()) throw ValidationError("ColumnMap value key not found");
+            return (*it).second;
+        }
+
+        class Iterator {
+            typename ArrayColumnType::ArrayValueView::Iterator data_iterator_;
+
+        public:
+            Iterator() = default;
+
+            Iterator(typename ArrayColumnType::ArrayValueView::Iterator data_iterator)
+                : data_iterator_(data_iterator) {}
+
+            using ValueType = std::pair<Key, Value>;
+            using difference_type = size_t;
+            using value_type = ValueType;
+            using pointer = void;
+            using reference = ValueType&;
+            using iterator_category = std::forward_iterator_tag;
+
+            inline auto operator*() const {
+                auto tuple = *data_iterator_;
+                return ValueType{std::get<0>(tuple), std::get<1>(tuple)};
+            }
+
+            inline Iterator& operator++() {
+                ++data_iterator_;
+                return *this;
+            }
+
+            inline bool operator==(const Iterator& other) const {
+                return this->data_iterator_ == other.data_iterator_;
+            }
+
+            inline bool operator!=(const Iterator& other) const { return !(*this == other); }
+        };
+
+        // minimalistic stl-like container interface, hence the lowercase
+        inline Iterator begin() const { return Iterator{data_.begin()}; }
+
+        inline Iterator cbegin() const { return Iterator{data_.cbegin()}; }
+
+        inline Iterator end() const { return Iterator{data_.end()}; }
+
+        inline Iterator cend() const { return Iterator{data_.cend()}; }
+
+        inline size_t size() const { return data_.size(); }
+
+        // It is ugly to have both size() and Size(), but it is for compatitability with both STL
+        // and rest of the clickhouse-cpp.
+        inline size_t Size() const { return data_.Size(); }
+
+        inline size_t Count(const Key& key) const {
+            size_t result = 0;
+            for (auto item : data_) {
+                if (std::get<0>(item) == key) {
+                    ++result;
+                }
+            }
+            return result;
+        }
+
+        inline Iterator Find(const Key& key) const {
+            for (auto it = data_.begin(); it != data_.end(); ++it) {
+                if (std::get<0>(*it) == key) {
+                    return Iterator{it};
+                }
+            }
+            return end();
+        }
+
+        inline bool operator==(const MapValueView& other) const {
+            if (size() != other.size()) {
+                return false;
+            }
+            const auto make_index = [](const auto& data) {
+                std::vector<size_t> result{data.Size()};
+                std::generate(result.begin(), result.end(), [i = 0] () mutable {return i++;});
+                std::sort(result.begin(), result.end(), [&data](size_t l, size_t r) {return data[l] < data[r];});
+                return result;
+            };
+            const auto index = make_index(data_);
+            for (const auto& val : other.data_) {
+                if (!std::binary_search(index.begin(), index.end(), val,
+                        [&data = data_](const auto& l, size_t r) {return l < data[r];})) {
+                    return false;
+                }
+            }
+            return true;
+        }
+
+        inline bool operator!=(const MapValueView& other) const { return !(*this == other); }
+    };
+
+    inline auto At(size_t index) const { return MapValueView{typed_data_->At(index)}; }
+
+    inline auto operator[](size_t index) const { return At(index); }
+
+    using ColumnMap::Append;
+
+    inline void Append(const MapValueView& value) { typed_data_->Append(value.data_); }
+
+    inline void Append(const std::vector<std::tuple<Key, Value>>& tuples) {
+        typed_data_->Append(tuples.begin(), tuples.end());
+    }
+
+    template <typename T>
+    inline void Append(const T& value) {
+        using BaseIter = decltype(value.begin());
+        using KeyOfT = decltype(std::declval<BaseIter>()->first);
+        using ValOfT = decltype(std::declval<BaseIter>()->second);
+        using Functor = std::function<std::tuple<KeyOfT, ValOfT>(const BaseIter&)>;
+        using Iterator = ProjectedIterator<Functor, BaseIter>;
+
+        Functor functor = [](const BaseIter& i) {
+            return std::make_tuple(std::cref(i->first), std::cref(i->second));
+        };
+
+        typed_data_->Append(Iterator{value.begin(), functor}, Iterator{value.end(), functor});
+    }
+
+    static auto Wrap(ColumnMap&& col) {
+        auto data = ArrayColumnType::Wrap(std::move(col.data_));
+        return std::make_shared<ColumnMapT<K, V>>(std::move(data));
+    }
+
+    static auto Wrap(Column&& col) { return Wrap(std::move(dynamic_cast<ColumnMap&&>(col))); }
+
+    // Helper to simplify integration with other APIs
+    static auto Wrap(ColumnRef&& col) { return Wrap(std::move(*col->AsStrict<ColumnMap>())); }
+
+private:
+    std::shared_ptr<ArrayColumnType> typed_data_;
+};
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/columns/nothing.h b/lib/clickhouse-cpp/clickhouse/columns/nothing.h
new file mode 100644
index 0000000..8e1a4e3
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/nothing.h
@@ -0,0 +1,87 @@
+
+#pragma once
+
+#include "column.h"
+#include "../base/input.h"
+
+#include <stdexcept>
+#include <utility>
+
+namespace clickhouse {
+
+/**
+ * Represents dummy column of NULLs.
+ */
+class ColumnNothing : public Column {
+public:
+    ColumnNothing()
+        : Column(Type::CreateNothing())
+        , size_(0)
+    {
+    }
+
+    explicit ColumnNothing(size_t n)
+        : Column(Type::CreateNothing())
+        , size_(n)
+    {
+    }
+
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t) override {};
+
+    /// Appends one element to the column.
+    void Append(std::unique_ptr<void*>) { ++size_; }
+
+    /// Returns element at given row number.
+    std::nullptr_t At(size_t) const { return nullptr; };
+
+    /// Returns element at given row number.
+    inline std::nullptr_t operator [] (size_t) const { return nullptr; };
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t, size_t len) const override {
+        return std::make_shared<ColumnNothing>(len);
+    }
+
+    ColumnRef CloneEmpty() const override {
+        return std::make_shared<ColumnNothing>();
+    }
+
+    ItemView GetItem(size_t /*index*/) const override { return ItemView{}; }
+
+public:
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override {
+        if (auto col = column->As<ColumnNothing>()) {
+            size_ += col->Size();
+        }
+    }
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override {
+        input->Skip(rows);
+        size_ += rows;
+        return true;
+    }
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream*) override {
+        throw UnimplementedError("method SaveBody is not supported for Nothing column");
+    }
+
+    /// Clear column data .
+    void Clear() override { size_ = 0; }
+
+    /// Returns count of rows in the column.
+    size_t Size() const override { return size_; }
+
+    void Swap(Column& other) override {
+        auto & col = dynamic_cast<ColumnNothing &>(other);
+        std::swap(size_, col.size_);
+    }
+
+private:
+    size_t size_;
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/nullable.cpp b/lib/clickhouse-cpp/clickhouse/columns/nullable.cpp
index f6e1792..23940c1 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/nullable.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/nullable.cpp
@@ -1,7 +1,7 @@
 #include "nullable.h"
 
-#include <stdexcept>
 #include <assert.h>
+#include <stdexcept>
 
 namespace clickhouse {
 
@@ -11,10 +11,16 @@ ColumnNullable::ColumnNullable(ColumnRef nested, ColumnRef nulls)
     , nulls_(nulls->As<ColumnUInt8>())
 {
     if (nested_->Size() != nulls->Size()) {
-        throw std::runtime_error("count of elements in nested and nulls should be the same");
+        throw ValidationError("count of elements in nested and nulls should be the same");
     }
 }
 
+void ColumnNullable::Append(bool isnull)
+{
+    nulls_->Append(isnull ? 1 : 0);
+}
+
+
 bool ColumnNullable::IsNull(size_t n) const {
     return nulls_->At(n) != 0;
 }
@@ -23,6 +29,16 @@ ColumnRef ColumnNullable::Nested() const {
     return nested_;
 }
 
+ColumnRef ColumnNullable::Nulls() const
+{
+       return nulls_;
+}
+
+void ColumnNullable::Reserve(size_t new_cap) {
+    nested_->Reserve(new_cap);
+    nulls_->Reserve(new_cap);
+}
+
 void ColumnNullable::Append(ColumnRef column) {
     if (auto col = column->As<ColumnNullable>()) {
         if (!col->nested_->Type()->IsEqual(nested_->Type())) {
@@ -39,28 +55,55 @@ void ColumnNullable::Clear() {
     nulls_->Clear();
 }
 
-bool ColumnNullable::Load(CodedInputStream* input, size_t rows) {
-    if (!nulls_->Load(input, rows)) {
+bool ColumnNullable::LoadPrefix(InputStream* input, size_t rows) {
+    return nested_->LoadPrefix(input, rows);
+}
+
+bool ColumnNullable::LoadBody(InputStream* input, size_t rows) {
+    if (!nulls_->LoadBody(input, rows)) {
         return false;
     }
-    if (!nested_->Load(input, rows)) {
+    if (!nested_->LoadBody(input, rows)) {
         return false;
     }
     return true;
 }
 
-void ColumnNullable::Save(CodedOutputStream* output) {
-    nulls_->Save(output);
-    nested_->Save(output);
+void ColumnNullable::SavePrefix(OutputStream* output) {
+    nested_->SavePrefix(output);
+}
+
+void ColumnNullable::SaveBody(OutputStream* output) {
+    nulls_->SaveBody(output);
+    nested_->SaveBody(output);
 }
 
 size_t ColumnNullable::Size() const {
-    assert(nested_->Size() == nulls_->Size());
     return nulls_->Size();
 }
 
-ColumnRef ColumnNullable::Slice(size_t begin, size_t len) {
+ColumnRef ColumnNullable::Slice(size_t begin, size_t len) const {
     return std::make_shared<ColumnNullable>(nested_->Slice(begin, len), nulls_->Slice(begin, len));
 }
 
+ColumnRef ColumnNullable::CloneEmpty() const {
+    return std::make_shared<ColumnNullable>(nested_->CloneEmpty(), nulls_->CloneEmpty());
+}
+
+void ColumnNullable::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnNullable &>(other);
+    if (!nested_->Type()->IsEqual(col.nested_->Type()))
+        throw ValidationError("Can't swap() Nullable columns of different types.");
+
+    nested_.swap(col.nested_);
+    nulls_.swap(col.nulls_);
+}
+
+ItemView ColumnNullable::GetItem(size_t index) const  {
+    if (IsNull(index))
+        return ItemView();
+
+    return nested_->GetItem(index);
+}
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/nullable.h b/lib/clickhouse-cpp/clickhouse/columns/nullable.h
index 3ecc541..1946e8b 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/nullable.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/nullable.h
@@ -3,6 +3,8 @@
 #include "column.h"
 #include "numeric.h"
 
+#include <optional>
+
 namespace clickhouse {
 
 /**
@@ -12,34 +14,140 @@ class ColumnNullable : public Column {
 public:
     ColumnNullable(ColumnRef nested, ColumnRef nulls);
 
+    /// Appends one null flag to the end of the column
+    void Append(bool isnull);
+
     /// Returns null flag at given row number.
     bool IsNull(size_t n) const;
 
     /// Returns nested column.
     ColumnRef Nested() const;
 
+    /// Returns nulls column.
+    ColumnRef Nulls() const;
+
 public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
+    /// Loads column prefix from input stream.
+    bool LoadPrefix(InputStream* input, size_t rows) override;
+
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column prefix to output stream.
+    void SavePrefix(OutputStream* output) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
+    void SaveBody(OutputStream* output) override;
 
     /// Clear column data .
     void Clear() override;
-    
+
     /// Returns count of rows in the column.
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column&) override;
+
+    ItemView GetItem(size_t) const override;
 
 private:
     ColumnRef nested_;
     std::shared_ptr<ColumnUInt8> nulls_;
 };
 
+template <typename ColumnType>
+class ColumnNullableT : public ColumnNullable {
+public:
+    using NestedColumnType = ColumnType;
+    using ValueType = std::optional<std::decay_t<decltype(std::declval<NestedColumnType>().At(0))>>;
+
+    ColumnNullableT(std::shared_ptr<NestedColumnType> data, std::shared_ptr<ColumnUInt8> nulls)
+        : ColumnNullable(data, nulls)
+        , typed_nested_data_(data)
+    {}
+
+    explicit ColumnNullableT(std::shared_ptr<NestedColumnType> data)
+        : ColumnNullableT(data, FillNulls(data->Size()))
+    {}
+
+    template <typename ...Args>
+    explicit ColumnNullableT(Args &&... args)
+        : ColumnNullableT(std::make_shared<NestedColumnType>(std::forward<Args>(args)...))
+    {}
+
+    inline ValueType At(size_t index) const {
+        return IsNull(index) ? ValueType{} : ValueType{typed_nested_data_->At(index)};
+    }
+
+    inline ValueType operator[](size_t index) const { return At(index); }
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override {
+        ColumnNullable::Append(std::move(column));
+    }
+
+    inline void Append(ValueType value) {
+        ColumnNullable::Append(!value.has_value());
+        if (value.has_value()) {
+            typed_nested_data_->Append(std::move(*value));
+        } else {
+            typed_nested_data_->Append(typename ValueType::value_type{});
+        }
+    }
+
+    /** Create a ColumnNullableT from a ColumnNullable, without copying data and offsets, but by
+     * 'stealing' those from `col`.
+     *
+     *  Ownership of column internals is transferred to returned object, original (argument) object
+     *  MUST NOT BE USED IN ANY WAY, it is only safe to dispose it.
+     *
+     *  Throws an exception if `col` is of wrong type, it is safe to use original col in this case.
+     *  This is a static method to make such conversion verbose.
+     */
+    static auto Wrap(ColumnNullable&& col) {
+        return std::make_shared<ColumnNullableT<NestedColumnType>>(
+            col.Nested()->AsStrict<NestedColumnType>(),
+            col.Nulls()->AsStrict<ColumnUInt8>()) ;
+    }
+
+    static auto Wrap(Column&& col) { return Wrap(std::move(dynamic_cast<ColumnNullable&&>(col))); }
+
+    // Helper to simplify integration with other APIs
+    static auto Wrap(ColumnRef&& col) { return Wrap(std::move(*col->AsStrict<ColumnNullable>())); }
+
+    ColumnRef Slice(size_t begin, size_t size) const override {
+        return Wrap(ColumnNullable::Slice(begin, size));
+    }
+
+    ColumnRef CloneEmpty() const override { return Wrap(ColumnNullable::CloneEmpty()); }
+
+    void Swap(Column& other) override {
+        auto& col = dynamic_cast<ColumnNullableT<NestedColumnType>&>(other);
+        typed_nested_data_.swap(col.typed_nested_data_);
+        ColumnNullable::Swap(other);
+    }
+
+private:
+    static inline auto FillNulls(size_t n){
+        auto result = std::make_shared<ColumnUInt8>();
+        for (size_t i = 0; i < n; ++i) {
+            result->Append(0);
+        }
+        return result;
+    }
+
+    std::shared_ptr<NestedColumnType> typed_nested_data_;
+};
+
+template <typename T>
+constexpr bool IsNullable = std::is_base_of_v<ColumnNullable, T>;
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/numeric.cpp b/lib/clickhouse-cpp/clickhouse/columns/numeric.cpp
index bbd516d..4819f37 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/numeric.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/numeric.cpp
@@ -1,6 +1,8 @@
 #include "numeric.h"
 #include "utils.h"
 
+#include "../base/wire_format.h"
+
 namespace clickhouse {
 
 template <typename T>
@@ -10,17 +12,47 @@ ColumnVector<T>::ColumnVector()
 }
 
 template <typename T>
-ColumnVector<T>::ColumnVector(const std::vector<T>& data)
+ColumnVector<T>::ColumnVector(const std::vector<T> & data)
     : Column(Type::CreateSimple<T>())
     , data_(data)
 {
 }
 
+template <typename T>
+ColumnVector<T>::ColumnVector(std::vector<T> && data)
+    : Column(Type::CreateSimple<T>())
+    , data_(std::move(data))
+{
+}
+
 template <typename T>
 void ColumnVector<T>::Append(const T& value) {
     data_.push_back(value);
 }
 
+template <typename T>
+void ColumnVector<T>::Erase(size_t pos, size_t count) {
+    const auto begin = std::min(pos, data_.size());
+    const auto last = begin + std::min(data_.size() - begin, count);
+
+    data_.erase(data_.begin() + begin, data_.begin() + last);
+}
+
+template <typename T>
+std::vector<T>& ColumnVector<T>::GetWritableData() {
+    return data_;
+}
+
+template <typename T>
+void ColumnVector<T>::Reserve(size_t new_cap) {
+    data_.reserve(new_cap);
+}
+
+template <typename T>
+size_t ColumnVector<T>::Capacity() const {
+    return data_.capacity();
+}
+
 template <typename T>
 void ColumnVector<T>::Clear() {
     data_.clear();
@@ -31,11 +63,6 @@ const T& ColumnVector<T>::At(size_t n) const {
     return data_.at(n);
 }
 
-template <typename T>
-const T& ColumnVector<T>::operator [] (size_t n) const {
-    return data_[n];
-}
-
 template <typename T>
 void ColumnVector<T>::Append(ColumnRef column) {
     if (auto col = column->As<ColumnVector<T>>()) {
@@ -44,15 +71,15 @@ void ColumnVector<T>::Append(ColumnRef column) {
 }
 
 template <typename T>
-bool ColumnVector<T>::Load(CodedInputStream* input, size_t rows) {
+bool ColumnVector<T>::LoadBody(InputStream* input, size_t rows) {
     data_.resize(rows);
 
-    return input->ReadRaw(data_.data(), data_.size() * sizeof(T));
+    return WireFormat::ReadBytes(*input, data_.data(), data_.size() * sizeof(T));
 }
 
 template <typename T>
-void ColumnVector<T>::Save(CodedOutputStream* output) {
-    output->WriteRaw(data_.data(), data_.size() * sizeof(T));
+void ColumnVector<T>::SaveBody(OutputStream* output) {
+    WireFormat::WriteBytes(*output, data_.data(), data_.size() * sizeof(T));
 }
 
 template <typename T>
@@ -61,10 +88,26 @@ size_t ColumnVector<T>::Size() const {
 }
 
 template <typename T>
-ColumnRef ColumnVector<T>::Slice(size_t begin, size_t len) {
+ColumnRef ColumnVector<T>::Slice(size_t begin, size_t len) const {
     return std::make_shared<ColumnVector<T>>(SliceVector(data_, begin, len));
 }
 
+template <typename T>
+ColumnRef ColumnVector<T>::CloneEmpty() const {
+    return std::make_shared<ColumnVector<T>>();
+}
+
+template <typename T>
+void ColumnVector<T>::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnVector<T> &>(other);
+    data_.swap(col.data_);
+}
+
+template <typename T>
+ItemView ColumnVector<T>::GetItem(size_t index) const  {
+    return ItemView{type_->GetCode(), data_[index]};
+}
+
 template class ColumnVector<int8_t>;
 template class ColumnVector<int16_t>;
 template class ColumnVector<int32_t>;
@@ -74,9 +117,10 @@ template class ColumnVector<uint8_t>;
 template class ColumnVector<uint16_t>;
 template class ColumnVector<uint32_t>;
 template class ColumnVector<uint64_t>;
+template class ColumnVector<Int128>;
+template class ColumnVector<UInt128>;
 
 template class ColumnVector<float>;
 template class ColumnVector<double>;
 
 }
-
diff --git a/lib/clickhouse-cpp/clickhouse/columns/numeric.h b/lib/clickhouse-cpp/clickhouse/columns/numeric.h
index 45e3e30..a277d6a 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/numeric.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/numeric.h
@@ -1,6 +1,9 @@
 #pragma once
 
 #include "column.h"
+#include "absl/numeric/int128.h"
+
+#include <cstdint>
 
 namespace clickhouse {
 
@@ -10,9 +13,16 @@ namespace clickhouse {
 template <typename T>
 class ColumnVector : public Column {
 public:
+    using DataType = T;
+    using ValueType = T;
+
     ColumnVector();
 
     explicit ColumnVector(const std::vector<T>& data);
+    explicit ColumnVector(std::vector<T> && data);
+
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
 
     /// Appends one element to the end of column.
     void Append(const T& value);
@@ -21,17 +31,25 @@ class ColumnVector : public Column {
     const T& At(size_t n) const;
 
     /// Returns element at given row number.
-    const T& operator [] (size_t n) const;
+    inline const T& operator [] (size_t n) const { return At(n); }
+
+    void Erase(size_t pos, size_t count = 1);
+
+    /// Get Raw Vector Contents
+    std::vector<T>& GetWritableData();
+
+    /// Returns the capacity of the column
+    size_t Capacity() const;
 
 public:
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
+    void SaveBody(OutputStream* output) override;
 
     /// Clear column data .
     void Clear() override;
@@ -40,21 +58,31 @@ class ColumnVector : public Column {
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
 
 private:
     std::vector<T> data_;
 };
 
+using Int128 = absl::int128;
+using UInt128 = absl::uint128;
+using Int64 = int64_t;
+
 using ColumnUInt8   = ColumnVector<uint8_t>;
 using ColumnUInt16  = ColumnVector<uint16_t>;
 using ColumnUInt32  = ColumnVector<uint32_t>;
 using ColumnUInt64  = ColumnVector<uint64_t>;
+using ColumnUInt128 = ColumnVector<UInt128>;
 
 using ColumnInt8    = ColumnVector<int8_t>;
 using ColumnInt16   = ColumnVector<int16_t>;
 using ColumnInt32   = ColumnVector<int32_t>;
 using ColumnInt64   = ColumnVector<int64_t>;
+using ColumnInt128  = ColumnVector<Int128>;
 
 using ColumnFloat32 = ColumnVector<float>;
 using ColumnFloat64 = ColumnVector<double>;
diff --git a/lib/clickhouse-cpp/clickhouse/columns/string.cpp b/lib/clickhouse-cpp/clickhouse/columns/string.cpp
index 0791224..22ce9a6 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/string.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/string.cpp
@@ -3,6 +3,25 @@
 
 #include "../base/wire_format.h"
 
+namespace {
+
+constexpr size_t DEFAULT_BLOCK_SIZE = 4096;
+
+template <typename Container>
+size_t ComputeTotalSize(const Container & strings, size_t begin = 0, size_t len = -1) {
+    size_t result = 0;
+    if (begin < strings.size()) {
+        len = std::min(len, strings.size() - begin);
+
+        for (size_t i = begin; i < begin + len; ++i)
+            result += strings[i].size();
+    }
+
+    return result;
+}
+
+}
+
 namespace clickhouse {
 
 ColumnFixedString::ColumnFixedString(size_t n)
@@ -11,21 +30,42 @@ ColumnFixedString::ColumnFixedString(size_t n)
 {
 }
 
-void ColumnFixedString::Append(const std::string& str) {
-    data_.push_back(str);
-    data_.back().resize(string_size_);
+void ColumnFixedString::Reserve(size_t new_cap) {
+    data_.reserve(string_size_ * new_cap);
+}
+
+void ColumnFixedString::Append(std::string_view str) {
+    if (str.size() > string_size_) {
+        throw ValidationError("Expected string of length not greater than "
+                                 + std::to_string(string_size_) + " bytes, received "
+                                 + std::to_string(str.size()) + " bytes.");
+    }
+
+    if (data_.capacity() - data_.size() < str.size()) {
+        // round up to the next block size
+        const auto new_size = (((data_.size() + string_size_) / DEFAULT_BLOCK_SIZE) + 1) * DEFAULT_BLOCK_SIZE;
+        data_.reserve(new_size);
+    }
+
+    data_.insert(data_.size(), str);
+    // Pad up to string_size_ with zeroes.
+    if (str.size() < string_size_) {
+        const auto padding_size = string_size_ - str.size();
+        data_.resize(data_.size() + padding_size, char(0));
+    }
 }
 
 void ColumnFixedString::Clear() {
     data_.clear();
 }
 
-const std::string& ColumnFixedString::At(size_t n) const {
-    return data_.at(n);
+std::string_view ColumnFixedString::At(size_t n) const {
+    const auto pos = n * string_size_;
+    return std::string_view(&data_.at(pos), string_size_);
 }
 
-const std::string& ColumnFixedString::operator [] (size_t n) const {
-    return data_[n];
+size_t ColumnFixedString::FixedSize() const {
+       return string_size_;
 }
 
 void ColumnFixedString::Append(ColumnRef column) {
@@ -36,101 +76,266 @@ void ColumnFixedString::Append(ColumnRef column) {
     }
 }
 
-bool ColumnFixedString::Load(CodedInputStream* input, size_t rows) {
-    for (size_t i = 0; i < rows; ++i) {
-        std::string s;
-        s.resize(string_size_);
-
-        if (!WireFormat::ReadBytes(input, &s[0], s.size())) {
-            return false;
-        }
-
-        data_.push_back(s);
+bool ColumnFixedString::LoadBody(InputStream * input, size_t rows) {
+    data_.resize(string_size_ * rows);
+    if (!WireFormat::ReadBytes(*input, &data_[0], data_.size())) {
+        return false;
     }
 
     return true;
 }
 
-void ColumnFixedString::Save(CodedOutputStream* output) {
-    for (size_t i = 0; i < data_.size(); ++i) {
-        WireFormat::WriteBytes(output, data_[i].data(), string_size_);
-    }
+void ColumnFixedString::SaveBody(OutputStream* output) {
+    WireFormat::WriteBytes(*output, data_.data(), data_.size());
 }
 
 size_t ColumnFixedString::Size() const {
-    return data_.size();
+    return data_.size() / string_size_;
 }
 
-ColumnRef ColumnFixedString::Slice(size_t begin, size_t len) {
+ColumnRef ColumnFixedString::Slice(size_t begin, size_t len) const {
     auto result = std::make_shared<ColumnFixedString>(string_size_);
 
-    if (begin < data_.size()) {
-        result->data_ = SliceVector(data_, begin, len);
+    if (begin < Size()) {
+        const auto b = begin * string_size_;
+        const auto l = len * string_size_;
+        result->data_ = data_.substr(b, std::min(data_.size() - b, l));
     }
 
     return result;
 }
 
+ColumnRef ColumnFixedString::CloneEmpty() const {
+    return std::make_shared<ColumnFixedString>(string_size_);
+}
+
+void ColumnFixedString::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnFixedString &>(other);
+    std::swap(string_size_, col.string_size_);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnFixedString::GetItem(size_t index) const {
+    return ItemView{Type::FixedString, this->At(index)};
+}
+
+struct ColumnString::Block
+{
+    using CharT = typename std::string::value_type;
+
+    explicit Block(size_t starting_capacity)
+        : size(0),
+        capacity(starting_capacity),
+        data_(new CharT[capacity])
+    {}
+
+    inline auto GetAvailable() const {
+        return capacity - size;
+    }
+
+    std::string_view AppendUnsafe(std::string_view str) {
+        const auto pos = &data_[size];
+
+        // memcpy declares the source pointer with __attribute__((nonnull)),
+        // so calling it with str.data() == NULL is undefined behavior under
+        // UBSan even when str.size() is 0 (empty string_view from an
+        // empty std::string is allowed to expose a null data pointer).
+        // Guard with a zero-size short-circuit so the column code can keep
+        // appending empty strings without tripping the sanitizer.
+        if (str.size() > 0) {
+            memcpy(pos, str.data(), str.size());
+            size += str.size();
+        }
+
+        return std::string_view(pos, str.size());
+    }
+
+    auto GetCurrentWritePos() {
+        return &data_[size];
+    }
+
+    std::string_view ConsumeTailAsStringViewUnsafe(size_t len) {
+        const auto start = &data_[size];
+        size += len;
+        return std::string_view(start, len);
+    }
+
+    size_t size;
+    const size_t capacity;
+    std::unique_ptr<CharT[]> data_;
+};
 
 ColumnString::ColumnString()
     : Column(Type::CreateString())
 {
 }
 
-ColumnString::ColumnString(const std::vector<std::string>& data)
+ColumnString::ColumnString(size_t element_count)
     : Column(Type::CreateString())
-    , data_(data)
 {
+    items_.reserve(element_count);
+    // 16 is arbitrary number, assumption that string values are about ~256 bytes long.
+    blocks_.reserve(std::max<size_t>(1, element_count / 16));
 }
 
-void ColumnString::Append(const std::string& str) {
-    data_.push_back(str);
+ColumnString::ColumnString(const std::vector<std::string>& data)
+    : ColumnString()
+{
+    items_.reserve(data.size());
+    blocks_.emplace_back(ComputeTotalSize(data));
+
+    for (const auto & s : data) {
+        AppendUnsafe(s);
+    }
 }
 
-void ColumnString::Clear() {
-    data_.clear();
+ColumnString::ColumnString(std::vector<std::string>&& data)
+    : ColumnString()
+{
+    items_.reserve(data.size());
+
+    for (auto&& d : data) {
+        append_data_.emplace_back(std::move(d));
+        auto& last_data = append_data_.back();
+        items_.emplace_back(std::string_view{ last_data.data(),last_data.length() });
+    }
+}
+
+ColumnString::~ColumnString()
+{}
+
+void ColumnString::Reserve(size_t new_cap) {
+    items_.reserve(new_cap);
+    // 16 is arbitrary number, assumption that string values are about ~256 bytes long.
+    blocks_.reserve(std::max<size_t>(1, new_cap / 16));
+}
+
+void ColumnString::Append(std::string_view str) {
+    if (blocks_.size() == 0 || blocks_.back().GetAvailable() < str.length()) {
+        blocks_.emplace_back(std::max(DEFAULT_BLOCK_SIZE, str.size()));
+    }
+
+    items_.emplace_back(blocks_.back().AppendUnsafe(str));
 }
 
-const std::string& ColumnString::At(size_t n) const {
-    return data_.at(n);
+void ColumnString::Append(const char* str) {
+    Append(std::string_view(str, strlen(str)));
 }
 
-const std::string& ColumnString::operator [] (size_t n) const {
-    return data_[n];
+void ColumnString::Append(std::string&& steal_value) {
+    append_data_.emplace_back(std::move(steal_value));
+    auto& last_data = append_data_.back();
+    items_.emplace_back(std::string_view{ last_data.data(),last_data.length() });
+}
+
+void ColumnString::AppendNoManagedLifetime(std::string_view str) {
+    items_.emplace_back(str);
+}
+
+void ColumnString::AppendUnsafe(std::string_view str) {
+    items_.emplace_back(blocks_.back().AppendUnsafe(str));
+}
+
+void ColumnString::Clear() {
+    items_.clear();
+    blocks_.clear();
+    append_data_.clear();
+}
+
+std::string_view ColumnString::At(size_t n) const {
+    return items_.at(n);
 }
 
 void ColumnString::Append(ColumnRef column) {
     if (auto col = column->As<ColumnString>()) {
-        data_.insert(data_.end(), col->data_.begin(), col->data_.end());
+        const auto total_size = ComputeTotalSize(col->items_);
+
+        // TODO: fill up existing block with some items and then add a new one for the rest of items
+        if (blocks_.size() == 0 || blocks_.back().GetAvailable() < total_size)
+            blocks_.emplace_back(std::max(DEFAULT_BLOCK_SIZE, total_size));
+
+        // Intentionally not doing items_.reserve() since that cripples performance.
+        for (size_t i = 0; i < column->Size(); ++i) {
+            this->AppendUnsafe((*col)[i]);
+        }
     }
 }
 
-bool ColumnString::Load(CodedInputStream* input, size_t rows) {
+bool ColumnString::LoadBody(InputStream* input, size_t rows) {
+    if (rows == 0) {
+        items_.clear();
+        blocks_.clear();
+
+        return true;
+    }
+
+    decltype(items_) new_items;
+    decltype(blocks_) new_blocks;
+
+    new_items.reserve(rows);
+
+    // Suboptimzal if the first row string is >DEFAULT_BLOCK_SIZE, but that must be a very rare case.
+    Block * block = &new_blocks.emplace_back(DEFAULT_BLOCK_SIZE);
+
     for (size_t i = 0; i < rows; ++i) {
-        std::string s;
+        uint64_t len;
+        if (!WireFormat::ReadUInt64(*input, &len))
+            return false;
+
+        if (len > block->GetAvailable())
+            block = &new_blocks.emplace_back(std::max<size_t>(DEFAULT_BLOCK_SIZE, len));
 
-        if (!WireFormat::ReadString(input, &s)) {
+        if (!WireFormat::ReadBytes(*input, block->GetCurrentWritePos(), len))
             return false;
-        }
 
-        data_.push_back(s);
+        new_items.emplace_back(block->ConsumeTailAsStringViewUnsafe(len));
     }
 
+    items_.swap(new_items);
+    blocks_.swap(new_blocks);
+
     return true;
 }
 
-void ColumnString::Save(CodedOutputStream* output) {
-    for (auto si = data_.begin(); si != data_.end(); ++si) {
-        WireFormat::WriteString(output, *si);
+void ColumnString::SaveBody(OutputStream* output) {
+    for (const auto & item : items_) {
+        WireFormat::WriteString(*output, item);
     }
 }
 
 size_t ColumnString::Size() const {
-    return data_.size();
+    return items_.size();
+}
+
+ColumnRef ColumnString::Slice(size_t begin, size_t len) const {
+    auto result = std::make_shared<ColumnString>();
+
+    if (begin < items_.size()) {
+        len = std::min(len, items_.size() - begin);
+        result->items_.reserve(len);
+
+        result->blocks_.emplace_back(ComputeTotalSize(items_, begin, len));
+        for (size_t i = begin; i < begin + len; ++i) {
+            result->Append(items_[i]);
+        }
+    }
+
+    return result;
+}
+
+ColumnRef ColumnString::CloneEmpty() const {
+    return std::make_shared<ColumnString>();
+}
+
+void ColumnString::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnString &>(other);
+    items_.swap(col.items_);
+    blocks_.swap(col.blocks_);
+    append_data_.swap(col.append_data_);
 }
 
-ColumnRef ColumnString::Slice(size_t begin, size_t len) {
-    return std::make_shared<ColumnString>(SliceVector(data_, begin, len));
+ItemView ColumnString::GetItem(size_t index) const {
+    return ItemView{Type::String, this->At(index)};
 }
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/string.h b/lib/clickhouse-cpp/clickhouse/columns/string.h
index 196ecba..d600655 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/string.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/string.h
@@ -2,6 +2,12 @@
 
 #include "column.h"
 
+#include <string>
+#include <string_view>
+#include <utility>
+#include <vector>
+#include <deque>
+
 namespace clickhouse {
 
 /**
@@ -9,27 +15,43 @@ namespace clickhouse {
  */
 class ColumnFixedString : public Column {
 public:
+    using ValueType = std::string_view;
+
     explicit ColumnFixedString(size_t n);
 
+    template <typename Values>
+    ColumnFixedString(size_t n, const Values & values)
+        : ColumnFixedString(n)
+    {
+        for (const auto & v : values)
+            Append(v);
+    }
+
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t) override;
+
     /// Appends one element to the column.
-    void Append(const std::string& str);
+    void Append(std::string_view str);
 
     /// Returns element at given row number.
-    const std::string& At(size_t n) const;
+    std::string_view At(size_t n) const;
 
     /// Returns element at given row number.
-    const std::string& operator [] (size_t n) const;
+    inline std::string_view operator [] (size_t n) const { return At(n); }
+
+    /// Returns the max size of the fixed string
+    size_t FixedSize() const;
 
 public:
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
-    
+    void SaveBody(OutputStream* output) override;
+
     /// Clear column data .
     void Clear() override;
 
@@ -37,41 +59,66 @@ class ColumnFixedString : public Column {
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t) const override;
 
 private:
-    const size_t string_size_;
-    std::vector<std::string> data_;
+    size_t string_size_;
+    std::string data_;
 };
 
-
 /**
  * Represents column of variable-length strings.
  */
 class ColumnString : public Column {
 public:
+    // Type this column takes as argument of Append and returns with At() and operator[]
+    using ValueType = std::string_view;
+
     ColumnString();
-    explicit ColumnString(const std::vector<std::string>& data);
+    ~ColumnString();
+
+    explicit ColumnString(size_t element_count);
+    explicit ColumnString(const std::vector<std::string> & data);
+    explicit ColumnString(std::vector<std::string>&& data);
+    ColumnString& operator=(const ColumnString&) = delete;
+    ColumnString(const ColumnString&) = delete;
+
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
 
     /// Appends one element to the column.
-    void Append(const std::string& str);
+    void Append(std::string_view str);
+
+    /// Appends one element to the column.
+    void Append(const char* str);
+
+    /// Appends one element to the column.
+    void Append(std::string&& steal_value);
+
+    /// Appends one element to the column.
+    /// If str lifetime is managed elsewhere and guaranteed to outlive the Block sent to the server
+    void AppendNoManagedLifetime(std::string_view str);
 
     /// Returns element at given row number.
-    const std::string& At(size_t n) const;
+    std::string_view At(size_t n) const;
 
     /// Returns element at given row number.
-    const std::string& operator [] (size_t n) const;
+    inline std::string_view operator [] (size_t n) const { return At(n); }
 
 public:
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
-    
+    void SaveBody(OutputStream* output) override;
+
     /// Clear column data .
     void Clear() override;
 
@@ -79,10 +126,20 @@ class ColumnString : public Column {
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+    ItemView GetItem(size_t) const override;
 
 private:
-    std::vector<std::string> data_;
+    void AppendUnsafe(std::string_view);
+
+private:
+    struct Block;
+
+    std::vector<std::string_view> items_;
+    std::vector<Block> blocks_;
+    std::deque<std::string> append_data_;
 };
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/time.cpp b/lib/clickhouse-cpp/clickhouse/columns/time.cpp
new file mode 100644
index 0000000..045efb2
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/time.cpp
@@ -0,0 +1,155 @@
+#include "time.h"
+
+namespace clickhouse {
+
+ColumnTime::ColumnTime() 
+    : ColumnTime(Type::CreateTime(), std::make_shared<ColumnInt32>()) {}
+
+ColumnTime::ColumnTime(std::vector<int32_t>&& data)
+    : ColumnTime(Type::CreateTime(), std::make_shared<ColumnInt32>(std::move(data))) {}
+
+ColumnTime::ColumnTime(TypeRef type, std::shared_ptr<ColumnInt32> data)
+    : Column(std::move(type)),
+      data_(std::move(data))
+{}
+
+void ColumnTime::Append(ValueType value) {
+    data_->Append(value);
+}
+
+ColumnTime::ValueType ColumnTime::At(size_t n) const {
+    return data_->At(n);
+}
+
+std::vector<int32_t>& ColumnTime::GetWritableData() {
+    return data_->GetWritableData();
+}
+
+void ColumnTime::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+size_t ColumnTime::Capacity() const {
+    return data_->Capacity();
+}
+
+void ColumnTime::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnTime>()) {
+        data_->Append(col->data_);
+    }
+}
+
+bool ColumnTime::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnTime::Clear() {
+    data_->Clear();
+}
+
+void ColumnTime::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+size_t ColumnTime::Size() const {
+    return data_->Size();
+}
+
+ColumnRef ColumnTime::Slice(size_t begin, size_t len) const {
+    auto sliced_data = data_->Slice(begin, len)->As<ColumnInt32>();
+    return ColumnRef{new ColumnTime(type_, sliced_data)};
+}
+
+ColumnRef ColumnTime::CloneEmpty() const {
+    return ColumnRef{new ColumnTime(type_, data_->CloneEmpty()->As<ColumnInt32>())};
+}
+
+void ColumnTime::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnTime &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnTime::GetItem(size_t index) const {
+    return ItemView{Type::Time, data_->GetItem(index)};
+}
+
+ColumnTime64::ColumnTime64(size_t precision) 
+    : ColumnTime64(Type::CreateTime64(precision), std::make_shared<ColumnInt64>())
+{}
+
+ColumnTime64::ColumnTime64(size_t precision, std::vector<int64_t>&& data)
+    : ColumnTime64(Type::CreateTime64(precision), std::make_shared<ColumnInt64>(std::move(data)))
+{}
+
+ColumnTime64::ColumnTime64(TypeRef type, std::shared_ptr<ColumnInt64> data)
+    : Column(std::move(type)),
+      data_(std::move(data)),
+      precision_{type_->As<Time64Type>()->GetPrecision()}
+{}
+
+void ColumnTime64::Append(ValueType value) {
+    data_->Append(value);
+}
+
+ColumnTime64::ValueType ColumnTime64::At(size_t n) const {
+    return data_->At(n);
+}
+
+std::vector<int64_t>& ColumnTime64::GetWritableData() {
+    return data_->GetWritableData();
+}
+
+void ColumnTime64::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
+}
+
+size_t ColumnTime64::Capacity() const {
+    return data_->Capacity();
+}
+
+void ColumnTime64::Append(ColumnRef column) {
+    if (auto col = column->As<ColumnTime64>()) {
+        data_->Append(col->data_);
+    }
+}
+
+bool ColumnTime64::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows);
+}
+
+void ColumnTime64::Clear() {
+    data_->Clear();
+}
+
+void ColumnTime64::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
+}
+
+size_t ColumnTime64::Size() const {
+    return data_->Size();
+}
+
+ColumnRef ColumnTime64::Slice(size_t begin, size_t len) const {
+    auto sliced_data = data_->Slice(begin, len)->As<ColumnInt64>();
+    return ColumnRef{new ColumnTime64(type_, sliced_data)};
+}
+
+ColumnRef ColumnTime64::CloneEmpty() const {
+    return ColumnRef{new ColumnTime64(type_, data_->CloneEmpty()->As<ColumnInt64>())};
+}
+
+void ColumnTime64::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnTime64 &>(other);
+    if (col.GetPrecision() != GetPrecision()) {
+        throw ValidationError("Can't swap Time64 columns when precisions are not the same: "
+            + std::to_string(GetPrecision()) + "(this) != "
+            + std::to_string(col.GetPrecision()) + "(that)");
+    }
+    data_.swap(col.data_);
+}
+
+ItemView ColumnTime64::GetItem(size_t index) const {
+    return ItemView{Type::Time64, data_->GetItem(index)};
+}
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/columns/time.h b/lib/clickhouse-cpp/clickhouse/columns/time.h
new file mode 100644
index 0000000..68ddead
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/columns/time.h
@@ -0,0 +1,118 @@
+#pragma once
+
+#include "column.h"
+#include "numeric.h"
+
+namespace clickhouse {
+
+class ColumnTime : public Column {
+public:
+    using ValueType = int32_t;
+
+    ColumnTime();
+    explicit ColumnTime(std::vector<int32_t>&& data);
+
+    /// Appends one element to the end of column.
+    void Append(ValueType value);
+
+    /// Returns element at given row number.
+    ValueType At(size_t n) const;
+    ValueType operator[](size_t n) const { return At(n); }
+
+    /// Get Raw Vector Contents
+    std::vector<int32_t>& GetWritableData();
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Returns the capacity of the column
+    size_t Capacity() const;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
+
+private:
+    ColumnTime(TypeRef type, std::shared_ptr<ColumnInt32> data);
+
+private:
+    std::shared_ptr<ColumnInt32> data_;
+};
+
+class ColumnTime64 : public Column {
+public:
+    using ValueType = int64_t;
+
+    explicit ColumnTime64(size_t precision);
+    ColumnTime64(size_t precision, std::vector<int64_t>&& data);
+
+    /// Appends one element to the end of column.
+    void Append(ValueType value);
+
+    /// Returns element at given row number.
+    ValueType At(size_t n) const;
+
+    ValueType operator[](size_t n) const { return At(n); }
+
+    /// Get Raw Vector Contents
+    std::vector<int64_t>& GetWritableData();
+
+public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
+    /// Returns the capacity of the column
+    size_t Capacity() const;
+
+    /// Appends content of given column to the end of current one.
+    void Append(ColumnRef column) override;
+
+    /// Loads column data from input stream.
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Clear column data .
+    void Clear() override;
+
+    /// Saves column data to output stream.
+    void SaveBody(OutputStream* output) override;
+
+    /// Returns count of rows in the column.
+    size_t Size() const override;
+
+    /// Makes slice of the current column.
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t index) const override;
+
+    size_t GetPrecision() const { return precision_; };
+
+private:
+    ColumnTime64(TypeRef type, std::shared_ptr<ColumnInt64> data);
+
+private:
+    std::shared_ptr<ColumnInt64> data_;
+    const size_t precision_;
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/tuple.cpp b/lib/clickhouse-cpp/clickhouse/columns/tuple.cpp
index 9a009e3..5685859 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/tuple.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/tuple.cpp
@@ -16,17 +16,55 @@ ColumnTuple::ColumnTuple(const std::vector<ColumnRef>& columns)
 {
 }
 
+size_t ColumnTuple::TupleSize() const {
+    return columns_.size();
+}
+
+void ColumnTuple::Reserve(size_t new_cap) {
+    for (auto& column : columns_) {
+        column->Reserve(new_cap);
+    }  
+}
+
+void ColumnTuple::Append(ColumnRef column) {
+    if (!this->Type()->IsEqual(column->Type())) {
+        throw ValidationError(
+            "can't append column of type " + column->Type()->GetName() + " "
+            "to column type " + this->Type()->GetName());
+    }
+    const auto & source_tuple_column = column->As<ColumnTuple>();
+    for (size_t ci = 0; ci < columns_.size(); ++ci) {
+        columns_[ci]->Append((*source_tuple_column)[ci]);
+    }
+}
 size_t ColumnTuple::Size() const {
     return columns_.empty() ? 0 : columns_[0]->Size();
 }
 
-size_t ColumnTuple::tupleSize() {
-    return columns_.empty() ? 0 : columns_.size();
+ColumnRef ColumnTuple::Slice(size_t begin, size_t len) const {
+    std::vector<ColumnRef> sliced_columns;
+    sliced_columns.reserve(columns_.size());
+    for(const auto &column : columns_) {
+        sliced_columns.push_back(column->Slice(begin, len));
+    }
+
+    return std::make_shared<ColumnTuple>(sliced_columns);
+}
+
+ColumnRef ColumnTuple::CloneEmpty() const {
+    std::vector<ColumnRef> result_columns;
+    result_columns.reserve(columns_.size());
+
+    for(const auto &column : columns_) {
+        result_columns.push_back(column->CloneEmpty());
+    }
+
+    return std::make_shared<ColumnTuple>(result_columns);
 }
 
-bool ColumnTuple::Load(CodedInputStream* input, size_t rows) {
+bool ColumnTuple::LoadPrefix(InputStream* input, size_t rows) {
     for (auto ci = columns_.begin(); ci != columns_.end(); ++ci) {
-        if (!(*ci)->Load(input, rows)) {
+        if (!(*ci)->LoadPrefix(input, rows)) {
             return false;
         }
     }
@@ -34,9 +72,25 @@ bool ColumnTuple::Load(CodedInputStream* input, size_t rows) {
     return true;
 }
 
-void ColumnTuple::Save(CodedOutputStream* output) {
+bool ColumnTuple::LoadBody(InputStream* input, size_t rows) {
     for (auto ci = columns_.begin(); ci != columns_.end(); ++ci) {
-        (*ci)->Save(output);
+        if (!(*ci)->LoadBody(input, rows)) {
+            return false;
+        }
+    }
+
+    return true;
+}
+
+void ColumnTuple::SavePrefix(OutputStream* output) {
+    for (auto & column : columns_) {
+        column->SavePrefix(output);
+    }
+}
+
+void ColumnTuple::SaveBody(OutputStream* output) {
+    for (auto & column : columns_) {
+        column->SaveBody(output);
     }
 }
 
@@ -44,4 +98,9 @@ void ColumnTuple::Clear() {
     columns_.clear();
 }
 
+void ColumnTuple::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnTuple &>(other);
+    columns_.swap(col.columns_);
+}
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/tuple.h b/lib/clickhouse-cpp/clickhouse/columns/tuple.h
index f261c52..ebc1b89 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/tuple.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/tuple.h
@@ -1,6 +1,7 @@
 #pragma once
 
 #include "column.h"
+#include "utils.h"
 
 #include <vector>
 
@@ -13,22 +14,36 @@ class ColumnTuple : public Column {
 public:
     ColumnTuple(const std::vector<ColumnRef>& columns);
 
-    ColumnRef operator [] (size_t n) const {
+    /// Returns count of columns in the tuple.
+    size_t TupleSize() const;
+
+    inline ColumnRef operator [] (size_t n) const {
         return columns_[n];
     }
 
-    size_t tupleSize();
+    inline ColumnRef At(size_t n) const {
+        return columns_[n];
+    }
 
 public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
     /// Appends content of given column to the end of current one.
-    void Append(ColumnRef) override { }
+    void Append(ColumnRef column) override;
+
+    /// Loads column prefix from input stream.
+    bool LoadPrefix(InputStream* input, size_t rows) override;
 
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
+
+    /// Saves column prefix to output stream.
+    void SavePrefix(OutputStream* output) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
-    
+    void SaveBody(OutputStream* output) override;
+
     /// Clear column data .
     void Clear() override;
 
@@ -36,10 +51,128 @@ class ColumnTuple : public Column {
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t, size_t) override { return ColumnRef(); }
+    ColumnRef Slice(size_t, size_t) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
 
 private:
     std::vector<ColumnRef> columns_;
 };
 
-}
+template <typename... Columns>
+class ColumnTupleT : public ColumnTuple {
+public:
+    using TupleOfColumns = std::tuple<std::shared_ptr<Columns>...>;
+
+    using ValueType = std::tuple<std::decay_t<decltype(std::declval<Columns>().At(0))>...>;
+
+    ColumnTupleT(std::tuple<std::shared_ptr<Columns>...> columns)
+        : ColumnTuple(TupleToVector(columns)), typed_columns_(std::move(columns)) {}
+
+    ColumnTupleT(std::vector<ColumnRef> columns)
+        : ColumnTuple(columns), typed_columns_(VectorToTuple(std::move(columns))) {}
+
+    ColumnTupleT(const std::initializer_list<ColumnRef> columns)
+        : ColumnTuple(columns), typed_columns_(VectorToTuple(std::move(columns))) {}
+
+    inline ValueType At(size_t index) const { return GetTupleOfValues(index); }
+
+    inline ValueType operator[](size_t index) const { return GetTupleOfValues(index); }
+
+    using ColumnTuple::Append;
+
+    template <typename... T>
+    inline void Append(std::tuple<T...> value) {
+        AppendTuple(std::move(value));
+    }
+
+    /** Create a ColumnTupleT from a ColumnTuple, without copying data and offsets, but by
+     * 'stealing' those from `col`.
+     *
+     *  Ownership of column internals is transferred to returned object, original (argument) object
+     *  MUST NOT BE USED IN ANY WAY, it is only safe to dispose it.
+     *
+     *  Throws an exception if `col` is of wrong type, it is safe to use original col in this case.
+     *  This is a static method to make such conversion verbose.
+     */
+    static auto Wrap(ColumnTuple&& col) {
+        if (col.TupleSize() != std::tuple_size_v<TupleOfColumns>) {
+            throw ValidationError("Can't wrap from " + col.GetType().GetName());
+        }
+        return std::make_shared<ColumnTupleT<Columns...>>(VectorToTuple(std::move(col)));
+    }
+
+    static auto Wrap(Column&& col) { return Wrap(std::move(dynamic_cast<ColumnTuple&&>(col))); }
+
+    // Helper to simplify integration with other APIs
+    static auto Wrap(ColumnRef&& col) { return Wrap(std::move(*col->AsStrict<ColumnTuple>())); }
+
+    ColumnRef Slice(size_t begin, size_t size) const override {
+        return Wrap(ColumnTuple::Slice(begin, size));
+    }
+
+    ColumnRef CloneEmpty() const override { return Wrap(ColumnTuple::CloneEmpty()); }
+
+    void Swap(Column& other) override {
+        auto& col = dynamic_cast<ColumnTupleT<Columns...>&>(other);
+        typed_columns_.swap(col.typed_columns_);
+        ColumnTuple::Swap(other);
+    }
+
+private:
+    template <typename T, size_t index = std::tuple_size_v<T>>
+    inline void AppendTuple([[maybe_unused]] T value) {
+        static_assert(index <= std::tuple_size_v<T>);
+        static_assert(std::tuple_size_v<TupleOfColumns> == std::tuple_size_v<T>);
+        if constexpr (index == 0) {
+            return;
+        } else {
+            std::get<index - 1>(typed_columns_)->Append(std::move(std::get<index - 1>(value)));
+            AppendTuple<T, index - 1>(std::move(value));
+        }
+    }
+
+    template <typename T, size_t index = std::tuple_size_v<T>>
+    inline static std::vector<ColumnRef> TupleToVector([[maybe_unused]] const T& value) {
+        static_assert(index <= std::tuple_size_v<T>);
+        if constexpr (index == 0) {
+            std::vector<ColumnRef> result;
+            result.reserve(std::tuple_size_v<T>);
+            return result;
+        } else {
+            auto result = TupleToVector<T, index - 1>(value);
+            result.push_back(std::get<index - 1>(value));
+            return result;
+        }
+    }
+
+    template <typename T, size_t column_index = std::tuple_size_v<TupleOfColumns>>
+    inline static auto VectorToTuple([[maybe_unused]] T columns) {
+        static_assert(column_index <= std::tuple_size_v<TupleOfColumns>);
+        if constexpr (column_index == 0) {
+            return std::make_tuple();
+        } else {
+            using ColumnType =
+                typename std::tuple_element<column_index - 1, TupleOfColumns>::type::element_type;
+            auto column = WrapColumn<ColumnType>(columns[column_index - 1]);
+            return std::tuple_cat(std::move(VectorToTuple<T, column_index - 1>(std::move(columns))),
+                                  std::make_tuple(std::move(column)));
+        }
+    }
+
+    template <size_t column_index = std::tuple_size_v<TupleOfColumns>>
+    inline auto GetTupleOfValues([[maybe_unused]]size_t index) const {
+        static_assert(column_index <= std::tuple_size_v<TupleOfColumns>);
+        if constexpr (column_index == 0) {
+            return std::make_tuple();
+        } else {
+            return std::tuple_cat(
+                std::move(GetTupleOfValues<column_index - 1>(index)),
+                std::move(std::make_tuple(std::get<column_index - 1>(typed_columns_)->At(index))));
+        }
+    }
+
+    TupleOfColumns typed_columns_;
+};
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/columns/utils.h b/lib/clickhouse-cpp/clickhouse/columns/utils.h
index bcc5b98..0fb8b99 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/utils.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/utils.h
@@ -2,6 +2,7 @@
 
 #include <algorithm>
 #include <vector>
+#include <memory>
 
 namespace clickhouse {
 
@@ -17,4 +18,24 @@ std::vector<T> SliceVector(const std::vector<T>& vec, size_t begin, size_t len)
     return result;
 }
 
+template <typename T>
+struct HasWrapMethod {
+private:
+    static int detect(...);
+    template <typename U>
+    static decltype(U::Wrap(std::move(std::declval<ColumnRef>()))) detect(const U&);
+
+public:
+    static constexpr bool value = !std::is_same<int, decltype(detect(std::declval<T>()))>::value;
+};
+
+template <typename T>
+inline std::shared_ptr<T> WrapColumn(ColumnRef&& column) {
+    if constexpr (HasWrapMethod<T>::value) {
+        return T::Wrap(std::move(column));
+    } else {
+        return column->template AsStrict<T>();
+    }
+}
+
 }
diff --git a/lib/clickhouse-cpp/clickhouse/columns/uuid.cpp b/lib/clickhouse-cpp/clickhouse/columns/uuid.cpp
index 9ac538a..fbaff97 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/uuid.cpp
+++ b/lib/clickhouse-cpp/clickhouse/columns/uuid.cpp
@@ -1,5 +1,6 @@
 #include "uuid.h"
 #include "utils.h"
+#include "../exceptions.h"
 
 #include <stdexcept>
 
@@ -15,12 +16,12 @@ ColumnUUID::ColumnUUID(ColumnRef data)
     : Column(Type::CreateUUID())
     , data_(data->As<ColumnUInt64>())
 {
-    if (data_->Size()%2 != 0) {
-        throw std::runtime_error("number of entries must be even (two 64-bit numbers for each UUID)");
+    if (data_->Size() % 2 != 0) {
+        throw ValidationError("number of entries must be even (two 64-bit numbers for each UUID)");
     }
 }
 
-void ColumnUUID::Append(const UInt128& value) {
+void ColumnUUID::Append(const UUID& value) {
     data_->Append(value.first);
     data_->Append(value.second);
 }
@@ -29,35 +30,51 @@ void ColumnUUID::Clear() {
     data_->Clear();
 }
 
-const UInt128 ColumnUUID::At(size_t n) const {
-    return UInt128(data_->At(n * 2), data_->At(n * 2 + 1));
+const UUID ColumnUUID::At(size_t n) const {
+    return UUID(data_->At(n * 2), data_->At(n * 2 + 1));
 }
 
-const UInt128 ColumnUUID::operator [] (size_t n) const {
-    return UInt128((*data_)[n * 2], (*data_)[n * 2 + 1]);
+void ColumnUUID::Reserve(size_t new_cap) {
+    data_->Reserve(new_cap);
 }
 
 void ColumnUUID::Append(ColumnRef column) {
     if (auto col = column->As<ColumnUUID>()) {
-        data_->Append(data_);
+        data_->Append(col->data_);
     }
 }
 
-bool ColumnUUID::Load(CodedInputStream* input, size_t rows) {
-    return data_->Load(input, rows * 2);
+bool ColumnUUID::LoadBody(InputStream* input, size_t rows) {
+    return data_->LoadBody(input, rows * 2);
 }
 
-void ColumnUUID::Save(CodedOutputStream* output) {
-    data_->Save(output);
+void ColumnUUID::SaveBody(OutputStream* output) {
+    data_->SaveBody(output);
 }
 
 size_t ColumnUUID::Size() const {
     return data_->Size() / 2;
 }
 
-ColumnRef ColumnUUID::Slice(size_t begin, size_t len) {
+ColumnRef ColumnUUID::Slice(size_t begin, size_t len) const {
     return std::make_shared<ColumnUUID>(data_->Slice(begin * 2, len * 2));
 }
 
+ColumnRef ColumnUUID::CloneEmpty() const {
+    return std::make_shared<ColumnUUID>();
 }
 
+void ColumnUUID::Swap(Column& other) {
+    auto & col = dynamic_cast<ColumnUUID &>(other);
+    data_.swap(col.data_);
+}
+
+ItemView ColumnUUID::GetItem(size_t index) const {
+    // We know that ColumnUInt64 stores it's data in continius memory region,
+    // and that every 2 values from data represent 1 UUID value.
+    const auto data_item_view = data_->GetItem(index * 2);
+
+    return ItemView{Type::UUID, std::string_view{data_item_view.data.data(), data_item_view.data.size() * 2}};
+}
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/columns/uuid.h b/lib/clickhouse-cpp/clickhouse/columns/uuid.h
index f5b5725..ccd03f8 100644
--- a/lib/clickhouse-cpp/clickhouse/columns/uuid.h
+++ b/lib/clickhouse-cpp/clickhouse/columns/uuid.h
@@ -1,11 +1,11 @@
 #pragma once
 
+#include "../base/uuid.h"
 #include "column.h"
 #include "numeric.h"
 
 namespace clickhouse {
 
-using UInt128 = std::pair<uint64_t, uint64_t>;
 
 /**
  * Represents a UUID column.
@@ -17,24 +17,27 @@ class ColumnUUID : public Column {
     explicit ColumnUUID(ColumnRef data);
 
     /// Appends one element to the end of column.
-    void Append(const UInt128& value);
+    void Append(const UUID& value);
 
     /// Returns element at given row number.
-    const UInt128 At(size_t n) const;
+    const UUID At(size_t n) const;
 
     /// Returns element at given row number.
-    const UInt128 operator [] (size_t n) const;
+    inline const UUID operator [] (size_t n) const { return At(n); }
 
 public:
+    /// Increase the capacity of the column for large block insertion.
+    void Reserve(size_t new_cap) override;
+
     /// Appends content of given column to the end of current one.
     void Append(ColumnRef column) override;
 
     /// Loads column data from input stream.
-    bool Load(CodedInputStream* input, size_t rows) override;
+    bool LoadBody(InputStream* input, size_t rows) override;
 
     /// Saves column data to output stream.
-    void Save(CodedOutputStream* output) override;
-    
+    void SaveBody(OutputStream* output) override;
+
     /// Clear column data .
     void Clear() override;
 
@@ -42,7 +45,11 @@ class ColumnUUID : public Column {
     size_t Size() const override;
 
     /// Makes slice of the current column.
-    ColumnRef Slice(size_t begin, size_t len) override;
+    ColumnRef Slice(size_t begin, size_t len) const override;
+    ColumnRef CloneEmpty() const override;
+    void Swap(Column& other) override;
+
+    ItemView GetItem(size_t) const override;
 
 private:
     std::shared_ptr<ColumnUInt64> data_;
diff --git a/lib/clickhouse-cpp/clickhouse/error_codes.h b/lib/clickhouse-cpp/clickhouse/error_codes.h
index 77e88ee..815a984 100644
--- a/lib/clickhouse-cpp/clickhouse/error_codes.h
+++ b/lib/clickhouse-cpp/clickhouse/error_codes.h
@@ -2,27 +2,593 @@
 
 namespace clickhouse {
 
+// based on https://github.com/ClickHouse/ClickHouse/blob/54ae88a859507722821624476c3818152f944055/src/Common/ErrorCodes.cpp
+// (master on 28 Feb 2024)
 enum ErrorCodes {
+    OK = 0,
+    UNSUPPORTED_METHOD = 1,
+    UNSUPPORTED_PARAMETER = 2,
+    UNEXPECTED_END_OF_FILE = 3,
+    EXPECTED_END_OF_FILE = 4,
+    CANNOT_PARSE_TEXT = 6,
+    INCORRECT_NUMBER_OF_COLUMNS = 7,
+    THERE_IS_NO_COLUMN = 8,
+    SIZES_OF_COLUMNS_DOESNT_MATCH = 9,
+    NOT_FOUND_COLUMN_IN_BLOCK = 10,
+    POSITION_OUT_OF_BOUND = 11,
+    PARAMETER_OUT_OF_BOUND = 12,
+    SIZES_OF_COLUMNS_IN_TUPLE_DOESNT_MATCH = 13,
+    DUPLICATE_COLUMN = 15,
+    NO_SUCH_COLUMN_IN_TABLE = 16,
+    SIZE_OF_FIXED_STRING_DOESNT_MATCH = 19,
+    NUMBER_OF_COLUMNS_DOESNT_MATCH = 20,
+    CANNOT_READ_FROM_ISTREAM = 23,
+    CANNOT_WRITE_TO_OSTREAM = 24,
+    CANNOT_PARSE_ESCAPE_SEQUENCE = 25,
+    CANNOT_PARSE_QUOTED_STRING = 26,
+    CANNOT_PARSE_INPUT_ASSERTION_FAILED = 27,
+    CANNOT_PRINT_FLOAT_OR_DOUBLE_NUMBER = 28,
+    ATTEMPT_TO_READ_AFTER_EOF = 32,
+    CANNOT_READ_ALL_DATA = 33,
+    TOO_MANY_ARGUMENTS_FOR_FUNCTION = 34,
+    TOO_FEW_ARGUMENTS_FOR_FUNCTION = 35,
+    BAD_ARGUMENTS = 36,
+    UNKNOWN_ELEMENT_IN_AST = 37,
+    CANNOT_PARSE_DATE = 38,
+    TOO_LARGE_SIZE_COMPRESSED = 39,
     CHECKSUM_DOESNT_MATCH = 40,
     CANNOT_PARSE_DATETIME = 41,
+    NUMBER_OF_ARGUMENTS_DOESNT_MATCH = 42,
+    ILLEGAL_TYPE_OF_ARGUMENT = 43,
+    ILLEGAL_COLUMN = 44,
     UNKNOWN_FUNCTION = 46,
     UNKNOWN_IDENTIFIER = 47,
+    NOT_IMPLEMENTED = 48,
+    LOGICAL_ERROR = 49,
+    UNKNOWN_TYPE = 50,
+    EMPTY_LIST_OF_COLUMNS_QUERIED = 51,
+    COLUMN_QUERIED_MORE_THAN_ONCE = 52,
+    TYPE_MISMATCH = 53,
+    STORAGE_REQUIRES_PARAMETER = 55,
+    UNKNOWN_STORAGE = 56,
     TABLE_ALREADY_EXISTS = 57,
+    TABLE_METADATA_ALREADY_EXISTS = 58,
+    ILLEGAL_TYPE_OF_COLUMN_FOR_FILTER = 59,
     UNKNOWN_TABLE = 60,
     SYNTAX_ERROR = 62,
+    UNKNOWN_AGGREGATE_FUNCTION = 63,
+    CANNOT_GET_SIZE_OF_FIELD = 68,
+    ARGUMENT_OUT_OF_BOUND = 69,
+    CANNOT_CONVERT_TYPE = 70,
+    CANNOT_WRITE_AFTER_END_OF_BUFFER = 71,
+    CANNOT_PARSE_NUMBER = 72,
+    UNKNOWN_FORMAT = 73,
+    CANNOT_READ_FROM_FILE_DESCRIPTOR = 74,
+    CANNOT_WRITE_TO_FILE_DESCRIPTOR = 75,
+    CANNOT_OPEN_FILE = 76,
+    CANNOT_CLOSE_FILE = 77,
+    UNKNOWN_TYPE_OF_QUERY = 78,
+    INCORRECT_FILE_NAME = 79,
+    INCORRECT_QUERY = 80,
     UNKNOWN_DATABASE = 81,
     DATABASE_ALREADY_EXISTS = 82,
+    DIRECTORY_DOESNT_EXIST = 83,
+    DIRECTORY_ALREADY_EXISTS = 84,
+    FORMAT_IS_NOT_SUITABLE_FOR_INPUT = 85,
+    RECEIVED_ERROR_FROM_REMOTE_IO_SERVER = 86,
+    CANNOT_SEEK_THROUGH_FILE = 87,
+    CANNOT_TRUNCATE_FILE = 88,
+    UNKNOWN_COMPRESSION_METHOD = 89,
+    EMPTY_LIST_OF_COLUMNS_PASSED = 90,
+    SIZES_OF_MARKS_FILES_ARE_INCONSISTENT = 91,
+    EMPTY_DATA_PASSED = 92,
+    UNKNOWN_AGGREGATED_DATA_VARIANT = 93,
+    CANNOT_MERGE_DIFFERENT_AGGREGATED_DATA_VARIANTS = 94,
+    CANNOT_READ_FROM_SOCKET = 95,
+    CANNOT_WRITE_TO_SOCKET = 96,
     UNKNOWN_PACKET_FROM_CLIENT = 99,
+    UNKNOWN_PACKET_FROM_SERVER = 100,
     UNEXPECTED_PACKET_FROM_CLIENT = 101,
-    RECEIVED_DATA_FOR_WRONG_QUERY_ID = 103,
+    UNEXPECTED_PACKET_FROM_SERVER = 102,
+    TOO_SMALL_BUFFER_SIZE = 104,
+    FILE_DOESNT_EXIST = 107,
+    NO_DATA_TO_INSERT = 108,
+    CANNOT_BLOCK_SIGNAL = 109,
+    CANNOT_UNBLOCK_SIGNAL = 110,
+    CANNOT_MANIPULATE_SIGSET = 111,
+    CANNOT_WAIT_FOR_SIGNAL = 112,
+    THERE_IS_NO_SESSION = 113,
+    CANNOT_CLOCK_GETTIME = 114,
+    UNKNOWN_SETTING = 115,
+    THERE_IS_NO_DEFAULT_VALUE = 116,
+    INCORRECT_DATA = 117,
     ENGINE_REQUIRED = 119,
+    CANNOT_INSERT_VALUE_OF_DIFFERENT_SIZE_INTO_TUPLE = 120,
+    UNSUPPORTED_JOIN_KEYS = 121,
+    INCOMPATIBLE_COLUMNS = 122,
+    UNKNOWN_TYPE_OF_AST_NODE = 123,
+    INCORRECT_ELEMENT_OF_SET = 124,
+    INCORRECT_RESULT_OF_SCALAR_SUBQUERY = 125,
+    ILLEGAL_INDEX = 127,
+    TOO_LARGE_ARRAY_SIZE = 128,
+    FUNCTION_IS_SPECIAL = 129,
+    CANNOT_READ_ARRAY_FROM_TEXT = 130,
+    TOO_LARGE_STRING_SIZE = 131,
+    AGGREGATE_FUNCTION_DOESNT_ALLOW_PARAMETERS = 133,
+    PARAMETERS_TO_AGGREGATE_FUNCTIONS_MUST_BE_LITERALS = 134,
+    ZERO_ARRAY_OR_TUPLE_INDEX = 135,
+    UNKNOWN_ELEMENT_IN_CONFIG = 137,
+    EXCESSIVE_ELEMENT_IN_CONFIG = 138,
+    NO_ELEMENTS_IN_CONFIG = 139,
+    SAMPLING_NOT_SUPPORTED = 141,
+    NOT_FOUND_NODE = 142,
+    UNKNOWN_OVERFLOW_MODE = 145,
+    UNKNOWN_DIRECTION_OF_SORTING = 152,
+    ILLEGAL_DIVISION = 153,
+    DICTIONARIES_WAS_NOT_LOADED = 156,
+    TOO_MANY_ROWS = 158,
+    TIMEOUT_EXCEEDED = 159,
+    TOO_SLOW = 160,
+    TOO_MANY_COLUMNS = 161,
+    TOO_DEEP_SUBQUERIES = 162,
     READONLY = 164,
+    TOO_MANY_TEMPORARY_COLUMNS = 165,
+    TOO_MANY_TEMPORARY_NON_CONST_COLUMNS = 166,
+    TOO_DEEP_AST = 167,
+    TOO_BIG_AST = 168,
+    BAD_TYPE_OF_FIELD = 169,
+    BAD_GET = 170,
+    CANNOT_CREATE_DIRECTORY = 172,
+    CANNOT_ALLOCATE_MEMORY = 173,
+    CYCLIC_ALIASES = 174,
+    MULTIPLE_EXPRESSIONS_FOR_ALIAS = 179,
+    THERE_IS_NO_PROFILE = 180,
+    ILLEGAL_FINAL = 181,
+    ILLEGAL_PREWHERE = 182,
+    UNEXPECTED_EXPRESSION = 183,
+    ILLEGAL_AGGREGATION = 184,
+    UNSUPPORTED_COLLATION_LOCALE = 186,
+    COLLATION_COMPARISON_FAILED = 187,
+    SIZES_OF_ARRAYS_DONT_MATCH = 190,
+    SET_SIZE_LIMIT_EXCEEDED = 191,
     UNKNOWN_USER = 192,
     WRONG_PASSWORD = 193,
     REQUIRED_PASSWORD = 194,
     IP_ADDRESS_NOT_ALLOWED = 195,
+    UNKNOWN_ADDRESS_PATTERN_TYPE = 196,
+    DNS_ERROR = 198,
+    UNKNOWN_QUOTA = 199,
+    QUOTA_EXCEEDED = 201,
+    TOO_MANY_SIMULTANEOUS_QUERIES = 202,
+    NO_FREE_CONNECTION = 203,
+    CANNOT_FSYNC = 204,
+    ALIAS_REQUIRED = 206,
+    AMBIGUOUS_IDENTIFIER = 207,
+    EMPTY_NESTED_TABLE = 208,
+    SOCKET_TIMEOUT = 209,
+    NETWORK_ERROR = 210,
+    EMPTY_QUERY = 211,
+    UNKNOWN_LOAD_BALANCING = 212,
+    UNKNOWN_TOTALS_MODE = 213,
+    CANNOT_STATVFS = 214,
+    NOT_AN_AGGREGATE = 215,
+    QUERY_WITH_SAME_ID_IS_ALREADY_RUNNING = 216,
+    CLIENT_HAS_CONNECTED_TO_WRONG_PORT = 217,
+    TABLE_IS_DROPPED = 218,
+    DATABASE_NOT_EMPTY = 219,
+    DUPLICATE_INTERSERVER_IO_ENDPOINT = 220,
+    NO_SUCH_INTERSERVER_IO_ENDPOINT = 221,
+    UNEXPECTED_AST_STRUCTURE = 223,
+    REPLICA_IS_ALREADY_ACTIVE = 224,
+    NO_ZOOKEEPER = 225,
+    NO_FILE_IN_DATA_PART = 226,
+    UNEXPECTED_FILE_IN_DATA_PART = 227,
+    BAD_SIZE_OF_FILE_IN_DATA_PART = 228,
+    QUERY_IS_TOO_LARGE = 229,
+    NOT_FOUND_EXPECTED_DATA_PART = 230,
+    TOO_MANY_UNEXPECTED_DATA_PARTS = 231,
+    NO_SUCH_DATA_PART = 232,
+    BAD_DATA_PART_NAME = 233,
+    NO_REPLICA_HAS_PART = 234,
+    DUPLICATE_DATA_PART = 235,
+    ABORTED = 236,
+    NO_REPLICA_NAME_GIVEN = 237,
+    FORMAT_VERSION_TOO_OLD = 238,
+    CANNOT_MUNMAP = 239,
+    CANNOT_MREMAP = 240,
+    MEMORY_LIMIT_EXCEEDED = 241,
+    TABLE_IS_READ_ONLY = 242,
+    NOT_ENOUGH_SPACE = 243,
+    UNEXPECTED_ZOOKEEPER_ERROR = 244,
+    CORRUPTED_DATA = 246,
+    INVALID_PARTITION_VALUE = 248,
+    NO_SUCH_REPLICA = 251,
+    TOO_MANY_PARTS = 252,
+    REPLICA_ALREADY_EXISTS = 253,
+    NO_ACTIVE_REPLICAS = 254,
+    TOO_MANY_RETRIES_TO_FETCH_PARTS = 255,
+    PARTITION_ALREADY_EXISTS = 256,
+    PARTITION_DOESNT_EXIST = 257,
+    UNION_ALL_RESULT_STRUCTURES_MISMATCH = 258,
+    CLIENT_OUTPUT_FORMAT_SPECIFIED = 260,
+    UNKNOWN_BLOCK_INFO_FIELD = 261,
+    BAD_COLLATION = 262,
+    CANNOT_COMPILE_CODE = 263,
+    INCOMPATIBLE_TYPE_OF_JOIN = 264,
+    NO_AVAILABLE_REPLICA = 265,
+    MISMATCH_REPLICAS_DATA_SOURCES = 266,
+    INFINITE_LOOP = 269,
+    CANNOT_COMPRESS = 270,
+    CANNOT_DECOMPRESS = 271,
+    CANNOT_IO_SUBMIT = 272,
+    CANNOT_IO_GETEVENTS = 273,
+    AIO_READ_ERROR = 274,
+    AIO_WRITE_ERROR = 275,
+    INDEX_NOT_USED = 277,
+    ALL_CONNECTION_TRIES_FAILED = 279,
+    NO_AVAILABLE_DATA = 280,
+    DICTIONARY_IS_EMPTY = 281,
+    INCORRECT_INDEX = 282,
+    UNKNOWN_DISTRIBUTED_PRODUCT_MODE = 283,
+    WRONG_GLOBAL_SUBQUERY = 284,
+    TOO_FEW_LIVE_REPLICAS = 285,
+    UNSATISFIED_QUORUM_FOR_PREVIOUS_WRITE = 286,
+    UNKNOWN_FORMAT_VERSION = 287,
+    DISTRIBUTED_IN_JOIN_SUBQUERY_DENIED = 288,
+    REPLICA_IS_NOT_IN_QUORUM = 289,
     LIMIT_EXCEEDED = 290,
+    DATABASE_ACCESS_DENIED = 291,
+    MONGODB_CANNOT_AUTHENTICATE = 293,
+    RECEIVED_EMPTY_DATA = 295,
+    SHARD_HAS_NO_CONNECTIONS = 297,
+    CANNOT_PIPE = 298,
+    CANNOT_FORK = 299,
+    CANNOT_DLSYM = 300,
+    CANNOT_CREATE_CHILD_PROCESS = 301,
+    CHILD_WAS_NOT_EXITED_NORMALLY = 302,
+    CANNOT_SELECT = 303,
+    CANNOT_WAITPID = 304,
+    TABLE_WAS_NOT_DROPPED = 305,
+    TOO_DEEP_RECURSION = 306,
+    TOO_MANY_BYTES = 307,
+    UNEXPECTED_NODE_IN_ZOOKEEPER = 308,
+    FUNCTION_CANNOT_HAVE_PARAMETERS = 309,
+    INVALID_CONFIG_PARAMETER = 318,
+    UNKNOWN_STATUS_OF_INSERT = 319,
+    VALUE_IS_OUT_OF_RANGE_OF_DATA_TYPE = 321,
     UNKNOWN_DATABASE_ENGINE = 336,
+    UNFINISHED = 341,
+    METADATA_MISMATCH = 342,
+    SUPPORT_IS_DISABLED = 344,
+    TABLE_DIFFERS_TOO_MUCH = 345,
+    CANNOT_CONVERT_CHARSET = 346,
+    CANNOT_LOAD_CONFIG = 347,
+    CANNOT_INSERT_NULL_IN_ORDINARY_COLUMN = 349,
+    AMBIGUOUS_COLUMN_NAME = 352,
+    INDEX_OF_POSITIONAL_ARGUMENT_IS_OUT_OF_RANGE = 353,
+    ZLIB_INFLATE_FAILED = 354,
+    ZLIB_DEFLATE_FAILED = 355,
+    INTO_OUTFILE_NOT_ALLOWED = 358,
+    TABLE_SIZE_EXCEEDS_MAX_DROP_SIZE_LIMIT = 359,
+    CANNOT_CREATE_CHARSET_CONVERTER = 360,
+    SEEK_POSITION_OUT_OF_BOUND = 361,
+    CURRENT_WRITE_BUFFER_IS_EXHAUSTED = 362,
+    CANNOT_CREATE_IO_BUFFER = 363,
+    RECEIVED_ERROR_TOO_MANY_REQUESTS = 364,
+    SIZES_OF_NESTED_COLUMNS_ARE_INCONSISTENT = 366,
+    ALL_REPLICAS_ARE_STALE = 369,
+    DATA_TYPE_CANNOT_BE_USED_IN_TABLES = 370,
+    INCONSISTENT_CLUSTER_DEFINITION = 371,
+    SESSION_NOT_FOUND = 372,
+    SESSION_IS_LOCKED = 373,
+    INVALID_SESSION_TIMEOUT = 374,
+    CANNOT_DLOPEN = 375,
+    CANNOT_PARSE_UUID = 376,
+    ILLEGAL_SYNTAX_FOR_DATA_TYPE = 377,
+    DATA_TYPE_CANNOT_HAVE_ARGUMENTS = 378,
+    CANNOT_KILL = 380,
+    HTTP_LENGTH_REQUIRED = 381,
+    CANNOT_LOAD_CATBOOST_MODEL = 382,
+    CANNOT_APPLY_CATBOOST_MODEL = 383,
+    PART_IS_TEMPORARILY_LOCKED = 384,
+    MULTIPLE_STREAMS_REQUIRED = 385,
+    NO_COMMON_TYPE = 386,
+    DICTIONARY_ALREADY_EXISTS = 387,
+    CANNOT_ASSIGN_OPTIMIZE = 388,
+    INSERT_WAS_DEDUPLICATED = 389,
+    CANNOT_GET_CREATE_TABLE_QUERY = 390,
+    EXTERNAL_LIBRARY_ERROR = 391,
+    QUERY_IS_PROHIBITED = 392,
+    THERE_IS_NO_QUERY = 393,
+    QUERY_WAS_CANCELLED = 394,
+    FUNCTION_THROW_IF_VALUE_IS_NON_ZERO = 395,
+    TOO_MANY_ROWS_OR_BYTES = 396,
+    QUERY_IS_NOT_SUPPORTED_IN_MATERIALIZED_VIEW = 397,
+    UNKNOWN_MUTATION_COMMAND = 398,
+    FORMAT_IS_NOT_SUITABLE_FOR_OUTPUT = 399,
+    CANNOT_STAT = 400,
+    FEATURE_IS_NOT_ENABLED_AT_BUILD_TIME = 401,
+    CANNOT_IOSETUP = 402,
+    INVALID_JOIN_ON_EXPRESSION = 403,
+    BAD_ODBC_CONNECTION_STRING = 404,
+    TOP_AND_LIMIT_TOGETHER = 406,
+    DECIMAL_OVERFLOW = 407,
+    BAD_REQUEST_PARAMETER = 408,
+    EXTERNAL_SERVER_IS_NOT_RESPONDING = 410,
+    PTHREAD_ERROR = 411,
+    NETLINK_ERROR = 412,
+    CANNOT_SET_SIGNAL_HANDLER = 413,
+    ALL_REPLICAS_LOST = 415,
+    REPLICA_STATUS_CHANGED = 416,
+    EXPECTED_ALL_OR_ANY = 417,
+    UNKNOWN_JOIN = 418,
+    MULTIPLE_ASSIGNMENTS_TO_COLUMN = 419,
+    CANNOT_UPDATE_COLUMN = 420,
+    CANNOT_ADD_DIFFERENT_AGGREGATE_STATES = 421,
+    UNSUPPORTED_URI_SCHEME = 422,
+    CANNOT_GETTIMEOFDAY = 423,
+    CANNOT_LINK = 424,
+    SYSTEM_ERROR = 425,
+    CANNOT_COMPILE_REGEXP = 427,
+    FAILED_TO_GETPWUID = 429,
+    MISMATCHING_USERS_FOR_PROCESS_AND_DATA = 430,
+    ILLEGAL_SYNTAX_FOR_CODEC_TYPE = 431,
+    UNKNOWN_CODEC = 432,
+    ILLEGAL_CODEC_PARAMETER = 433,
+    CANNOT_PARSE_PROTOBUF_SCHEMA = 434,
+    NO_COLUMN_SERIALIZED_TO_REQUIRED_PROTOBUF_FIELD = 435,
+    PROTOBUF_BAD_CAST = 436,
+    PROTOBUF_FIELD_NOT_REPEATED = 437,
+    DATA_TYPE_CANNOT_BE_PROMOTED = 438,
+    CANNOT_SCHEDULE_TASK = 439,
+    INVALID_LIMIT_EXPRESSION = 440,
+    CANNOT_PARSE_DOMAIN_VALUE_FROM_STRING = 441,
+    BAD_DATABASE_FOR_TEMPORARY_TABLE = 442,
+    NO_COLUMNS_SERIALIZED_TO_PROTOBUF_FIELDS = 443,
+    UNKNOWN_PROTOBUF_FORMAT = 444,
+    CANNOT_MPROTECT = 445,
+    FUNCTION_NOT_ALLOWED = 446,
+    HYPERSCAN_CANNOT_SCAN_TEXT = 447,
+    BROTLI_READ_FAILED = 448,
+    BROTLI_WRITE_FAILED = 449,
+    BAD_TTL_EXPRESSION = 450,
+    BAD_TTL_FILE = 451,
+    SETTING_CONSTRAINT_VIOLATION = 452,
+    MYSQL_CLIENT_INSUFFICIENT_CAPABILITIES = 453,
+    OPENSSL_ERROR = 454,
+    SUSPICIOUS_TYPE_FOR_LOW_CARDINALITY = 455,
+    UNKNOWN_QUERY_PARAMETER = 456,
+    BAD_QUERY_PARAMETER = 457,
+    CANNOT_UNLINK = 458,
+    CANNOT_SET_THREAD_PRIORITY = 459,
+    CANNOT_CREATE_TIMER = 460,
+    CANNOT_SET_TIMER_PERIOD = 461,
+    CANNOT_FCNTL = 463,
+    CANNOT_PARSE_ELF = 464,
+    CANNOT_PARSE_DWARF = 465,
+    INSECURE_PATH = 466,
+    CANNOT_PARSE_BOOL = 467,
+    CANNOT_PTHREAD_ATTR = 468,
+    VIOLATED_CONSTRAINT = 469,
+    QUERY_IS_NOT_SUPPORTED_IN_LIVE_VIEW = 470,
+    INVALID_SETTING_VALUE = 471,
+    READONLY_SETTING = 472,
+    DEADLOCK_AVOIDED = 473,
+    INVALID_TEMPLATE_FORMAT = 474,
+    INVALID_WITH_FILL_EXPRESSION = 475,
+    WITH_TIES_WITHOUT_ORDER_BY = 476,
+    INVALID_USAGE_OF_INPUT = 477,
+    UNKNOWN_POLICY = 478,
+    UNKNOWN_DISK = 479,
+    UNKNOWN_PROTOCOL = 480,
+    PATH_ACCESS_DENIED = 481,
+    DICTIONARY_ACCESS_DENIED = 482,
+    TOO_MANY_REDIRECTS = 483,
+    INTERNAL_REDIS_ERROR = 484,
+    CANNOT_GET_CREATE_DICTIONARY_QUERY = 487,
+    INCORRECT_DICTIONARY_DEFINITION = 489,
+    CANNOT_FORMAT_DATETIME = 490,
+    UNACCEPTABLE_URL = 491,
+    ACCESS_ENTITY_NOT_FOUND = 492,
+    ACCESS_ENTITY_ALREADY_EXISTS = 493,
+    ACCESS_STORAGE_READONLY = 495,
+    QUOTA_REQUIRES_CLIENT_KEY = 496,
+    ACCESS_DENIED = 497,
+    LIMIT_BY_WITH_TIES_IS_NOT_SUPPORTED = 498,
+    S3_ERROR = 499,
+    AZURE_BLOB_STORAGE_ERROR = 500,
+    CANNOT_CREATE_DATABASE = 501,
+    CANNOT_SIGQUEUE = 502,
+    AGGREGATE_FUNCTION_THROW = 503,
+    FILE_ALREADY_EXISTS = 504,
+    UNABLE_TO_SKIP_UNUSED_SHARDS = 507,
+    UNKNOWN_ACCESS_TYPE = 508,
+    INVALID_GRANT = 509,
+    CACHE_DICTIONARY_UPDATE_FAIL = 510,
+    UNKNOWN_ROLE = 511,
+    SET_NON_GRANTED_ROLE = 512,
+    UNKNOWN_PART_TYPE = 513,
+    ACCESS_STORAGE_FOR_INSERTION_NOT_FOUND = 514,
+    INCORRECT_ACCESS_ENTITY_DEFINITION = 515,
+    AUTHENTICATION_FAILED = 516,
+    CANNOT_ASSIGN_ALTER = 517,
+    CANNOT_COMMIT_OFFSET = 518,
+    NO_REMOTE_SHARD_AVAILABLE = 519,
+    CANNOT_DETACH_DICTIONARY_AS_TABLE = 520,
+    ATOMIC_RENAME_FAIL = 521,
+    UNKNOWN_ROW_POLICY = 523,
+    ALTER_OF_COLUMN_IS_FORBIDDEN = 524,
+    INCORRECT_DISK_INDEX = 525,
+    NO_SUITABLE_FUNCTION_IMPLEMENTATION = 527,
+    CASSANDRA_INTERNAL_ERROR = 528,
+    NOT_A_LEADER = 529,
+    CANNOT_CONNECT_RABBITMQ = 530,
+    CANNOT_FSTAT = 531,
+    LDAP_ERROR = 532,
+    UNKNOWN_RAID_TYPE = 535,
+    CANNOT_RESTORE_FROM_FIELD_DUMP = 536,
+    ILLEGAL_MYSQL_VARIABLE = 537,
+    MYSQL_SYNTAX_ERROR = 538,
+    CANNOT_BIND_RABBITMQ_EXCHANGE = 539,
+    CANNOT_DECLARE_RABBITMQ_EXCHANGE = 540,
+    CANNOT_CREATE_RABBITMQ_QUEUE_BINDING = 541,
+    CANNOT_REMOVE_RABBITMQ_EXCHANGE = 542,
+    UNKNOWN_MYSQL_DATATYPES_SUPPORT_LEVEL = 543,
+    ROW_AND_ROWS_TOGETHER = 544,
+    FIRST_AND_NEXT_TOGETHER = 545,
+    NO_ROW_DELIMITER = 546,
+    INVALID_RAID_TYPE = 547,
+    UNKNOWN_VOLUME = 548,
+    DATA_TYPE_CANNOT_BE_USED_IN_KEY = 549,
+    UNRECOGNIZED_ARGUMENTS = 552,
+    LZMA_STREAM_ENCODER_FAILED = 553,
+    LZMA_STREAM_DECODER_FAILED = 554,
+    ROCKSDB_ERROR = 555,
+    SYNC_MYSQL_USER_ACCESS_ERROR = 556,
+    UNKNOWN_UNION = 557,
+    EXPECTED_ALL_OR_DISTINCT = 558,
+    INVALID_GRPC_QUERY_INFO = 559,
+    ZSTD_ENCODER_FAILED = 560,
+    ZSTD_DECODER_FAILED = 561,
+    TLD_LIST_NOT_FOUND = 562,
+    CANNOT_READ_MAP_FROM_TEXT = 563,
+    INTERSERVER_SCHEME_DOESNT_MATCH = 564,
+    TOO_MANY_PARTITIONS = 565,
+    CANNOT_RMDIR = 566,
+    DUPLICATED_PART_UUIDS = 567,
+    RAFT_ERROR = 568,
+    MULTIPLE_COLUMNS_SERIALIZED_TO_SAME_PROTOBUF_FIELD = 569,
+    DATA_TYPE_INCOMPATIBLE_WITH_PROTOBUF_FIELD = 570,
+    DATABASE_REPLICATION_FAILED = 571,
+    TOO_MANY_QUERY_PLAN_OPTIMIZATIONS = 572,
+    EPOLL_ERROR = 573,
+    DISTRIBUTED_TOO_MANY_PENDING_BYTES = 574,
+    UNKNOWN_SNAPSHOT = 575,
+    KERBEROS_ERROR = 576,
+    INVALID_SHARD_ID = 577,
+    INVALID_FORMAT_INSERT_QUERY_WITH_DATA = 578,
+    INCORRECT_PART_TYPE = 579,
+    CANNOT_SET_ROUNDING_MODE = 580,
+    TOO_LARGE_DISTRIBUTED_DEPTH = 581,
+    NO_SUCH_PROJECTION_IN_TABLE = 582,
+    ILLEGAL_PROJECTION = 583,
+    PROJECTION_NOT_USED = 584,
+    CANNOT_PARSE_YAML = 585,
+    CANNOT_CREATE_FILE = 586,
+    CONCURRENT_ACCESS_NOT_SUPPORTED = 587,
+    DISTRIBUTED_BROKEN_BATCH_INFO = 588,
+    DISTRIBUTED_BROKEN_BATCH_FILES = 589,
+    CANNOT_SYSCONF = 590,
+    SQLITE_ENGINE_ERROR = 591,
+    DATA_ENCRYPTION_ERROR = 592,
+    ZERO_COPY_REPLICATION_ERROR = 593,
+    BZIP2_STREAM_DECODER_FAILED = 594,
+    BZIP2_STREAM_ENCODER_FAILED = 595,
+    INTERSECT_OR_EXCEPT_RESULT_STRUCTURES_MISMATCH = 596,
+    NO_SUCH_ERROR_CODE = 597,
+    BACKUP_ALREADY_EXISTS = 598,
+    BACKUP_NOT_FOUND = 599,
+    BACKUP_VERSION_NOT_SUPPORTED = 600,
+    BACKUP_DAMAGED = 601,
+    NO_BASE_BACKUP = 602,
+    WRONG_BASE_BACKUP = 603,
+    BACKUP_ENTRY_ALREADY_EXISTS = 604,
+    BACKUP_ENTRY_NOT_FOUND = 605,
+    BACKUP_IS_EMPTY = 606,
+    CANNOT_RESTORE_DATABASE = 607,
+    CANNOT_RESTORE_TABLE = 608,
+    FUNCTION_ALREADY_EXISTS = 609,
+    CANNOT_DROP_FUNCTION = 610,
+    CANNOT_CREATE_RECURSIVE_FUNCTION = 611,
+    POSTGRESQL_CONNECTION_FAILURE = 614,
+    CANNOT_ADVISE = 615,
+    UNKNOWN_READ_METHOD = 616,
+    LZ4_ENCODER_FAILED = 617,
+    LZ4_DECODER_FAILED = 618,
+    POSTGRESQL_REPLICATION_INTERNAL_ERROR = 619,
+    QUERY_NOT_ALLOWED = 620,
+    CANNOT_NORMALIZE_STRING = 621,
+    CANNOT_PARSE_CAPN_PROTO_SCHEMA = 622,
+    CAPN_PROTO_BAD_CAST = 623,
+    BAD_FILE_TYPE = 624,
+    IO_SETUP_ERROR = 625,
+    CANNOT_SKIP_UNKNOWN_FIELD = 626,
+    BACKUP_ENGINE_NOT_FOUND = 627,
+    OFFSET_FETCH_WITHOUT_ORDER_BY = 628,
+    HTTP_RANGE_NOT_SATISFIABLE = 629,
+    HAVE_DEPENDENT_OBJECTS = 630,
+    UNKNOWN_FILE_SIZE = 631,
+    UNEXPECTED_DATA_AFTER_PARSED_VALUE = 632,
+    QUERY_IS_NOT_SUPPORTED_IN_WINDOW_VIEW = 633,
+    MONGODB_ERROR = 634,
+    CANNOT_POLL = 635,
+    CANNOT_EXTRACT_TABLE_STRUCTURE = 636,
+    INVALID_TABLE_OVERRIDE = 637,
+    SNAPPY_UNCOMPRESS_FAILED = 638,
+    SNAPPY_COMPRESS_FAILED = 639,
+    NO_HIVEMETASTORE = 640,
+    CANNOT_APPEND_TO_FILE = 641,
+    CANNOT_PACK_ARCHIVE = 642,
+    CANNOT_UNPACK_ARCHIVE = 643,
+    NUMBER_OF_DIMENSIONS_MISMATCHED = 645,
+    CANNOT_BACKUP_TABLE = 647,
+    WRONG_DDL_RENAMING_SETTINGS = 648,
+    INVALID_TRANSACTION = 649,
+    SERIALIZATION_ERROR = 650,
+    CAPN_PROTO_BAD_TYPE = 651,
+    ONLY_NULLS_WHILE_READING_SCHEMA = 652,
+    CANNOT_PARSE_BACKUP_SETTINGS = 653,
+    WRONG_BACKUP_SETTINGS = 654,
+    FAILED_TO_SYNC_BACKUP_OR_RESTORE = 655,
+    UNKNOWN_STATUS_OF_TRANSACTION = 659,
+    HDFS_ERROR = 660,
+    CANNOT_SEND_SIGNAL = 661,
+    FS_METADATA_ERROR = 662,
+    INCONSISTENT_METADATA_FOR_BACKUP = 663,
+    ACCESS_STORAGE_DOESNT_ALLOW_BACKUP = 664,
+    CANNOT_CONNECT_NATS = 665,
+    NOT_INITIALIZED = 667,
+    INVALID_STATE = 668,
+    NAMED_COLLECTION_DOESNT_EXIST = 669,
+    NAMED_COLLECTION_ALREADY_EXISTS = 670,
+    NAMED_COLLECTION_IS_IMMUTABLE = 671,
+    INVALID_SCHEDULER_NODE = 672,
+    RESOURCE_ACCESS_DENIED = 673,
+    RESOURCE_NOT_FOUND = 674,
+    CANNOT_PARSE_IPV4 = 675,
+    CANNOT_PARSE_IPV6 = 676,
+    THREAD_WAS_CANCELED = 677,
+    IO_URING_INIT_FAILED = 678,
+    IO_URING_SUBMIT_ERROR = 679,
+    MIXED_ACCESS_PARAMETER_TYPES = 690,
+    UNKNOWN_ELEMENT_OF_ENUM = 691,
+    TOO_MANY_MUTATIONS = 692,
+    AWS_ERROR = 693,
+    ASYNC_LOAD_CYCLE = 694,
+    ASYNC_LOAD_FAILED = 695,
+    ASYNC_LOAD_CANCELED = 696,
+    CANNOT_RESTORE_TO_NONENCRYPTED_DISK = 697,
+    INVALID_REDIS_STORAGE_TYPE = 698,
+    INVALID_REDIS_TABLE_STRUCTURE = 699,
+    USER_SESSION_LIMIT_EXCEEDED = 700,
+    CLUSTER_DOESNT_EXIST = 701,
+    CLIENT_INFO_DOES_NOT_MATCH = 702,
+    INVALID_IDENTIFIER = 703,
+    QUERY_CACHE_USED_WITH_NONDETERMINISTIC_FUNCTIONS = 704,
+    // TABLE_NOT_EMPTY = 705,
+    // LIBSSH_ERROR = 706,
+    // GCP_ERROR = 707,
+    // ILLEGAL_STATISTIC = 708,
+    // CANNOT_GET_REPLICATED_DATABASE_SNAPSHOT = 709,
+    // FAULT_INJECTED = 710,
+    // FILECACHE_ACCESS_DENIED = 711,
+    // TOO_MANY_MATERIALIZED_VIEWS = 712,
+    // UNEXPECTED_CLUSTER = 714,
+    // CANNOT_DETECT_FORMAT = 715,
+    // CANNOT_FORGET_PARTITION = 716,
+    KEEPER_EXCEPTION = 999,
+    POCO_EXCEPTION = 1000,
+    STD_EXCEPTION = 1001,
     UNKNOWN_EXCEPTION = 1002,
 };
 
diff --git a/lib/clickhouse-cpp/clickhouse/exceptions.h b/lib/clickhouse-cpp/clickhouse/exceptions.h
index 44f400f..69cc87d 100644
--- a/lib/clickhouse-cpp/clickhouse/exceptions.h
+++ b/lib/clickhouse-cpp/clickhouse/exceptions.h
@@ -1,16 +1,48 @@
 #pragma once
 
-#include "query.h"
+#include "server_exception.h"
 
 #include <stdexcept>
 
 namespace clickhouse {
 
-class ServerException : public std::runtime_error {
+class Error : public std::runtime_error {
+    using std::runtime_error::runtime_error;
+};
+
+// Caused by any user-related code, like invalid column types or arguments passed to any method.
+class ValidationError : public Error {
+    using Error::Error;
+};
+
+// Buffers+IO errors, failure to serialize/deserialize, checksum mismatches, etc.
+class ProtocolError : public Error {
+    using Error::Error;
+};
+
+class UnimplementedError : public Error {
+    using Error::Error;
+};
+
+// Internal validation error.
+class AssertionError : public Error {
+    using Error::Error;
+};
+
+class OpenSSLError : public Error {
+    using Error::Error;
+};
+
+class CompressionError : public Error {
+    using Error::Error;
+};
+
+// Exception received from server.
+class ServerException : public Error {
 public:
-    ServerException(std::unique_ptr<Exception> e)
-        : runtime_error(std::string())
-        , exception_(std::move(e))
+    ServerException(std::shared_ptr<Exception> e)
+        : Error(std::string())
+        , exception_(e)
     {
     }
 
@@ -27,7 +59,8 @@ class ServerException : public std::runtime_error {
     }
 
 private:
-    std::unique_ptr<Exception> exception_;
+    std::shared_ptr<Exception> exception_;
 };
+using ServerError = ServerException;
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/protocol.h b/lib/clickhouse-cpp/clickhouse/protocol.h
index 46ae7ce..8d36193 100644
--- a/lib/clickhouse-cpp/clickhouse/protocol.h
+++ b/lib/clickhouse-cpp/clickhouse/protocol.h
@@ -2,36 +2,43 @@
 
 namespace clickhouse {
 
-    /// То, что передаёт сервер.
+    /// Types of packets received from server
     namespace ServerCodes {
         enum {
-            Hello       = 0,    /// Имя, версия, ревизия.
-            Data        = 1,    /// Блок данных со сжатием или без.
-            Exception   = 2,    /// Исключение во время обработки запроса.
-            Progress    = 3,    /// Прогресс выполнения запроса: строк считано, байт считано.
-            Pong        = 4,    /// Ответ на Ping.
-            EndOfStream = 5,    /// Все пакеты были переданы.
-            ProfileInfo = 6,    /// Пакет с профайлинговой информацией.
-            Totals      = 7,    /// Блок данных с тотальными значениями, со сжатием или без.
-            Extremes    = 8,    /// Блок данных с минимумами и максимумами, аналогично.
+            Hello                = 0,    /// Name, version, revision.
+            Data                 = 1,    /// `Block` of data, may be compressed.
+            Exception            = 2,    /// Exception that occurred on server side during query execution.
+            Progress             = 3,    /// Query execcution progress: rows and bytes read.
+            Pong                 = 4,    /// response to Ping sent by client.
+            EndOfStream          = 5,    /// All packets were sent.
+            ProfileInfo          = 6,    /// Profiling data
+            Totals               = 7,    /// Block of totals, may be compressed.
+            Extremes             = 8,    /// Block of mins and maxs, may be compressed.
+            TablesStatusResponse = 9,    /// Response to TableStatus.
+            Log                  = 10,   /// Query execution log.
+            TableColumns         = 11,   /// Columns' description for default values calculation
+            PartUUIDs            = 12,   /// List of unique parts ids.
+            ReadTaskRequest      = 13,   /// String (UUID) describes a request for which next task is needed
+                                         /// This is such an inverted logic, where server sends requests
+                                         /// And client returns back response
+            ProfileEvents        = 14,   /// Packet with profile events from server.
         };
     }
 
-    /// То, что передаёт клиент.
+    /// Types of packets sent by client.
     namespace ClientCodes {
         enum {
-            Hello       = 0,    /// Имя, версия, ревизия, БД по-умолчанию.
-            Query       = 1,    /** Идентификатор запроса, настройки на отдельный запрос,
-                                  * информация, до какой стадии исполнять запрос,
-                                  * использовать ли сжатие, текст запроса (без данных для INSERT-а).
+            Hello       = 0,    /// Name, version, default database name.
+            Query       = 1,    /** Query id, query settings, query processing stage,
+                                  * compression status, and query text (no INSERT data).
                                   */
-            Data        = 2,    /// Блок данных со сжатием или без.
-            Cancel      = 3,    /// Отменить выполнение запроса.
-            Ping        = 4,    /// Проверка живости соединения с сервером.
+            Data        = 2,    /// Data `Block` (e.g. INSERT data), may be compressed.
+            Cancel      = 3,    /// Cancel query.
+            Ping        = 4,    /// Check server connection.
         };
     }
 
-    /// Использовать ли сжатие.
+    /// Should we compress `Block`s of data
     namespace CompressionState {
         enum {
             Disable     = 0,
diff --git a/lib/clickhouse-cpp/clickhouse/query.cpp b/lib/clickhouse-cpp/clickhouse/query.cpp
index 89a9037..3986064 100644
--- a/lib/clickhouse-cpp/clickhouse/query.cpp
+++ b/lib/clickhouse-cpp/clickhouse/query.cpp
@@ -2,16 +2,20 @@
 
 namespace clickhouse {
 
+const std::string Query::default_query_id = {};
+
 Query::Query()
 { }
 
-Query::Query(const char* query)
+Query::Query(const char* query, const char* query_id)
     : query_(query)
+    , query_id_(query_id ? std::string(query_id): default_query_id)
 {
 }
 
-Query::Query(const std::string& query)
+Query::Query(const std::string& query, const std::string& query_id)
     : query_(query)
+    , query_id_(query_id)
 {
 }
 
diff --git a/lib/clickhouse-cpp/clickhouse/query.h b/lib/clickhouse-cpp/clickhouse/query.h
index c9d7e68..500cdae 100644
--- a/lib/clickhouse-cpp/clickhouse/query.h
+++ b/lib/clickhouse-cpp/clickhouse/query.h
@@ -1,47 +1,33 @@
 #pragma once
 
 #include "block.h"
+#include "server_exception.h"
+
+#include "base/open_telemetry.h"
 
 #include <cstdint>
 #include <functional>
 #include <memory>
+#include <optional>
 #include <string>
+#include <unordered_map>
 
 namespace clickhouse {
 
-/**
- * Settings of individual query.
- */
-struct QuerySettings {
-    /// Максимальное количество потоков выполнения запроса. По-умолчанию - определять автоматически.
-    int max_threads = 0;
-    /// Считать минимумы и максимумы столбцов результата.
-    bool extremes = false;
-    /// Тихо пропускать недоступные шарды.
-    bool skip_unavailable_shards = false;
-    /// Write statistics about read rows, bytes, time elapsed, etc.
-    bool output_format_write_statistics = true;
-    /// Use client timezone for interpreting DateTime string values, instead of adopting server timezone.
-    bool use_client_time_zone = false;
-
-    // connect_timeout
-    // max_block_size
-    // distributed_group_by_no_merge = false
-    // strict_insert_defaults = 0
-    // network_compression_method = LZ4
-    // priority = 0
-};
-
-
-struct Exception {
-    int code = 0;
-    std::string name;
-    std::string display_text;
-    std::string stack_trace;
-    /// Pointer to nested exception.
-    std::unique_ptr<Exception> nested;
+struct QuerySettingsField {
+    enum Flags : uint64_t
+    {
+        IMPORTANT = 0x01,
+        CUSTOM = 0x02,
+        OBSOLETE = 0x04,
+    };
+    std::string value;
+    uint64_t flags{0};
 };
 
+using QuerySettings = std::unordered_map<std::string, QuerySettingsField>;
+using QueryParamValue = std::optional<std::string>;
+using QueryParams = std::unordered_map<std::string, QueryParamValue>;
 
 struct Profile {
     uint64_t rows = 0;
@@ -57,6 +43,8 @@ struct Progress {
     uint64_t rows = 0;
     uint64_t bytes = 0;
     uint64_t total_rows = 0;
+    uint64_t written_rows = 0;
+    uint64_t written_bytes = 0;
 };
 
 
@@ -75,6 +63,15 @@ class QueryEvents {
 
     virtual void OnProgress(const Progress& progress) = 0;
 
+    /** Handle query execution logs provided by server.
+     *  Amount of logs regulated by `send_logs_level` setting.
+     *  By-default only `fatal` log events are sent to the client side.
+     */
+    virtual void OnServerLog(const Block& block) = 0;
+
+    /// Handle query execution profile events.
+    virtual void OnProfileEvents(const Block& block) = 0;
+
     virtual void OnFinish() = 0;
 };
 
@@ -83,44 +80,107 @@ using ExceptionCallback        = std::function<void(const Exception& e)>;
 using ProgressCallback         = std::function<void(const Progress& progress)>;
 using SelectCallback           = std::function<void(const Block& block)>;
 using SelectCancelableCallback = std::function<bool(const Block& block)>;
+using SelectServerLogCallback  = std::function<bool(const Block& block)>;
+using ProfileEventsCallback    = std::function<bool(const Block& block)>;
+using ProfileCallback          = std::function<void(const Profile& profile)>;
 
 
 class Query : public QueryEvents {
 public:
      Query();
-     Query(const char* query);
-     Query(const std::string& query);
-    ~Query();
+     Query(const char* query, const char* query_id = nullptr);
+     Query(const std::string& query, const std::string& query_id = default_query_id);
+    ~Query() override;
 
     ///
-    inline std::string GetText() const {
+    inline const std::string& GetText() const {
         return query_;
     }
 
+    inline const std::string& GetQueryID() const {
+        return query_id_;
+    }
+
+    inline const QuerySettings& GetQuerySettings() const {
+        return query_settings_;
+    }
+
+    /// Set per query settings
+    inline Query& SetQuerySettings(QuerySettings query_settings) {
+        query_settings_ = std::move(query_settings);
+        return *this;
+    }
+
+    /// Set per query setting
+    inline Query& SetSetting(const std::string& key, const QuerySettingsField& value) {
+        query_settings_[key] = value;
+        return *this;
+    }
+
+    inline const QueryParams& GetParams() const { return query_params_; }
+
+    inline Query& SetParams(QueryParams query_params) {
+        query_params_ = std::move(query_params);
+        return *this;
+    }
+
+    inline Query& SetParam(const std::string& name, const QueryParamValue& value) {
+        query_params_[name] = value;
+        return *this;
+    }
+
+    inline const std::optional<open_telemetry::TracingContext>& GetTracingContext() const {
+        return tracing_context_;
+    }
+
+    /// Set tracing context for open telemetry signals
+    inline Query& SetTracingContext(open_telemetry::TracingContext tracing_context) {
+        tracing_context_ = std::move(tracing_context);
+        return *this;
+    }
+
     /// Set handler for receiving result data.
     inline Query& OnData(SelectCallback cb) {
-        select_cb_ = cb;
+        select_cb_ = std::move(cb);
         return *this;
     }
 
     inline Query& OnDataCancelable(SelectCancelableCallback cb) {
-        select_cancelable_cb_ = cb;
+        select_cancelable_cb_ = std::move(cb);
         return *this;
     }
 
     /// Set handler for receiving server's exception.
     inline Query& OnException(ExceptionCallback cb) {
-        exception_cb_ = cb;
+        exception_cb_ = std::move(cb);
         return *this;
     }
 
-
-    /// Set handler for receiving a progress of query exceution.
+    /// Set handler for receiving a progress of query execution.
     inline Query& OnProgress(ProgressCallback cb) {
-        progress_cb_ = cb;
+        progress_cb_ = std::move(cb);
+        return *this;
+    }
+
+    /// Set handler for receiving a server log of query exceution.
+    inline Query& OnServerLog(SelectServerLogCallback cb) {
+        select_server_log_cb_ = std::move(cb);
+        return *this;
+    }
+
+    /// Set handler for receiving profile events.
+    inline Query& OnProfileEvents(ProfileEventsCallback cb) {
+        profile_events_callback_cb_ = std::move(cb);
+        return *this;
+    }
+
+    inline Query& OnProfile(ProfileCallback cb) {
+        profile_callback_cb_ = std::move(cb);
         return *this;
     }
 
+    static const std::string default_query_id;
+
 private:
     void OnData(const Block& block) override {
         if (select_cb_) {
@@ -143,7 +203,8 @@ class Query : public QueryEvents {
     }
 
     void OnProfile(const Profile& profile) override {
-        (void)profile;
+        if (profile_callback_cb_)
+            profile_callback_cb_(profile);
     }
 
     void OnProgress(const Progress& progress) override {
@@ -152,15 +213,34 @@ class Query : public QueryEvents {
         }
     }
 
+    void OnServerLog(const Block& block) override {
+        if (select_server_log_cb_) {
+            select_server_log_cb_(block);
+        }
+    }
+
+    void OnProfileEvents(const Block& block) override {
+        if (profile_events_callback_cb_) {
+            profile_events_callback_cb_(block);
+        }
+    }
+
     void OnFinish() override {
     }
 
 private:
-    std::string query_;
+    const std::string query_;
+    const std::string query_id_;
+    std::optional<open_telemetry::TracingContext> tracing_context_;
+    QuerySettings query_settings_;
+    QueryParams query_params_;
     ExceptionCallback exception_cb_;
     ProgressCallback progress_cb_;
     SelectCallback select_cb_;
     SelectCancelableCallback select_cancelable_cb_;
+    SelectServerLogCallback select_server_log_cb_;
+    ProfileEventsCallback profile_events_callback_cb_;
+    ProfileCallback profile_callback_cb_;
 };
 
 }
diff --git a/lib/clickhouse-cpp/clickhouse/server_exception.h b/lib/clickhouse-cpp/clickhouse/server_exception.h
new file mode 100644
index 0000000..dcc97c5
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/server_exception.h
@@ -0,0 +1,16 @@
+#pragma once
+
+#include <string>
+#include <memory>
+
+namespace clickhouse {
+struct Exception {
+    int code = 0;
+    std::string name;
+    std::string display_text;
+    std::string stack_trace;
+    /// Pointer to nested exception.
+    std::unique_ptr<Exception> nested;
+};
+
+}
diff --git a/lib/clickhouse-cpp/clickhouse/types/type_parser.cpp b/lib/clickhouse-cpp/clickhouse/types/type_parser.cpp
index 0b1fa20..d488a07 100644
--- a/lib/clickhouse-cpp/clickhouse/types/type_parser.cpp
+++ b/lib/clickhouse-cpp/clickhouse/types/type_parser.cpp
@@ -1,15 +1,38 @@
 #include "type_parser.h"
-#include "../base/string_utils.h"
 
+#include "clickhouse/exceptions.h"
+#include "clickhouse/base/platform.h" // for _win_
+
+#include <algorithm>
+#include <cmath>
+#include <map>
+#include <mutex>
 #include <unordered_map>
 
+#if defined _win_
+#include <string.h>
+#else
+#include <strings.h>
+#endif
+
+
 namespace clickhouse {
 
+bool TypeAst::operator==(const TypeAst & other) const {
+    return meta == other.meta
+        && code == other.code
+        && name == other.name
+        && value == other.value
+        && std::equal(elements.begin(), elements.end(), other.elements.begin(), other.elements.end());
+}
+
 static const std::unordered_map<std::string, Type::Code> kTypeCode = {
+    { "Void",        Type::Void },
     { "Int8",        Type::Int8 },
     { "Int16",       Type::Int16 },
     { "Int32",       Type::Int32 },
     { "Int64",       Type::Int64 },
+    { "Bool",        Type::UInt8 },
     { "UInt8",       Type::UInt8 },
     { "UInt16",      Type::UInt16 },
     { "UInt32",      Type::UInt32 },
@@ -19,20 +42,52 @@ static const std::unordered_map<std::string, Type::Code> kTypeCode = {
     { "String",      Type::String },
     { "FixedString", Type::FixedString },
     { "DateTime",    Type::DateTime },
+    { "DateTime64",  Type::DateTime64 },
     { "Date",        Type::Date },
+    { "Date32",      Type::Date32 },
     { "Array",       Type::Array },
     { "Nullable",    Type::Nullable },
     { "Tuple",       Type::Tuple },
     { "Enum8",       Type::Enum8 },
     { "Enum16",      Type::Enum16 },
     { "UUID",        Type::UUID },
+    { "IPv4",        Type::IPv4 },
+    { "IPv6",        Type::IPv6 },
+    { "Int128",      Type::Int128 },
+    { "UInt128",     Type::UInt128 },
+    { "Decimal",     Type::Decimal },
+    { "Decimal32",   Type::Decimal32 },
+    { "Decimal64",   Type::Decimal64 },
+    { "Decimal128",  Type::Decimal128 },
+    { "LowCardinality", Type::LowCardinality },
+    { "Map",         Type::Map },
+    { "Point",       Type::Point },
+    { "Ring",        Type::Ring },
+    { "Polygon",     Type::Polygon },
+    { "MultiPolygon", Type::MultiPolygon },
+    { "Time",        Type::Time },
+    { "Time64",      Type::Time64 },
 };
 
+template <typename L, typename R>
+inline int CompateStringsCaseInsensitive(const L& left, const R& right) {
+    int64_t size_diff = left.size() - right.size();
+    if (size_diff != 0)
+        return size_diff > 0 ? 1 : -1;
+
+#if defined _win_
+    return _strnicmp(left.data(), right.data(), left.size());
+#else
+    return strncasecmp(left.data(), right.data(), left.size());
+#endif
+}
+
 static Type::Code GetTypeCode(const std::string& name) {
     auto it = kTypeCode.find(name);
     if (it != kTypeCode.end()) {
         return it->second;
     }
+
     return Type::Void;
 }
 
@@ -57,9 +112,32 @@ static TypeAst::Meta GetTypeMeta(const StringView& name) {
         return TypeAst::Enum;
     }
 
+    if (name == "LowCardinality") {
+        return TypeAst::LowCardinality;
+    }
+
+    if (name == "SimpleAggregateFunction") {
+        return TypeAst::SimpleAggregateFunction;
+    }
+
+    if (name == "Map") {
+        return TypeAst::Map;
+    }
+
     return TypeAst::Terminal;
 }
 
+bool ValidateAST(const TypeAst& ast) {
+    // Void terminal that is not actually "void" produced when unknown type is encountered.
+    if (ast.meta == TypeAst::Terminal
+            && ast.code == Type::Void
+            && CompateStringsCaseInsensitive(ast.name, std::string_view("void")) != 0)
+        //throw UnimplementedError("Unsupported type: " + ast.name);
+        return false;
+
+    return true;
+}
+
 
 TypeParser::TypeParser(const StringView& name)
     : cur_(name.data())
@@ -74,10 +152,20 @@ bool TypeParser::Parse(TypeAst* type) {
     type_ = type;
     open_elements_.push(type_);
 
+    size_t processed_tokens = 0;
     do {
-        const Token& token = NextToken();
-
+        const Token & token = NextToken();
         switch (token.type) {
+            case Token::QuotedString:
+            {
+                type_->meta = TypeAst::Terminal;
+                if (token.value.length() < 1)
+                    type_->value_string = {};
+                else
+                    type_->value_string = token.value.substr(1, token.value.length() - 2).to_string();
+                type_->code = Type::String;
+                break;
+            }
             case Token::Name:
                 type_->meta = GetTypeMeta(token.value);
                 type_->name = token.value.to_string();
@@ -87,6 +175,10 @@ bool TypeParser::Parse(TypeAst* type) {
                 type_->meta = TypeAst::Number;
                 type_->value = std::stol(token.value.to_string());
                 break;
+            case Token::String:
+                type_->meta = TypeAst::String;
+                type_->value_string = std::string(token.value);
+                break;
             case Token::LPar:
                 type_->elements.emplace_back(TypeAst());
                 open_elements_.push(type_);
@@ -96,6 +188,7 @@ bool TypeParser::Parse(TypeAst* type) {
                 type_ = open_elements_.top();
                 open_elements_.pop();
                 break;
+            case Token::Assign:
             case Token::Comma:
                 type_ = open_elements_.top();
                 open_elements_.pop();
@@ -104,10 +197,21 @@ bool TypeParser::Parse(TypeAst* type) {
                 type_ = &type_->elements.back();
                 break;
             case Token::EOS:
-                return true;
+            {
+                // Ubalanced braces, brackets, etc is an error.
+                if (open_elements_.size() != 1)
+                    return false;
+
+                // Empty input string, no tokens produced
+                if (processed_tokens == 0)
+                    return false;
+
+                return ValidateAST(*type);
+            }
             case Token::Invalid:
                 return false;
         }
+        ++processed_tokens;
     } while (true);
 }
 
@@ -119,21 +223,44 @@ TypeParser::Token TypeParser::NextToken() {
             case '\t':
             case '\0':
                 continue;
-
             case '=':
-            case '\'':
-                continue;
-
+                return Token{Token::Assign, StringView(cur_++, 1)};
             case '(':
                 return Token{Token::LPar, StringView(cur_++, 1)};
             case ')':
                 return Token{Token::RPar, StringView(cur_++, 1)};
             case ',':
                 return Token{Token::Comma, StringView(cur_++, 1)};
+            case '\'':
+            {
+                const auto end_quote_length = 1;
+                const StringView end_quote{cur_, end_quote_length};
+                // Fast forward to the closing quote.
+                const auto start = cur_++;
+                for (; cur_ < end_ - end_quote_length; ++cur_) {
+                    // TODO (nemkov): handle escaping ?
+                    if (end_quote == StringView{cur_, end_quote_length}) {
+                        cur_ += end_quote_length;
+
+                        return Token{Token::QuotedString, StringView{start, cur_}};
+                    }
+                }
+                return Token{Token::QuotedString, StringView(cur_++, 1)};
+            }
 
             default: {
                 const char* st = cur_;
 
+                if (*cur_ == '\'') {
+                    for (st = ++cur_; cur_ < end_; ++cur_) {
+                        if (*cur_ == '\'') {
+                            return Token{Token::String, StringView(st, cur_++ - st)};
+                        }
+                    }
+
+                    return Token{Token::Invalid, StringView()};
+                }
+
                 if (isalpha(*cur_) || *cur_ == '_') {
                     for (; cur_ < end_; ++cur_) {
                         if (!isalpha(*cur_) && !isdigit(*cur_) && *cur_ != '_') {
@@ -167,8 +294,10 @@ const TypeAst* ParseTypeName(const std::string& type_name) {
     // Cache for type_name.
     // Usually we won't have too many type names in the cache, so do not try to
     // limit cache size.
-    static std::unordered_map<std::string, TypeAst> ast_cache;
+    static std::map<std::string, TypeAst> ast_cache;
+    static std::mutex lock;
 
+    std::lock_guard<std::mutex> guard(lock);
     auto it = ast_cache.find(type_name);
     if (it != ast_cache.end()) {
         return &it->second;
diff --git a/lib/clickhouse-cpp/clickhouse/types/type_parser.h b/lib/clickhouse-cpp/clickhouse/types/type_parser.h
index 31c0e1c..2f8f2f6 100644
--- a/lib/clickhouse-cpp/clickhouse/types/type_parser.h
+++ b/lib/clickhouse-cpp/clickhouse/types/type_parser.h
@@ -12,12 +12,17 @@ namespace clickhouse {
 struct TypeAst {
     enum Meta {
         Array,
+        Assign,
         Null,
         Nullable,
         Number,
+        String,
         Terminal,
         Tuple,
-        Enum
+        Enum,
+        LowCardinality,
+        SimpleAggregateFunction,
+        Map
     };
 
     /// Type's category.
@@ -29,9 +34,15 @@ struct TypeAst {
     /// Value associated with the node,
     /// used for fixed-width types and enum values.
     int64_t value = 0;
+    std::string value_string;
     /// Subelements of the type.
     /// Used to store enum's names and values as well.
-    std::list<TypeAst> elements;
+    std::vector<TypeAst> elements;
+
+    bool operator==(const TypeAst & other) const;
+    inline bool operator!=(const TypeAst & other) const {
+        return !(*this == other);
+    }
 };
 
 
@@ -40,11 +51,14 @@ class TypeParser {
     struct Token {
         enum Type {
             Invalid = 0,
+            Assign,
             Name,
             Number,
+            String,
             LPar,
             RPar,
             Comma,
+            QuotedString, // string with quotation marks included
             EOS,
         };
 
diff --git a/lib/clickhouse-cpp/clickhouse/types/types.cpp b/lib/clickhouse-cpp/clickhouse/types/types.cpp
index 09b815c..a5588c6 100644
--- a/lib/clickhouse-cpp/clickhouse/types/types.cpp
+++ b/lib/clickhouse-cpp/clickhouse/types/types.cpp
@@ -1,157 +1,234 @@
 #include "types.h"
 
-#include <assert.h>
+#include "../exceptions.h"
+
+#include <city.h>
+
+#include <stdexcept>
 
 namespace clickhouse {
 
 Type::Type(const Code code)
     : code_(code)
-{
-    if (code_ == Array) {
-        array_ = new ArrayImpl;
-    } else if (code_ == Tuple) {
-        tuple_ = new TupleImpl;
-    } else if (code_ == Nullable) {
-        nullable_ = new NullableImpl;
-    } else if (code_ == Enum8 || code_ == Enum16) {
-        enum_ = new EnumImpl;
-    }
-}
-
-Type::~Type() {
-    if (code_ == Array) {
-        delete array_;
-    } else if (code_ == Tuple) {
-        delete tuple_;
-    } else if (code_ == Nullable) {
-        delete nullable_;
-    } else if (code_ == Enum8 || code_ == Enum16) {
-        delete enum_;
+    , type_unique_id_(0)
+{}
+
+const char* Type::TypeName(Type::Code code) {
+    switch (code) {
+        case Type::Code::Void:           return "Void";
+        case Type::Code::Int8:           return "Int8";
+        case Type::Code::Int16:          return "Int16";
+        case Type::Code::Int32:          return "Int32";
+        case Type::Code::Int64:          return "Int64";
+        case Type::Code::UInt8:          return "UInt8";
+        case Type::Code::UInt16:         return "UInt16";
+        case Type::Code::UInt32:         return "UInt32";
+        case Type::Code::UInt64:         return "UInt64";
+        case Type::Code::Float32:        return "Float32";
+        case Type::Code::Float64:        return "Float64";
+        case Type::Code::String:         return "String";
+        case Type::Code::FixedString:    return "FixedString";
+        case Type::Code::DateTime:       return "DateTime";
+        case Type::Code::Date:           return "Date";
+        case Type::Code::Array:          return "Array";
+        case Type::Code::Nullable:       return "Nullable";
+        case Type::Code::Tuple:          return "Tuple";
+        case Type::Code::Enum8:          return "Enum8";
+        case Type::Code::Enum16:         return "Enum16";
+        case Type::Code::UUID:           return "UUID";
+        case Type::Code::IPv4:           return "IPv4";
+        case Type::Code::IPv6:           return "IPv6";
+        case Type::Code::Int128:         return "Int128";
+        case Type::Code::UInt128:        return "UInt128";
+        case Type::Code::Decimal:        return "Decimal";
+        case Type::Code::Decimal32:      return "Decimal32";
+        case Type::Code::Decimal64:      return "Decimal64";
+        case Type::Code::Decimal128:     return "Decimal128";
+        case Type::Code::LowCardinality: return "LowCardinality";
+        case Type::Code::DateTime64:     return "DateTime64";
+        case Type::Code::Date32:         return "Date32";
+        case Type::Code::Map:            return "Map";
+        case Type::Code::Point:          return "Point";
+        case Type::Code::Ring:           return "Ring";
+        case Type::Code::Polygon:        return "Polygon";
+        case Type::Code::MultiPolygon:   return "MultiPolygon";
+        case Type::Code::Time:           return "Time";
+        case Type::Code::Time64:         return "Time64";
     }
-}
 
-Type::Code Type::GetCode() const {
-    return code_;
-}
-
-TypeRef Type::GetItemType() const {
-    if (code_ == Array) {
-        return array_->item_type;
-    }
-    return TypeRef();
+    return "Unknown type";
 }
 
-TypeRef Type::GetNestedType() const {
-    if (code_ == Nullable) {
-        return nullable_->nested_type;
+std::string Type::GetName() const {
+    switch (code_) {
+        case Void:
+        case Int8:
+        case Int16:
+        case Int32:
+        case Int64:
+        case Int128:
+        case UInt8:
+        case UInt16:
+        case UInt32:
+        case UInt64:
+        case UInt128:
+        case UUID:
+        case Float32:
+        case Float64:
+        case String:
+        case IPv4:
+        case IPv6:
+        case Date:
+        case Date32:
+        case Time:
+        case Point:
+        case Ring:
+        case Polygon:
+        case MultiPolygon:
+            return TypeName(code_);
+        case Time64:
+            return As<Time64Type>()->GetName();
+        case FixedString:
+            return As<FixedStringType>()->GetName();
+        case DateTime:
+            return As<DateTimeType>()->GetName();
+        case DateTime64:
+            return As<DateTime64Type>()->GetName();
+        case Array:
+            return As<ArrayType>()->GetName();
+        case Nullable:
+            return As<NullableType>()->GetName();
+        case Tuple:
+            return As<TupleType>()->GetName();
+        case Enum8:
+        case Enum16:
+            return As<EnumType>()->GetName();
+        case Decimal:
+        case Decimal32:
+        case Decimal64:
+        case Decimal128:
+            return As<DecimalType>()->GetName();
+        case LowCardinality:
+            return As<LowCardinalityType>()->GetName();
+        case Map:
+            return As<MapType>()->GetName();
     }
-    return TypeRef();
-}
 
-std::vector<TypeRef> Type::GetTupleType() const {
-    return tuple_->item_types;
+    // XXX: NOT REACHED!
+    return std::string();
 }
 
-std::string Type::GetName() const {
+uint64_t Type::GetTypeUniqueId() const {
+    // Helper method to optimize equality checks of types with Type::IsEqual(),
+    // base invariant: types with same names produce same unique id (and hence considered equal).
+    // As an optimization, full type name is constructed at most once, and only for complex types.
     switch (code_) {
         case Void:
-            return "Void";
         case Int8:
-            return "Int8";
         case Int16:
-            return "Int16";
         case Int32:
-            return "Int32";
         case Int64:
-            return "Int64";
+        case Int128:
         case UInt8:
-            return "UInt8";
         case UInt16:
-            return "UInt16";
         case UInt32:
-            return "UInt32";
         case UInt64:
-            return "UInt64";
+        case UInt128:
         case UUID:
-            return "UUID";
         case Float32:
-            return "Float32";
         case Float64:
-            return "Float64";
         case String:
-            return "String";
+        case IPv4:
+        case IPv6:
+        case Date:
+        case Date32:
+        case Point:
+        case Ring:
+        case Polygon:
+        case MultiPolygon:
+            // For simple types, unique ID is the same as Type::Code
+            return code_;
+
         case FixedString:
-            return "FixedString(" + std::to_string(string_size_) + ")";
+        case Time:
+        case Time64:
         case DateTime:
-            return "DateTime";
-        case Date:
-            return "Date";
+        case DateTime64:
         case Array:
-            return std::string("Array(") + array_->item_type->GetName() +")";
         case Nullable:
-            return std::string("Nullable(") + nullable_->nested_type->GetName() + ")";
-        case Tuple: {
-            std::string result("Tuple(");
-            for (size_t i = 0; i < tuple_->item_types.size(); ++i) {
-                result += tuple_->item_types[i]->GetName();
-
-                if (i + 1 != tuple_->item_types.size()) {
-                    result += ", ";
-                }
-            }
-            result += ")";
-            return result;
-        }
+        case Tuple:
         case Enum8:
-        case Enum16: {
-            std::string result;
-            if (code_ == Enum8) {
-                result = "Enum8(";
-            } else {
-                result = "Enum16(";
-            }
-            for (auto ei = enum_->value_to_name.begin(); ; ) {
-                result += "'";
-                result += ei->second;
-                result += "' = ";
-                result += std::to_string(ei->first);
-
-                if (++ei != enum_->value_to_name.end()) {
-                    result += ", ";
-                } else {
-                    break;
-                }
+        case Enum16:
+        case Decimal:
+        case Decimal32:
+        case Decimal64:
+        case Decimal128:
+        case LowCardinality:
+        case Map: {
+            // For complex types, exact unique ID depends on nested types and/or parameters,
+            // the easiest way is to lazy-compute unique ID from name once.
+            // Here we do not care if multiple threads are computing value simultaneosly since it is both:
+            //   1. going to be the same
+            //   2. going to be stored atomically
+
+            if (type_unique_id_.load(std::memory_order_relaxed) == 0) {
+                const auto name = GetName();
+                type_unique_id_.store(CityHash64WithSeed(name.c_str(), name.size(), code_), std::memory_order_relaxed);
             }
-            result += ")";
-            return result;
+
+            return type_unique_id_;
         }
     }
-
-    return std::string();
-}
-
-bool Type::IsEqual(const TypeRef& other) const {
-    return this->GetName() == other->GetName();
+    assert(false);
+    return 0;
 }
 
 TypeRef Type::CreateArray(TypeRef item_type) {
-    TypeRef type(new Type(Type::Array));
-    type->array_->item_type = item_type;
-    return type;
+    return TypeRef(new ArrayType(item_type));
 }
 
 TypeRef Type::CreateDate() {
     return TypeRef(new Type(Type::Date));
 }
 
-TypeRef Type::CreateDateTime() {
-    return TypeRef(new Type(Type::DateTime));
+TypeRef Type::CreateDate32() {
+    return TypeRef(new Type(Type::Date32));
+}
+
+TypeRef Type::CreateTime() {
+    return TypeRef(new Type(Type::Time));
+}
+
+TypeRef Type::CreateTime64(size_t precision) {
+    return TypeRef(new Time64Type(precision));
+}
+
+TypeRef Type::CreateDateTime(std::string timezone) {
+    return TypeRef(new DateTimeType(std::move(timezone)));
+}
+
+TypeRef Type::CreateDateTime64(size_t precision, std::string timezone) {
+    return TypeRef(new DateTime64Type(precision, std::move(timezone)));
+}
+
+TypeRef Type::CreateDecimal(size_t precision, size_t scale) {
+    return TypeRef(new DecimalType(precision, scale));
+}
+
+TypeRef Type::CreateIPv4() {
+    return TypeRef(new Type(Type::IPv4));
+}
+
+TypeRef Type::CreateIPv6() {
+    return TypeRef(new Type(Type::IPv6));
+}
+
+TypeRef Type::CreateNothing() {
+    return TypeRef(new Type(Type::Void));
 }
 
 TypeRef Type::CreateNullable(TypeRef nested_type) {
-    TypeRef type(new Type(Type::Nullable));
-    type->nullable_->nested_type = nested_type;
-    return type;
+    return TypeRef(new NullableType(nested_type));
 }
 
 TypeRef Type::CreateString() {
@@ -159,69 +236,249 @@ TypeRef Type::CreateString() {
 }
 
 TypeRef Type::CreateString(size_t n) {
-    TypeRef type(new Type(Type::FixedString));
-    type->string_size_ = n;
-    return type;
+    return TypeRef(new FixedStringType(n));
 }
 
 TypeRef Type::CreateTuple(const std::vector<TypeRef>& item_types) {
-    TypeRef type(new Type(Type::Tuple));
-    type->tuple_->item_types.assign(item_types.begin(), item_types.end());
-    return type;
+    return TypeRef(new TupleType(item_types));
 }
 
 TypeRef Type::CreateEnum8(const std::vector<EnumItem>& enum_items) {
-    TypeRef type(new Type(Type::Enum8));
-    for (const auto& item : enum_items) {
-        type->enum_->value_to_name[item.value] = item.name;
-        type->enum_->name_to_value[item.name] = item.value;
-    }
-    return type;
+    return TypeRef(new EnumType(Type::Enum8, enum_items));
 }
 
 TypeRef Type::CreateEnum16(const std::vector<EnumItem>& enum_items) {
-    TypeRef type(new Type(Type::Enum16));
-    for (const auto& item : enum_items) {
-        type->enum_->value_to_name[item.value] = item.name;
-        type->enum_->name_to_value[item.name] = item.value;
-    }
-    return type;
+    return TypeRef(new EnumType(Type::Enum16, enum_items));
 }
 
 TypeRef Type::CreateUUID() {
     return TypeRef(new Type(Type::UUID));
 }
 
+TypeRef Type::CreateLowCardinality(TypeRef item_type) {
+    return std::make_shared<LowCardinalityType>(item_type);
+}
 
-EnumType::EnumType(const TypeRef& type)
-    : type_(type)
-{
-    assert(type_->GetCode() == Type::Enum8 ||
-           type_->GetCode() == Type::Enum16);
+TypeRef Type::CreateMap(TypeRef key_type, TypeRef value_type) {
+    return std::make_shared<MapType>(key_type, value_type);
 }
 
-const std::string& EnumType::GetEnumName(int16_t value) const {
-    return type_->enum_->value_to_name[value];
+TypeRef Type::CreatePoint() {
+    return TypeRef(new Type(Type::Point));
+}
+
+TypeRef Type::CreateRing() {
+    return TypeRef(new Type(Type::Ring));
+}
+
+TypeRef Type::CreatePolygon() {
+    return TypeRef(new Type(Type::Polygon));
+}
+
+TypeRef Type::CreateMultiPolygon() {
+    return TypeRef(new Type(Type::MultiPolygon));
+}
+
+/// class ArrayType
+
+ArrayType::ArrayType(TypeRef item_type) : Type(Array), item_type_(item_type) {
+}
+
+/// class DecimalType
+
+DecimalType::DecimalType(size_t precision, size_t scale)
+    : Type(Decimal),
+      precision_(precision),
+      scale_(scale) {
+    // TODO: assert(precision <= 38 && precision > 0);
+}
+
+std::string DecimalType::GetName() const {
+    switch (GetCode()) {
+        case Decimal:
+            return "Decimal(" + std::to_string(precision_) + "," + std::to_string(scale_) + ")";
+        case Decimal32:
+            return "Decimal32(" + std::to_string(scale_) + ")";
+        case Decimal64:
+            return "Decimal64(" + std::to_string(scale_) + ")";
+        case Decimal128:
+            return "Decimal128(" + std::to_string(scale_) + ")";
+        default:
+            /// XXX: NOT REACHED!
+            return "";
+    }
+}
+
+/// class EnumType
+
+EnumType::EnumType(Type::Code type, const std::vector<EnumItem>& items) : Type(type) {
+    for (const auto& item : items) {
+        auto result = name_to_value_.insert(item);
+        value_to_name_[item.second] = result.first->first;
+    }
+}
+
+std::string EnumType::GetName() const {
+    std::string result;
+
+    if (GetCode() == Enum8) {
+        result = "Enum8(";
+    } else {
+        result = "Enum16(";
+    }
+
+    for (auto ei = value_to_name_.begin(); ei != value_to_name_.end();) {
+        result += "'";
+        result += ei->second;
+        result += "' = ";
+        result += std::to_string(ei->first);
+
+        if (++ei != value_to_name_.end()) {
+            result += ", ";
+        } else {
+            break;
+        }
+    }
+
+    result += ")";
+
+    return result;
+}
+
+std::string_view EnumType::GetEnumName(int16_t value) const {
+    return value_to_name_.at(value);
 }
 
 int16_t EnumType::GetEnumValue(const std::string& name) const {
-    return type_->enum_->name_to_value[name];
+    return name_to_value_.at(name);
 }
 
 bool EnumType::HasEnumName(const std::string& name) const {
-    return type_->enum_->name_to_value.find(name) != type_->enum_->name_to_value.end();
+    return name_to_value_.find(name) != name_to_value_.end();
 }
 
 bool EnumType::HasEnumValue(int16_t value) const {
-    return type_->enum_->value_to_name.find(value) != type_->enum_->value_to_name.end();
+    return value_to_name_.find(value) != value_to_name_.end();
 }
 
 EnumType::ValueToNameIterator EnumType::BeginValueToName() const {
-    return type_->enum_->value_to_name.begin();
+    return value_to_name_.begin();
 }
 
 EnumType::ValueToNameIterator EnumType::EndValueToName() const {
-    return type_->enum_->value_to_name.end();
+    return value_to_name_.end();
+}
+
+
+namespace details
+{
+TypeWithTimeZoneMixin::TypeWithTimeZoneMixin(std::string timezone)
+    : timezone_(std::move(timezone)) {
+}
+
+const std::string & TypeWithTimeZoneMixin::Timezone() const {
+    return timezone_;
 }
+}
+
+/// class Time64
+Time64Type::Time64Type(size_t precision)
+    : Type(Time64), precision_{precision} {
+    if (precision_ > 9) {
+        throw ValidationError("Time64 precision is > 9");
+    }
+}
+
+std::string Time64Type::GetName() const {
+    return "Time64(" + std::to_string(precision_) + ")";
+}
+
+/// class DateTimeType
+DateTimeType::DateTimeType(std::string timezone)
+    : Type(DateTime), details::TypeWithTimeZoneMixin(std::move(timezone)) {
+}
+
+std::string DateTimeType::GetName() const {
+    std::string datetime_representation = "DateTime";
+    const auto & timezone = Timezone();
+    if (!timezone.empty())
+        datetime_representation += "('" + timezone + "')";
 
+    return datetime_representation;
 }
+
+/// class DateTime64Type
+
+DateTime64Type::DateTime64Type(size_t precision, std::string timezone)
+    : Type(DateTime64), details::TypeWithTimeZoneMixin(std::move(timezone)), precision_(precision) {
+
+    if (precision_ > 18) {
+        throw ValidationError("DateTime64 precision is > 18");
+    }
+}
+
+std::string DateTime64Type::GetName() const {
+    std::string datetime64_representation;
+    datetime64_representation.reserve(14);
+    datetime64_representation += "DateTime64(";
+    datetime64_representation += std::to_string(precision_);
+
+    const auto & timezone = Timezone();
+    if (!timezone.empty()) {
+        datetime64_representation += ", '" + timezone + "'";
+    }
+
+    datetime64_representation += ")";
+    return datetime64_representation;
+}
+
+/// class FixedStringType
+
+FixedStringType::FixedStringType(size_t n) : Type(FixedString), size_(n) {
+}
+
+/// class NullableType
+
+NullableType::NullableType(TypeRef nested_type) : Type(Nullable), nested_type_(nested_type) {
+}
+
+/// class TupleType
+
+TupleType::TupleType(const std::vector<TypeRef>& item_types) : Type(Tuple), item_types_(item_types) {
+}
+
+/// class LowCardinalityType
+LowCardinalityType::LowCardinalityType(TypeRef nested_type) : Type(LowCardinality), nested_type_(nested_type) {
+}
+
+LowCardinalityType::~LowCardinalityType() {
+}
+
+std::string TupleType::GetName() const {
+    std::string result("Tuple(");
+
+    if (!item_types_.empty()) {
+        result += item_types_[0]->GetName();
+    }
+
+    for (size_t i = 1; i < item_types_.size(); ++i) {
+        result += ", " + item_types_[i]->GetName();
+    }
+
+    result += ")";
+
+    return result;
+}
+
+/// class MapType
+MapType::MapType(TypeRef key_type, TypeRef value_type)
+    : Type(Map)
+    , key_type_(key_type)
+    , value_type_(value_type) {
+}
+
+std::string MapType::GetName() const {
+    return std::string("Map(") + key_type_->GetName() + ", " +value_type_->GetName() + ")";
+}
+
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/types/types.h b/lib/clickhouse-cpp/clickhouse/types/types.h
index e8d6c99..1727c8e 100644
--- a/lib/clickhouse-cpp/clickhouse/types/types.h
+++ b/lib/clickhouse-cpp/clickhouse/types/types.h
@@ -1,12 +1,21 @@
 #pragma once
 
+#include "absl/numeric/int128.h"
+
+#include <atomic>
+#include <cstdint>
 #include <map>
 #include <memory>
 #include <string>
 #include <vector>
+#include <stdexcept>
 
 namespace clickhouse {
 
+using Int128 = absl::int128;
+using UInt128 = absl::uint128;
+using Int64 = int64_t;
+
 using TypeRef = std::shared_ptr<class Type>;
 
 class Type {
@@ -33,40 +42,79 @@ class Type {
         Enum8,
         Enum16,
         UUID,
+        IPv4,
+        IPv6,
+        Int128,
+        UInt128,
+        Decimal,
+        Decimal32,
+        Decimal64,
+        Decimal128,
+        LowCardinality,
+        DateTime64,
+        Date32,
+        Map,
+        Point,
+        Ring,
+        Polygon,
+        MultiPolygon,
+        Time,
+        Time64,
     };
 
-    struct EnumItem {
-        std::string name;
-        int16_t value;
-    };
+    using EnumItem = std::pair<std::string /* name */, int16_t /* value */>;
 
-    /// Destructor
-    ~Type();
-
-    /// Type's code.
-    Code GetCode() const;
+protected:
+    Type(const Code code);
 
-    /// Type of array's elements.
-    TypeRef GetItemType() const;
+public:
+    template <typename Derived>
+    auto* As() {
+        return static_cast<Derived*>(this);
+    }
 
-    /// Type of nested nullable element.
-    TypeRef GetNestedType() const;
+    template <typename Derived>
+    const auto* As() const {
+        return static_cast<const Derived*>(this);
+    }
 
-    /// Type of nested Tuple element type.
-    std::vector<TypeRef> GetTupleType() const;
+    /// Type's code.
+    Code GetCode() const { return code_; }
 
     /// String representation of the type.
     std::string GetName() const;
 
     /// Is given type same as current one.
-    bool IsEqual(const TypeRef& other) const;
+    bool IsEqual(const Type& other) const {
+        // Types are equal only if both code_ and type_unique_id_ are equal.
+        return this == &other
+                // GetTypeUniqueId() is relatively heavy, so avoid calling it when comparing obviously different types.
+                || (this->GetCode() == other.GetCode() && this->GetTypeUniqueId() == other.GetTypeUniqueId());
+    }
+
+    bool IsEqual(const TypeRef& other) const { return IsEqual(*other); }
+
+    /// Simple name, doesn't depend on parameters and\or nested types, caller MUST NOT free returned value.
+    static const char* TypeName(Code);
 
 public:
     static TypeRef CreateArray(TypeRef item_type);
 
     static TypeRef CreateDate();
 
-    static TypeRef CreateDateTime();
+    static TypeRef CreateDate32();
+
+    static TypeRef CreateDateTime(std::string timezone = std::string());
+
+    static TypeRef CreateDateTime64(size_t precision, std::string timezone = std::string());
+
+    static TypeRef CreateDecimal(size_t precision, size_t scale);
+
+    static TypeRef CreateIPv4();
+
+    static TypeRef CreateIPv6();
+
+    static TypeRef CreateNothing();
 
     static TypeRef CreateNullable(TypeRef nested_type);
 
@@ -85,61 +133,206 @@ class Type {
 
     static TypeRef CreateUUID();
 
+    static TypeRef CreateLowCardinality(TypeRef item_type);
+
+    static TypeRef CreateMap(TypeRef key_type, TypeRef value_type);
+
+    static TypeRef CreatePoint();
+
+    static TypeRef CreateRing();
+
+    static TypeRef CreatePolygon();
+
+    static TypeRef CreateMultiPolygon();
+
+    static TypeRef CreateTime();
+
+    static TypeRef CreateTime64(size_t precision);
+
 private:
-    Type(const Code code);
+    uint64_t GetTypeUniqueId() const;
 
-    struct ArrayImpl {
-        TypeRef item_type;
-    };
+    const Code code_;
+    mutable std::atomic<uint64_t> type_unique_id_;
+};
 
-    struct NullableImpl {
-        TypeRef nested_type;
-    };
+inline bool operator==(const Type & left, const Type & right) {
+    if (&left == &right)
+        return true;
+    if (typeid(left) == typeid(right))
+        return left.IsEqual(right);
+    return false;
+}
 
-    struct TupleImpl {
-        std::vector<TypeRef> item_types;
-    };
+inline bool operator==(const TypeRef & left, const TypeRef & right) {
+    return *left == *right;
+}
 
-    struct EnumImpl {
-        using ValueToNameType = std::map<int16_t, std::string>;
-        using NameToValueType = std::map<std::string, int16_t>;
-        ValueToNameType value_to_name;
-        NameToValueType name_to_value;
-    };
+class ArrayType : public Type {
+public:
+    explicit ArrayType(TypeRef item_type);
 
+    std::string GetName() const { return std::string("Array(") + item_type_->GetName() + ")"; }
+
+    /// Type of array's elements.
+    inline TypeRef GetItemType() const { return item_type_; }
+
+private:
+    TypeRef item_type_;
+};
+
+class DecimalType : public Type {
+public:
+    DecimalType(size_t precision, size_t scale);
+
+    std::string GetName() const;
     friend class EnumType;
+    friend class DateTimeType;
 
+    inline size_t GetScale() const { return scale_; }
+    inline size_t GetPrecision() const { return precision_; }
 
-    const Code code_;
-    union {
-        ArrayImpl* array_;
-        NullableImpl* nullable_;
-        TupleImpl* tuple_;
-        EnumImpl* enum_;
-        int string_size_;
-    };
+private:
+    const size_t precision_, scale_;
 };
 
-class EnumType {
+namespace details
+{
+class TypeWithTimeZoneMixin
+{
 public:
-    explicit EnumType(const TypeRef& type);
+    TypeWithTimeZoneMixin(std::string timezone);
+
+    /// Timezone associated with a data column.
+    const std::string & Timezone() const;
+
+private:
+    std::string timezone_;
+};
+}
+
+class Time64Type : public Type {
+public:
+    explicit Time64Type(size_t precision);
+
+    std::string GetName() const;
+
+    inline size_t GetPrecision() const { return precision_; }
+
+private:
+    size_t precision_;
+};
+
+class DateTimeType : public Type, public details::TypeWithTimeZoneMixin {
+public:
+    explicit DateTimeType(std::string timezone);
+
+    std::string GetName() const;
+};
+
+class DateTime64Type: public Type, public details::TypeWithTimeZoneMixin {
+public:
+    explicit DateTime64Type(size_t precision, std::string timezone_);
+
+    std::string GetName() const;
+
+    inline size_t GetPrecision() const { return precision_; }
+private:
+    size_t precision_;
+};
+
+class EnumType : public Type {
+public:
+    EnumType(Type::Code type, const std::vector<EnumItem>& items);
+
+    std::string GetName() const;
 
-    std::string GetName() const {
-        return type_->GetName();
-    }
     /// Methods to work with enum types.
-    const std::string& GetEnumName(int16_t value) const;
+    std::string_view GetEnumName(int16_t value) const;
     int16_t GetEnumValue(const std::string& name) const;
     bool HasEnumName(const std::string& name) const;
     bool HasEnumValue(int16_t value) const;
 
-    /// Iterator for enum elements.
-    using ValueToNameIterator = Type::EnumImpl::ValueToNameType::const_iterator;
+private:
+    using ValueToNameType     = std::map<int16_t, std::string_view>;
+    using NameToValueType     = std::map<std::string, int16_t>;
+    using ValueToNameIterator = ValueToNameType::const_iterator;
+
+    ValueToNameType value_to_name_;
+    NameToValueType name_to_value_;
+
+public:
     ValueToNameIterator BeginValueToName() const;
     ValueToNameIterator EndValueToName() const;
+};
+
+class FixedStringType : public Type {
+public:
+    explicit FixedStringType(size_t n);
+
+    std::string GetName() const { return std::string("FixedString(") + std::to_string(size_) + ")"; }
+
+    inline size_t GetSize() const { return size_; }
+
+private:
+    size_t size_;
+};
+
+class NullableType : public Type {
+public:
+    explicit NullableType(TypeRef nested_type);
+
+    std::string GetName() const { return std::string("Nullable(") + nested_type_->GetName() + ")"; }
+
+    /// Type of nested nullable element.
+    TypeRef GetNestedType() const { return nested_type_; }
 
 private:
-    TypeRef type_;
+    TypeRef nested_type_;
+};
+
+class TupleType : public Type {
+public:
+    explicit TupleType(const std::vector<TypeRef>& item_types);
+
+    std::string GetName() const;
+
+    /// Type of nested Tuple element type.
+    std::vector<TypeRef> GetTupleType() const { return item_types_; }
+
+private:
+    std::vector<TypeRef> item_types_;
+};
+
+class LowCardinalityType : public Type {
+public:
+    explicit LowCardinalityType(TypeRef nested_type);
+    ~LowCardinalityType();
+
+    std::string GetName() const { return std::string("LowCardinality(") + nested_type_->GetName() + ")"; }
+
+    /// Type of nested nullable element.
+    TypeRef GetNestedType() const { return nested_type_; }
+
+private:
+    TypeRef nested_type_;
+};
+
+class MapType : public Type {
+public:
+    explicit MapType(TypeRef key_type, TypeRef value_type);
+
+    std::string GetName() const;
+
+    /// Type of keys.
+    TypeRef GetKeyType() const { return key_type_; }
+
+    /// Type of values.
+    TypeRef GetValueType() const { return value_type_; }
+
+private:
+    TypeRef key_type_;
+    TypeRef value_type_;
 };
 
 template <>
@@ -162,6 +355,16 @@ inline TypeRef Type::CreateSimple<int64_t>() {
     return TypeRef(new Type(Int64));
 }
 
+template <>
+inline TypeRef Type::CreateSimple<Int128>() {
+    return TypeRef(new Type(Int128));
+}
+
+template <>
+inline TypeRef Type::CreateSimple<UInt128>() {
+    return TypeRef(new Type(UInt128));
+}
+
 template <>
 inline TypeRef Type::CreateSimple<uint8_t>() {
     return TypeRef(new Type(UInt8));
@@ -192,4 +395,4 @@ inline TypeRef Type::CreateSimple<double>() {
     return TypeRef(new Type(Float64));
 }
 
-}
+}  // namespace clickhouse
diff --git a/lib/clickhouse-cpp/clickhouse/version.h b/lib/clickhouse-cpp/clickhouse/version.h
new file mode 100644
index 0000000..99bbd19
--- /dev/null
+++ b/lib/clickhouse-cpp/clickhouse/version.h
@@ -0,0 +1,14 @@
+#pragma once
+
+#define CLICKHOUSE_CPP_VERSION_MAJOR 2
+#define CLICKHOUSE_CPP_VERSION_MINOR 6
+#define CLICKHOUSE_CPP_VERSION_PATCH 1
+
+#define CLICKHOUSE_CPP_VERSION_BUILD 0
+
+// Expecting each version component to be less than 99
+#define CLICKHOUSE_CPP_VERSION \
+    CLICKHOUSE_CPP_VERSION_MAJOR *   100 * 100 * 100 \
+    + CLICKHOUSE_CPP_VERSION_MINOR * 100 * 100 \
+    + CLICKHOUSE_CPP_VERSION_PATCH * 100 \
+    + CLICKHOUSE_CPP_VERSION_BUILD
diff --git a/lib/clickhouse-cpp/cmake/cpp11.cmake b/lib/clickhouse-cpp/cmake/cpp11.cmake
deleted file mode 100644
index bd47583..0000000
--- a/lib/clickhouse-cpp/cmake/cpp11.cmake
+++ /dev/null
@@ -1,8 +0,0 @@
-MACRO (USE_CXX11)
-    IF (CMAKE_VERSION VERSION_LESS "3.1")
-      SET (CMAKE_CXX_FLAGS "-std=c++11 ${CMAKE_CXX_FLAGS}")
-    ELSE ()
-      SET (CMAKE_CXX_STANDARD 11)
-      SET (CMAKE_CXX_STANDARD_REQUIRED ON)
-    ENDIF ()
-ENDMACRO (USE_CXX11)
diff --git a/lib/clickhouse-cpp/cmake/subdirs.cmake b/lib/clickhouse-cpp/cmake/subdirs.cmake
deleted file mode 100644
index 270c579..0000000
--- a/lib/clickhouse-cpp/cmake/subdirs.cmake
+++ /dev/null
@@ -1,5 +0,0 @@
-FUNCTION (SUBDIRS)
-    FOREACH (dir ${ARGV})
-        ADD_SUBDIRECTORY (${dir})
-    ENDFOREACH(dir)
-ENDFUNCTION (SUBDIRS)
\ No newline at end of file
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/CMakeLists.txt b/lib/clickhouse-cpp/contrib/absl/absl/CMakeLists.txt
new file mode 100644
index 0000000..69f088f
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/CMakeLists.txt
@@ -0,0 +1,9 @@
+ADD_LIBRARY (absl_int128 STATIC
+    numeric/int128.cc
+)
+
+TARGET_INCLUDE_DIRECTORIES (absl_int128
+    PUBLIC ${PROJECT_SOURCE_DIR}/contrib/absl
+)
+
+ADD_LIBRARY (absl::int128 ALIAS absl_int128)
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/attributes.h b/lib/clickhouse-cpp/contrib/absl/absl/base/attributes.h
new file mode 100644
index 0000000..046fbea
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/attributes.h
@@ -0,0 +1,682 @@
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// This header file defines macros for declaring attributes for functions,
+// types, and variables.
+//
+// These macros are used within Abseil and allow the compiler to optimize, where
+// applicable, certain function calls.
+//
+// This file is used for both C and C++!
+//
+// Most macros here are exposing GCC or Clang features, and are stubbed out for
+// other compilers.
+//
+// GCC attributes documentation:
+//   https://gcc.gnu.org/onlinedocs/gcc-4.7.0/gcc/Function-Attributes.html
+//   https://gcc.gnu.org/onlinedocs/gcc-4.7.0/gcc/Variable-Attributes.html
+//   https://gcc.gnu.org/onlinedocs/gcc-4.7.0/gcc/Type-Attributes.html
+//
+// Most attributes in this file are already supported by GCC 4.7. However, some
+// of them are not supported in older version of Clang. Thus, we check
+// `__has_attribute()` first. If the check fails, we check if we are on GCC and
+// assume the attribute exists on GCC (which is verified on GCC 4.7).
+
+#ifndef ABSL_BASE_ATTRIBUTES_H_
+#define ABSL_BASE_ATTRIBUTES_H_
+
+#include "absl/base/config.h"
+
+// ABSL_HAVE_ATTRIBUTE
+//
+// A function-like feature checking macro that is a wrapper around
+// `__has_attribute`, which is defined by GCC 5+ and Clang and evaluates to a
+// nonzero constant integer if the attribute is supported or 0 if not.
+//
+// It evaluates to zero if `__has_attribute` is not defined by the compiler.
+//
+// GCC: https://gcc.gnu.org/gcc-5/changes.html
+// Clang: https://clang.llvm.org/docs/LanguageExtensions.html
+#ifdef __has_attribute
+#define ABSL_HAVE_ATTRIBUTE(x) __has_attribute(x)
+#else
+#define ABSL_HAVE_ATTRIBUTE(x) 0
+#endif
+
+// ABSL_HAVE_CPP_ATTRIBUTE
+//
+// A function-like feature checking macro that accepts C++11 style attributes.
+// It's a wrapper around `__has_cpp_attribute`, defined by ISO C++ SD-6
+// (https://en.cppreference.com/w/cpp/experimental/feature_test). If we don't
+// find `__has_cpp_attribute`, will evaluate to 0.
+#if defined(__cplusplus) && defined(__has_cpp_attribute)
+// NOTE: requiring __cplusplus above should not be necessary, but
+// works around https://bugs.llvm.org/show_bug.cgi?id=23435.
+#define ABSL_HAVE_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+#else
+#define ABSL_HAVE_CPP_ATTRIBUTE(x) 0
+#endif
+
+// -----------------------------------------------------------------------------
+// Function Attributes
+// -----------------------------------------------------------------------------
+//
+// GCC: https://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html
+// Clang: https://clang.llvm.org/docs/AttributeReference.html
+
+// ABSL_PRINTF_ATTRIBUTE
+// ABSL_SCANF_ATTRIBUTE
+//
+// Tells the compiler to perform `printf` format string checking if the
+// compiler supports it; see the 'format' attribute in
+// <https://gcc.gnu.org/onlinedocs/gcc-4.7.0/gcc/Function-Attributes.html>.
+//
+// Note: As the GCC manual states, "[s]ince non-static C++ methods
+// have an implicit 'this' argument, the arguments of such methods
+// should be counted from two, not one."
+#if ABSL_HAVE_ATTRIBUTE(format) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_PRINTF_ATTRIBUTE(string_index, first_to_check) \
+  __attribute__((__format__(__printf__, string_index, first_to_check)))
+#define ABSL_SCANF_ATTRIBUTE(string_index, first_to_check) \
+  __attribute__((__format__(__scanf__, string_index, first_to_check)))
+#else
+#define ABSL_PRINTF_ATTRIBUTE(string_index, first_to_check)
+#define ABSL_SCANF_ATTRIBUTE(string_index, first_to_check)
+#endif
+
+// ABSL_ATTRIBUTE_ALWAYS_INLINE
+// ABSL_ATTRIBUTE_NOINLINE
+//
+// Forces functions to either inline or not inline. Introduced in gcc 3.1.
+#if ABSL_HAVE_ATTRIBUTE(always_inline) || \
+    (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_ALWAYS_INLINE __attribute__((always_inline))
+#define ABSL_HAVE_ATTRIBUTE_ALWAYS_INLINE 1
+#else
+#define ABSL_ATTRIBUTE_ALWAYS_INLINE
+#endif
+
+#if ABSL_HAVE_ATTRIBUTE(noinline) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_NOINLINE __attribute__((noinline))
+#define ABSL_HAVE_ATTRIBUTE_NOINLINE 1
+#else
+#define ABSL_ATTRIBUTE_NOINLINE
+#endif
+
+// ABSL_ATTRIBUTE_NO_TAIL_CALL
+//
+// Prevents the compiler from optimizing away stack frames for functions which
+// end in a call to another function.
+#if ABSL_HAVE_ATTRIBUTE(disable_tail_calls)
+#define ABSL_HAVE_ATTRIBUTE_NO_TAIL_CALL 1
+#define ABSL_ATTRIBUTE_NO_TAIL_CALL __attribute__((disable_tail_calls))
+#elif defined(__GNUC__) && !defined(__clang__)
+#define ABSL_HAVE_ATTRIBUTE_NO_TAIL_CALL 1
+#define ABSL_ATTRIBUTE_NO_TAIL_CALL \
+  __attribute__((optimize("no-optimize-sibling-calls")))
+#else
+#define ABSL_ATTRIBUTE_NO_TAIL_CALL
+#define ABSL_HAVE_ATTRIBUTE_NO_TAIL_CALL 0
+#endif
+
+// ABSL_ATTRIBUTE_WEAK
+//
+// Tags a function as weak for the purposes of compilation and linking.
+// Weak attributes currently do not work properly in LLVM's Windows backend,
+// so disable them there. See https://bugs.llvm.org/show_bug.cgi?id=37598
+// for further information.
+// The MinGW compiler doesn't complain about the weak attribute until the link
+// step, presumably because Windows doesn't use ELF binaries.
+#if (ABSL_HAVE_ATTRIBUTE(weak) ||                   \
+     (defined(__GNUC__) && !defined(__clang__))) && \
+    !(defined(__llvm__) && defined(_WIN32)) && !defined(__MINGW32__)
+#undef ABSL_ATTRIBUTE_WEAK
+#define ABSL_ATTRIBUTE_WEAK __attribute__((weak))
+#define ABSL_HAVE_ATTRIBUTE_WEAK 1
+#else
+#define ABSL_ATTRIBUTE_WEAK
+#define ABSL_HAVE_ATTRIBUTE_WEAK 0
+#endif
+
+// ABSL_ATTRIBUTE_NONNULL
+//
+// Tells the compiler either (a) that a particular function parameter
+// should be a non-null pointer, or (b) that all pointer arguments should
+// be non-null.
+//
+// Note: As the GCC manual states, "[s]ince non-static C++ methods
+// have an implicit 'this' argument, the arguments of such methods
+// should be counted from two, not one."
+//
+// Args are indexed starting at 1.
+//
+// For non-static class member functions, the implicit `this` argument
+// is arg 1, and the first explicit argument is arg 2. For static class member
+// functions, there is no implicit `this`, and the first explicit argument is
+// arg 1.
+//
+// Example:
+//
+//   /* arg_a cannot be null, but arg_b can */
+//   void Function(void* arg_a, void* arg_b) ABSL_ATTRIBUTE_NONNULL(1);
+//
+//   class C {
+//     /* arg_a cannot be null, but arg_b can */
+//     void Method(void* arg_a, void* arg_b) ABSL_ATTRIBUTE_NONNULL(2);
+//
+//     /* arg_a cannot be null, but arg_b can */
+//     static void StaticMethod(void* arg_a, void* arg_b)
+//     ABSL_ATTRIBUTE_NONNULL(1);
+//   };
+//
+// If no arguments are provided, then all pointer arguments should be non-null.
+//
+//  /* No pointer arguments may be null. */
+//  void Function(void* arg_a, void* arg_b, int arg_c) ABSL_ATTRIBUTE_NONNULL();
+//
+// NOTE: The GCC nonnull attribute actually accepts a list of arguments, but
+// ABSL_ATTRIBUTE_NONNULL does not.
+#if ABSL_HAVE_ATTRIBUTE(nonnull) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_NONNULL(arg_index) __attribute__((nonnull(arg_index)))
+#else
+#define ABSL_ATTRIBUTE_NONNULL(...)
+#endif
+
+// ABSL_ATTRIBUTE_NORETURN
+//
+// Tells the compiler that a given function never returns.
+#if ABSL_HAVE_ATTRIBUTE(noreturn) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_NORETURN __attribute__((noreturn))
+#elif defined(_MSC_VER)
+#define ABSL_ATTRIBUTE_NORETURN __declspec(noreturn)
+#else
+#define ABSL_ATTRIBUTE_NORETURN
+#endif
+
+// ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS
+//
+// Tells the AddressSanitizer (or other memory testing tools) to ignore a given
+// function. Useful for cases when a function reads random locations on stack,
+// calls _exit from a cloned subprocess, deliberately accesses buffer
+// out of bounds or does other scary things with memory.
+// NOTE: GCC supports AddressSanitizer(asan) since 4.8.
+// https://gcc.gnu.org/gcc-4.8/changes.html
+#if ABSL_HAVE_ATTRIBUTE(no_sanitize_address)
+#define ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS __attribute__((no_sanitize_address))
+#else
+#define ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS
+#endif
+
+// ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY
+//
+// Tells the MemorySanitizer to relax the handling of a given function. All "Use
+// of uninitialized value" warnings from such functions will be suppressed, and
+// all values loaded from memory will be considered fully initialized.  This
+// attribute is similar to the ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS attribute
+// above, but deals with initialized-ness rather than addressability issues.
+// NOTE: MemorySanitizer(msan) is supported by Clang but not GCC.
+#if ABSL_HAVE_ATTRIBUTE(no_sanitize_memory)
+#define ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY __attribute__((no_sanitize_memory))
+#else
+#define ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY
+#endif
+
+// ABSL_ATTRIBUTE_NO_SANITIZE_THREAD
+//
+// Tells the ThreadSanitizer to not instrument a given function.
+// NOTE: GCC supports ThreadSanitizer(tsan) since 4.8.
+// https://gcc.gnu.org/gcc-4.8/changes.html
+#if ABSL_HAVE_ATTRIBUTE(no_sanitize_thread)
+#define ABSL_ATTRIBUTE_NO_SANITIZE_THREAD __attribute__((no_sanitize_thread))
+#else
+#define ABSL_ATTRIBUTE_NO_SANITIZE_THREAD
+#endif
+
+// ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED
+//
+// Tells the UndefinedSanitizer to ignore a given function. Useful for cases
+// where certain behavior (eg. division by zero) is being used intentionally.
+// NOTE: GCC supports UndefinedBehaviorSanitizer(ubsan) since 4.9.
+// https://gcc.gnu.org/gcc-4.9/changes.html
+#if ABSL_HAVE_ATTRIBUTE(no_sanitize_undefined)
+#define ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED \
+  __attribute__((no_sanitize_undefined))
+#elif ABSL_HAVE_ATTRIBUTE(no_sanitize)
+#define ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED \
+  __attribute__((no_sanitize("undefined")))
+#else
+#define ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED
+#endif
+
+// ABSL_ATTRIBUTE_NO_SANITIZE_CFI
+//
+// Tells the ControlFlowIntegrity sanitizer to not instrument a given function.
+// See https://clang.llvm.org/docs/ControlFlowIntegrity.html for details.
+#if ABSL_HAVE_ATTRIBUTE(no_sanitize)
+#define ABSL_ATTRIBUTE_NO_SANITIZE_CFI __attribute__((no_sanitize("cfi")))
+#else
+#define ABSL_ATTRIBUTE_NO_SANITIZE_CFI
+#endif
+
+// ABSL_ATTRIBUTE_NO_SANITIZE_SAFESTACK
+//
+// Tells the SafeStack to not instrument a given function.
+// See https://clang.llvm.org/docs/SafeStack.html for details.
+#if ABSL_HAVE_ATTRIBUTE(no_sanitize)
+#define ABSL_ATTRIBUTE_NO_SANITIZE_SAFESTACK \
+  __attribute__((no_sanitize("safe-stack")))
+#else
+#define ABSL_ATTRIBUTE_NO_SANITIZE_SAFESTACK
+#endif
+
+// ABSL_ATTRIBUTE_RETURNS_NONNULL
+//
+// Tells the compiler that a particular function never returns a null pointer.
+#if ABSL_HAVE_ATTRIBUTE(returns_nonnull) || \
+    (defined(__GNUC__) && \
+     (__GNUC__ > 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9)) && \
+     !defined(__clang__))
+#define ABSL_ATTRIBUTE_RETURNS_NONNULL __attribute__((returns_nonnull))
+#else
+#define ABSL_ATTRIBUTE_RETURNS_NONNULL
+#endif
+
+// ABSL_HAVE_ATTRIBUTE_SECTION
+//
+// Indicates whether labeled sections are supported. Weak symbol support is
+// a prerequisite. Labeled sections are not supported on Darwin/iOS.
+#ifdef ABSL_HAVE_ATTRIBUTE_SECTION
+#error ABSL_HAVE_ATTRIBUTE_SECTION cannot be directly set
+#elif (ABSL_HAVE_ATTRIBUTE(section) ||                \
+       (defined(__GNUC__) && !defined(__clang__))) && \
+    !defined(__APPLE__) && ABSL_HAVE_ATTRIBUTE_WEAK
+#define ABSL_HAVE_ATTRIBUTE_SECTION 1
+
+// ABSL_ATTRIBUTE_SECTION
+//
+// Tells the compiler/linker to put a given function into a section and define
+// `__start_ ## name` and `__stop_ ## name` symbols to bracket the section.
+// This functionality is supported by GNU linker.  Any function annotated with
+// `ABSL_ATTRIBUTE_SECTION` must not be inlined, or it will be placed into
+// whatever section its caller is placed into.
+//
+#ifndef ABSL_ATTRIBUTE_SECTION
+#define ABSL_ATTRIBUTE_SECTION(name) \
+  __attribute__((section(#name))) __attribute__((noinline))
+#endif
+
+
+// ABSL_ATTRIBUTE_SECTION_VARIABLE
+//
+// Tells the compiler/linker to put a given variable into a section and define
+// `__start_ ## name` and `__stop_ ## name` symbols to bracket the section.
+// This functionality is supported by GNU linker.
+#ifndef ABSL_ATTRIBUTE_SECTION_VARIABLE
+#define ABSL_ATTRIBUTE_SECTION_VARIABLE(name) __attribute__((section(#name)))
+#endif
+
+// ABSL_DECLARE_ATTRIBUTE_SECTION_VARS
+//
+// A weak section declaration to be used as a global declaration
+// for ABSL_ATTRIBUTE_SECTION_START|STOP(name) to compile and link
+// even without functions with ABSL_ATTRIBUTE_SECTION(name).
+// ABSL_DEFINE_ATTRIBUTE_SECTION should be in the exactly one file; it's
+// a no-op on ELF but not on Mach-O.
+//
+#ifndef ABSL_DECLARE_ATTRIBUTE_SECTION_VARS
+#define ABSL_DECLARE_ATTRIBUTE_SECTION_VARS(name) \
+  extern char __start_##name[] ABSL_ATTRIBUTE_WEAK;    \
+  extern char __stop_##name[] ABSL_ATTRIBUTE_WEAK
+#endif
+#ifndef ABSL_DEFINE_ATTRIBUTE_SECTION_VARS
+#define ABSL_INIT_ATTRIBUTE_SECTION_VARS(name)
+#define ABSL_DEFINE_ATTRIBUTE_SECTION_VARS(name)
+#endif
+
+// ABSL_ATTRIBUTE_SECTION_START
+//
+// Returns `void*` pointers to start/end of a section of code with
+// functions having ABSL_ATTRIBUTE_SECTION(name).
+// Returns 0 if no such functions exist.
+// One must ABSL_DECLARE_ATTRIBUTE_SECTION_VARS(name) for this to compile and
+// link.
+//
+#define ABSL_ATTRIBUTE_SECTION_START(name) \
+  (reinterpret_cast<void *>(__start_##name))
+#define ABSL_ATTRIBUTE_SECTION_STOP(name) \
+  (reinterpret_cast<void *>(__stop_##name))
+
+#else  // !ABSL_HAVE_ATTRIBUTE_SECTION
+
+#define ABSL_HAVE_ATTRIBUTE_SECTION 0
+
+// provide dummy definitions
+#define ABSL_ATTRIBUTE_SECTION(name)
+#define ABSL_ATTRIBUTE_SECTION_VARIABLE(name)
+#define ABSL_INIT_ATTRIBUTE_SECTION_VARS(name)
+#define ABSL_DEFINE_ATTRIBUTE_SECTION_VARS(name)
+#define ABSL_DECLARE_ATTRIBUTE_SECTION_VARS(name)
+#define ABSL_ATTRIBUTE_SECTION_START(name) (reinterpret_cast<void *>(0))
+#define ABSL_ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast<void *>(0))
+
+#endif  // ABSL_ATTRIBUTE_SECTION
+
+// ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC
+//
+// Support for aligning the stack on 32-bit x86.
+#if ABSL_HAVE_ATTRIBUTE(force_align_arg_pointer) || \
+    (defined(__GNUC__) && !defined(__clang__))
+#if defined(__i386__)
+#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC \
+  __attribute__((force_align_arg_pointer))
+#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (0)
+#elif defined(__x86_64__)
+#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (1)
+#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC
+#else  // !__i386__ && !__x86_64
+#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (0)
+#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC
+#endif  // __i386__
+#else
+#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC
+#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (0)
+#endif
+
+// ABSL_MUST_USE_RESULT
+//
+// Tells the compiler to warn about unused results.
+//
+// When annotating a function, it must appear as the first part of the
+// declaration or definition. The compiler will warn if the return value from
+// such a function is unused:
+//
+//   ABSL_MUST_USE_RESULT Sprocket* AllocateSprocket();
+//   AllocateSprocket();  // Triggers a warning.
+//
+// When annotating a class, it is equivalent to annotating every function which
+// returns an instance.
+//
+//   class ABSL_MUST_USE_RESULT Sprocket {};
+//   Sprocket();  // Triggers a warning.
+//
+//   Sprocket MakeSprocket();
+//   MakeSprocket();  // Triggers a warning.
+//
+// Note that references and pointers are not instances:
+//
+//   Sprocket* SprocketPointer();
+//   SprocketPointer();  // Does *not* trigger a warning.
+//
+// ABSL_MUST_USE_RESULT allows using cast-to-void to suppress the unused result
+// warning. For that, warn_unused_result is used only for clang but not for gcc.
+// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66425
+//
+// Note: past advice was to place the macro after the argument list.
+#if ABSL_HAVE_ATTRIBUTE(nodiscard)
+#define ABSL_MUST_USE_RESULT [[nodiscard]]
+#elif defined(__clang__) && ABSL_HAVE_ATTRIBUTE(warn_unused_result)
+#define ABSL_MUST_USE_RESULT __attribute__((warn_unused_result))
+#else
+#define ABSL_MUST_USE_RESULT
+#endif
+
+// ABSL_ATTRIBUTE_HOT, ABSL_ATTRIBUTE_COLD
+//
+// Tells GCC that a function is hot or cold. GCC can use this information to
+// improve static analysis, i.e. a conditional branch to a cold function
+// is likely to be not-taken.
+// This annotation is used for function declarations.
+//
+// Example:
+//
+//   int foo() ABSL_ATTRIBUTE_HOT;
+#if ABSL_HAVE_ATTRIBUTE(hot) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_HOT __attribute__((hot))
+#else
+#define ABSL_ATTRIBUTE_HOT
+#endif
+
+#if ABSL_HAVE_ATTRIBUTE(cold) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_COLD __attribute__((cold))
+#else
+#define ABSL_ATTRIBUTE_COLD
+#endif
+
+// ABSL_XRAY_ALWAYS_INSTRUMENT, ABSL_XRAY_NEVER_INSTRUMENT, ABSL_XRAY_LOG_ARGS
+//
+// We define the ABSL_XRAY_ALWAYS_INSTRUMENT and ABSL_XRAY_NEVER_INSTRUMENT
+// macro used as an attribute to mark functions that must always or never be
+// instrumented by XRay. Currently, this is only supported in Clang/LLVM.
+//
+// For reference on the LLVM XRay instrumentation, see
+// http://llvm.org/docs/XRay.html.
+//
+// A function with the XRAY_ALWAYS_INSTRUMENT macro attribute in its declaration
+// will always get the XRay instrumentation sleds. These sleds may introduce
+// some binary size and runtime overhead and must be used sparingly.
+//
+// These attributes only take effect when the following conditions are met:
+//
+//   * The file/target is built in at least C++11 mode, with a Clang compiler
+//     that supports XRay attributes.
+//   * The file/target is built with the -fxray-instrument flag set for the
+//     Clang/LLVM compiler.
+//   * The function is defined in the translation unit (the compiler honors the
+//     attribute in either the definition or the declaration, and must match).
+//
+// There are cases when, even when building with XRay instrumentation, users
+// might want to control specifically which functions are instrumented for a
+// particular build using special-case lists provided to the compiler. These
+// special case lists are provided to Clang via the
+// -fxray-always-instrument=... and -fxray-never-instrument=... flags. The
+// attributes in source take precedence over these special-case lists.
+//
+// To disable the XRay attributes at build-time, users may define
+// ABSL_NO_XRAY_ATTRIBUTES. Do NOT define ABSL_NO_XRAY_ATTRIBUTES on specific
+// packages/targets, as this may lead to conflicting definitions of functions at
+// link-time.
+//
+// XRay isn't currently supported on Android:
+// https://github.com/android/ndk/issues/368
+#if ABSL_HAVE_CPP_ATTRIBUTE(clang::xray_always_instrument) && \
+    !defined(ABSL_NO_XRAY_ATTRIBUTES) && !defined(__ANDROID__)
+#define ABSL_XRAY_ALWAYS_INSTRUMENT [[clang::xray_always_instrument]]
+#define ABSL_XRAY_NEVER_INSTRUMENT [[clang::xray_never_instrument]]
+#if ABSL_HAVE_CPP_ATTRIBUTE(clang::xray_log_args)
+#define ABSL_XRAY_LOG_ARGS(N) \
+    [[clang::xray_always_instrument, clang::xray_log_args(N)]]
+#else
+#define ABSL_XRAY_LOG_ARGS(N) [[clang::xray_always_instrument]]
+#endif
+#else
+#define ABSL_XRAY_ALWAYS_INSTRUMENT
+#define ABSL_XRAY_NEVER_INSTRUMENT
+#define ABSL_XRAY_LOG_ARGS(N)
+#endif
+
+// ABSL_ATTRIBUTE_REINITIALIZES
+//
+// Indicates that a member function reinitializes the entire object to a known
+// state, independent of the previous state of the object.
+//
+// The clang-tidy check bugprone-use-after-move allows member functions marked
+// with this attribute to be called on objects that have been moved from;
+// without the attribute, this would result in a use-after-move warning.
+#if ABSL_HAVE_CPP_ATTRIBUTE(clang::reinitializes)
+#define ABSL_ATTRIBUTE_REINITIALIZES [[clang::reinitializes]]
+#else
+#define ABSL_ATTRIBUTE_REINITIALIZES
+#endif
+
+// -----------------------------------------------------------------------------
+// Variable Attributes
+// -----------------------------------------------------------------------------
+
+// ABSL_ATTRIBUTE_UNUSED
+//
+// Prevents the compiler from complaining about variables that appear unused.
+#if ABSL_HAVE_ATTRIBUTE(unused) || (defined(__GNUC__) && !defined(__clang__))
+#undef ABSL_ATTRIBUTE_UNUSED
+#define ABSL_ATTRIBUTE_UNUSED __attribute__((__unused__))
+#else
+#define ABSL_ATTRIBUTE_UNUSED
+#endif
+
+// ABSL_ATTRIBUTE_INITIAL_EXEC
+//
+// Tells the compiler to use "initial-exec" mode for a thread-local variable.
+// See http://people.redhat.com/drepper/tls.pdf for the gory details.
+#if ABSL_HAVE_ATTRIBUTE(tls_model) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_INITIAL_EXEC __attribute__((tls_model("initial-exec")))
+#else
+#define ABSL_ATTRIBUTE_INITIAL_EXEC
+#endif
+
+// ABSL_ATTRIBUTE_PACKED
+//
+// Instructs the compiler not to use natural alignment for a tagged data
+// structure, but instead to reduce its alignment to 1. This attribute can
+// either be applied to members of a structure or to a structure in its
+// entirety. Applying this attribute (judiciously) to a structure in its
+// entirety to optimize the memory footprint of very commonly-used structs is
+// fine. Do not apply this attribute to a structure in its entirety if the
+// purpose is to control the offsets of the members in the structure. Instead,
+// apply this attribute only to structure members that need it.
+//
+// When applying ABSL_ATTRIBUTE_PACKED only to specific structure members the
+// natural alignment of structure members not annotated is preserved. Aligned
+// member accesses are faster than non-aligned member accesses even if the
+// targeted microprocessor supports non-aligned accesses.
+#if ABSL_HAVE_ATTRIBUTE(packed) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_PACKED __attribute__((__packed__))
+#else
+#define ABSL_ATTRIBUTE_PACKED
+#endif
+
+// ABSL_ATTRIBUTE_FUNC_ALIGN
+//
+// Tells the compiler to align the function start at least to certain
+// alignment boundary
+#if ABSL_HAVE_ATTRIBUTE(aligned) || (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_ATTRIBUTE_FUNC_ALIGN(bytes) __attribute__((aligned(bytes)))
+#else
+#define ABSL_ATTRIBUTE_FUNC_ALIGN(bytes)
+#endif
+
+// ABSL_FALLTHROUGH_INTENDED
+//
+// Annotates implicit fall-through between switch labels, allowing a case to
+// indicate intentional fallthrough and turn off warnings about any lack of a
+// `break` statement. The ABSL_FALLTHROUGH_INTENDED macro should be followed by
+// a semicolon and can be used in most places where `break` can, provided that
+// no statements exist between it and the next switch label.
+//
+// Example:
+//
+//  switch (x) {
+//    case 40:
+//    case 41:
+//      if (truth_is_out_there) {
+//        ++x;
+//        ABSL_FALLTHROUGH_INTENDED;  // Use instead of/along with annotations
+//                                    // in comments
+//      } else {
+//        return x;
+//      }
+//    case 42:
+//      ...
+//
+// Notes: when compiled with clang in C++11 mode, the ABSL_FALLTHROUGH_INTENDED
+// macro is expanded to the [[clang::fallthrough]] attribute, which is analysed
+// when  performing switch labels fall-through diagnostic
+// (`-Wimplicit-fallthrough`). See clang documentation on language extensions
+// for details:
+// https://clang.llvm.org/docs/AttributeReference.html#fallthrough-clang-fallthrough
+//
+// When used with unsupported compilers, the ABSL_FALLTHROUGH_INTENDED macro
+// has no effect on diagnostics. In any case this macro has no effect on runtime
+// behavior and performance of code.
+#ifdef ABSL_FALLTHROUGH_INTENDED
+#error "ABSL_FALLTHROUGH_INTENDED should not be defined."
+#endif
+
+// TODO(zhangxy): Use c++17 standard [[fallthrough]] macro, when supported.
+#if defined(__clang__) && defined(__has_warning)
+#if __has_feature(cxx_attributes) && __has_warning("-Wimplicit-fallthrough")
+#define ABSL_FALLTHROUGH_INTENDED [[clang::fallthrough]]
+#endif
+#elif defined(__GNUC__) && __GNUC__ >= 7
+#define ABSL_FALLTHROUGH_INTENDED [[gnu::fallthrough]]
+#endif
+
+#ifndef ABSL_FALLTHROUGH_INTENDED
+#define ABSL_FALLTHROUGH_INTENDED \
+  do {                            \
+  } while (0)
+#endif
+
+// ABSL_DEPRECATED()
+//
+// Marks a deprecated class, struct, enum, function, method and variable
+// declarations. The macro argument is used as a custom diagnostic message (e.g.
+// suggestion of a better alternative).
+//
+// Examples:
+//
+//   class ABSL_DEPRECATED("Use Bar instead") Foo {...};
+//
+//   ABSL_DEPRECATED("Use Baz() instead") void Bar() {...}
+//
+//   template <typename T>
+//   ABSL_DEPRECATED("Use DoThat() instead")
+//   void DoThis();
+//
+// Every usage of a deprecated entity will trigger a warning when compiled with
+// clang's `-Wdeprecated-declarations` option. This option is turned off by
+// default, but the warnings will be reported by clang-tidy.
+#if defined(__clang__) && __cplusplus >= 201103L
+#define ABSL_DEPRECATED(message) __attribute__((deprecated(message)))
+#endif
+
+#ifndef ABSL_DEPRECATED
+#define ABSL_DEPRECATED(message)
+#endif
+
+// ABSL_CONST_INIT
+//
+// A variable declaration annotated with the `ABSL_CONST_INIT` attribute will
+// not compile (on supported platforms) unless the variable has a constant
+// initializer. This is useful for variables with static and thread storage
+// duration, because it guarantees that they will not suffer from the so-called
+// "static init order fiasco".  Prefer to put this attribute on the most visible
+// declaration of the variable, if there's more than one, because code that
+// accesses the variable can then use the attribute for optimization.
+//
+// Example:
+//
+//   class MyClass {
+//    public:
+//     ABSL_CONST_INIT static MyType my_var;
+//   };
+//
+//   MyType MyClass::my_var = MakeMyType(...);
+//
+// Note that this attribute is redundant if the variable is declared constexpr.
+#if ABSL_HAVE_CPP_ATTRIBUTE(clang::require_constant_initialization)
+#define ABSL_CONST_INIT [[clang::require_constant_initialization]]
+#else
+#define ABSL_CONST_INIT
+#endif  // ABSL_HAVE_CPP_ATTRIBUTE(clang::require_constant_initialization)
+
+#endif  // ABSL_BASE_ATTRIBUTES_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/config.h b/lib/clickhouse-cpp/contrib/absl/absl/base/config.h
new file mode 100644
index 0000000..c1d0494
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/config.h
@@ -0,0 +1,714 @@
+//
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// -----------------------------------------------------------------------------
+// File: config.h
+// -----------------------------------------------------------------------------
+//
+// This header file defines a set of macros for checking the presence of
+// important compiler and platform features. Such macros can be used to
+// produce portable code by parameterizing compilation based on the presence or
+// lack of a given feature.
+//
+// We define a "feature" as some interface we wish to program to: for example,
+// a library function or system call. A value of `1` indicates support for
+// that feature; any other value indicates the feature support is undefined.
+//
+// Example:
+//
+// Suppose a programmer wants to write a program that uses the 'mmap()' system
+// call. The Abseil macro for that feature (`ABSL_HAVE_MMAP`) allows you to
+// selectively include the `mmap.h` header and bracket code using that feature
+// in the macro:
+//
+//   #include "absl/base/config.h"
+//
+//   #ifdef ABSL_HAVE_MMAP
+//   #include "sys/mman.h"
+//   #endif  //ABSL_HAVE_MMAP
+//
+//   ...
+//   #ifdef ABSL_HAVE_MMAP
+//   void *ptr = mmap(...);
+//   ...
+//   #endif  // ABSL_HAVE_MMAP
+
+#ifndef ABSL_BASE_CONFIG_H_
+#define ABSL_BASE_CONFIG_H_
+
+// Included for the __GLIBC__ macro (or similar macros on other systems).
+#include <limits.h>
+
+#ifdef __cplusplus
+// Included for __GLIBCXX__, _LIBCPP_VERSION
+#include <cstddef>
+#endif  // __cplusplus
+
+#if defined(__APPLE__)
+// Included for TARGET_OS_IPHONE, __IPHONE_OS_VERSION_MIN_REQUIRED,
+// __IPHONE_8_0.
+#include <Availability.h>
+#include <TargetConditionals.h>
+#endif
+
+#include "absl/base/options.h"
+#include "absl/base/policy_checks.h"
+
+// Helper macro to convert a CPP variable to a string literal.
+#define ABSL_INTERNAL_DO_TOKEN_STR(x) #x
+#define ABSL_INTERNAL_TOKEN_STR(x) ABSL_INTERNAL_DO_TOKEN_STR(x)
+
+// -----------------------------------------------------------------------------
+// Abseil namespace annotations
+// -----------------------------------------------------------------------------
+
+// ABSL_NAMESPACE_BEGIN/ABSL_NAMESPACE_END
+//
+// An annotation placed at the beginning/end of each `namespace absl` scope.
+// This is used to inject an inline namespace.
+//
+// The proper way to write Abseil code in the `absl` namespace is:
+//
+// namespace absl {
+// ABSL_NAMESPACE_BEGIN
+//
+// void Foo();  // absl::Foo().
+//
+// ABSL_NAMESPACE_END
+// }  // namespace absl
+//
+// Users of Abseil should not use these macros, because users of Abseil should
+// not write `namespace absl {` in their own code for any reason.  (Abseil does
+// not support forward declarations of its own types, nor does it support
+// user-provided specialization of Abseil templates.  Code that violates these
+// rules may be broken without warning.)
+#if !defined(ABSL_OPTION_USE_INLINE_NAMESPACE) || \
+    !defined(ABSL_OPTION_INLINE_NAMESPACE_NAME)
+#error options.h is misconfigured.
+#endif
+
+// Check that ABSL_OPTION_INLINE_NAMESPACE_NAME is neither "head" nor ""
+#if defined(__cplusplus) && ABSL_OPTION_USE_INLINE_NAMESPACE == 1
+
+#define ABSL_INTERNAL_INLINE_NAMESPACE_STR \
+  ABSL_INTERNAL_TOKEN_STR(ABSL_OPTION_INLINE_NAMESPACE_NAME)
+
+static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != '\0',
+              "options.h misconfigured: ABSL_OPTION_INLINE_NAMESPACE_NAME must "
+              "not be empty.");
+static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != 'h' ||
+                  ABSL_INTERNAL_INLINE_NAMESPACE_STR[1] != 'e' ||
+                  ABSL_INTERNAL_INLINE_NAMESPACE_STR[2] != 'a' ||
+                  ABSL_INTERNAL_INLINE_NAMESPACE_STR[3] != 'd' ||
+                  ABSL_INTERNAL_INLINE_NAMESPACE_STR[4] != '\0',
+              "options.h misconfigured: ABSL_OPTION_INLINE_NAMESPACE_NAME must "
+              "be changed to a new, unique identifier name.");
+
+#endif
+
+#if ABSL_OPTION_USE_INLINE_NAMESPACE == 0
+#define ABSL_NAMESPACE_BEGIN
+#define ABSL_NAMESPACE_END
+#elif ABSL_OPTION_USE_INLINE_NAMESPACE == 1
+#define ABSL_NAMESPACE_BEGIN \
+  inline namespace ABSL_OPTION_INLINE_NAMESPACE_NAME {
+#define ABSL_NAMESPACE_END }
+#else
+#error options.h is misconfigured.
+#endif
+
+// -----------------------------------------------------------------------------
+// Compiler Feature Checks
+// -----------------------------------------------------------------------------
+
+// ABSL_HAVE_BUILTIN()
+//
+// Checks whether the compiler supports a Clang Feature Checking Macro, and if
+// so, checks whether it supports the provided builtin function "x" where x
+// is one of the functions noted in
+// https://clang.llvm.org/docs/LanguageExtensions.html
+//
+// Note: Use this macro to avoid an extra level of #ifdef __has_builtin check.
+// http://releases.llvm.org/3.3/tools/clang/docs/LanguageExtensions.html
+#ifdef __has_builtin
+#define ABSL_HAVE_BUILTIN(x) __has_builtin(x)
+#else
+#define ABSL_HAVE_BUILTIN(x) 0
+#endif
+
+#if defined(__is_identifier)
+#define ABSL_INTERNAL_HAS_KEYWORD(x) !(__is_identifier(x))
+#else
+#define ABSL_INTERNAL_HAS_KEYWORD(x) 0
+#endif
+
+#ifdef __has_feature
+#define ABSL_HAVE_FEATURE(f) __has_feature(f)
+#else
+#define ABSL_HAVE_FEATURE(f) 0
+#endif
+
+// ABSL_HAVE_TLS is defined to 1 when __thread should be supported.
+// We assume __thread is supported on Linux when compiled with Clang or compiled
+// against libstdc++ with _GLIBCXX_HAVE_TLS defined.
+#ifdef ABSL_HAVE_TLS
+#error ABSL_HAVE_TLS cannot be directly set
+#elif defined(__linux__) && (defined(__clang__) || defined(_GLIBCXX_HAVE_TLS))
+#define ABSL_HAVE_TLS 1
+#endif
+
+// ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE
+//
+// Checks whether `std::is_trivially_destructible<T>` is supported.
+//
+// Notes: All supported compilers using libc++ support this feature, as does
+// gcc >= 4.8.1 using libstdc++, and Visual Studio.
+#ifdef ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE
+#error ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE cannot be directly set
+#elif defined(_LIBCPP_VERSION) ||                                        \
+    (!defined(__clang__) && defined(__GNUC__) && defined(__GLIBCXX__) && \
+     (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8))) ||        \
+    defined(_MSC_VER)
+#define ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE 1
+#endif
+
+// ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE
+//
+// Checks whether `std::is_trivially_default_constructible<T>` and
+// `std::is_trivially_copy_constructible<T>` are supported.
+
+// ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE
+//
+// Checks whether `std::is_trivially_copy_assignable<T>` is supported.
+
+// Notes: Clang with libc++ supports these features, as does gcc >= 5.1 with
+// either libc++ or libstdc++, and Visual Studio (but not NVCC).
+#if defined(ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE)
+#error ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE cannot be directly set
+#elif defined(ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE)
+#error ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE cannot directly set
+#elif (defined(__clang__) && defined(_LIBCPP_VERSION)) ||        \
+    (!defined(__clang__) && defined(__GNUC__) &&                 \
+     (__GNUC__ > 7 || (__GNUC__ == 7 && __GNUC_MINOR__ >= 4)) && \
+     (defined(_LIBCPP_VERSION) || defined(__GLIBCXX__))) ||      \
+    (defined(_MSC_VER) && !defined(__NVCC__))
+#define ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE 1
+#define ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE 1
+#endif
+
+// ABSL_HAVE_SOURCE_LOCATION_CURRENT
+//
+// Indicates whether `absl::SourceLocation::current()` will return useful
+// information in some contexts.
+#ifndef ABSL_HAVE_SOURCE_LOCATION_CURRENT
+#if ABSL_INTERNAL_HAS_KEYWORD(__builtin_LINE) && \
+    ABSL_INTERNAL_HAS_KEYWORD(__builtin_FILE)
+#define ABSL_HAVE_SOURCE_LOCATION_CURRENT 1
+#endif
+#endif
+
+// ABSL_HAVE_THREAD_LOCAL
+//
+// Checks whether C++11's `thread_local` storage duration specifier is
+// supported.
+#ifdef ABSL_HAVE_THREAD_LOCAL
+#error ABSL_HAVE_THREAD_LOCAL cannot be directly set
+#elif defined(__APPLE__)
+// Notes:
+// * Xcode's clang did not support `thread_local` until version 8, and
+//   even then not for all iOS < 9.0.
+// * Xcode 9.3 started disallowing `thread_local` for 32-bit iOS simulator
+//   targeting iOS 9.x.
+// * Xcode 10 moves the deployment target check for iOS < 9.0 to link time
+//   making ABSL_HAVE_FEATURE unreliable there.
+//
+#if ABSL_HAVE_FEATURE(cxx_thread_local) && \
+    !(TARGET_OS_IPHONE && __IPHONE_OS_VERSION_MIN_REQUIRED < __IPHONE_9_0)
+#define ABSL_HAVE_THREAD_LOCAL 1
+#endif
+#else  // !defined(__APPLE__)
+#define ABSL_HAVE_THREAD_LOCAL 1
+#endif
+
+// There are platforms for which TLS should not be used even though the compiler
+// makes it seem like it's supported (Android NDK < r12b for example).
+// This is primarily because of linker problems and toolchain misconfiguration:
+// Abseil does not intend to support this indefinitely. Currently, the newest
+// toolchain that we intend to support that requires this behavior is the
+// r11 NDK - allowing for a 5 year support window on that means this option
+// is likely to be removed around June of 2021.
+// TLS isn't supported until NDK r12b per
+// https://developer.android.com/ndk/downloads/revision_history.html
+// Since NDK r16, `__NDK_MAJOR__` and `__NDK_MINOR__` are defined in
+// <android/ndk-version.h>. For NDK < r16, users should define these macros,
+// e.g. `-D__NDK_MAJOR__=11 -D__NKD_MINOR__=0` for NDK r11.
+#if defined(__ANDROID__) && defined(__clang__)
+#if __has_include(<android/ndk-version.h>)
+#include <android/ndk-version.h>
+#endif  // __has_include(<android/ndk-version.h>)
+#if defined(__ANDROID__) && defined(__clang__) && defined(__NDK_MAJOR__) && \
+    defined(__NDK_MINOR__) &&                                               \
+    ((__NDK_MAJOR__ < 12) || ((__NDK_MAJOR__ == 12) && (__NDK_MINOR__ < 1)))
+#undef ABSL_HAVE_TLS
+#undef ABSL_HAVE_THREAD_LOCAL
+#endif
+#endif  // defined(__ANDROID__) && defined(__clang__)
+
+// ABSL_HAVE_INTRINSIC_INT128
+//
+// Checks whether the __int128 compiler extension for a 128-bit integral type is
+// supported.
+//
+// Note: __SIZEOF_INT128__ is defined by Clang and GCC when __int128 is
+// supported, but we avoid using it in certain cases:
+// * On Clang:
+//   * Building using Clang for Windows, where the Clang runtime library has
+//     128-bit support only on LP64 architectures, but Windows is LLP64.
+// * On Nvidia's nvcc:
+//   * nvcc also defines __GNUC__ and __SIZEOF_INT128__, but not all versions
+//     actually support __int128.
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+#error ABSL_HAVE_INTRINSIC_INT128 cannot be directly set
+#elif defined(__SIZEOF_INT128__)
+#if (defined(__clang__) && !defined(_WIN32)) || \
+    (defined(__CUDACC__) && __CUDACC_VER_MAJOR__ >= 9) ||                \
+    (defined(__GNUC__) && !defined(__clang__) && !defined(__CUDACC__))
+#define ABSL_HAVE_INTRINSIC_INT128 1
+#elif defined(__CUDACC__)
+// __CUDACC_VER__ is a full version number before CUDA 9, and is defined to a
+// string explaining that it has been removed starting with CUDA 9. We use
+// nested #ifs because there is no short-circuiting in the preprocessor.
+// NOTE: `__CUDACC__` could be undefined while `__CUDACC_VER__` is defined.
+#if __CUDACC_VER__ >= 70000
+#define ABSL_HAVE_INTRINSIC_INT128 1
+#endif  // __CUDACC_VER__ >= 70000
+#endif  // defined(__CUDACC__)
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+// ABSL_HAVE_EXCEPTIONS
+//
+// Checks whether the compiler both supports and enables exceptions. Many
+// compilers support a "no exceptions" mode that disables exceptions.
+//
+// Generally, when ABSL_HAVE_EXCEPTIONS is not defined:
+//
+// * Code using `throw` and `try` may not compile.
+// * The `noexcept` specifier will still compile and behave as normal.
+// * The `noexcept` operator may still return `false`.
+//
+// For further details, consult the compiler's documentation.
+#ifdef ABSL_HAVE_EXCEPTIONS
+#error ABSL_HAVE_EXCEPTIONS cannot be directly set.
+
+#elif defined(__clang__)
+
+#if __clang_major__ > 3 || (__clang_major__ == 3 && __clang_minor__ >= 6)
+// Clang >= 3.6
+#if ABSL_HAVE_FEATURE(cxx_exceptions)
+#define ABSL_HAVE_EXCEPTIONS 1
+#endif  // ABSL_HAVE_FEATURE(cxx_exceptions)
+#else
+// Clang < 3.6
+// http://releases.llvm.org/3.6.0/tools/clang/docs/ReleaseNotes.html#the-exceptions-macro
+#if defined(__EXCEPTIONS) && ABSL_HAVE_FEATURE(cxx_exceptions)
+#define ABSL_HAVE_EXCEPTIONS 1
+#endif  // defined(__EXCEPTIONS) && ABSL_HAVE_FEATURE(cxx_exceptions)
+#endif  // __clang_major__ > 3 || (__clang_major__ == 3 && __clang_minor__ >= 6)
+
+// Handle remaining special cases and default to exceptions being supported.
+#elif !(defined(__GNUC__) && (__GNUC__ < 5) && !defined(__EXCEPTIONS)) &&    \
+    !(defined(__GNUC__) && (__GNUC__ >= 5) && !defined(__cpp_exceptions)) && \
+    !(defined(_MSC_VER) && !defined(_CPPUNWIND))
+#define ABSL_HAVE_EXCEPTIONS 1
+#endif
+
+// -----------------------------------------------------------------------------
+// Platform Feature Checks
+// -----------------------------------------------------------------------------
+
+// Currently supported operating systems and associated preprocessor
+// symbols:
+//
+//   Linux and Linux-derived           __linux__
+//   Android                           __ANDROID__ (implies __linux__)
+//   Linux (non-Android)               __linux__ && !__ANDROID__
+//   Darwin (macOS and iOS)            __APPLE__
+//   Akaros (http://akaros.org)        __ros__
+//   Windows                           _WIN32
+//   NaCL                              __native_client__
+//   AsmJS                             __asmjs__
+//   WebAssembly                       __wasm__
+//   Fuchsia                           __Fuchsia__
+//
+// Note that since Android defines both __ANDROID__ and __linux__, one
+// may probe for either Linux or Android by simply testing for __linux__.
+
+// ABSL_HAVE_MMAP
+//
+// Checks whether the platform has an mmap(2) implementation as defined in
+// POSIX.1-2001.
+#ifdef ABSL_HAVE_MMAP
+#error ABSL_HAVE_MMAP cannot be directly set
+#elif defined(__linux__) || defined(__APPLE__) || defined(__FreeBSD__) ||   \
+    defined(__ros__) || defined(__native_client__) || defined(__asmjs__) || \
+    defined(__wasm__) || defined(__Fuchsia__) || defined(__sun) || \
+    defined(__ASYLO__)
+#define ABSL_HAVE_MMAP 1
+#endif
+
+// ABSL_HAVE_PTHREAD_GETSCHEDPARAM
+//
+// Checks whether the platform implements the pthread_(get|set)schedparam(3)
+// functions as defined in POSIX.1-2001.
+#ifdef ABSL_HAVE_PTHREAD_GETSCHEDPARAM
+#error ABSL_HAVE_PTHREAD_GETSCHEDPARAM cannot be directly set
+#elif defined(__linux__) || defined(__APPLE__) || defined(__FreeBSD__) || \
+    defined(__ros__)
+#define ABSL_HAVE_PTHREAD_GETSCHEDPARAM 1
+#endif
+
+// ABSL_HAVE_SCHED_YIELD
+//
+// Checks whether the platform implements sched_yield(2) as defined in
+// POSIX.1-2001.
+#ifdef ABSL_HAVE_SCHED_YIELD
+#error ABSL_HAVE_SCHED_YIELD cannot be directly set
+#elif defined(__linux__) || defined(__ros__) || defined(__native_client__)
+#define ABSL_HAVE_SCHED_YIELD 1
+#endif
+
+// ABSL_HAVE_SEMAPHORE_H
+//
+// Checks whether the platform supports the <semaphore.h> header and sem_init(3)
+// family of functions as standardized in POSIX.1-2001.
+//
+// Note: While Apple provides <semaphore.h> for both iOS and macOS, it is
+// explicitly deprecated and will cause build failures if enabled for those
+// platforms.  We side-step the issue by not defining it here for Apple
+// platforms.
+#ifdef ABSL_HAVE_SEMAPHORE_H
+#error ABSL_HAVE_SEMAPHORE_H cannot be directly set
+#elif defined(__linux__) || defined(__ros__)
+#define ABSL_HAVE_SEMAPHORE_H 1
+#endif
+
+// ABSL_HAVE_ALARM
+//
+// Checks whether the platform supports the <signal.h> header and alarm(2)
+// function as standardized in POSIX.1-2001.
+#ifdef ABSL_HAVE_ALARM
+#error ABSL_HAVE_ALARM cannot be directly set
+#elif defined(__GOOGLE_GRTE_VERSION__)
+// feature tests for Google's GRTE
+#define ABSL_HAVE_ALARM 1
+#elif defined(__GLIBC__)
+// feature test for glibc
+#define ABSL_HAVE_ALARM 1
+#elif defined(_MSC_VER)
+// feature tests for Microsoft's library
+#elif defined(__MINGW32__)
+// mingw32 doesn't provide alarm(2):
+// https://osdn.net/projects/mingw/scm/git/mingw-org-wsl/blobs/5.2-trunk/mingwrt/include/unistd.h
+// mingw-w64 provides a no-op implementation:
+// https://sourceforge.net/p/mingw-w64/mingw-w64/ci/master/tree/mingw-w64-crt/misc/alarm.c
+#elif defined(__EMSCRIPTEN__)
+// emscripten doesn't support signals
+#elif defined(__Fuchsia__)
+// Signals don't exist on fuchsia.
+#elif defined(__native_client__)
+#else
+// other standard libraries
+#define ABSL_HAVE_ALARM 1
+#endif
+
+// ABSL_IS_LITTLE_ENDIAN
+// ABSL_IS_BIG_ENDIAN
+//
+// Checks the endianness of the platform.
+//
+// Notes: uses the built in endian macros provided by GCC (since 4.6) and
+// Clang (since 3.2); see
+// https://gcc.gnu.org/onlinedocs/cpp/Common-Predefined-Macros.html.
+// Otherwise, if _WIN32, assume little endian. Otherwise, bail with an error.
+#if defined(ABSL_IS_BIG_ENDIAN)
+#error "ABSL_IS_BIG_ENDIAN cannot be directly set."
+#endif
+#if defined(ABSL_IS_LITTLE_ENDIAN)
+#error "ABSL_IS_LITTLE_ENDIAN cannot be directly set."
+#endif
+
+#if (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && \
+     __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+#define ABSL_IS_LITTLE_ENDIAN 1
+#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && \
+    __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+#define ABSL_IS_BIG_ENDIAN 1
+#elif defined(_WIN32)
+#define ABSL_IS_LITTLE_ENDIAN 1
+#else
+#error "absl endian detection needs to be set up for your compiler"
+#endif
+
+// macOS 10.13 and iOS 10.11 don't let you use <any>, <optional>, or <variant>
+// even though the headers exist and are publicly noted to work.  See
+// https://github.com/abseil/abseil-cpp/issues/207 and
+// https://developer.apple.com/documentation/xcode_release_notes/xcode_10_release_notes
+// libc++ spells out the availability requirements in the file
+// llvm-project/libcxx/include/__config via the #define
+// _LIBCPP_AVAILABILITY_BAD_OPTIONAL_ACCESS.
+#if defined(__APPLE__) && defined(_LIBCPP_VERSION) && \
+  ((defined(__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__) && \
+   __ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ < 101400) || \
+  (defined(__ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__) && \
+   __ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__ < 120000) || \
+  (defined(__ENVIRONMENT_WATCH_OS_VERSION_MIN_REQUIRED__) && \
+   __ENVIRONMENT_WATCH_OS_VERSION_MIN_REQUIRED__ < 50000) || \
+  (defined(__ENVIRONMENT_TV_OS_VERSION_MIN_REQUIRED__) && \
+   __ENVIRONMENT_TV_OS_VERSION_MIN_REQUIRED__ < 120000))
+#define ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE 1
+#else
+#define ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE 0
+#endif
+
+// ABSL_HAVE_STD_ANY
+//
+// Checks whether C++17 std::any is available by checking whether <any> exists.
+#ifdef ABSL_HAVE_STD_ANY
+#error "ABSL_HAVE_STD_ANY cannot be directly set."
+#endif
+
+#ifdef __has_include
+#if __has_include(<any>) && __cplusplus >= 201703L && \
+    !ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE
+#define ABSL_HAVE_STD_ANY 1
+#endif
+#endif
+
+// ABSL_HAVE_STD_OPTIONAL
+//
+// Checks whether C++17 std::optional is available.
+#ifdef ABSL_HAVE_STD_OPTIONAL
+#error "ABSL_HAVE_STD_OPTIONAL cannot be directly set."
+#endif
+
+#ifdef __has_include
+#if __has_include(<optional>) && __cplusplus >= 201703L && \
+    !ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE
+#define ABSL_HAVE_STD_OPTIONAL 1
+#endif
+#endif
+
+// ABSL_HAVE_STD_VARIANT
+//
+// Checks whether C++17 std::variant is available.
+#ifdef ABSL_HAVE_STD_VARIANT
+#error "ABSL_HAVE_STD_VARIANT cannot be directly set."
+#endif
+
+#ifdef __has_include
+#if __has_include(<variant>) && __cplusplus >= 201703L && \
+    !ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE
+#define ABSL_HAVE_STD_VARIANT 1
+#endif
+#endif
+
+// ABSL_HAVE_STD_STRING_VIEW
+//
+// Checks whether C++17 std::string_view is available.
+#ifdef ABSL_HAVE_STD_STRING_VIEW
+#error "ABSL_HAVE_STD_STRING_VIEW cannot be directly set."
+#endif
+
+#ifdef __has_include
+#if __has_include(<string_view>) && __cplusplus >= 201703L
+#define ABSL_HAVE_STD_STRING_VIEW 1
+#endif
+#endif
+
+// For MSVC, `__has_include` is supported in VS 2017 15.3, which is later than
+// the support for <optional>, <any>, <string_view>, <variant>. So we use
+// _MSC_VER to check whether we have VS 2017 RTM (when <optional>, <any>,
+// <string_view>, <variant> is implemented) or higher. Also, `__cplusplus` is
+// not correctly set by MSVC, so we use `_MSVC_LANG` to check the language
+// version.
+// TODO(zhangxy): fix tests before enabling aliasing for `std::any`.
+#if defined(_MSC_VER) && _MSC_VER >= 1910 && \
+    ((defined(_MSVC_LANG) && _MSVC_LANG > 201402) || __cplusplus > 201402)
+// #define ABSL_HAVE_STD_ANY 1
+#define ABSL_HAVE_STD_OPTIONAL 1
+#define ABSL_HAVE_STD_VARIANT 1
+#define ABSL_HAVE_STD_STRING_VIEW 1
+#endif
+
+// ABSL_USES_STD_ANY
+//
+// Indicates whether absl::any is an alias for std::any.
+#if !defined(ABSL_OPTION_USE_STD_ANY)
+#error options.h is misconfigured.
+#elif ABSL_OPTION_USE_STD_ANY == 0 || \
+    (ABSL_OPTION_USE_STD_ANY == 2 && !defined(ABSL_HAVE_STD_ANY))
+#undef ABSL_USES_STD_ANY
+#elif ABSL_OPTION_USE_STD_ANY == 1 || \
+    (ABSL_OPTION_USE_STD_ANY == 2 && defined(ABSL_HAVE_STD_ANY))
+#define ABSL_USES_STD_ANY 1
+#else
+#error options.h is misconfigured.
+#endif
+
+// ABSL_USES_STD_OPTIONAL
+//
+// Indicates whether absl::optional is an alias for std::optional.
+#if !defined(ABSL_OPTION_USE_STD_OPTIONAL)
+#error options.h is misconfigured.
+#elif ABSL_OPTION_USE_STD_OPTIONAL == 0 || \
+    (ABSL_OPTION_USE_STD_OPTIONAL == 2 && !defined(ABSL_HAVE_STD_OPTIONAL))
+#undef ABSL_USES_STD_OPTIONAL
+#elif ABSL_OPTION_USE_STD_OPTIONAL == 1 || \
+    (ABSL_OPTION_USE_STD_OPTIONAL == 2 && defined(ABSL_HAVE_STD_OPTIONAL))
+#define ABSL_USES_STD_OPTIONAL 1
+#else
+#error options.h is misconfigured.
+#endif
+
+// ABSL_USES_STD_VARIANT
+//
+// Indicates whether absl::variant is an alias for std::variant.
+#if !defined(ABSL_OPTION_USE_STD_VARIANT)
+#error options.h is misconfigured.
+#elif ABSL_OPTION_USE_STD_VARIANT == 0 || \
+    (ABSL_OPTION_USE_STD_VARIANT == 2 && !defined(ABSL_HAVE_STD_VARIANT))
+#undef ABSL_USES_STD_VARIANT
+#elif ABSL_OPTION_USE_STD_VARIANT == 1 || \
+    (ABSL_OPTION_USE_STD_VARIANT == 2 && defined(ABSL_HAVE_STD_VARIANT))
+#define ABSL_USES_STD_VARIANT 1
+#else
+#error options.h is misconfigured.
+#endif
+
+// ABSL_USES_STD_STRING_VIEW
+//
+// Indicates whether absl::string_view is an alias for std::string_view.
+#if !defined(ABSL_OPTION_USE_STD_STRING_VIEW)
+#error options.h is misconfigured.
+#elif ABSL_OPTION_USE_STD_STRING_VIEW == 0 || \
+    (ABSL_OPTION_USE_STD_STRING_VIEW == 2 &&  \
+     !defined(ABSL_HAVE_STD_STRING_VIEW))
+#undef ABSL_USES_STD_STRING_VIEW
+#elif ABSL_OPTION_USE_STD_STRING_VIEW == 1 || \
+    (ABSL_OPTION_USE_STD_STRING_VIEW == 2 &&  \
+     defined(ABSL_HAVE_STD_STRING_VIEW))
+#define ABSL_USES_STD_STRING_VIEW 1
+#else
+#error options.h is misconfigured.
+#endif
+
+// In debug mode, MSVC 2017's std::variant throws a EXCEPTION_ACCESS_VIOLATION
+// SEH exception from emplace for variant<SomeStruct> when constructing the
+// struct can throw. This defeats some of variant_test and
+// variant_exception_safety_test.
+#if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_DEBUG)
+#define ABSL_INTERNAL_MSVC_2017_DBG_MODE
+#endif
+
+// ABSL_INTERNAL_MANGLED_NS
+// ABSL_INTERNAL_MANGLED_BACKREFERENCE
+//
+// Internal macros for building up mangled names in our internal fork of CCTZ.
+// This implementation detail is only needed and provided for the MSVC build.
+//
+// These macros both expand to string literals.  ABSL_INTERNAL_MANGLED_NS is
+// the mangled spelling of the `absl` namespace, and
+// ABSL_INTERNAL_MANGLED_BACKREFERENCE is a back-reference integer representing
+// the proper count to skip past the CCTZ fork namespace names.  (This number
+// is one larger when there is an inline namespace name to skip.)
+#if defined(_MSC_VER)
+#if ABSL_OPTION_USE_INLINE_NAMESPACE == 0
+#define ABSL_INTERNAL_MANGLED_NS "absl"
+#define ABSL_INTERNAL_MANGLED_BACKREFERENCE "5"
+#else
+#define ABSL_INTERNAL_MANGLED_NS \
+  ABSL_INTERNAL_TOKEN_STR(ABSL_OPTION_INLINE_NAMESPACE_NAME) "@absl"
+#define ABSL_INTERNAL_MANGLED_BACKREFERENCE "6"
+#endif
+#endif
+
+#undef ABSL_INTERNAL_HAS_KEYWORD
+
+// ABSL_DLL
+//
+// When building Abseil as a DLL, this macro expands to `__declspec(dllexport)`
+// so we can annotate symbols appropriately as being exported. When used in
+// headers consuming a DLL, this macro expands to `__declspec(dllimport)` so
+// that consumers know the symbol is defined inside the DLL. In all other cases,
+// the macro expands to nothing.
+#if defined(_MSC_VER)
+#if defined(ABSL_BUILD_DLL)
+#define ABSL_DLL __declspec(dllexport)
+#elif defined(ABSL_CONSUME_DLL)
+#define ABSL_DLL __declspec(dllimport)
+#else
+#define ABSL_DLL
+#endif
+#else
+#define ABSL_DLL
+#endif  // defined(_MSC_VER)
+
+// ABSL_HAVE_MEMORY_SANITIZER
+//
+// MemorySanitizer (MSan) is a detector of uninitialized reads. It consists of
+// a compiler instrumentation module and a run-time library.
+#ifdef ABSL_HAVE_MEMORY_SANITIZER
+#error "ABSL_HAVE_MEMORY_SANITIZER cannot be directly set."
+#elif defined(MEMORY_SANITIZER)
+// The MEMORY_SANITIZER macro is deprecated but we will continue to honor it
+// for now.
+#define ABSL_HAVE_MEMORY_SANITIZER 1
+#elif defined(__SANITIZE_MEMORY__)
+#define ABSL_HAVE_MEMORY_SANITIZER 1
+#elif !defined(__native_client__) && ABSL_HAVE_FEATURE(memory_sanitizer)
+#define ABSL_HAVE_MEMORY_SANITIZER 1
+#endif
+
+// ABSL_HAVE_THREAD_SANITIZER
+//
+// ThreadSanitizer (TSan) is a fast data race detector.
+#ifdef ABSL_HAVE_THREAD_SANITIZER
+#error "ABSL_HAVE_THREAD_SANITIZER cannot be directly set."
+#elif defined(THREAD_SANITIZER)
+// The THREAD_SANITIZER macro is deprecated but we will continue to honor it
+// for now.
+#define ABSL_HAVE_THREAD_SANITIZER 1
+#elif defined(__SANITIZE_THREAD__)
+#define ABSL_HAVE_THREAD_SANITIZER 1
+#elif ABSL_HAVE_FEATURE(thread_sanitizer)
+#define ABSL_HAVE_THREAD_SANITIZER 1
+#endif
+
+// ABSL_HAVE_ADDRESS_SANITIZER
+//
+// AddressSanitizer (ASan) is a fast memory error detector.
+#ifdef ABSL_HAVE_ADDRESS_SANITIZER
+#error "ABSL_HAVE_ADDRESS_SANITIZER cannot be directly set."
+#elif defined(ADDRESS_SANITIZER)
+// The ADDRESS_SANITIZER macro is deprecated but we will continue to honor it
+// for now.
+#define ABSL_HAVE_ADDRESS_SANITIZER 1
+#elif defined(__SANITIZE_ADDRESS__)
+#define ABSL_HAVE_ADDRESS_SANITIZER 1
+#elif ABSL_HAVE_FEATURE(address_sanitizer)
+#define ABSL_HAVE_ADDRESS_SANITIZER 1
+#endif
+
+#endif  // ABSL_BASE_CONFIG_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/internal/bits.h b/lib/clickhouse-cpp/contrib/absl/absl/base/internal/bits.h
new file mode 100644
index 0000000..81648e2
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/internal/bits.h
@@ -0,0 +1,219 @@
+// Copyright 2018 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef ABSL_BASE_INTERNAL_BITS_H_
+#define ABSL_BASE_INTERNAL_BITS_H_
+
+// This file contains bitwise ops which are implementation details of various
+// absl libraries.
+
+#include <cstdint>
+
+#include "absl/base/config.h"
+
+// Clang on Windows has __builtin_clzll; otherwise we need to use the
+// windows intrinsic functions.
+#if defined(_MSC_VER) && !defined(__clang__)
+#include <intrin.h>
+#if defined(_M_X64)
+#pragma intrinsic(_BitScanReverse64)
+#pragma intrinsic(_BitScanForward64)
+#endif
+#pragma intrinsic(_BitScanReverse)
+#pragma intrinsic(_BitScanForward)
+#endif
+
+#include "absl/base/attributes.h"
+
+#if defined(_MSC_VER) && !defined(__clang__)
+// We can achieve something similar to attribute((always_inline)) with MSVC by
+// using the __forceinline keyword, however this is not perfect. MSVC is
+// much less aggressive about inlining, and even with the __forceinline keyword.
+#define ABSL_BASE_INTERNAL_FORCEINLINE __forceinline
+#else
+// Use default attribute inline.
+#define ABSL_BASE_INTERNAL_FORCEINLINE inline ABSL_ATTRIBUTE_ALWAYS_INLINE
+#endif
+
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace base_internal {
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountLeadingZeros64Slow(uint64_t n) {
+  int zeroes = 60;
+  if (n >> 32) {
+    zeroes -= 32;
+    n >>= 32;
+  }
+  if (n >> 16) {
+    zeroes -= 16;
+    n >>= 16;
+  }
+  if (n >> 8) {
+    zeroes -= 8;
+    n >>= 8;
+  }
+  if (n >> 4) {
+    zeroes -= 4;
+    n >>= 4;
+  }
+  return "\4\3\2\2\1\1\1\1\0\0\0\0\0\0\0"[n] + zeroes;
+}
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountLeadingZeros64(uint64_t n) {
+#if defined(_MSC_VER) && !defined(__clang__) && defined(_M_X64)
+  // MSVC does not have __buitin_clzll. Use _BitScanReverse64.
+  unsigned long result = 0;  // NOLINT(runtime/int)
+  if (_BitScanReverse64(&result, n)) {
+    return 63 - result;
+  }
+  return 64;
+#elif defined(_MSC_VER) && !defined(__clang__)
+  // MSVC does not have __buitin_clzll. Compose two calls to _BitScanReverse
+  unsigned long result = 0;  // NOLINT(runtime/int)
+  if ((n >> 32) &&
+      _BitScanReverse(&result, static_cast<unsigned long>(n >> 32))) {
+    return 31 - result;
+  }
+  if (_BitScanReverse(&result, static_cast<unsigned long>(n))) {
+    return 63 - result;
+  }
+  return 64;
+#elif defined(__GNUC__) || defined(__clang__)
+  // Use __builtin_clzll, which uses the following instructions:
+  //  x86: bsr
+  //  ARM64: clz
+  //  PPC: cntlzd
+  static_assert(sizeof(unsigned long long) == sizeof(n),  // NOLINT(runtime/int)
+                "__builtin_clzll does not take 64-bit arg");
+
+  // Handle 0 as a special case because __builtin_clzll(0) is undefined.
+  if (n == 0) {
+    return 64;
+  }
+  return __builtin_clzll(n);
+#else
+  return CountLeadingZeros64Slow(n);
+#endif
+}
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountLeadingZeros32Slow(uint64_t n) {
+  int zeroes = 28;
+  if (n >> 16) {
+    zeroes -= 16;
+    n >>= 16;
+  }
+  if (n >> 8) {
+    zeroes -= 8;
+    n >>= 8;
+  }
+  if (n >> 4) {
+    zeroes -= 4;
+    n >>= 4;
+  }
+  return "\4\3\2\2\1\1\1\1\0\0\0\0\0\0\0"[n] + zeroes;
+}
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountLeadingZeros32(uint32_t n) {
+#if defined(_MSC_VER) && !defined(__clang__)
+  unsigned long result = 0;  // NOLINT(runtime/int)
+  if (_BitScanReverse(&result, n)) {
+    return 31 - result;
+  }
+  return 32;
+#elif defined(__GNUC__) || defined(__clang__)
+  // Use __builtin_clz, which uses the following instructions:
+  //  x86: bsr
+  //  ARM64: clz
+  //  PPC: cntlzd
+  static_assert(sizeof(int) == sizeof(n),
+                "__builtin_clz does not take 32-bit arg");
+
+  // Handle 0 as a special case because __builtin_clz(0) is undefined.
+  if (n == 0) {
+    return 32;
+  }
+  return __builtin_clz(n);
+#else
+  return CountLeadingZeros32Slow(n);
+#endif
+}
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountTrailingZerosNonZero64Slow(uint64_t n) {
+  int c = 63;
+  n &= ~n + 1;
+  if (n & 0x00000000FFFFFFFF) c -= 32;
+  if (n & 0x0000FFFF0000FFFF) c -= 16;
+  if (n & 0x00FF00FF00FF00FF) c -= 8;
+  if (n & 0x0F0F0F0F0F0F0F0F) c -= 4;
+  if (n & 0x3333333333333333) c -= 2;
+  if (n & 0x5555555555555555) c -= 1;
+  return c;
+}
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountTrailingZerosNonZero64(uint64_t n) {
+#if defined(_MSC_VER) && !defined(__clang__) && defined(_M_X64)
+  unsigned long result = 0;  // NOLINT(runtime/int)
+  _BitScanForward64(&result, n);
+  return result;
+#elif defined(_MSC_VER) && !defined(__clang__)
+  unsigned long result = 0;  // NOLINT(runtime/int)
+  if (static_cast<uint32_t>(n) == 0) {
+    _BitScanForward(&result, static_cast<unsigned long>(n >> 32));
+    return result + 32;
+  }
+  _BitScanForward(&result, static_cast<unsigned long>(n));
+  return result;
+#elif defined(__GNUC__) || defined(__clang__)
+  static_assert(sizeof(unsigned long long) == sizeof(n),  // NOLINT(runtime/int)
+                "__builtin_ctzll does not take 64-bit arg");
+  return __builtin_ctzll(n);
+#else
+  return CountTrailingZerosNonZero64Slow(n);
+#endif
+}
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountTrailingZerosNonZero32Slow(uint32_t n) {
+  int c = 31;
+  n &= ~n + 1;
+  if (n & 0x0000FFFF) c -= 16;
+  if (n & 0x00FF00FF) c -= 8;
+  if (n & 0x0F0F0F0F) c -= 4;
+  if (n & 0x33333333) c -= 2;
+  if (n & 0x55555555) c -= 1;
+  return c;
+}
+
+ABSL_BASE_INTERNAL_FORCEINLINE int CountTrailingZerosNonZero32(uint32_t n) {
+#if defined(_MSC_VER) && !defined(__clang__)
+  unsigned long result = 0;  // NOLINT(runtime/int)
+  _BitScanForward(&result, n);
+  return result;
+#elif defined(__GNUC__) || defined(__clang__)
+  static_assert(sizeof(int) == sizeof(n),
+                "__builtin_ctz does not take 32-bit arg");
+  return __builtin_ctz(n);
+#else
+  return CountTrailingZerosNonZero32Slow(n);
+#endif
+}
+
+#undef ABSL_BASE_INTERNAL_FORCEINLINE
+
+}  // namespace base_internal
+ABSL_NAMESPACE_END
+}  // namespace absl
+
+#endif  // ABSL_BASE_INTERNAL_BITS_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/macros.h b/lib/clickhouse-cpp/contrib/absl/absl/base/macros.h
new file mode 100644
index 0000000..02dd9ff
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/macros.h
@@ -0,0 +1,147 @@
+//
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// -----------------------------------------------------------------------------
+// File: macros.h
+// -----------------------------------------------------------------------------
+//
+// This header file defines the set of language macros used within Abseil code.
+// For the set of macros used to determine supported compilers and platforms,
+// see absl/base/config.h instead.
+//
+// This code is compiled directly on many platforms, including client
+// platforms like Windows, Mac, and embedded systems.  Before making
+// any changes here, make sure that you're not breaking any platforms.
+
+#ifndef ABSL_BASE_MACROS_H_
+#define ABSL_BASE_MACROS_H_
+
+#include <cassert>
+#include <cstddef>
+
+#include "absl/base/attributes.h"
+#include "absl/base/config.h"
+#include "absl/base/optimization.h"
+#include "absl/base/port.h"
+
+// ABSL_ARRAYSIZE()
+//
+// Returns the number of elements in an array as a compile-time constant, which
+// can be used in defining new arrays. If you use this macro on a pointer by
+// mistake, you will get a compile-time error.
+#define ABSL_ARRAYSIZE(array) \
+  (sizeof(::absl::macros_internal::ArraySizeHelper(array)))
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace macros_internal {
+// Note: this internal template function declaration is used by ABSL_ARRAYSIZE.
+// The function doesn't need a definition, as we only use its type.
+template <typename T, size_t N>
+auto ArraySizeHelper(const T (&array)[N]) -> char (&)[N];
+}  // namespace macros_internal
+ABSL_NAMESPACE_END
+}  // namespace absl
+
+// ABSL_BAD_CALL_IF()
+//
+// Used on a function overload to trap bad calls: any call that matches the
+// overload will cause a compile-time error. This macro uses a clang-specific
+// "enable_if" attribute, as described at
+// https://clang.llvm.org/docs/AttributeReference.html#enable-if
+//
+// Overloads which use this macro should be bracketed by
+// `#ifdef ABSL_BAD_CALL_IF`.
+//
+// Example:
+//
+//   int isdigit(int c);
+//   #ifdef ABSL_BAD_CALL_IF
+//   int isdigit(int c)
+//     ABSL_BAD_CALL_IF(c <= -1 || c > 255,
+//                       "'c' must have the value of an unsigned char or EOF");
+//   #endif // ABSL_BAD_CALL_IF
+#if ABSL_HAVE_ATTRIBUTE(enable_if)
+#define ABSL_BAD_CALL_IF(expr, msg) \
+  __attribute__((enable_if(expr, "Bad call trap"), unavailable(msg)))
+#endif
+
+// ABSL_ASSERT()
+//
+// In C++11, `assert` can't be used portably within constexpr functions.
+// ABSL_ASSERT functions as a runtime assert but works in C++11 constexpr
+// functions.  Example:
+//
+// constexpr double Divide(double a, double b) {
+//   return ABSL_ASSERT(b != 0), a / b;
+// }
+//
+// This macro is inspired by
+// https://akrzemi1.wordpress.com/2017/05/18/asserts-in-constexpr-functions/
+#if defined(NDEBUG)
+#define ABSL_ASSERT(expr) \
+  (false ? static_cast<void>(expr) : static_cast<void>(0))
+#else
+#define ABSL_ASSERT(expr)                           \
+  (ABSL_PREDICT_TRUE((expr)) ? static_cast<void>(0) \
+                             : [] { assert(false && #expr); }())  // NOLINT
+#endif
+
+// `ABSL_INTERNAL_HARDENING_ABORT()` controls how `ABSL_HARDENING_ASSERT()`
+// aborts the program in release mode (when NDEBUG is defined). The
+// implementation should abort the program as quickly as possible and ideally it
+// should not be possible to ignore the abort request.
+#if (ABSL_HAVE_BUILTIN(__builtin_trap) &&         \
+     ABSL_HAVE_BUILTIN(__builtin_unreachable)) || \
+    (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_INTERNAL_HARDENING_ABORT() \
+  do {                                  \
+    __builtin_trap();                   \
+    __builtin_unreachable();            \
+  } while (false)
+#else
+#define ABSL_INTERNAL_HARDENING_ABORT() abort()
+#endif
+
+// ABSL_HARDENING_ASSERT()
+//
+// `ABSL_HARDENING_ASSERT()` is like `ABSL_ASSERT()`, but used to implement
+// runtime assertions that should be enabled in hardened builds even when
+// `NDEBUG` is defined.
+//
+// When `NDEBUG` is not defined, `ABSL_HARDENING_ASSERT()` is identical to
+// `ABSL_ASSERT()`.
+//
+// See `ABSL_OPTION_HARDENED` in `absl/base/options.h` for more information on
+// hardened mode.
+#if ABSL_OPTION_HARDENED == 1 && defined(NDEBUG)
+#define ABSL_HARDENING_ASSERT(expr)                 \
+  (ABSL_PREDICT_TRUE((expr)) ? static_cast<void>(0) \
+                             : [] { ABSL_INTERNAL_HARDENING_ABORT(); }())
+#else
+#define ABSL_HARDENING_ASSERT(expr) ABSL_ASSERT(expr)
+#endif
+
+#ifdef ABSL_HAVE_EXCEPTIONS
+#define ABSL_INTERNAL_TRY try
+#define ABSL_INTERNAL_CATCH_ANY catch (...)
+#define ABSL_INTERNAL_RETHROW do { throw; } while (false)
+#else  // ABSL_HAVE_EXCEPTIONS
+#define ABSL_INTERNAL_TRY if (true)
+#define ABSL_INTERNAL_CATCH_ANY else if (false)
+#define ABSL_INTERNAL_RETHROW do {} while (false)
+#endif  // ABSL_HAVE_EXCEPTIONS
+
+#endif  // ABSL_BASE_MACROS_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/optimization.h b/lib/clickhouse-cpp/contrib/absl/absl/base/optimization.h
new file mode 100644
index 0000000..2e31376
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/optimization.h
@@ -0,0 +1,241 @@
+//
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// -----------------------------------------------------------------------------
+// File: optimization.h
+// -----------------------------------------------------------------------------
+//
+// This header file defines portable macros for performance optimization.
+
+#ifndef ABSL_BASE_OPTIMIZATION_H_
+#define ABSL_BASE_OPTIMIZATION_H_
+
+#include "absl/base/config.h"
+
+// ABSL_BLOCK_TAIL_CALL_OPTIMIZATION
+//
+// Instructs the compiler to avoid optimizing tail-call recursion. Use of this
+// macro is useful when you wish to preserve the existing function order within
+// a stack trace for logging, debugging, or profiling purposes.
+//
+// Example:
+//
+//   int f() {
+//     int result = g();
+//     ABSL_BLOCK_TAIL_CALL_OPTIMIZATION();
+//     return result;
+//   }
+#if defined(__pnacl__)
+#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() if (volatile int x = 0) { (void)x; }
+#elif defined(__clang__)
+// Clang will not tail call given inline volatile assembly.
+#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() __asm__ __volatile__("")
+#elif defined(__GNUC__)
+// GCC will not tail call given inline volatile assembly.
+#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() __asm__ __volatile__("")
+#elif defined(_MSC_VER)
+#include <intrin.h>
+// The __nop() intrinsic blocks the optimisation.
+#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() __nop()
+#else
+#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() if (volatile int x = 0) { (void)x; }
+#endif
+
+// ABSL_CACHELINE_SIZE
+//
+// Explicitly defines the size of the L1 cache for purposes of alignment.
+// Setting the cacheline size allows you to specify that certain objects be
+// aligned on a cacheline boundary with `ABSL_CACHELINE_ALIGNED` declarations.
+// (See below.)
+//
+// NOTE: this macro should be replaced with the following C++17 features, when
+// those are generally available:
+//
+//   * `std::hardware_constructive_interference_size`
+//   * `std::hardware_destructive_interference_size`
+//
+// See http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0154r1.html
+// for more information.
+#if defined(__GNUC__)
+// Cache line alignment
+#if defined(__i386__) || defined(__x86_64__)
+#define ABSL_CACHELINE_SIZE 64
+#elif defined(__powerpc64__)
+#define ABSL_CACHELINE_SIZE 128
+#elif defined(__aarch64__)
+// We would need to read special register ctr_el0 to find out L1 dcache size.
+// This value is a good estimate based on a real aarch64 machine.
+#define ABSL_CACHELINE_SIZE 64
+#elif defined(__arm__)
+// Cache line sizes for ARM: These values are not strictly correct since
+// cache line sizes depend on implementations, not architectures.  There
+// are even implementations with cache line sizes configurable at boot
+// time.
+#if defined(__ARM_ARCH_5T__)
+#define ABSL_CACHELINE_SIZE 32
+#elif defined(__ARM_ARCH_7A__)
+#define ABSL_CACHELINE_SIZE 64
+#endif
+#endif
+
+#ifndef ABSL_CACHELINE_SIZE
+// A reasonable default guess.  Note that overestimates tend to waste more
+// space, while underestimates tend to waste more time.
+#define ABSL_CACHELINE_SIZE 64
+#endif
+
+// ABSL_CACHELINE_ALIGNED
+//
+// Indicates that the declared object be cache aligned using
+// `ABSL_CACHELINE_SIZE` (see above). Cacheline aligning objects allows you to
+// load a set of related objects in the L1 cache for performance improvements.
+// Cacheline aligning objects properly allows constructive memory sharing and
+// prevents destructive (or "false") memory sharing.
+//
+// NOTE: this macro should be replaced with usage of `alignas()` using
+// `std::hardware_constructive_interference_size` and/or
+// `std::hardware_destructive_interference_size` when available within C++17.
+//
+// See http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0154r1.html
+// for more information.
+//
+// On some compilers, `ABSL_CACHELINE_ALIGNED` expands to an `__attribute__`
+// or `__declspec` attribute. For compilers where this is not known to work,
+// the macro expands to nothing.
+//
+// No further guarantees are made here. The result of applying the macro
+// to variables and types is always implementation-defined.
+//
+// WARNING: It is easy to use this attribute incorrectly, even to the point
+// of causing bugs that are difficult to diagnose, crash, etc. It does not
+// of itself guarantee that objects are aligned to a cache line.
+//
+// NOTE: Some compilers are picky about the locations of annotations such as
+// this attribute, so prefer to put it at the beginning of your declaration.
+// For example,
+//
+//   ABSL_CACHELINE_ALIGNED static Foo* foo = ...
+//
+//   class ABSL_CACHELINE_ALIGNED Bar { ...
+//
+// Recommendations:
+//
+// 1) Consult compiler documentation; this comment is not kept in sync as
+//    toolchains evolve.
+// 2) Verify your use has the intended effect. This often requires inspecting
+//    the generated machine code.
+// 3) Prefer applying this attribute to individual variables. Avoid
+//    applying it to types. This tends to localize the effect.
+#define ABSL_CACHELINE_ALIGNED __attribute__((aligned(ABSL_CACHELINE_SIZE)))
+#elif defined(_MSC_VER)
+#define ABSL_CACHELINE_SIZE 64
+#define ABSL_CACHELINE_ALIGNED __declspec(align(ABSL_CACHELINE_SIZE))
+#else
+#define ABSL_CACHELINE_SIZE 64
+#define ABSL_CACHELINE_ALIGNED
+#endif
+
+// ABSL_PREDICT_TRUE, ABSL_PREDICT_FALSE
+//
+// Enables the compiler to prioritize compilation using static analysis for
+// likely paths within a boolean branch.
+//
+// Example:
+//
+//   if (ABSL_PREDICT_TRUE(expression)) {
+//     return result;                        // Faster if more likely
+//   } else {
+//     return 0;
+//   }
+//
+// Compilers can use the information that a certain branch is not likely to be
+// taken (for instance, a CHECK failure) to optimize for the common case in
+// the absence of better information (ie. compiling gcc with `-fprofile-arcs`).
+//
+// Recommendation: Modern CPUs dynamically predict branch execution paths,
+// typically with accuracy greater than 97%. As a result, annotating every
+// branch in a codebase is likely counterproductive; however, annotating
+// specific branches that are both hot and consistently mispredicted is likely
+// to yield performance improvements.
+#if ABSL_HAVE_BUILTIN(__builtin_expect) || \
+    (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_PREDICT_FALSE(x) (__builtin_expect(false || (x), false))
+#define ABSL_PREDICT_TRUE(x) (__builtin_expect(false || (x), true))
+#else
+#define ABSL_PREDICT_FALSE(x) (x)
+#define ABSL_PREDICT_TRUE(x) (x)
+#endif
+
+// ABSL_INTERNAL_ASSUME(cond)
+// Informs the compiler than a condition is always true and that it can assume
+// it to be true for optimization purposes. The call has undefined behavior if
+// the condition is false.
+// In !NDEBUG mode, the condition is checked with an assert().
+// NOTE: The expression must not have side effects, as it will only be evaluated
+// in some compilation modes and not others.
+//
+// Example:
+//
+//   int x = ...;
+//   ABSL_INTERNAL_ASSUME(x >= 0);
+//   // The compiler can optimize the division to a simple right shift using the
+//   // assumption specified above.
+//   int y = x / 16;
+//
+#if !defined(NDEBUG)
+#define ABSL_INTERNAL_ASSUME(cond) assert(cond)
+#elif ABSL_HAVE_BUILTIN(__builtin_assume)
+#define ABSL_INTERNAL_ASSUME(cond) __builtin_assume(cond)
+#elif defined(__GNUC__) || ABSL_HAVE_BUILTIN(__builtin_unreachable)
+#define ABSL_INTERNAL_ASSUME(cond)        \
+  do {                                    \
+    if (!(cond)) __builtin_unreachable(); \
+  } while (0)
+#elif defined(_MSC_VER)
+#define ABSL_INTERNAL_ASSUME(cond) __assume(cond)
+#else
+#define ABSL_INTERNAL_ASSUME(cond)      \
+  do {                                  \
+    static_cast<void>(false && (cond)); \
+  } while (0)
+#endif
+
+// ABSL_INTERNAL_UNIQUE_SMALL_NAME(cond)
+// This macro forces small unique name on a static file level symbols like
+// static local variables or static functions. This is intended to be used in
+// macro definitions to optimize the cost of generated code. Do NOT use it on
+// symbols exported from translation unit since it may casue a link time
+// conflict.
+//
+// Example:
+//
+// #define MY_MACRO(txt)
+// namespace {
+//  char VeryVeryLongVarName[] ABSL_INTERNAL_UNIQUE_SMALL_NAME() = txt;
+//  const char* VeryVeryLongFuncName() ABSL_INTERNAL_UNIQUE_SMALL_NAME();
+//  const char* VeryVeryLongFuncName() { return txt; }
+// }
+//
+
+#if defined(__GNUC__)
+#define ABSL_INTERNAL_UNIQUE_SMALL_NAME2(x) #x
+#define ABSL_INTERNAL_UNIQUE_SMALL_NAME1(x) ABSL_INTERNAL_UNIQUE_SMALL_NAME2(x)
+#define ABSL_INTERNAL_UNIQUE_SMALL_NAME() \
+  asm(ABSL_INTERNAL_UNIQUE_SMALL_NAME1(.absl.__COUNTER__))
+#else
+#define ABSL_INTERNAL_UNIQUE_SMALL_NAME()
+#endif
+
+#endif  // ABSL_BASE_OPTIMIZATION_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/options.h b/lib/clickhouse-cpp/contrib/absl/absl/base/options.h
new file mode 100644
index 0000000..230bf1e
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/options.h
@@ -0,0 +1,238 @@
+// Copyright 2019 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// -----------------------------------------------------------------------------
+// File: options.h
+// -----------------------------------------------------------------------------
+//
+// This file contains Abseil configuration options for setting specific
+// implementations instead of letting Abseil determine which implementation to
+// use at compile-time. Setting these options may be useful for package or build
+// managers who wish to guarantee ABI stability within binary builds (which are
+// otherwise difficult to enforce).
+//
+// *** IMPORTANT NOTICE FOR PACKAGE MANAGERS:  It is important that
+// maintainers of package managers who wish to package Abseil read and
+// understand this file! ***
+//
+// Abseil contains a number of possible configuration endpoints, based on
+// parameters such as the detected platform, language version, or command-line
+// flags used to invoke the underlying binary. As is the case with all
+// libraries, binaries which contain Abseil code must ensure that separate
+// packages use the same compiled copy of Abseil to avoid a diamond dependency
+// problem, which can occur if two packages built with different Abseil
+// configuration settings are linked together. Diamond dependency problems in
+// C++ may manifest as violations to the One Definition Rule (ODR) (resulting in
+// linker errors), or undefined behavior (resulting in crashes).
+//
+// Diamond dependency problems can be avoided if all packages utilize the same
+// exact version of Abseil. Building from source code with the same compilation
+// parameters is the easiest way to avoid such dependency problems. However, for
+// package managers who cannot control such compilation parameters, we are
+// providing the file to allow you to inject ABI (Application Binary Interface)
+// stability across builds. Settings options in this file will neither change
+// API nor ABI, providing a stable copy of Abseil between packages.
+//
+// Care must be taken to keep options within these configurations isolated
+// from any other dynamic settings, such as command-line flags which could alter
+// these options. This file is provided specifically to help build and package
+// managers provide a stable copy of Abseil within their libraries and binaries;
+// other developers should not have need to alter the contents of this file.
+//
+// -----------------------------------------------------------------------------
+// Usage
+// -----------------------------------------------------------------------------
+//
+// For any particular package release, set the appropriate definitions within
+// this file to whatever value makes the most sense for your package(s). Note
+// that, by default, most of these options, at the moment, affect the
+// implementation of types; future options may affect other implementation
+// details.
+//
+// NOTE: the defaults within this file all assume that Abseil can select the
+// proper Abseil implementation at compile-time, which will not be sufficient
+// to guarantee ABI stability to package managers.
+
+#ifndef ABSL_BASE_OPTIONS_H_
+#define ABSL_BASE_OPTIONS_H_
+
+// Include a standard library header to allow configuration based on the
+// standard library in use.
+#ifdef __cplusplus
+#include <ciso646>
+#endif
+
+// -----------------------------------------------------------------------------
+// Type Compatibility Options
+// -----------------------------------------------------------------------------
+//
+// ABSL_OPTION_USE_STD_ANY
+//
+// This option controls whether absl::any is implemented as an alias to
+// std::any, or as an independent implementation.
+//
+// A value of 0 means to use Abseil's implementation.  This requires only C++11
+// support, and is expected to work on every toolchain we support.
+//
+// A value of 1 means to use an alias to std::any.  This requires that all code
+// using Abseil is built in C++17 mode or later.
+//
+// A value of 2 means to detect the C++ version being used to compile Abseil,
+// and use an alias only if a working std::any is available.  This option is
+// useful when you are building your entire program, including all of its
+// dependencies, from source.  It should not be used otherwise -- for example,
+// if you are distributing Abseil in a binary package manager -- since in
+// mode 2, absl::any will name a different type, with a different mangled name
+// and binary layout, depending on the compiler flags passed by the end user.
+// For more info, see https://abseil.io/about/design/dropin-types.
+//
+// User code should not inspect this macro.  To check in the preprocessor if
+// absl::any is a typedef of std::any, use the feature macro ABSL_USES_STD_ANY.
+
+#define ABSL_OPTION_USE_STD_ANY 2
+
+
+// ABSL_OPTION_USE_STD_OPTIONAL
+//
+// This option controls whether absl::optional is implemented as an alias to
+// std::optional, or as an independent implementation.
+//
+// A value of 0 means to use Abseil's implementation.  This requires only C++11
+// support, and is expected to work on every toolchain we support.
+//
+// A value of 1 means to use an alias to std::optional.  This requires that all
+// code using Abseil is built in C++17 mode or later.
+//
+// A value of 2 means to detect the C++ version being used to compile Abseil,
+// and use an alias only if a working std::optional is available.  This option
+// is useful when you are building your program from source.  It should not be
+// used otherwise -- for example, if you are distributing Abseil in a binary
+// package manager -- since in mode 2, absl::optional will name a different
+// type, with a different mangled name and binary layout, depending on the
+// compiler flags passed by the end user.  For more info, see
+// https://abseil.io/about/design/dropin-types.
+
+// User code should not inspect this macro.  To check in the preprocessor if
+// absl::optional is a typedef of std::optional, use the feature macro
+// ABSL_USES_STD_OPTIONAL.
+
+#define ABSL_OPTION_USE_STD_OPTIONAL 2
+
+
+// ABSL_OPTION_USE_STD_STRING_VIEW
+//
+// This option controls whether absl::string_view is implemented as an alias to
+// std::string_view, or as an independent implementation.
+//
+// A value of 0 means to use Abseil's implementation.  This requires only C++11
+// support, and is expected to work on every toolchain we support.
+//
+// A value of 1 means to use an alias to std::string_view.  This requires that
+// all code using Abseil is built in C++17 mode or later.
+//
+// A value of 2 means to detect the C++ version being used to compile Abseil,
+// and use an alias only if a working std::string_view is available.  This
+// option is useful when you are building your program from source.  It should
+// not be used otherwise -- for example, if you are distributing Abseil in a
+// binary package manager -- since in mode 2, absl::string_view will name a
+// different type, with a different mangled name and binary layout, depending on
+// the compiler flags passed by the end user.  For more info, see
+// https://abseil.io/about/design/dropin-types.
+//
+// User code should not inspect this macro.  To check in the preprocessor if
+// absl::string_view is a typedef of std::string_view, use the feature macro
+// ABSL_USES_STD_STRING_VIEW.
+
+#define ABSL_OPTION_USE_STD_STRING_VIEW 2
+
+// ABSL_OPTION_USE_STD_VARIANT
+//
+// This option controls whether absl::variant is implemented as an alias to
+// std::variant, or as an independent implementation.
+//
+// A value of 0 means to use Abseil's implementation.  This requires only C++11
+// support, and is expected to work on every toolchain we support.
+//
+// A value of 1 means to use an alias to std::variant.  This requires that all
+// code using Abseil is built in C++17 mode or later.
+//
+// A value of 2 means to detect the C++ version being used to compile Abseil,
+// and use an alias only if a working std::variant is available.  This option
+// is useful when you are building your program from source.  It should not be
+// used otherwise -- for example, if you are distributing Abseil in a binary
+// package manager -- since in mode 2, absl::variant will name a different
+// type, with a different mangled name and binary layout, depending on the
+// compiler flags passed by the end user.  For more info, see
+// https://abseil.io/about/design/dropin-types.
+//
+// User code should not inspect this macro.  To check in the preprocessor if
+// absl::variant is a typedef of std::variant, use the feature macro
+// ABSL_USES_STD_VARIANT.
+
+#define ABSL_OPTION_USE_STD_VARIANT 2
+
+
+// ABSL_OPTION_USE_INLINE_NAMESPACE
+// ABSL_OPTION_INLINE_NAMESPACE_NAME
+//
+// These options controls whether all entities in the absl namespace are
+// contained within an inner inline namespace.  This does not affect the
+// user-visible API of Abseil, but it changes the mangled names of all symbols.
+//
+// This can be useful as a version tag if you are distributing Abseil in
+// precompiled form.  This will prevent a binary library build of Abseil with
+// one inline namespace being used with headers configured with a different
+// inline namespace name.  Binary packagers are reminded that Abseil does not
+// guarantee any ABI stability in Abseil, so any update of Abseil or
+// configuration change in such a binary package should be combined with a
+// new, unique value for the inline namespace name.
+//
+// A value of 0 means not to use inline namespaces.
+//
+// A value of 1 means to use an inline namespace with the given name inside
+// namespace absl.  If this is set, ABSL_OPTION_INLINE_NAMESPACE_NAME must also
+// be changed to a new, unique identifier name.  In particular "head" is not
+// allowed.
+
+#define ABSL_OPTION_USE_INLINE_NAMESPACE 0
+#define ABSL_OPTION_INLINE_NAMESPACE_NAME head
+
+// ABSL_OPTION_HARDENED
+//
+// This option enables a "hardened" build in release mode (in this context,
+// release mode is defined as a build where the `NDEBUG` macro is defined).
+//
+// A value of 0 means that "hardened" mode is not enabled.
+//
+// A value of 1 means that "hardened" mode is enabled.
+//
+// Hardened builds have additional security checks enabled when `NDEBUG` is
+// defined. Defining `NDEBUG` is normally used to turn `assert()` macro into a
+// no-op, as well as disabling other bespoke program consistency checks. By
+// defining ABSL_OPTION_HARDENED to 1, a select set of checks remain enabled in
+// release mode. These checks guard against programming errors that may lead to
+// security vulnerabilities. In release mode, when one of these programming
+// errors is encountered, the program will immediately abort, possibly without
+// any attempt at logging.
+//
+// The checks enabled by this option are not free; they do incur runtime cost.
+//
+// The checks enabled by this option are always active when `NDEBUG` is not
+// defined, even in the case when ABSL_OPTION_HARDENED is defined to 0. The
+// checks enabled by this option may abort the program in a different way and
+// log additional information when `NDEBUG` is not defined.
+
+#define ABSL_OPTION_HARDENED 0
+
+#endif  // ABSL_BASE_OPTIONS_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/policy_checks.h b/lib/clickhouse-cpp/contrib/absl/absl/base/policy_checks.h
new file mode 100644
index 0000000..ffdd8b3
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/policy_checks.h
@@ -0,0 +1,111 @@
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// -----------------------------------------------------------------------------
+// File: policy_checks.h
+// -----------------------------------------------------------------------------
+//
+// This header enforces a minimum set of policies at build time, such as the
+// supported compiler and library versions. Unsupported configurations are
+// reported with `#error`. This enforcement is best effort, so successfully
+// compiling this header does not guarantee a supported configuration.
+
+#ifndef ABSL_BASE_POLICY_CHECKS_H_
+#define ABSL_BASE_POLICY_CHECKS_H_
+
+// Included for the __GLIBC_PREREQ macro used below.
+#include <limits.h>
+
+// Included for the _STLPORT_VERSION macro used below.
+#if defined(__cplusplus)
+#include <cstddef>
+#endif
+
+// -----------------------------------------------------------------------------
+// Operating System Check
+// -----------------------------------------------------------------------------
+
+#if 0 // defined(__CYGWIN__) // it looks like int128 part of absl we use works correctly on cygwin env.
+#error "Cygwin is not supported."
+#endif
+
+// -----------------------------------------------------------------------------
+// Toolchain Check
+// -----------------------------------------------------------------------------
+
+// We support MSVC++ 14.0 update 2 and later.
+// This minimum will go up.
+#if defined(_MSC_FULL_VER) && _MSC_FULL_VER < 190023918 && !defined(__clang__)
+#error "This package requires Visual Studio 2015 Update 2 or higher."
+#endif
+
+// We support gcc 4.7 and later.
+// This minimum will go up.
+#if defined(__GNUC__) && !defined(__clang__)
+#if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 7)
+#error "This package requires gcc 4.7 or higher."
+#endif
+#endif
+
+// We support Apple Xcode clang 4.2.1 (version 421.11.65) and later.
+// This corresponds to Apple Xcode version 4.5.
+// This minimum will go up.
+#if defined(__apple_build_version__) && __apple_build_version__ < 4211165
+#error "This package requires __apple_build_version__ of 4211165 or higher."
+#endif
+
+// -----------------------------------------------------------------------------
+// C++ Version Check
+// -----------------------------------------------------------------------------
+
+// Enforce C++11 as the minimum.  Note that Visual Studio has not
+// advanced __cplusplus despite being good enough for our purposes, so
+// so we exempt it from the check.
+#if defined(__cplusplus) && !defined(_MSC_VER)
+#if __cplusplus < 201103L
+#error "C++ versions less than C++11 are not supported."
+#endif
+#endif
+
+// -----------------------------------------------------------------------------
+// Standard Library Check
+// -----------------------------------------------------------------------------
+
+#if defined(_STLPORT_VERSION)
+#error "STLPort is not supported."
+#endif
+
+// -----------------------------------------------------------------------------
+// `char` Size Check
+// -----------------------------------------------------------------------------
+
+// Abseil currently assumes CHAR_BIT == 8. If you would like to use Abseil on a
+// platform where this is not the case, please provide us with the details about
+// your platform so we can consider relaxing this requirement.
+#if CHAR_BIT != 8
+#error "Abseil assumes CHAR_BIT == 8."
+#endif
+
+// -----------------------------------------------------------------------------
+// `int` Size Check
+// -----------------------------------------------------------------------------
+
+// Abseil currently assumes that an int is 4 bytes. If you would like to use
+// Abseil on a platform where this is not the case, please provide us with the
+// details about your platform so we can consider relaxing this requirement.
+#if INT_MAX < 2147483647
+#error "Abseil assumes that int is at least 4 bytes. "
+#endif
+
+#endif  // ABSL_BASE_POLICY_CHECKS_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/base/port.h b/lib/clickhouse-cpp/contrib/absl/absl/base/port.h
new file mode 100644
index 0000000..6c28068
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/base/port.h
@@ -0,0 +1,26 @@
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// This files is a forwarding header for other headers containing various
+// portability macros and functions.
+// This file is used for both C and C++!
+
+#ifndef ABSL_BASE_PORT_H_
+#define ABSL_BASE_PORT_H_
+
+#include "absl/base/attributes.h"
+#include "absl/base/config.h"
+#include "absl/base/optimization.h"
+
+#endif  // ABSL_BASE_PORT_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128.cc b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128.cc
new file mode 100644
index 0000000..d3696ef
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128.cc
@@ -0,0 +1,390 @@
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "int128.h"
+
+#include <stddef.h>
+
+#include <cassert>
+#include <iomanip>
+#include <ostream>  // NOLINT(readability/streams)
+#include <sstream>
+#include <string>
+#include <type_traits>
+
+#include "absl/base/internal/bits.h"
+#include "absl/base/optimization.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+ABSL_DLL const uint128 kuint128max = MakeUint128(
+    std::numeric_limits<uint64_t>::max(), std::numeric_limits<uint64_t>::max());
+
+namespace {
+
+// Returns the 0-based position of the last set bit (i.e., most significant bit)
+// in the given uint128. The argument is not 0.
+//
+// For example:
+//   Given: 5 (decimal) == 101 (binary)
+//   Returns: 2
+inline ABSL_ATTRIBUTE_ALWAYS_INLINE int Fls128(uint128 n) {
+  if (uint64_t hi = Uint128High64(n)) {
+    ABSL_INTERNAL_ASSUME(hi != 0);
+    return 127 - base_internal::CountLeadingZeros64(hi);
+  }
+  const uint64_t low = Uint128Low64(n);
+  ABSL_INTERNAL_ASSUME(low != 0);
+  return 63 - base_internal::CountLeadingZeros64(low);
+}
+
+// Long division/modulo for uint128 implemented using the shift-subtract
+// division algorithm adapted from:
+// https://stackoverflow.com/questions/5386377/division-without-using
+inline void DivModImpl(uint128 dividend, uint128 divisor, uint128* quotient_ret,
+                       uint128* remainder_ret) {
+  assert(divisor != 0);
+
+  if (divisor > dividend) {
+    *quotient_ret = 0;
+    *remainder_ret = dividend;
+    return;
+  }
+
+  if (divisor == dividend) {
+    *quotient_ret = 1;
+    *remainder_ret = 0;
+    return;
+  }
+
+  uint128 denominator = divisor;
+  uint128 quotient = 0;
+
+  // Left aligns the MSB of the denominator and the dividend.
+  const int shift = Fls128(dividend) - Fls128(denominator);
+  denominator <<= shift;
+
+  // Uses shift-subtract algorithm to divide dividend by denominator. The
+  // remainder will be left in dividend.
+  for (int i = 0; i <= shift; ++i) {
+    quotient <<= 1;
+    if (dividend >= denominator) {
+      dividend -= denominator;
+      quotient |= 1;
+    }
+    denominator >>= 1;
+  }
+
+  *quotient_ret = quotient;
+  *remainder_ret = dividend;
+}
+
+template <typename T>
+uint128 MakeUint128FromFloat(T v) {
+  static_assert(std::is_floating_point<T>::value, "");
+
+  // Rounding behavior is towards zero, same as for built-in types.
+
+  // Undefined behavior if v is NaN or cannot fit into uint128.
+  assert(std::isfinite(v) && v > -1 &&
+         (std::numeric_limits<T>::max_exponent <= 128 ||
+          v < std::ldexp(static_cast<T>(1), 128)));
+
+  if (v >= std::ldexp(static_cast<T>(1), 64)) {
+    uint64_t hi = static_cast<uint64_t>(std::ldexp(v, -64));
+    uint64_t lo = static_cast<uint64_t>(v - std::ldexp(static_cast<T>(hi), 64));
+    return MakeUint128(hi, lo);
+  }
+
+  return MakeUint128(0, static_cast<uint64_t>(v));
+}
+
+#if defined(__clang__) && !defined(__SSE3__)
+// Workaround for clang bug: https://bugs.llvm.org/show_bug.cgi?id=38289
+// Casting from long double to uint64_t is miscompiled and drops bits.
+// It is more work, so only use when we need the workaround.
+uint128 MakeUint128FromFloat(long double v) {
+  // Go 50 bits at a time, that fits in a double
+  static_assert(std::numeric_limits<double>::digits >= 50, "");
+  static_assert(std::numeric_limits<long double>::digits <= 150, "");
+  // Undefined behavior if v is not finite or cannot fit into uint128.
+  assert(std::isfinite(v) && v > -1 && v < std::ldexp(1.0L, 128));
+
+  v = std::ldexp(v, -100);
+  uint64_t w0 = static_cast<uint64_t>(static_cast<double>(std::trunc(v)));
+  v = std::ldexp(v - static_cast<double>(w0), 50);
+  uint64_t w1 = static_cast<uint64_t>(static_cast<double>(std::trunc(v)));
+  v = std::ldexp(v - static_cast<double>(w1), 50);
+  uint64_t w2 = static_cast<uint64_t>(static_cast<double>(std::trunc(v)));
+  return (static_cast<uint128>(w0) << 100) | (static_cast<uint128>(w1) << 50) |
+         static_cast<uint128>(w2);
+}
+#endif  // __clang__ && !__SSE3__
+}  // namespace
+
+uint128::uint128(float v) : uint128(MakeUint128FromFloat(v)) {}
+uint128::uint128(double v) : uint128(MakeUint128FromFloat(v)) {}
+uint128::uint128(long double v) : uint128(MakeUint128FromFloat(v)) {}
+
+uint128 operator/(uint128 lhs, uint128 rhs) {
+#if defined(ABSL_HAVE_INTRINSIC_INT128)
+  return static_cast<unsigned __int128>(lhs) /
+         static_cast<unsigned __int128>(rhs);
+#else  // ABSL_HAVE_INTRINSIC_INT128
+  uint128 quotient = 0;
+  uint128 remainder = 0;
+  DivModImpl(lhs, rhs, &quotient, &remainder);
+  return quotient;
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+}
+uint128 operator%(uint128 lhs, uint128 rhs) {
+#if defined(ABSL_HAVE_INTRINSIC_INT128)
+  return static_cast<unsigned __int128>(lhs) %
+         static_cast<unsigned __int128>(rhs);
+#else  // ABSL_HAVE_INTRINSIC_INT128
+  uint128 quotient = 0;
+  uint128 remainder = 0;
+  DivModImpl(lhs, rhs, &quotient, &remainder);
+  return remainder;
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+}
+
+namespace {
+
+std::string Uint128ToFormattedString(uint128 v, std::ios_base::fmtflags flags) {
+  // Select a divisor which is the largest power of the base < 2^64.
+  uint128 div;
+  int div_base_log;
+  switch (flags & std::ios::basefield) {
+    case std::ios::hex:
+      div = 0x1000000000000000;  // 16^15
+      div_base_log = 15;
+      break;
+    case std::ios::oct:
+      div = 01000000000000000000000;  // 8^21
+      div_base_log = 21;
+      break;
+    default:  // std::ios::dec
+      div = 10000000000000000000u;  // 10^19
+      div_base_log = 19;
+      break;
+  }
+
+  // Now piece together the uint128 representation from three chunks of the
+  // original value, each less than "div" and therefore representable as a
+  // uint64_t.
+  std::ostringstream os;
+  std::ios_base::fmtflags copy_mask =
+      std::ios::basefield | std::ios::showbase | std::ios::uppercase;
+  os.setf(flags & copy_mask, copy_mask);
+  uint128 high = v;
+  uint128 low;
+  DivModImpl(high, div, &high, &low);
+  uint128 mid;
+  DivModImpl(high, div, &high, &mid);
+  if (Uint128Low64(high) != 0) {
+    os << Uint128Low64(high);
+    os << std::noshowbase << std::setfill('0') << std::setw(div_base_log);
+    os << Uint128Low64(mid);
+    os << std::setw(div_base_log);
+  } else if (Uint128Low64(mid) != 0) {
+    os << Uint128Low64(mid);
+    os << std::noshowbase << std::setfill('0') << std::setw(div_base_log);
+  }
+  os << Uint128Low64(low);
+  return os.str();
+}
+
+}  // namespace
+
+std::ostream& operator<<(std::ostream& os, uint128 v) {
+  std::ios_base::fmtflags flags = os.flags();
+  std::string rep = Uint128ToFormattedString(v, flags);
+
+  // Add the requisite padding.
+  std::streamsize width = os.width(0);
+  if (static_cast<size_t>(width) > rep.size()) {
+    std::ios::fmtflags adjustfield = flags & std::ios::adjustfield;
+    if (adjustfield == std::ios::left) {
+      rep.append(width - rep.size(), os.fill());
+    } else if (adjustfield == std::ios::internal &&
+               (flags & std::ios::showbase) &&
+               (flags & std::ios::basefield) == std::ios::hex && v != 0) {
+      rep.insert(2, width - rep.size(), os.fill());
+    } else {
+      rep.insert(0, width - rep.size(), os.fill());
+    }
+  }
+
+  return os << rep;
+}
+
+namespace {
+
+uint128 UnsignedAbsoluteValue(int128 v) {
+  // Cast to uint128 before possibly negating because -Int128Min() is undefined.
+  return Int128High64(v) < 0 ? -uint128(v) : uint128(v);
+}
+
+}  // namespace
+
+#if !defined(ABSL_HAVE_INTRINSIC_INT128)
+namespace {
+
+template <typename T>
+int128 MakeInt128FromFloat(T v) {
+  // Conversion when v is NaN or cannot fit into int128 would be undefined
+  // behavior if using an intrinsic 128-bit integer.
+  assert(std::isfinite(v) && (std::numeric_limits<T>::max_exponent <= 127 ||
+                              (v >= -std::ldexp(static_cast<T>(1), 127) &&
+                               v < std::ldexp(static_cast<T>(1), 127))));
+
+  // We must convert the absolute value and then negate as needed, because
+  // floating point types are typically sign-magnitude. Otherwise, the
+  // difference between the high and low 64 bits when interpreted as two's
+  // complement overwhelms the precision of the mantissa.
+  uint128 result = v < 0 ? -MakeUint128FromFloat(-v) : MakeUint128FromFloat(v);
+  return MakeInt128(int128_internal::BitCastToSigned(Uint128High64(result)),
+                    Uint128Low64(result));
+}
+
+}  // namespace
+
+int128::int128(float v) : int128(MakeInt128FromFloat(v)) {}
+int128::int128(double v) : int128(MakeInt128FromFloat(v)) {}
+int128::int128(long double v) : int128(MakeInt128FromFloat(v)) {}
+
+int128 operator/(int128 lhs, int128 rhs) {
+  assert(lhs != Int128Min() || rhs != -1);  // UB on two's complement.
+
+  uint128 quotient = 0;
+  uint128 remainder = 0;
+  DivModImpl(UnsignedAbsoluteValue(lhs), UnsignedAbsoluteValue(rhs),
+             &quotient, &remainder);
+  if ((Int128High64(lhs) < 0) != (Int128High64(rhs) < 0)) quotient = -quotient;
+  return MakeInt128(int128_internal::BitCastToSigned(Uint128High64(quotient)),
+                    Uint128Low64(quotient));
+}
+
+int128 operator%(int128 lhs, int128 rhs) {
+  assert(lhs != Int128Min() || rhs != -1);  // UB on two's complement.
+
+  uint128 quotient = 0;
+  uint128 remainder = 0;
+  DivModImpl(UnsignedAbsoluteValue(lhs), UnsignedAbsoluteValue(rhs),
+             &quotient, &remainder);
+  if (Int128High64(lhs) < 0) remainder = -remainder;
+  return MakeInt128(int128_internal::BitCastToSigned(Uint128High64(remainder)),
+                    Uint128Low64(remainder));
+}
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+std::ostream& operator<<(std::ostream& os, int128 v) {
+  std::ios_base::fmtflags flags = os.flags();
+  std::string rep;
+
+  // Add the sign if needed.
+  bool print_as_decimal =
+      (flags & std::ios::basefield) == std::ios::dec ||
+      (flags & std::ios::basefield) == std::ios_base::fmtflags();
+  if (print_as_decimal) {
+    if (Int128High64(v) < 0) {
+      rep = "-";
+    } else if (flags & std::ios::showpos) {
+      rep = "+";
+    }
+  }
+
+  rep.append(Uint128ToFormattedString(
+      print_as_decimal ? UnsignedAbsoluteValue(v) : uint128(v), os.flags()));
+
+  // Add the requisite padding.
+  std::streamsize width = os.width(0);
+  if (static_cast<size_t>(width) > rep.size()) {
+    switch (flags & std::ios::adjustfield) {
+      case std::ios::left:
+        rep.append(width - rep.size(), os.fill());
+        break;
+      case std::ios::internal:
+        if (print_as_decimal && (rep[0] == '+' || rep[0] == '-')) {
+          rep.insert(1, width - rep.size(), os.fill());
+        } else if ((flags & std::ios::basefield) == std::ios::hex &&
+                   (flags & std::ios::showbase) && v != 0) {
+          rep.insert(2, width - rep.size(), os.fill());
+        } else {
+          rep.insert(0, width - rep.size(), os.fill());
+        }
+        break;
+      default:  // std::ios::right
+        rep.insert(0, width - rep.size(), os.fill());
+        break;
+    }
+  }
+
+  return os << rep;
+}
+
+ABSL_NAMESPACE_END
+}  // namespace absl
+
+namespace std {
+constexpr bool numeric_limits<absl::uint128>::is_specialized;
+constexpr bool numeric_limits<absl::uint128>::is_signed;
+constexpr bool numeric_limits<absl::uint128>::is_integer;
+constexpr bool numeric_limits<absl::uint128>::is_exact;
+constexpr bool numeric_limits<absl::uint128>::has_infinity;
+constexpr bool numeric_limits<absl::uint128>::has_quiet_NaN;
+constexpr bool numeric_limits<absl::uint128>::has_signaling_NaN;
+constexpr float_denorm_style numeric_limits<absl::uint128>::has_denorm;
+constexpr bool numeric_limits<absl::uint128>::has_denorm_loss;
+constexpr float_round_style numeric_limits<absl::uint128>::round_style;
+constexpr bool numeric_limits<absl::uint128>::is_iec559;
+constexpr bool numeric_limits<absl::uint128>::is_bounded;
+constexpr bool numeric_limits<absl::uint128>::is_modulo;
+constexpr int numeric_limits<absl::uint128>::digits;
+constexpr int numeric_limits<absl::uint128>::digits10;
+constexpr int numeric_limits<absl::uint128>::max_digits10;
+constexpr int numeric_limits<absl::uint128>::radix;
+constexpr int numeric_limits<absl::uint128>::min_exponent;
+constexpr int numeric_limits<absl::uint128>::min_exponent10;
+constexpr int numeric_limits<absl::uint128>::max_exponent;
+constexpr int numeric_limits<absl::uint128>::max_exponent10;
+constexpr bool numeric_limits<absl::uint128>::traps;
+constexpr bool numeric_limits<absl::uint128>::tinyness_before;
+
+constexpr bool numeric_limits<absl::int128>::is_specialized;
+constexpr bool numeric_limits<absl::int128>::is_signed;
+constexpr bool numeric_limits<absl::int128>::is_integer;
+constexpr bool numeric_limits<absl::int128>::is_exact;
+constexpr bool numeric_limits<absl::int128>::has_infinity;
+constexpr bool numeric_limits<absl::int128>::has_quiet_NaN;
+constexpr bool numeric_limits<absl::int128>::has_signaling_NaN;
+constexpr float_denorm_style numeric_limits<absl::int128>::has_denorm;
+constexpr bool numeric_limits<absl::int128>::has_denorm_loss;
+constexpr float_round_style numeric_limits<absl::int128>::round_style;
+constexpr bool numeric_limits<absl::int128>::is_iec559;
+constexpr bool numeric_limits<absl::int128>::is_bounded;
+constexpr bool numeric_limits<absl::int128>::is_modulo;
+constexpr int numeric_limits<absl::int128>::digits;
+constexpr int numeric_limits<absl::int128>::digits10;
+constexpr int numeric_limits<absl::int128>::max_digits10;
+constexpr int numeric_limits<absl::int128>::radix;
+constexpr int numeric_limits<absl::int128>::min_exponent;
+constexpr int numeric_limits<absl::int128>::min_exponent10;
+constexpr int numeric_limits<absl::int128>::max_exponent;
+constexpr int numeric_limits<absl::int128>::max_exponent10;
+constexpr bool numeric_limits<absl::int128>::traps;
+constexpr bool numeric_limits<absl::int128>::tinyness_before;
+}  // namespace std
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128.h b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128.h
new file mode 100644
index 0000000..0dd814a
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128.h
@@ -0,0 +1,1092 @@
+//
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// -----------------------------------------------------------------------------
+// File: int128.h
+// -----------------------------------------------------------------------------
+//
+// This header file defines 128-bit integer types, `uint128` and `int128`.
+
+#ifndef ABSL_NUMERIC_INT128_H_
+#define ABSL_NUMERIC_INT128_H_
+
+#include <cassert>
+#include <cmath>
+#include <cstdint>
+#include <cstring>
+#include <iosfwd>
+#include <limits>
+#include <utility>
+
+#include "absl/base/config.h"
+#include "absl/base/macros.h"
+#include "absl/base/port.h"
+
+#if defined(_MSC_VER)
+// In very old versions of MSVC and when the /Zc:wchar_t flag is off, wchar_t is
+// a typedef for unsigned short.  Otherwise wchar_t is mapped to the __wchar_t
+// builtin type.  We need to make sure not to define operator wchar_t()
+// alongside operator unsigned short() in these instances.
+#define ABSL_INTERNAL_WCHAR_T __wchar_t
+#if defined(_M_X64)
+#include <intrin.h>
+#pragma intrinsic(_umul128)
+#endif  // defined(_M_X64)
+#else   // defined(_MSC_VER)
+#define ABSL_INTERNAL_WCHAR_T wchar_t
+#endif  // defined(_MSC_VER)
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+class int128;
+
+// uint128
+//
+// An unsigned 128-bit integer type. The API is meant to mimic an intrinsic type
+// as closely as is practical, including exhibiting undefined behavior in
+// analogous cases (e.g. division by zero). This type is intended to be a
+// drop-in replacement once C++ supports an intrinsic `uint128_t` type; when
+// that occurs, existing well-behaved uses of `uint128` will continue to work
+// using that new type.
+//
+// Note: code written with this type will continue to compile once `uint128_t`
+// is introduced, provided the replacement helper functions
+// `Uint128(Low|High)64()` and `MakeUint128()` are made.
+//
+// A `uint128` supports the following:
+//
+//   * Implicit construction from integral types
+//   * Explicit conversion to integral types
+//
+// Additionally, if your compiler supports `__int128`, `uint128` is
+// interoperable with that type. (Abseil checks for this compatibility through
+// the `ABSL_HAVE_INTRINSIC_INT128` macro.)
+//
+// However, a `uint128` differs from intrinsic integral types in the following
+// ways:
+//
+//   * Errors on implicit conversions that do not preserve value (such as
+//     loss of precision when converting to float values).
+//   * Requires explicit construction from and conversion to floating point
+//     types.
+//   * Conversion to integral types requires an explicit static_cast() to
+//     mimic use of the `-Wnarrowing` compiler flag.
+//   * The alignment requirement of `uint128` may differ from that of an
+//     intrinsic 128-bit integer type depending on platform and build
+//     configuration.
+//
+// Example:
+//
+//     float y = absl::Uint128Max();  // Error. uint128 cannot be implicitly
+//                                    // converted to float.
+//
+//     absl::uint128 v;
+//     uint64_t i = v;                         // Error
+//     uint64_t i = static_cast<uint64_t>(v);  // OK
+//
+class
+#if defined(ABSL_HAVE_INTRINSIC_INT128)
+    alignas(unsigned __int128)
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+        uint128 {
+ public:
+  uint128() = default;
+
+  // Constructors from arithmetic types
+  constexpr uint128(int v);                 // NOLINT(runtime/explicit)
+  constexpr uint128(unsigned int v);        // NOLINT(runtime/explicit)
+  constexpr uint128(long v);                // NOLINT(runtime/int)
+  constexpr uint128(unsigned long v);       // NOLINT(runtime/int)
+  constexpr uint128(long long v);           // NOLINT(runtime/int)
+  constexpr uint128(unsigned long long v);  // NOLINT(runtime/int)
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  constexpr uint128(__int128 v);           // NOLINT(runtime/explicit)
+  constexpr uint128(unsigned __int128 v);  // NOLINT(runtime/explicit)
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+  constexpr uint128(int128 v);  // NOLINT(runtime/explicit)
+  explicit uint128(float v);
+  explicit uint128(double v);
+  explicit uint128(long double v);
+
+  // Assignment operators from arithmetic types
+  uint128& operator=(int v);
+  uint128& operator=(unsigned int v);
+  uint128& operator=(long v);                // NOLINT(runtime/int)
+  uint128& operator=(unsigned long v);       // NOLINT(runtime/int)
+  uint128& operator=(long long v);           // NOLINT(runtime/int)
+  uint128& operator=(unsigned long long v);  // NOLINT(runtime/int)
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  uint128& operator=(__int128 v);
+  uint128& operator=(unsigned __int128 v);
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+  uint128& operator=(int128 v);
+
+  // Conversion operators to other arithmetic types
+  constexpr explicit operator bool() const;
+  constexpr explicit operator char() const;
+  constexpr explicit operator signed char() const;
+  constexpr explicit operator unsigned char() const;
+  constexpr explicit operator char16_t() const;
+  constexpr explicit operator char32_t() const;
+  constexpr explicit operator ABSL_INTERNAL_WCHAR_T() const;
+  constexpr explicit operator short() const;  // NOLINT(runtime/int)
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator unsigned short() const;
+  constexpr explicit operator int() const;
+  constexpr explicit operator unsigned int() const;
+  constexpr explicit operator long() const;  // NOLINT(runtime/int)
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator unsigned long() const;
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator long long() const;
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator unsigned long long() const;
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  constexpr explicit operator __int128() const;
+  constexpr explicit operator unsigned __int128() const;
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+  explicit operator float() const;
+  explicit operator double() const;
+  explicit operator long double() const;
+
+  // Trivial copy constructor, assignment operator and destructor.
+
+  // Arithmetic operators.
+  uint128& operator+=(uint128 other);
+  uint128& operator-=(uint128 other);
+  uint128& operator*=(uint128 other);
+  // Long division/modulo for uint128.
+  uint128& operator/=(uint128 other);
+  uint128& operator%=(uint128 other);
+  uint128 operator++(int);
+  uint128 operator--(int);
+  uint128& operator<<=(int);
+  uint128& operator>>=(int);
+  uint128& operator&=(uint128 other);
+  uint128& operator|=(uint128 other);
+  uint128& operator^=(uint128 other);
+  uint128& operator++();
+  uint128& operator--();
+
+  // Uint128Low64()
+  //
+  // Returns the lower 64-bit value of a `uint128` value.
+  friend constexpr uint64_t Uint128Low64(uint128 v);
+
+  // Uint128High64()
+  //
+  // Returns the higher 64-bit value of a `uint128` value.
+  friend constexpr uint64_t Uint128High64(uint128 v);
+
+  // MakeUInt128()
+  //
+  // Constructs a `uint128` numeric value from two 64-bit unsigned integers.
+  // Note that this factory function is the only way to construct a `uint128`
+  // from integer values greater than 2^64.
+  //
+  // Example:
+  //
+  //   absl::uint128 big = absl::MakeUint128(1, 0);
+  friend constexpr uint128 MakeUint128(uint64_t high, uint64_t low);
+
+  // Uint128Max()
+  //
+  // Returns the highest value for a 128-bit unsigned integer.
+  friend constexpr uint128 Uint128Max();
+
+  // Support for absl::Hash.
+  template <typename H>
+  friend H AbslHashValue(H h, uint128 v) {
+    return H::combine(std::move(h), Uint128High64(v), Uint128Low64(v));
+  }
+
+ private:
+  constexpr uint128(uint64_t high, uint64_t low);
+
+  // TODO(strel) Update implementation to use __int128 once all users of
+  // uint128 are fixed to not depend on alignof(uint128) == 8. Also add
+  // alignas(16) to class definition to keep alignment consistent across
+  // platforms.
+#if defined(ABSL_IS_LITTLE_ENDIAN)
+  uint64_t lo_;
+  uint64_t hi_;
+#elif defined(ABSL_IS_BIG_ENDIAN)
+  uint64_t hi_;
+  uint64_t lo_;
+#else  // byte order
+#error "Unsupported byte order: must be little-endian or big-endian."
+#endif  // byte order
+};
+
+// Prefer to use the constexpr `Uint128Max()`.
+//
+// TODO(absl-team) deprecate kuint128max once migration tool is released.
+ABSL_DLL extern const uint128 kuint128max;
+
+// allow uint128 to be logged
+std::ostream& operator<<(std::ostream& os, uint128 v);
+
+// TODO(strel) add operator>>(std::istream&, uint128)
+
+constexpr uint128 Uint128Max() {
+  return uint128((std::numeric_limits<uint64_t>::max)(),
+                 (std::numeric_limits<uint64_t>::max)());
+}
+
+ABSL_NAMESPACE_END
+}  // namespace absl
+
+// Specialized numeric_limits for uint128.
+namespace std {
+template <>
+class numeric_limits<absl::uint128> {
+ public:
+  static constexpr bool is_specialized = true;
+  static constexpr bool is_signed = false;
+  static constexpr bool is_integer = true;
+  static constexpr bool is_exact = true;
+  static constexpr bool has_infinity = false;
+  static constexpr bool has_quiet_NaN = false;
+  static constexpr bool has_signaling_NaN = false;
+  static constexpr float_denorm_style has_denorm = denorm_absent;
+  static constexpr bool has_denorm_loss = false;
+  static constexpr float_round_style round_style = round_toward_zero;
+  static constexpr bool is_iec559 = false;
+  static constexpr bool is_bounded = true;
+  static constexpr bool is_modulo = true;
+  static constexpr int digits = 128;
+  static constexpr int digits10 = 38;
+  static constexpr int max_digits10 = 0;
+  static constexpr int radix = 2;
+  static constexpr int min_exponent = 0;
+  static constexpr int min_exponent10 = 0;
+  static constexpr int max_exponent = 0;
+  static constexpr int max_exponent10 = 0;
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  static constexpr bool traps = numeric_limits<unsigned __int128>::traps;
+#else   // ABSL_HAVE_INTRINSIC_INT128
+  static constexpr bool traps = numeric_limits<uint64_t>::traps;
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+  static constexpr bool tinyness_before = false;
+
+  static constexpr absl::uint128 (min)() { return 0; }
+  static constexpr absl::uint128 lowest() { return 0; }
+  static constexpr absl::uint128 (max)() { return absl::Uint128Max(); }
+  static constexpr absl::uint128 epsilon() { return 0; }
+  static constexpr absl::uint128 round_error() { return 0; }
+  static constexpr absl::uint128 infinity() { return 0; }
+  static constexpr absl::uint128 quiet_NaN() { return 0; }
+  static constexpr absl::uint128 signaling_NaN() { return 0; }
+  static constexpr absl::uint128 denorm_min() { return 0; }
+};
+}  // namespace std
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+// int128
+//
+// A signed 128-bit integer type. The API is meant to mimic an intrinsic
+// integral type as closely as is practical, including exhibiting undefined
+// behavior in analogous cases (e.g. division by zero).
+//
+// An `int128` supports the following:
+//
+//   * Implicit construction from integral types
+//   * Explicit conversion to integral types
+//
+// However, an `int128` differs from intrinsic integral types in the following
+// ways:
+//
+//   * It is not implicitly convertible to other integral types.
+//   * Requires explicit construction from and conversion to floating point
+//     types.
+
+// Additionally, if your compiler supports `__int128`, `int128` is
+// interoperable with that type. (Abseil checks for this compatibility through
+// the `ABSL_HAVE_INTRINSIC_INT128` macro.)
+//
+// The design goal for `int128` is that it will be compatible with a future
+// `int128_t`, if that type becomes a part of the standard.
+//
+// Example:
+//
+//     float y = absl::int128(17);  // Error. int128 cannot be implicitly
+//                                  // converted to float.
+//
+//     absl::int128 v;
+//     int64_t i = v;                        // Error
+//     int64_t i = static_cast<int64_t>(v);  // OK
+//
+class int128 {
+ public:
+  int128() = default;
+
+  // Constructors from arithmetic types
+  constexpr int128(int v);                 // NOLINT(runtime/explicit)
+  constexpr int128(unsigned int v);        // NOLINT(runtime/explicit)
+  constexpr int128(long v);                // NOLINT(runtime/int)
+  constexpr int128(unsigned long v);       // NOLINT(runtime/int)
+  constexpr int128(long long v);           // NOLINT(runtime/int)
+  constexpr int128(unsigned long long v);  // NOLINT(runtime/int)
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  constexpr int128(__int128 v);  // NOLINT(runtime/explicit)
+  constexpr explicit int128(unsigned __int128 v);
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+  constexpr explicit int128(uint128 v);
+  explicit int128(float v);
+  explicit int128(double v);
+  explicit int128(long double v);
+
+  // Assignment operators from arithmetic types
+  int128& operator=(int v);
+  int128& operator=(unsigned int v);
+  int128& operator=(long v);                // NOLINT(runtime/int)
+  int128& operator=(unsigned long v);       // NOLINT(runtime/int)
+  int128& operator=(long long v);           // NOLINT(runtime/int)
+  int128& operator=(unsigned long long v);  // NOLINT(runtime/int)
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  int128& operator=(__int128 v);
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+  // Conversion operators to other arithmetic types
+  constexpr explicit operator bool() const;
+  constexpr explicit operator char() const;
+  constexpr explicit operator signed char() const;
+  constexpr explicit operator unsigned char() const;
+  constexpr explicit operator char16_t() const;
+  constexpr explicit operator char32_t() const;
+  constexpr explicit operator ABSL_INTERNAL_WCHAR_T() const;
+  constexpr explicit operator short() const;  // NOLINT(runtime/int)
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator unsigned short() const;
+  constexpr explicit operator int() const;
+  constexpr explicit operator unsigned int() const;
+  constexpr explicit operator long() const;  // NOLINT(runtime/int)
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator unsigned long() const;
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator long long() const;
+  // NOLINTNEXTLINE(runtime/int)
+  constexpr explicit operator unsigned long long() const;
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  constexpr explicit operator __int128() const;
+  constexpr explicit operator unsigned __int128() const;
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+  explicit operator float() const;
+  explicit operator double() const;
+  explicit operator long double() const;
+
+  // Trivial copy constructor, assignment operator and destructor.
+
+  // Arithmetic operators
+  int128& operator+=(int128 other);
+  int128& operator-=(int128 other);
+  int128& operator*=(int128 other);
+  int128& operator/=(int128 other);
+  int128& operator%=(int128 other);
+  int128 operator++(int);  // postfix increment: i++
+  int128 operator--(int);  // postfix decrement: i--
+  int128& operator++();    // prefix increment:  ++i
+  int128& operator--();    // prefix decrement:  --i
+  int128& operator&=(int128 other);
+  int128& operator|=(int128 other);
+  int128& operator^=(int128 other);
+  int128& operator<<=(int amount);
+  int128& operator>>=(int amount);
+
+  // Int128Low64()
+  //
+  // Returns the lower 64-bit value of a `int128` value.
+  friend constexpr uint64_t Int128Low64(int128 v);
+
+  // Int128High64()
+  //
+  // Returns the higher 64-bit value of a `int128` value.
+  friend constexpr int64_t Int128High64(int128 v);
+
+  // MakeInt128()
+  //
+  // Constructs a `int128` numeric value from two 64-bit integers. Note that
+  // signedness is conveyed in the upper `high` value.
+  //
+  //   (absl::int128(1) << 64) * high + low
+  //
+  // Note that this factory function is the only way to construct a `int128`
+  // from integer values greater than 2^64 or less than -2^64.
+  //
+  // Example:
+  //
+  //   absl::int128 big = absl::MakeInt128(1, 0);
+  //   absl::int128 big_n = absl::MakeInt128(-1, 0);
+  friend constexpr int128 MakeInt128(int64_t high, uint64_t low);
+
+  // Int128Max()
+  //
+  // Returns the maximum value for a 128-bit signed integer.
+  friend constexpr int128 Int128Max();
+
+  // Int128Min()
+  //
+  // Returns the minimum value for a 128-bit signed integer.
+  friend constexpr int128 Int128Min();
+
+  // Support for absl::Hash.
+  template <typename H>
+  friend H AbslHashValue(H h, int128 v) {
+    return H::combine(std::move(h), Int128High64(v), Int128Low64(v));
+  }
+
+ private:
+  constexpr int128(int64_t high, uint64_t low);
+
+#if defined(ABSL_HAVE_INTRINSIC_INT128)
+  __int128 v_;
+#else  // ABSL_HAVE_INTRINSIC_INT128
+#if defined(ABSL_IS_LITTLE_ENDIAN)
+  uint64_t lo_;
+  int64_t hi_;
+#elif defined(ABSL_IS_BIG_ENDIAN)
+  int64_t hi_;
+  uint64_t lo_;
+#else  // byte order
+#error "Unsupported byte order: must be little-endian or big-endian."
+#endif  // byte order
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+};
+
+std::ostream& operator<<(std::ostream& os, int128 v);
+
+// TODO(absl-team) add operator>>(std::istream&, int128)
+
+constexpr int128 Int128Max() {
+  return int128((std::numeric_limits<int64_t>::max)(),
+                (std::numeric_limits<uint64_t>::max)());
+}
+
+constexpr int128 Int128Min() {
+  return int128((std::numeric_limits<int64_t>::min)(), 0);
+}
+
+ABSL_NAMESPACE_END
+}  // namespace absl
+
+// Specialized numeric_limits for int128.
+namespace std {
+template <>
+class numeric_limits<absl::int128> {
+ public:
+  static constexpr bool is_specialized = true;
+  static constexpr bool is_signed = true;
+  static constexpr bool is_integer = true;
+  static constexpr bool is_exact = true;
+  static constexpr bool has_infinity = false;
+  static constexpr bool has_quiet_NaN = false;
+  static constexpr bool has_signaling_NaN = false;
+  static constexpr float_denorm_style has_denorm = denorm_absent;
+  static constexpr bool has_denorm_loss = false;
+  static constexpr float_round_style round_style = round_toward_zero;
+  static constexpr bool is_iec559 = false;
+  static constexpr bool is_bounded = true;
+  static constexpr bool is_modulo = false;
+  static constexpr int digits = 127;
+  static constexpr int digits10 = 38;
+  static constexpr int max_digits10 = 0;
+  static constexpr int radix = 2;
+  static constexpr int min_exponent = 0;
+  static constexpr int min_exponent10 = 0;
+  static constexpr int max_exponent = 0;
+  static constexpr int max_exponent10 = 0;
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  static constexpr bool traps = numeric_limits<__int128>::traps;
+#else   // ABSL_HAVE_INTRINSIC_INT128
+  static constexpr bool traps = numeric_limits<uint64_t>::traps;
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+  static constexpr bool tinyness_before = false;
+
+  static constexpr absl::int128 (min)() { return absl::Int128Min(); }
+  static constexpr absl::int128 lowest() { return absl::Int128Min(); }
+  static constexpr absl::int128 (max)() { return absl::Int128Max(); }
+  static constexpr absl::int128 epsilon() { return 0; }
+  static constexpr absl::int128 round_error() { return 0; }
+  static constexpr absl::int128 infinity() { return 0; }
+  static constexpr absl::int128 quiet_NaN() { return 0; }
+  static constexpr absl::int128 signaling_NaN() { return 0; }
+  static constexpr absl::int128 denorm_min() { return 0; }
+};
+}  // namespace std
+
+// --------------------------------------------------------------------------
+//                      Implementation details follow
+// --------------------------------------------------------------------------
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+constexpr uint128 MakeUint128(uint64_t high, uint64_t low) {
+  return uint128(high, low);
+}
+
+// Assignment from integer types.
+
+inline uint128& uint128::operator=(int v) { return *this = uint128(v); }
+
+inline uint128& uint128::operator=(unsigned int v) {
+  return *this = uint128(v);
+}
+
+inline uint128& uint128::operator=(long v) {  // NOLINT(runtime/int)
+  return *this = uint128(v);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+inline uint128& uint128::operator=(unsigned long v) {
+  return *this = uint128(v);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+inline uint128& uint128::operator=(long long v) {
+  return *this = uint128(v);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+inline uint128& uint128::operator=(unsigned long long v) {
+  return *this = uint128(v);
+}
+
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+inline uint128& uint128::operator=(__int128 v) {
+  return *this = uint128(v);
+}
+
+inline uint128& uint128::operator=(unsigned __int128 v) {
+  return *this = uint128(v);
+}
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+inline uint128& uint128::operator=(int128 v) {
+  return *this = uint128(v);
+}
+
+// Arithmetic operators.
+
+uint128 operator<<(uint128 lhs, int amount);
+uint128 operator>>(uint128 lhs, int amount);
+uint128 operator+(uint128 lhs, uint128 rhs);
+uint128 operator-(uint128 lhs, uint128 rhs);
+uint128 operator*(uint128 lhs, uint128 rhs);
+uint128 operator/(uint128 lhs, uint128 rhs);
+uint128 operator%(uint128 lhs, uint128 rhs);
+
+inline uint128& uint128::operator<<=(int amount) {
+  *this = *this << amount;
+  return *this;
+}
+
+inline uint128& uint128::operator>>=(int amount) {
+  *this = *this >> amount;
+  return *this;
+}
+
+inline uint128& uint128::operator+=(uint128 other) {
+  *this = *this + other;
+  return *this;
+}
+
+inline uint128& uint128::operator-=(uint128 other) {
+  *this = *this - other;
+  return *this;
+}
+
+inline uint128& uint128::operator*=(uint128 other) {
+  *this = *this * other;
+  return *this;
+}
+
+inline uint128& uint128::operator/=(uint128 other) {
+  *this = *this / other;
+  return *this;
+}
+
+inline uint128& uint128::operator%=(uint128 other) {
+  *this = *this % other;
+  return *this;
+}
+
+constexpr uint64_t Uint128Low64(uint128 v) { return v.lo_; }
+
+constexpr uint64_t Uint128High64(uint128 v) { return v.hi_; }
+
+// Constructors from integer types.
+
+#if defined(ABSL_IS_LITTLE_ENDIAN)
+
+constexpr uint128::uint128(uint64_t high, uint64_t low)
+    : lo_{low}, hi_{high} {}
+
+constexpr uint128::uint128(int v)
+    : lo_{static_cast<uint64_t>(v)},
+      hi_{v < 0 ? (std::numeric_limits<uint64_t>::max)() : 0} {}
+constexpr uint128::uint128(long v)  // NOLINT(runtime/int)
+    : lo_{static_cast<uint64_t>(v)},
+      hi_{v < 0 ? (std::numeric_limits<uint64_t>::max)() : 0} {}
+constexpr uint128::uint128(long long v)  // NOLINT(runtime/int)
+    : lo_{static_cast<uint64_t>(v)},
+      hi_{v < 0 ? (std::numeric_limits<uint64_t>::max)() : 0} {}
+
+constexpr uint128::uint128(unsigned int v) : lo_{v}, hi_{0} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr uint128::uint128(unsigned long v) : lo_{v}, hi_{0} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr uint128::uint128(unsigned long long v) : lo_{v}, hi_{0} {}
+
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+constexpr uint128::uint128(__int128 v)
+    : lo_{static_cast<uint64_t>(v & ~uint64_t{0})},
+      hi_{static_cast<uint64_t>(static_cast<unsigned __int128>(v) >> 64)} {}
+constexpr uint128::uint128(unsigned __int128 v)
+    : lo_{static_cast<uint64_t>(v & ~uint64_t{0})},
+      hi_{static_cast<uint64_t>(v >> 64)} {}
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+constexpr uint128::uint128(int128 v)
+    : lo_{Int128Low64(v)}, hi_{static_cast<uint64_t>(Int128High64(v))} {}
+
+#elif defined(ABSL_IS_BIG_ENDIAN)
+
+constexpr uint128::uint128(uint64_t high, uint64_t low)
+    : hi_{high}, lo_{low} {}
+
+constexpr uint128::uint128(int v)
+    : hi_{v < 0 ? (std::numeric_limits<uint64_t>::max)() : 0},
+      lo_{static_cast<uint64_t>(v)} {}
+constexpr uint128::uint128(long v)  // NOLINT(runtime/int)
+    : hi_{v < 0 ? (std::numeric_limits<uint64_t>::max)() : 0},
+      lo_{static_cast<uint64_t>(v)} {}
+constexpr uint128::uint128(long long v)  // NOLINT(runtime/int)
+    : hi_{v < 0 ? (std::numeric_limits<uint64_t>::max)() : 0},
+      lo_{static_cast<uint64_t>(v)} {}
+
+constexpr uint128::uint128(unsigned int v) : hi_{0}, lo_{v} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr uint128::uint128(unsigned long v) : hi_{0}, lo_{v} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr uint128::uint128(unsigned long long v) : hi_{0}, lo_{v} {}
+
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+constexpr uint128::uint128(__int128 v)
+    : hi_{static_cast<uint64_t>(static_cast<unsigned __int128>(v) >> 64)},
+      lo_{static_cast<uint64_t>(v & ~uint64_t{0})} {}
+constexpr uint128::uint128(unsigned __int128 v)
+    : hi_{static_cast<uint64_t>(v >> 64)},
+      lo_{static_cast<uint64_t>(v & ~uint64_t{0})} {}
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+constexpr uint128::uint128(int128 v)
+    : hi_{static_cast<uint64_t>(Int128High64(v))}, lo_{Int128Low64(v)} {}
+
+#else  // byte order
+#error "Unsupported byte order: must be little-endian or big-endian."
+#endif  // byte order
+
+// Conversion operators to integer types.
+
+constexpr uint128::operator bool() const { return lo_ || hi_; }
+
+constexpr uint128::operator char() const { return static_cast<char>(lo_); }
+
+constexpr uint128::operator signed char() const {
+  return static_cast<signed char>(lo_);
+}
+
+constexpr uint128::operator unsigned char() const {
+  return static_cast<unsigned char>(lo_);
+}
+
+constexpr uint128::operator char16_t() const {
+  return static_cast<char16_t>(lo_);
+}
+
+constexpr uint128::operator char32_t() const {
+  return static_cast<char32_t>(lo_);
+}
+
+constexpr uint128::operator ABSL_INTERNAL_WCHAR_T() const {
+  return static_cast<ABSL_INTERNAL_WCHAR_T>(lo_);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+constexpr uint128::operator short() const { return static_cast<short>(lo_); }
+
+constexpr uint128::operator unsigned short() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned short>(lo_);            // NOLINT(runtime/int)
+}
+
+constexpr uint128::operator int() const { return static_cast<int>(lo_); }
+
+constexpr uint128::operator unsigned int() const {
+  return static_cast<unsigned int>(lo_);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+constexpr uint128::operator long() const { return static_cast<long>(lo_); }
+
+constexpr uint128::operator unsigned long() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned long>(lo_);            // NOLINT(runtime/int)
+}
+
+constexpr uint128::operator long long() const {  // NOLINT(runtime/int)
+  return static_cast<long long>(lo_);            // NOLINT(runtime/int)
+}
+
+constexpr uint128::operator unsigned long long() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned long long>(lo_);            // NOLINT(runtime/int)
+}
+
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+constexpr uint128::operator __int128() const {
+  return (static_cast<__int128>(hi_) << 64) + lo_;
+}
+
+constexpr uint128::operator unsigned __int128() const {
+  return (static_cast<unsigned __int128>(hi_) << 64) + lo_;
+}
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+// Conversion operators to floating point types.
+
+inline uint128::operator float() const {
+  return static_cast<float>(lo_) + std::ldexp(static_cast<float>(hi_), 64);
+}
+
+inline uint128::operator double() const {
+  return static_cast<double>(lo_) + std::ldexp(static_cast<double>(hi_), 64);
+}
+
+inline uint128::operator long double() const {
+  return static_cast<long double>(lo_) +
+         std::ldexp(static_cast<long double>(hi_), 64);
+}
+
+// Comparison operators.
+
+inline bool operator==(uint128 lhs, uint128 rhs) {
+  return (Uint128Low64(lhs) == Uint128Low64(rhs) &&
+          Uint128High64(lhs) == Uint128High64(rhs));
+}
+
+inline bool operator!=(uint128 lhs, uint128 rhs) {
+  return !(lhs == rhs);
+}
+
+inline bool operator<(uint128 lhs, uint128 rhs) {
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  return static_cast<unsigned __int128>(lhs) <
+         static_cast<unsigned __int128>(rhs);
+#else
+  return (Uint128High64(lhs) == Uint128High64(rhs))
+             ? (Uint128Low64(lhs) < Uint128Low64(rhs))
+             : (Uint128High64(lhs) < Uint128High64(rhs));
+#endif
+}
+
+inline bool operator>(uint128 lhs, uint128 rhs) { return rhs < lhs; }
+
+inline bool operator<=(uint128 lhs, uint128 rhs) { return !(rhs < lhs); }
+
+inline bool operator>=(uint128 lhs, uint128 rhs) { return !(lhs < rhs); }
+
+// Unary operators.
+
+inline uint128 operator-(uint128 val) {
+  uint64_t hi = ~Uint128High64(val);
+  uint64_t lo = ~Uint128Low64(val) + 1;
+  if (lo == 0) ++hi;  // carry
+  return MakeUint128(hi, lo);
+}
+
+inline bool operator!(uint128 val) {
+  return !Uint128High64(val) && !Uint128Low64(val);
+}
+
+// Logical operators.
+
+inline uint128 operator~(uint128 val) {
+  return MakeUint128(~Uint128High64(val), ~Uint128Low64(val));
+}
+
+inline uint128 operator|(uint128 lhs, uint128 rhs) {
+  return MakeUint128(Uint128High64(lhs) | Uint128High64(rhs),
+                           Uint128Low64(lhs) | Uint128Low64(rhs));
+}
+
+inline uint128 operator&(uint128 lhs, uint128 rhs) {
+  return MakeUint128(Uint128High64(lhs) & Uint128High64(rhs),
+                           Uint128Low64(lhs) & Uint128Low64(rhs));
+}
+
+inline uint128 operator^(uint128 lhs, uint128 rhs) {
+  return MakeUint128(Uint128High64(lhs) ^ Uint128High64(rhs),
+                           Uint128Low64(lhs) ^ Uint128Low64(rhs));
+}
+
+inline uint128& uint128::operator|=(uint128 other) {
+  hi_ |= other.hi_;
+  lo_ |= other.lo_;
+  return *this;
+}
+
+inline uint128& uint128::operator&=(uint128 other) {
+  hi_ &= other.hi_;
+  lo_ &= other.lo_;
+  return *this;
+}
+
+inline uint128& uint128::operator^=(uint128 other) {
+  hi_ ^= other.hi_;
+  lo_ ^= other.lo_;
+  return *this;
+}
+
+// Arithmetic operators.
+
+inline uint128 operator<<(uint128 lhs, int amount) {
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  return static_cast<unsigned __int128>(lhs) << amount;
+#else
+  // uint64_t shifts of >= 64 are undefined, so we will need some
+  // special-casing.
+  if (amount < 64) {
+    if (amount != 0) {
+      return MakeUint128(
+          (Uint128High64(lhs) << amount) | (Uint128Low64(lhs) >> (64 - amount)),
+          Uint128Low64(lhs) << amount);
+    }
+    return lhs;
+  }
+  return MakeUint128(Uint128Low64(lhs) << (amount - 64), 0);
+#endif
+}
+
+inline uint128 operator>>(uint128 lhs, int amount) {
+#ifdef ABSL_HAVE_INTRINSIC_INT128
+  return static_cast<unsigned __int128>(lhs) >> amount;
+#else
+  // uint64_t shifts of >= 64 are undefined, so we will need some
+  // special-casing.
+  if (amount < 64) {
+    if (amount != 0) {
+      return MakeUint128(Uint128High64(lhs) >> amount,
+                         (Uint128Low64(lhs) >> amount) |
+                             (Uint128High64(lhs) << (64 - amount)));
+    }
+    return lhs;
+  }
+  return MakeUint128(0, Uint128High64(lhs) >> (amount - 64));
+#endif
+}
+
+inline uint128 operator+(uint128 lhs, uint128 rhs) {
+  uint128 result = MakeUint128(Uint128High64(lhs) + Uint128High64(rhs),
+                               Uint128Low64(lhs) + Uint128Low64(rhs));
+  if (Uint128Low64(result) < Uint128Low64(lhs)) {  // check for carry
+    return MakeUint128(Uint128High64(result) + 1, Uint128Low64(result));
+  }
+  return result;
+}
+
+inline uint128 operator-(uint128 lhs, uint128 rhs) {
+  uint128 result = MakeUint128(Uint128High64(lhs) - Uint128High64(rhs),
+                               Uint128Low64(lhs) - Uint128Low64(rhs));
+  if (Uint128Low64(lhs) < Uint128Low64(rhs)) {  // check for carry
+    return MakeUint128(Uint128High64(result) - 1, Uint128Low64(result));
+  }
+  return result;
+}
+
+inline uint128 operator*(uint128 lhs, uint128 rhs) {
+#if defined(ABSL_HAVE_INTRINSIC_INT128)
+  // TODO(strel) Remove once alignment issues are resolved and unsigned __int128
+  // can be used for uint128 storage.
+  return static_cast<unsigned __int128>(lhs) *
+         static_cast<unsigned __int128>(rhs);
+#elif defined(_MSC_VER) && defined(_M_X64)
+  uint64_t carry;
+  uint64_t low = _umul128(Uint128Low64(lhs), Uint128Low64(rhs), &carry);
+  return MakeUint128(Uint128Low64(lhs) * Uint128High64(rhs) +
+                         Uint128High64(lhs) * Uint128Low64(rhs) + carry,
+                     low);
+#else   // ABSL_HAVE_INTRINSIC128
+  uint64_t a32 = Uint128Low64(lhs) >> 32;
+  uint64_t a00 = Uint128Low64(lhs) & 0xffffffff;
+  uint64_t b32 = Uint128Low64(rhs) >> 32;
+  uint64_t b00 = Uint128Low64(rhs) & 0xffffffff;
+  uint128 result =
+      MakeUint128(Uint128High64(lhs) * Uint128Low64(rhs) +
+                      Uint128Low64(lhs) * Uint128High64(rhs) + a32 * b32,
+                  a00 * b00);
+  result += uint128(a32 * b00) << 32;
+  result += uint128(a00 * b32) << 32;
+  return result;
+#endif  // ABSL_HAVE_INTRINSIC128
+}
+
+// Increment/decrement operators.
+
+inline uint128 uint128::operator++(int) {
+  uint128 tmp(*this);
+  *this += 1;
+  return tmp;
+}
+
+inline uint128 uint128::operator--(int) {
+  uint128 tmp(*this);
+  *this -= 1;
+  return tmp;
+}
+
+inline uint128& uint128::operator++() {
+  *this += 1;
+  return *this;
+}
+
+inline uint128& uint128::operator--() {
+  *this -= 1;
+  return *this;
+}
+
+constexpr int128 MakeInt128(int64_t high, uint64_t low) {
+  return int128(high, low);
+}
+
+// Assignment from integer types.
+inline int128& int128::operator=(int v) {
+  return *this = int128(v);
+}
+
+inline int128& int128::operator=(unsigned int v) {
+  return *this = int128(v);
+}
+
+inline int128& int128::operator=(long v) {  // NOLINT(runtime/int)
+  return *this = int128(v);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+inline int128& int128::operator=(unsigned long v) {
+  return *this = int128(v);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+inline int128& int128::operator=(long long v) {
+  return *this = int128(v);
+}
+
+// NOLINTNEXTLINE(runtime/int)
+inline int128& int128::operator=(unsigned long long v) {
+  return *this = int128(v);
+}
+
+// Arithmetic operators.
+
+int128 operator+(int128 lhs, int128 rhs);
+int128 operator-(int128 lhs, int128 rhs);
+int128 operator*(int128 lhs, int128 rhs);
+int128 operator/(int128 lhs, int128 rhs);
+int128 operator%(int128 lhs, int128 rhs);
+int128 operator|(int128 lhs, int128 rhs);
+int128 operator&(int128 lhs, int128 rhs);
+int128 operator^(int128 lhs, int128 rhs);
+int128 operator<<(int128 lhs, int amount);
+int128 operator>>(int128 lhs, int amount);
+
+inline int128& int128::operator+=(int128 other) {
+  *this = *this + other;
+  return *this;
+}
+
+inline int128& int128::operator-=(int128 other) {
+  *this = *this - other;
+  return *this;
+}
+
+inline int128& int128::operator*=(int128 other) {
+  *this = *this * other;
+  return *this;
+}
+
+inline int128& int128::operator/=(int128 other) {
+  *this = *this / other;
+  return *this;
+}
+
+inline int128& int128::operator%=(int128 other) {
+  *this = *this % other;
+  return *this;
+}
+
+inline int128& int128::operator|=(int128 other) {
+  *this = *this | other;
+  return *this;
+}
+
+inline int128& int128::operator&=(int128 other) {
+  *this = *this & other;
+  return *this;
+}
+
+inline int128& int128::operator^=(int128 other) {
+  *this = *this ^ other;
+  return *this;
+}
+
+inline int128& int128::operator<<=(int amount) {
+  *this = *this << amount;
+  return *this;
+}
+
+inline int128& int128::operator>>=(int amount) {
+  *this = *this >> amount;
+  return *this;
+}
+
+namespace int128_internal {
+
+// Casts from unsigned to signed while preserving the underlying binary
+// representation.
+constexpr int64_t BitCastToSigned(uint64_t v) {
+  // Casting an unsigned integer to a signed integer of the same
+  // width is implementation defined behavior if the source value would not fit
+  // in the destination type. We step around it with a roundtrip bitwise not
+  // operation to make sure this function remains constexpr. Clang, GCC, and
+  // MSVC optimize this to a no-op on x86-64.
+  return v & (uint64_t{1} << 63) ? ~static_cast<int64_t>(~v)
+                                 : static_cast<int64_t>(v);
+}
+
+}  // namespace int128_internal
+
+#if defined(ABSL_HAVE_INTRINSIC_INT128)
+#include "absl/numeric/int128_have_intrinsic.inc"  // IWYU pragma: export
+#else  // ABSL_HAVE_INTRINSIC_INT128
+#include "absl/numeric/int128_no_intrinsic.inc"  // IWYU pragma: export
+#endif  // ABSL_HAVE_INTRINSIC_INT128
+
+ABSL_NAMESPACE_END
+}  // namespace absl
+
+#undef ABSL_INTERNAL_WCHAR_T
+
+#endif  // ABSL_NUMERIC_INT128_H_
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128_have_intrinsic.inc b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128_have_intrinsic.inc
new file mode 100644
index 0000000..d6c76dd
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128_have_intrinsic.inc
@@ -0,0 +1,302 @@
+//
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// This file contains :int128 implementation details that depend on internal
+// representation when ABSL_HAVE_INTRINSIC_INT128 is defined. This file is
+// included by int128.h and relies on ABSL_INTERNAL_WCHAR_T being defined.
+
+namespace int128_internal {
+
+// Casts from unsigned to signed while preserving the underlying binary
+// representation.
+constexpr __int128 BitCastToSigned(unsigned __int128 v) {
+  // Casting an unsigned integer to a signed integer of the same
+  // width is implementation defined behavior if the source value would not fit
+  // in the destination type. We step around it with a roundtrip bitwise not
+  // operation to make sure this function remains constexpr. Clang and GCC
+  // optimize this to a no-op on x86-64.
+  return v & (static_cast<unsigned __int128>(1) << 127)
+             ? ~static_cast<__int128>(~v)
+             : static_cast<__int128>(v);
+}
+
+}  // namespace int128_internal
+
+inline int128& int128::operator=(__int128 v) {
+  v_ = v;
+  return *this;
+}
+
+constexpr uint64_t Int128Low64(int128 v) {
+  return static_cast<uint64_t>(v.v_ & ~uint64_t{0});
+}
+
+constexpr int64_t Int128High64(int128 v) {
+  // Initially cast to unsigned to prevent a right shift on a negative value.
+  return int128_internal::BitCastToSigned(
+      static_cast<uint64_t>(static_cast<unsigned __int128>(v.v_) >> 64));
+}
+
+constexpr int128::int128(int64_t high, uint64_t low)
+    // Initially cast to unsigned to prevent a left shift that overflows.
+    : v_(int128_internal::BitCastToSigned(static_cast<unsigned __int128>(high)
+                                           << 64) |
+         low) {}
+
+
+constexpr int128::int128(int v) : v_{v} {}
+
+constexpr int128::int128(long v) : v_{v} {}       // NOLINT(runtime/int)
+
+constexpr int128::int128(long long v) : v_{v} {}  // NOLINT(runtime/int)
+
+constexpr int128::int128(__int128 v) : v_{v} {}
+
+constexpr int128::int128(unsigned int v) : v_{v} {}
+
+constexpr int128::int128(unsigned long v) : v_{v} {}  // NOLINT(runtime/int)
+
+// NOLINTNEXTLINE(runtime/int)
+constexpr int128::int128(unsigned long long v) : v_{v} {}
+
+constexpr int128::int128(unsigned __int128 v) : v_{static_cast<__int128>(v)} {}
+
+inline int128::int128(float v) {
+  v_ = static_cast<__int128>(v);
+}
+
+inline int128::int128(double v) {
+  v_ = static_cast<__int128>(v);
+}
+
+inline int128::int128(long double v) {
+  v_ = static_cast<__int128>(v);
+}
+
+constexpr int128::int128(uint128 v) : v_{static_cast<__int128>(v)} {}
+
+constexpr int128::operator bool() const { return static_cast<bool>(v_); }
+
+constexpr int128::operator char() const { return static_cast<char>(v_); }
+
+constexpr int128::operator signed char() const {
+  return static_cast<signed char>(v_);
+}
+
+constexpr int128::operator unsigned char() const {
+  return static_cast<unsigned char>(v_);
+}
+
+constexpr int128::operator char16_t() const {
+  return static_cast<char16_t>(v_);
+}
+
+constexpr int128::operator char32_t() const {
+  return static_cast<char32_t>(v_);
+}
+
+constexpr int128::operator ABSL_INTERNAL_WCHAR_T() const {
+  return static_cast<ABSL_INTERNAL_WCHAR_T>(v_);
+}
+
+constexpr int128::operator short() const {  // NOLINT(runtime/int)
+  return static_cast<short>(v_);            // NOLINT(runtime/int)
+}
+
+constexpr int128::operator unsigned short() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned short>(v_);            // NOLINT(runtime/int)
+}
+
+constexpr int128::operator int() const {
+  return static_cast<int>(v_);
+}
+
+constexpr int128::operator unsigned int() const {
+  return static_cast<unsigned int>(v_);
+}
+
+constexpr int128::operator long() const {  // NOLINT(runtime/int)
+  return static_cast<long>(v_);            // NOLINT(runtime/int)
+}
+
+constexpr int128::operator unsigned long() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned long>(v_);            // NOLINT(runtime/int)
+}
+
+constexpr int128::operator long long() const {  // NOLINT(runtime/int)
+  return static_cast<long long>(v_);            // NOLINT(runtime/int)
+}
+
+constexpr int128::operator unsigned long long() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned long long>(v_);            // NOLINT(runtime/int)
+}
+
+constexpr int128::operator __int128() const { return v_; }
+
+constexpr int128::operator unsigned __int128() const {
+  return static_cast<unsigned __int128>(v_);
+}
+
+// Clang on PowerPC sometimes produces incorrect __int128 to floating point
+// conversions. In that case, we do the conversion with a similar implementation
+// to the conversion operators in int128_no_intrinsic.inc.
+#if defined(__clang__) && !defined(__ppc64__)
+inline int128::operator float() const { return static_cast<float>(v_); }
+
+inline int128::operator double () const { return static_cast<double>(v_); }
+
+inline int128::operator long double() const {
+  return static_cast<long double>(v_);
+}
+
+#else  // Clang on PowerPC
+// Forward declaration for conversion operators to floating point types.
+int128 operator-(int128 v);
+bool operator!=(int128 lhs, int128 rhs);
+
+inline int128::operator float() const {
+  // We must convert the absolute value and then negate as needed, because
+  // floating point types are typically sign-magnitude. Otherwise, the
+  // difference between the high and low 64 bits when interpreted as two's
+  // complement overwhelms the precision of the mantissa.
+  //
+  // Also check to make sure we don't negate Int128Min()
+  return v_ < 0 && *this != Int128Min()
+             ? -static_cast<float>(-*this)
+             : static_cast<float>(Int128Low64(*this)) +
+                   std::ldexp(static_cast<float>(Int128High64(*this)), 64);
+}
+
+inline int128::operator double() const {
+  // See comment in int128::operator float() above.
+  return v_ < 0 && *this != Int128Min()
+             ? -static_cast<double>(-*this)
+             : static_cast<double>(Int128Low64(*this)) +
+                   std::ldexp(static_cast<double>(Int128High64(*this)), 64);
+}
+
+inline int128::operator long double() const {
+  // See comment in int128::operator float() above.
+  return v_ < 0 && *this != Int128Min()
+             ? -static_cast<long double>(-*this)
+             : static_cast<long double>(Int128Low64(*this)) +
+                   std::ldexp(static_cast<long double>(Int128High64(*this)),
+                              64);
+}
+#endif  // Clang on PowerPC
+
+// Comparison operators.
+
+inline bool operator==(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) == static_cast<__int128>(rhs);
+}
+
+inline bool operator!=(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) != static_cast<__int128>(rhs);
+}
+
+inline bool operator<(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) < static_cast<__int128>(rhs);
+}
+
+inline bool operator>(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) > static_cast<__int128>(rhs);
+}
+
+inline bool operator<=(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) <= static_cast<__int128>(rhs);
+}
+
+inline bool operator>=(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) >= static_cast<__int128>(rhs);
+}
+
+// Unary operators.
+
+inline int128 operator-(int128 v) {
+  return -static_cast<__int128>(v);
+}
+
+inline bool operator!(int128 v) {
+  return !static_cast<__int128>(v);
+}
+
+inline int128 operator~(int128 val) {
+  return ~static_cast<__int128>(val);
+}
+
+// Arithmetic operators.
+
+inline int128 operator+(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) + static_cast<__int128>(rhs);
+}
+
+inline int128 operator-(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) - static_cast<__int128>(rhs);
+}
+
+inline int128 operator*(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) * static_cast<__int128>(rhs);
+}
+
+inline int128 operator/(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) / static_cast<__int128>(rhs);
+}
+
+inline int128 operator%(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) % static_cast<__int128>(rhs);
+}
+
+inline int128 int128::operator++(int) {
+  int128 tmp(*this);
+  ++v_;
+  return tmp;
+}
+
+inline int128 int128::operator--(int) {
+  int128 tmp(*this);
+  --v_;
+  return tmp;
+}
+
+inline int128& int128::operator++() {
+  ++v_;
+  return *this;
+}
+
+inline int128& int128::operator--() {
+  --v_;
+  return *this;
+}
+
+inline int128 operator|(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) | static_cast<__int128>(rhs);
+}
+
+inline int128 operator&(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) & static_cast<__int128>(rhs);
+}
+
+inline int128 operator^(int128 lhs, int128 rhs) {
+  return static_cast<__int128>(lhs) ^ static_cast<__int128>(rhs);
+}
+
+inline int128 operator<<(int128 lhs, int amount) {
+  return static_cast<__int128>(lhs) << amount;
+}
+
+inline int128 operator>>(int128 lhs, int amount) {
+  return static_cast<__int128>(lhs) >> amount;
+}
diff --git a/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128_no_intrinsic.inc b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128_no_intrinsic.inc
new file mode 100644
index 0000000..c753771
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/absl/absl/numeric/int128_no_intrinsic.inc
@@ -0,0 +1,308 @@
+//
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// This file contains :int128 implementation details that depend on internal
+// representation when ABSL_HAVE_INTRINSIC_INT128 is *not* defined. This file
+// is included by int128.h and relies on ABSL_INTERNAL_WCHAR_T being defined.
+
+constexpr uint64_t Int128Low64(int128 v) { return v.lo_; }
+
+constexpr int64_t Int128High64(int128 v) { return v.hi_; }
+
+#if defined(ABSL_IS_LITTLE_ENDIAN)
+
+constexpr int128::int128(int64_t high, uint64_t low) :
+    lo_(low), hi_(high) {}
+
+constexpr int128::int128(int v)
+    : lo_{static_cast<uint64_t>(v)}, hi_{v < 0 ? ~int64_t{0} : 0} {}
+constexpr int128::int128(long v)  // NOLINT(runtime/int)
+    : lo_{static_cast<uint64_t>(v)}, hi_{v < 0 ? ~int64_t{0} : 0} {}
+constexpr int128::int128(long long v)  // NOLINT(runtime/int)
+    : lo_{static_cast<uint64_t>(v)}, hi_{v < 0 ? ~int64_t{0} : 0} {}
+
+constexpr int128::int128(unsigned int v) : lo_{v}, hi_{0} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr int128::int128(unsigned long v) : lo_{v}, hi_{0} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr int128::int128(unsigned long long v) : lo_{v}, hi_{0} {}
+
+constexpr int128::int128(uint128 v)
+    : lo_{Uint128Low64(v)}, hi_{static_cast<int64_t>(Uint128High64(v))} {}
+
+#elif defined(ABSL_IS_BIG_ENDIAN)
+
+constexpr int128::int128(int64_t high, uint64_t low) :
+    hi_{high}, lo_{low} {}
+
+constexpr int128::int128(int v)
+    : hi_{v < 0 ? ~int64_t{0} : 0}, lo_{static_cast<uint64_t>(v)} {}
+constexpr int128::int128(long v)  // NOLINT(runtime/int)
+    : hi_{v < 0 ? ~int64_t{0} : 0}, lo_{static_cast<uint64_t>(v)} {}
+constexpr int128::int128(long long v)  // NOLINT(runtime/int)
+    : hi_{v < 0 ? ~int64_t{0} : 0}, lo_{static_cast<uint64_t>(v)} {}
+
+constexpr int128::int128(unsigned int v) : hi_{0}, lo_{v} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr int128::int128(unsigned long v) : hi_{0}, lo_{v} {}
+// NOLINTNEXTLINE(runtime/int)
+constexpr int128::int128(unsigned long long v) : hi_{0}, lo_{v} {}
+
+constexpr int128::int128(uint128 v)
+    : hi_{static_cast<int64_t>(Uint128High64(v))}, lo_{Uint128Low64(v)} {}
+
+#else  // byte order
+#error "Unsupported byte order: must be little-endian or big-endian."
+#endif  // byte order
+
+constexpr int128::operator bool() const { return lo_ || hi_; }
+
+constexpr int128::operator char() const {
+  // NOLINTNEXTLINE(runtime/int)
+  return static_cast<char>(static_cast<long long>(*this));
+}
+
+constexpr int128::operator signed char() const {
+  // NOLINTNEXTLINE(runtime/int)
+  return static_cast<signed char>(static_cast<long long>(*this));
+}
+
+constexpr int128::operator unsigned char() const {
+  return static_cast<unsigned char>(lo_);
+}
+
+constexpr int128::operator char16_t() const {
+  return static_cast<char16_t>(lo_);
+}
+
+constexpr int128::operator char32_t() const {
+  return static_cast<char32_t>(lo_);
+}
+
+constexpr int128::operator ABSL_INTERNAL_WCHAR_T() const {
+  // NOLINTNEXTLINE(runtime/int)
+  return static_cast<ABSL_INTERNAL_WCHAR_T>(static_cast<long long>(*this));
+}
+
+constexpr int128::operator short() const {  // NOLINT(runtime/int)
+  // NOLINTNEXTLINE(runtime/int)
+  return static_cast<short>(static_cast<long long>(*this));
+}
+
+constexpr int128::operator unsigned short() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned short>(lo_);           // NOLINT(runtime/int)
+}
+
+constexpr int128::operator int() const {
+  // NOLINTNEXTLINE(runtime/int)
+  return static_cast<int>(static_cast<long long>(*this));
+}
+
+constexpr int128::operator unsigned int() const {
+  return static_cast<unsigned int>(lo_);
+}
+
+constexpr int128::operator long() const {  // NOLINT(runtime/int)
+  // NOLINTNEXTLINE(runtime/int)
+  return static_cast<long>(static_cast<long long>(*this));
+}
+
+constexpr int128::operator unsigned long() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned long>(lo_);           // NOLINT(runtime/int)
+}
+
+constexpr int128::operator long long() const {  // NOLINT(runtime/int)
+  // We don't bother checking the value of hi_. If *this < 0, lo_'s high bit
+  // must be set in order for the value to fit into a long long. Conversely, if
+  // lo_'s high bit is set, *this must be < 0 for the value to fit.
+  return int128_internal::BitCastToSigned(lo_);
+}
+
+constexpr int128::operator unsigned long long() const {  // NOLINT(runtime/int)
+  return static_cast<unsigned long long>(lo_);           // NOLINT(runtime/int)
+}
+
+// Forward declaration for conversion operators to floating point types.
+int128 operator-(int128 v);
+bool operator!=(int128 lhs, int128 rhs);
+
+inline int128::operator float() const {
+  // We must convert the absolute value and then negate as needed, because
+  // floating point types are typically sign-magnitude. Otherwise, the
+  // difference between the high and low 64 bits when interpreted as two's
+  // complement overwhelms the precision of the mantissa.
+  //
+  // Also check to make sure we don't negate Int128Min()
+  return hi_ < 0 && *this != Int128Min()
+             ? -static_cast<float>(-*this)
+             : static_cast<float>(lo_) +
+                   std::ldexp(static_cast<float>(hi_), 64);
+}
+
+inline int128::operator double() const {
+  // See comment in int128::operator float() above.
+  return hi_ < 0 && *this != Int128Min()
+             ? -static_cast<double>(-*this)
+             : static_cast<double>(lo_) +
+                   std::ldexp(static_cast<double>(hi_), 64);
+}
+
+inline int128::operator long double() const {
+  // See comment in int128::operator float() above.
+  return hi_ < 0 && *this != Int128Min()
+             ? -static_cast<long double>(-*this)
+             : static_cast<long double>(lo_) +
+                   std::ldexp(static_cast<long double>(hi_), 64);
+}
+
+// Comparison operators.
+
+inline bool operator==(int128 lhs, int128 rhs) {
+  return (Int128Low64(lhs) == Int128Low64(rhs) &&
+          Int128High64(lhs) == Int128High64(rhs));
+}
+
+inline bool operator!=(int128 lhs, int128 rhs) {
+  return !(lhs == rhs);
+}
+
+inline bool operator<(int128 lhs, int128 rhs) {
+  return (Int128High64(lhs) == Int128High64(rhs))
+             ? (Int128Low64(lhs) < Int128Low64(rhs))
+             : (Int128High64(lhs) < Int128High64(rhs));
+}
+
+inline bool operator>(int128 lhs, int128 rhs) {
+  return (Int128High64(lhs) == Int128High64(rhs))
+             ? (Int128Low64(lhs) > Int128Low64(rhs))
+             : (Int128High64(lhs) > Int128High64(rhs));
+}
+
+inline bool operator<=(int128 lhs, int128 rhs) {
+  return !(lhs > rhs);
+}
+
+inline bool operator>=(int128 lhs, int128 rhs) {
+  return !(lhs < rhs);
+}
+
+// Unary operators.
+
+inline int128 operator-(int128 v) {
+  int64_t hi = ~Int128High64(v);
+  uint64_t lo = ~Int128Low64(v) + 1;
+  if (lo == 0) ++hi;  // carry
+  return MakeInt128(hi, lo);
+}
+
+inline bool operator!(int128 v) {
+  return !Int128Low64(v) && !Int128High64(v);
+}
+
+inline int128 operator~(int128 val) {
+  return MakeInt128(~Int128High64(val), ~Int128Low64(val));
+}
+
+// Arithmetic operators.
+
+inline int128 operator+(int128 lhs, int128 rhs) {
+  int128 result = MakeInt128(Int128High64(lhs) + Int128High64(rhs),
+                             Int128Low64(lhs) + Int128Low64(rhs));
+  if (Int128Low64(result) < Int128Low64(lhs)) {  // check for carry
+    return MakeInt128(Int128High64(result) + 1, Int128Low64(result));
+  }
+  return result;
+}
+
+inline int128 operator-(int128 lhs, int128 rhs) {
+  int128 result = MakeInt128(Int128High64(lhs) - Int128High64(rhs),
+                             Int128Low64(lhs) - Int128Low64(rhs));
+  if (Int128Low64(lhs) < Int128Low64(rhs)) {  // check for carry
+    return MakeInt128(Int128High64(result) - 1, Int128Low64(result));
+  }
+  return result;
+}
+
+inline int128 operator*(int128 lhs, int128 rhs) {
+  uint128 result = uint128(lhs) * rhs;
+  return MakeInt128(int128_internal::BitCastToSigned(Uint128High64(result)),
+                    Uint128Low64(result));
+}
+
+inline int128 int128::operator++(int) {
+  int128 tmp(*this);
+  *this += 1;
+  return tmp;
+}
+
+inline int128 int128::operator--(int) {
+  int128 tmp(*this);
+  *this -= 1;
+  return tmp;
+}
+
+inline int128& int128::operator++() {
+  *this += 1;
+  return *this;
+}
+
+inline int128& int128::operator--() {
+  *this -= 1;
+  return *this;
+}
+
+inline int128 operator|(int128 lhs, int128 rhs) {
+  return MakeInt128(Int128High64(lhs) | Int128High64(rhs),
+                    Int128Low64(lhs) | Int128Low64(rhs));
+}
+
+inline int128 operator&(int128 lhs, int128 rhs) {
+  return MakeInt128(Int128High64(lhs) & Int128High64(rhs),
+                    Int128Low64(lhs) & Int128Low64(rhs));
+}
+
+inline int128 operator^(int128 lhs, int128 rhs) {
+  return MakeInt128(Int128High64(lhs) ^ Int128High64(rhs),
+                    Int128Low64(lhs) ^ Int128Low64(rhs));
+}
+
+inline int128 operator<<(int128 lhs, int amount) {
+  // uint64_t shifts of >= 64 are undefined, so we need some special-casing.
+  if (amount < 64) {
+    if (amount != 0) {
+      return MakeInt128(
+          (Int128High64(lhs) << amount) |
+              static_cast<int64_t>(Int128Low64(lhs) >> (64 - amount)),
+          Int128Low64(lhs) << amount);
+    }
+    return lhs;
+  }
+  return MakeInt128(static_cast<int64_t>(Int128Low64(lhs) << (amount - 64)), 0);
+}
+
+inline int128 operator>>(int128 lhs, int amount) {
+  // uint64_t shifts of >= 64 are undefined, so we need some special-casing.
+  if (amount < 64) {
+    if (amount != 0) {
+      return MakeInt128(
+          Int128High64(lhs) >> amount,
+          (Int128Low64(lhs) >> amount) |
+              (static_cast<uint64_t>(Int128High64(lhs)) << (64 - amount)));
+    }
+    return lhs;
+  }
+  return MakeInt128(0,
+                    static_cast<uint64_t>(Int128High64(lhs) >> (amount - 64)));
+}
diff --git a/lib/clickhouse-cpp/contrib/cityhash/CMakeLists.txt b/lib/clickhouse-cpp/contrib/cityhash/CMakeLists.txt
deleted file mode 100644
index e31bc24..0000000
--- a/lib/clickhouse-cpp/contrib/cityhash/CMakeLists.txt
+++ /dev/null
@@ -1,5 +0,0 @@
-ADD_LIBRARY (cityhash-lib STATIC
-    city.cc
-)
-
-set_property(TARGET cityhash-lib PROPERTY POSITION_INDEPENDENT_CODE ON)
diff --git a/lib/clickhouse-cpp/contrib/cityhash/BUCK b/lib/clickhouse-cpp/contrib/cityhash/cityhash/BUCK
similarity index 100%
rename from lib/clickhouse-cpp/contrib/cityhash/BUCK
rename to lib/clickhouse-cpp/contrib/cityhash/cityhash/BUCK
diff --git a/lib/clickhouse-cpp/contrib/cityhash/cityhash/CMakeLists.txt b/lib/clickhouse-cpp/contrib/cityhash/cityhash/CMakeLists.txt
new file mode 100644
index 0000000..d131156
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/cityhash/cityhash/CMakeLists.txt
@@ -0,0 +1,7 @@
+ADD_LIBRARY (cityhash STATIC
+    city.cc
+)
+
+set_property(TARGET cityhash PROPERTY POSITION_INDEPENDENT_CODE ON)
+
+ADD_LIBRARY (cityhash::cityhash ALIAS cityhash)
diff --git a/lib/clickhouse-cpp/contrib/cityhash/COPYING b/lib/clickhouse-cpp/contrib/cityhash/cityhash/COPYING
similarity index 100%
rename from lib/clickhouse-cpp/contrib/cityhash/COPYING
rename to lib/clickhouse-cpp/contrib/cityhash/cityhash/COPYING
diff --git a/lib/clickhouse-cpp/contrib/cityhash/city.cc b/lib/clickhouse-cpp/contrib/cityhash/cityhash/city.cc
similarity index 99%
rename from lib/clickhouse-cpp/contrib/cityhash/city.cc
rename to lib/clickhouse-cpp/contrib/cityhash/cityhash/city.cc
index e157ea0..e4ec85b 100644
--- a/lib/clickhouse-cpp/contrib/cityhash/city.cc
+++ b/lib/clickhouse-cpp/contrib/cityhash/cityhash/city.cc
@@ -27,7 +27,9 @@
 // possible hash functions, by using SIMD instructions, or by
 // compromising on hash quality.
 
+#ifdef HAVE_CONFIG_H
 #include "config.h"
+#endif
 #include "city.h"
 
 #include <algorithm>
diff --git a/lib/clickhouse-cpp/contrib/cityhash/city.h b/lib/clickhouse-cpp/contrib/cityhash/cityhash/city.h
similarity index 100%
rename from lib/clickhouse-cpp/contrib/cityhash/city.h
rename to lib/clickhouse-cpp/contrib/cityhash/cityhash/city.h
diff --git a/lib/clickhouse-cpp/contrib/cityhash/citycrc.h b/lib/clickhouse-cpp/contrib/cityhash/cityhash/citycrc.h
similarity index 100%
rename from lib/clickhouse-cpp/contrib/cityhash/citycrc.h
rename to lib/clickhouse-cpp/contrib/cityhash/cityhash/citycrc.h
diff --git a/lib/clickhouse-cpp/contrib/cityhash/config.h b/lib/clickhouse-cpp/contrib/cityhash/config.h
deleted file mode 100644
index 8062998..0000000
--- a/lib/clickhouse-cpp/contrib/cityhash/config.h
+++ /dev/null
@@ -1,118 +0,0 @@
-/* config.h.  Generated from config.h.in by configure.  */
-/* config.h.in.  Generated from configure.ac by autoheader.  */
-
-/* Define if building universal (internal helper macro) */
-/* #undef AC_APPLE_UNIVERSAL_BUILD */
-
-/* Define to 1 if the compiler supports __builtin_expect. */
-#if WIN32 || WIN64
-#	define HAVE_BUILTIN_EXPECT 0
-#else
-#	define HAVE_BUILTIN_EXPECT 1
-#endif
-
-/* Define to 1 if you have the <dlfcn.h> header file. */
-#define HAVE_DLFCN_H 1
-
-/* Define to 1 if you have the <inttypes.h> header file. */
-#define HAVE_INTTYPES_H 1
-
-/* Define to 1 if you have the <memory.h> header file. */
-#define HAVE_MEMORY_H 1
-
-/* Define to 1 if you have the <stdint.h> header file. */
-#define HAVE_STDINT_H 1
-
-/* Define to 1 if you have the <stdlib.h> header file. */
-#define HAVE_STDLIB_H 1
-
-/* Define to 1 if you have the <strings.h> header file. */
-#define HAVE_STRINGS_H 1
-
-/* Define to 1 if you have the <string.h> header file. */
-#define HAVE_STRING_H 1
-
-/* Define to 1 if you have the <sys/stat.h> header file. */
-#define HAVE_SYS_STAT_H 1
-
-/* Define to 1 if you have the <sys/types.h> header file. */
-#define HAVE_SYS_TYPES_H 1
-
-/* Define to 1 if you have the <unistd.h> header file. */
-#define HAVE_UNISTD_H 1
-
-/* Define to the sub-directory in which libtool stores uninstalled libraries.
-   */
-#define LT_OBJDIR ".libs/"
-
-/* Define to the address where bug reports for this package should be sent. */
-#define PACKAGE_BUGREPORT "cityhash-discuss@googlegroups.com"
-
-/* Define to the full name of this package. */
-#define PACKAGE_NAME "CityHash"
-
-/* Define to the full name and version of this package. */
-#define PACKAGE_STRING "CityHash 1.0.2"
-
-/* Define to the one symbol short name of this package. */
-#define PACKAGE_TARNAME "cityhash"
-
-/* Define to the home page for this package. */
-#define PACKAGE_URL ""
-
-/* Define to the version of this package. */
-#define PACKAGE_VERSION "1.0.2"
-
-/* Define to 1 if you have the ANSI C header files. */
-#define STDC_HEADERS 1
-
-/* Define WORDS_BIGENDIAN to 1 if your processor stores words with the most
-   significant byte first (like Motorola and SPARC, unlike Intel). */
-#if defined AC_APPLE_UNIVERSAL_BUILD
-# if defined __BIG_ENDIAN__
-#  define WORDS_BIGENDIAN 1
-# endif
-#else
-# ifndef WORDS_BIGENDIAN
-/* #  undef WORDS_BIGENDIAN */
-# endif
-#endif
-
-/* Define for Solaris 2.5.1 so the uint32_t typedef from <sys/synch.h>,
-   <pthread.h>, or <semaphore.h> is not used. If the typedef were allowed, the
-   #define below would cause a syntax error. */
-/* #undef _UINT32_T */
-
-/* Define for Solaris 2.5.1 so the uint64_t typedef from <sys/synch.h>,
-   <pthread.h>, or <semaphore.h> is not used. If the typedef were allowed, the
-   #define below would cause a syntax error. */
-/* #undef _UINT64_T */
-
-/* Define for Solaris 2.5.1 so the uint8_t typedef from <sys/synch.h>,
-   <pthread.h>, or <semaphore.h> is not used. If the typedef were allowed, the
-   #define below would cause a syntax error. */
-/* #undef _UINT8_T */
-
-/* Define to `__inline__' or `__inline' if that's what the C compiler
-   calls it, or to nothing if 'inline' is not supported under any name.  */
-#ifndef __cplusplus
-/* #undef inline */
-#endif
-
-/* Define to `unsigned int' if <sys/types.h> does not define. */
-/* #undef size_t */
-
-/* Define to `int' if <sys/types.h> does not define. */
-/* #undef ssize_t */
-
-/* Define to the type of an unsigned integer type of width exactly 32 bits if
-   such a type exists and the standard includes do not define it. */
-/* #undef uint32_t */
-
-/* Define to the type of an unsigned integer type of width exactly 64 bits if
-   such a type exists and the standard includes do not define it. */
-/* #undef uint64_t */
-
-/* Define to the type of an unsigned integer type of width exactly 8 bits if
-   such a type exists and the standard includes do not define it. */
-/* #undef uint8_t */
diff --git a/lib/clickhouse-cpp/contrib/gtest/BUCK b/lib/clickhouse-cpp/contrib/gtest/BUCK
deleted file mode 100644
index d888dbe..0000000
--- a/lib/clickhouse-cpp/contrib/gtest/BUCK
+++ /dev/null
@@ -1,14 +0,0 @@
-cxx_library(
-  name = 'gtest',
-  # This is a bit weird, but this is how the tests use the headers... 
-  header_namespace = 'contrib/gtest',
-  exported_headers = subdir_glob([
-    ('', '*.h'),
-  ]),
-  srcs = glob([
-    '*.cc',
-  ]),
-  visibility = [
-    '//...',
-  ],
-)
diff --git a/lib/clickhouse-cpp/contrib/gtest/CMakeLists.txt b/lib/clickhouse-cpp/contrib/gtest/CMakeLists.txt
deleted file mode 100644
index b824308..0000000
--- a/lib/clickhouse-cpp/contrib/gtest/CMakeLists.txt
+++ /dev/null
@@ -1,3 +0,0 @@
-ADD_LIBRARY (gtest-lib STATIC
-    gtest-all.cc
-)
diff --git a/lib/clickhouse-cpp/contrib/gtest/gtest-all.cc b/lib/clickhouse-cpp/contrib/gtest/gtest-all.cc
deleted file mode 100644
index 213f9ad..0000000
--- a/lib/clickhouse-cpp/contrib/gtest/gtest-all.cc
+++ /dev/null
@@ -1,10403 +0,0 @@
-// Copyright 2008, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: mheule@google.com (Markus Heule)
-//
-// Google C++ Testing Framework (Google Test)
-//
-// Sometimes it's desirable to build Google Test by compiling a single file.
-// This file serves this purpose.
-
-// This line ensures that gtest.h can be compiled on its own, even
-// when it's fused.
-#include "gtest.h"
-
-// The following lines pull in the real gtest *.cc files.
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-//
-// The Google C++ Testing Framework (Google Test)
-
-// Copyright 2007, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-//
-// Utilities for testing Google Test itself and code that uses Google Test
-// (e.g. frameworks built on top of Google Test).
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_SPI_H_
-#define GTEST_INCLUDE_GTEST_GTEST_SPI_H_
-
-
-namespace testing {
-
-// This helper class can be used to mock out Google Test failure reporting
-// so that we can test Google Test or code that builds on Google Test.
-//
-// An object of this class appends a TestPartResult object to the
-// TestPartResultArray object given in the constructor whenever a Google Test
-// failure is reported. It can either intercept only failures that are
-// generated in the same thread that created this object or it can intercept
-// all generated failures. The scope of this mock object can be controlled with
-// the second argument to the two arguments constructor.
-class GTEST_API_ ScopedFakeTestPartResultReporter
-    : public TestPartResultReporterInterface {
- public:
-  // The two possible mocking modes of this object.
-  enum InterceptMode {
-    INTERCEPT_ONLY_CURRENT_THREAD,  // Intercepts only thread local failures.
-    INTERCEPT_ALL_THREADS           // Intercepts all failures.
-  };
-
-  // The c'tor sets this object as the test part result reporter used
-  // by Google Test.  The 'result' parameter specifies where to report the
-  // results. This reporter will only catch failures generated in the current
-  // thread. DEPRECATED
-  explicit ScopedFakeTestPartResultReporter(TestPartResultArray* result);
-
-  // Same as above, but you can choose the interception scope of this object.
-  ScopedFakeTestPartResultReporter(InterceptMode intercept_mode,
-                                   TestPartResultArray* result);
-
-  // The d'tor restores the previous test part result reporter.
-  virtual ~ScopedFakeTestPartResultReporter();
-
-  // Appends the TestPartResult object to the TestPartResultArray
-  // received in the constructor.
-  //
-  // This method is from the TestPartResultReporterInterface
-  // interface.
-  virtual void ReportTestPartResult(const TestPartResult& result);
- private:
-  void Init();
-
-  const InterceptMode intercept_mode_;
-  TestPartResultReporterInterface* old_reporter_;
-  TestPartResultArray* const result_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ScopedFakeTestPartResultReporter);
-};
-
-namespace internal {
-
-// A helper class for implementing EXPECT_FATAL_FAILURE() and
-// EXPECT_NONFATAL_FAILURE().  Its destructor verifies that the given
-// TestPartResultArray contains exactly one failure that has the given
-// type and contains the given substring.  If that's not the case, a
-// non-fatal failure will be generated.
-class GTEST_API_ SingleFailureChecker {
- public:
-  // The constructor remembers the arguments.
-  SingleFailureChecker(const TestPartResultArray* results,
-                       TestPartResult::Type type,
-                       const string& substr);
-  ~SingleFailureChecker();
- private:
-  const TestPartResultArray* const results_;
-  const TestPartResult::Type type_;
-  const string substr_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(SingleFailureChecker);
-};
-
-}  // namespace internal
-
-}  // namespace testing
-
-// A set of macros for testing Google Test assertions or code that's expected
-// to generate Google Test fatal failures.  It verifies that the given
-// statement will cause exactly one fatal Google Test failure with 'substr'
-// being part of the failure message.
-//
-// There are two different versions of this macro. EXPECT_FATAL_FAILURE only
-// affects and considers failures generated in the current thread and
-// EXPECT_FATAL_FAILURE_ON_ALL_THREADS does the same but for all threads.
-//
-// The verification of the assertion is done correctly even when the statement
-// throws an exception or aborts the current function.
-//
-// Known restrictions:
-//   - 'statement' cannot reference local non-static variables or
-//     non-static members of the current object.
-//   - 'statement' cannot return a value.
-//   - You cannot stream a failure message to this macro.
-//
-// Note that even though the implementations of the following two
-// macros are much alike, we cannot refactor them to use a common
-// helper macro, due to some peculiarity in how the preprocessor
-// works.  The AcceptsMacroThatExpandsToUnprotectedComma test in
-// gtest_unittest.cc will fail to compile if we do that.
-#define EXPECT_FATAL_FAILURE(statement, substr) \
-  do { \
-    class GTestExpectFatalFailureHelper {\
-     public:\
-      static void Execute() { statement; }\
-    };\
-    ::testing::TestPartResultArray gtest_failures;\
-    ::testing::internal::SingleFailureChecker gtest_checker(\
-        &gtest_failures, ::testing::TestPartResult::kFatalFailure, (substr));\
-    {\
-      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
-          ::testing::ScopedFakeTestPartResultReporter:: \
-          INTERCEPT_ONLY_CURRENT_THREAD, &gtest_failures);\
-      GTestExpectFatalFailureHelper::Execute();\
-    }\
-  } while (::testing::internal::AlwaysFalse())
-
-#define EXPECT_FATAL_FAILURE_ON_ALL_THREADS(statement, substr) \
-  do { \
-    class GTestExpectFatalFailureHelper {\
-     public:\
-      static void Execute() { statement; }\
-    };\
-    ::testing::TestPartResultArray gtest_failures;\
-    ::testing::internal::SingleFailureChecker gtest_checker(\
-        &gtest_failures, ::testing::TestPartResult::kFatalFailure, (substr));\
-    {\
-      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
-          ::testing::ScopedFakeTestPartResultReporter:: \
-          INTERCEPT_ALL_THREADS, &gtest_failures);\
-      GTestExpectFatalFailureHelper::Execute();\
-    }\
-  } while (::testing::internal::AlwaysFalse())
-
-// A macro for testing Google Test assertions or code that's expected to
-// generate Google Test non-fatal failures.  It asserts that the given
-// statement will cause exactly one non-fatal Google Test failure with 'substr'
-// being part of the failure message.
-//
-// There are two different versions of this macro. EXPECT_NONFATAL_FAILURE only
-// affects and considers failures generated in the current thread and
-// EXPECT_NONFATAL_FAILURE_ON_ALL_THREADS does the same but for all threads.
-//
-// 'statement' is allowed to reference local variables and members of
-// the current object.
-//
-// The verification of the assertion is done correctly even when the statement
-// throws an exception or aborts the current function.
-//
-// Known restrictions:
-//   - You cannot stream a failure message to this macro.
-//
-// Note that even though the implementations of the following two
-// macros are much alike, we cannot refactor them to use a common
-// helper macro, due to some peculiarity in how the preprocessor
-// works.  If we do that, the code won't compile when the user gives
-// EXPECT_NONFATAL_FAILURE() a statement that contains a macro that
-// expands to code containing an unprotected comma.  The
-// AcceptsMacroThatExpandsToUnprotectedComma test in gtest_unittest.cc
-// catches that.
-//
-// For the same reason, we have to write
-//   if (::testing::internal::AlwaysTrue()) { statement; }
-// instead of
-//   GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement)
-// to avoid an MSVC warning on unreachable code.
-#define EXPECT_NONFATAL_FAILURE(statement, substr) \
-  do {\
-    ::testing::TestPartResultArray gtest_failures;\
-    ::testing::internal::SingleFailureChecker gtest_checker(\
-        &gtest_failures, ::testing::TestPartResult::kNonFatalFailure, \
-        (substr));\
-    {\
-      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
-          ::testing::ScopedFakeTestPartResultReporter:: \
-          INTERCEPT_ONLY_CURRENT_THREAD, &gtest_failures);\
-      if (::testing::internal::AlwaysTrue()) { statement; }\
-    }\
-  } while (::testing::internal::AlwaysFalse())
-
-#define EXPECT_NONFATAL_FAILURE_ON_ALL_THREADS(statement, substr) \
-  do {\
-    ::testing::TestPartResultArray gtest_failures;\
-    ::testing::internal::SingleFailureChecker gtest_checker(\
-        &gtest_failures, ::testing::TestPartResult::kNonFatalFailure, \
-        (substr));\
-    {\
-      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
-          ::testing::ScopedFakeTestPartResultReporter::INTERCEPT_ALL_THREADS, \
-          &gtest_failures);\
-      if (::testing::internal::AlwaysTrue()) { statement; }\
-    }\
-  } while (::testing::internal::AlwaysFalse())
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_SPI_H_
-
-#include <ctype.h>
-#include <math.h>
-#include <stdarg.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <time.h>
-#include <wchar.h>
-#include <wctype.h>
-
-#include <algorithm>
-#include <iomanip>
-#include <limits>
-#include <list>
-#include <map>
-#include <ostream>  // NOLINT
-#include <sstream>
-#include <vector>
-
-#if GTEST_OS_LINUX
-
-// TODO(kenton@google.com): Use autoconf to detect availability of
-// gettimeofday().
-# define GTEST_HAS_GETTIMEOFDAY_ 1
-
-# include <fcntl.h>  // NOLINT
-# include <limits.h>  // NOLINT
-# include <sched.h>  // NOLINT
-// Declares vsnprintf().  This header is not available on Windows.
-# include <strings.h>  // NOLINT
-# include <sys/mman.h>  // NOLINT
-# include <sys/time.h>  // NOLINT
-# include <unistd.h>  // NOLINT
-# include <string>
-
-#elif GTEST_OS_SYMBIAN
-# define GTEST_HAS_GETTIMEOFDAY_ 1
-# include <sys/time.h>  // NOLINT
-
-#elif GTEST_OS_ZOS
-# define GTEST_HAS_GETTIMEOFDAY_ 1
-# include <sys/time.h>  // NOLINT
-
-// On z/OS we additionally need strings.h for strcasecmp.
-# include <strings.h>  // NOLINT
-
-#elif GTEST_OS_WINDOWS_MOBILE  // We are on Windows CE.
-
-# include <windows.h>  // NOLINT
-# undef min
-
-#elif GTEST_OS_WINDOWS  // We are on Windows proper.
-
-# include <io.h>  // NOLINT
-# include <sys/timeb.h>  // NOLINT
-# include <sys/types.h>  // NOLINT
-# include <sys/stat.h>  // NOLINT
-
-# if GTEST_OS_WINDOWS_MINGW
-// MinGW has gettimeofday() but not _ftime64().
-// TODO(kenton@google.com): Use autoconf to detect availability of
-//   gettimeofday().
-// TODO(kenton@google.com): There are other ways to get the time on
-//   Windows, like GetTickCount() or GetSystemTimeAsFileTime().  MinGW
-//   supports these.  consider using them instead.
-#  define GTEST_HAS_GETTIMEOFDAY_ 1
-#  include <sys/time.h>  // NOLINT
-# endif  // GTEST_OS_WINDOWS_MINGW
-
-// cpplint thinks that the header is already included, so we want to
-// silence it.
-# include <windows.h>  // NOLINT
-# undef min
-
-#else
-
-// Assume other platforms have gettimeofday().
-// TODO(kenton@google.com): Use autoconf to detect availability of
-//   gettimeofday().
-# define GTEST_HAS_GETTIMEOFDAY_ 1
-
-// cpplint thinks that the header is already included, so we want to
-// silence it.
-# include <sys/time.h>  // NOLINT
-# include <unistd.h>  // NOLINT
-
-#endif  // GTEST_OS_LINUX
-
-#if GTEST_HAS_EXCEPTIONS
-# include <stdexcept>
-#endif
-
-#if GTEST_CAN_STREAM_RESULTS_
-# include <arpa/inet.h>  // NOLINT
-# include <netdb.h>  // NOLINT
-# include <sys/socket.h>  // NOLINT
-# include <sys/types.h>  // NOLINT
-#endif
-
-// Indicates that this translation unit is part of Google Test's
-// implementation.  It must come before gtest-internal-inl.h is
-// included, or there will be a compiler error.  This trick is to
-// prevent a user from accidentally including gtest-internal-inl.h in
-// his code.
-#define GTEST_IMPLEMENTATION_ 1
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-// Utility functions and classes used by the Google C++ testing framework.
-//
-// Author: wan@google.com (Zhanyong Wan)
-//
-// This file contains purely Google Test's internal implementation.  Please
-// DO NOT #INCLUDE IT IN A USER PROGRAM.
-
-#ifndef GTEST_SRC_GTEST_INTERNAL_INL_H_
-#define GTEST_SRC_GTEST_INTERNAL_INL_H_
-
-// GTEST_IMPLEMENTATION_ is defined to 1 iff the current translation unit is
-// part of Google Test's implementation; otherwise it's undefined.
-#if !GTEST_IMPLEMENTATION_
-// If this file is included from the user's code, just say no.
-# error "gtest-internal-inl.h is part of Google Test's internal implementation."
-# error "It must not be included except by Google Test itself."
-#endif  // GTEST_IMPLEMENTATION_
-
-#ifndef _WIN32_WCE
-# include <errno.h>
-#endif  // !_WIN32_WCE
-#include <stddef.h>
-#include <stdlib.h>  // For strtoll/_strtoul64/malloc/free.
-#include <string.h>  // For memmove.
-
-#include <algorithm>
-#include <string>
-#include <vector>
-
-
-#if GTEST_CAN_STREAM_RESULTS_
-# include <arpa/inet.h>  // NOLINT
-# include <netdb.h>  // NOLINT
-#endif
-
-#if GTEST_OS_WINDOWS
-# include <windows.h>  // NOLINT
-#endif  // GTEST_OS_WINDOWS
-
-
-namespace testing {
-
-// Declares the flags.
-//
-// We don't want the users to modify this flag in the code, but want
-// Google Test's own unit tests to be able to access it. Therefore we
-// declare it here as opposed to in gtest.h.
-GTEST_DECLARE_bool_(death_test_use_fork);
-
-namespace internal {
-
-// The value of GetTestTypeId() as seen from within the Google Test
-// library.  This is solely for testing GetTestTypeId().
-GTEST_API_ extern const TypeId kTestTypeIdInGoogleTest;
-
-// Names of the flags (needed for parsing Google Test flags).
-const char kAlsoRunDisabledTestsFlag[] = "also_run_disabled_tests";
-const char kBreakOnFailureFlag[] = "break_on_failure";
-const char kCatchExceptionsFlag[] = "catch_exceptions";
-const char kColorFlag[] = "color";
-const char kFilterFlag[] = "filter";
-const char kListTestsFlag[] = "list_tests";
-const char kOutputFlag[] = "output";
-const char kPrintTimeFlag[] = "print_time";
-const char kRandomSeedFlag[] = "random_seed";
-const char kRepeatFlag[] = "repeat";
-const char kShuffleFlag[] = "shuffle";
-const char kStackTraceDepthFlag[] = "stack_trace_depth";
-const char kStreamResultToFlag[] = "stream_result_to";
-const char kThrowOnFailureFlag[] = "throw_on_failure";
-const char kFlagfileFlag[] = "flagfile";
-
-// A valid random seed must be in [1, kMaxRandomSeed].
-const int kMaxRandomSeed = 99999;
-
-// g_help_flag is true iff the --help flag or an equivalent form is
-// specified on the command line.
-GTEST_API_ extern bool g_help_flag;
-
-// Returns the current time in milliseconds.
-GTEST_API_ TimeInMillis GetTimeInMillis();
-
-// Returns true iff Google Test should use colors in the output.
-GTEST_API_ bool ShouldUseColor(bool stdout_is_tty);
-
-// Formats the given time in milliseconds as seconds.
-GTEST_API_ std::string FormatTimeInMillisAsSeconds(TimeInMillis ms);
-
-// Converts the given time in milliseconds to a date string in the ISO 8601
-// format, without the timezone information.  N.B.: due to the use the
-// non-reentrant localtime() function, this function is not thread safe.  Do
-// not use it in any code that can be called from multiple threads.
-GTEST_API_ std::string FormatEpochTimeInMillisAsIso8601(TimeInMillis ms);
-
-// Parses a string for an Int32 flag, in the form of "--flag=value".
-//
-// On success, stores the value of the flag in *value, and returns
-// true.  On failure, returns false without changing *value.
-GTEST_API_ bool ParseInt32Flag(
-    const char* str, const char* flag, Int32* value);
-
-// Returns a random seed in range [1, kMaxRandomSeed] based on the
-// given --gtest_random_seed flag value.
-inline int GetRandomSeedFromFlag(Int32 random_seed_flag) {
-  const unsigned int raw_seed = (random_seed_flag == 0) ?
-      static_cast<unsigned int>(GetTimeInMillis()) :
-      static_cast<unsigned int>(random_seed_flag);
-
-  // Normalizes the actual seed to range [1, kMaxRandomSeed] such that
-  // it's easy to type.
-  const int normalized_seed =
-      static_cast<int>((raw_seed - 1U) %
-                       static_cast<unsigned int>(kMaxRandomSeed)) + 1;
-  return normalized_seed;
-}
-
-// Returns the first valid random seed after 'seed'.  The behavior is
-// undefined if 'seed' is invalid.  The seed after kMaxRandomSeed is
-// considered to be 1.
-inline int GetNextRandomSeed(int seed) {
-  GTEST_CHECK_(1 <= seed && seed <= kMaxRandomSeed)
-      << "Invalid random seed " << seed << " - must be in [1, "
-      << kMaxRandomSeed << "].";
-  const int next_seed = seed + 1;
-  return (next_seed > kMaxRandomSeed) ? 1 : next_seed;
-}
-
-// This class saves the values of all Google Test flags in its c'tor, and
-// restores them in its d'tor.
-class GTestFlagSaver {
- public:
-  // The c'tor.
-  GTestFlagSaver() {
-    also_run_disabled_tests_ = GTEST_FLAG(also_run_disabled_tests);
-    break_on_failure_ = GTEST_FLAG(break_on_failure);
-    catch_exceptions_ = GTEST_FLAG(catch_exceptions);
-    color_ = GTEST_FLAG(color);
-    death_test_style_ = GTEST_FLAG(death_test_style);
-    death_test_use_fork_ = GTEST_FLAG(death_test_use_fork);
-    filter_ = GTEST_FLAG(filter);
-    internal_run_death_test_ = GTEST_FLAG(internal_run_death_test);
-    list_tests_ = GTEST_FLAG(list_tests);
-    output_ = GTEST_FLAG(output);
-    print_time_ = GTEST_FLAG(print_time);
-    random_seed_ = GTEST_FLAG(random_seed);
-    repeat_ = GTEST_FLAG(repeat);
-    shuffle_ = GTEST_FLAG(shuffle);
-    stack_trace_depth_ = GTEST_FLAG(stack_trace_depth);
-    stream_result_to_ = GTEST_FLAG(stream_result_to);
-    throw_on_failure_ = GTEST_FLAG(throw_on_failure);
-  }
-
-  // The d'tor is not virtual.  DO NOT INHERIT FROM THIS CLASS.
-  ~GTestFlagSaver() {
-    GTEST_FLAG(also_run_disabled_tests) = also_run_disabled_tests_;
-    GTEST_FLAG(break_on_failure) = break_on_failure_;
-    GTEST_FLAG(catch_exceptions) = catch_exceptions_;
-    GTEST_FLAG(color) = color_;
-    GTEST_FLAG(death_test_style) = death_test_style_;
-    GTEST_FLAG(death_test_use_fork) = death_test_use_fork_;
-    GTEST_FLAG(filter) = filter_;
-    GTEST_FLAG(internal_run_death_test) = internal_run_death_test_;
-    GTEST_FLAG(list_tests) = list_tests_;
-    GTEST_FLAG(output) = output_;
-    GTEST_FLAG(print_time) = print_time_;
-    GTEST_FLAG(random_seed) = random_seed_;
-    GTEST_FLAG(repeat) = repeat_;
-    GTEST_FLAG(shuffle) = shuffle_;
-    GTEST_FLAG(stack_trace_depth) = stack_trace_depth_;
-    GTEST_FLAG(stream_result_to) = stream_result_to_;
-    GTEST_FLAG(throw_on_failure) = throw_on_failure_;
-  }
-
- private:
-  // Fields for saving the original values of flags.
-  bool also_run_disabled_tests_;
-  bool break_on_failure_;
-  bool catch_exceptions_;
-  std::string color_;
-  std::string death_test_style_;
-  bool death_test_use_fork_;
-  std::string filter_;
-  std::string internal_run_death_test_;
-  bool list_tests_;
-  std::string output_;
-  bool print_time_;
-  internal::Int32 random_seed_;
-  internal::Int32 repeat_;
-  bool shuffle_;
-  internal::Int32 stack_trace_depth_;
-  std::string stream_result_to_;
-  bool throw_on_failure_;
-} GTEST_ATTRIBUTE_UNUSED_;
-
-// Converts a Unicode code point to a narrow string in UTF-8 encoding.
-// code_point parameter is of type UInt32 because wchar_t may not be
-// wide enough to contain a code point.
-// If the code_point is not a valid Unicode code point
-// (i.e. outside of Unicode range U+0 to U+10FFFF) it will be converted
-// to "(Invalid Unicode 0xXXXXXXXX)".
-GTEST_API_ std::string CodePointToUtf8(UInt32 code_point);
-
-// Converts a wide string to a narrow string in UTF-8 encoding.
-// The wide string is assumed to have the following encoding:
-//   UTF-16 if sizeof(wchar_t) == 2 (on Windows, Cygwin, Symbian OS)
-//   UTF-32 if sizeof(wchar_t) == 4 (on Linux)
-// Parameter str points to a null-terminated wide string.
-// Parameter num_chars may additionally limit the number
-// of wchar_t characters processed. -1 is used when the entire string
-// should be processed.
-// If the string contains code points that are not valid Unicode code points
-// (i.e. outside of Unicode range U+0 to U+10FFFF) they will be output
-// as '(Invalid Unicode 0xXXXXXXXX)'. If the string is in UTF16 encoding
-// and contains invalid UTF-16 surrogate pairs, values in those pairs
-// will be encoded as individual Unicode characters from Basic Normal Plane.
-GTEST_API_ std::string WideStringToUtf8(const wchar_t* str, int num_chars);
-
-// Reads the GTEST_SHARD_STATUS_FILE environment variable, and creates the file
-// if the variable is present. If a file already exists at this location, this
-// function will write over it. If the variable is present, but the file cannot
-// be created, prints an error and exits.
-void WriteToShardStatusFileIfNeeded();
-
-// Checks whether sharding is enabled by examining the relevant
-// environment variable values. If the variables are present,
-// but inconsistent (e.g., shard_index >= total_shards), prints
-// an error and exits. If in_subprocess_for_death_test, sharding is
-// disabled because it must only be applied to the original test
-// process. Otherwise, we could filter out death tests we intended to execute.
-GTEST_API_ bool ShouldShard(const char* total_shards_str,
-                            const char* shard_index_str,
-                            bool in_subprocess_for_death_test);
-
-// Parses the environment variable var as an Int32. If it is unset,
-// returns default_val. If it is not an Int32, prints an error and
-// and aborts.
-GTEST_API_ Int32 Int32FromEnvOrDie(const char* env_var, Int32 default_val);
-
-// Given the total number of shards, the shard index, and the test id,
-// returns true iff the test should be run on this shard. The test id is
-// some arbitrary but unique non-negative integer assigned to each test
-// method. Assumes that 0 <= shard_index < total_shards.
-GTEST_API_ bool ShouldRunTestOnShard(
-    int total_shards, int shard_index, int test_id);
-
-// STL container utilities.
-
-// Returns the number of elements in the given container that satisfy
-// the given predicate.
-template <class Container, typename Predicate>
-inline int CountIf(const Container& c, Predicate predicate) {
-  // Implemented as an explicit loop since std::count_if() in libCstd on
-  // Solaris has a non-standard signature.
-  int count = 0;
-  for (typename Container::const_iterator it = c.begin(); it != c.end(); ++it) {
-    if (predicate(*it))
-      ++count;
-  }
-  return count;
-}
-
-// Applies a function/functor to each element in the container.
-template <class Container, typename Functor>
-void ForEach(const Container& c, Functor functor) {
-  std::for_each(c.begin(), c.end(), functor);
-}
-
-// Returns the i-th element of the vector, or default_value if i is not
-// in range [0, v.size()).
-template <typename E>
-inline E GetElementOr(const std::vector<E>& v, int i, E default_value) {
-  return (i < 0 || i >= static_cast<int>(v.size())) ? default_value : v[i];
-}
-
-// Performs an in-place shuffle of a range of the vector's elements.
-// 'begin' and 'end' are element indices as an STL-style range;
-// i.e. [begin, end) are shuffled, where 'end' == size() means to
-// shuffle to the end of the vector.
-template <typename E>
-void ShuffleRange(internal::Random* random, int begin, int end,
-                  std::vector<E>* v) {
-  const int size = static_cast<int>(v->size());
-  GTEST_CHECK_(0 <= begin && begin <= size)
-      << "Invalid shuffle range start " << begin << ": must be in range [0, "
-      << size << "].";
-  GTEST_CHECK_(begin <= end && end <= size)
-      << "Invalid shuffle range finish " << end << ": must be in range ["
-      << begin << ", " << size << "].";
-
-  // Fisher-Yates shuffle, from
-  // http://en.wikipedia.org/wiki/Fisher-Yates_shuffle
-  for (int range_width = end - begin; range_width >= 2; range_width--) {
-    const int last_in_range = begin + range_width - 1;
-    const int selected = begin + random->Generate(range_width);
-    std::swap((*v)[selected], (*v)[last_in_range]);
-  }
-}
-
-// Performs an in-place shuffle of the vector's elements.
-template <typename E>
-inline void Shuffle(internal::Random* random, std::vector<E>* v) {
-  ShuffleRange(random, 0, static_cast<int>(v->size()), v);
-}
-
-// A function for deleting an object.  Handy for being used as a
-// functor.
-template <typename T>
-static void Delete(T* x) {
-  delete x;
-}
-
-// A predicate that checks the key of a TestProperty against a known key.
-//
-// TestPropertyKeyIs is copyable.
-class TestPropertyKeyIs {
- public:
-  // Constructor.
-  //
-  // TestPropertyKeyIs has NO default constructor.
-  explicit TestPropertyKeyIs(const std::string& key) : key_(key) {}
-
-  // Returns true iff the test name of test property matches on key_.
-  bool operator()(const TestProperty& test_property) const {
-    return test_property.key() == key_;
-  }
-
- private:
-  std::string key_;
-};
-
-// Class UnitTestOptions.
-//
-// This class contains functions for processing options the user
-// specifies when running the tests.  It has only static members.
-//
-// In most cases, the user can specify an option using either an
-// environment variable or a command line flag.  E.g. you can set the
-// test filter using either GTEST_FILTER or --gtest_filter.  If both
-// the variable and the flag are present, the latter overrides the
-// former.
-class GTEST_API_ UnitTestOptions {
- public:
-  // Functions for processing the gtest_output flag.
-
-  // Returns the output format, or "" for normal printed output.
-  static std::string GetOutputFormat();
-
-  // Returns the absolute path of the requested output file, or the
-  // default (test_detail.xml in the original working directory) if
-  // none was explicitly specified.
-  static std::string GetAbsolutePathToOutputFile();
-
-  // Functions for processing the gtest_filter flag.
-
-  // Returns true iff the wildcard pattern matches the string.  The
-  // first ':' or '\0' character in pattern marks the end of it.
-  //
-  // This recursive algorithm isn't very efficient, but is clear and
-  // works well enough for matching test names, which are short.
-  static bool PatternMatchesString(const char *pattern, const char *str);
-
-  // Returns true iff the user-specified filter matches the test case
-  // name and the test name.
-  static bool FilterMatchesTest(const std::string &test_case_name,
-                                const std::string &test_name);
-
-#if GTEST_OS_WINDOWS
-  // Function for supporting the gtest_catch_exception flag.
-
-  // Returns EXCEPTION_EXECUTE_HANDLER if Google Test should handle the
-  // given SEH exception, or EXCEPTION_CONTINUE_SEARCH otherwise.
-  // This function is useful as an __except condition.
-  static int GTestShouldProcessSEH(DWORD exception_code);
-#endif  // GTEST_OS_WINDOWS
-
-  // Returns true if "name" matches the ':' separated list of glob-style
-  // filters in "filter".
-  static bool MatchesFilter(const std::string& name, const char* filter);
-};
-
-// Returns the current application's name, removing directory path if that
-// is present.  Used by UnitTestOptions::GetOutputFile.
-GTEST_API_ FilePath GetCurrentExecutableName();
-
-// The role interface for getting the OS stack trace as a string.
-class OsStackTraceGetterInterface {
- public:
-  OsStackTraceGetterInterface() {}
-  virtual ~OsStackTraceGetterInterface() {}
-
-  // Returns the current OS stack trace as an std::string.  Parameters:
-  //
-  //   max_depth  - the maximum number of stack frames to be included
-  //                in the trace.
-  //   skip_count - the number of top frames to be skipped; doesn't count
-  //                against max_depth.
-  virtual string CurrentStackTrace(int max_depth, int skip_count) = 0;
-
-  // UponLeavingGTest() should be called immediately before Google Test calls
-  // user code. It saves some information about the current stack that
-  // CurrentStackTrace() will use to find and hide Google Test stack frames.
-  virtual void UponLeavingGTest() = 0;
-
-  // This string is inserted in place of stack frames that are part of
-  // Google Test's implementation.
-  static const char* const kElidedFramesMarker;
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(OsStackTraceGetterInterface);
-};
-
-// A working implementation of the OsStackTraceGetterInterface interface.
-class OsStackTraceGetter : public OsStackTraceGetterInterface {
- public:
-  OsStackTraceGetter() {}
-
-  virtual string CurrentStackTrace(int max_depth, int skip_count);
-  virtual void UponLeavingGTest();
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(OsStackTraceGetter);
-};
-
-// Information about a Google Test trace point.
-struct TraceInfo {
-  const char* file;
-  int line;
-  std::string message;
-};
-
-// This is the default global test part result reporter used in UnitTestImpl.
-// This class should only be used by UnitTestImpl.
-class DefaultGlobalTestPartResultReporter
-  : public TestPartResultReporterInterface {
- public:
-  explicit DefaultGlobalTestPartResultReporter(UnitTestImpl* unit_test);
-  // Implements the TestPartResultReporterInterface. Reports the test part
-  // result in the current test.
-  virtual void ReportTestPartResult(const TestPartResult& result);
-
- private:
-  UnitTestImpl* const unit_test_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(DefaultGlobalTestPartResultReporter);
-};
-
-// This is the default per thread test part result reporter used in
-// UnitTestImpl. This class should only be used by UnitTestImpl.
-class DefaultPerThreadTestPartResultReporter
-    : public TestPartResultReporterInterface {
- public:
-  explicit DefaultPerThreadTestPartResultReporter(UnitTestImpl* unit_test);
-  // Implements the TestPartResultReporterInterface. The implementation just
-  // delegates to the current global test part result reporter of *unit_test_.
-  virtual void ReportTestPartResult(const TestPartResult& result);
-
- private:
-  UnitTestImpl* const unit_test_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(DefaultPerThreadTestPartResultReporter);
-};
-
-// The private implementation of the UnitTest class.  We don't protect
-// the methods under a mutex, as this class is not accessible by a
-// user and the UnitTest class that delegates work to this class does
-// proper locking.
-class GTEST_API_ UnitTestImpl {
- public:
-  explicit UnitTestImpl(UnitTest* parent);
-  virtual ~UnitTestImpl();
-
-  // There are two different ways to register your own TestPartResultReporter.
-  // You can register your own repoter to listen either only for test results
-  // from the current thread or for results from all threads.
-  // By default, each per-thread test result repoter just passes a new
-  // TestPartResult to the global test result reporter, which registers the
-  // test part result for the currently running test.
-
-  // Returns the global test part result reporter.
-  TestPartResultReporterInterface* GetGlobalTestPartResultReporter();
-
-  // Sets the global test part result reporter.
-  void SetGlobalTestPartResultReporter(
-      TestPartResultReporterInterface* reporter);
-
-  // Returns the test part result reporter for the current thread.
-  TestPartResultReporterInterface* GetTestPartResultReporterForCurrentThread();
-
-  // Sets the test part result reporter for the current thread.
-  void SetTestPartResultReporterForCurrentThread(
-      TestPartResultReporterInterface* reporter);
-
-  // Gets the number of successful test cases.
-  int successful_test_case_count() const;
-
-  // Gets the number of failed test cases.
-  int failed_test_case_count() const;
-
-  // Gets the number of all test cases.
-  int total_test_case_count() const;
-
-  // Gets the number of all test cases that contain at least one test
-  // that should run.
-  int test_case_to_run_count() const;
-
-  // Gets the number of successful tests.
-  int successful_test_count() const;
-
-  // Gets the number of failed tests.
-  int failed_test_count() const;
-
-  // Gets the number of disabled tests that will be reported in the XML report.
-  int reportable_disabled_test_count() const;
-
-  // Gets the number of disabled tests.
-  int disabled_test_count() const;
-
-  // Gets the number of tests to be printed in the XML report.
-  int reportable_test_count() const;
-
-  // Gets the number of all tests.
-  int total_test_count() const;
-
-  // Gets the number of tests that should run.
-  int test_to_run_count() const;
-
-  // Gets the time of the test program start, in ms from the start of the
-  // UNIX epoch.
-  TimeInMillis start_timestamp() const { return start_timestamp_; }
-
-  // Gets the elapsed time, in milliseconds.
-  TimeInMillis elapsed_time() const { return elapsed_time_; }
-
-  // Returns true iff the unit test passed (i.e. all test cases passed).
-  bool Passed() const { return !Failed(); }
-
-  // Returns true iff the unit test failed (i.e. some test case failed
-  // or something outside of all tests failed).
-  bool Failed() const {
-    return failed_test_case_count() > 0 || ad_hoc_test_result()->Failed();
-  }
-
-  // Gets the i-th test case among all the test cases. i can range from 0 to
-  // total_test_case_count() - 1. If i is not in that range, returns NULL.
-  const TestCase* GetTestCase(int i) const {
-    const int index = GetElementOr(test_case_indices_, i, -1);
-    return index < 0 ? NULL : test_cases_[i];
-  }
-
-  // Gets the i-th test case among all the test cases. i can range from 0 to
-  // total_test_case_count() - 1. If i is not in that range, returns NULL.
-  TestCase* GetMutableTestCase(int i) {
-    const int index = GetElementOr(test_case_indices_, i, -1);
-    return index < 0 ? NULL : test_cases_[index];
-  }
-
-  // Provides access to the event listener list.
-  TestEventListeners* listeners() { return &listeners_; }
-
-  // Returns the TestResult for the test that's currently running, or
-  // the TestResult for the ad hoc test if no test is running.
-  TestResult* current_test_result();
-
-  // Returns the TestResult for the ad hoc test.
-  const TestResult* ad_hoc_test_result() const { return &ad_hoc_test_result_; }
-
-  // Sets the OS stack trace getter.
-  //
-  // Does nothing if the input and the current OS stack trace getter
-  // are the same; otherwise, deletes the old getter and makes the
-  // input the current getter.
-  void set_os_stack_trace_getter(OsStackTraceGetterInterface* getter);
-
-  // Returns the current OS stack trace getter if it is not NULL;
-  // otherwise, creates an OsStackTraceGetter, makes it the current
-  // getter, and returns it.
-  OsStackTraceGetterInterface* os_stack_trace_getter();
-
-  // Returns the current OS stack trace as an std::string.
-  //
-  // The maximum number of stack frames to be included is specified by
-  // the gtest_stack_trace_depth flag.  The skip_count parameter
-  // specifies the number of top frames to be skipped, which doesn't
-  // count against the number of frames to be included.
-  //
-  // For example, if Foo() calls Bar(), which in turn calls
-  // CurrentOsStackTraceExceptTop(1), Foo() will be included in the
-  // trace but Bar() and CurrentOsStackTraceExceptTop() won't.
-  std::string CurrentOsStackTraceExceptTop(int skip_count) GTEST_NO_INLINE_;
-
-  // Finds and returns a TestCase with the given name.  If one doesn't
-  // exist, creates one and returns it.
-  //
-  // Arguments:
-  //
-  //   test_case_name: name of the test case
-  //   type_param:     the name of the test's type parameter, or NULL if
-  //                   this is not a typed or a type-parameterized test.
-  //   set_up_tc:      pointer to the function that sets up the test case
-  //   tear_down_tc:   pointer to the function that tears down the test case
-  TestCase* GetTestCase(const char* test_case_name,
-                        const char* type_param,
-                        Test::SetUpTestCaseFunc set_up_tc,
-                        Test::TearDownTestCaseFunc tear_down_tc);
-
-  // Adds a TestInfo to the unit test.
-  //
-  // Arguments:
-  //
-  //   set_up_tc:    pointer to the function that sets up the test case
-  //   tear_down_tc: pointer to the function that tears down the test case
-  //   test_info:    the TestInfo object
-  void AddTestInfo(Test::SetUpTestCaseFunc set_up_tc,
-                   Test::TearDownTestCaseFunc tear_down_tc,
-                   TestInfo* test_info) {
-    // In order to support thread-safe death tests, we need to
-    // remember the original working directory when the test program
-    // was first invoked.  We cannot do this in RUN_ALL_TESTS(), as
-    // the user may have changed the current directory before calling
-    // RUN_ALL_TESTS().  Therefore we capture the current directory in
-    // AddTestInfo(), which is called to register a TEST or TEST_F
-    // before main() is reached.
-    if (original_working_dir_.IsEmpty()) {
-      original_working_dir_.Set(FilePath::GetCurrentDir());
-      GTEST_CHECK_(!original_working_dir_.IsEmpty())
-          << "Failed to get the current working directory.";
-    }
-
-    GetTestCase(test_info->test_case_name(),
-                test_info->type_param(),
-                set_up_tc,
-                tear_down_tc)->AddTestInfo(test_info);
-  }
-
-#if GTEST_HAS_PARAM_TEST
-  // Returns ParameterizedTestCaseRegistry object used to keep track of
-  // value-parameterized tests and instantiate and register them.
-  internal::ParameterizedTestCaseRegistry& parameterized_test_registry() {
-    return parameterized_test_registry_;
-  }
-#endif  // GTEST_HAS_PARAM_TEST
-
-  // Sets the TestCase object for the test that's currently running.
-  void set_current_test_case(TestCase* a_current_test_case) {
-    current_test_case_ = a_current_test_case;
-  }
-
-  // Sets the TestInfo object for the test that's currently running.  If
-  // current_test_info is NULL, the assertion results will be stored in
-  // ad_hoc_test_result_.
-  void set_current_test_info(TestInfo* a_current_test_info) {
-    current_test_info_ = a_current_test_info;
-  }
-
-  // Registers all parameterized tests defined using TEST_P and
-  // INSTANTIATE_TEST_CASE_P, creating regular tests for each test/parameter
-  // combination. This method can be called more then once; it has guards
-  // protecting from registering the tests more then once.  If
-  // value-parameterized tests are disabled, RegisterParameterizedTests is
-  // present but does nothing.
-  void RegisterParameterizedTests();
-
-  // Runs all tests in this UnitTest object, prints the result, and
-  // returns true if all tests are successful.  If any exception is
-  // thrown during a test, this test is considered to be failed, but
-  // the rest of the tests will still be run.
-  bool RunAllTests();
-
-  // Clears the results of all tests, except the ad hoc tests.
-  void ClearNonAdHocTestResult() {
-    ForEach(test_cases_, TestCase::ClearTestCaseResult);
-  }
-
-  // Clears the results of ad-hoc test assertions.
-  void ClearAdHocTestResult() {
-    ad_hoc_test_result_.Clear();
-  }
-
-  // Adds a TestProperty to the current TestResult object when invoked in a
-  // context of a test or a test case, or to the global property set. If the
-  // result already contains a property with the same key, the value will be
-  // updated.
-  void RecordProperty(const TestProperty& test_property);
-
-  enum ReactionToSharding {
-    HONOR_SHARDING_PROTOCOL,
-    IGNORE_SHARDING_PROTOCOL
-  };
-
-  // Matches the full name of each test against the user-specified
-  // filter to decide whether the test should run, then records the
-  // result in each TestCase and TestInfo object.
-  // If shard_tests == HONOR_SHARDING_PROTOCOL, further filters tests
-  // based on sharding variables in the environment.
-  // Returns the number of tests that should run.
-  int FilterTests(ReactionToSharding shard_tests);
-
-  // Prints the names of the tests matching the user-specified filter flag.
-  void ListTestsMatchingFilter();
-
-  const TestCase* current_test_case() const { return current_test_case_; }
-  TestInfo* current_test_info() { return current_test_info_; }
-  const TestInfo* current_test_info() const { return current_test_info_; }
-
-  // Returns the vector of environments that need to be set-up/torn-down
-  // before/after the tests are run.
-  std::vector<Environment*>& environments() { return environments_; }
-
-  // Getters for the per-thread Google Test trace stack.
-  std::vector<TraceInfo>& gtest_trace_stack() {
-    return *(gtest_trace_stack_.pointer());
-  }
-  const std::vector<TraceInfo>& gtest_trace_stack() const {
-    return gtest_trace_stack_.get();
-  }
-
-#if GTEST_HAS_DEATH_TEST
-  void InitDeathTestSubprocessControlInfo() {
-    internal_run_death_test_flag_.reset(ParseInternalRunDeathTestFlag());
-  }
-  // Returns a pointer to the parsed --gtest_internal_run_death_test
-  // flag, or NULL if that flag was not specified.
-  // This information is useful only in a death test child process.
-  // Must not be called before a call to InitGoogleTest.
-  const InternalRunDeathTestFlag* internal_run_death_test_flag() const {
-    return internal_run_death_test_flag_.get();
-  }
-
-  // Returns a pointer to the current death test factory.
-  internal::DeathTestFactory* death_test_factory() {
-    return death_test_factory_.get();
-  }
-
-  void SuppressTestEventsIfInSubprocess();
-
-  friend class ReplaceDeathTestFactory;
-#endif  // GTEST_HAS_DEATH_TEST
-
-  // Initializes the event listener performing XML output as specified by
-  // UnitTestOptions. Must not be called before InitGoogleTest.
-  void ConfigureXmlOutput();
-
-#if GTEST_CAN_STREAM_RESULTS_
-  // Initializes the event listener for streaming test results to a socket.
-  // Must not be called before InitGoogleTest.
-  void ConfigureStreamingOutput();
-#endif
-
-  // Performs initialization dependent upon flag values obtained in
-  // ParseGoogleTestFlagsOnly.  Is called from InitGoogleTest after the call to
-  // ParseGoogleTestFlagsOnly.  In case a user neglects to call InitGoogleTest
-  // this function is also called from RunAllTests.  Since this function can be
-  // called more than once, it has to be idempotent.
-  void PostFlagParsingInit();
-
-  // Gets the random seed used at the start of the current test iteration.
-  int random_seed() const { return random_seed_; }
-
-  // Gets the random number generator.
-  internal::Random* random() { return &random_; }
-
-  // Shuffles all test cases, and the tests within each test case,
-  // making sure that death tests are still run first.
-  void ShuffleTests();
-
-  // Restores the test cases and tests to their order before the first shuffle.
-  void UnshuffleTests();
-
-  // Returns the value of GTEST_FLAG(catch_exceptions) at the moment
-  // UnitTest::Run() starts.
-  bool catch_exceptions() const { return catch_exceptions_; }
-
- private:
-  friend class ::testing::UnitTest;
-
-  // Used by UnitTest::Run() to capture the state of
-  // GTEST_FLAG(catch_exceptions) at the moment it starts.
-  void set_catch_exceptions(bool value) { catch_exceptions_ = value; }
-
-  // The UnitTest object that owns this implementation object.
-  UnitTest* const parent_;
-
-  // The working directory when the first TEST() or TEST_F() was
-  // executed.
-  internal::FilePath original_working_dir_;
-
-  // The default test part result reporters.
-  DefaultGlobalTestPartResultReporter default_global_test_part_result_reporter_;
-  DefaultPerThreadTestPartResultReporter
-      default_per_thread_test_part_result_reporter_;
-
-  // Points to (but doesn't own) the global test part result reporter.
-  TestPartResultReporterInterface* global_test_part_result_repoter_;
-
-  // Protects read and write access to global_test_part_result_reporter_.
-  internal::Mutex global_test_part_result_reporter_mutex_;
-
-  // Points to (but doesn't own) the per-thread test part result reporter.
-  internal::ThreadLocal<TestPartResultReporterInterface*>
-      per_thread_test_part_result_reporter_;
-
-  // The vector of environments that need to be set-up/torn-down
-  // before/after the tests are run.
-  std::vector<Environment*> environments_;
-
-  // The vector of TestCases in their original order.  It owns the
-  // elements in the vector.
-  std::vector<TestCase*> test_cases_;
-
-  // Provides a level of indirection for the test case list to allow
-  // easy shuffling and restoring the test case order.  The i-th
-  // element of this vector is the index of the i-th test case in the
-  // shuffled order.
-  std::vector<int> test_case_indices_;
-
-#if GTEST_HAS_PARAM_TEST
-  // ParameterizedTestRegistry object used to register value-parameterized
-  // tests.
-  internal::ParameterizedTestCaseRegistry parameterized_test_registry_;
-
-  // Indicates whether RegisterParameterizedTests() has been called already.
-  bool parameterized_tests_registered_;
-#endif  // GTEST_HAS_PARAM_TEST
-
-  // Index of the last death test case registered.  Initially -1.
-  int last_death_test_case_;
-
-  // This points to the TestCase for the currently running test.  It
-  // changes as Google Test goes through one test case after another.
-  // When no test is running, this is set to NULL and Google Test
-  // stores assertion results in ad_hoc_test_result_.  Initially NULL.
-  TestCase* current_test_case_;
-
-  // This points to the TestInfo for the currently running test.  It
-  // changes as Google Test goes through one test after another.  When
-  // no test is running, this is set to NULL and Google Test stores
-  // assertion results in ad_hoc_test_result_.  Initially NULL.
-  TestInfo* current_test_info_;
-
-  // Normally, a user only writes assertions inside a TEST or TEST_F,
-  // or inside a function called by a TEST or TEST_F.  Since Google
-  // Test keeps track of which test is current running, it can
-  // associate such an assertion with the test it belongs to.
-  //
-  // If an assertion is encountered when no TEST or TEST_F is running,
-  // Google Test attributes the assertion result to an imaginary "ad hoc"
-  // test, and records the result in ad_hoc_test_result_.
-  TestResult ad_hoc_test_result_;
-
-  // The list of event listeners that can be used to track events inside
-  // Google Test.
-  TestEventListeners listeners_;
-
-  // The OS stack trace getter.  Will be deleted when the UnitTest
-  // object is destructed.  By default, an OsStackTraceGetter is used,
-  // but the user can set this field to use a custom getter if that is
-  // desired.
-  OsStackTraceGetterInterface* os_stack_trace_getter_;
-
-  // True iff PostFlagParsingInit() has been called.
-  bool post_flag_parse_init_performed_;
-
-  // The random number seed used at the beginning of the test run.
-  int random_seed_;
-
-  // Our random number generator.
-  internal::Random random_;
-
-  // The time of the test program start, in ms from the start of the
-  // UNIX epoch.
-  TimeInMillis start_timestamp_;
-
-  // How long the test took to run, in milliseconds.
-  TimeInMillis elapsed_time_;
-
-#if GTEST_HAS_DEATH_TEST
-  // The decomposed components of the gtest_internal_run_death_test flag,
-  // parsed when RUN_ALL_TESTS is called.
-  internal::scoped_ptr<InternalRunDeathTestFlag> internal_run_death_test_flag_;
-  internal::scoped_ptr<internal::DeathTestFactory> death_test_factory_;
-#endif  // GTEST_HAS_DEATH_TEST
-
-  // A per-thread stack of traces created by the SCOPED_TRACE() macro.
-  internal::ThreadLocal<std::vector<TraceInfo> > gtest_trace_stack_;
-
-  // The value of GTEST_FLAG(catch_exceptions) at the moment RunAllTests()
-  // starts.
-  bool catch_exceptions_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(UnitTestImpl);
-};  // class UnitTestImpl
-
-// Convenience function for accessing the global UnitTest
-// implementation object.
-inline UnitTestImpl* GetUnitTestImpl() {
-  return UnitTest::GetInstance()->impl();
-}
-
-#if GTEST_USES_SIMPLE_RE
-
-// Internal helper functions for implementing the simple regular
-// expression matcher.
-GTEST_API_ bool IsInSet(char ch, const char* str);
-GTEST_API_ bool IsAsciiDigit(char ch);
-GTEST_API_ bool IsAsciiPunct(char ch);
-GTEST_API_ bool IsRepeat(char ch);
-GTEST_API_ bool IsAsciiWhiteSpace(char ch);
-GTEST_API_ bool IsAsciiWordChar(char ch);
-GTEST_API_ bool IsValidEscape(char ch);
-GTEST_API_ bool AtomMatchesChar(bool escaped, char pattern, char ch);
-GTEST_API_ bool ValidateRegex(const char* regex);
-GTEST_API_ bool MatchRegexAtHead(const char* regex, const char* str);
-GTEST_API_ bool MatchRepetitionAndRegexAtHead(
-    bool escaped, char ch, char repeat, const char* regex, const char* str);
-GTEST_API_ bool MatchRegexAnywhere(const char* regex, const char* str);
-
-#endif  // GTEST_USES_SIMPLE_RE
-
-// Parses the command line for Google Test flags, without initializing
-// other parts of Google Test.
-GTEST_API_ void ParseGoogleTestFlagsOnly(int* argc, char** argv);
-GTEST_API_ void ParseGoogleTestFlagsOnly(int* argc, wchar_t** argv);
-
-#if GTEST_HAS_DEATH_TEST
-
-// Returns the message describing the last system error, regardless of the
-// platform.
-GTEST_API_ std::string GetLastErrnoDescription();
-
-// Attempts to parse a string into a positive integer pointed to by the
-// number parameter.  Returns true if that is possible.
-// GTEST_HAS_DEATH_TEST implies that we have ::std::string, so we can use
-// it here.
-template <typename Integer>
-bool ParseNaturalNumber(const ::std::string& str, Integer* number) {
-  // Fail fast if the given string does not begin with a digit;
-  // this bypasses strtoXXX's "optional leading whitespace and plus
-  // or minus sign" semantics, which are undesirable here.
-  if (str.empty() || !IsDigit(str[0])) {
-    return false;
-  }
-  errno = 0;
-
-  char* end;
-  // BiggestConvertible is the largest integer type that system-provided
-  // string-to-number conversion routines can return.
-
-# if GTEST_OS_WINDOWS && !defined(__GNUC__)
-
-  // MSVC and C++ Builder define __int64 instead of the standard long long.
-  typedef unsigned __int64 BiggestConvertible;
-  const BiggestConvertible parsed = _strtoui64(str.c_str(), &end, 10);
-
-# else
-
-  typedef unsigned long long BiggestConvertible;  // NOLINT
-  const BiggestConvertible parsed = strtoull(str.c_str(), &end, 10);
-
-# endif  // GTEST_OS_WINDOWS && !defined(__GNUC__)
-
-  const bool parse_success = *end == '\0' && errno == 0;
-
-  // TODO(vladl@google.com): Convert this to compile time assertion when it is
-  // available.
-  GTEST_CHECK_(sizeof(Integer) <= sizeof(parsed));
-
-  const Integer result = static_cast<Integer>(parsed);
-  if (parse_success && static_cast<BiggestConvertible>(result) == parsed) {
-    *number = result;
-    return true;
-  }
-  return false;
-}
-#endif  // GTEST_HAS_DEATH_TEST
-
-// TestResult contains some private methods that should be hidden from
-// Google Test user but are required for testing. This class allow our tests
-// to access them.
-//
-// This class is supplied only for the purpose of testing Google Test's own
-// constructs. Do not use it in user tests, either directly or indirectly.
-class TestResultAccessor {
- public:
-  static void RecordProperty(TestResult* test_result,
-                             const std::string& xml_element,
-                             const TestProperty& property) {
-    test_result->RecordProperty(xml_element, property);
-  }
-
-  static void ClearTestPartResults(TestResult* test_result) {
-    test_result->ClearTestPartResults();
-  }
-
-  static const std::vector<testing::TestPartResult>& test_part_results(
-      const TestResult& test_result) {
-    return test_result.test_part_results();
-  }
-};
-
-#if GTEST_CAN_STREAM_RESULTS_
-
-// Streams test results to the given port on the given host machine.
-class GTEST_API_ StreamingListener : public EmptyTestEventListener {
- public:
-  // Abstract base class for writing strings to a socket.
-  class AbstractSocketWriter {
-   public:
-    virtual ~AbstractSocketWriter() {}
-
-    // Sends a string to the socket.
-    virtual void Send(const string& message) = 0;
-
-    // Closes the socket.
-    virtual void CloseConnection() {}
-
-    // Sends a string and a newline to the socket.
-    void SendLn(const string& message) {
-      Send(message + "\n");
-    }
-  };
-
-  // Concrete class for actually writing strings to a socket.
-  class SocketWriter : public AbstractSocketWriter {
-   public:
-    SocketWriter(const string& host, const string& port)
-        : sockfd_(-1), host_name_(host), port_num_(port) {
-      MakeConnection();
-    }
-
-    virtual ~SocketWriter() {
-      if (sockfd_ != -1)
-        CloseConnection();
-    }
-
-    // Sends a string to the socket.
-    virtual void Send(const string& message) {
-      GTEST_CHECK_(sockfd_ != -1)
-          << "Send() can be called only when there is a connection.";
-
-      const int len = static_cast<int>(message.length());
-      if (write(sockfd_, message.c_str(), len) != len) {
-        GTEST_LOG_(WARNING)
-            << "stream_result_to: failed to stream to "
-            << host_name_ << ":" << port_num_;
-      }
-    }
-
-   private:
-    // Creates a client socket and connects to the server.
-    void MakeConnection();
-
-    // Closes the socket.
-    void CloseConnection() {
-      GTEST_CHECK_(sockfd_ != -1)
-          << "CloseConnection() can be called only when there is a connection.";
-
-      close(sockfd_);
-      sockfd_ = -1;
-    }
-
-    int sockfd_;  // socket file descriptor
-    const string host_name_;
-    const string port_num_;
-
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(SocketWriter);
-  };  // class SocketWriter
-
-  // Escapes '=', '&', '%', and '\n' characters in str as "%xx".
-  static string UrlEncode(const char* str);
-
-  StreamingListener(const string& host, const string& port)
-      : socket_writer_(new SocketWriter(host, port)) { Start(); }
-
-  explicit StreamingListener(AbstractSocketWriter* socket_writer)
-      : socket_writer_(socket_writer) { Start(); }
-
-  void OnTestProgramStart(const UnitTest& /* unit_test */) {
-    SendLn("event=TestProgramStart");
-  }
-
-  void OnTestProgramEnd(const UnitTest& unit_test) {
-    // Note that Google Test current only report elapsed time for each
-    // test iteration, not for the entire test program.
-    SendLn("event=TestProgramEnd&passed=" + FormatBool(unit_test.Passed()));
-
-    // Notify the streaming server to stop.
-    socket_writer_->CloseConnection();
-  }
-
-  void OnTestIterationStart(const UnitTest& /* unit_test */, int iteration) {
-    SendLn("event=TestIterationStart&iteration=" +
-           StreamableToString(iteration));
-  }
-
-  void OnTestIterationEnd(const UnitTest& unit_test, int /* iteration */) {
-    SendLn("event=TestIterationEnd&passed=" +
-           FormatBool(unit_test.Passed()) + "&elapsed_time=" +
-           StreamableToString(unit_test.elapsed_time()) + "ms");
-  }
-
-  void OnTestCaseStart(const TestCase& test_case) {
-    SendLn(std::string("event=TestCaseStart&name=") + test_case.name());
-  }
-
-  void OnTestCaseEnd(const TestCase& test_case) {
-    SendLn("event=TestCaseEnd&passed=" + FormatBool(test_case.Passed())
-           + "&elapsed_time=" + StreamableToString(test_case.elapsed_time())
-           + "ms");
-  }
-
-  void OnTestStart(const TestInfo& test_info) {
-    SendLn(std::string("event=TestStart&name=") + test_info.name());
-  }
-
-  void OnTestEnd(const TestInfo& test_info) {
-    SendLn("event=TestEnd&passed=" +
-           FormatBool((test_info.result())->Passed()) +
-           "&elapsed_time=" +
-           StreamableToString((test_info.result())->elapsed_time()) + "ms");
-  }
-
-  void OnTestPartResult(const TestPartResult& test_part_result) {
-    const char* file_name = test_part_result.file_name();
-    if (file_name == NULL)
-      file_name = "";
-    SendLn("event=TestPartResult&file=" + UrlEncode(file_name) +
-           "&line=" + StreamableToString(test_part_result.line_number()) +
-           "&message=" + UrlEncode(test_part_result.message()));
-  }
-
- private:
-  // Sends the given message and a newline to the socket.
-  void SendLn(const string& message) { socket_writer_->SendLn(message); }
-
-  // Called at the start of streaming to notify the receiver what
-  // protocol we are using.
-  void Start() { SendLn("gtest_streaming_protocol_version=1.0"); }
-
-  string FormatBool(bool value) { return value ? "1" : "0"; }
-
-  const scoped_ptr<AbstractSocketWriter> socket_writer_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(StreamingListener);
-};  // class StreamingListener
-
-#endif  // GTEST_CAN_STREAM_RESULTS_
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_SRC_GTEST_INTERNAL_INL_H_
-#undef GTEST_IMPLEMENTATION_
-
-#if GTEST_OS_WINDOWS
-# define vsnprintf _vsnprintf
-#endif  // GTEST_OS_WINDOWS
-
-namespace testing {
-
-using internal::CountIf;
-using internal::ForEach;
-using internal::GetElementOr;
-using internal::Shuffle;
-
-// Constants.
-
-// A test whose test case name or test name matches this filter is
-// disabled and not run.
-static const char kDisableTestFilter[] = "DISABLED_*:*/DISABLED_*";
-
-// A test case whose name matches this filter is considered a death
-// test case and will be run before test cases whose name doesn't
-// match this filter.
-static const char kDeathTestCaseFilter[] = "*DeathTest:*DeathTest/*";
-
-// A test filter that matches everything.
-static const char kUniversalFilter[] = "*";
-
-// The default output file for XML output.
-static const char kDefaultOutputFile[] = "test_detail.xml";
-
-// The environment variable name for the test shard index.
-static const char kTestShardIndex[] = "GTEST_SHARD_INDEX";
-// The environment variable name for the total number of test shards.
-static const char kTestTotalShards[] = "GTEST_TOTAL_SHARDS";
-// The environment variable name for the test shard status file.
-static const char kTestShardStatusFile[] = "GTEST_SHARD_STATUS_FILE";
-
-namespace internal {
-
-// The text used in failure messages to indicate the start of the
-// stack trace.
-const char kStackTraceMarker[] = "\nStack trace:\n";
-
-// g_help_flag is true iff the --help flag or an equivalent form is
-// specified on the command line.
-bool g_help_flag = false;
-
-}  // namespace internal
-
-static const char* GetDefaultFilter() {
-#ifdef GTEST_TEST_FILTER_ENV_VAR_
-  const char* const testbridge_test_only = getenv(GTEST_TEST_FILTER_ENV_VAR_);
-  if (testbridge_test_only != NULL) {
-    return testbridge_test_only;
-  }
-#endif  // GTEST_TEST_FILTER_ENV_VAR_
-  return kUniversalFilter;
-}
-
-GTEST_DEFINE_bool_(
-    also_run_disabled_tests,
-    internal::BoolFromGTestEnv("also_run_disabled_tests", false),
-    "Run disabled tests too, in addition to the tests normally being run.");
-
-GTEST_DEFINE_bool_(
-    break_on_failure,
-    internal::BoolFromGTestEnv("break_on_failure", false),
-    "True iff a failed assertion should be a debugger break-point.");
-
-GTEST_DEFINE_bool_(
-    catch_exceptions,
-    internal::BoolFromGTestEnv("catch_exceptions", true),
-    "True iff " GTEST_NAME_
-    " should catch exceptions and treat them as test failures.");
-
-GTEST_DEFINE_string_(
-    color,
-    internal::StringFromGTestEnv("color", "auto"),
-    "Whether to use colors in the output.  Valid values: yes, no, "
-    "and auto.  'auto' means to use colors if the output is "
-    "being sent to a terminal and the TERM environment variable "
-    "is set to a terminal type that supports colors.");
-
-GTEST_DEFINE_string_(
-    filter,
-    internal::StringFromGTestEnv("filter", GetDefaultFilter()),
-    "A colon-separated list of glob (not regex) patterns "
-    "for filtering the tests to run, optionally followed by a "
-    "'-' and a : separated list of negative patterns (tests to "
-    "exclude).  A test is run if it matches one of the positive "
-    "patterns and does not match any of the negative patterns.");
-
-GTEST_DEFINE_bool_(list_tests, false,
-                   "List all tests without running them.");
-
-GTEST_DEFINE_string_(
-    output,
-    internal::StringFromGTestEnv("output", ""),
-    "A format (currently must be \"xml\"), optionally followed "
-    "by a colon and an output file name or directory. A directory "
-    "is indicated by a trailing pathname separator. "
-    "Examples: \"xml:filename.xml\", \"xml::directoryname/\". "
-    "If a directory is specified, output files will be created "
-    "within that directory, with file-names based on the test "
-    "executable's name and, if necessary, made unique by adding "
-    "digits.");
-
-GTEST_DEFINE_bool_(
-    print_time,
-    internal::BoolFromGTestEnv("print_time", true),
-    "True iff " GTEST_NAME_
-    " should display elapsed time in text output.");
-
-GTEST_DEFINE_int32_(
-    random_seed,
-    internal::Int32FromGTestEnv("random_seed", 0),
-    "Random number seed to use when shuffling test orders.  Must be in range "
-    "[1, 99999], or 0 to use a seed based on the current time.");
-
-GTEST_DEFINE_int32_(
-    repeat,
-    internal::Int32FromGTestEnv("repeat", 1),
-    "How many times to repeat each test.  Specify a negative number "
-    "for repeating forever.  Useful for shaking out flaky tests.");
-
-GTEST_DEFINE_bool_(
-    show_internal_stack_frames, false,
-    "True iff " GTEST_NAME_ " should include internal stack frames when "
-    "printing test failure stack traces.");
-
-GTEST_DEFINE_bool_(
-    shuffle,
-    internal::BoolFromGTestEnv("shuffle", false),
-    "True iff " GTEST_NAME_
-    " should randomize tests' order on every run.");
-
-GTEST_DEFINE_int32_(
-    stack_trace_depth,
-    internal::Int32FromGTestEnv("stack_trace_depth", kMaxStackTraceDepth),
-    "The maximum number of stack frames to print when an "
-    "assertion fails.  The valid range is 0 through 100, inclusive.");
-
-GTEST_DEFINE_string_(
-    stream_result_to,
-    internal::StringFromGTestEnv("stream_result_to", ""),
-    "This flag specifies the host name and the port number on which to stream "
-    "test results. Example: \"localhost:555\". The flag is effective only on "
-    "Linux.");
-
-GTEST_DEFINE_bool_(
-    throw_on_failure,
-    internal::BoolFromGTestEnv("throw_on_failure", false),
-    "When this flag is specified, a failed assertion will throw an exception "
-    "if exceptions are enabled or exit the program with a non-zero code "
-    "otherwise.");
-
-#if GTEST_USE_OWN_FLAGFILE_FLAG_
-GTEST_DEFINE_string_(
-    flagfile,
-    internal::StringFromGTestEnv("flagfile", ""),
-    "This flag specifies the flagfile to read command-line flags from.");
-#endif  // GTEST_USE_OWN_FLAGFILE_FLAG_
-
-namespace internal {
-
-// Generates a random number from [0, range), using a Linear
-// Congruential Generator (LCG).  Crashes if 'range' is 0 or greater
-// than kMaxRange.
-UInt32 Random::Generate(UInt32 range) {
-  // These constants are the same as are used in glibc's rand(3).
-  state_ = (1103515245U*state_ + 12345U) % kMaxRange;
-
-  GTEST_CHECK_(range > 0)
-      << "Cannot generate a number in the range [0, 0).";
-  GTEST_CHECK_(range <= kMaxRange)
-      << "Generation of a number in [0, " << range << ") was requested, "
-      << "but this can only generate numbers in [0, " << kMaxRange << ").";
-
-  // Converting via modulus introduces a bit of downward bias, but
-  // it's simple, and a linear congruential generator isn't too good
-  // to begin with.
-  return state_ % range;
-}
-
-// GTestIsInitialized() returns true iff the user has initialized
-// Google Test.  Useful for catching the user mistake of not initializing
-// Google Test before calling RUN_ALL_TESTS().
-static bool GTestIsInitialized() { return GetArgvs().size() > 0; }
-
-// Iterates over a vector of TestCases, keeping a running sum of the
-// results of calling a given int-returning method on each.
-// Returns the sum.
-static int SumOverTestCaseList(const std::vector<TestCase*>& case_list,
-                               int (TestCase::*method)() const) {
-  int sum = 0;
-  for (size_t i = 0; i < case_list.size(); i++) {
-    sum += (case_list[i]->*method)();
-  }
-  return sum;
-}
-
-// Returns true iff the test case passed.
-static bool TestCasePassed(const TestCase* test_case) {
-  return test_case->should_run() && test_case->Passed();
-}
-
-// Returns true iff the test case failed.
-static bool TestCaseFailed(const TestCase* test_case) {
-  return test_case->should_run() && test_case->Failed();
-}
-
-// Returns true iff test_case contains at least one test that should
-// run.
-static bool ShouldRunTestCase(const TestCase* test_case) {
-  return test_case->should_run();
-}
-
-// AssertHelper constructor.
-AssertHelper::AssertHelper(TestPartResult::Type type,
-                           const char* file,
-                           int line,
-                           const char* message)
-    : data_(new AssertHelperData(type, file, line, message)) {
-}
-
-AssertHelper::~AssertHelper() {
-  delete data_;
-}
-
-// Message assignment, for assertion streaming support.
-void AssertHelper::operator=(const Message& message) const {
-  UnitTest::GetInstance()->
-    AddTestPartResult(data_->type, data_->file, data_->line,
-                      AppendUserMessage(data_->message, message),
-                      UnitTest::GetInstance()->impl()
-                      ->CurrentOsStackTraceExceptTop(1)
-                      // Skips the stack frame for this function itself.
-                      );  // NOLINT
-}
-
-// Mutex for linked pointers.
-GTEST_API_ GTEST_DEFINE_STATIC_MUTEX_(g_linked_ptr_mutex);
-
-// A copy of all command line arguments.  Set by InitGoogleTest().
-::std::vector<testing::internal::string> g_argvs;
-
-const ::std::vector<testing::internal::string>& GetArgvs() {
-#if defined(GTEST_CUSTOM_GET_ARGVS_)
-  return GTEST_CUSTOM_GET_ARGVS_();
-#else  // defined(GTEST_CUSTOM_GET_ARGVS_)
-  return g_argvs;
-#endif  // defined(GTEST_CUSTOM_GET_ARGVS_)
-}
-
-// Returns the current application's name, removing directory path if that
-// is present.
-FilePath GetCurrentExecutableName() {
-  FilePath result;
-
-#if GTEST_OS_WINDOWS
-  result.Set(FilePath(GetArgvs()[0]).RemoveExtension("exe"));
-#else
-  result.Set(FilePath(GetArgvs()[0]));
-#endif  // GTEST_OS_WINDOWS
-
-  return result.RemoveDirectoryName();
-}
-
-// Functions for processing the gtest_output flag.
-
-// Returns the output format, or "" for normal printed output.
-std::string UnitTestOptions::GetOutputFormat() {
-  const char* const gtest_output_flag = GTEST_FLAG(output).c_str();
-  if (gtest_output_flag == NULL) return std::string("");
-
-  const char* const colon = strchr(gtest_output_flag, ':');
-  return (colon == NULL) ?
-      std::string(gtest_output_flag) :
-      std::string(gtest_output_flag, colon - gtest_output_flag);
-}
-
-// Returns the name of the requested output file, or the default if none
-// was explicitly specified.
-std::string UnitTestOptions::GetAbsolutePathToOutputFile() {
-  const char* const gtest_output_flag = GTEST_FLAG(output).c_str();
-  if (gtest_output_flag == NULL)
-    return "";
-
-  const char* const colon = strchr(gtest_output_flag, ':');
-  if (colon == NULL)
-    return internal::FilePath::ConcatPaths(
-        internal::FilePath(
-            UnitTest::GetInstance()->original_working_dir()),
-        internal::FilePath(kDefaultOutputFile)).string();
-
-  internal::FilePath output_name(colon + 1);
-  if (!output_name.IsAbsolutePath())
-    // TODO(wan@google.com): on Windows \some\path is not an absolute
-    // path (as its meaning depends on the current drive), yet the
-    // following logic for turning it into an absolute path is wrong.
-    // Fix it.
-    output_name = internal::FilePath::ConcatPaths(
-        internal::FilePath(UnitTest::GetInstance()->original_working_dir()),
-        internal::FilePath(colon + 1));
-
-  if (!output_name.IsDirectory())
-    return output_name.string();
-
-  internal::FilePath result(internal::FilePath::GenerateUniqueFileName(
-      output_name, internal::GetCurrentExecutableName(),
-      GetOutputFormat().c_str()));
-  return result.string();
-}
-
-// Returns true iff the wildcard pattern matches the string.  The
-// first ':' or '\0' character in pattern marks the end of it.
-//
-// This recursive algorithm isn't very efficient, but is clear and
-// works well enough for matching test names, which are short.
-bool UnitTestOptions::PatternMatchesString(const char *pattern,
-                                           const char *str) {
-  switch (*pattern) {
-    case '\0':
-    case ':':  // Either ':' or '\0' marks the end of the pattern.
-      return *str == '\0';
-    case '?':  // Matches any single character.
-      return *str != '\0' && PatternMatchesString(pattern + 1, str + 1);
-    case '*':  // Matches any string (possibly empty) of characters.
-      return (*str != '\0' && PatternMatchesString(pattern, str + 1)) ||
-          PatternMatchesString(pattern + 1, str);
-    default:  // Non-special character.  Matches itself.
-      return *pattern == *str &&
-          PatternMatchesString(pattern + 1, str + 1);
-  }
-}
-
-bool UnitTestOptions::MatchesFilter(
-    const std::string& name, const char* filter) {
-  const char *cur_pattern = filter;
-  for (;;) {
-    if (PatternMatchesString(cur_pattern, name.c_str())) {
-      return true;
-    }
-
-    // Finds the next pattern in the filter.
-    cur_pattern = strchr(cur_pattern, ':');
-
-    // Returns if no more pattern can be found.
-    if (cur_pattern == NULL) {
-      return false;
-    }
-
-    // Skips the pattern separater (the ':' character).
-    cur_pattern++;
-  }
-}
-
-// Returns true iff the user-specified filter matches the test case
-// name and the test name.
-bool UnitTestOptions::FilterMatchesTest(const std::string &test_case_name,
-                                        const std::string &test_name) {
-  const std::string& full_name = test_case_name + "." + test_name.c_str();
-
-  // Split --gtest_filter at '-', if there is one, to separate into
-  // positive filter and negative filter portions
-  const char* const p = GTEST_FLAG(filter).c_str();
-  const char* const dash = strchr(p, '-');
-  std::string positive;
-  std::string negative;
-  if (dash == NULL) {
-    positive = GTEST_FLAG(filter).c_str();  // Whole string is a positive filter
-    negative = "";
-  } else {
-    positive = std::string(p, dash);   // Everything up to the dash
-    negative = std::string(dash + 1);  // Everything after the dash
-    if (positive.empty()) {
-      // Treat '-test1' as the same as '*-test1'
-      positive = kUniversalFilter;
-    }
-  }
-
-  // A filter is a colon-separated list of patterns.  It matches a
-  // test if any pattern in it matches the test.
-  return (MatchesFilter(full_name, positive.c_str()) &&
-          !MatchesFilter(full_name, negative.c_str()));
-}
-
-#if GTEST_HAS_SEH
-// Returns EXCEPTION_EXECUTE_HANDLER if Google Test should handle the
-// given SEH exception, or EXCEPTION_CONTINUE_SEARCH otherwise.
-// This function is useful as an __except condition.
-int UnitTestOptions::GTestShouldProcessSEH(DWORD exception_code) {
-  // Google Test should handle a SEH exception if:
-  //   1. the user wants it to, AND
-  //   2. this is not a breakpoint exception, AND
-  //   3. this is not a C++ exception (VC++ implements them via SEH,
-  //      apparently).
-  //
-  // SEH exception code for C++ exceptions.
-  // (see http://support.microsoft.com/kb/185294 for more information).
-  const DWORD kCxxExceptionCode = 0xe06d7363;
-
-  bool should_handle = true;
-
-  if (!GTEST_FLAG(catch_exceptions))
-    should_handle = false;
-  else if (exception_code == EXCEPTION_BREAKPOINT)
-    should_handle = false;
-  else if (exception_code == kCxxExceptionCode)
-    should_handle = false;
-
-  return should_handle ? EXCEPTION_EXECUTE_HANDLER : EXCEPTION_CONTINUE_SEARCH;
-}
-#endif  // GTEST_HAS_SEH
-
-}  // namespace internal
-
-// The c'tor sets this object as the test part result reporter used by
-// Google Test.  The 'result' parameter specifies where to report the
-// results. Intercepts only failures from the current thread.
-ScopedFakeTestPartResultReporter::ScopedFakeTestPartResultReporter(
-    TestPartResultArray* result)
-    : intercept_mode_(INTERCEPT_ONLY_CURRENT_THREAD),
-      result_(result) {
-  Init();
-}
-
-// The c'tor sets this object as the test part result reporter used by
-// Google Test.  The 'result' parameter specifies where to report the
-// results.
-ScopedFakeTestPartResultReporter::ScopedFakeTestPartResultReporter(
-    InterceptMode intercept_mode, TestPartResultArray* result)
-    : intercept_mode_(intercept_mode),
-      result_(result) {
-  Init();
-}
-
-void ScopedFakeTestPartResultReporter::Init() {
-  internal::UnitTestImpl* const impl = internal::GetUnitTestImpl();
-  if (intercept_mode_ == INTERCEPT_ALL_THREADS) {
-    old_reporter_ = impl->GetGlobalTestPartResultReporter();
-    impl->SetGlobalTestPartResultReporter(this);
-  } else {
-    old_reporter_ = impl->GetTestPartResultReporterForCurrentThread();
-    impl->SetTestPartResultReporterForCurrentThread(this);
-  }
-}
-
-// The d'tor restores the test part result reporter used by Google Test
-// before.
-ScopedFakeTestPartResultReporter::~ScopedFakeTestPartResultReporter() {
-  internal::UnitTestImpl* const impl = internal::GetUnitTestImpl();
-  if (intercept_mode_ == INTERCEPT_ALL_THREADS) {
-    impl->SetGlobalTestPartResultReporter(old_reporter_);
-  } else {
-    impl->SetTestPartResultReporterForCurrentThread(old_reporter_);
-  }
-}
-
-// Increments the test part result count and remembers the result.
-// This method is from the TestPartResultReporterInterface interface.
-void ScopedFakeTestPartResultReporter::ReportTestPartResult(
-    const TestPartResult& result) {
-  result_->Append(result);
-}
-
-namespace internal {
-
-// Returns the type ID of ::testing::Test.  We should always call this
-// instead of GetTypeId< ::testing::Test>() to get the type ID of
-// testing::Test.  This is to work around a suspected linker bug when
-// using Google Test as a framework on Mac OS X.  The bug causes
-// GetTypeId< ::testing::Test>() to return different values depending
-// on whether the call is from the Google Test framework itself or
-// from user test code.  GetTestTypeId() is guaranteed to always
-// return the same value, as it always calls GetTypeId<>() from the
-// gtest.cc, which is within the Google Test framework.
-TypeId GetTestTypeId() {
-  return GetTypeId<Test>();
-}
-
-// The value of GetTestTypeId() as seen from within the Google Test
-// library.  This is solely for testing GetTestTypeId().
-extern const TypeId kTestTypeIdInGoogleTest = GetTestTypeId();
-
-// This predicate-formatter checks that 'results' contains a test part
-// failure of the given type and that the failure message contains the
-// given substring.
-AssertionResult HasOneFailure(const char* /* results_expr */,
-                              const char* /* type_expr */,
-                              const char* /* substr_expr */,
-                              const TestPartResultArray& results,
-                              TestPartResult::Type type,
-                              const string& substr) {
-  const std::string expected(type == TestPartResult::kFatalFailure ?
-                        "1 fatal failure" :
-                        "1 non-fatal failure");
-  Message msg;
-  if (results.size() != 1) {
-    msg << "Expected: " << expected << "\n"
-        << "  Actual: " << results.size() << " failures";
-    for (int i = 0; i < results.size(); i++) {
-      msg << "\n" << results.GetTestPartResult(i);
-    }
-    return AssertionFailure() << msg;
-  }
-
-  const TestPartResult& r = results.GetTestPartResult(0);
-  if (r.type() != type) {
-    return AssertionFailure() << "Expected: " << expected << "\n"
-                              << "  Actual:\n"
-                              << r;
-  }
-
-  if (strstr(r.message(), substr.c_str()) == NULL) {
-    return AssertionFailure() << "Expected: " << expected << " containing \""
-                              << substr << "\"\n"
-                              << "  Actual:\n"
-                              << r;
-  }
-
-  return AssertionSuccess();
-}
-
-// The constructor of SingleFailureChecker remembers where to look up
-// test part results, what type of failure we expect, and what
-// substring the failure message should contain.
-SingleFailureChecker:: SingleFailureChecker(
-    const TestPartResultArray* results,
-    TestPartResult::Type type,
-    const string& substr)
-    : results_(results),
-      type_(type),
-      substr_(substr) {}
-
-// The destructor of SingleFailureChecker verifies that the given
-// TestPartResultArray contains exactly one failure that has the given
-// type and contains the given substring.  If that's not the case, a
-// non-fatal failure will be generated.
-SingleFailureChecker::~SingleFailureChecker() {
-  EXPECT_PRED_FORMAT3(HasOneFailure, *results_, type_, substr_);
-}
-
-DefaultGlobalTestPartResultReporter::DefaultGlobalTestPartResultReporter(
-    UnitTestImpl* unit_test) : unit_test_(unit_test) {}
-
-void DefaultGlobalTestPartResultReporter::ReportTestPartResult(
-    const TestPartResult& result) {
-  unit_test_->current_test_result()->AddTestPartResult(result);
-  unit_test_->listeners()->repeater()->OnTestPartResult(result);
-}
-
-DefaultPerThreadTestPartResultReporter::DefaultPerThreadTestPartResultReporter(
-    UnitTestImpl* unit_test) : unit_test_(unit_test) {}
-
-void DefaultPerThreadTestPartResultReporter::ReportTestPartResult(
-    const TestPartResult& result) {
-  unit_test_->GetGlobalTestPartResultReporter()->ReportTestPartResult(result);
-}
-
-// Returns the global test part result reporter.
-TestPartResultReporterInterface*
-UnitTestImpl::GetGlobalTestPartResultReporter() {
-  internal::MutexLock lock(&global_test_part_result_reporter_mutex_);
-  return global_test_part_result_repoter_;
-}
-
-// Sets the global test part result reporter.
-void UnitTestImpl::SetGlobalTestPartResultReporter(
-    TestPartResultReporterInterface* reporter) {
-  internal::MutexLock lock(&global_test_part_result_reporter_mutex_);
-  global_test_part_result_repoter_ = reporter;
-}
-
-// Returns the test part result reporter for the current thread.
-TestPartResultReporterInterface*
-UnitTestImpl::GetTestPartResultReporterForCurrentThread() {
-  return per_thread_test_part_result_reporter_.get();
-}
-
-// Sets the test part result reporter for the current thread.
-void UnitTestImpl::SetTestPartResultReporterForCurrentThread(
-    TestPartResultReporterInterface* reporter) {
-  per_thread_test_part_result_reporter_.set(reporter);
-}
-
-// Gets the number of successful test cases.
-int UnitTestImpl::successful_test_case_count() const {
-  return CountIf(test_cases_, TestCasePassed);
-}
-
-// Gets the number of failed test cases.
-int UnitTestImpl::failed_test_case_count() const {
-  return CountIf(test_cases_, TestCaseFailed);
-}
-
-// Gets the number of all test cases.
-int UnitTestImpl::total_test_case_count() const {
-  return static_cast<int>(test_cases_.size());
-}
-
-// Gets the number of all test cases that contain at least one test
-// that should run.
-int UnitTestImpl::test_case_to_run_count() const {
-  return CountIf(test_cases_, ShouldRunTestCase);
-}
-
-// Gets the number of successful tests.
-int UnitTestImpl::successful_test_count() const {
-  return SumOverTestCaseList(test_cases_, &TestCase::successful_test_count);
-}
-
-// Gets the number of failed tests.
-int UnitTestImpl::failed_test_count() const {
-  return SumOverTestCaseList(test_cases_, &TestCase::failed_test_count);
-}
-
-// Gets the number of disabled tests that will be reported in the XML report.
-int UnitTestImpl::reportable_disabled_test_count() const {
-  return SumOverTestCaseList(test_cases_,
-                             &TestCase::reportable_disabled_test_count);
-}
-
-// Gets the number of disabled tests.
-int UnitTestImpl::disabled_test_count() const {
-  return SumOverTestCaseList(test_cases_, &TestCase::disabled_test_count);
-}
-
-// Gets the number of tests to be printed in the XML report.
-int UnitTestImpl::reportable_test_count() const {
-  return SumOverTestCaseList(test_cases_, &TestCase::reportable_test_count);
-}
-
-// Gets the number of all tests.
-int UnitTestImpl::total_test_count() const {
-  return SumOverTestCaseList(test_cases_, &TestCase::total_test_count);
-}
-
-// Gets the number of tests that should run.
-int UnitTestImpl::test_to_run_count() const {
-  return SumOverTestCaseList(test_cases_, &TestCase::test_to_run_count);
-}
-
-// Returns the current OS stack trace as an std::string.
-//
-// The maximum number of stack frames to be included is specified by
-// the gtest_stack_trace_depth flag.  The skip_count parameter
-// specifies the number of top frames to be skipped, which doesn't
-// count against the number of frames to be included.
-//
-// For example, if Foo() calls Bar(), which in turn calls
-// CurrentOsStackTraceExceptTop(1), Foo() will be included in the
-// trace but Bar() and CurrentOsStackTraceExceptTop() won't.
-std::string UnitTestImpl::CurrentOsStackTraceExceptTop(int skip_count) {
-  return os_stack_trace_getter()->CurrentStackTrace(
-      static_cast<int>(GTEST_FLAG(stack_trace_depth)),
-      skip_count + 1
-      // Skips the user-specified number of frames plus this function
-      // itself.
-      );  // NOLINT
-}
-
-// Returns the current time in milliseconds.
-TimeInMillis GetTimeInMillis() {
-#if GTEST_OS_WINDOWS_MOBILE || defined(__BORLANDC__)
-  // Difference between 1970-01-01 and 1601-01-01 in milliseconds.
-  // http://analogous.blogspot.com/2005/04/epoch.html
-  const TimeInMillis kJavaEpochToWinFileTimeDelta =
-    static_cast<TimeInMillis>(116444736UL) * 100000UL;
-  const DWORD kTenthMicrosInMilliSecond = 10000;
-
-  SYSTEMTIME now_systime;
-  FILETIME now_filetime;
-  ULARGE_INTEGER now_int64;
-  // TODO(kenton@google.com): Shouldn't this just use
-  //   GetSystemTimeAsFileTime()?
-  GetSystemTime(&now_systime);
-  if (SystemTimeToFileTime(&now_systime, &now_filetime)) {
-    now_int64.LowPart = now_filetime.dwLowDateTime;
-    now_int64.HighPart = now_filetime.dwHighDateTime;
-    now_int64.QuadPart = (now_int64.QuadPart / kTenthMicrosInMilliSecond) -
-      kJavaEpochToWinFileTimeDelta;
-    return now_int64.QuadPart;
-  }
-  return 0;
-#elif GTEST_OS_WINDOWS && !GTEST_HAS_GETTIMEOFDAY_
-  __timeb64 now;
-
-  // MSVC 8 deprecates _ftime64(), so we want to suppress warning 4996
-  // (deprecated function) there.
-  // TODO(kenton@google.com): Use GetTickCount()?  Or use
-  //   SystemTimeToFileTime()
-  GTEST_DISABLE_MSC_WARNINGS_PUSH_(4996)
-  _ftime64(&now);
-  GTEST_DISABLE_MSC_WARNINGS_POP_()
-
-  return static_cast<TimeInMillis>(now.time) * 1000 + now.millitm;
-#elif GTEST_HAS_GETTIMEOFDAY_
-  struct timeval now;
-  gettimeofday(&now, NULL);
-  return static_cast<TimeInMillis>(now.tv_sec) * 1000 + now.tv_usec / 1000;
-#else
-# error "Don't know how to get the current time on your system."
-#endif
-}
-
-// Utilities
-
-// class String.
-
-#if GTEST_OS_WINDOWS_MOBILE
-// Creates a UTF-16 wide string from the given ANSI string, allocating
-// memory using new. The caller is responsible for deleting the return
-// value using delete[]. Returns the wide string, or NULL if the
-// input is NULL.
-LPCWSTR String::AnsiToUtf16(const char* ansi) {
-  if (!ansi) return NULL;
-  const int length = strlen(ansi);
-  const int unicode_length =
-      MultiByteToWideChar(CP_ACP, 0, ansi, length,
-                          NULL, 0);
-  WCHAR* unicode = new WCHAR[unicode_length + 1];
-  MultiByteToWideChar(CP_ACP, 0, ansi, length,
-                      unicode, unicode_length);
-  unicode[unicode_length] = 0;
-  return unicode;
-}
-
-// Creates an ANSI string from the given wide string, allocating
-// memory using new. The caller is responsible for deleting the return
-// value using delete[]. Returns the ANSI string, or NULL if the
-// input is NULL.
-const char* String::Utf16ToAnsi(LPCWSTR utf16_str)  {
-  if (!utf16_str) return NULL;
-  const int ansi_length =
-      WideCharToMultiByte(CP_ACP, 0, utf16_str, -1,
-                          NULL, 0, NULL, NULL);
-  char* ansi = new char[ansi_length + 1];
-  WideCharToMultiByte(CP_ACP, 0, utf16_str, -1,
-                      ansi, ansi_length, NULL, NULL);
-  ansi[ansi_length] = 0;
-  return ansi;
-}
-
-#endif  // GTEST_OS_WINDOWS_MOBILE
-
-// Compares two C strings.  Returns true iff they have the same content.
-//
-// Unlike strcmp(), this function can handle NULL argument(s).  A NULL
-// C string is considered different to any non-NULL C string,
-// including the empty string.
-bool String::CStringEquals(const char * lhs, const char * rhs) {
-  if ( lhs == NULL ) return rhs == NULL;
-
-  if ( rhs == NULL ) return false;
-
-  return strcmp(lhs, rhs) == 0;
-}
-
-#if GTEST_HAS_STD_WSTRING || GTEST_HAS_GLOBAL_WSTRING
-
-// Converts an array of wide chars to a narrow string using the UTF-8
-// encoding, and streams the result to the given Message object.
-static void StreamWideCharsToMessage(const wchar_t* wstr, size_t length,
-                                     Message* msg) {
-  for (size_t i = 0; i != length; ) {  // NOLINT
-    if (wstr[i] != L'\0') {
-      *msg << WideStringToUtf8(wstr + i, static_cast<int>(length - i));
-      while (i != length && wstr[i] != L'\0')
-        i++;
-    } else {
-      *msg << '\0';
-      i++;
-    }
-  }
-}
-
-#endif  // GTEST_HAS_STD_WSTRING || GTEST_HAS_GLOBAL_WSTRING
-
-void SplitString(const ::std::string& str, char delimiter,
-                 ::std::vector< ::std::string>* dest) {
-  ::std::vector< ::std::string> parsed;
-  ::std::string::size_type pos = 0;
-  while (::testing::internal::AlwaysTrue()) {
-    const ::std::string::size_type colon = str.find(delimiter, pos);
-    if (colon == ::std::string::npos) {
-      parsed.push_back(str.substr(pos));
-      break;
-    } else {
-      parsed.push_back(str.substr(pos, colon - pos));
-      pos = colon + 1;
-    }
-  }
-  dest->swap(parsed);
-}
-
-}  // namespace internal
-
-// Constructs an empty Message.
-// We allocate the stringstream separately because otherwise each use of
-// ASSERT/EXPECT in a procedure adds over 200 bytes to the procedure's
-// stack frame leading to huge stack frames in some cases; gcc does not reuse
-// the stack space.
-Message::Message() : ss_(new ::std::stringstream) {
-  // By default, we want there to be enough precision when printing
-  // a double to a Message.
-  *ss_ << std::setprecision(std::numeric_limits<double>::digits10 + 2);
-}
-
-// These two overloads allow streaming a wide C string to a Message
-// using the UTF-8 encoding.
-Message& Message::operator <<(const wchar_t* wide_c_str) {
-  return *this << internal::String::ShowWideCString(wide_c_str);
-}
-Message& Message::operator <<(wchar_t* wide_c_str) {
-  return *this << internal::String::ShowWideCString(wide_c_str);
-}
-
-#if GTEST_HAS_STD_WSTRING
-// Converts the given wide string to a narrow string using the UTF-8
-// encoding, and streams the result to this Message object.
-Message& Message::operator <<(const ::std::wstring& wstr) {
-  internal::StreamWideCharsToMessage(wstr.c_str(), wstr.length(), this);
-  return *this;
-}
-#endif  // GTEST_HAS_STD_WSTRING
-
-#if GTEST_HAS_GLOBAL_WSTRING
-// Converts the given wide string to a narrow string using the UTF-8
-// encoding, and streams the result to this Message object.
-Message& Message::operator <<(const ::wstring& wstr) {
-  internal::StreamWideCharsToMessage(wstr.c_str(), wstr.length(), this);
-  return *this;
-}
-#endif  // GTEST_HAS_GLOBAL_WSTRING
-
-// Gets the text streamed to this object so far as an std::string.
-// Each '\0' character in the buffer is replaced with "\\0".
-std::string Message::GetString() const {
-  return internal::StringStreamToString(ss_.get());
-}
-
-// AssertionResult constructors.
-// Used in EXPECT_TRUE/FALSE(assertion_result).
-AssertionResult::AssertionResult(const AssertionResult& other)
-    : success_(other.success_),
-      message_(other.message_.get() != NULL ?
-               new ::std::string(*other.message_) :
-               static_cast< ::std::string*>(NULL)) {
-}
-
-// Swaps two AssertionResults.
-void AssertionResult::swap(AssertionResult& other) {
-  using std::swap;
-  swap(success_, other.success_);
-  swap(message_, other.message_);
-}
-
-// Returns the assertion's negation. Used with EXPECT/ASSERT_FALSE.
-AssertionResult AssertionResult::operator!() const {
-  AssertionResult negation(!success_);
-  if (message_.get() != NULL)
-    negation << *message_;
-  return negation;
-}
-
-// Makes a successful assertion result.
-AssertionResult AssertionSuccess() {
-  return AssertionResult(true);
-}
-
-// Makes a failed assertion result.
-AssertionResult AssertionFailure() {
-  return AssertionResult(false);
-}
-
-// Makes a failed assertion result with the given failure message.
-// Deprecated; use AssertionFailure() << message.
-AssertionResult AssertionFailure(const Message& message) {
-  return AssertionFailure() << message;
-}
-
-namespace internal {
-
-namespace edit_distance {
-std::vector<EditType> CalculateOptimalEdits(const std::vector<size_t>& left,
-                                            const std::vector<size_t>& right) {
-  std::vector<std::vector<double> > costs(
-      left.size() + 1, std::vector<double>(right.size() + 1));
-  std::vector<std::vector<EditType> > best_move(
-      left.size() + 1, std::vector<EditType>(right.size() + 1));
-
-  // Populate for empty right.
-  for (size_t l_i = 0; l_i < costs.size(); ++l_i) {
-    costs[l_i][0] = static_cast<double>(l_i);
-    best_move[l_i][0] = kRemove;
-  }
-  // Populate for empty left.
-  for (size_t r_i = 1; r_i < costs[0].size(); ++r_i) {
-    costs[0][r_i] = static_cast<double>(r_i);
-    best_move[0][r_i] = kAdd;
-  }
-
-  for (size_t l_i = 0; l_i < left.size(); ++l_i) {
-    for (size_t r_i = 0; r_i < right.size(); ++r_i) {
-      if (left[l_i] == right[r_i]) {
-        // Found a match. Consume it.
-        costs[l_i + 1][r_i + 1] = costs[l_i][r_i];
-        best_move[l_i + 1][r_i + 1] = kMatch;
-        continue;
-      }
-
-      const double add = costs[l_i + 1][r_i];
-      const double remove = costs[l_i][r_i + 1];
-      const double replace = costs[l_i][r_i];
-      if (add < remove && add < replace) {
-        costs[l_i + 1][r_i + 1] = add + 1;
-        best_move[l_i + 1][r_i + 1] = kAdd;
-      } else if (remove < add && remove < replace) {
-        costs[l_i + 1][r_i + 1] = remove + 1;
-        best_move[l_i + 1][r_i + 1] = kRemove;
-      } else {
-        // We make replace a little more expensive than add/remove to lower
-        // their priority.
-        costs[l_i + 1][r_i + 1] = replace + 1.00001;
-        best_move[l_i + 1][r_i + 1] = kReplace;
-      }
-    }
-  }
-
-  // Reconstruct the best path. We do it in reverse order.
-  std::vector<EditType> best_path;
-  for (size_t l_i = left.size(), r_i = right.size(); l_i > 0 || r_i > 0;) {
-    EditType move = best_move[l_i][r_i];
-    best_path.push_back(move);
-    l_i -= move != kAdd;
-    r_i -= move != kRemove;
-  }
-  std::reverse(best_path.begin(), best_path.end());
-  return best_path;
-}
-
-namespace {
-
-// Helper class to convert string into ids with deduplication.
-class InternalStrings {
- public:
-  size_t GetId(const std::string& str) {
-    IdMap::iterator it = ids_.find(str);
-    if (it != ids_.end()) return it->second;
-    size_t id = ids_.size();
-    return ids_[str] = id;
-  }
-
- private:
-  typedef std::map<std::string, size_t> IdMap;
-  IdMap ids_;
-};
-
-}  // namespace
-
-std::vector<EditType> CalculateOptimalEdits(
-    const std::vector<std::string>& left,
-    const std::vector<std::string>& right) {
-  std::vector<size_t> left_ids, right_ids;
-  {
-    InternalStrings intern_table;
-    for (size_t i = 0; i < left.size(); ++i) {
-      left_ids.push_back(intern_table.GetId(left[i]));
-    }
-    for (size_t i = 0; i < right.size(); ++i) {
-      right_ids.push_back(intern_table.GetId(right[i]));
-    }
-  }
-  return CalculateOptimalEdits(left_ids, right_ids);
-}
-
-namespace {
-
-// Helper class that holds the state for one hunk and prints it out to the
-// stream.
-// It reorders adds/removes when possible to group all removes before all
-// adds. It also adds the hunk header before printint into the stream.
-class Hunk {
- public:
-  Hunk(size_t left_start, size_t right_start)
-      : left_start_(left_start),
-        right_start_(right_start),
-        adds_(),
-        removes_(),
-        common_() {}
-
-  void PushLine(char edit, const char* line) {
-    switch (edit) {
-      case ' ':
-        ++common_;
-        FlushEdits();
-        hunk_.push_back(std::make_pair(' ', line));
-        break;
-      case '-':
-        ++removes_;
-        hunk_removes_.push_back(std::make_pair('-', line));
-        break;
-      case '+':
-        ++adds_;
-        hunk_adds_.push_back(std::make_pair('+', line));
-        break;
-    }
-  }
-
-  void PrintTo(std::ostream* os) {
-    PrintHeader(os);
-    FlushEdits();
-    for (std::list<std::pair<char, const char*> >::const_iterator it =
-             hunk_.begin();
-         it != hunk_.end(); ++it) {
-      *os << it->first << it->second << "\n";
-    }
-  }
-
-  bool has_edits() const { return adds_ || removes_; }
-
- private:
-  void FlushEdits() {
-    hunk_.splice(hunk_.end(), hunk_removes_);
-    hunk_.splice(hunk_.end(), hunk_adds_);
-  }
-
-  // Print a unified diff header for one hunk.
-  // The format is
-  //   "@@ -<left_start>,<left_length> +<right_start>,<right_length> @@"
-  // where the left/right parts are ommitted if unnecessary.
-  void PrintHeader(std::ostream* ss) const {
-    *ss << "@@ ";
-    if (removes_) {
-      *ss << "-" << left_start_ << "," << (removes_ + common_);
-    }
-    if (removes_ && adds_) {
-      *ss << " ";
-    }
-    if (adds_) {
-      *ss << "+" << right_start_ << "," << (adds_ + common_);
-    }
-    *ss << " @@\n";
-  }
-
-  size_t left_start_, right_start_;
-  size_t adds_, removes_, common_;
-  std::list<std::pair<char, const char*> > hunk_, hunk_adds_, hunk_removes_;
-};
-
-}  // namespace
-
-// Create a list of diff hunks in Unified diff format.
-// Each hunk has a header generated by PrintHeader above plus a body with
-// lines prefixed with ' ' for no change, '-' for deletion and '+' for
-// addition.
-// 'context' represents the desired unchanged prefix/suffix around the diff.
-// If two hunks are close enough that their contexts overlap, then they are
-// joined into one hunk.
-std::string CreateUnifiedDiff(const std::vector<std::string>& left,
-                              const std::vector<std::string>& right,
-                              size_t context) {
-  const std::vector<EditType> edits = CalculateOptimalEdits(left, right);
-
-  size_t l_i = 0, r_i = 0, edit_i = 0;
-  std::stringstream ss;
-  while (edit_i < edits.size()) {
-    // Find first edit.
-    while (edit_i < edits.size() && edits[edit_i] == kMatch) {
-      ++l_i;
-      ++r_i;
-      ++edit_i;
-    }
-
-    // Find the first line to include in the hunk.
-    const size_t prefix_context = std::min(l_i, context);
-    Hunk hunk(l_i - prefix_context + 1, r_i - prefix_context + 1);
-    for (size_t i = prefix_context; i > 0; --i) {
-      hunk.PushLine(' ', left[l_i - i].c_str());
-    }
-
-    // Iterate the edits until we found enough suffix for the hunk or the input
-    // is over.
-    size_t n_suffix = 0;
-    for (; edit_i < edits.size(); ++edit_i) {
-      if (n_suffix >= context) {
-        // Continue only if the next hunk is very close.
-        std::vector<EditType>::const_iterator it = edits.begin() + edit_i;
-        while (it != edits.end() && *it == kMatch) ++it;
-        if (it == edits.end() || (it - edits.begin()) - edit_i >= context) {
-          // There is no next edit or it is too far away.
-          break;
-        }
-      }
-
-      EditType edit = edits[edit_i];
-      // Reset count when a non match is found.
-      n_suffix = edit == kMatch ? n_suffix + 1 : 0;
-
-      if (edit == kMatch || edit == kRemove || edit == kReplace) {
-        hunk.PushLine(edit == kMatch ? ' ' : '-', left[l_i].c_str());
-      }
-      if (edit == kAdd || edit == kReplace) {
-        hunk.PushLine('+', right[r_i].c_str());
-      }
-
-      // Advance indices, depending on edit type.
-      l_i += edit != kAdd;
-      r_i += edit != kRemove;
-    }
-
-    if (!hunk.has_edits()) {
-      // We are done. We don't want this hunk.
-      break;
-    }
-
-    hunk.PrintTo(&ss);
-  }
-  return ss.str();
-}
-
-}  // namespace edit_distance
-
-namespace {
-
-// The string representation of the values received in EqFailure() are already
-// escaped. Split them on escaped '\n' boundaries. Leave all other escaped
-// characters the same.
-std::vector<std::string> SplitEscapedString(const std::string& str) {
-  std::vector<std::string> lines;
-  size_t start = 0, end = str.size();
-  if (end > 2 && str[0] == '"' && str[end - 1] == '"') {
-    ++start;
-    --end;
-  }
-  bool escaped = false;
-  for (size_t i = start; i + 1 < end; ++i) {
-    if (escaped) {
-      escaped = false;
-      if (str[i] == 'n') {
-        lines.push_back(str.substr(start, i - start - 1));
-        start = i + 1;
-      }
-    } else {
-      escaped = str[i] == '\\';
-    }
-  }
-  lines.push_back(str.substr(start, end - start));
-  return lines;
-}
-
-}  // namespace
-
-// Constructs and returns the message for an equality assertion
-// (e.g. ASSERT_EQ, EXPECT_STREQ, etc) failure.
-//
-// The first four parameters are the expressions used in the assertion
-// and their values, as strings.  For example, for ASSERT_EQ(foo, bar)
-// where foo is 5 and bar is 6, we have:
-//
-//   lhs_expression: "foo"
-//   rhs_expression: "bar"
-//   lhs_value:      "5"
-//   rhs_value:      "6"
-//
-// The ignoring_case parameter is true iff the assertion is a
-// *_STRCASEEQ*.  When it's true, the string "Ignoring case" will
-// be inserted into the message.
-AssertionResult EqFailure(const char* lhs_expression,
-                          const char* rhs_expression,
-                          const std::string& lhs_value,
-                          const std::string& rhs_value,
-                          bool ignoring_case) {
-  Message msg;
-  msg << "      Expected: " << lhs_expression;
-  if (lhs_value != lhs_expression) {
-    msg << "\n      Which is: " << lhs_value;
-  }
-  msg << "\nTo be equal to: " << rhs_expression;
-  if (rhs_value != rhs_expression) {
-    msg << "\n      Which is: " << rhs_value;
-  }
-
-  if (ignoring_case) {
-    msg << "\nIgnoring case";
-  }
-
-  if (!lhs_value.empty() && !rhs_value.empty()) {
-    const std::vector<std::string> lhs_lines =
-        SplitEscapedString(lhs_value);
-    const std::vector<std::string> rhs_lines =
-        SplitEscapedString(rhs_value);
-    if (lhs_lines.size() > 1 || rhs_lines.size() > 1) {
-      msg << "\nWith diff:\n"
-          << edit_distance::CreateUnifiedDiff(lhs_lines, rhs_lines);
-    }
-  }
-
-  return AssertionFailure() << msg;
-}
-
-// Constructs a failure message for Boolean assertions such as EXPECT_TRUE.
-std::string GetBoolAssertionFailureMessage(
-    const AssertionResult& assertion_result,
-    const char* expression_text,
-    const char* actual_predicate_value,
-    const char* expected_predicate_value) {
-  const char* actual_message = assertion_result.message();
-  Message msg;
-  msg << "Value of: " << expression_text
-      << "\n  Actual: " << actual_predicate_value;
-  if (actual_message[0] != '\0')
-    msg << " (" << actual_message << ")";
-  msg << "\nExpected: " << expected_predicate_value;
-  return msg.GetString();
-}
-
-// Helper function for implementing ASSERT_NEAR.
-AssertionResult DoubleNearPredFormat(const char* expr1,
-                                     const char* expr2,
-                                     const char* abs_error_expr,
-                                     double val1,
-                                     double val2,
-                                     double abs_error) {
-  const double diff = fabs(val1 - val2);
-  if (diff <= abs_error) return AssertionSuccess();
-
-  // TODO(wan): do not print the value of an expression if it's
-  // already a literal.
-  return AssertionFailure()
-      << "The difference between " << expr1 << " and " << expr2
-      << " is " << diff << ", which exceeds " << abs_error_expr << ", where\n"
-      << expr1 << " evaluates to " << val1 << ",\n"
-      << expr2 << " evaluates to " << val2 << ", and\n"
-      << abs_error_expr << " evaluates to " << abs_error << ".";
-}
-
-
-// Helper template for implementing FloatLE() and DoubleLE().
-template <typename RawType>
-AssertionResult FloatingPointLE(const char* expr1,
-                                const char* expr2,
-                                RawType val1,
-                                RawType val2) {
-  // Returns success if val1 is less than val2,
-  if (val1 < val2) {
-    return AssertionSuccess();
-  }
-
-  // or if val1 is almost equal to val2.
-  const FloatingPoint<RawType> lhs(val1), rhs(val2);
-  if (lhs.AlmostEquals(rhs)) {
-    return AssertionSuccess();
-  }
-
-  // Note that the above two checks will both fail if either val1 or
-  // val2 is NaN, as the IEEE floating-point standard requires that
-  // any predicate involving a NaN must return false.
-
-  ::std::stringstream val1_ss;
-  val1_ss << std::setprecision(std::numeric_limits<RawType>::digits10 + 2)
-          << val1;
-
-  ::std::stringstream val2_ss;
-  val2_ss << std::setprecision(std::numeric_limits<RawType>::digits10 + 2)
-          << val2;
-
-  return AssertionFailure()
-      << "Expected: (" << expr1 << ") <= (" << expr2 << ")\n"
-      << "  Actual: " << StringStreamToString(&val1_ss) << " vs "
-      << StringStreamToString(&val2_ss);
-}
-
-}  // namespace internal
-
-// Asserts that val1 is less than, or almost equal to, val2.  Fails
-// otherwise.  In particular, it fails if either val1 or val2 is NaN.
-AssertionResult FloatLE(const char* expr1, const char* expr2,
-                        float val1, float val2) {
-  return internal::FloatingPointLE<float>(expr1, expr2, val1, val2);
-}
-
-// Asserts that val1 is less than, or almost equal to, val2.  Fails
-// otherwise.  In particular, it fails if either val1 or val2 is NaN.
-AssertionResult DoubleLE(const char* expr1, const char* expr2,
-                         double val1, double val2) {
-  return internal::FloatingPointLE<double>(expr1, expr2, val1, val2);
-}
-
-namespace internal {
-
-// The helper function for {ASSERT|EXPECT}_EQ with int or enum
-// arguments.
-AssertionResult CmpHelperEQ(const char* lhs_expression,
-                            const char* rhs_expression,
-                            BiggestInt lhs,
-                            BiggestInt rhs) {
-  if (lhs == rhs) {
-    return AssertionSuccess();
-  }
-
-  return EqFailure(lhs_expression,
-                   rhs_expression,
-                   FormatForComparisonFailureMessage(lhs, rhs),
-                   FormatForComparisonFailureMessage(rhs, lhs),
-                   false);
-}
-
-// A macro for implementing the helper functions needed to implement
-// ASSERT_?? and EXPECT_?? with integer or enum arguments.  It is here
-// just to avoid copy-and-paste of similar code.
-#define GTEST_IMPL_CMP_HELPER_(op_name, op)\
-AssertionResult CmpHelper##op_name(const char* expr1, const char* expr2, \
-                                   BiggestInt val1, BiggestInt val2) {\
-  if (val1 op val2) {\
-    return AssertionSuccess();\
-  } else {\
-    return AssertionFailure() \
-        << "Expected: (" << expr1 << ") " #op " (" << expr2\
-        << "), actual: " << FormatForComparisonFailureMessage(val1, val2)\
-        << " vs " << FormatForComparisonFailureMessage(val2, val1);\
-  }\
-}
-
-// Implements the helper function for {ASSERT|EXPECT}_NE with int or
-// enum arguments.
-GTEST_IMPL_CMP_HELPER_(NE, !=)
-// Implements the helper function for {ASSERT|EXPECT}_LE with int or
-// enum arguments.
-GTEST_IMPL_CMP_HELPER_(LE, <=)
-// Implements the helper function for {ASSERT|EXPECT}_LT with int or
-// enum arguments.
-GTEST_IMPL_CMP_HELPER_(LT, < )
-// Implements the helper function for {ASSERT|EXPECT}_GE with int or
-// enum arguments.
-GTEST_IMPL_CMP_HELPER_(GE, >=)
-// Implements the helper function for {ASSERT|EXPECT}_GT with int or
-// enum arguments.
-GTEST_IMPL_CMP_HELPER_(GT, > )
-
-#undef GTEST_IMPL_CMP_HELPER_
-
-// The helper function for {ASSERT|EXPECT}_STREQ.
-AssertionResult CmpHelperSTREQ(const char* lhs_expression,
-                               const char* rhs_expression,
-                               const char* lhs,
-                               const char* rhs) {
-  if (String::CStringEquals(lhs, rhs)) {
-    return AssertionSuccess();
-  }
-
-  return EqFailure(lhs_expression,
-                   rhs_expression,
-                   PrintToString(lhs),
-                   PrintToString(rhs),
-                   false);
-}
-
-// The helper function for {ASSERT|EXPECT}_STRCASEEQ.
-AssertionResult CmpHelperSTRCASEEQ(const char* lhs_expression,
-                                   const char* rhs_expression,
-                                   const char* lhs,
-                                   const char* rhs) {
-  if (String::CaseInsensitiveCStringEquals(lhs, rhs)) {
-    return AssertionSuccess();
-  }
-
-  return EqFailure(lhs_expression,
-                   rhs_expression,
-                   PrintToString(lhs),
-                   PrintToString(rhs),
-                   true);
-}
-
-// The helper function for {ASSERT|EXPECT}_STRNE.
-AssertionResult CmpHelperSTRNE(const char* s1_expression,
-                               const char* s2_expression,
-                               const char* s1,
-                               const char* s2) {
-  if (!String::CStringEquals(s1, s2)) {
-    return AssertionSuccess();
-  } else {
-    return AssertionFailure() << "Expected: (" << s1_expression << ") != ("
-                              << s2_expression << "), actual: \""
-                              << s1 << "\" vs \"" << s2 << "\"";
-  }
-}
-
-// The helper function for {ASSERT|EXPECT}_STRCASENE.
-AssertionResult CmpHelperSTRCASENE(const char* s1_expression,
-                                   const char* s2_expression,
-                                   const char* s1,
-                                   const char* s2) {
-  if (!String::CaseInsensitiveCStringEquals(s1, s2)) {
-    return AssertionSuccess();
-  } else {
-    return AssertionFailure()
-        << "Expected: (" << s1_expression << ") != ("
-        << s2_expression << ") (ignoring case), actual: \""
-        << s1 << "\" vs \"" << s2 << "\"";
-  }
-}
-
-}  // namespace internal
-
-namespace {
-
-// Helper functions for implementing IsSubString() and IsNotSubstring().
-
-// This group of overloaded functions return true iff needle is a
-// substring of haystack.  NULL is considered a substring of itself
-// only.
-
-bool IsSubstringPred(const char* needle, const char* haystack) {
-  if (needle == NULL || haystack == NULL)
-    return needle == haystack;
-
-  return strstr(haystack, needle) != NULL;
-}
-
-bool IsSubstringPred(const wchar_t* needle, const wchar_t* haystack) {
-  if (needle == NULL || haystack == NULL)
-    return needle == haystack;
-
-  return wcsstr(haystack, needle) != NULL;
-}
-
-// StringType here can be either ::std::string or ::std::wstring.
-template <typename StringType>
-bool IsSubstringPred(const StringType& needle,
-                     const StringType& haystack) {
-  return haystack.find(needle) != StringType::npos;
-}
-
-// This function implements either IsSubstring() or IsNotSubstring(),
-// depending on the value of the expected_to_be_substring parameter.
-// StringType here can be const char*, const wchar_t*, ::std::string,
-// or ::std::wstring.
-template <typename StringType>
-AssertionResult IsSubstringImpl(
-    bool expected_to_be_substring,
-    const char* needle_expr, const char* haystack_expr,
-    const StringType& needle, const StringType& haystack) {
-  if (IsSubstringPred(needle, haystack) == expected_to_be_substring)
-    return AssertionSuccess();
-
-  const bool is_wide_string = sizeof(needle[0]) > 1;
-  const char* const begin_string_quote = is_wide_string ? "L\"" : "\"";
-  return AssertionFailure()
-      << "Value of: " << needle_expr << "\n"
-      << "  Actual: " << begin_string_quote << needle << "\"\n"
-      << "Expected: " << (expected_to_be_substring ? "" : "not ")
-      << "a substring of " << haystack_expr << "\n"
-      << "Which is: " << begin_string_quote << haystack << "\"";
-}
-
-}  // namespace
-
-// IsSubstring() and IsNotSubstring() check whether needle is a
-// substring of haystack (NULL is considered a substring of itself
-// only), and return an appropriate error message when they fail.
-
-AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const char* needle, const char* haystack) {
-  return IsSubstringImpl(true, needle_expr, haystack_expr, needle, haystack);
-}
-
-AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const wchar_t* needle, const wchar_t* haystack) {
-  return IsSubstringImpl(true, needle_expr, haystack_expr, needle, haystack);
-}
-
-AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const char* needle, const char* haystack) {
-  return IsSubstringImpl(false, needle_expr, haystack_expr, needle, haystack);
-}
-
-AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const wchar_t* needle, const wchar_t* haystack) {
-  return IsSubstringImpl(false, needle_expr, haystack_expr, needle, haystack);
-}
-
-AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::string& needle, const ::std::string& haystack) {
-  return IsSubstringImpl(true, needle_expr, haystack_expr, needle, haystack);
-}
-
-AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::string& needle, const ::std::string& haystack) {
-  return IsSubstringImpl(false, needle_expr, haystack_expr, needle, haystack);
-}
-
-#if GTEST_HAS_STD_WSTRING
-AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::wstring& needle, const ::std::wstring& haystack) {
-  return IsSubstringImpl(true, needle_expr, haystack_expr, needle, haystack);
-}
-
-AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::wstring& needle, const ::std::wstring& haystack) {
-  return IsSubstringImpl(false, needle_expr, haystack_expr, needle, haystack);
-}
-#endif  // GTEST_HAS_STD_WSTRING
-
-namespace internal {
-
-#if GTEST_OS_WINDOWS
-
-namespace {
-
-// Helper function for IsHRESULT{SuccessFailure} predicates
-AssertionResult HRESULTFailureHelper(const char* expr,
-                                     const char* expected,
-                                     long hr) {  // NOLINT
-# if GTEST_OS_WINDOWS_MOBILE
-
-  // Windows CE doesn't support FormatMessage.
-  const char error_text[] = "";
-
-# else
-
-  // Looks up the human-readable system message for the HRESULT code
-  // and since we're not passing any params to FormatMessage, we don't
-  // want inserts expanded.
-  const DWORD kFlags = FORMAT_MESSAGE_FROM_SYSTEM |
-                       FORMAT_MESSAGE_IGNORE_INSERTS;
-  const DWORD kBufSize = 4096;
-  // Gets the system's human readable message string for this HRESULT.
-  char error_text[kBufSize] = { '\0' };
-  DWORD message_length = ::FormatMessageA(kFlags,
-                                          0,  // no source, we're asking system
-                                          hr,  // the error
-                                          0,  // no line width restrictions
-                                          error_text,  // output buffer
-                                          kBufSize,  // buf size
-                                          NULL);  // no arguments for inserts
-  // Trims tailing white space (FormatMessage leaves a trailing CR-LF)
-  for (; message_length && IsSpace(error_text[message_length - 1]);
-          --message_length) {
-    error_text[message_length - 1] = '\0';
-  }
-
-# endif  // GTEST_OS_WINDOWS_MOBILE
-
-  const std::string error_hex("0x" + String::FormatHexInt(hr));
-  return ::testing::AssertionFailure()
-      << "Expected: " << expr << " " << expected << ".\n"
-      << "  Actual: " << error_hex << " " << error_text << "\n";
-}
-
-}  // namespace
-
-AssertionResult IsHRESULTSuccess(const char* expr, long hr) {  // NOLINT
-  if (SUCCEEDED(hr)) {
-    return AssertionSuccess();
-  }
-  return HRESULTFailureHelper(expr, "succeeds", hr);
-}
-
-AssertionResult IsHRESULTFailure(const char* expr, long hr) {  // NOLINT
-  if (FAILED(hr)) {
-    return AssertionSuccess();
-  }
-  return HRESULTFailureHelper(expr, "fails", hr);
-}
-
-#endif  // GTEST_OS_WINDOWS
-
-// Utility functions for encoding Unicode text (wide strings) in
-// UTF-8.
-
-// A Unicode code-point can have upto 21 bits, and is encoded in UTF-8
-// like this:
-//
-// Code-point length   Encoding
-//   0 -  7 bits       0xxxxxxx
-//   8 - 11 bits       110xxxxx 10xxxxxx
-//  12 - 16 bits       1110xxxx 10xxxxxx 10xxxxxx
-//  17 - 21 bits       11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
-
-// The maximum code-point a one-byte UTF-8 sequence can represent.
-const UInt32 kMaxCodePoint1 = (static_cast<UInt32>(1) <<  7) - 1;
-
-// The maximum code-point a two-byte UTF-8 sequence can represent.
-const UInt32 kMaxCodePoint2 = (static_cast<UInt32>(1) << (5 + 6)) - 1;
-
-// The maximum code-point a three-byte UTF-8 sequence can represent.
-const UInt32 kMaxCodePoint3 = (static_cast<UInt32>(1) << (4 + 2*6)) - 1;
-
-// The maximum code-point a four-byte UTF-8 sequence can represent.
-const UInt32 kMaxCodePoint4 = (static_cast<UInt32>(1) << (3 + 3*6)) - 1;
-
-// Chops off the n lowest bits from a bit pattern.  Returns the n
-// lowest bits.  As a side effect, the original bit pattern will be
-// shifted to the right by n bits.
-inline UInt32 ChopLowBits(UInt32* bits, int n) {
-  const UInt32 low_bits = *bits & ((static_cast<UInt32>(1) << n) - 1);
-  *bits >>= n;
-  return low_bits;
-}
-
-// Converts a Unicode code point to a narrow string in UTF-8 encoding.
-// code_point parameter is of type UInt32 because wchar_t may not be
-// wide enough to contain a code point.
-// If the code_point is not a valid Unicode code point
-// (i.e. outside of Unicode range U+0 to U+10FFFF) it will be converted
-// to "(Invalid Unicode 0xXXXXXXXX)".
-std::string CodePointToUtf8(UInt32 code_point) {
-  if (code_point > kMaxCodePoint4) {
-    return "(Invalid Unicode 0x" + String::FormatHexInt(code_point) + ")";
-  }
-
-  char str[5];  // Big enough for the largest valid code point.
-  if (code_point <= kMaxCodePoint1) {
-    str[1] = '\0';
-    str[0] = static_cast<char>(code_point);                          // 0xxxxxxx
-  } else if (code_point <= kMaxCodePoint2) {
-    str[2] = '\0';
-    str[1] = static_cast<char>(0x80 | ChopLowBits(&code_point, 6));  // 10xxxxxx
-    str[0] = static_cast<char>(0xC0 | code_point);                   // 110xxxxx
-  } else if (code_point <= kMaxCodePoint3) {
-    str[3] = '\0';
-    str[2] = static_cast<char>(0x80 | ChopLowBits(&code_point, 6));  // 10xxxxxx
-    str[1] = static_cast<char>(0x80 | ChopLowBits(&code_point, 6));  // 10xxxxxx
-    str[0] = static_cast<char>(0xE0 | code_point);                   // 1110xxxx
-  } else {  // code_point <= kMaxCodePoint4
-    str[4] = '\0';
-    str[3] = static_cast<char>(0x80 | ChopLowBits(&code_point, 6));  // 10xxxxxx
-    str[2] = static_cast<char>(0x80 | ChopLowBits(&code_point, 6));  // 10xxxxxx
-    str[1] = static_cast<char>(0x80 | ChopLowBits(&code_point, 6));  // 10xxxxxx
-    str[0] = static_cast<char>(0xF0 | code_point);                   // 11110xxx
-  }
-  return str;
-}
-
-// The following two functions only make sense if the the system
-// uses UTF-16 for wide string encoding. All supported systems
-// with 16 bit wchar_t (Windows, Cygwin, Symbian OS) do use UTF-16.
-
-// Determines if the arguments constitute UTF-16 surrogate pair
-// and thus should be combined into a single Unicode code point
-// using CreateCodePointFromUtf16SurrogatePair.
-inline bool IsUtf16SurrogatePair(wchar_t first, wchar_t second) {
-  return sizeof(wchar_t) == 2 &&
-      (first & 0xFC00) == 0xD800 && (second & 0xFC00) == 0xDC00;
-}
-
-// Creates a Unicode code point from UTF16 surrogate pair.
-inline UInt32 CreateCodePointFromUtf16SurrogatePair(wchar_t first,
-                                                    wchar_t second) {
-  const UInt32 mask = (1 << 10) - 1;
-  return (sizeof(wchar_t) == 2) ?
-      (((first & mask) << 10) | (second & mask)) + 0x10000 :
-      // This function should not be called when the condition is
-      // false, but we provide a sensible default in case it is.
-      static_cast<UInt32>(first);
-}
-
-// Converts a wide string to a narrow string in UTF-8 encoding.
-// The wide string is assumed to have the following encoding:
-//   UTF-16 if sizeof(wchar_t) == 2 (on Windows, Cygwin, Symbian OS)
-//   UTF-32 if sizeof(wchar_t) == 4 (on Linux)
-// Parameter str points to a null-terminated wide string.
-// Parameter num_chars may additionally limit the number
-// of wchar_t characters processed. -1 is used when the entire string
-// should be processed.
-// If the string contains code points that are not valid Unicode code points
-// (i.e. outside of Unicode range U+0 to U+10FFFF) they will be output
-// as '(Invalid Unicode 0xXXXXXXXX)'. If the string is in UTF16 encoding
-// and contains invalid UTF-16 surrogate pairs, values in those pairs
-// will be encoded as individual Unicode characters from Basic Normal Plane.
-std::string WideStringToUtf8(const wchar_t* str, int num_chars) {
-  if (num_chars == -1)
-    num_chars = static_cast<int>(wcslen(str));
-
-  ::std::stringstream stream;
-  for (int i = 0; i < num_chars; ++i) {
-    UInt32 unicode_code_point;
-
-    if (str[i] == L'\0') {
-      break;
-    } else if (i + 1 < num_chars && IsUtf16SurrogatePair(str[i], str[i + 1])) {
-      unicode_code_point = CreateCodePointFromUtf16SurrogatePair(str[i],
-                                                                 str[i + 1]);
-      i++;
-    } else {
-      unicode_code_point = static_cast<UInt32>(str[i]);
-    }
-
-    stream << CodePointToUtf8(unicode_code_point);
-  }
-  return StringStreamToString(&stream);
-}
-
-// Converts a wide C string to an std::string using the UTF-8 encoding.
-// NULL will be converted to "(null)".
-std::string String::ShowWideCString(const wchar_t * wide_c_str) {
-  if (wide_c_str == NULL)  return "(null)";
-
-  return internal::WideStringToUtf8(wide_c_str, -1);
-}
-
-// Compares two wide C strings.  Returns true iff they have the same
-// content.
-//
-// Unlike wcscmp(), this function can handle NULL argument(s).  A NULL
-// C string is considered different to any non-NULL C string,
-// including the empty string.
-bool String::WideCStringEquals(const wchar_t * lhs, const wchar_t * rhs) {
-  if (lhs == NULL) return rhs == NULL;
-
-  if (rhs == NULL) return false;
-
-  return wcscmp(lhs, rhs) == 0;
-}
-
-// Helper function for *_STREQ on wide strings.
-AssertionResult CmpHelperSTREQ(const char* lhs_expression,
-                               const char* rhs_expression,
-                               const wchar_t* lhs,
-                               const wchar_t* rhs) {
-  if (String::WideCStringEquals(lhs, rhs)) {
-    return AssertionSuccess();
-  }
-
-  return EqFailure(lhs_expression,
-                   rhs_expression,
-                   PrintToString(lhs),
-                   PrintToString(rhs),
-                   false);
-}
-
-// Helper function for *_STRNE on wide strings.
-AssertionResult CmpHelperSTRNE(const char* s1_expression,
-                               const char* s2_expression,
-                               const wchar_t* s1,
-                               const wchar_t* s2) {
-  if (!String::WideCStringEquals(s1, s2)) {
-    return AssertionSuccess();
-  }
-
-  return AssertionFailure() << "Expected: (" << s1_expression << ") != ("
-                            << s2_expression << "), actual: "
-                            << PrintToString(s1)
-                            << " vs " << PrintToString(s2);
-}
-
-// Compares two C strings, ignoring case.  Returns true iff they have
-// the same content.
-//
-// Unlike strcasecmp(), this function can handle NULL argument(s).  A
-// NULL C string is considered different to any non-NULL C string,
-// including the empty string.
-bool String::CaseInsensitiveCStringEquals(const char * lhs, const char * rhs) {
-  if (lhs == NULL)
-    return rhs == NULL;
-  if (rhs == NULL)
-    return false;
-  return posix::StrCaseCmp(lhs, rhs) == 0;
-}
-
-  // Compares two wide C strings, ignoring case.  Returns true iff they
-  // have the same content.
-  //
-  // Unlike wcscasecmp(), this function can handle NULL argument(s).
-  // A NULL C string is considered different to any non-NULL wide C string,
-  // including the empty string.
-  // NB: The implementations on different platforms slightly differ.
-  // On windows, this method uses _wcsicmp which compares according to LC_CTYPE
-  // environment variable. On GNU platform this method uses wcscasecmp
-  // which compares according to LC_CTYPE category of the current locale.
-  // On MacOS X, it uses towlower, which also uses LC_CTYPE category of the
-  // current locale.
-bool String::CaseInsensitiveWideCStringEquals(const wchar_t* lhs,
-                                              const wchar_t* rhs) {
-  if (lhs == NULL) return rhs == NULL;
-
-  if (rhs == NULL) return false;
-
-#if GTEST_OS_WINDOWS
-  return _wcsicmp(lhs, rhs) == 0;
-#elif GTEST_OS_LINUX && !GTEST_OS_LINUX_ANDROID
-  return wcscasecmp(lhs, rhs) == 0;
-#else
-  // Android, Mac OS X and Cygwin don't define wcscasecmp.
-  // Other unknown OSes may not define it either.
-  wint_t left, right;
-  do {
-    left = towlower(*lhs++);
-    right = towlower(*rhs++);
-  } while (left && left == right);
-  return left == right;
-#endif  // OS selector
-}
-
-// Returns true iff str ends with the given suffix, ignoring case.
-// Any string is considered to end with an empty suffix.
-bool String::EndsWithCaseInsensitive(
-    const std::string& str, const std::string& suffix) {
-  const size_t str_len = str.length();
-  const size_t suffix_len = suffix.length();
-  return (str_len >= suffix_len) &&
-         CaseInsensitiveCStringEquals(str.c_str() + str_len - suffix_len,
-                                      suffix.c_str());
-}
-
-// Formats an int value as "%02d".
-std::string String::FormatIntWidth2(int value) {
-  std::stringstream ss;
-  ss << std::setfill('0') << std::setw(2) << value;
-  return ss.str();
-}
-
-// Formats an int value as "%X".
-std::string String::FormatHexInt(int value) {
-  std::stringstream ss;
-  ss << std::hex << std::uppercase << value;
-  return ss.str();
-}
-
-// Formats a byte as "%02X".
-std::string String::FormatByte(unsigned char value) {
-  std::stringstream ss;
-  ss << std::setfill('0') << std::setw(2) << std::hex << std::uppercase
-     << static_cast<unsigned int>(value);
-  return ss.str();
-}
-
-// Converts the buffer in a stringstream to an std::string, converting NUL
-// bytes to "\\0" along the way.
-std::string StringStreamToString(::std::stringstream* ss) {
-  const ::std::string& str = ss->str();
-  const char* const start = str.c_str();
-  const char* const end = start + str.length();
-
-  std::string result;
-  result.reserve(2 * (end - start));
-  for (const char* ch = start; ch != end; ++ch) {
-    if (*ch == '\0') {
-      result += "\\0";  // Replaces NUL with "\\0";
-    } else {
-      result += *ch;
-    }
-  }
-
-  return result;
-}
-
-// Appends the user-supplied message to the Google-Test-generated message.
-std::string AppendUserMessage(const std::string& gtest_msg,
-                              const Message& user_msg) {
-  // Appends the user message if it's non-empty.
-  const std::string user_msg_string = user_msg.GetString();
-  if (user_msg_string.empty()) {
-    return gtest_msg;
-  }
-
-  return gtest_msg + "\n" + user_msg_string;
-}
-
-}  // namespace internal
-
-// class TestResult
-
-// Creates an empty TestResult.
-TestResult::TestResult()
-    : death_test_count_(0),
-      elapsed_time_(0) {
-}
-
-// D'tor.
-TestResult::~TestResult() {
-}
-
-// Returns the i-th test part result among all the results. i can
-// range from 0 to total_part_count() - 1. If i is not in that range,
-// aborts the program.
-const TestPartResult& TestResult::GetTestPartResult(int i) const {
-  if (i < 0 || i >= total_part_count())
-    internal::posix::Abort();
-  return test_part_results_.at(i);
-}
-
-// Returns the i-th test property. i can range from 0 to
-// test_property_count() - 1. If i is not in that range, aborts the
-// program.
-const TestProperty& TestResult::GetTestProperty(int i) const {
-  if (i < 0 || i >= test_property_count())
-    internal::posix::Abort();
-  return test_properties_.at(i);
-}
-
-// Clears the test part results.
-void TestResult::ClearTestPartResults() {
-  test_part_results_.clear();
-}
-
-// Adds a test part result to the list.
-void TestResult::AddTestPartResult(const TestPartResult& test_part_result) {
-  test_part_results_.push_back(test_part_result);
-}
-
-// Adds a test property to the list. If a property with the same key as the
-// supplied property is already represented, the value of this test_property
-// replaces the old value for that key.
-void TestResult::RecordProperty(const std::string& xml_element,
-                                const TestProperty& test_property) {
-  if (!ValidateTestProperty(xml_element, test_property)) {
-    return;
-  }
-  internal::MutexLock lock(&test_properites_mutex_);
-  const std::vector<TestProperty>::iterator property_with_matching_key =
-      std::find_if(test_properties_.begin(), test_properties_.end(),
-                   internal::TestPropertyKeyIs(test_property.key()));
-  if (property_with_matching_key == test_properties_.end()) {
-    test_properties_.push_back(test_property);
-    return;
-  }
-  property_with_matching_key->SetValue(test_property.value());
-}
-
-// The list of reserved attributes used in the <testsuites> element of XML
-// output.
-static const char* const kReservedTestSuitesAttributes[] = {
-  "disabled",
-  "errors",
-  "failures",
-  "name",
-  "random_seed",
-  "tests",
-  "time",
-  "timestamp"
-};
-
-// The list of reserved attributes used in the <testsuite> element of XML
-// output.
-static const char* const kReservedTestSuiteAttributes[] = {
-  "disabled",
-  "errors",
-  "failures",
-  "name",
-  "tests",
-  "time"
-};
-
-// The list of reserved attributes used in the <testcase> element of XML output.
-static const char* const kReservedTestCaseAttributes[] = {
-  "classname",
-  "name",
-  "status",
-  "time",
-  "type_param",
-  "value_param"
-};
-
-template <int kSize>
-std::vector<std::string> ArrayAsVector(const char* const (&array)[kSize]) {
-  return std::vector<std::string>(array, array + kSize);
-}
-
-static std::vector<std::string> GetReservedAttributesForElement(
-    const std::string& xml_element) {
-  if (xml_element == "testsuites") {
-    return ArrayAsVector(kReservedTestSuitesAttributes);
-  } else if (xml_element == "testsuite") {
-    return ArrayAsVector(kReservedTestSuiteAttributes);
-  } else if (xml_element == "testcase") {
-    return ArrayAsVector(kReservedTestCaseAttributes);
-  } else {
-    GTEST_CHECK_(false) << "Unrecognized xml_element provided: " << xml_element;
-  }
-  // This code is unreachable but some compilers may not realizes that.
-  return std::vector<std::string>();
-}
-
-static std::string FormatWordList(const std::vector<std::string>& words) {
-  Message word_list;
-  for (size_t i = 0; i < words.size(); ++i) {
-    if (i > 0 && words.size() > 2) {
-      word_list << ", ";
-    }
-    if (i == words.size() - 1) {
-      word_list << "and ";
-    }
-    word_list << "'" << words[i] << "'";
-  }
-  return word_list.GetString();
-}
-
-bool ValidateTestPropertyName(const std::string& property_name,
-                              const std::vector<std::string>& reserved_names) {
-  if (std::find(reserved_names.begin(), reserved_names.end(), property_name) !=
-          reserved_names.end()) {
-    ADD_FAILURE() << "Reserved key used in RecordProperty(): " << property_name
-                  << " (" << FormatWordList(reserved_names)
-                  << " are reserved by " << GTEST_NAME_ << ")";
-    return false;
-  }
-  return true;
-}
-
-// Adds a failure if the key is a reserved attribute of the element named
-// xml_element.  Returns true if the property is valid.
-bool TestResult::ValidateTestProperty(const std::string& xml_element,
-                                      const TestProperty& test_property) {
-  return ValidateTestPropertyName(test_property.key(),
-                                  GetReservedAttributesForElement(xml_element));
-}
-
-// Clears the object.
-void TestResult::Clear() {
-  test_part_results_.clear();
-  test_properties_.clear();
-  death_test_count_ = 0;
-  elapsed_time_ = 0;
-}
-
-// Returns true iff the test failed.
-bool TestResult::Failed() const {
-  for (int i = 0; i < total_part_count(); ++i) {
-    if (GetTestPartResult(i).failed())
-      return true;
-  }
-  return false;
-}
-
-// Returns true iff the test part fatally failed.
-static bool TestPartFatallyFailed(const TestPartResult& result) {
-  return result.fatally_failed();
-}
-
-// Returns true iff the test fatally failed.
-bool TestResult::HasFatalFailure() const {
-  return CountIf(test_part_results_, TestPartFatallyFailed) > 0;
-}
-
-// Returns true iff the test part non-fatally failed.
-static bool TestPartNonfatallyFailed(const TestPartResult& result) {
-  return result.nonfatally_failed();
-}
-
-// Returns true iff the test has a non-fatal failure.
-bool TestResult::HasNonfatalFailure() const {
-  return CountIf(test_part_results_, TestPartNonfatallyFailed) > 0;
-}
-
-// Gets the number of all test parts.  This is the sum of the number
-// of successful test parts and the number of failed test parts.
-int TestResult::total_part_count() const {
-  return static_cast<int>(test_part_results_.size());
-}
-
-// Returns the number of the test properties.
-int TestResult::test_property_count() const {
-  return static_cast<int>(test_properties_.size());
-}
-
-// class Test
-
-// Creates a Test object.
-
-// The c'tor saves the states of all flags.
-Test::Test()
-    : gtest_flag_saver_(new GTEST_FLAG_SAVER_) {
-}
-
-// The d'tor restores the states of all flags.  The actual work is
-// done by the d'tor of the gtest_flag_saver_ field, and thus not
-// visible here.
-Test::~Test() {
-}
-
-// Sets up the test fixture.
-//
-// A sub-class may override this.
-void Test::SetUp() {
-}
-
-// Tears down the test fixture.
-//
-// A sub-class may override this.
-void Test::TearDown() {
-}
-
-// Allows user supplied key value pairs to be recorded for later output.
-void Test::RecordProperty(const std::string& key, const std::string& value) {
-  UnitTest::GetInstance()->RecordProperty(key, value);
-}
-
-// Allows user supplied key value pairs to be recorded for later output.
-void Test::RecordProperty(const std::string& key, int value) {
-  Message value_message;
-  value_message << value;
-  RecordProperty(key, value_message.GetString().c_str());
-}
-
-namespace internal {
-
-void ReportFailureInUnknownLocation(TestPartResult::Type result_type,
-                                    const std::string& message) {
-  // This function is a friend of UnitTest and as such has access to
-  // AddTestPartResult.
-  UnitTest::GetInstance()->AddTestPartResult(
-      result_type,
-      NULL,  // No info about the source file where the exception occurred.
-      -1,    // We have no info on which line caused the exception.
-      message,
-      "");   // No stack trace, either.
-}
-
-}  // namespace internal
-
-// Google Test requires all tests in the same test case to use the same test
-// fixture class.  This function checks if the current test has the
-// same fixture class as the first test in the current test case.  If
-// yes, it returns true; otherwise it generates a Google Test failure and
-// returns false.
-bool Test::HasSameFixtureClass() {
-  internal::UnitTestImpl* const impl = internal::GetUnitTestImpl();
-  const TestCase* const test_case = impl->current_test_case();
-
-  // Info about the first test in the current test case.
-  const TestInfo* const first_test_info = test_case->test_info_list()[0];
-  const internal::TypeId first_fixture_id = first_test_info->fixture_class_id_;
-  const char* const first_test_name = first_test_info->name();
-
-  // Info about the current test.
-  const TestInfo* const this_test_info = impl->current_test_info();
-  const internal::TypeId this_fixture_id = this_test_info->fixture_class_id_;
-  const char* const this_test_name = this_test_info->name();
-
-  if (this_fixture_id != first_fixture_id) {
-    // Is the first test defined using TEST?
-    const bool first_is_TEST = first_fixture_id == internal::GetTestTypeId();
-    // Is this test defined using TEST?
-    const bool this_is_TEST = this_fixture_id == internal::GetTestTypeId();
-
-    if (first_is_TEST || this_is_TEST) {
-      // Both TEST and TEST_F appear in same test case, which is incorrect.
-      // Tell the user how to fix this.
-
-      // Gets the name of the TEST and the name of the TEST_F.  Note
-      // that first_is_TEST and this_is_TEST cannot both be true, as
-      // the fixture IDs are different for the two tests.
-      const char* const TEST_name =
-          first_is_TEST ? first_test_name : this_test_name;
-      const char* const TEST_F_name =
-          first_is_TEST ? this_test_name : first_test_name;
-
-      ADD_FAILURE()
-          << "All tests in the same test case must use the same test fixture\n"
-          << "class, so mixing TEST_F and TEST in the same test case is\n"
-          << "illegal.  In test case " << this_test_info->test_case_name()
-          << ",\n"
-          << "test " << TEST_F_name << " is defined using TEST_F but\n"
-          << "test " << TEST_name << " is defined using TEST.  You probably\n"
-          << "want to change the TEST to TEST_F or move it to another test\n"
-          << "case.";
-    } else {
-      // Two fixture classes with the same name appear in two different
-      // namespaces, which is not allowed. Tell the user how to fix this.
-      ADD_FAILURE()
-          << "All tests in the same test case must use the same test fixture\n"
-          << "class.  However, in test case "
-          << this_test_info->test_case_name() << ",\n"
-          << "you defined test " << first_test_name
-          << " and test " << this_test_name << "\n"
-          << "using two different test fixture classes.  This can happen if\n"
-          << "the two classes are from different namespaces or translation\n"
-          << "units and have the same name.  You should probably rename one\n"
-          << "of the classes to put the tests into different test cases.";
-    }
-    return false;
-  }
-
-  return true;
-}
-
-#if GTEST_HAS_SEH
-
-// Adds an "exception thrown" fatal failure to the current test.  This
-// function returns its result via an output parameter pointer because VC++
-// prohibits creation of objects with destructors on stack in functions
-// using __try (see error C2712).
-static std::string* FormatSehExceptionMessage(DWORD exception_code,
-                                              const char* location) {
-  Message message;
-  message << "SEH exception with code 0x" << std::setbase(16) <<
-    exception_code << std::setbase(10) << " thrown in " << location << ".";
-
-  return new std::string(message.GetString());
-}
-
-#endif  // GTEST_HAS_SEH
-
-namespace internal {
-
-#if GTEST_HAS_EXCEPTIONS
-
-// Adds an "exception thrown" fatal failure to the current test.
-static std::string FormatCxxExceptionMessage(const char* description,
-                                             const char* location) {
-  Message message;
-  if (description != NULL) {
-    message << "C++ exception with description \"" << description << "\"";
-  } else {
-    message << "Unknown C++ exception";
-  }
-  message << " thrown in " << location << ".";
-
-  return message.GetString();
-}
-
-static std::string PrintTestPartResultToString(
-    const TestPartResult& test_part_result);
-
-GoogleTestFailureException::GoogleTestFailureException(
-    const TestPartResult& failure)
-    : ::std::runtime_error(PrintTestPartResultToString(failure).c_str()) {}
-
-#endif  // GTEST_HAS_EXCEPTIONS
-
-// We put these helper functions in the internal namespace as IBM's xlC
-// compiler rejects the code if they were declared static.
-
-// Runs the given method and handles SEH exceptions it throws, when
-// SEH is supported; returns the 0-value for type Result in case of an
-// SEH exception.  (Microsoft compilers cannot handle SEH and C++
-// exceptions in the same function.  Therefore, we provide a separate
-// wrapper function for handling SEH exceptions.)
-template <class T, typename Result>
-Result HandleSehExceptionsInMethodIfSupported(
-    T* object, Result (T::*method)(), const char* location) {
-#if GTEST_HAS_SEH
-  __try {
-    return (object->*method)();
-  } __except (internal::UnitTestOptions::GTestShouldProcessSEH(  // NOLINT
-      GetExceptionCode())) {
-    // We create the exception message on the heap because VC++ prohibits
-    // creation of objects with destructors on stack in functions using __try
-    // (see error C2712).
-    std::string* exception_message = FormatSehExceptionMessage(
-        GetExceptionCode(), location);
-    internal::ReportFailureInUnknownLocation(TestPartResult::kFatalFailure,
-                                             *exception_message);
-    delete exception_message;
-    return static_cast<Result>(0);
-  }
-#else
-  (void)location;
-  return (object->*method)();
-#endif  // GTEST_HAS_SEH
-}
-
-// Runs the given method and catches and reports C++ and/or SEH-style
-// exceptions, if they are supported; returns the 0-value for type
-// Result in case of an SEH exception.
-template <class T, typename Result>
-Result HandleExceptionsInMethodIfSupported(
-    T* object, Result (T::*method)(), const char* location) {
-  // NOTE: The user code can affect the way in which Google Test handles
-  // exceptions by setting GTEST_FLAG(catch_exceptions), but only before
-  // RUN_ALL_TESTS() starts. It is technically possible to check the flag
-  // after the exception is caught and either report or re-throw the
-  // exception based on the flag's value:
-  //
-  // try {
-  //   // Perform the test method.
-  // } catch (...) {
-  //   if (GTEST_FLAG(catch_exceptions))
-  //     // Report the exception as failure.
-  //   else
-  //     throw;  // Re-throws the original exception.
-  // }
-  //
-  // However, the purpose of this flag is to allow the program to drop into
-  // the debugger when the exception is thrown. On most platforms, once the
-  // control enters the catch block, the exception origin information is
-  // lost and the debugger will stop the program at the point of the
-  // re-throw in this function -- instead of at the point of the original
-  // throw statement in the code under test.  For this reason, we perform
-  // the check early, sacrificing the ability to affect Google Test's
-  // exception handling in the method where the exception is thrown.
-  if (internal::GetUnitTestImpl()->catch_exceptions()) {
-#if GTEST_HAS_EXCEPTIONS
-    try {
-      return HandleSehExceptionsInMethodIfSupported(object, method, location);
-    } catch (const internal::GoogleTestFailureException&) {  // NOLINT
-      // This exception type can only be thrown by a failed Google
-      // Test assertion with the intention of letting another testing
-      // framework catch it.  Therefore we just re-throw it.
-      throw;
-    } catch (const std::exception& e) {  // NOLINT
-      internal::ReportFailureInUnknownLocation(
-          TestPartResult::kFatalFailure,
-          FormatCxxExceptionMessage(e.what(), location));
-    } catch (...) {  // NOLINT
-      internal::ReportFailureInUnknownLocation(
-          TestPartResult::kFatalFailure,
-          FormatCxxExceptionMessage(NULL, location));
-    }
-    return static_cast<Result>(0);
-#else
-    return HandleSehExceptionsInMethodIfSupported(object, method, location);
-#endif  // GTEST_HAS_EXCEPTIONS
-  } else {
-    return (object->*method)();
-  }
-}
-
-}  // namespace internal
-
-// Runs the test and updates the test result.
-void Test::Run() {
-  if (!HasSameFixtureClass()) return;
-
-  internal::UnitTestImpl* const impl = internal::GetUnitTestImpl();
-  impl->os_stack_trace_getter()->UponLeavingGTest();
-  internal::HandleExceptionsInMethodIfSupported(this, &Test::SetUp, "SetUp()");
-  // We will run the test only if SetUp() was successful.
-  if (!HasFatalFailure()) {
-    impl->os_stack_trace_getter()->UponLeavingGTest();
-    internal::HandleExceptionsInMethodIfSupported(
-        this, &Test::TestBody, "the test body");
-  }
-
-  // However, we want to clean up as much as possible.  Hence we will
-  // always call TearDown(), even if SetUp() or the test body has
-  // failed.
-  impl->os_stack_trace_getter()->UponLeavingGTest();
-  internal::HandleExceptionsInMethodIfSupported(
-      this, &Test::TearDown, "TearDown()");
-}
-
-// Returns true iff the current test has a fatal failure.
-bool Test::HasFatalFailure() {
-  return internal::GetUnitTestImpl()->current_test_result()->HasFatalFailure();
-}
-
-// Returns true iff the current test has a non-fatal failure.
-bool Test::HasNonfatalFailure() {
-  return internal::GetUnitTestImpl()->current_test_result()->
-      HasNonfatalFailure();
-}
-
-// class TestInfo
-
-// Constructs a TestInfo object. It assumes ownership of the test factory
-// object.
-TestInfo::TestInfo(const std::string& a_test_case_name,
-                   const std::string& a_name,
-                   const char* a_type_param,
-                   const char* a_value_param,
-                   internal::CodeLocation a_code_location,
-                   internal::TypeId fixture_class_id,
-                   internal::TestFactoryBase* factory)
-    : test_case_name_(a_test_case_name),
-      name_(a_name),
-      type_param_(a_type_param ? new std::string(a_type_param) : NULL),
-      value_param_(a_value_param ? new std::string(a_value_param) : NULL),
-      location_(a_code_location),
-      fixture_class_id_(fixture_class_id),
-      should_run_(false),
-      is_disabled_(false),
-      matches_filter_(false),
-      factory_(factory),
-      result_() {}
-
-// Destructs a TestInfo object.
-TestInfo::~TestInfo() { delete factory_; }
-
-namespace internal {
-
-// Creates a new TestInfo object and registers it with Google Test;
-// returns the created object.
-//
-// Arguments:
-//
-//   test_case_name:   name of the test case
-//   name:             name of the test
-//   type_param:       the name of the test's type parameter, or NULL if
-//                     this is not a typed or a type-parameterized test.
-//   value_param:      text representation of the test's value parameter,
-//                     or NULL if this is not a value-parameterized test.
-//   code_location:    code location where the test is defined
-//   fixture_class_id: ID of the test fixture class
-//   set_up_tc:        pointer to the function that sets up the test case
-//   tear_down_tc:     pointer to the function that tears down the test case
-//   factory:          pointer to the factory that creates a test object.
-//                     The newly created TestInfo instance will assume
-//                     ownership of the factory object.
-TestInfo* MakeAndRegisterTestInfo(
-    const char* test_case_name,
-    const char* name,
-    const char* type_param,
-    const char* value_param,
-    CodeLocation code_location,
-    TypeId fixture_class_id,
-    SetUpTestCaseFunc set_up_tc,
-    TearDownTestCaseFunc tear_down_tc,
-    TestFactoryBase* factory) {
-  TestInfo* const test_info =
-      new TestInfo(test_case_name, name, type_param, value_param,
-                   code_location, fixture_class_id, factory);
-  GetUnitTestImpl()->AddTestInfo(set_up_tc, tear_down_tc, test_info);
-  return test_info;
-}
-
-#if GTEST_HAS_PARAM_TEST
-void ReportInvalidTestCaseType(const char* test_case_name,
-                               CodeLocation code_location) {
-  Message errors;
-  errors
-      << "Attempted redefinition of test case " << test_case_name << ".\n"
-      << "All tests in the same test case must use the same test fixture\n"
-      << "class.  However, in test case " << test_case_name << ", you tried\n"
-      << "to define a test using a fixture class different from the one\n"
-      << "used earlier. This can happen if the two fixture classes are\n"
-      << "from different namespaces and have the same name. You should\n"
-      << "probably rename one of the classes to put the tests into different\n"
-      << "test cases.";
-
-  fprintf(stderr, "%s %s",
-          FormatFileLocation(code_location.file.c_str(),
-                             code_location.line).c_str(),
-          errors.GetString().c_str());
-}
-#endif  // GTEST_HAS_PARAM_TEST
-
-}  // namespace internal
-
-namespace {
-
-// A predicate that checks the test name of a TestInfo against a known
-// value.
-//
-// This is used for implementation of the TestCase class only.  We put
-// it in the anonymous namespace to prevent polluting the outer
-// namespace.
-//
-// TestNameIs is copyable.
-class TestNameIs {
- public:
-  // Constructor.
-  //
-  // TestNameIs has NO default constructor.
-  explicit TestNameIs(const char* name)
-      : name_(name) {}
-
-  // Returns true iff the test name of test_info matches name_.
-  bool operator()(const TestInfo * test_info) const {
-    return test_info && test_info->name() == name_;
-  }
-
- private:
-  std::string name_;
-};
-
-}  // namespace
-
-namespace internal {
-
-// This method expands all parameterized tests registered with macros TEST_P
-// and INSTANTIATE_TEST_CASE_P into regular tests and registers those.
-// This will be done just once during the program runtime.
-void UnitTestImpl::RegisterParameterizedTests() {
-#if GTEST_HAS_PARAM_TEST
-  if (!parameterized_tests_registered_) {
-    parameterized_test_registry_.RegisterTests();
-    parameterized_tests_registered_ = true;
-  }
-#endif
-}
-
-}  // namespace internal
-
-// Creates the test object, runs it, records its result, and then
-// deletes it.
-void TestInfo::Run() {
-  if (!should_run_) return;
-
-  // Tells UnitTest where to store test result.
-  internal::UnitTestImpl* const impl = internal::GetUnitTestImpl();
-  impl->set_current_test_info(this);
-
-  TestEventListener* repeater = UnitTest::GetInstance()->listeners().repeater();
-
-  // Notifies the unit test event listeners that a test is about to start.
-  repeater->OnTestStart(*this);
-
-  const TimeInMillis start = internal::GetTimeInMillis();
-
-  impl->os_stack_trace_getter()->UponLeavingGTest();
-
-  // Creates the test object.
-  Test* const test = internal::HandleExceptionsInMethodIfSupported(
-      factory_, &internal::TestFactoryBase::CreateTest,
-      "the test fixture's constructor");
-
-  // Runs the test only if the test object was created and its
-  // constructor didn't generate a fatal failure.
-  if ((test != NULL) && !Test::HasFatalFailure()) {
-    // This doesn't throw as all user code that can throw are wrapped into
-    // exception handling code.
-    test->Run();
-  }
-
-  // Deletes the test object.
-  impl->os_stack_trace_getter()->UponLeavingGTest();
-  internal::HandleExceptionsInMethodIfSupported(
-      test, &Test::DeleteSelf_, "the test fixture's destructor");
-
-  result_.set_elapsed_time(internal::GetTimeInMillis() - start);
-
-  // Notifies the unit test event listener that a test has just finished.
-  repeater->OnTestEnd(*this);
-
-  // Tells UnitTest to stop associating assertion results to this
-  // test.
-  impl->set_current_test_info(NULL);
-}
-
-// class TestCase
-
-// Gets the number of successful tests in this test case.
-int TestCase::successful_test_count() const {
-  return CountIf(test_info_list_, TestPassed);
-}
-
-// Gets the number of failed tests in this test case.
-int TestCase::failed_test_count() const {
-  return CountIf(test_info_list_, TestFailed);
-}
-
-// Gets the number of disabled tests that will be reported in the XML report.
-int TestCase::reportable_disabled_test_count() const {
-  return CountIf(test_info_list_, TestReportableDisabled);
-}
-
-// Gets the number of disabled tests in this test case.
-int TestCase::disabled_test_count() const {
-  return CountIf(test_info_list_, TestDisabled);
-}
-
-// Gets the number of tests to be printed in the XML report.
-int TestCase::reportable_test_count() const {
-  return CountIf(test_info_list_, TestReportable);
-}
-
-// Get the number of tests in this test case that should run.
-int TestCase::test_to_run_count() const {
-  return CountIf(test_info_list_, ShouldRunTest);
-}
-
-// Gets the number of all tests.
-int TestCase::total_test_count() const {
-  return static_cast<int>(test_info_list_.size());
-}
-
-// Creates a TestCase with the given name.
-//
-// Arguments:
-//
-//   name:         name of the test case
-//   a_type_param: the name of the test case's type parameter, or NULL if
-//                 this is not a typed or a type-parameterized test case.
-//   set_up_tc:    pointer to the function that sets up the test case
-//   tear_down_tc: pointer to the function that tears down the test case
-TestCase::TestCase(const char* a_name, const char* a_type_param,
-                   Test::SetUpTestCaseFunc set_up_tc,
-                   Test::TearDownTestCaseFunc tear_down_tc)
-    : name_(a_name),
-      type_param_(a_type_param ? new std::string(a_type_param) : NULL),
-      set_up_tc_(set_up_tc),
-      tear_down_tc_(tear_down_tc),
-      should_run_(false),
-      elapsed_time_(0) {
-}
-
-// Destructor of TestCase.
-TestCase::~TestCase() {
-  // Deletes every Test in the collection.
-  ForEach(test_info_list_, internal::Delete<TestInfo>);
-}
-
-// Returns the i-th test among all the tests. i can range from 0 to
-// total_test_count() - 1. If i is not in that range, returns NULL.
-const TestInfo* TestCase::GetTestInfo(int i) const {
-  const int index = GetElementOr(test_indices_, i, -1);
-  return index < 0 ? NULL : test_info_list_[index];
-}
-
-// Returns the i-th test among all the tests. i can range from 0 to
-// total_test_count() - 1. If i is not in that range, returns NULL.
-TestInfo* TestCase::GetMutableTestInfo(int i) {
-  const int index = GetElementOr(test_indices_, i, -1);
-  return index < 0 ? NULL : test_info_list_[index];
-}
-
-// Adds a test to this test case.  Will delete the test upon
-// destruction of the TestCase object.
-void TestCase::AddTestInfo(TestInfo * test_info) {
-  test_info_list_.push_back(test_info);
-  test_indices_.push_back(static_cast<int>(test_indices_.size()));
-}
-
-// Runs every test in this TestCase.
-void TestCase::Run() {
-  if (!should_run_) return;
-
-  internal::UnitTestImpl* const impl = internal::GetUnitTestImpl();
-  impl->set_current_test_case(this);
-
-  TestEventListener* repeater = UnitTest::GetInstance()->listeners().repeater();
-
-  repeater->OnTestCaseStart(*this);
-  impl->os_stack_trace_getter()->UponLeavingGTest();
-  internal::HandleExceptionsInMethodIfSupported(
-      this, &TestCase::RunSetUpTestCase, "SetUpTestCase()");
-
-  const internal::TimeInMillis start = internal::GetTimeInMillis();
-  for (int i = 0; i < total_test_count(); i++) {
-    GetMutableTestInfo(i)->Run();
-  }
-  elapsed_time_ = internal::GetTimeInMillis() - start;
-
-  impl->os_stack_trace_getter()->UponLeavingGTest();
-  internal::HandleExceptionsInMethodIfSupported(
-      this, &TestCase::RunTearDownTestCase, "TearDownTestCase()");
-
-  repeater->OnTestCaseEnd(*this);
-  impl->set_current_test_case(NULL);
-}
-
-// Clears the results of all tests in this test case.
-void TestCase::ClearResult() {
-  ad_hoc_test_result_.Clear();
-  ForEach(test_info_list_, TestInfo::ClearTestResult);
-}
-
-// Shuffles the tests in this test case.
-void TestCase::ShuffleTests(internal::Random* random) {
-  Shuffle(random, &test_indices_);
-}
-
-// Restores the test order to before the first shuffle.
-void TestCase::UnshuffleTests() {
-  for (size_t i = 0; i < test_indices_.size(); i++) {
-    test_indices_[i] = static_cast<int>(i);
-  }
-}
-
-// Formats a countable noun.  Depending on its quantity, either the
-// singular form or the plural form is used. e.g.
-//
-// FormatCountableNoun(1, "formula", "formuli") returns "1 formula".
-// FormatCountableNoun(5, "book", "books") returns "5 books".
-static std::string FormatCountableNoun(int count,
-                                       const char * singular_form,
-                                       const char * plural_form) {
-  return internal::StreamableToString(count) + " " +
-      (count == 1 ? singular_form : plural_form);
-}
-
-// Formats the count of tests.
-static std::string FormatTestCount(int test_count) {
-  return FormatCountableNoun(test_count, "test", "tests");
-}
-
-// Formats the count of test cases.
-static std::string FormatTestCaseCount(int test_case_count) {
-  return FormatCountableNoun(test_case_count, "test case", "test cases");
-}
-
-// Converts a TestPartResult::Type enum to human-friendly string
-// representation.  Both kNonFatalFailure and kFatalFailure are translated
-// to "Failure", as the user usually doesn't care about the difference
-// between the two when viewing the test result.
-static const char * TestPartResultTypeToString(TestPartResult::Type type) {
-  switch (type) {
-    case TestPartResult::kSuccess:
-      return "Success";
-
-    case TestPartResult::kNonFatalFailure:
-    case TestPartResult::kFatalFailure:
-#ifdef _MSC_VER
-      return "error: ";
-#else
-      return "Failure\n";
-#endif
-    default:
-      return "Unknown result type";
-  }
-}
-
-namespace internal {
-
-// Prints a TestPartResult to an std::string.
-static std::string PrintTestPartResultToString(
-    const TestPartResult& test_part_result) {
-  return (Message()
-          << internal::FormatFileLocation(test_part_result.file_name(),
-                                          test_part_result.line_number())
-          << " " << TestPartResultTypeToString(test_part_result.type())
-          << test_part_result.message()).GetString();
-}
-
-// Prints a TestPartResult.
-static void PrintTestPartResult(const TestPartResult& test_part_result) {
-  const std::string& result =
-      PrintTestPartResultToString(test_part_result);
-  printf("%s\n", result.c_str());
-  fflush(stdout);
-  // If the test program runs in Visual Studio or a debugger, the
-  // following statements add the test part result message to the Output
-  // window such that the user can double-click on it to jump to the
-  // corresponding source code location; otherwise they do nothing.
-#if GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_MOBILE
-  // We don't call OutputDebugString*() on Windows Mobile, as printing
-  // to stdout is done by OutputDebugString() there already - we don't
-  // want the same message printed twice.
-  ::OutputDebugStringA(result.c_str());
-  ::OutputDebugStringA("\n");
-#endif
-}
-
-// class PrettyUnitTestResultPrinter
-
-enum GTestColor {
-  COLOR_DEFAULT,
-  COLOR_RED,
-  COLOR_GREEN,
-  COLOR_YELLOW
-};
-
-#if GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_MOBILE && \
-    !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT
-
-// Returns the character attribute for the given color.
-WORD GetColorAttribute(GTestColor color) {
-  switch (color) {
-    case COLOR_RED:    return FOREGROUND_RED;
-    case COLOR_GREEN:  return FOREGROUND_GREEN;
-    case COLOR_YELLOW: return FOREGROUND_RED | FOREGROUND_GREEN;
-    default:           return 0;
-  }
-}
-
-#else
-
-// Returns the ANSI color code for the given color.  COLOR_DEFAULT is
-// an invalid input.
-const char* GetAnsiColorCode(GTestColor color) {
-  switch (color) {
-    case COLOR_RED:     return "1";
-    case COLOR_GREEN:   return "2";
-    case COLOR_YELLOW:  return "3";
-    default:            return NULL;
-  };
-}
-
-#endif  // GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_MOBILE
-
-// Returns true iff Google Test should use colors in the output.
-bool ShouldUseColor(bool stdout_is_tty) {
-  const char* const gtest_color = GTEST_FLAG(color).c_str();
-
-  if (String::CaseInsensitiveCStringEquals(gtest_color, "auto")) {
-#if GTEST_OS_WINDOWS
-    // On Windows the TERM variable is usually not set, but the
-    // console there does support colors.
-    return stdout_is_tty;
-#else
-    // On non-Windows platforms, we rely on the TERM variable.
-    const char* const term = posix::GetEnv("TERM");
-    const bool term_supports_color =
-        String::CStringEquals(term, "xterm") ||
-        String::CStringEquals(term, "xterm-color") ||
-        String::CStringEquals(term, "xterm-256color") ||
-        String::CStringEquals(term, "screen") ||
-        String::CStringEquals(term, "screen-256color") ||
-        String::CStringEquals(term, "tmux") ||
-        String::CStringEquals(term, "tmux-256color") ||
-        String::CStringEquals(term, "rxvt-unicode") ||
-        String::CStringEquals(term, "rxvt-unicode-256color") ||
-        String::CStringEquals(term, "linux") ||
-        String::CStringEquals(term, "cygwin");
-    return stdout_is_tty && term_supports_color;
-#endif  // GTEST_OS_WINDOWS
-  }
-
-  return String::CaseInsensitiveCStringEquals(gtest_color, "yes") ||
-      String::CaseInsensitiveCStringEquals(gtest_color, "true") ||
-      String::CaseInsensitiveCStringEquals(gtest_color, "t") ||
-      String::CStringEquals(gtest_color, "1");
-  // We take "yes", "true", "t", and "1" as meaning "yes".  If the
-  // value is neither one of these nor "auto", we treat it as "no" to
-  // be conservative.
-}
-
-// Helpers for printing colored strings to stdout. Note that on Windows, we
-// cannot simply emit special characters and have the terminal change colors.
-// This routine must actually emit the characters rather than return a string
-// that would be colored when printed, as can be done on Linux.
-void ColoredPrintf(GTestColor color, const char* fmt, ...) {
-  va_list args;
-  va_start(args, fmt);
-
-#if GTEST_OS_WINDOWS_MOBILE || GTEST_OS_SYMBIAN || GTEST_OS_ZOS || \
-    GTEST_OS_IOS || GTEST_OS_WINDOWS_PHONE || GTEST_OS_WINDOWS_RT
-  const bool use_color = AlwaysFalse();
-#else
-  static const bool in_color_mode =
-      ShouldUseColor(posix::IsATTY(posix::FileNo(stdout)) != 0);
-  const bool use_color = in_color_mode && (color != COLOR_DEFAULT);
-#endif  // GTEST_OS_WINDOWS_MOBILE || GTEST_OS_SYMBIAN || GTEST_OS_ZOS
-  // The '!= 0' comparison is necessary to satisfy MSVC 7.1.
-
-  if (!use_color) {
-    vprintf(fmt, args);
-    va_end(args);
-    return;
-  }
-
-#if GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_MOBILE && \
-    !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT
-  const HANDLE stdout_handle = GetStdHandle(STD_OUTPUT_HANDLE);
-
-  // Gets the current text color.
-  CONSOLE_SCREEN_BUFFER_INFO buffer_info;
-  GetConsoleScreenBufferInfo(stdout_handle, &buffer_info);
-  const WORD old_color_attrs = buffer_info.wAttributes;
-
-  // We need to flush the stream buffers into the console before each
-  // SetConsoleTextAttribute call lest it affect the text that is already
-  // printed but has not yet reached the console.
-  fflush(stdout);
-  SetConsoleTextAttribute(stdout_handle,
-                          GetColorAttribute(color) | FOREGROUND_INTENSITY);
-  vprintf(fmt, args);
-
-  fflush(stdout);
-  // Restores the text color.
-  SetConsoleTextAttribute(stdout_handle, old_color_attrs);
-#else
-  printf("\033[0;3%sm", GetAnsiColorCode(color));
-  vprintf(fmt, args);
-  printf("\033[m");  // Resets the terminal to default.
-#endif  // GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_MOBILE
-  va_end(args);
-}
-
-// Text printed in Google Test's text output and --gunit_list_tests
-// output to label the type parameter and value parameter for a test.
-static const char kTypeParamLabel[] = "TypeParam";
-static const char kValueParamLabel[] = "GetParam()";
-
-void PrintFullTestCommentIfPresent(const TestInfo& test_info) {
-  const char* const type_param = test_info.type_param();
-  const char* const value_param = test_info.value_param();
-
-  if (type_param != NULL || value_param != NULL) {
-    printf(", where ");
-    if (type_param != NULL) {
-      printf("%s = %s", kTypeParamLabel, type_param);
-      if (value_param != NULL)
-        printf(" and ");
-    }
-    if (value_param != NULL) {
-      printf("%s = %s", kValueParamLabel, value_param);
-    }
-  }
-}
-
-// This class implements the TestEventListener interface.
-//
-// Class PrettyUnitTestResultPrinter is copyable.
-class PrettyUnitTestResultPrinter : public TestEventListener {
- public:
-  PrettyUnitTestResultPrinter() {}
-  static void PrintTestName(const char * test_case, const char * test) {
-    printf("%s.%s", test_case, test);
-  }
-
-  // The following methods override what's in the TestEventListener class.
-  virtual void OnTestProgramStart(const UnitTest& /*unit_test*/) {}
-  virtual void OnTestIterationStart(const UnitTest& unit_test, int iteration);
-  virtual void OnEnvironmentsSetUpStart(const UnitTest& unit_test);
-  virtual void OnEnvironmentsSetUpEnd(const UnitTest& /*unit_test*/) {}
-  virtual void OnTestCaseStart(const TestCase& test_case);
-  virtual void OnTestStart(const TestInfo& test_info);
-  virtual void OnTestPartResult(const TestPartResult& result);
-  virtual void OnTestEnd(const TestInfo& test_info);
-  virtual void OnTestCaseEnd(const TestCase& test_case);
-  virtual void OnEnvironmentsTearDownStart(const UnitTest& unit_test);
-  virtual void OnEnvironmentsTearDownEnd(const UnitTest& /*unit_test*/) {}
-  virtual void OnTestIterationEnd(const UnitTest& unit_test, int iteration);
-  virtual void OnTestProgramEnd(const UnitTest& /*unit_test*/) {}
-
- private:
-  static void PrintFailedTests(const UnitTest& unit_test);
-};
-
-  // Fired before each iteration of tests starts.
-void PrettyUnitTestResultPrinter::OnTestIterationStart(
-    const UnitTest& unit_test, int iteration) {
-  if (GTEST_FLAG(repeat) != 1)
-    printf("\nRepeating all tests (iteration %d) . . .\n\n", iteration + 1);
-
-  const char* const filter = GTEST_FLAG(filter).c_str();
-
-  // Prints the filter if it's not *.  This reminds the user that some
-  // tests may be skipped.
-  if (!String::CStringEquals(filter, kUniversalFilter)) {
-    ColoredPrintf(COLOR_YELLOW,
-                  "Note: %s filter = %s\n", GTEST_NAME_, filter);
-  }
-
-  if (internal::ShouldShard(kTestTotalShards, kTestShardIndex, false)) {
-    const Int32 shard_index = Int32FromEnvOrDie(kTestShardIndex, -1);
-    ColoredPrintf(COLOR_YELLOW,
-                  "Note: This is test shard %d of %s.\n",
-                  static_cast<int>(shard_index) + 1,
-                  internal::posix::GetEnv(kTestTotalShards));
-  }
-
-  if (GTEST_FLAG(shuffle)) {
-    ColoredPrintf(COLOR_YELLOW,
-                  "Note: Randomizing tests' orders with a seed of %d .\n",
-                  unit_test.random_seed());
-  }
-
-  ColoredPrintf(COLOR_GREEN,  "[==========] ");
-  printf("Running %s from %s.\n",
-         FormatTestCount(unit_test.test_to_run_count()).c_str(),
-         FormatTestCaseCount(unit_test.test_case_to_run_count()).c_str());
-  fflush(stdout);
-}
-
-void PrettyUnitTestResultPrinter::OnEnvironmentsSetUpStart(
-    const UnitTest& /*unit_test*/) {
-  ColoredPrintf(COLOR_GREEN,  "[----------] ");
-  printf("Global test environment set-up.\n");
-  fflush(stdout);
-}
-
-void PrettyUnitTestResultPrinter::OnTestCaseStart(const TestCase& test_case) {
-  const std::string counts =
-      FormatCountableNoun(test_case.test_to_run_count(), "test", "tests");
-  ColoredPrintf(COLOR_GREEN, "[----------] ");
-  printf("%s from %s", counts.c_str(), test_case.name());
-  if (test_case.type_param() == NULL) {
-    printf("\n");
-  } else {
-    printf(", where %s = %s\n", kTypeParamLabel, test_case.type_param());
-  }
-  fflush(stdout);
-}
-
-void PrettyUnitTestResultPrinter::OnTestStart(const TestInfo& test_info) {
-  ColoredPrintf(COLOR_GREEN,  "[ RUN      ] ");
-  PrintTestName(test_info.test_case_name(), test_info.name());
-  printf("\n");
-  fflush(stdout);
-}
-
-// Called after an assertion failure.
-void PrettyUnitTestResultPrinter::OnTestPartResult(
-    const TestPartResult& result) {
-  // If the test part succeeded, we don't need to do anything.
-  if (result.type() == TestPartResult::kSuccess)
-    return;
-
-  // Print failure message from the assertion (e.g. expected this and got that).
-  PrintTestPartResult(result);
-  fflush(stdout);
-}
-
-void PrettyUnitTestResultPrinter::OnTestEnd(const TestInfo& test_info) {
-  if (test_info.result()->Passed()) {
-    ColoredPrintf(COLOR_GREEN, "[       OK ] ");
-  } else {
-    ColoredPrintf(COLOR_RED, "[  FAILED  ] ");
-  }
-  PrintTestName(test_info.test_case_name(), test_info.name());
-  if (test_info.result()->Failed())
-    PrintFullTestCommentIfPresent(test_info);
-
-  if (GTEST_FLAG(print_time)) {
-    printf(" (%s ms)\n", internal::StreamableToString(
-           test_info.result()->elapsed_time()).c_str());
-  } else {
-    printf("\n");
-  }
-  fflush(stdout);
-}
-
-void PrettyUnitTestResultPrinter::OnTestCaseEnd(const TestCase& test_case) {
-  if (!GTEST_FLAG(print_time)) return;
-
-  const std::string counts =
-      FormatCountableNoun(test_case.test_to_run_count(), "test", "tests");
-  ColoredPrintf(COLOR_GREEN, "[----------] ");
-  printf("%s from %s (%s ms total)\n\n",
-         counts.c_str(), test_case.name(),
-         internal::StreamableToString(test_case.elapsed_time()).c_str());
-  fflush(stdout);
-}
-
-void PrettyUnitTestResultPrinter::OnEnvironmentsTearDownStart(
-    const UnitTest& /*unit_test*/) {
-  ColoredPrintf(COLOR_GREEN,  "[----------] ");
-  printf("Global test environment tear-down\n");
-  fflush(stdout);
-}
-
-// Internal helper for printing the list of failed tests.
-void PrettyUnitTestResultPrinter::PrintFailedTests(const UnitTest& unit_test) {
-  const int failed_test_count = unit_test.failed_test_count();
-  if (failed_test_count == 0) {
-    return;
-  }
-
-  for (int i = 0; i < unit_test.total_test_case_count(); ++i) {
-    const TestCase& test_case = *unit_test.GetTestCase(i);
-    if (!test_case.should_run() || (test_case.failed_test_count() == 0)) {
-      continue;
-    }
-    for (int j = 0; j < test_case.total_test_count(); ++j) {
-      const TestInfo& test_info = *test_case.GetTestInfo(j);
-      if (!test_info.should_run() || test_info.result()->Passed()) {
-        continue;
-      }
-      ColoredPrintf(COLOR_RED, "[  FAILED  ] ");
-      printf("%s.%s", test_case.name(), test_info.name());
-      PrintFullTestCommentIfPresent(test_info);
-      printf("\n");
-    }
-  }
-}
-
-void PrettyUnitTestResultPrinter::OnTestIterationEnd(const UnitTest& unit_test,
-                                                     int /*iteration*/) {
-  ColoredPrintf(COLOR_GREEN,  "[==========] ");
-  printf("%s from %s ran.",
-         FormatTestCount(unit_test.test_to_run_count()).c_str(),
-         FormatTestCaseCount(unit_test.test_case_to_run_count()).c_str());
-  if (GTEST_FLAG(print_time)) {
-    printf(" (%s ms total)",
-           internal::StreamableToString(unit_test.elapsed_time()).c_str());
-  }
-  printf("\n");
-  ColoredPrintf(COLOR_GREEN,  "[  PASSED  ] ");
-  printf("%s.\n", FormatTestCount(unit_test.successful_test_count()).c_str());
-
-  int num_failures = unit_test.failed_test_count();
-  if (!unit_test.Passed()) {
-    const int failed_test_count = unit_test.failed_test_count();
-    ColoredPrintf(COLOR_RED,  "[  FAILED  ] ");
-    printf("%s, listed below:\n", FormatTestCount(failed_test_count).c_str());
-    PrintFailedTests(unit_test);
-    printf("\n%2d FAILED %s\n", num_failures,
-                        num_failures == 1 ? "TEST" : "TESTS");
-  }
-
-  int num_disabled = unit_test.reportable_disabled_test_count();
-  if (num_disabled && !GTEST_FLAG(also_run_disabled_tests)) {
-    if (!num_failures) {
-      printf("\n");  // Add a spacer if no FAILURE banner is displayed.
-    }
-    ColoredPrintf(COLOR_YELLOW,
-                  "  YOU HAVE %d DISABLED %s\n\n",
-                  num_disabled,
-                  num_disabled == 1 ? "TEST" : "TESTS");
-  }
-  // Ensure that Google Test output is printed before, e.g., heapchecker output.
-  fflush(stdout);
-}
-
-// End PrettyUnitTestResultPrinter
-
-// class TestEventRepeater
-//
-// This class forwards events to other event listeners.
-class TestEventRepeater : public TestEventListener {
- public:
-  TestEventRepeater() : forwarding_enabled_(true) {}
-  virtual ~TestEventRepeater();
-  void Append(TestEventListener *listener);
-  TestEventListener* Release(TestEventListener* listener);
-
-  // Controls whether events will be forwarded to listeners_. Set to false
-  // in death test child processes.
-  bool forwarding_enabled() const { return forwarding_enabled_; }
-  void set_forwarding_enabled(bool enable) { forwarding_enabled_ = enable; }
-
-  virtual void OnTestProgramStart(const UnitTest& unit_test);
-  virtual void OnTestIterationStart(const UnitTest& unit_test, int iteration);
-  virtual void OnEnvironmentsSetUpStart(const UnitTest& unit_test);
-  virtual void OnEnvironmentsSetUpEnd(const UnitTest& unit_test);
-  virtual void OnTestCaseStart(const TestCase& test_case);
-  virtual void OnTestStart(const TestInfo& test_info);
-  virtual void OnTestPartResult(const TestPartResult& result);
-  virtual void OnTestEnd(const TestInfo& test_info);
-  virtual void OnTestCaseEnd(const TestCase& test_case);
-  virtual void OnEnvironmentsTearDownStart(const UnitTest& unit_test);
-  virtual void OnEnvironmentsTearDownEnd(const UnitTest& unit_test);
-  virtual void OnTestIterationEnd(const UnitTest& unit_test, int iteration);
-  virtual void OnTestProgramEnd(const UnitTest& unit_test);
-
- private:
-  // Controls whether events will be forwarded to listeners_. Set to false
-  // in death test child processes.
-  bool forwarding_enabled_;
-  // The list of listeners that receive events.
-  std::vector<TestEventListener*> listeners_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestEventRepeater);
-};
-
-TestEventRepeater::~TestEventRepeater() {
-  ForEach(listeners_, Delete<TestEventListener>);
-}
-
-void TestEventRepeater::Append(TestEventListener *listener) {
-  listeners_.push_back(listener);
-}
-
-// TODO(vladl@google.com): Factor the search functionality into Vector::Find.
-TestEventListener* TestEventRepeater::Release(TestEventListener *listener) {
-  for (size_t i = 0; i < listeners_.size(); ++i) {
-    if (listeners_[i] == listener) {
-      listeners_.erase(listeners_.begin() + i);
-      return listener;
-    }
-  }
-
-  return NULL;
-}
-
-// Since most methods are very similar, use macros to reduce boilerplate.
-// This defines a member that forwards the call to all listeners.
-#define GTEST_REPEATER_METHOD_(Name, Type) \
-void TestEventRepeater::Name(const Type& parameter) { \
-  if (forwarding_enabled_) { \
-    for (size_t i = 0; i < listeners_.size(); i++) { \
-      listeners_[i]->Name(parameter); \
-    } \
-  } \
-}
-// This defines a member that forwards the call to all listeners in reverse
-// order.
-#define GTEST_REVERSE_REPEATER_METHOD_(Name, Type) \
-void TestEventRepeater::Name(const Type& parameter) { \
-  if (forwarding_enabled_) { \
-    for (int i = static_cast<int>(listeners_.size()) - 1; i >= 0; i--) { \
-      listeners_[i]->Name(parameter); \
-    } \
-  } \
-}
-
-GTEST_REPEATER_METHOD_(OnTestProgramStart, UnitTest)
-GTEST_REPEATER_METHOD_(OnEnvironmentsSetUpStart, UnitTest)
-GTEST_REPEATER_METHOD_(OnTestCaseStart, TestCase)
-GTEST_REPEATER_METHOD_(OnTestStart, TestInfo)
-GTEST_REPEATER_METHOD_(OnTestPartResult, TestPartResult)
-GTEST_REPEATER_METHOD_(OnEnvironmentsTearDownStart, UnitTest)
-GTEST_REVERSE_REPEATER_METHOD_(OnEnvironmentsSetUpEnd, UnitTest)
-GTEST_REVERSE_REPEATER_METHOD_(OnEnvironmentsTearDownEnd, UnitTest)
-GTEST_REVERSE_REPEATER_METHOD_(OnTestEnd, TestInfo)
-GTEST_REVERSE_REPEATER_METHOD_(OnTestCaseEnd, TestCase)
-GTEST_REVERSE_REPEATER_METHOD_(OnTestProgramEnd, UnitTest)
-
-#undef GTEST_REPEATER_METHOD_
-#undef GTEST_REVERSE_REPEATER_METHOD_
-
-void TestEventRepeater::OnTestIterationStart(const UnitTest& unit_test,
-                                             int iteration) {
-  if (forwarding_enabled_) {
-    for (size_t i = 0; i < listeners_.size(); i++) {
-      listeners_[i]->OnTestIterationStart(unit_test, iteration);
-    }
-  }
-}
-
-void TestEventRepeater::OnTestIterationEnd(const UnitTest& unit_test,
-                                           int iteration) {
-  if (forwarding_enabled_) {
-    for (int i = static_cast<int>(listeners_.size()) - 1; i >= 0; i--) {
-      listeners_[i]->OnTestIterationEnd(unit_test, iteration);
-    }
-  }
-}
-
-// End TestEventRepeater
-
-// This class generates an XML output file.
-class XmlUnitTestResultPrinter : public EmptyTestEventListener {
- public:
-  explicit XmlUnitTestResultPrinter(const char* output_file);
-
-  virtual void OnTestIterationEnd(const UnitTest& unit_test, int iteration);
-
- private:
-  // Is c a whitespace character that is normalized to a space character
-  // when it appears in an XML attribute value?
-  static bool IsNormalizableWhitespace(char c) {
-    return c == 0x9 || c == 0xA || c == 0xD;
-  }
-
-  // May c appear in a well-formed XML document?
-  static bool IsValidXmlCharacter(char c) {
-    return IsNormalizableWhitespace(c) || c >= 0x20;
-  }
-
-  // Returns an XML-escaped copy of the input string str.  If
-  // is_attribute is true, the text is meant to appear as an attribute
-  // value, and normalizable whitespace is preserved by replacing it
-  // with character references.
-  static std::string EscapeXml(const std::string& str, bool is_attribute);
-
-  // Returns the given string with all characters invalid in XML removed.
-  static std::string RemoveInvalidXmlCharacters(const std::string& str);
-
-  // Convenience wrapper around EscapeXml when str is an attribute value.
-  static std::string EscapeXmlAttribute(const std::string& str) {
-    return EscapeXml(str, true);
-  }
-
-  // Convenience wrapper around EscapeXml when str is not an attribute value.
-  static std::string EscapeXmlText(const char* str) {
-    return EscapeXml(str, false);
-  }
-
-  // Verifies that the given attribute belongs to the given element and
-  // streams the attribute as XML.
-  static void OutputXmlAttribute(std::ostream* stream,
-                                 const std::string& element_name,
-                                 const std::string& name,
-                                 const std::string& value);
-
-  // Streams an XML CDATA section, escaping invalid CDATA sequences as needed.
-  static void OutputXmlCDataSection(::std::ostream* stream, const char* data);
-
-  // Streams an XML representation of a TestInfo object.
-  static void OutputXmlTestInfo(::std::ostream* stream,
-                                const char* test_case_name,
-                                const TestInfo& test_info);
-
-  // Prints an XML representation of a TestCase object
-  static void PrintXmlTestCase(::std::ostream* stream,
-                               const TestCase& test_case);
-
-  // Prints an XML summary of unit_test to output stream out.
-  static void PrintXmlUnitTest(::std::ostream* stream,
-                               const UnitTest& unit_test);
-
-  // Produces a string representing the test properties in a result as space
-  // delimited XML attributes based on the property key="value" pairs.
-  // When the std::string is not empty, it includes a space at the beginning,
-  // to delimit this attribute from prior attributes.
-  static std::string TestPropertiesAsXmlAttributes(const TestResult& result);
-
-  // The output file.
-  const std::string output_file_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(XmlUnitTestResultPrinter);
-};
-
-// Creates a new XmlUnitTestResultPrinter.
-XmlUnitTestResultPrinter::XmlUnitTestResultPrinter(const char* output_file)
-    : output_file_(output_file) {
-  if (output_file_.c_str() == NULL || output_file_.empty()) {
-    fprintf(stderr, "XML output file may not be null\n");
-    fflush(stderr);
-    exit(EXIT_FAILURE);
-  }
-}
-
-// Called after the unit test ends.
-void XmlUnitTestResultPrinter::OnTestIterationEnd(const UnitTest& unit_test,
-                                                  int /*iteration*/) {
-  FILE* xmlout = NULL;
-  FilePath output_file(output_file_);
-  FilePath output_dir(output_file.RemoveFileName());
-
-  if (output_dir.CreateDirectoriesRecursively()) {
-    xmlout = posix::FOpen(output_file_.c_str(), "w");
-  }
-  if (xmlout == NULL) {
-    // TODO(wan): report the reason of the failure.
-    //
-    // We don't do it for now as:
-    //
-    //   1. There is no urgent need for it.
-    //   2. It's a bit involved to make the errno variable thread-safe on
-    //      all three operating systems (Linux, Windows, and Mac OS).
-    //   3. To interpret the meaning of errno in a thread-safe way,
-    //      we need the strerror_r() function, which is not available on
-    //      Windows.
-    fprintf(stderr,
-            "Unable to open file \"%s\"\n",
-            output_file_.c_str());
-    fflush(stderr);
-    exit(EXIT_FAILURE);
-  }
-  std::stringstream stream;
-  PrintXmlUnitTest(&stream, unit_test);
-  fprintf(xmlout, "%s", StringStreamToString(&stream).c_str());
-  fclose(xmlout);
-}
-
-// Returns an XML-escaped copy of the input string str.  If is_attribute
-// is true, the text is meant to appear as an attribute value, and
-// normalizable whitespace is preserved by replacing it with character
-// references.
-//
-// Invalid XML characters in str, if any, are stripped from the output.
-// It is expected that most, if not all, of the text processed by this
-// module will consist of ordinary English text.
-// If this module is ever modified to produce version 1.1 XML output,
-// most invalid characters can be retained using character references.
-// TODO(wan): It might be nice to have a minimally invasive, human-readable
-// escaping scheme for invalid characters, rather than dropping them.
-std::string XmlUnitTestResultPrinter::EscapeXml(
-    const std::string& str, bool is_attribute) {
-  Message m;
-
-  for (size_t i = 0; i < str.size(); ++i) {
-    const char ch = str[i];
-    switch (ch) {
-      case '<':
-        m << "&lt;";
-        break;
-      case '>':
-        m << "&gt;";
-        break;
-      case '&':
-        m << "&amp;";
-        break;
-      case '\'':
-        if (is_attribute)
-          m << "&apos;";
-        else
-          m << '\'';
-        break;
-      case '"':
-        if (is_attribute)
-          m << "&quot;";
-        else
-          m << '"';
-        break;
-      default:
-        if (IsValidXmlCharacter(ch)) {
-          if (is_attribute && IsNormalizableWhitespace(ch))
-            m << "&#x" << String::FormatByte(static_cast<unsigned char>(ch))
-              << ";";
-          else
-            m << ch;
-        }
-        break;
-    }
-  }
-
-  return m.GetString();
-}
-
-// Returns the given string with all characters invalid in XML removed.
-// Currently invalid characters are dropped from the string. An
-// alternative is to replace them with certain characters such as . or ?.
-std::string XmlUnitTestResultPrinter::RemoveInvalidXmlCharacters(
-    const std::string& str) {
-  std::string output;
-  output.reserve(str.size());
-  for (std::string::const_iterator it = str.begin(); it != str.end(); ++it)
-    if (IsValidXmlCharacter(*it))
-      output.push_back(*it);
-
-  return output;
-}
-
-// The following routines generate an XML representation of a UnitTest
-// object.
-//
-// This is how Google Test concepts map to the DTD:
-//
-// <testsuites name="AllTests">        <-- corresponds to a UnitTest object
-//   <testsuite name="testcase-name">  <-- corresponds to a TestCase object
-//     <testcase name="test-name">     <-- corresponds to a TestInfo object
-//       <failure message="...">...</failure>
-//       <failure message="...">...</failure>
-//       <failure message="...">...</failure>
-//                                     <-- individual assertion failures
-//     </testcase>
-//   </testsuite>
-// </testsuites>
-
-// Formats the given time in milliseconds as seconds.
-std::string FormatTimeInMillisAsSeconds(TimeInMillis ms) {
-  ::std::stringstream ss;
-  ss << (static_cast<double>(ms) * 1e-3);
-  return ss.str();
-}
-
-static bool PortableLocaltime(time_t seconds, struct tm* out) {
-#if defined(_MSC_VER)
-  return localtime_s(out, &seconds) == 0;
-#elif defined(__MINGW32__) || defined(__MINGW64__)
-  // MINGW <time.h> provides neither localtime_r nor localtime_s, but uses
-  // Windows' localtime(), which has a thread-local tm buffer.
-  struct tm* tm_ptr = localtime(&seconds);  // NOLINT
-  if (tm_ptr == NULL)
-    return false;
-  *out = *tm_ptr;
-  return true;
-#else
-  return localtime_r(&seconds, out) != NULL;
-#endif
-}
-
-// Converts the given epoch time in milliseconds to a date string in the ISO
-// 8601 format, without the timezone information.
-std::string FormatEpochTimeInMillisAsIso8601(TimeInMillis ms) {
-  struct tm time_struct;
-  if (!PortableLocaltime(static_cast<time_t>(ms / 1000), &time_struct))
-    return "";
-  // YYYY-MM-DDThh:mm:ss
-  return StreamableToString(time_struct.tm_year + 1900) + "-" +
-      String::FormatIntWidth2(time_struct.tm_mon + 1) + "-" +
-      String::FormatIntWidth2(time_struct.tm_mday) + "T" +
-      String::FormatIntWidth2(time_struct.tm_hour) + ":" +
-      String::FormatIntWidth2(time_struct.tm_min) + ":" +
-      String::FormatIntWidth2(time_struct.tm_sec);
-}
-
-// Streams an XML CDATA section, escaping invalid CDATA sequences as needed.
-void XmlUnitTestResultPrinter::OutputXmlCDataSection(::std::ostream* stream,
-                                                     const char* data) {
-  const char* segment = data;
-  *stream << "<![CDATA[";
-  for (;;) {
-    const char* const next_segment = strstr(segment, "]]>");
-    if (next_segment != NULL) {
-      stream->write(
-          segment, static_cast<std::streamsize>(next_segment - segment));
-      *stream << "]]>]]&gt;<![CDATA[";
-      segment = next_segment + strlen("]]>");
-    } else {
-      *stream << segment;
-      break;
-    }
-  }
-  *stream << "]]>";
-}
-
-void XmlUnitTestResultPrinter::OutputXmlAttribute(
-    std::ostream* stream,
-    const std::string& element_name,
-    const std::string& name,
-    const std::string& value) {
-  const std::vector<std::string>& allowed_names =
-      GetReservedAttributesForElement(element_name);
-
-  GTEST_CHECK_(std::find(allowed_names.begin(), allowed_names.end(), name) !=
-                   allowed_names.end())
-      << "Attribute " << name << " is not allowed for element <" << element_name
-      << ">.";
-
-  *stream << " " << name << "=\"" << EscapeXmlAttribute(value) << "\"";
-}
-
-// Prints an XML representation of a TestInfo object.
-// TODO(wan): There is also value in printing properties with the plain printer.
-void XmlUnitTestResultPrinter::OutputXmlTestInfo(::std::ostream* stream,
-                                                 const char* test_case_name,
-                                                 const TestInfo& test_info) {
-  const TestResult& result = *test_info.result();
-  const std::string kTestcase = "testcase";
-
-  *stream << "    <testcase";
-  OutputXmlAttribute(stream, kTestcase, "name", test_info.name());
-
-  if (test_info.value_param() != NULL) {
-    OutputXmlAttribute(stream, kTestcase, "value_param",
-                       test_info.value_param());
-  }
-  if (test_info.type_param() != NULL) {
-    OutputXmlAttribute(stream, kTestcase, "type_param", test_info.type_param());
-  }
-
-  OutputXmlAttribute(stream, kTestcase, "status",
-                     test_info.should_run() ? "run" : "notrun");
-  OutputXmlAttribute(stream, kTestcase, "time",
-                     FormatTimeInMillisAsSeconds(result.elapsed_time()));
-  OutputXmlAttribute(stream, kTestcase, "classname", test_case_name);
-  *stream << TestPropertiesAsXmlAttributes(result);
-
-  int failures = 0;
-  for (int i = 0; i < result.total_part_count(); ++i) {
-    const TestPartResult& part = result.GetTestPartResult(i);
-    if (part.failed()) {
-      if (++failures == 1) {
-        *stream << ">\n";
-      }
-      const string location = internal::FormatCompilerIndependentFileLocation(
-          part.file_name(), part.line_number());
-      const string summary = location + "\n" + part.summary();
-      *stream << "      <failure message=\""
-              << EscapeXmlAttribute(summary.c_str())
-              << "\" type=\"\">";
-      const string detail = location + "\n" + part.message();
-      OutputXmlCDataSection(stream, RemoveInvalidXmlCharacters(detail).c_str());
-      *stream << "</failure>\n";
-    }
-  }
-
-  if (failures == 0)
-    *stream << " />\n";
-  else
-    *stream << "    </testcase>\n";
-}
-
-// Prints an XML representation of a TestCase object
-void XmlUnitTestResultPrinter::PrintXmlTestCase(std::ostream* stream,
-                                                const TestCase& test_case) {
-  const std::string kTestsuite = "testsuite";
-  *stream << "  <" << kTestsuite;
-  OutputXmlAttribute(stream, kTestsuite, "name", test_case.name());
-  OutputXmlAttribute(stream, kTestsuite, "tests",
-                     StreamableToString(test_case.reportable_test_count()));
-  OutputXmlAttribute(stream, kTestsuite, "failures",
-                     StreamableToString(test_case.failed_test_count()));
-  OutputXmlAttribute(
-      stream, kTestsuite, "disabled",
-      StreamableToString(test_case.reportable_disabled_test_count()));
-  OutputXmlAttribute(stream, kTestsuite, "errors", "0");
-  OutputXmlAttribute(stream, kTestsuite, "time",
-                     FormatTimeInMillisAsSeconds(test_case.elapsed_time()));
-  *stream << TestPropertiesAsXmlAttributes(test_case.ad_hoc_test_result())
-          << ">\n";
-
-  for (int i = 0; i < test_case.total_test_count(); ++i) {
-    if (test_case.GetTestInfo(i)->is_reportable())
-      OutputXmlTestInfo(stream, test_case.name(), *test_case.GetTestInfo(i));
-  }
-  *stream << "  </" << kTestsuite << ">\n";
-}
-
-// Prints an XML summary of unit_test to output stream out.
-void XmlUnitTestResultPrinter::PrintXmlUnitTest(std::ostream* stream,
-                                                const UnitTest& unit_test) {
-  const std::string kTestsuites = "testsuites";
-
-  *stream << "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n";
-  *stream << "<" << kTestsuites;
-
-  OutputXmlAttribute(stream, kTestsuites, "tests",
-                     StreamableToString(unit_test.reportable_test_count()));
-  OutputXmlAttribute(stream, kTestsuites, "failures",
-                     StreamableToString(unit_test.failed_test_count()));
-  OutputXmlAttribute(
-      stream, kTestsuites, "disabled",
-      StreamableToString(unit_test.reportable_disabled_test_count()));
-  OutputXmlAttribute(stream, kTestsuites, "errors", "0");
-  OutputXmlAttribute(
-      stream, kTestsuites, "timestamp",
-      FormatEpochTimeInMillisAsIso8601(unit_test.start_timestamp()));
-  OutputXmlAttribute(stream, kTestsuites, "time",
-                     FormatTimeInMillisAsSeconds(unit_test.elapsed_time()));
-
-  if (GTEST_FLAG(shuffle)) {
-    OutputXmlAttribute(stream, kTestsuites, "random_seed",
-                       StreamableToString(unit_test.random_seed()));
-  }
-
-  *stream << TestPropertiesAsXmlAttributes(unit_test.ad_hoc_test_result());
-
-  OutputXmlAttribute(stream, kTestsuites, "name", "AllTests");
-  *stream << ">\n";
-
-  for (int i = 0; i < unit_test.total_test_case_count(); ++i) {
-    if (unit_test.GetTestCase(i)->reportable_test_count() > 0)
-      PrintXmlTestCase(stream, *unit_test.GetTestCase(i));
-  }
-  *stream << "</" << kTestsuites << ">\n";
-}
-
-// Produces a string representing the test properties in a result as space
-// delimited XML attributes based on the property key="value" pairs.
-std::string XmlUnitTestResultPrinter::TestPropertiesAsXmlAttributes(
-    const TestResult& result) {
-  Message attributes;
-  for (int i = 0; i < result.test_property_count(); ++i) {
-    const TestProperty& property = result.GetTestProperty(i);
-    attributes << " " << property.key() << "="
-        << "\"" << EscapeXmlAttribute(property.value()) << "\"";
-  }
-  return attributes.GetString();
-}
-
-// End XmlUnitTestResultPrinter
-
-#if GTEST_CAN_STREAM_RESULTS_
-
-// Checks if str contains '=', '&', '%' or '\n' characters. If yes,
-// replaces them by "%xx" where xx is their hexadecimal value. For
-// example, replaces "=" with "%3D".  This algorithm is O(strlen(str))
-// in both time and space -- important as the input str may contain an
-// arbitrarily long test failure message and stack trace.
-string StreamingListener::UrlEncode(const char* str) {
-  string result;
-  result.reserve(strlen(str) + 1);
-  for (char ch = *str; ch != '\0'; ch = *++str) {
-    switch (ch) {
-      case '%':
-      case '=':
-      case '&':
-      case '\n':
-        result.append("%" + String::FormatByte(static_cast<unsigned char>(ch)));
-        break;
-      default:
-        result.push_back(ch);
-        break;
-    }
-  }
-  return result;
-}
-
-void StreamingListener::SocketWriter::MakeConnection() {
-  GTEST_CHECK_(sockfd_ == -1)
-      << "MakeConnection() can't be called when there is already a connection.";
-
-  addrinfo hints;
-  memset(&hints, 0, sizeof(hints));
-  hints.ai_family = AF_UNSPEC;    // To allow both IPv4 and IPv6 addresses.
-  hints.ai_socktype = SOCK_STREAM;
-  addrinfo* servinfo = NULL;
-
-  // Use the getaddrinfo() to get a linked list of IP addresses for
-  // the given host name.
-  const int error_num = getaddrinfo(
-      host_name_.c_str(), port_num_.c_str(), &hints, &servinfo);
-  if (error_num != 0) {
-    GTEST_LOG_(WARNING) << "stream_result_to: getaddrinfo() failed: "
-                        << gai_strerror(error_num);
-  }
-
-  // Loop through all the results and connect to the first we can.
-  for (addrinfo* cur_addr = servinfo; sockfd_ == -1 && cur_addr != NULL;
-       cur_addr = cur_addr->ai_next) {
-    sockfd_ = socket(
-        cur_addr->ai_family, cur_addr->ai_socktype, cur_addr->ai_protocol);
-    if (sockfd_ != -1) {
-      // Connect the client socket to the server socket.
-      if (connect(sockfd_, cur_addr->ai_addr, cur_addr->ai_addrlen) == -1) {
-        close(sockfd_);
-        sockfd_ = -1;
-      }
-    }
-  }
-
-  freeaddrinfo(servinfo);  // all done with this structure
-
-  if (sockfd_ == -1) {
-    GTEST_LOG_(WARNING) << "stream_result_to: failed to connect to "
-                        << host_name_ << ":" << port_num_;
-  }
-}
-
-// End of class Streaming Listener
-#endif  // GTEST_CAN_STREAM_RESULTS__
-
-// Class ScopedTrace
-
-// Pushes the given source file location and message onto a per-thread
-// trace stack maintained by Google Test.
-ScopedTrace::ScopedTrace(const char* file, int line, const Message& message)
-    GTEST_LOCK_EXCLUDED_(&UnitTest::mutex_) {
-  TraceInfo trace;
-  trace.file = file;
-  trace.line = line;
-  trace.message = message.GetString();
-
-  UnitTest::GetInstance()->PushGTestTrace(trace);
-}
-
-// Pops the info pushed by the c'tor.
-ScopedTrace::~ScopedTrace()
-    GTEST_LOCK_EXCLUDED_(&UnitTest::mutex_) {
-  UnitTest::GetInstance()->PopGTestTrace();
-}
-
-
-// class OsStackTraceGetter
-
-const char* const OsStackTraceGetterInterface::kElidedFramesMarker =
-    "... " GTEST_NAME_ " internal frames ...";
-
-string OsStackTraceGetter::CurrentStackTrace(int /*max_depth*/,
-                                             int /*skip_count*/) {
-  return "";
-}
-
-void OsStackTraceGetter::UponLeavingGTest() {}
-
-// A helper class that creates the premature-exit file in its
-// constructor and deletes the file in its destructor.
-class ScopedPrematureExitFile {
- public:
-  explicit ScopedPrematureExitFile(const char* premature_exit_filepath)
-      : premature_exit_filepath_(premature_exit_filepath) {
-    // If a path to the premature-exit file is specified...
-    if (premature_exit_filepath != NULL && *premature_exit_filepath != '\0') {
-      // create the file with a single "0" character in it.  I/O
-      // errors are ignored as there's nothing better we can do and we
-      // don't want to fail the test because of this.
-      FILE* pfile = posix::FOpen(premature_exit_filepath, "w");
-      fwrite("0", 1, 1, pfile);
-      fclose(pfile);
-    }
-  }
-
-  ~ScopedPrematureExitFile() {
-    if (premature_exit_filepath_ != NULL && *premature_exit_filepath_ != '\0') {
-      remove(premature_exit_filepath_);
-    }
-  }
-
- private:
-  const char* const premature_exit_filepath_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ScopedPrematureExitFile);
-};
-
-}  // namespace internal
-
-// class TestEventListeners
-
-TestEventListeners::TestEventListeners()
-    : repeater_(new internal::TestEventRepeater()),
-      default_result_printer_(NULL),
-      default_xml_generator_(NULL) {
-}
-
-TestEventListeners::~TestEventListeners() { delete repeater_; }
-
-// Returns the standard listener responsible for the default console
-// output.  Can be removed from the listeners list to shut down default
-// console output.  Note that removing this object from the listener list
-// with Release transfers its ownership to the user.
-void TestEventListeners::Append(TestEventListener* listener) {
-  repeater_->Append(listener);
-}
-
-// Removes the given event listener from the list and returns it.  It then
-// becomes the caller's responsibility to delete the listener. Returns
-// NULL if the listener is not found in the list.
-TestEventListener* TestEventListeners::Release(TestEventListener* listener) {
-  if (listener == default_result_printer_)
-    default_result_printer_ = NULL;
-  else if (listener == default_xml_generator_)
-    default_xml_generator_ = NULL;
-  return repeater_->Release(listener);
-}
-
-// Returns repeater that broadcasts the TestEventListener events to all
-// subscribers.
-TestEventListener* TestEventListeners::repeater() { return repeater_; }
-
-// Sets the default_result_printer attribute to the provided listener.
-// The listener is also added to the listener list and previous
-// default_result_printer is removed from it and deleted. The listener can
-// also be NULL in which case it will not be added to the list. Does
-// nothing if the previous and the current listener objects are the same.
-void TestEventListeners::SetDefaultResultPrinter(TestEventListener* listener) {
-  if (default_result_printer_ != listener) {
-    // It is an error to pass this method a listener that is already in the
-    // list.
-    delete Release(default_result_printer_);
-    default_result_printer_ = listener;
-    if (listener != NULL)
-      Append(listener);
-  }
-}
-
-// Sets the default_xml_generator attribute to the provided listener.  The
-// listener is also added to the listener list and previous
-// default_xml_generator is removed from it and deleted. The listener can
-// also be NULL in which case it will not be added to the list. Does
-// nothing if the previous and the current listener objects are the same.
-void TestEventListeners::SetDefaultXmlGenerator(TestEventListener* listener) {
-  if (default_xml_generator_ != listener) {
-    // It is an error to pass this method a listener that is already in the
-    // list.
-    delete Release(default_xml_generator_);
-    default_xml_generator_ = listener;
-    if (listener != NULL)
-      Append(listener);
-  }
-}
-
-// Controls whether events will be forwarded by the repeater to the
-// listeners in the list.
-bool TestEventListeners::EventForwardingEnabled() const {
-  return repeater_->forwarding_enabled();
-}
-
-void TestEventListeners::SuppressEventForwarding() {
-  repeater_->set_forwarding_enabled(false);
-}
-
-// class UnitTest
-
-// Gets the singleton UnitTest object.  The first time this method is
-// called, a UnitTest object is constructed and returned.  Consecutive
-// calls will return the same object.
-//
-// We don't protect this under mutex_ as a user is not supposed to
-// call this before main() starts, from which point on the return
-// value will never change.
-UnitTest* UnitTest::GetInstance() {
-  // When compiled with MSVC 7.1 in optimized mode, destroying the
-  // UnitTest object upon exiting the program messes up the exit code,
-  // causing successful tests to appear failed.  We have to use a
-  // different implementation in this case to bypass the compiler bug.
-  // This implementation makes the compiler happy, at the cost of
-  // leaking the UnitTest object.
-
-  // CodeGear C++Builder insists on a public destructor for the
-  // default implementation.  Use this implementation to keep good OO
-  // design with private destructor.
-
-#if (_MSC_VER == 1310 && !defined(_DEBUG)) || defined(__BORLANDC__)
-  static UnitTest* const instance = new UnitTest;
-  return instance;
-#else
-  static UnitTest instance;
-  return &instance;
-#endif  // (_MSC_VER == 1310 && !defined(_DEBUG)) || defined(__BORLANDC__)
-}
-
-// Gets the number of successful test cases.
-int UnitTest::successful_test_case_count() const {
-  return impl()->successful_test_case_count();
-}
-
-// Gets the number of failed test cases.
-int UnitTest::failed_test_case_count() const {
-  return impl()->failed_test_case_count();
-}
-
-// Gets the number of all test cases.
-int UnitTest::total_test_case_count() const {
-  return impl()->total_test_case_count();
-}
-
-// Gets the number of all test cases that contain at least one test
-// that should run.
-int UnitTest::test_case_to_run_count() const {
-  return impl()->test_case_to_run_count();
-}
-
-// Gets the number of successful tests.
-int UnitTest::successful_test_count() const {
-  return impl()->successful_test_count();
-}
-
-// Gets the number of failed tests.
-int UnitTest::failed_test_count() const { return impl()->failed_test_count(); }
-
-// Gets the number of disabled tests that will be reported in the XML report.
-int UnitTest::reportable_disabled_test_count() const {
-  return impl()->reportable_disabled_test_count();
-}
-
-// Gets the number of disabled tests.
-int UnitTest::disabled_test_count() const {
-  return impl()->disabled_test_count();
-}
-
-// Gets the number of tests to be printed in the XML report.
-int UnitTest::reportable_test_count() const {
-  return impl()->reportable_test_count();
-}
-
-// Gets the number of all tests.
-int UnitTest::total_test_count() const { return impl()->total_test_count(); }
-
-// Gets the number of tests that should run.
-int UnitTest::test_to_run_count() const { return impl()->test_to_run_count(); }
-
-// Gets the time of the test program start, in ms from the start of the
-// UNIX epoch.
-internal::TimeInMillis UnitTest::start_timestamp() const {
-    return impl()->start_timestamp();
-}
-
-// Gets the elapsed time, in milliseconds.
-internal::TimeInMillis UnitTest::elapsed_time() const {
-  return impl()->elapsed_time();
-}
-
-// Returns true iff the unit test passed (i.e. all test cases passed).
-bool UnitTest::Passed() const { return impl()->Passed(); }
-
-// Returns true iff the unit test failed (i.e. some test case failed
-// or something outside of all tests failed).
-bool UnitTest::Failed() const { return impl()->Failed(); }
-
-// Gets the i-th test case among all the test cases. i can range from 0 to
-// total_test_case_count() - 1. If i is not in that range, returns NULL.
-const TestCase* UnitTest::GetTestCase(int i) const {
-  return impl()->GetTestCase(i);
-}
-
-// Returns the TestResult containing information on test failures and
-// properties logged outside of individual test cases.
-const TestResult& UnitTest::ad_hoc_test_result() const {
-  return *impl()->ad_hoc_test_result();
-}
-
-// Gets the i-th test case among all the test cases. i can range from 0 to
-// total_test_case_count() - 1. If i is not in that range, returns NULL.
-TestCase* UnitTest::GetMutableTestCase(int i) {
-  return impl()->GetMutableTestCase(i);
-}
-
-// Returns the list of event listeners that can be used to track events
-// inside Google Test.
-TestEventListeners& UnitTest::listeners() {
-  return *impl()->listeners();
-}
-
-// Registers and returns a global test environment.  When a test
-// program is run, all global test environments will be set-up in the
-// order they were registered.  After all tests in the program have
-// finished, all global test environments will be torn-down in the
-// *reverse* order they were registered.
-//
-// The UnitTest object takes ownership of the given environment.
-//
-// We don't protect this under mutex_, as we only support calling it
-// from the main thread.
-Environment* UnitTest::AddEnvironment(Environment* env) {
-  if (env == NULL) {
-    return NULL;
-  }
-
-  impl_->environments().push_back(env);
-  return env;
-}
-
-// Adds a TestPartResult to the current TestResult object.  All Google Test
-// assertion macros (e.g. ASSERT_TRUE, EXPECT_EQ, etc) eventually call
-// this to report their results.  The user code should use the
-// assertion macros instead of calling this directly.
-void UnitTest::AddTestPartResult(
-    TestPartResult::Type result_type,
-    const char* file_name,
-    int line_number,
-    const std::string& message,
-    const std::string& os_stack_trace) GTEST_LOCK_EXCLUDED_(mutex_) {
-  Message msg;
-  msg << message;
-
-  internal::MutexLock lock(&mutex_);
-  if (impl_->gtest_trace_stack().size() > 0) {
-    msg << "\n" << GTEST_NAME_ << " trace:";
-
-    for (int i = static_cast<int>(impl_->gtest_trace_stack().size());
-         i > 0; --i) {
-      const internal::TraceInfo& trace = impl_->gtest_trace_stack()[i - 1];
-      msg << "\n" << internal::FormatFileLocation(trace.file, trace.line)
-          << " " << trace.message;
-    }
-  }
-
-  if (os_stack_trace.c_str() != NULL && !os_stack_trace.empty()) {
-    msg << internal::kStackTraceMarker << os_stack_trace;
-  }
-
-  const TestPartResult result =
-    TestPartResult(result_type, file_name, line_number,
-                   msg.GetString().c_str());
-  impl_->GetTestPartResultReporterForCurrentThread()->
-      ReportTestPartResult(result);
-
-  if (result_type != TestPartResult::kSuccess) {
-    // gtest_break_on_failure takes precedence over
-    // gtest_throw_on_failure.  This allows a user to set the latter
-    // in the code (perhaps in order to use Google Test assertions
-    // with another testing framework) and specify the former on the
-    // command line for debugging.
-    if (GTEST_FLAG(break_on_failure)) {
-#if GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT
-      // Using DebugBreak on Windows allows gtest to still break into a debugger
-      // when a failure happens and both the --gtest_break_on_failure and
-      // the --gtest_catch_exceptions flags are specified.
-      DebugBreak();
-#else
-      // Dereference NULL through a volatile pointer to prevent the compiler
-      // from removing. We use this rather than abort() or __builtin_trap() for
-      // portability: Symbian doesn't implement abort() well, and some debuggers
-      // don't correctly trap abort().
-      *static_cast<volatile int*>(NULL) = 1;
-#endif  // GTEST_OS_WINDOWS
-    } else if (GTEST_FLAG(throw_on_failure)) {
-#if GTEST_HAS_EXCEPTIONS
-      throw internal::GoogleTestFailureException(result);
-#else
-      // We cannot call abort() as it generates a pop-up in debug mode
-      // that cannot be suppressed in VC 7.1 or below.
-      exit(1);
-#endif
-    }
-  }
-}
-
-// Adds a TestProperty to the current TestResult object when invoked from
-// inside a test, to current TestCase's ad_hoc_test_result_ when invoked
-// from SetUpTestCase or TearDownTestCase, or to the global property set
-// when invoked elsewhere.  If the result already contains a property with
-// the same key, the value will be updated.
-void UnitTest::RecordProperty(const std::string& key,
-                              const std::string& value) {
-  impl_->RecordProperty(TestProperty(key, value));
-}
-
-// Runs all tests in this UnitTest object and prints the result.
-// Returns 0 if successful, or 1 otherwise.
-//
-// We don't protect this under mutex_, as we only support calling it
-// from the main thread.
-int UnitTest::Run() {
-  const bool in_death_test_child_process =
-      internal::GTEST_FLAG(internal_run_death_test).length() > 0;
-
-  // Google Test implements this protocol for catching that a test
-  // program exits before returning control to Google Test:
-  //
-  //   1. Upon start, Google Test creates a file whose absolute path
-  //      is specified by the environment variable
-  //      TEST_PREMATURE_EXIT_FILE.
-  //   2. When Google Test has finished its work, it deletes the file.
-  //
-  // This allows a test runner to set TEST_PREMATURE_EXIT_FILE before
-  // running a Google-Test-based test program and check the existence
-  // of the file at the end of the test execution to see if it has
-  // exited prematurely.
-
-  // If we are in the child process of a death test, don't
-  // create/delete the premature exit file, as doing so is unnecessary
-  // and will confuse the parent process.  Otherwise, create/delete
-  // the file upon entering/leaving this function.  If the program
-  // somehow exits before this function has a chance to return, the
-  // premature-exit file will be left undeleted, causing a test runner
-  // that understands the premature-exit-file protocol to report the
-  // test as having failed.
-  const internal::ScopedPrematureExitFile premature_exit_file(
-      in_death_test_child_process ?
-      NULL : internal::posix::GetEnv("TEST_PREMATURE_EXIT_FILE"));
-
-  // Captures the value of GTEST_FLAG(catch_exceptions).  This value will be
-  // used for the duration of the program.
-  impl()->set_catch_exceptions(GTEST_FLAG(catch_exceptions));
-
-#if GTEST_HAS_SEH
-  // Either the user wants Google Test to catch exceptions thrown by the
-  // tests or this is executing in the context of death test child
-  // process. In either case the user does not want to see pop-up dialogs
-  // about crashes - they are expected.
-  if (impl()->catch_exceptions() || in_death_test_child_process) {
-# if !GTEST_OS_WINDOWS_MOBILE && !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT
-    // SetErrorMode doesn't exist on CE.
-    SetErrorMode(SEM_FAILCRITICALERRORS | SEM_NOALIGNMENTFAULTEXCEPT |
-                 SEM_NOGPFAULTERRORBOX | SEM_NOOPENFILEERRORBOX);
-# endif  // !GTEST_OS_WINDOWS_MOBILE
-
-# if (defined(_MSC_VER) || GTEST_OS_WINDOWS_MINGW) && !GTEST_OS_WINDOWS_MOBILE
-    // Death test children can be terminated with _abort().  On Windows,
-    // _abort() can show a dialog with a warning message.  This forces the
-    // abort message to go to stderr instead.
-    _set_error_mode(_OUT_TO_STDERR);
-# endif
-
-# if _MSC_VER >= 1400 && !GTEST_OS_WINDOWS_MOBILE
-    // In the debug version, Visual Studio pops up a separate dialog
-    // offering a choice to debug the aborted program. We need to suppress
-    // this dialog or it will pop up for every EXPECT/ASSERT_DEATH statement
-    // executed. Google Test will notify the user of any unexpected
-    // failure via stderr.
-    //
-    // VC++ doesn't define _set_abort_behavior() prior to the version 8.0.
-    // Users of prior VC versions shall suffer the agony and pain of
-    // clicking through the countless debug dialogs.
-    // TODO(vladl@google.com): find a way to suppress the abort dialog() in the
-    // debug mode when compiled with VC 7.1 or lower.
-    if (!GTEST_FLAG(break_on_failure))
-      _set_abort_behavior(
-          0x0,                                    // Clear the following flags:
-          _WRITE_ABORT_MSG | _CALL_REPORTFAULT);  // pop-up window, core dump.
-# endif
-  }
-#endif  // GTEST_HAS_SEH
-
-  return internal::HandleExceptionsInMethodIfSupported(
-      impl(),
-      &internal::UnitTestImpl::RunAllTests,
-      "auxiliary test code (environments or event listeners)") ? 0 : 1;
-}
-
-// Returns the working directory when the first TEST() or TEST_F() was
-// executed.
-const char* UnitTest::original_working_dir() const {
-  return impl_->original_working_dir_.c_str();
-}
-
-// Returns the TestCase object for the test that's currently running,
-// or NULL if no test is running.
-const TestCase* UnitTest::current_test_case() const
-    GTEST_LOCK_EXCLUDED_(mutex_) {
-  internal::MutexLock lock(&mutex_);
-  return impl_->current_test_case();
-}
-
-// Returns the TestInfo object for the test that's currently running,
-// or NULL if no test is running.
-const TestInfo* UnitTest::current_test_info() const
-    GTEST_LOCK_EXCLUDED_(mutex_) {
-  internal::MutexLock lock(&mutex_);
-  return impl_->current_test_info();
-}
-
-// Returns the random seed used at the start of the current test run.
-int UnitTest::random_seed() const { return impl_->random_seed(); }
-
-#if GTEST_HAS_PARAM_TEST
-// Returns ParameterizedTestCaseRegistry object used to keep track of
-// value-parameterized tests and instantiate and register them.
-internal::ParameterizedTestCaseRegistry&
-    UnitTest::parameterized_test_registry()
-        GTEST_LOCK_EXCLUDED_(mutex_) {
-  return impl_->parameterized_test_registry();
-}
-#endif  // GTEST_HAS_PARAM_TEST
-
-// Creates an empty UnitTest.
-UnitTest::UnitTest() {
-  impl_ = new internal::UnitTestImpl(this);
-}
-
-// Destructor of UnitTest.
-UnitTest::~UnitTest() {
-  delete impl_;
-}
-
-// Pushes a trace defined by SCOPED_TRACE() on to the per-thread
-// Google Test trace stack.
-void UnitTest::PushGTestTrace(const internal::TraceInfo& trace)
-    GTEST_LOCK_EXCLUDED_(mutex_) {
-  internal::MutexLock lock(&mutex_);
-  impl_->gtest_trace_stack().push_back(trace);
-}
-
-// Pops a trace from the per-thread Google Test trace stack.
-void UnitTest::PopGTestTrace()
-    GTEST_LOCK_EXCLUDED_(mutex_) {
-  internal::MutexLock lock(&mutex_);
-  impl_->gtest_trace_stack().pop_back();
-}
-
-namespace internal {
-
-UnitTestImpl::UnitTestImpl(UnitTest* parent)
-    : parent_(parent),
-      GTEST_DISABLE_MSC_WARNINGS_PUSH_(4355 /* using this in initializer */)
-      default_global_test_part_result_reporter_(this),
-      default_per_thread_test_part_result_reporter_(this),
-      GTEST_DISABLE_MSC_WARNINGS_POP_()
-      global_test_part_result_repoter_(
-          &default_global_test_part_result_reporter_),
-      per_thread_test_part_result_reporter_(
-          &default_per_thread_test_part_result_reporter_),
-#if GTEST_HAS_PARAM_TEST
-      parameterized_test_registry_(),
-      parameterized_tests_registered_(false),
-#endif  // GTEST_HAS_PARAM_TEST
-      last_death_test_case_(-1),
-      current_test_case_(NULL),
-      current_test_info_(NULL),
-      ad_hoc_test_result_(),
-      os_stack_trace_getter_(NULL),
-      post_flag_parse_init_performed_(false),
-      random_seed_(0),  // Will be overridden by the flag before first use.
-      random_(0),  // Will be reseeded before first use.
-      start_timestamp_(0),
-      elapsed_time_(0),
-#if GTEST_HAS_DEATH_TEST
-      death_test_factory_(new DefaultDeathTestFactory),
-#endif
-      // Will be overridden by the flag before first use.
-      catch_exceptions_(false) {
-  listeners()->SetDefaultResultPrinter(new PrettyUnitTestResultPrinter);
-}
-
-UnitTestImpl::~UnitTestImpl() {
-  // Deletes every TestCase.
-  ForEach(test_cases_, internal::Delete<TestCase>);
-
-  // Deletes every Environment.
-  ForEach(environments_, internal::Delete<Environment>);
-
-  delete os_stack_trace_getter_;
-}
-
-// Adds a TestProperty to the current TestResult object when invoked in a
-// context of a test, to current test case's ad_hoc_test_result when invoke
-// from SetUpTestCase/TearDownTestCase, or to the global property set
-// otherwise.  If the result already contains a property with the same key,
-// the value will be updated.
-void UnitTestImpl::RecordProperty(const TestProperty& test_property) {
-  std::string xml_element;
-  TestResult* test_result;  // TestResult appropriate for property recording.
-
-  if (current_test_info_ != NULL) {
-    xml_element = "testcase";
-    test_result = &(current_test_info_->result_);
-  } else if (current_test_case_ != NULL) {
-    xml_element = "testsuite";
-    test_result = &(current_test_case_->ad_hoc_test_result_);
-  } else {
-    xml_element = "testsuites";
-    test_result = &ad_hoc_test_result_;
-  }
-  test_result->RecordProperty(xml_element, test_property);
-}
-
-#if GTEST_HAS_DEATH_TEST
-// Disables event forwarding if the control is currently in a death test
-// subprocess. Must not be called before InitGoogleTest.
-void UnitTestImpl::SuppressTestEventsIfInSubprocess() {
-  if (internal_run_death_test_flag_.get() != NULL)
-    listeners()->SuppressEventForwarding();
-}
-#endif  // GTEST_HAS_DEATH_TEST
-
-// Initializes event listeners performing XML output as specified by
-// UnitTestOptions. Must not be called before InitGoogleTest.
-void UnitTestImpl::ConfigureXmlOutput() {
-  const std::string& output_format = UnitTestOptions::GetOutputFormat();
-  if (output_format == "xml") {
-    listeners()->SetDefaultXmlGenerator(new XmlUnitTestResultPrinter(
-        UnitTestOptions::GetAbsolutePathToOutputFile().c_str()));
-  } else if (output_format != "") {
-    printf("WARNING: unrecognized output format \"%s\" ignored.\n",
-           output_format.c_str());
-    fflush(stdout);
-  }
-}
-
-#if GTEST_CAN_STREAM_RESULTS_
-// Initializes event listeners for streaming test results in string form.
-// Must not be called before InitGoogleTest.
-void UnitTestImpl::ConfigureStreamingOutput() {
-  const std::string& target = GTEST_FLAG(stream_result_to);
-  if (!target.empty()) {
-    const size_t pos = target.find(':');
-    if (pos != std::string::npos) {
-      listeners()->Append(new StreamingListener(target.substr(0, pos),
-                                                target.substr(pos+1)));
-    } else {
-      printf("WARNING: unrecognized streaming target \"%s\" ignored.\n",
-             target.c_str());
-      fflush(stdout);
-    }
-  }
-}
-#endif  // GTEST_CAN_STREAM_RESULTS_
-
-// Performs initialization dependent upon flag values obtained in
-// ParseGoogleTestFlagsOnly.  Is called from InitGoogleTest after the call to
-// ParseGoogleTestFlagsOnly.  In case a user neglects to call InitGoogleTest
-// this function is also called from RunAllTests.  Since this function can be
-// called more than once, it has to be idempotent.
-void UnitTestImpl::PostFlagParsingInit() {
-  // Ensures that this function does not execute more than once.
-  if (!post_flag_parse_init_performed_) {
-    post_flag_parse_init_performed_ = true;
-
-#if defined(GTEST_CUSTOM_TEST_EVENT_LISTENER_)
-    // Register to send notifications about key process state changes.
-    listeners()->Append(new GTEST_CUSTOM_TEST_EVENT_LISTENER_());
-#endif  // defined(GTEST_CUSTOM_TEST_EVENT_LISTENER_)
-
-#if GTEST_HAS_DEATH_TEST
-    InitDeathTestSubprocessControlInfo();
-    SuppressTestEventsIfInSubprocess();
-#endif  // GTEST_HAS_DEATH_TEST
-
-    // Registers parameterized tests. This makes parameterized tests
-    // available to the UnitTest reflection API without running
-    // RUN_ALL_TESTS.
-    RegisterParameterizedTests();
-
-    // Configures listeners for XML output. This makes it possible for users
-    // to shut down the default XML output before invoking RUN_ALL_TESTS.
-    ConfigureXmlOutput();
-
-#if GTEST_CAN_STREAM_RESULTS_
-    // Configures listeners for streaming test results to the specified server.
-    ConfigureStreamingOutput();
-#endif  // GTEST_CAN_STREAM_RESULTS_
-  }
-}
-
-// A predicate that checks the name of a TestCase against a known
-// value.
-//
-// This is used for implementation of the UnitTest class only.  We put
-// it in the anonymous namespace to prevent polluting the outer
-// namespace.
-//
-// TestCaseNameIs is copyable.
-class TestCaseNameIs {
- public:
-  // Constructor.
-  explicit TestCaseNameIs(const std::string& name)
-      : name_(name) {}
-
-  // Returns true iff the name of test_case matches name_.
-  bool operator()(const TestCase* test_case) const {
-    return test_case != NULL && strcmp(test_case->name(), name_.c_str()) == 0;
-  }
-
- private:
-  std::string name_;
-};
-
-// Finds and returns a TestCase with the given name.  If one doesn't
-// exist, creates one and returns it.  It's the CALLER'S
-// RESPONSIBILITY to ensure that this function is only called WHEN THE
-// TESTS ARE NOT SHUFFLED.
-//
-// Arguments:
-//
-//   test_case_name: name of the test case
-//   type_param:     the name of the test case's type parameter, or NULL if
-//                   this is not a typed or a type-parameterized test case.
-//   set_up_tc:      pointer to the function that sets up the test case
-//   tear_down_tc:   pointer to the function that tears down the test case
-TestCase* UnitTestImpl::GetTestCase(const char* test_case_name,
-                                    const char* type_param,
-                                    Test::SetUpTestCaseFunc set_up_tc,
-                                    Test::TearDownTestCaseFunc tear_down_tc) {
-  // Can we find a TestCase with the given name?
-  const std::vector<TestCase*>::const_iterator test_case =
-      std::find_if(test_cases_.begin(), test_cases_.end(),
-                   TestCaseNameIs(test_case_name));
-
-  if (test_case != test_cases_.end())
-    return *test_case;
-
-  // No.  Let's create one.
-  TestCase* const new_test_case =
-      new TestCase(test_case_name, type_param, set_up_tc, tear_down_tc);
-
-  // Is this a death test case?
-  if (internal::UnitTestOptions::MatchesFilter(test_case_name,
-                                               kDeathTestCaseFilter)) {
-    // Yes.  Inserts the test case after the last death test case
-    // defined so far.  This only works when the test cases haven't
-    // been shuffled.  Otherwise we may end up running a death test
-    // after a non-death test.
-    ++last_death_test_case_;
-    test_cases_.insert(test_cases_.begin() + last_death_test_case_,
-                       new_test_case);
-  } else {
-    // No.  Appends to the end of the list.
-    test_cases_.push_back(new_test_case);
-  }
-
-  test_case_indices_.push_back(static_cast<int>(test_case_indices_.size()));
-  return new_test_case;
-}
-
-// Helpers for setting up / tearing down the given environment.  They
-// are for use in the ForEach() function.
-static void SetUpEnvironment(Environment* env) { env->SetUp(); }
-static void TearDownEnvironment(Environment* env) { env->TearDown(); }
-
-// Runs all tests in this UnitTest object, prints the result, and
-// returns true if all tests are successful.  If any exception is
-// thrown during a test, the test is considered to be failed, but the
-// rest of the tests will still be run.
-//
-// When parameterized tests are enabled, it expands and registers
-// parameterized tests first in RegisterParameterizedTests().
-// All other functions called from RunAllTests() may safely assume that
-// parameterized tests are ready to be counted and run.
-bool UnitTestImpl::RunAllTests() {
-  // Makes sure InitGoogleTest() was called.
-  if (!GTestIsInitialized()) {
-    printf("%s",
-           "\nThis test program did NOT call ::testing::InitGoogleTest "
-           "before calling RUN_ALL_TESTS().  Please fix it.\n");
-    return false;
-  }
-
-  // Do not run any test if the --help flag was specified.
-  if (g_help_flag)
-    return true;
-
-  // Repeats the call to the post-flag parsing initialization in case the
-  // user didn't call InitGoogleTest.
-  PostFlagParsingInit();
-
-  // Even if sharding is not on, test runners may want to use the
-  // GTEST_SHARD_STATUS_FILE to query whether the test supports the sharding
-  // protocol.
-  internal::WriteToShardStatusFileIfNeeded();
-
-  // True iff we are in a subprocess for running a thread-safe-style
-  // death test.
-  bool in_subprocess_for_death_test = false;
-
-#if GTEST_HAS_DEATH_TEST
-  in_subprocess_for_death_test = (internal_run_death_test_flag_.get() != NULL);
-# if defined(GTEST_EXTRA_DEATH_TEST_CHILD_SETUP_)
-  if (in_subprocess_for_death_test) {
-    GTEST_EXTRA_DEATH_TEST_CHILD_SETUP_();
-  }
-# endif  // defined(GTEST_EXTRA_DEATH_TEST_CHILD_SETUP_)
-#endif  // GTEST_HAS_DEATH_TEST
-
-  const bool should_shard = ShouldShard(kTestTotalShards, kTestShardIndex,
-                                        in_subprocess_for_death_test);
-
-  // Compares the full test names with the filter to decide which
-  // tests to run.
-  const bool has_tests_to_run = FilterTests(should_shard
-                                              ? HONOR_SHARDING_PROTOCOL
-                                              : IGNORE_SHARDING_PROTOCOL) > 0;
-
-  // Lists the tests and exits if the --gtest_list_tests flag was specified.
-  if (GTEST_FLAG(list_tests)) {
-    // This must be called *after* FilterTests() has been called.
-    ListTestsMatchingFilter();
-    return true;
-  }
-
-  random_seed_ = GTEST_FLAG(shuffle) ?
-      GetRandomSeedFromFlag(GTEST_FLAG(random_seed)) : 0;
-
-  // True iff at least one test has failed.
-  bool failed = false;
-
-  TestEventListener* repeater = listeners()->repeater();
-
-  start_timestamp_ = GetTimeInMillis();
-  repeater->OnTestProgramStart(*parent_);
-
-  // How many times to repeat the tests?  We don't want to repeat them
-  // when we are inside the subprocess of a death test.
-  const int repeat = in_subprocess_for_death_test ? 1 : GTEST_FLAG(repeat);
-  // Repeats forever if the repeat count is negative.
-  const bool forever = repeat < 0;
-  for (int i = 0; forever || i != repeat; i++) {
-    // We want to preserve failures generated by ad-hoc test
-    // assertions executed before RUN_ALL_TESTS().
-    ClearNonAdHocTestResult();
-
-    const TimeInMillis start = GetTimeInMillis();
-
-    // Shuffles test cases and tests if requested.
-    if (has_tests_to_run && GTEST_FLAG(shuffle)) {
-      random()->Reseed(random_seed_);
-      // This should be done before calling OnTestIterationStart(),
-      // such that a test event listener can see the actual test order
-      // in the event.
-      ShuffleTests();
-    }
-
-    // Tells the unit test event listeners that the tests are about to start.
-    repeater->OnTestIterationStart(*parent_, i);
-
-    // Runs each test case if there is at least one test to run.
-    if (has_tests_to_run) {
-      // Sets up all environments beforehand.
-      repeater->OnEnvironmentsSetUpStart(*parent_);
-      ForEach(environments_, SetUpEnvironment);
-      repeater->OnEnvironmentsSetUpEnd(*parent_);
-
-      // Runs the tests only if there was no fatal failure during global
-      // set-up.
-      if (!Test::HasFatalFailure()) {
-        for (int test_index = 0; test_index < total_test_case_count();
-             test_index++) {
-          GetMutableTestCase(test_index)->Run();
-        }
-      }
-
-      // Tears down all environments in reverse order afterwards.
-      repeater->OnEnvironmentsTearDownStart(*parent_);
-      std::for_each(environments_.rbegin(), environments_.rend(),
-                    TearDownEnvironment);
-      repeater->OnEnvironmentsTearDownEnd(*parent_);
-    }
-
-    elapsed_time_ = GetTimeInMillis() - start;
-
-    // Tells the unit test event listener that the tests have just finished.
-    repeater->OnTestIterationEnd(*parent_, i);
-
-    // Gets the result and clears it.
-    if (!Passed()) {
-      failed = true;
-    }
-
-    // Restores the original test order after the iteration.  This
-    // allows the user to quickly repro a failure that happens in the
-    // N-th iteration without repeating the first (N - 1) iterations.
-    // This is not enclosed in "if (GTEST_FLAG(shuffle)) { ... }", in
-    // case the user somehow changes the value of the flag somewhere
-    // (it's always safe to unshuffle the tests).
-    UnshuffleTests();
-
-    if (GTEST_FLAG(shuffle)) {
-      // Picks a new random seed for each iteration.
-      random_seed_ = GetNextRandomSeed(random_seed_);
-    }
-  }
-
-  repeater->OnTestProgramEnd(*parent_);
-
-  return !failed;
-}
-
-// Reads the GTEST_SHARD_STATUS_FILE environment variable, and creates the file
-// if the variable is present. If a file already exists at this location, this
-// function will write over it. If the variable is present, but the file cannot
-// be created, prints an error and exits.
-void WriteToShardStatusFileIfNeeded() {
-  const char* const test_shard_file = posix::GetEnv(kTestShardStatusFile);
-  if (test_shard_file != NULL) {
-    FILE* const file = posix::FOpen(test_shard_file, "w");
-    if (file == NULL) {
-      ColoredPrintf(COLOR_RED,
-                    "Could not write to the test shard status file \"%s\" "
-                    "specified by the %s environment variable.\n",
-                    test_shard_file, kTestShardStatusFile);
-      fflush(stdout);
-      exit(EXIT_FAILURE);
-    }
-    fclose(file);
-  }
-}
-
-// Checks whether sharding is enabled by examining the relevant
-// environment variable values. If the variables are present,
-// but inconsistent (i.e., shard_index >= total_shards), prints
-// an error and exits. If in_subprocess_for_death_test, sharding is
-// disabled because it must only be applied to the original test
-// process. Otherwise, we could filter out death tests we intended to execute.
-bool ShouldShard(const char* total_shards_env,
-                 const char* shard_index_env,
-                 bool in_subprocess_for_death_test) {
-  if (in_subprocess_for_death_test) {
-    return false;
-  }
-
-  const Int32 total_shards = Int32FromEnvOrDie(total_shards_env, -1);
-  const Int32 shard_index = Int32FromEnvOrDie(shard_index_env, -1);
-
-  if (total_shards == -1 && shard_index == -1) {
-    return false;
-  } else if (total_shards == -1 && shard_index != -1) {
-    const Message msg = Message()
-      << "Invalid environment variables: you have "
-      << kTestShardIndex << " = " << shard_index
-      << ", but have left " << kTestTotalShards << " unset.\n";
-    ColoredPrintf(COLOR_RED, msg.GetString().c_str());
-    fflush(stdout);
-    exit(EXIT_FAILURE);
-  } else if (total_shards != -1 && shard_index == -1) {
-    const Message msg = Message()
-      << "Invalid environment variables: you have "
-      << kTestTotalShards << " = " << total_shards
-      << ", but have left " << kTestShardIndex << " unset.\n";
-    ColoredPrintf(COLOR_RED, msg.GetString().c_str());
-    fflush(stdout);
-    exit(EXIT_FAILURE);
-  } else if (shard_index < 0 || shard_index >= total_shards) {
-    const Message msg = Message()
-      << "Invalid environment variables: we require 0 <= "
-      << kTestShardIndex << " < " << kTestTotalShards
-      << ", but you have " << kTestShardIndex << "=" << shard_index
-      << ", " << kTestTotalShards << "=" << total_shards << ".\n";
-    ColoredPrintf(COLOR_RED, msg.GetString().c_str());
-    fflush(stdout);
-    exit(EXIT_FAILURE);
-  }
-
-  return total_shards > 1;
-}
-
-// Parses the environment variable var as an Int32. If it is unset,
-// returns default_val. If it is not an Int32, prints an error
-// and aborts.
-Int32 Int32FromEnvOrDie(const char* var, Int32 default_val) {
-  const char* str_val = posix::GetEnv(var);
-  if (str_val == NULL) {
-    return default_val;
-  }
-
-  Int32 result;
-  if (!ParseInt32(Message() << "The value of environment variable " << var,
-                  str_val, &result)) {
-    exit(EXIT_FAILURE);
-  }
-  return result;
-}
-
-// Given the total number of shards, the shard index, and the test id,
-// returns true iff the test should be run on this shard. The test id is
-// some arbitrary but unique non-negative integer assigned to each test
-// method. Assumes that 0 <= shard_index < total_shards.
-bool ShouldRunTestOnShard(int total_shards, int shard_index, int test_id) {
-  return (test_id % total_shards) == shard_index;
-}
-
-// Compares the name of each test with the user-specified filter to
-// decide whether the test should be run, then records the result in
-// each TestCase and TestInfo object.
-// If shard_tests == true, further filters tests based on sharding
-// variables in the environment - see
-// http://code.google.com/p/googletest/wiki/GoogleTestAdvancedGuide.
-// Returns the number of tests that should run.
-int UnitTestImpl::FilterTests(ReactionToSharding shard_tests) {
-  const Int32 total_shards = shard_tests == HONOR_SHARDING_PROTOCOL ?
-      Int32FromEnvOrDie(kTestTotalShards, -1) : -1;
-  const Int32 shard_index = shard_tests == HONOR_SHARDING_PROTOCOL ?
-      Int32FromEnvOrDie(kTestShardIndex, -1) : -1;
-
-  // num_runnable_tests are the number of tests that will
-  // run across all shards (i.e., match filter and are not disabled).
-  // num_selected_tests are the number of tests to be run on
-  // this shard.
-  int num_runnable_tests = 0;
-  int num_selected_tests = 0;
-  for (size_t i = 0; i < test_cases_.size(); i++) {
-    TestCase* const test_case = test_cases_[i];
-    const std::string &test_case_name = test_case->name();
-    test_case->set_should_run(false);
-
-    for (size_t j = 0; j < test_case->test_info_list().size(); j++) {
-      TestInfo* const test_info = test_case->test_info_list()[j];
-      const std::string test_name(test_info->name());
-      // A test is disabled if test case name or test name matches
-      // kDisableTestFilter.
-      const bool is_disabled =
-          internal::UnitTestOptions::MatchesFilter(test_case_name,
-                                                   kDisableTestFilter) ||
-          internal::UnitTestOptions::MatchesFilter(test_name,
-                                                   kDisableTestFilter);
-      test_info->is_disabled_ = is_disabled;
-
-      const bool matches_filter =
-          internal::UnitTestOptions::FilterMatchesTest(test_case_name,
-                                                       test_name);
-      test_info->matches_filter_ = matches_filter;
-
-      const bool is_runnable =
-          (GTEST_FLAG(also_run_disabled_tests) || !is_disabled) &&
-          matches_filter;
-
-      const bool is_selected = is_runnable &&
-          (shard_tests == IGNORE_SHARDING_PROTOCOL ||
-           ShouldRunTestOnShard(total_shards, shard_index,
-                                num_runnable_tests));
-
-      num_runnable_tests += is_runnable;
-      num_selected_tests += is_selected;
-
-      test_info->should_run_ = is_selected;
-      test_case->set_should_run(test_case->should_run() || is_selected);
-    }
-  }
-  return num_selected_tests;
-}
-
-// Prints the given C-string on a single line by replacing all '\n'
-// characters with string "\\n".  If the output takes more than
-// max_length characters, only prints the first max_length characters
-// and "...".
-static void PrintOnOneLine(const char* str, int max_length) {
-  if (str != NULL) {
-    for (int i = 0; *str != '\0'; ++str) {
-      if (i >= max_length) {
-        printf("...");
-        break;
-      }
-      if (*str == '\n') {
-        printf("\\n");
-        i += 2;
-      } else {
-        printf("%c", *str);
-        ++i;
-      }
-    }
-  }
-}
-
-// Prints the names of the tests matching the user-specified filter flag.
-void UnitTestImpl::ListTestsMatchingFilter() {
-  // Print at most this many characters for each type/value parameter.
-  const int kMaxParamLength = 250;
-
-  for (size_t i = 0; i < test_cases_.size(); i++) {
-    const TestCase* const test_case = test_cases_[i];
-    bool printed_test_case_name = false;
-
-    for (size_t j = 0; j < test_case->test_info_list().size(); j++) {
-      const TestInfo* const test_info =
-          test_case->test_info_list()[j];
-      if (test_info->matches_filter_) {
-        if (!printed_test_case_name) {
-          printed_test_case_name = true;
-          printf("%s.", test_case->name());
-          if (test_case->type_param() != NULL) {
-            printf("  # %s = ", kTypeParamLabel);
-            // We print the type parameter on a single line to make
-            // the output easy to parse by a program.
-            PrintOnOneLine(test_case->type_param(), kMaxParamLength);
-          }
-          printf("\n");
-        }
-        printf("  %s", test_info->name());
-        if (test_info->value_param() != NULL) {
-          printf("  # %s = ", kValueParamLabel);
-          // We print the value parameter on a single line to make the
-          // output easy to parse by a program.
-          PrintOnOneLine(test_info->value_param(), kMaxParamLength);
-        }
-        printf("\n");
-      }
-    }
-  }
-  fflush(stdout);
-}
-
-// Sets the OS stack trace getter.
-//
-// Does nothing if the input and the current OS stack trace getter are
-// the same; otherwise, deletes the old getter and makes the input the
-// current getter.
-void UnitTestImpl::set_os_stack_trace_getter(
-    OsStackTraceGetterInterface* getter) {
-  if (os_stack_trace_getter_ != getter) {
-    delete os_stack_trace_getter_;
-    os_stack_trace_getter_ = getter;
-  }
-}
-
-// Returns the current OS stack trace getter if it is not NULL;
-// otherwise, creates an OsStackTraceGetter, makes it the current
-// getter, and returns it.
-OsStackTraceGetterInterface* UnitTestImpl::os_stack_trace_getter() {
-  if (os_stack_trace_getter_ == NULL) {
-#ifdef GTEST_OS_STACK_TRACE_GETTER_
-    os_stack_trace_getter_ = new GTEST_OS_STACK_TRACE_GETTER_;
-#else
-    os_stack_trace_getter_ = new OsStackTraceGetter;
-#endif  // GTEST_OS_STACK_TRACE_GETTER_
-  }
-
-  return os_stack_trace_getter_;
-}
-
-// Returns the TestResult for the test that's currently running, or
-// the TestResult for the ad hoc test if no test is running.
-TestResult* UnitTestImpl::current_test_result() {
-  return current_test_info_ ?
-      &(current_test_info_->result_) : &ad_hoc_test_result_;
-}
-
-// Shuffles all test cases, and the tests within each test case,
-// making sure that death tests are still run first.
-void UnitTestImpl::ShuffleTests() {
-  // Shuffles the death test cases.
-  ShuffleRange(random(), 0, last_death_test_case_ + 1, &test_case_indices_);
-
-  // Shuffles the non-death test cases.
-  ShuffleRange(random(), last_death_test_case_ + 1,
-               static_cast<int>(test_cases_.size()), &test_case_indices_);
-
-  // Shuffles the tests inside each test case.
-  for (size_t i = 0; i < test_cases_.size(); i++) {
-    test_cases_[i]->ShuffleTests(random());
-  }
-}
-
-// Restores the test cases and tests to their order before the first shuffle.
-void UnitTestImpl::UnshuffleTests() {
-  for (size_t i = 0; i < test_cases_.size(); i++) {
-    // Unshuffles the tests in each test case.
-    test_cases_[i]->UnshuffleTests();
-    // Resets the index of each test case.
-    test_case_indices_[i] = static_cast<int>(i);
-  }
-}
-
-// Returns the current OS stack trace as an std::string.
-//
-// The maximum number of stack frames to be included is specified by
-// the gtest_stack_trace_depth flag.  The skip_count parameter
-// specifies the number of top frames to be skipped, which doesn't
-// count against the number of frames to be included.
-//
-// For example, if Foo() calls Bar(), which in turn calls
-// GetCurrentOsStackTraceExceptTop(..., 1), Foo() will be included in
-// the trace but Bar() and GetCurrentOsStackTraceExceptTop() won't.
-std::string GetCurrentOsStackTraceExceptTop(UnitTest* /*unit_test*/,
-                                            int skip_count) {
-  // We pass skip_count + 1 to skip this wrapper function in addition
-  // to what the user really wants to skip.
-  return GetUnitTestImpl()->CurrentOsStackTraceExceptTop(skip_count + 1);
-}
-
-// Used by the GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_ macro to
-// suppress unreachable code warnings.
-namespace {
-class ClassUniqueToAlwaysTrue {};
-}
-
-bool IsTrue(bool condition) { return condition; }
-
-bool AlwaysTrue() {
-#if GTEST_HAS_EXCEPTIONS
-  // This condition is always false so AlwaysTrue() never actually throws,
-  // but it makes the compiler think that it may throw.
-  if (IsTrue(false))
-    throw ClassUniqueToAlwaysTrue();
-#endif  // GTEST_HAS_EXCEPTIONS
-  return true;
-}
-
-// If *pstr starts with the given prefix, modifies *pstr to be right
-// past the prefix and returns true; otherwise leaves *pstr unchanged
-// and returns false.  None of pstr, *pstr, and prefix can be NULL.
-bool SkipPrefix(const char* prefix, const char** pstr) {
-  const size_t prefix_len = strlen(prefix);
-  if (strncmp(*pstr, prefix, prefix_len) == 0) {
-    *pstr += prefix_len;
-    return true;
-  }
-  return false;
-}
-
-// Parses a string as a command line flag.  The string should have
-// the format "--flag=value".  When def_optional is true, the "=value"
-// part can be omitted.
-//
-// Returns the value of the flag, or NULL if the parsing failed.
-const char* ParseFlagValue(const char* str,
-                           const char* flag,
-                           bool def_optional) {
-  // str and flag must not be NULL.
-  if (str == NULL || flag == NULL) return NULL;
-
-  // The flag must start with "--" followed by GTEST_FLAG_PREFIX_.
-  const std::string flag_str = std::string("--") + GTEST_FLAG_PREFIX_ + flag;
-  const size_t flag_len = flag_str.length();
-  if (strncmp(str, flag_str.c_str(), flag_len) != 0) return NULL;
-
-  // Skips the flag name.
-  const char* flag_end = str + flag_len;
-
-  // When def_optional is true, it's OK to not have a "=value" part.
-  if (def_optional && (flag_end[0] == '\0')) {
-    return flag_end;
-  }
-
-  // If def_optional is true and there are more characters after the
-  // flag name, or if def_optional is false, there must be a '=' after
-  // the flag name.
-  if (flag_end[0] != '=') return NULL;
-
-  // Returns the string after "=".
-  return flag_end + 1;
-}
-
-// Parses a string for a bool flag, in the form of either
-// "--flag=value" or "--flag".
-//
-// In the former case, the value is taken as true as long as it does
-// not start with '0', 'f', or 'F'.
-//
-// In the latter case, the value is taken as true.
-//
-// On success, stores the value of the flag in *value, and returns
-// true.  On failure, returns false without changing *value.
-bool ParseBoolFlag(const char* str, const char* flag, bool* value) {
-  // Gets the value of the flag as a string.
-  const char* const value_str = ParseFlagValue(str, flag, true);
-
-  // Aborts if the parsing failed.
-  if (value_str == NULL) return false;
-
-  // Converts the string value to a bool.
-  *value = !(*value_str == '0' || *value_str == 'f' || *value_str == 'F');
-  return true;
-}
-
-// Parses a string for an Int32 flag, in the form of
-// "--flag=value".
-//
-// On success, stores the value of the flag in *value, and returns
-// true.  On failure, returns false without changing *value.
-bool ParseInt32Flag(const char* str, const char* flag, Int32* value) {
-  // Gets the value of the flag as a string.
-  const char* const value_str = ParseFlagValue(str, flag, false);
-
-  // Aborts if the parsing failed.
-  if (value_str == NULL) return false;
-
-  // Sets *value to the value of the flag.
-  return ParseInt32(Message() << "The value of flag --" << flag,
-                    value_str, value);
-}
-
-// Parses a string for a string flag, in the form of
-// "--flag=value".
-//
-// On success, stores the value of the flag in *value, and returns
-// true.  On failure, returns false without changing *value.
-bool ParseStringFlag(const char* str, const char* flag, std::string* value) {
-  // Gets the value of the flag as a string.
-  const char* const value_str = ParseFlagValue(str, flag, false);
-
-  // Aborts if the parsing failed.
-  if (value_str == NULL) return false;
-
-  // Sets *value to the value of the flag.
-  *value = value_str;
-  return true;
-}
-
-// Determines whether a string has a prefix that Google Test uses for its
-// flags, i.e., starts with GTEST_FLAG_PREFIX_ or GTEST_FLAG_PREFIX_DASH_.
-// If Google Test detects that a command line flag has its prefix but is not
-// recognized, it will print its help message. Flags starting with
-// GTEST_INTERNAL_PREFIX_ followed by "internal_" are considered Google Test
-// internal flags and do not trigger the help message.
-static bool HasGoogleTestFlagPrefix(const char* str) {
-  return (SkipPrefix("--", &str) ||
-          SkipPrefix("-", &str) ||
-          SkipPrefix("/", &str)) &&
-         !SkipPrefix(GTEST_FLAG_PREFIX_ "internal_", &str) &&
-         (SkipPrefix(GTEST_FLAG_PREFIX_, &str) ||
-          SkipPrefix(GTEST_FLAG_PREFIX_DASH_, &str));
-}
-
-// Prints a string containing code-encoded text.  The following escape
-// sequences can be used in the string to control the text color:
-//
-//   @@    prints a single '@' character.
-//   @R    changes the color to red.
-//   @G    changes the color to green.
-//   @Y    changes the color to yellow.
-//   @D    changes to the default terminal text color.
-//
-// TODO(wan@google.com): Write tests for this once we add stdout
-// capturing to Google Test.
-static void PrintColorEncoded(const char* str) {
-  GTestColor color = COLOR_DEFAULT;  // The current color.
-
-  // Conceptually, we split the string into segments divided by escape
-  // sequences.  Then we print one segment at a time.  At the end of
-  // each iteration, the str pointer advances to the beginning of the
-  // next segment.
-  for (;;) {
-    const char* p = strchr(str, '@');
-    if (p == NULL) {
-      ColoredPrintf(color, "%s", str);
-      return;
-    }
-
-    ColoredPrintf(color, "%s", std::string(str, p).c_str());
-
-    const char ch = p[1];
-    str = p + 2;
-    if (ch == '@') {
-      ColoredPrintf(color, "@");
-    } else if (ch == 'D') {
-      color = COLOR_DEFAULT;
-    } else if (ch == 'R') {
-      color = COLOR_RED;
-    } else if (ch == 'G') {
-      color = COLOR_GREEN;
-    } else if (ch == 'Y') {
-      color = COLOR_YELLOW;
-    } else {
-      --str;
-    }
-  }
-}
-
-static const char kColorEncodedHelpMessage[] =
-"This program contains tests written using " GTEST_NAME_ ". You can use the\n"
-"following command line flags to control its behavior:\n"
-"\n"
-"Test Selection:\n"
-"  @G--" GTEST_FLAG_PREFIX_ "list_tests@D\n"
-"      List the names of all tests instead of running them. The name of\n"
-"      TEST(Foo, Bar) is \"Foo.Bar\".\n"
-"  @G--" GTEST_FLAG_PREFIX_ "filter=@YPOSTIVE_PATTERNS"
-    "[@G-@YNEGATIVE_PATTERNS]@D\n"
-"      Run only the tests whose name matches one of the positive patterns but\n"
-"      none of the negative patterns. '?' matches any single character; '*'\n"
-"      matches any substring; ':' separates two patterns.\n"
-"  @G--" GTEST_FLAG_PREFIX_ "also_run_disabled_tests@D\n"
-"      Run all disabled tests too.\n"
-"\n"
-"Test Execution:\n"
-"  @G--" GTEST_FLAG_PREFIX_ "repeat=@Y[COUNT]@D\n"
-"      Run the tests repeatedly; use a negative count to repeat forever.\n"
-"  @G--" GTEST_FLAG_PREFIX_ "shuffle@D\n"
-"      Randomize tests' orders on every iteration.\n"
-"  @G--" GTEST_FLAG_PREFIX_ "random_seed=@Y[NUMBER]@D\n"
-"      Random number seed to use for shuffling test orders (between 1 and\n"
-"      99999, or 0 to use a seed based on the current time).\n"
-"\n"
-"Test Output:\n"
-"  @G--" GTEST_FLAG_PREFIX_ "color=@Y(@Gyes@Y|@Gno@Y|@Gauto@Y)@D\n"
-"      Enable/disable colored output. The default is @Gauto@D.\n"
-"  -@G-" GTEST_FLAG_PREFIX_ "print_time=0@D\n"
-"      Don't print the elapsed time of each test.\n"
-"  @G--" GTEST_FLAG_PREFIX_ "output=xml@Y[@G:@YDIRECTORY_PATH@G"
-    GTEST_PATH_SEP_ "@Y|@G:@YFILE_PATH]@D\n"
-"      Generate an XML report in the given directory or with the given file\n"
-"      name. @YFILE_PATH@D defaults to @Gtest_details.xml@D.\n"
-#if GTEST_CAN_STREAM_RESULTS_
-"  @G--" GTEST_FLAG_PREFIX_ "stream_result_to=@YHOST@G:@YPORT@D\n"
-"      Stream test results to the given server.\n"
-#endif  // GTEST_CAN_STREAM_RESULTS_
-"\n"
-"Assertion Behavior:\n"
-#if GTEST_HAS_DEATH_TEST && !GTEST_OS_WINDOWS
-"  @G--" GTEST_FLAG_PREFIX_ "death_test_style=@Y(@Gfast@Y|@Gthreadsafe@Y)@D\n"
-"      Set the default death test style.\n"
-#endif  // GTEST_HAS_DEATH_TEST && !GTEST_OS_WINDOWS
-"  @G--" GTEST_FLAG_PREFIX_ "break_on_failure@D\n"
-"      Turn assertion failures into debugger break-points.\n"
-"  @G--" GTEST_FLAG_PREFIX_ "throw_on_failure@D\n"
-"      Turn assertion failures into C++ exceptions.\n"
-"  @G--" GTEST_FLAG_PREFIX_ "catch_exceptions=0@D\n"
-"      Do not report exceptions as test failures. Instead, allow them\n"
-"      to crash the program or throw a pop-up (on Windows).\n"
-"\n"
-"Except for @G--" GTEST_FLAG_PREFIX_ "list_tests@D, you can alternatively set "
-    "the corresponding\n"
-"environment variable of a flag (all letters in upper-case). For example, to\n"
-"disable colored text output, you can either specify @G--" GTEST_FLAG_PREFIX_
-    "color=no@D or set\n"
-"the @G" GTEST_FLAG_PREFIX_UPPER_ "COLOR@D environment variable to @Gno@D.\n"
-"\n"
-"For more information, please read the " GTEST_NAME_ " documentation at\n"
-"@G" GTEST_PROJECT_URL_ "@D. If you find a bug in " GTEST_NAME_ "\n"
-"(not one in your own code or tests), please report it to\n"
-"@G<" GTEST_DEV_EMAIL_ ">@D.\n";
-
-bool ParseGoogleTestFlag(const char* const arg) {
-  return ParseBoolFlag(arg, kAlsoRunDisabledTestsFlag,
-                       &GTEST_FLAG(also_run_disabled_tests)) ||
-      ParseBoolFlag(arg, kBreakOnFailureFlag,
-                    &GTEST_FLAG(break_on_failure)) ||
-      ParseBoolFlag(arg, kCatchExceptionsFlag,
-                    &GTEST_FLAG(catch_exceptions)) ||
-      ParseStringFlag(arg, kColorFlag, &GTEST_FLAG(color)) ||
-      ParseStringFlag(arg, kDeathTestStyleFlag,
-                      &GTEST_FLAG(death_test_style)) ||
-      ParseBoolFlag(arg, kDeathTestUseFork,
-                    &GTEST_FLAG(death_test_use_fork)) ||
-      ParseStringFlag(arg, kFilterFlag, &GTEST_FLAG(filter)) ||
-      ParseStringFlag(arg, kInternalRunDeathTestFlag,
-                      &GTEST_FLAG(internal_run_death_test)) ||
-      ParseBoolFlag(arg, kListTestsFlag, &GTEST_FLAG(list_tests)) ||
-      ParseStringFlag(arg, kOutputFlag, &GTEST_FLAG(output)) ||
-      ParseBoolFlag(arg, kPrintTimeFlag, &GTEST_FLAG(print_time)) ||
-      ParseInt32Flag(arg, kRandomSeedFlag, &GTEST_FLAG(random_seed)) ||
-      ParseInt32Flag(arg, kRepeatFlag, &GTEST_FLAG(repeat)) ||
-      ParseBoolFlag(arg, kShuffleFlag, &GTEST_FLAG(shuffle)) ||
-      ParseInt32Flag(arg, kStackTraceDepthFlag,
-                     &GTEST_FLAG(stack_trace_depth)) ||
-      ParseStringFlag(arg, kStreamResultToFlag,
-                      &GTEST_FLAG(stream_result_to)) ||
-      ParseBoolFlag(arg, kThrowOnFailureFlag,
-                    &GTEST_FLAG(throw_on_failure));
-}
-
-#if GTEST_USE_OWN_FLAGFILE_FLAG_
-void LoadFlagsFromFile(const std::string& path) {
-  FILE* flagfile = posix::FOpen(path.c_str(), "r");
-  if (!flagfile) {
-    fprintf(stderr,
-            "Unable to open file \"%s\"\n",
-            GTEST_FLAG(flagfile).c_str());
-    fflush(stderr);
-    exit(EXIT_FAILURE);
-  }
-  std::string contents(ReadEntireFile(flagfile));
-  posix::FClose(flagfile);
-  std::vector<std::string> lines;
-  SplitString(contents, '\n', &lines);
-  for (size_t i = 0; i < lines.size(); ++i) {
-    if (lines[i].empty())
-      continue;
-    if (!ParseGoogleTestFlag(lines[i].c_str()))
-      g_help_flag = true;
-  }
-}
-#endif  // GTEST_USE_OWN_FLAGFILE_FLAG_
-
-// Parses the command line for Google Test flags, without initializing
-// other parts of Google Test.  The type parameter CharType can be
-// instantiated to either char or wchar_t.
-template <typename CharType>
-void ParseGoogleTestFlagsOnlyImpl(int* argc, CharType** argv) {
-  for (int i = 1; i < *argc; i++) {
-    const std::string arg_string = StreamableToString(argv[i]);
-    const char* const arg = arg_string.c_str();
-
-    using internal::ParseBoolFlag;
-    using internal::ParseInt32Flag;
-    using internal::ParseStringFlag;
-
-    bool remove_flag = false;
-    if (ParseGoogleTestFlag(arg)) {
-      remove_flag = true;
-#if GTEST_USE_OWN_FLAGFILE_FLAG_
-    } else if (ParseStringFlag(arg, kFlagfileFlag, &GTEST_FLAG(flagfile))) {
-      LoadFlagsFromFile(GTEST_FLAG(flagfile));
-      remove_flag = true;
-#endif  // GTEST_USE_OWN_FLAGFILE_FLAG_
-    } else if (arg_string == "--help" || arg_string == "-h" ||
-               arg_string == "-?" || arg_string == "/?" ||
-               HasGoogleTestFlagPrefix(arg)) {
-      // Both help flag and unrecognized Google Test flags (excluding
-      // internal ones) trigger help display.
-      g_help_flag = true;
-    }
-
-    if (remove_flag) {
-      // Shift the remainder of the argv list left by one.  Note
-      // that argv has (*argc + 1) elements, the last one always being
-      // NULL.  The following loop moves the trailing NULL element as
-      // well.
-      for (int j = i; j != *argc; j++) {
-        argv[j] = argv[j + 1];
-      }
-
-      // Decrements the argument count.
-      (*argc)--;
-
-      // We also need to decrement the iterator as we just removed
-      // an element.
-      i--;
-    }
-  }
-
-  if (g_help_flag) {
-    // We print the help here instead of in RUN_ALL_TESTS(), as the
-    // latter may not be called at all if the user is using Google
-    // Test with another testing framework.
-    PrintColorEncoded(kColorEncodedHelpMessage);
-  }
-}
-
-// Parses the command line for Google Test flags, without initializing
-// other parts of Google Test.
-void ParseGoogleTestFlagsOnly(int* argc, char** argv) {
-  ParseGoogleTestFlagsOnlyImpl(argc, argv);
-}
-void ParseGoogleTestFlagsOnly(int* argc, wchar_t** argv) {
-  ParseGoogleTestFlagsOnlyImpl(argc, argv);
-}
-
-// The internal implementation of InitGoogleTest().
-//
-// The type parameter CharType can be instantiated to either char or
-// wchar_t.
-template <typename CharType>
-void InitGoogleTestImpl(int* argc, CharType** argv) {
-  // We don't want to run the initialization code twice.
-  if (GTestIsInitialized()) return;
-
-  if (*argc <= 0) return;
-
-  g_argvs.clear();
-  for (int i = 0; i != *argc; i++) {
-    g_argvs.push_back(StreamableToString(argv[i]));
-  }
-
-  ParseGoogleTestFlagsOnly(argc, argv);
-  GetUnitTestImpl()->PostFlagParsingInit();
-}
-
-}  // namespace internal
-
-// Initializes Google Test.  This must be called before calling
-// RUN_ALL_TESTS().  In particular, it parses a command line for the
-// flags that Google Test recognizes.  Whenever a Google Test flag is
-// seen, it is removed from argv, and *argc is decremented.
-//
-// No value is returned.  Instead, the Google Test flag variables are
-// updated.
-//
-// Calling the function for the second time has no user-visible effect.
-void InitGoogleTest(int* argc, char** argv) {
-#if defined(GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_)
-  GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_(argc, argv);
-#else  // defined(GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_)
-  internal::InitGoogleTestImpl(argc, argv);
-#endif  // defined(GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_)
-}
-
-// This overloaded version can be used in Windows programs compiled in
-// UNICODE mode.
-void InitGoogleTest(int* argc, wchar_t** argv) {
-#if defined(GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_)
-  GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_(argc, argv);
-#else  // defined(GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_)
-  internal::InitGoogleTestImpl(argc, argv);
-#endif  // defined(GTEST_CUSTOM_INIT_GOOGLE_TEST_FUNCTION_)
-}
-
-}  // namespace testing
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan), vladl@google.com (Vlad Losev)
-//
-// This file implements death tests.
-
-
-#if GTEST_HAS_DEATH_TEST
-
-# if GTEST_OS_MAC
-#  include <crt_externs.h>
-# endif  // GTEST_OS_MAC
-
-# include <errno.h>
-# include <fcntl.h>
-# include <limits.h>
-
-# if GTEST_OS_LINUX
-#  include <signal.h>
-# endif  // GTEST_OS_LINUX
-
-# include <stdarg.h>
-
-# if GTEST_OS_WINDOWS
-#  include <windows.h>
-# else
-#  include <sys/mman.h>
-#  include <sys/wait.h>
-# endif  // GTEST_OS_WINDOWS
-
-# if GTEST_OS_QNX
-#  include <spawn.h>
-# endif  // GTEST_OS_QNX
-
-#endif  // GTEST_HAS_DEATH_TEST
-
-
-// Indicates that this translation unit is part of Google Test's
-// implementation.  It must come before gtest-internal-inl.h is
-// included, or there will be a compiler error.  This trick exists to
-// prevent the accidental inclusion of gtest-internal-inl.h in the
-// user's code.
-#define GTEST_IMPLEMENTATION_ 1
-#undef GTEST_IMPLEMENTATION_
-
-namespace testing {
-
-// Constants.
-
-// The default death test style.
-static const char kDefaultDeathTestStyle[] = "fast";
-
-GTEST_DEFINE_string_(
-    death_test_style,
-    internal::StringFromGTestEnv("death_test_style", kDefaultDeathTestStyle),
-    "Indicates how to run a death test in a forked child process: "
-    "\"threadsafe\" (child process re-executes the test binary "
-    "from the beginning, running only the specific death test) or "
-    "\"fast\" (child process runs the death test immediately "
-    "after forking).");
-
-GTEST_DEFINE_bool_(
-    death_test_use_fork,
-    internal::BoolFromGTestEnv("death_test_use_fork", false),
-    "Instructs to use fork()/_exit() instead of clone() in death tests. "
-    "Ignored and always uses fork() on POSIX systems where clone() is not "
-    "implemented. Useful when running under valgrind or similar tools if "
-    "those do not support clone(). Valgrind 3.3.1 will just fail if "
-    "it sees an unsupported combination of clone() flags. "
-    "It is not recommended to use this flag w/o valgrind though it will "
-    "work in 99% of the cases. Once valgrind is fixed, this flag will "
-    "most likely be removed.");
-
-namespace internal {
-GTEST_DEFINE_string_(
-    internal_run_death_test, "",
-    "Indicates the file, line number, temporal index of "
-    "the single death test to run, and a file descriptor to "
-    "which a success code may be sent, all separated by "
-    "the '|' characters.  This flag is specified if and only if the current "
-    "process is a sub-process launched for running a thread-safe "
-    "death test.  FOR INTERNAL USE ONLY.");
-}  // namespace internal
-
-#if GTEST_HAS_DEATH_TEST
-
-namespace internal {
-
-// Valid only for fast death tests. Indicates the code is running in the
-// child process of a fast style death test.
-# if !GTEST_OS_WINDOWS
-static bool g_in_fast_death_test_child = false;
-# endif
-
-// Returns a Boolean value indicating whether the caller is currently
-// executing in the context of the death test child process.  Tools such as
-// Valgrind heap checkers may need this to modify their behavior in death
-// tests.  IMPORTANT: This is an internal utility.  Using it may break the
-// implementation of death tests.  User code MUST NOT use it.
-bool InDeathTestChild() {
-# if GTEST_OS_WINDOWS
-
-  // On Windows, death tests are thread-safe regardless of the value of the
-  // death_test_style flag.
-  return !GTEST_FLAG(internal_run_death_test).empty();
-
-# else
-
-  if (GTEST_FLAG(death_test_style) == "threadsafe")
-    return !GTEST_FLAG(internal_run_death_test).empty();
-  else
-    return g_in_fast_death_test_child;
-#endif
-}
-
-}  // namespace internal
-
-// ExitedWithCode constructor.
-ExitedWithCode::ExitedWithCode(int exit_code) : exit_code_(exit_code) {
-}
-
-// ExitedWithCode function-call operator.
-bool ExitedWithCode::operator()(int exit_status) const {
-# if GTEST_OS_WINDOWS
-
-  return exit_status == exit_code_;
-
-# else
-
-  return WIFEXITED(exit_status) && WEXITSTATUS(exit_status) == exit_code_;
-
-# endif  // GTEST_OS_WINDOWS
-}
-
-# if !GTEST_OS_WINDOWS
-// KilledBySignal constructor.
-KilledBySignal::KilledBySignal(int signum) : signum_(signum) {
-}
-
-// KilledBySignal function-call operator.
-bool KilledBySignal::operator()(int exit_status) const {
-#  if defined(GTEST_KILLED_BY_SIGNAL_OVERRIDE_)
-  {
-    bool result;
-    if (GTEST_KILLED_BY_SIGNAL_OVERRIDE_(signum_, exit_status, &result)) {
-      return result;
-    }
-  }
-#  endif  // defined(GTEST_KILLED_BY_SIGNAL_OVERRIDE_)
-  return WIFSIGNALED(exit_status) && WTERMSIG(exit_status) == signum_;
-}
-# endif  // !GTEST_OS_WINDOWS
-
-namespace internal {
-
-// Utilities needed for death tests.
-
-// Generates a textual description of a given exit code, in the format
-// specified by wait(2).
-static std::string ExitSummary(int exit_code) {
-  Message m;
-
-# if GTEST_OS_WINDOWS
-
-  m << "Exited with exit status " << exit_code;
-
-# else
-
-  if (WIFEXITED(exit_code)) {
-    m << "Exited with exit status " << WEXITSTATUS(exit_code);
-  } else if (WIFSIGNALED(exit_code)) {
-    m << "Terminated by signal " << WTERMSIG(exit_code);
-  }
-#  ifdef WCOREDUMP
-  if (WCOREDUMP(exit_code)) {
-    m << " (core dumped)";
-  }
-#  endif
-# endif  // GTEST_OS_WINDOWS
-
-  return m.GetString();
-}
-
-// Returns true if exit_status describes a process that was terminated
-// by a signal, or exited normally with a nonzero exit code.
-bool ExitedUnsuccessfully(int exit_status) {
-  return !ExitedWithCode(0)(exit_status);
-}
-
-# if !GTEST_OS_WINDOWS
-// Generates a textual failure message when a death test finds more than
-// one thread running, or cannot determine the number of threads, prior
-// to executing the given statement.  It is the responsibility of the
-// caller not to pass a thread_count of 1.
-static std::string DeathTestThreadWarning(size_t thread_count) {
-  Message msg;
-  msg << "Death tests use fork(), which is unsafe particularly"
-      << " in a threaded context. For this test, " << GTEST_NAME_ << " ";
-  if (thread_count == 0)
-    msg << "couldn't detect the number of threads.";
-  else
-    msg << "detected " << thread_count << " threads.";
-  return msg.GetString();
-}
-# endif  // !GTEST_OS_WINDOWS
-
-// Flag characters for reporting a death test that did not die.
-static const char kDeathTestLived = 'L';
-static const char kDeathTestReturned = 'R';
-static const char kDeathTestThrew = 'T';
-static const char kDeathTestInternalError = 'I';
-
-// An enumeration describing all of the possible ways that a death test can
-// conclude.  DIED means that the process died while executing the test
-// code; LIVED means that process lived beyond the end of the test code;
-// RETURNED means that the test statement attempted to execute a return
-// statement, which is not allowed; THREW means that the test statement
-// returned control by throwing an exception.  IN_PROGRESS means the test
-// has not yet concluded.
-// TODO(vladl@google.com): Unify names and possibly values for
-// AbortReason, DeathTestOutcome, and flag characters above.
-enum DeathTestOutcome { IN_PROGRESS, DIED, LIVED, RETURNED, THREW };
-
-// Routine for aborting the program which is safe to call from an
-// exec-style death test child process, in which case the error
-// message is propagated back to the parent process.  Otherwise, the
-// message is simply printed to stderr.  In either case, the program
-// then exits with status 1.
-void DeathTestAbort(const std::string& message) {
-  // On a POSIX system, this function may be called from a threadsafe-style
-  // death test child process, which operates on a very small stack.  Use
-  // the heap for any additional non-minuscule memory requirements.
-  const InternalRunDeathTestFlag* const flag =
-      GetUnitTestImpl()->internal_run_death_test_flag();
-  if (flag != NULL) {
-    FILE* parent = posix::FDOpen(flag->write_fd(), "w");
-    fputc(kDeathTestInternalError, parent);
-    fprintf(parent, "%s", message.c_str());
-    fflush(parent);
-    _exit(1);
-  } else {
-    fprintf(stderr, "%s", message.c_str());
-    fflush(stderr);
-    posix::Abort();
-  }
-}
-
-// A replacement for CHECK that calls DeathTestAbort if the assertion
-// fails.
-# define GTEST_DEATH_TEST_CHECK_(expression) \
-  do { \
-    if (!::testing::internal::IsTrue(expression)) { \
-      DeathTestAbort( \
-          ::std::string("CHECK failed: File ") + __FILE__ +  ", line " \
-          + ::testing::internal::StreamableToString(__LINE__) + ": " \
-          + #expression); \
-    } \
-  } while (::testing::internal::AlwaysFalse())
-
-// This macro is similar to GTEST_DEATH_TEST_CHECK_, but it is meant for
-// evaluating any system call that fulfills two conditions: it must return
-// -1 on failure, and set errno to EINTR when it is interrupted and
-// should be tried again.  The macro expands to a loop that repeatedly
-// evaluates the expression as long as it evaluates to -1 and sets
-// errno to EINTR.  If the expression evaluates to -1 but errno is
-// something other than EINTR, DeathTestAbort is called.
-# define GTEST_DEATH_TEST_CHECK_SYSCALL_(expression) \
-  do { \
-    int gtest_retval; \
-    do { \
-      gtest_retval = (expression); \
-    } while (gtest_retval == -1 && errno == EINTR); \
-    if (gtest_retval == -1) { \
-      DeathTestAbort( \
-          ::std::string("CHECK failed: File ") + __FILE__ + ", line " \
-          + ::testing::internal::StreamableToString(__LINE__) + ": " \
-          + #expression + " != -1"); \
-    } \
-  } while (::testing::internal::AlwaysFalse())
-
-// Returns the message describing the last system error in errno.
-std::string GetLastErrnoDescription() {
-    return errno == 0 ? "" : posix::StrError(errno);
-}
-
-// This is called from a death test parent process to read a failure
-// message from the death test child process and log it with the FATAL
-// severity. On Windows, the message is read from a pipe handle. On other
-// platforms, it is read from a file descriptor.
-static void FailFromInternalError(int fd) {
-  Message error;
-  char buffer[256];
-  int num_read;
-
-  do {
-    while ((num_read = posix::Read(fd, buffer, 255)) > 0) {
-      buffer[num_read] = '\0';
-      error << buffer;
-    }
-  } while (num_read == -1 && errno == EINTR);
-
-  if (num_read == 0) {
-    GTEST_LOG_(FATAL) << error.GetString();
-  } else {
-    const int last_error = errno;
-    GTEST_LOG_(FATAL) << "Error while reading death test internal: "
-                      << GetLastErrnoDescription() << " [" << last_error << "]";
-  }
-}
-
-// Death test constructor.  Increments the running death test count
-// for the current test.
-DeathTest::DeathTest() {
-  TestInfo* const info = GetUnitTestImpl()->current_test_info();
-  if (info == NULL) {
-    DeathTestAbort("Cannot run a death test outside of a TEST or "
-                   "TEST_F construct");
-  }
-}
-
-// Creates and returns a death test by dispatching to the current
-// death test factory.
-bool DeathTest::Create(const char* statement, const RE* regex,
-                       const char* file, int line, DeathTest** test) {
-  return GetUnitTestImpl()->death_test_factory()->Create(
-      statement, regex, file, line, test);
-}
-
-const char* DeathTest::LastMessage() {
-  return last_death_test_message_.c_str();
-}
-
-void DeathTest::set_last_death_test_message(const std::string& message) {
-  last_death_test_message_ = message;
-}
-
-std::string DeathTest::last_death_test_message_;
-
-// Provides cross platform implementation for some death functionality.
-class DeathTestImpl : public DeathTest {
- protected:
-  DeathTestImpl(const char* a_statement, const RE* a_regex)
-      : statement_(a_statement),
-        regex_(a_regex),
-        spawned_(false),
-        status_(-1),
-        outcome_(IN_PROGRESS),
-        read_fd_(-1),
-        write_fd_(-1) {}
-
-  // read_fd_ is expected to be closed and cleared by a derived class.
-  ~DeathTestImpl() { GTEST_DEATH_TEST_CHECK_(read_fd_ == -1); }
-
-  void Abort(AbortReason reason);
-  virtual bool Passed(bool status_ok);
-
-  const char* statement() const { return statement_; }
-  const RE* regex() const { return regex_; }
-  bool spawned() const { return spawned_; }
-  void set_spawned(bool is_spawned) { spawned_ = is_spawned; }
-  int status() const { return status_; }
-  void set_status(int a_status) { status_ = a_status; }
-  DeathTestOutcome outcome() const { return outcome_; }
-  void set_outcome(DeathTestOutcome an_outcome) { outcome_ = an_outcome; }
-  int read_fd() const { return read_fd_; }
-  void set_read_fd(int fd) { read_fd_ = fd; }
-  int write_fd() const { return write_fd_; }
-  void set_write_fd(int fd) { write_fd_ = fd; }
-
-  // Called in the parent process only. Reads the result code of the death
-  // test child process via a pipe, interprets it to set the outcome_
-  // member, and closes read_fd_.  Outputs diagnostics and terminates in
-  // case of unexpected codes.
-  void ReadAndInterpretStatusByte();
-
- private:
-  // The textual content of the code this object is testing.  This class
-  // doesn't own this string and should not attempt to delete it.
-  const char* const statement_;
-  // The regular expression which test output must match.  DeathTestImpl
-  // doesn't own this object and should not attempt to delete it.
-  const RE* const regex_;
-  // True if the death test child process has been successfully spawned.
-  bool spawned_;
-  // The exit status of the child process.
-  int status_;
-  // How the death test concluded.
-  DeathTestOutcome outcome_;
-  // Descriptor to the read end of the pipe to the child process.  It is
-  // always -1 in the child process.  The child keeps its write end of the
-  // pipe in write_fd_.
-  int read_fd_;
-  // Descriptor to the child's write end of the pipe to the parent process.
-  // It is always -1 in the parent process.  The parent keeps its end of the
-  // pipe in read_fd_.
-  int write_fd_;
-};
-
-// Called in the parent process only. Reads the result code of the death
-// test child process via a pipe, interprets it to set the outcome_
-// member, and closes read_fd_.  Outputs diagnostics and terminates in
-// case of unexpected codes.
-void DeathTestImpl::ReadAndInterpretStatusByte() {
-  char flag;
-  int bytes_read;
-
-  // The read() here blocks until data is available (signifying the
-  // failure of the death test) or until the pipe is closed (signifying
-  // its success), so it's okay to call this in the parent before
-  // the child process has exited.
-  do {
-    bytes_read = posix::Read(read_fd(), &flag, 1);
-  } while (bytes_read == -1 && errno == EINTR);
-
-  if (bytes_read == 0) {
-    set_outcome(DIED);
-  } else if (bytes_read == 1) {
-    switch (flag) {
-      case kDeathTestReturned:
-        set_outcome(RETURNED);
-        break;
-      case kDeathTestThrew:
-        set_outcome(THREW);
-        break;
-      case kDeathTestLived:
-        set_outcome(LIVED);
-        break;
-      case kDeathTestInternalError:
-        FailFromInternalError(read_fd());  // Does not return.
-        break;
-      default:
-        GTEST_LOG_(FATAL) << "Death test child process reported "
-                          << "unexpected status byte ("
-                          << static_cast<unsigned int>(flag) << ")";
-    }
-  } else {
-    GTEST_LOG_(FATAL) << "Read from death test child process failed: "
-                      << GetLastErrnoDescription();
-  }
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(posix::Close(read_fd()));
-  set_read_fd(-1);
-}
-
-// Signals that the death test code which should have exited, didn't.
-// Should be called only in a death test child process.
-// Writes a status byte to the child's status file descriptor, then
-// calls _exit(1).
-void DeathTestImpl::Abort(AbortReason reason) {
-  // The parent process considers the death test to be a failure if
-  // it finds any data in our pipe.  So, here we write a single flag byte
-  // to the pipe, then exit.
-  const char status_ch =
-      reason == TEST_DID_NOT_DIE ? kDeathTestLived :
-      reason == TEST_THREW_EXCEPTION ? kDeathTestThrew : kDeathTestReturned;
-
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(posix::Write(write_fd(), &status_ch, 1));
-  // We are leaking the descriptor here because on some platforms (i.e.,
-  // when built as Windows DLL), destructors of global objects will still
-  // run after calling _exit(). On such systems, write_fd_ will be
-  // indirectly closed from the destructor of UnitTestImpl, causing double
-  // close if it is also closed here. On debug configurations, double close
-  // may assert. As there are no in-process buffers to flush here, we are
-  // relying on the OS to close the descriptor after the process terminates
-  // when the destructors are not run.
-  _exit(1);  // Exits w/o any normal exit hooks (we were supposed to crash)
-}
-
-// Returns an indented copy of stderr output for a death test.
-// This makes distinguishing death test output lines from regular log lines
-// much easier.
-static ::std::string FormatDeathTestOutput(const ::std::string& output) {
-  ::std::string ret;
-  for (size_t at = 0; ; ) {
-    const size_t line_end = output.find('\n', at);
-    ret += "[  DEATH   ] ";
-    if (line_end == ::std::string::npos) {
-      ret += output.substr(at);
-      break;
-    }
-    ret += output.substr(at, line_end + 1 - at);
-    at = line_end + 1;
-  }
-  return ret;
-}
-
-// Assesses the success or failure of a death test, using both private
-// members which have previously been set, and one argument:
-//
-// Private data members:
-//   outcome:  An enumeration describing how the death test
-//             concluded: DIED, LIVED, THREW, or RETURNED.  The death test
-//             fails in the latter three cases.
-//   status:   The exit status of the child process. On *nix, it is in the
-//             in the format specified by wait(2). On Windows, this is the
-//             value supplied to the ExitProcess() API or a numeric code
-//             of the exception that terminated the program.
-//   regex:    A regular expression object to be applied to
-//             the test's captured standard error output; the death test
-//             fails if it does not match.
-//
-// Argument:
-//   status_ok: true if exit_status is acceptable in the context of
-//              this particular death test, which fails if it is false
-//
-// Returns true iff all of the above conditions are met.  Otherwise, the
-// first failing condition, in the order given above, is the one that is
-// reported. Also sets the last death test message string.
-bool DeathTestImpl::Passed(bool status_ok) {
-  if (!spawned())
-    return false;
-
-  const std::string error_message = GetCapturedStderr();
-
-  bool success = false;
-  Message buffer;
-
-  buffer << "Death test: " << statement() << "\n";
-  switch (outcome()) {
-    case LIVED:
-      buffer << "    Result: failed to die.\n"
-             << " Error msg:\n" << FormatDeathTestOutput(error_message);
-      break;
-    case THREW:
-      buffer << "    Result: threw an exception.\n"
-             << " Error msg:\n" << FormatDeathTestOutput(error_message);
-      break;
-    case RETURNED:
-      buffer << "    Result: illegal return in test statement.\n"
-             << " Error msg:\n" << FormatDeathTestOutput(error_message);
-      break;
-    case DIED:
-      if (status_ok) {
-        const bool matched = RE::PartialMatch(error_message.c_str(), *regex());
-        if (matched) {
-          success = true;
-        } else {
-          buffer << "    Result: died but not with expected error.\n"
-                 << "  Expected: " << regex()->pattern() << "\n"
-                 << "Actual msg:\n" << FormatDeathTestOutput(error_message);
-        }
-      } else {
-        buffer << "    Result: died but not with expected exit code:\n"
-               << "            " << ExitSummary(status()) << "\n"
-               << "Actual msg:\n" << FormatDeathTestOutput(error_message);
-      }
-      break;
-    case IN_PROGRESS:
-    default:
-      GTEST_LOG_(FATAL)
-          << "DeathTest::Passed somehow called before conclusion of test";
-  }
-
-  DeathTest::set_last_death_test_message(buffer.GetString());
-  return success;
-}
-
-# if GTEST_OS_WINDOWS
-// WindowsDeathTest implements death tests on Windows. Due to the
-// specifics of starting new processes on Windows, death tests there are
-// always threadsafe, and Google Test considers the
-// --gtest_death_test_style=fast setting to be equivalent to
-// --gtest_death_test_style=threadsafe there.
-//
-// A few implementation notes:  Like the Linux version, the Windows
-// implementation uses pipes for child-to-parent communication. But due to
-// the specifics of pipes on Windows, some extra steps are required:
-//
-// 1. The parent creates a communication pipe and stores handles to both
-//    ends of it.
-// 2. The parent starts the child and provides it with the information
-//    necessary to acquire the handle to the write end of the pipe.
-// 3. The child acquires the write end of the pipe and signals the parent
-//    using a Windows event.
-// 4. Now the parent can release the write end of the pipe on its side. If
-//    this is done before step 3, the object's reference count goes down to
-//    0 and it is destroyed, preventing the child from acquiring it. The
-//    parent now has to release it, or read operations on the read end of
-//    the pipe will not return when the child terminates.
-// 5. The parent reads child's output through the pipe (outcome code and
-//    any possible error messages) from the pipe, and its stderr and then
-//    determines whether to fail the test.
-//
-// Note: to distinguish Win32 API calls from the local method and function
-// calls, the former are explicitly resolved in the global namespace.
-//
-class WindowsDeathTest : public DeathTestImpl {
- public:
-  WindowsDeathTest(const char* a_statement,
-                   const RE* a_regex,
-                   const char* file,
-                   int line)
-      : DeathTestImpl(a_statement, a_regex), file_(file), line_(line) {}
-
-  // All of these virtual functions are inherited from DeathTest.
-  virtual int Wait();
-  virtual TestRole AssumeRole();
-
- private:
-  // The name of the file in which the death test is located.
-  const char* const file_;
-  // The line number on which the death test is located.
-  const int line_;
-  // Handle to the write end of the pipe to the child process.
-  AutoHandle write_handle_;
-  // Child process handle.
-  AutoHandle child_handle_;
-  // Event the child process uses to signal the parent that it has
-  // acquired the handle to the write end of the pipe. After seeing this
-  // event the parent can release its own handles to make sure its
-  // ReadFile() calls return when the child terminates.
-  AutoHandle event_handle_;
-};
-
-// Waits for the child in a death test to exit, returning its exit
-// status, or 0 if no child process exists.  As a side effect, sets the
-// outcome data member.
-int WindowsDeathTest::Wait() {
-  if (!spawned())
-    return 0;
-
-  // Wait until the child either signals that it has acquired the write end
-  // of the pipe or it dies.
-  const HANDLE wait_handles[2] = { child_handle_.Get(), event_handle_.Get() };
-  switch (::WaitForMultipleObjects(2,
-                                   wait_handles,
-                                   FALSE,  // Waits for any of the handles.
-                                   INFINITE)) {
-    case WAIT_OBJECT_0:
-    case WAIT_OBJECT_0 + 1:
-      break;
-    default:
-      GTEST_DEATH_TEST_CHECK_(false);  // Should not get here.
-  }
-
-  // The child has acquired the write end of the pipe or exited.
-  // We release the handle on our side and continue.
-  write_handle_.Reset();
-  event_handle_.Reset();
-
-  ReadAndInterpretStatusByte();
-
-  // Waits for the child process to exit if it haven't already. This
-  // returns immediately if the child has already exited, regardless of
-  // whether previous calls to WaitForMultipleObjects synchronized on this
-  // handle or not.
-  GTEST_DEATH_TEST_CHECK_(
-      WAIT_OBJECT_0 == ::WaitForSingleObject(child_handle_.Get(),
-                                             INFINITE));
-  DWORD status_code;
-  GTEST_DEATH_TEST_CHECK_(
-      ::GetExitCodeProcess(child_handle_.Get(), &status_code) != FALSE);
-  child_handle_.Reset();
-  set_status(static_cast<int>(status_code));
-  return status();
-}
-
-// The AssumeRole process for a Windows death test.  It creates a child
-// process with the same executable as the current process to run the
-// death test.  The child process is given the --gtest_filter and
-// --gtest_internal_run_death_test flags such that it knows to run the
-// current death test only.
-DeathTest::TestRole WindowsDeathTest::AssumeRole() {
-  const UnitTestImpl* const impl = GetUnitTestImpl();
-  const InternalRunDeathTestFlag* const flag =
-      impl->internal_run_death_test_flag();
-  const TestInfo* const info = impl->current_test_info();
-  const int death_test_index = info->result()->death_test_count();
-
-  if (flag != NULL) {
-    // ParseInternalRunDeathTestFlag() has performed all the necessary
-    // processing.
-    set_write_fd(flag->write_fd());
-    return EXECUTE_TEST;
-  }
-
-  // WindowsDeathTest uses an anonymous pipe to communicate results of
-  // a death test.
-  SECURITY_ATTRIBUTES handles_are_inheritable = {
-    sizeof(SECURITY_ATTRIBUTES), NULL, TRUE };
-  HANDLE read_handle, write_handle;
-  GTEST_DEATH_TEST_CHECK_(
-      ::CreatePipe(&read_handle, &write_handle, &handles_are_inheritable,
-                   0)  // Default buffer size.
-      != FALSE);
-  set_read_fd(::_open_osfhandle(reinterpret_cast<intptr_t>(read_handle),
-                                O_RDONLY));
-  write_handle_.Reset(write_handle);
-  event_handle_.Reset(::CreateEvent(
-      &handles_are_inheritable,
-      TRUE,    // The event will automatically reset to non-signaled state.
-      FALSE,   // The initial state is non-signalled.
-      NULL));  // The even is unnamed.
-  GTEST_DEATH_TEST_CHECK_(event_handle_.Get() != NULL);
-  const std::string filter_flag =
-      std::string("--") + GTEST_FLAG_PREFIX_ + kFilterFlag + "=" +
-      info->test_case_name() + "." + info->name();
-  const std::string internal_flag =
-      std::string("--") + GTEST_FLAG_PREFIX_ + kInternalRunDeathTestFlag +
-      "=" + file_ + "|" + StreamableToString(line_) + "|" +
-      StreamableToString(death_test_index) + "|" +
-      StreamableToString(static_cast<unsigned int>(::GetCurrentProcessId())) +
-      // size_t has the same width as pointers on both 32-bit and 64-bit
-      // Windows platforms.
-      // See http://msdn.microsoft.com/en-us/library/tcxf1dw6.aspx.
-      "|" + StreamableToString(reinterpret_cast<size_t>(write_handle)) +
-      "|" + StreamableToString(reinterpret_cast<size_t>(event_handle_.Get()));
-
-  char executable_path[_MAX_PATH + 1];  // NOLINT
-  GTEST_DEATH_TEST_CHECK_(
-      _MAX_PATH + 1 != ::GetModuleFileNameA(NULL,
-                                            executable_path,
-                                            _MAX_PATH));
-
-  std::string command_line =
-      std::string(::GetCommandLineA()) + " " + filter_flag + " \"" +
-      internal_flag + "\"";
-
-  DeathTest::set_last_death_test_message("");
-
-  CaptureStderr();
-  // Flush the log buffers since the log streams are shared with the child.
-  FlushInfoLog();
-
-  // The child process will share the standard handles with the parent.
-  STARTUPINFOA startup_info;
-  memset(&startup_info, 0, sizeof(STARTUPINFO));
-  startup_info.dwFlags = STARTF_USESTDHANDLES;
-  startup_info.hStdInput = ::GetStdHandle(STD_INPUT_HANDLE);
-  startup_info.hStdOutput = ::GetStdHandle(STD_OUTPUT_HANDLE);
-  startup_info.hStdError = ::GetStdHandle(STD_ERROR_HANDLE);
-
-  PROCESS_INFORMATION process_info;
-  GTEST_DEATH_TEST_CHECK_(::CreateProcessA(
-      executable_path,
-      const_cast<char*>(command_line.c_str()),
-      NULL,   // Retuned process handle is not inheritable.
-      NULL,   // Retuned thread handle is not inheritable.
-      TRUE,   // Child inherits all inheritable handles (for write_handle_).
-      0x0,    // Default creation flags.
-      NULL,   // Inherit the parent's environment.
-      UnitTest::GetInstance()->original_working_dir(),
-      &startup_info,
-      &process_info) != FALSE);
-  child_handle_.Reset(process_info.hProcess);
-  ::CloseHandle(process_info.hThread);
-  set_spawned(true);
-  return OVERSEE_TEST;
-}
-# else  // We are not on Windows.
-
-// ForkingDeathTest provides implementations for most of the abstract
-// methods of the DeathTest interface.  Only the AssumeRole method is
-// left undefined.
-class ForkingDeathTest : public DeathTestImpl {
- public:
-  ForkingDeathTest(const char* statement, const RE* regex);
-
-  // All of these virtual functions are inherited from DeathTest.
-  virtual int Wait();
-
- protected:
-  void set_child_pid(pid_t child_pid) { child_pid_ = child_pid; }
-
- private:
-  // PID of child process during death test; 0 in the child process itself.
-  pid_t child_pid_;
-};
-
-// Constructs a ForkingDeathTest.
-ForkingDeathTest::ForkingDeathTest(const char* a_statement, const RE* a_regex)
-    : DeathTestImpl(a_statement, a_regex),
-      child_pid_(-1) {}
-
-// Waits for the child in a death test to exit, returning its exit
-// status, or 0 if no child process exists.  As a side effect, sets the
-// outcome data member.
-int ForkingDeathTest::Wait() {
-  if (!spawned())
-    return 0;
-
-  ReadAndInterpretStatusByte();
-
-  int status_value;
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(waitpid(child_pid_, &status_value, 0));
-  set_status(status_value);
-  return status_value;
-}
-
-// A concrete death test class that forks, then immediately runs the test
-// in the child process.
-class NoExecDeathTest : public ForkingDeathTest {
- public:
-  NoExecDeathTest(const char* a_statement, const RE* a_regex) :
-      ForkingDeathTest(a_statement, a_regex) { }
-  virtual TestRole AssumeRole();
-};
-
-// The AssumeRole process for a fork-and-run death test.  It implements a
-// straightforward fork, with a simple pipe to transmit the status byte.
-DeathTest::TestRole NoExecDeathTest::AssumeRole() {
-  const size_t thread_count = GetThreadCount();
-  if (thread_count != 1) {
-    GTEST_LOG_(WARNING) << DeathTestThreadWarning(thread_count);
-  }
-
-  int pipe_fd[2];
-  GTEST_DEATH_TEST_CHECK_(pipe(pipe_fd) != -1);
-
-  DeathTest::set_last_death_test_message("");
-  CaptureStderr();
-  // When we fork the process below, the log file buffers are copied, but the
-  // file descriptors are shared.  We flush all log files here so that closing
-  // the file descriptors in the child process doesn't throw off the
-  // synchronization between descriptors and buffers in the parent process.
-  // This is as close to the fork as possible to avoid a race condition in case
-  // there are multiple threads running before the death test, and another
-  // thread writes to the log file.
-  FlushInfoLog();
-
-  const pid_t child_pid = fork();
-  GTEST_DEATH_TEST_CHECK_(child_pid != -1);
-  set_child_pid(child_pid);
-  if (child_pid == 0) {
-    GTEST_DEATH_TEST_CHECK_SYSCALL_(close(pipe_fd[0]));
-    set_write_fd(pipe_fd[1]);
-    // Redirects all logging to stderr in the child process to prevent
-    // concurrent writes to the log files.  We capture stderr in the parent
-    // process and append the child process' output to a log.
-    LogToStderr();
-    // Event forwarding to the listeners of event listener API mush be shut
-    // down in death test subprocesses.
-    GetUnitTestImpl()->listeners()->SuppressEventForwarding();
-    g_in_fast_death_test_child = true;
-    return EXECUTE_TEST;
-  } else {
-    GTEST_DEATH_TEST_CHECK_SYSCALL_(close(pipe_fd[1]));
-    set_read_fd(pipe_fd[0]);
-    set_spawned(true);
-    return OVERSEE_TEST;
-  }
-}
-
-// A concrete death test class that forks and re-executes the main
-// program from the beginning, with command-line flags set that cause
-// only this specific death test to be run.
-class ExecDeathTest : public ForkingDeathTest {
- public:
-  ExecDeathTest(const char* a_statement, const RE* a_regex,
-                const char* file, int line) :
-      ForkingDeathTest(a_statement, a_regex), file_(file), line_(line) { }
-  virtual TestRole AssumeRole();
- private:
-  static ::std::vector<testing::internal::string>
-  GetArgvsForDeathTestChildProcess() {
-    ::std::vector<testing::internal::string> args = GetInjectableArgvs();
-#  if defined(GTEST_EXTRA_DEATH_TEST_COMMAND_LINE_ARGS_)
-    ::std::vector<testing::internal::string> extra_args =
-        GTEST_EXTRA_DEATH_TEST_COMMAND_LINE_ARGS_();
-    args.insert(args.end(), extra_args.begin(), extra_args.end());
-#  endif  // defined(GTEST_EXTRA_DEATH_TEST_COMMAND_LINE_ARGS_)
-    return args;
-  }
-  // The name of the file in which the death test is located.
-  const char* const file_;
-  // The line number on which the death test is located.
-  const int line_;
-};
-
-// Utility class for accumulating command-line arguments.
-class Arguments {
- public:
-  Arguments() {
-    args_.push_back(NULL);
-  }
-
-  ~Arguments() {
-    for (std::vector<char*>::iterator i = args_.begin(); i != args_.end();
-         ++i) {
-      free(*i);
-    }
-  }
-  void AddArgument(const char* argument) {
-    args_.insert(args_.end() - 1, posix::StrDup(argument));
-  }
-
-  template <typename Str>
-  void AddArguments(const ::std::vector<Str>& arguments) {
-    for (typename ::std::vector<Str>::const_iterator i = arguments.begin();
-         i != arguments.end();
-         ++i) {
-      args_.insert(args_.end() - 1, posix::StrDup(i->c_str()));
-    }
-  }
-  char* const* Argv() {
-    return &args_[0];
-  }
-
- private:
-  std::vector<char*> args_;
-};
-
-// A struct that encompasses the arguments to the child process of a
-// threadsafe-style death test process.
-struct ExecDeathTestArgs {
-  char* const* argv;  // Command-line arguments for the child's call to exec
-  int close_fd;       // File descriptor to close; the read end of a pipe
-};
-
-#  if GTEST_OS_MAC
-inline char** GetEnviron() {
-  // When Google Test is built as a framework on MacOS X, the environ variable
-  // is unavailable. Apple's documentation (man environ) recommends using
-  // _NSGetEnviron() instead.
-  return *_NSGetEnviron();
-}
-#  else
-// Some POSIX platforms expect you to declare environ. extern "C" makes
-// it reside in the global namespace.
-extern "C" char** environ;
-inline char** GetEnviron() { return environ; }
-#  endif  // GTEST_OS_MAC
-
-#  if !GTEST_OS_QNX
-// The main function for a threadsafe-style death test child process.
-// This function is called in a clone()-ed process and thus must avoid
-// any potentially unsafe operations like malloc or libc functions.
-static int ExecDeathTestChildMain(void* child_arg) {
-  ExecDeathTestArgs* const args = static_cast<ExecDeathTestArgs*>(child_arg);
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(close(args->close_fd));
-
-  // We need to execute the test program in the same environment where
-  // it was originally invoked.  Therefore we change to the original
-  // working directory first.
-  const char* const original_dir =
-      UnitTest::GetInstance()->original_working_dir();
-  // We can safely call chdir() as it's a direct system call.
-  if (chdir(original_dir) != 0) {
-    DeathTestAbort(std::string("chdir(\"") + original_dir + "\") failed: " +
-                   GetLastErrnoDescription());
-    return EXIT_FAILURE;
-  }
-
-  // We can safely call execve() as it's a direct system call.  We
-  // cannot use execvp() as it's a libc function and thus potentially
-  // unsafe.  Since execve() doesn't search the PATH, the user must
-  // invoke the test program via a valid path that contains at least
-  // one path separator.
-  execve(args->argv[0], args->argv, GetEnviron());
-  DeathTestAbort(std::string("execve(") + args->argv[0] + ", ...) in " +
-                 original_dir + " failed: " +
-                 GetLastErrnoDescription());
-  return EXIT_FAILURE;
-}
-#  endif  // !GTEST_OS_QNX
-
-// Two utility routines that together determine the direction the stack
-// grows.
-// This could be accomplished more elegantly by a single recursive
-// function, but we want to guard against the unlikely possibility of
-// a smart compiler optimizing the recursion away.
-//
-// GTEST_NO_INLINE_ is required to prevent GCC 4.6 from inlining
-// StackLowerThanAddress into StackGrowsDown, which then doesn't give
-// correct answer.
-void StackLowerThanAddress(const void* ptr, bool* result) GTEST_NO_INLINE_;
-void StackLowerThanAddress(const void* ptr, bool* result) {
-  int dummy;
-  *result = (&dummy < ptr);
-}
-
-// Make sure AddressSanitizer does not tamper with the stack here.
-GTEST_ATTRIBUTE_NO_SANITIZE_ADDRESS_
-bool StackGrowsDown() {
-  int dummy;
-  bool result;
-  StackLowerThanAddress(&dummy, &result);
-  return result;
-}
-
-// Spawns a child process with the same executable as the current process in
-// a thread-safe manner and instructs it to run the death test.  The
-// implementation uses fork(2) + exec.  On systems where clone(2) is
-// available, it is used instead, being slightly more thread-safe.  On QNX,
-// fork supports only single-threaded environments, so this function uses
-// spawn(2) there instead.  The function dies with an error message if
-// anything goes wrong.
-static pid_t ExecDeathTestSpawnChild(char* const* argv, int close_fd) {
-  ExecDeathTestArgs args = { argv, close_fd };
-  pid_t child_pid = -1;
-
-#  if GTEST_OS_QNX
-  // Obtains the current directory and sets it to be closed in the child
-  // process.
-  const int cwd_fd = open(".", O_RDONLY);
-  GTEST_DEATH_TEST_CHECK_(cwd_fd != -1);
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(fcntl(cwd_fd, F_SETFD, FD_CLOEXEC));
-  // We need to execute the test program in the same environment where
-  // it was originally invoked.  Therefore we change to the original
-  // working directory first.
-  const char* const original_dir =
-      UnitTest::GetInstance()->original_working_dir();
-  // We can safely call chdir() as it's a direct system call.
-  if (chdir(original_dir) != 0) {
-    DeathTestAbort(std::string("chdir(\"") + original_dir + "\") failed: " +
-                   GetLastErrnoDescription());
-    return EXIT_FAILURE;
-  }
-
-  int fd_flags;
-  // Set close_fd to be closed after spawn.
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(fd_flags = fcntl(close_fd, F_GETFD));
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(fcntl(close_fd, F_SETFD,
-                                        fd_flags | FD_CLOEXEC));
-  struct inheritance inherit = {0};
-  // spawn is a system call.
-  child_pid = spawn(args.argv[0], 0, NULL, &inherit, args.argv, GetEnviron());
-  // Restores the current working directory.
-  GTEST_DEATH_TEST_CHECK_(fchdir(cwd_fd) != -1);
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(close(cwd_fd));
-
-#  else   // GTEST_OS_QNX
-#   if GTEST_OS_LINUX
-  // When a SIGPROF signal is received while fork() or clone() are executing,
-  // the process may hang. To avoid this, we ignore SIGPROF here and re-enable
-  // it after the call to fork()/clone() is complete.
-  struct sigaction saved_sigprof_action;
-  struct sigaction ignore_sigprof_action;
-  memset(&ignore_sigprof_action, 0, sizeof(ignore_sigprof_action));
-  sigemptyset(&ignore_sigprof_action.sa_mask);
-  ignore_sigprof_action.sa_handler = SIG_IGN;
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(sigaction(
-      SIGPROF, &ignore_sigprof_action, &saved_sigprof_action));
-#   endif  // GTEST_OS_LINUX
-
-#   if GTEST_HAS_CLONE
-  const bool use_fork = GTEST_FLAG(death_test_use_fork);
-
-  if (!use_fork) {
-    static const bool stack_grows_down = StackGrowsDown();
-    const size_t stack_size = getpagesize();
-    // MMAP_ANONYMOUS is not defined on Mac, so we use MAP_ANON instead.
-    void* const stack = mmap(NULL, stack_size, PROT_READ | PROT_WRITE,
-                             MAP_ANON | MAP_PRIVATE, -1, 0);
-    GTEST_DEATH_TEST_CHECK_(stack != MAP_FAILED);
-
-    // Maximum stack alignment in bytes:  For a downward-growing stack, this
-    // amount is subtracted from size of the stack space to get an address
-    // that is within the stack space and is aligned on all systems we care
-    // about.  As far as I know there is no ABI with stack alignment greater
-    // than 64.  We assume stack and stack_size already have alignment of
-    // kMaxStackAlignment.
-    const size_t kMaxStackAlignment = 64;
-    void* const stack_top =
-        static_cast<char*>(stack) +
-            (stack_grows_down ? stack_size - kMaxStackAlignment : 0);
-    GTEST_DEATH_TEST_CHECK_(stack_size > kMaxStackAlignment &&
-        reinterpret_cast<intptr_t>(stack_top) % kMaxStackAlignment == 0);
-
-    child_pid = clone(&ExecDeathTestChildMain, stack_top, SIGCHLD, &args);
-
-    GTEST_DEATH_TEST_CHECK_(munmap(stack, stack_size) != -1);
-  }
-#   else
-  const bool use_fork = true;
-#   endif  // GTEST_HAS_CLONE
-
-  if (use_fork && (child_pid = fork()) == 0) {
-      ExecDeathTestChildMain(&args);
-      _exit(0);
-  }
-#  endif  // GTEST_OS_QNX
-#  if GTEST_OS_LINUX
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(
-      sigaction(SIGPROF, &saved_sigprof_action, NULL));
-#  endif  // GTEST_OS_LINUX
-
-  GTEST_DEATH_TEST_CHECK_(child_pid != -1);
-  return child_pid;
-}
-
-// The AssumeRole process for a fork-and-exec death test.  It re-executes the
-// main program from the beginning, setting the --gtest_filter
-// and --gtest_internal_run_death_test flags to cause only the current
-// death test to be re-run.
-DeathTest::TestRole ExecDeathTest::AssumeRole() {
-  const UnitTestImpl* const impl = GetUnitTestImpl();
-  const InternalRunDeathTestFlag* const flag =
-      impl->internal_run_death_test_flag();
-  const TestInfo* const info = impl->current_test_info();
-  const int death_test_index = info->result()->death_test_count();
-
-  if (flag != NULL) {
-    set_write_fd(flag->write_fd());
-    return EXECUTE_TEST;
-  }
-
-  int pipe_fd[2];
-  GTEST_DEATH_TEST_CHECK_(pipe(pipe_fd) != -1);
-  // Clear the close-on-exec flag on the write end of the pipe, lest
-  // it be closed when the child process does an exec:
-  GTEST_DEATH_TEST_CHECK_(fcntl(pipe_fd[1], F_SETFD, 0) != -1);
-
-  const std::string filter_flag =
-      std::string("--") + GTEST_FLAG_PREFIX_ + kFilterFlag + "="
-      + info->test_case_name() + "." + info->name();
-  const std::string internal_flag =
-      std::string("--") + GTEST_FLAG_PREFIX_ + kInternalRunDeathTestFlag + "="
-      + file_ + "|" + StreamableToString(line_) + "|"
-      + StreamableToString(death_test_index) + "|"
-      + StreamableToString(pipe_fd[1]);
-  Arguments args;
-  args.AddArguments(GetArgvsForDeathTestChildProcess());
-  args.AddArgument(filter_flag.c_str());
-  args.AddArgument(internal_flag.c_str());
-
-  DeathTest::set_last_death_test_message("");
-
-  CaptureStderr();
-  // See the comment in NoExecDeathTest::AssumeRole for why the next line
-  // is necessary.
-  FlushInfoLog();
-
-  const pid_t child_pid = ExecDeathTestSpawnChild(args.Argv(), pipe_fd[0]);
-  GTEST_DEATH_TEST_CHECK_SYSCALL_(close(pipe_fd[1]));
-  set_child_pid(child_pid);
-  set_read_fd(pipe_fd[0]);
-  set_spawned(true);
-  return OVERSEE_TEST;
-}
-
-# endif  // !GTEST_OS_WINDOWS
-
-// Creates a concrete DeathTest-derived class that depends on the
-// --gtest_death_test_style flag, and sets the pointer pointed to
-// by the "test" argument to its address.  If the test should be
-// skipped, sets that pointer to NULL.  Returns true, unless the
-// flag is set to an invalid value.
-bool DefaultDeathTestFactory::Create(const char* statement, const RE* regex,
-                                     const char* file, int line,
-                                     DeathTest** test) {
-  UnitTestImpl* const impl = GetUnitTestImpl();
-  const InternalRunDeathTestFlag* const flag =
-      impl->internal_run_death_test_flag();
-  const int death_test_index = impl->current_test_info()
-      ->increment_death_test_count();
-
-  if (flag != NULL) {
-    if (death_test_index > flag->index()) {
-      DeathTest::set_last_death_test_message(
-          "Death test count (" + StreamableToString(death_test_index)
-          + ") somehow exceeded expected maximum ("
-          + StreamableToString(flag->index()) + ")");
-      return false;
-    }
-
-    if (!(flag->file() == file && flag->line() == line &&
-          flag->index() == death_test_index)) {
-      *test = NULL;
-      return true;
-    }
-  }
-
-# if GTEST_OS_WINDOWS
-
-  if (GTEST_FLAG(death_test_style) == "threadsafe" ||
-      GTEST_FLAG(death_test_style) == "fast") {
-    *test = new WindowsDeathTest(statement, regex, file, line);
-  }
-
-# else
-
-  if (GTEST_FLAG(death_test_style) == "threadsafe") {
-    *test = new ExecDeathTest(statement, regex, file, line);
-  } else if (GTEST_FLAG(death_test_style) == "fast") {
-    *test = new NoExecDeathTest(statement, regex);
-  }
-
-# endif  // GTEST_OS_WINDOWS
-
-  else {  // NOLINT - this is more readable than unbalanced brackets inside #if.
-    DeathTest::set_last_death_test_message(
-        "Unknown death test style \"" + GTEST_FLAG(death_test_style)
-        + "\" encountered");
-    return false;
-  }
-
-  return true;
-}
-
-# if GTEST_OS_WINDOWS
-// Recreates the pipe and event handles from the provided parameters,
-// signals the event, and returns a file descriptor wrapped around the pipe
-// handle. This function is called in the child process only.
-int GetStatusFileDescriptor(unsigned int parent_process_id,
-                            size_t write_handle_as_size_t,
-                            size_t event_handle_as_size_t) {
-  AutoHandle parent_process_handle(::OpenProcess(PROCESS_DUP_HANDLE,
-                                                   FALSE,  // Non-inheritable.
-                                                   parent_process_id));
-  if (parent_process_handle.Get() == INVALID_HANDLE_VALUE) {
-    DeathTestAbort("Unable to open parent process " +
-                   StreamableToString(parent_process_id));
-  }
-
-  // TODO(vladl@google.com): Replace the following check with a
-  // compile-time assertion when available.
-  GTEST_CHECK_(sizeof(HANDLE) <= sizeof(size_t));
-
-  const HANDLE write_handle =
-      reinterpret_cast<HANDLE>(write_handle_as_size_t);
-  HANDLE dup_write_handle;
-
-  // The newly initialized handle is accessible only in in the parent
-  // process. To obtain one accessible within the child, we need to use
-  // DuplicateHandle.
-  if (!::DuplicateHandle(parent_process_handle.Get(), write_handle,
-                         ::GetCurrentProcess(), &dup_write_handle,
-                         0x0,    // Requested privileges ignored since
-                                 // DUPLICATE_SAME_ACCESS is used.
-                         FALSE,  // Request non-inheritable handler.
-                         DUPLICATE_SAME_ACCESS)) {
-    DeathTestAbort("Unable to duplicate the pipe handle " +
-                   StreamableToString(write_handle_as_size_t) +
-                   " from the parent process " +
-                   StreamableToString(parent_process_id));
-  }
-
-  const HANDLE event_handle = reinterpret_cast<HANDLE>(event_handle_as_size_t);
-  HANDLE dup_event_handle;
-
-  if (!::DuplicateHandle(parent_process_handle.Get(), event_handle,
-                         ::GetCurrentProcess(), &dup_event_handle,
-                         0x0,
-                         FALSE,
-                         DUPLICATE_SAME_ACCESS)) {
-    DeathTestAbort("Unable to duplicate the event handle " +
-                   StreamableToString(event_handle_as_size_t) +
-                   " from the parent process " +
-                   StreamableToString(parent_process_id));
-  }
-
-  const int write_fd =
-      ::_open_osfhandle(reinterpret_cast<intptr_t>(dup_write_handle), O_APPEND);
-  if (write_fd == -1) {
-    DeathTestAbort("Unable to convert pipe handle " +
-                   StreamableToString(write_handle_as_size_t) +
-                   " to a file descriptor");
-  }
-
-  // Signals the parent that the write end of the pipe has been acquired
-  // so the parent can release its own write end.
-  ::SetEvent(dup_event_handle);
-
-  return write_fd;
-}
-# endif  // GTEST_OS_WINDOWS
-
-// Returns a newly created InternalRunDeathTestFlag object with fields
-// initialized from the GTEST_FLAG(internal_run_death_test) flag if
-// the flag is specified; otherwise returns NULL.
-InternalRunDeathTestFlag* ParseInternalRunDeathTestFlag() {
-  if (GTEST_FLAG(internal_run_death_test) == "") return NULL;
-
-  // GTEST_HAS_DEATH_TEST implies that we have ::std::string, so we
-  // can use it here.
-  int line = -1;
-  int index = -1;
-  ::std::vector< ::std::string> fields;
-  SplitString(GTEST_FLAG(internal_run_death_test).c_str(), '|', &fields);
-  int write_fd = -1;
-
-# if GTEST_OS_WINDOWS
-
-  unsigned int parent_process_id = 0;
-  size_t write_handle_as_size_t = 0;
-  size_t event_handle_as_size_t = 0;
-
-  if (fields.size() != 6
-      || !ParseNaturalNumber(fields[1], &line)
-      || !ParseNaturalNumber(fields[2], &index)
-      || !ParseNaturalNumber(fields[3], &parent_process_id)
-      || !ParseNaturalNumber(fields[4], &write_handle_as_size_t)
-      || !ParseNaturalNumber(fields[5], &event_handle_as_size_t)) {
-    DeathTestAbort("Bad --gtest_internal_run_death_test flag: " +
-                   GTEST_FLAG(internal_run_death_test));
-  }
-  write_fd = GetStatusFileDescriptor(parent_process_id,
-                                     write_handle_as_size_t,
-                                     event_handle_as_size_t);
-# else
-
-  if (fields.size() != 4
-      || !ParseNaturalNumber(fields[1], &line)
-      || !ParseNaturalNumber(fields[2], &index)
-      || !ParseNaturalNumber(fields[3], &write_fd)) {
-    DeathTestAbort("Bad --gtest_internal_run_death_test flag: "
-        + GTEST_FLAG(internal_run_death_test));
-  }
-
-# endif  // GTEST_OS_WINDOWS
-
-  return new InternalRunDeathTestFlag(fields[0], line, index, write_fd);
-}
-
-}  // namespace internal
-
-#endif  // GTEST_HAS_DEATH_TEST
-
-}  // namespace testing
-// Copyright 2008, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Authors: keith.ray@gmail.com (Keith Ray)
-
-
-#include <stdlib.h>
-
-#if GTEST_OS_WINDOWS_MOBILE
-# include <windows.h>
-#elif GTEST_OS_WINDOWS
-# include <direct.h>
-# include <io.h>
-#elif GTEST_OS_SYMBIAN
-// Symbian OpenC has PATH_MAX in sys/syslimits.h
-# include <sys/syslimits.h>
-#else
-# include <limits.h>
-# include <climits>  // Some Linux distributions define PATH_MAX here.
-#endif  // GTEST_OS_WINDOWS_MOBILE
-
-#if GTEST_OS_WINDOWS
-# define GTEST_PATH_MAX_ _MAX_PATH
-#elif defined(PATH_MAX)
-# define GTEST_PATH_MAX_ PATH_MAX
-#elif defined(_XOPEN_PATH_MAX)
-# define GTEST_PATH_MAX_ _XOPEN_PATH_MAX
-#else
-# define GTEST_PATH_MAX_ _POSIX_PATH_MAX
-#endif  // GTEST_OS_WINDOWS
-
-
-namespace testing {
-namespace internal {
-
-#if GTEST_OS_WINDOWS
-// On Windows, '\\' is the standard path separator, but many tools and the
-// Windows API also accept '/' as an alternate path separator. Unless otherwise
-// noted, a file path can contain either kind of path separators, or a mixture
-// of them.
-const char kPathSeparator = '\\';
-const char kAlternatePathSeparator = '/';
-const char kAlternatePathSeparatorString[] = "/";
-# if GTEST_OS_WINDOWS_MOBILE
-// Windows CE doesn't have a current directory. You should not use
-// the current directory in tests on Windows CE, but this at least
-// provides a reasonable fallback.
-const char kCurrentDirectoryString[] = "\\";
-// Windows CE doesn't define INVALID_FILE_ATTRIBUTES
-const DWORD kInvalidFileAttributes = 0xffffffff;
-# else
-const char kCurrentDirectoryString[] = ".\\";
-# endif  // GTEST_OS_WINDOWS_MOBILE
-#else
-const char kPathSeparator = '/';
-const char kCurrentDirectoryString[] = "./";
-#endif  // GTEST_OS_WINDOWS
-
-// Returns whether the given character is a valid path separator.
-static bool IsPathSeparator(char c) {
-#if GTEST_HAS_ALT_PATH_SEP_
-  return (c == kPathSeparator) || (c == kAlternatePathSeparator);
-#else
-  return c == kPathSeparator;
-#endif
-}
-
-// Returns the current working directory, or "" if unsuccessful.
-FilePath FilePath::GetCurrentDir() {
-#if GTEST_OS_WINDOWS_MOBILE || GTEST_OS_WINDOWS_PHONE || GTEST_OS_WINDOWS_RT
-  // Windows CE doesn't have a current directory, so we just return
-  // something reasonable.
-  return FilePath(kCurrentDirectoryString);
-#elif GTEST_OS_WINDOWS
-  char cwd[GTEST_PATH_MAX_ + 1] = { '\0' };
-  return FilePath(_getcwd(cwd, sizeof(cwd)) == NULL ? "" : cwd);
-#else
-  char cwd[GTEST_PATH_MAX_ + 1] = { '\0' };
-  char* result = getcwd(cwd, sizeof(cwd));
-# if GTEST_OS_NACL
-  // getcwd will likely fail in NaCl due to the sandbox, so return something
-  // reasonable. The user may have provided a shim implementation for getcwd,
-  // however, so fallback only when failure is detected.
-  return FilePath(result == NULL ? kCurrentDirectoryString : cwd);
-# endif  // GTEST_OS_NACL
-  return FilePath(result == NULL ? "" : cwd);
-#endif  // GTEST_OS_WINDOWS_MOBILE
-}
-
-// Returns a copy of the FilePath with the case-insensitive extension removed.
-// Example: FilePath("dir/file.exe").RemoveExtension("EXE") returns
-// FilePath("dir/file"). If a case-insensitive extension is not
-// found, returns a copy of the original FilePath.
-FilePath FilePath::RemoveExtension(const char* extension) const {
-  const std::string dot_extension = std::string(".") + extension;
-  if (String::EndsWithCaseInsensitive(pathname_, dot_extension)) {
-    return FilePath(pathname_.substr(
-        0, pathname_.length() - dot_extension.length()));
-  }
-  return *this;
-}
-
-// Returns a pointer to the last occurence of a valid path separator in
-// the FilePath. On Windows, for example, both '/' and '\' are valid path
-// separators. Returns NULL if no path separator was found.
-const char* FilePath::FindLastPathSeparator() const {
-  const char* const last_sep = strrchr(c_str(), kPathSeparator);
-#if GTEST_HAS_ALT_PATH_SEP_
-  const char* const last_alt_sep = strrchr(c_str(), kAlternatePathSeparator);
-  // Comparing two pointers of which only one is NULL is undefined.
-  if (last_alt_sep != NULL &&
-      (last_sep == NULL || last_alt_sep > last_sep)) {
-    return last_alt_sep;
-  }
-#endif
-  return last_sep;
-}
-
-// Returns a copy of the FilePath with the directory part removed.
-// Example: FilePath("path/to/file").RemoveDirectoryName() returns
-// FilePath("file"). If there is no directory part ("just_a_file"), it returns
-// the FilePath unmodified. If there is no file part ("just_a_dir/") it
-// returns an empty FilePath ("").
-// On Windows platform, '\' is the path separator, otherwise it is '/'.
-FilePath FilePath::RemoveDirectoryName() const {
-  const char* const last_sep = FindLastPathSeparator();
-  return last_sep ? FilePath(last_sep + 1) : *this;
-}
-
-// RemoveFileName returns the directory path with the filename removed.
-// Example: FilePath("path/to/file").RemoveFileName() returns "path/to/".
-// If the FilePath is "a_file" or "/a_file", RemoveFileName returns
-// FilePath("./") or, on Windows, FilePath(".\\"). If the filepath does
-// not have a file, like "just/a/dir/", it returns the FilePath unmodified.
-// On Windows platform, '\' is the path separator, otherwise it is '/'.
-FilePath FilePath::RemoveFileName() const {
-  const char* const last_sep = FindLastPathSeparator();
-  std::string dir;
-  if (last_sep) {
-    dir = std::string(c_str(), last_sep + 1 - c_str());
-  } else {
-    dir = kCurrentDirectoryString;
-  }
-  return FilePath(dir);
-}
-
-// Helper functions for naming files in a directory for xml output.
-
-// Given directory = "dir", base_name = "test", number = 0,
-// extension = "xml", returns "dir/test.xml". If number is greater
-// than zero (e.g., 12), returns "dir/test_12.xml".
-// On Windows platform, uses \ as the separator rather than /.
-FilePath FilePath::MakeFileName(const FilePath& directory,
-                                const FilePath& base_name,
-                                int number,
-                                const char* extension) {
-  std::string file;
-  if (number == 0) {
-    file = base_name.string() + "." + extension;
-  } else {
-    file = base_name.string() + "_" + StreamableToString(number)
-        + "." + extension;
-  }
-  return ConcatPaths(directory, FilePath(file));
-}
-
-// Given directory = "dir", relative_path = "test.xml", returns "dir/test.xml".
-// On Windows, uses \ as the separator rather than /.
-FilePath FilePath::ConcatPaths(const FilePath& directory,
-                               const FilePath& relative_path) {
-  if (directory.IsEmpty())
-    return relative_path;
-  const FilePath dir(directory.RemoveTrailingPathSeparator());
-  return FilePath(dir.string() + kPathSeparator + relative_path.string());
-}
-
-// Returns true if pathname describes something findable in the file-system,
-// either a file, directory, or whatever.
-bool FilePath::FileOrDirectoryExists() const {
-#if GTEST_OS_WINDOWS_MOBILE
-  LPCWSTR unicode = String::AnsiToUtf16(pathname_.c_str());
-  const DWORD attributes = GetFileAttributes(unicode);
-  delete [] unicode;
-  return attributes != kInvalidFileAttributes;
-#else
-  posix::StatStruct file_stat;
-  return posix::Stat(pathname_.c_str(), &file_stat) == 0;
-#endif  // GTEST_OS_WINDOWS_MOBILE
-}
-
-// Returns true if pathname describes a directory in the file-system
-// that exists.
-bool FilePath::DirectoryExists() const {
-  bool result = false;
-#if GTEST_OS_WINDOWS
-  // Don't strip off trailing separator if path is a root directory on
-  // Windows (like "C:\\").
-  const FilePath& path(IsRootDirectory() ? *this :
-                                           RemoveTrailingPathSeparator());
-#else
-  const FilePath& path(*this);
-#endif
-
-#if GTEST_OS_WINDOWS_MOBILE
-  LPCWSTR unicode = String::AnsiToUtf16(path.c_str());
-  const DWORD attributes = GetFileAttributes(unicode);
-  delete [] unicode;
-  if ((attributes != kInvalidFileAttributes) &&
-      (attributes & FILE_ATTRIBUTE_DIRECTORY)) {
-    result = true;
-  }
-#else
-  posix::StatStruct file_stat;
-  result = posix::Stat(path.c_str(), &file_stat) == 0 &&
-      posix::IsDir(file_stat);
-#endif  // GTEST_OS_WINDOWS_MOBILE
-
-  return result;
-}
-
-// Returns true if pathname describes a root directory. (Windows has one
-// root directory per disk drive.)
-bool FilePath::IsRootDirectory() const {
-#if GTEST_OS_WINDOWS
-  // TODO(wan@google.com): on Windows a network share like
-  // \\server\share can be a root directory, although it cannot be the
-  // current directory.  Handle this properly.
-  return pathname_.length() == 3 && IsAbsolutePath();
-#else
-  return pathname_.length() == 1 && IsPathSeparator(pathname_.c_str()[0]);
-#endif
-}
-
-// Returns true if pathname describes an absolute path.
-bool FilePath::IsAbsolutePath() const {
-  const char* const name = pathname_.c_str();
-#if GTEST_OS_WINDOWS
-  return pathname_.length() >= 3 &&
-     ((name[0] >= 'a' && name[0] <= 'z') ||
-      (name[0] >= 'A' && name[0] <= 'Z')) &&
-     name[1] == ':' &&
-     IsPathSeparator(name[2]);
-#else
-  return IsPathSeparator(name[0]);
-#endif
-}
-
-// Returns a pathname for a file that does not currently exist. The pathname
-// will be directory/base_name.extension or
-// directory/base_name_<number>.extension if directory/base_name.extension
-// already exists. The number will be incremented until a pathname is found
-// that does not already exist.
-// Examples: 'dir/foo_test.xml' or 'dir/foo_test_1.xml'.
-// There could be a race condition if two or more processes are calling this
-// function at the same time -- they could both pick the same filename.
-FilePath FilePath::GenerateUniqueFileName(const FilePath& directory,
-                                          const FilePath& base_name,
-                                          const char* extension) {
-  FilePath full_pathname;
-  int number = 0;
-  do {
-    full_pathname.Set(MakeFileName(directory, base_name, number++, extension));
-  } while (full_pathname.FileOrDirectoryExists());
-  return full_pathname;
-}
-
-// Returns true if FilePath ends with a path separator, which indicates that
-// it is intended to represent a directory. Returns false otherwise.
-// This does NOT check that a directory (or file) actually exists.
-bool FilePath::IsDirectory() const {
-  return !pathname_.empty() &&
-         IsPathSeparator(pathname_.c_str()[pathname_.length() - 1]);
-}
-
-// Create directories so that path exists. Returns true if successful or if
-// the directories already exist; returns false if unable to create directories
-// for any reason.
-bool FilePath::CreateDirectoriesRecursively() const {
-  if (!this->IsDirectory()) {
-    return false;
-  }
-
-  if (pathname_.length() == 0 || this->DirectoryExists()) {
-    return true;
-  }
-
-  const FilePath parent(this->RemoveTrailingPathSeparator().RemoveFileName());
-  return parent.CreateDirectoriesRecursively() && this->CreateFolder();
-}
-
-// Create the directory so that path exists. Returns true if successful or
-// if the directory already exists; returns false if unable to create the
-// directory for any reason, including if the parent directory does not
-// exist. Not named "CreateDirectory" because that's a macro on Windows.
-bool FilePath::CreateFolder() const {
-#if GTEST_OS_WINDOWS_MOBILE
-  FilePath removed_sep(this->RemoveTrailingPathSeparator());
-  LPCWSTR unicode = String::AnsiToUtf16(removed_sep.c_str());
-  int result = CreateDirectory(unicode, NULL) ? 0 : -1;
-  delete [] unicode;
-#elif GTEST_OS_WINDOWS
-  int result = _mkdir(pathname_.c_str());
-#else
-  int result = mkdir(pathname_.c_str(), 0777);
-#endif  // GTEST_OS_WINDOWS_MOBILE
-
-  if (result == -1) {
-    return this->DirectoryExists();  // An error is OK if the directory exists.
-  }
-  return true;  // No error.
-}
-
-// If input name has a trailing separator character, remove it and return the
-// name, otherwise return the name string unmodified.
-// On Windows platform, uses \ as the separator, other platforms use /.
-FilePath FilePath::RemoveTrailingPathSeparator() const {
-  return IsDirectory()
-      ? FilePath(pathname_.substr(0, pathname_.length() - 1))
-      : *this;
-}
-
-// Removes any redundant separators that might be in the pathname.
-// For example, "bar///foo" becomes "bar/foo". Does not eliminate other
-// redundancies that might be in a pathname involving "." or "..".
-// TODO(wan@google.com): handle Windows network shares (e.g. \\server\share).
-void FilePath::Normalize() {
-  if (pathname_.c_str() == NULL) {
-    pathname_ = "";
-    return;
-  }
-  const char* src = pathname_.c_str();
-  char* const dest = new char[pathname_.length() + 1];
-  char* dest_ptr = dest;
-  memset(dest_ptr, 0, pathname_.length() + 1);
-
-  while (*src != '\0') {
-    *dest_ptr = *src;
-    if (!IsPathSeparator(*src)) {
-      src++;
-    } else {
-#if GTEST_HAS_ALT_PATH_SEP_
-      if (*dest_ptr == kAlternatePathSeparator) {
-        *dest_ptr = kPathSeparator;
-      }
-#endif
-      while (IsPathSeparator(*src))
-        src++;
-    }
-    dest_ptr++;
-  }
-  *dest_ptr = '\0';
-  pathname_ = dest;
-  delete[] dest;
-}
-
-}  // namespace internal
-}  // namespace testing
-// Copyright 2008, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-
-
-#include <limits.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include <fstream>
-
-#if GTEST_OS_WINDOWS
-# include <windows.h>
-# include <io.h>
-# include <sys/stat.h>
-# include <map>  // Used in ThreadLocal.
-#else
-# include <unistd.h>
-#endif  // GTEST_OS_WINDOWS
-
-#if GTEST_OS_MAC
-# include <mach/mach_init.h>
-# include <mach/task.h>
-# include <mach/vm_map.h>
-#endif  // GTEST_OS_MAC
-
-#if GTEST_OS_QNX
-# include <devctl.h>
-# include <fcntl.h>
-# include <sys/procfs.h>
-#endif  // GTEST_OS_QNX
-
-#if GTEST_OS_AIX
-# include <procinfo.h>
-# include <sys/types.h>
-#endif  // GTEST_OS_AIX
-
-
-// Indicates that this translation unit is part of Google Test's
-// implementation.  It must come before gtest-internal-inl.h is
-// included, or there will be a compiler error.  This trick exists to
-// prevent the accidental inclusion of gtest-internal-inl.h in the
-// user's code.
-#define GTEST_IMPLEMENTATION_ 1
-#undef GTEST_IMPLEMENTATION_
-
-namespace testing {
-namespace internal {
-
-#if defined(_MSC_VER) || defined(__BORLANDC__)
-// MSVC and C++Builder do not provide a definition of STDERR_FILENO.
-const int kStdOutFileno = 1;
-const int kStdErrFileno = 2;
-#else
-const int kStdOutFileno = STDOUT_FILENO;
-const int kStdErrFileno = STDERR_FILENO;
-#endif  // _MSC_VER
-
-#if GTEST_OS_LINUX
-
-namespace {
-template <typename T>
-T ReadProcFileField(const string& filename, int field) {
-  std::string dummy;
-  std::ifstream file(filename.c_str());
-  while (field-- > 0) {
-    file >> dummy;
-  }
-  T output = 0;
-  file >> output;
-  return output;
-}
-}  // namespace
-
-// Returns the number of active threads, or 0 when there is an error.
-size_t GetThreadCount() {
-  const string filename =
-      (Message() << "/proc/" << getpid() << "/stat").GetString();
-  return ReadProcFileField<int>(filename, 19);
-}
-
-#elif GTEST_OS_MAC
-
-size_t GetThreadCount() {
-  const task_t task = mach_task_self();
-  mach_msg_type_number_t thread_count;
-  thread_act_array_t thread_list;
-  const kern_return_t status = task_threads(task, &thread_list, &thread_count);
-  if (status == KERN_SUCCESS) {
-    // task_threads allocates resources in thread_list and we need to free them
-    // to avoid leaks.
-    vm_deallocate(task,
-                  reinterpret_cast<vm_address_t>(thread_list),
-                  sizeof(thread_t) * thread_count);
-    return static_cast<size_t>(thread_count);
-  } else {
-    return 0;
-  }
-}
-
-#elif GTEST_OS_QNX
-
-// Returns the number of threads running in the process, or 0 to indicate that
-// we cannot detect it.
-size_t GetThreadCount() {
-  const int fd = open("/proc/self/as", O_RDONLY);
-  if (fd < 0) {
-    return 0;
-  }
-  procfs_info process_info;
-  const int status =
-      devctl(fd, DCMD_PROC_INFO, &process_info, sizeof(process_info), NULL);
-  close(fd);
-  if (status == EOK) {
-    return static_cast<size_t>(process_info.num_threads);
-  } else {
-    return 0;
-  }
-}
-
-#elif GTEST_OS_AIX
-
-size_t GetThreadCount() {
-  struct procentry64 entry;
-  pid_t pid = getpid();
-  int status = getprocs64(&entry, sizeof(entry), NULL, 0, &pid, 1);
-  if (status == 1) {
-    return entry.pi_thcount;
-  } else {
-    return 0;
-  }
-}
-
-#else
-
-size_t GetThreadCount() {
-  // There's no portable way to detect the number of threads, so we just
-  // return 0 to indicate that we cannot detect it.
-  return 0;
-}
-
-#endif  // GTEST_OS_LINUX
-
-#if GTEST_IS_THREADSAFE && GTEST_OS_WINDOWS
-
-void SleepMilliseconds(int n) {
-  ::Sleep(n);
-}
-
-AutoHandle::AutoHandle()
-    : handle_(INVALID_HANDLE_VALUE) {}
-
-AutoHandle::AutoHandle(Handle handle)
-    : handle_(handle) {}
-
-AutoHandle::~AutoHandle() {
-  Reset();
-}
-
-AutoHandle::Handle AutoHandle::Get() const {
-  return handle_;
-}
-
-void AutoHandle::Reset() {
-  Reset(INVALID_HANDLE_VALUE);
-}
-
-void AutoHandle::Reset(HANDLE handle) {
-  // Resetting with the same handle we already own is invalid.
-  if (handle_ != handle) {
-    if (IsCloseable()) {
-      ::CloseHandle(handle_);
-    }
-    handle_ = handle;
-  } else {
-    GTEST_CHECK_(!IsCloseable())
-        << "Resetting a valid handle to itself is likely a programmer error "
-            "and thus not allowed.";
-  }
-}
-
-bool AutoHandle::IsCloseable() const {
-  // Different Windows APIs may use either of these values to represent an
-  // invalid handle.
-  return handle_ != NULL && handle_ != INVALID_HANDLE_VALUE;
-}
-
-Notification::Notification()
-    : event_(::CreateEvent(NULL,   // Default security attributes.
-                           TRUE,   // Do not reset automatically.
-                           FALSE,  // Initially unset.
-                           NULL)) {  // Anonymous event.
-  GTEST_CHECK_(event_.Get() != NULL);
-}
-
-void Notification::Notify() {
-  GTEST_CHECK_(::SetEvent(event_.Get()) != FALSE);
-}
-
-void Notification::WaitForNotification() {
-  GTEST_CHECK_(
-      ::WaitForSingleObject(event_.Get(), INFINITE) == WAIT_OBJECT_0);
-}
-
-Mutex::Mutex()
-    : owner_thread_id_(0),
-      type_(kDynamic),
-      critical_section_init_phase_(0),
-      critical_section_(new CRITICAL_SECTION) {
-  ::InitializeCriticalSection(critical_section_);
-}
-
-Mutex::~Mutex() {
-  // Static mutexes are leaked intentionally. It is not thread-safe to try
-  // to clean them up.
-  // TODO(yukawa): Switch to Slim Reader/Writer (SRW) Locks, which requires
-  // nothing to clean it up but is available only on Vista and later.
-  // http://msdn.microsoft.com/en-us/library/windows/desktop/aa904937.aspx
-  if (type_ == kDynamic) {
-    ::DeleteCriticalSection(critical_section_);
-    delete critical_section_;
-    critical_section_ = NULL;
-  }
-}
-
-void Mutex::Lock() {
-  ThreadSafeLazyInit();
-  ::EnterCriticalSection(critical_section_);
-  owner_thread_id_ = ::GetCurrentThreadId();
-}
-
-void Mutex::Unlock() {
-  ThreadSafeLazyInit();
-  // We don't protect writing to owner_thread_id_ here, as it's the
-  // caller's responsibility to ensure that the current thread holds the
-  // mutex when this is called.
-  owner_thread_id_ = 0;
-  ::LeaveCriticalSection(critical_section_);
-}
-
-// Does nothing if the current thread holds the mutex. Otherwise, crashes
-// with high probability.
-void Mutex::AssertHeld() {
-  ThreadSafeLazyInit();
-  GTEST_CHECK_(owner_thread_id_ == ::GetCurrentThreadId())
-      << "The current thread is not holding the mutex @" << this;
-}
-
-// Initializes owner_thread_id_ and critical_section_ in static mutexes.
-void Mutex::ThreadSafeLazyInit() {
-  // Dynamic mutexes are initialized in the constructor.
-  if (type_ == kStatic) {
-    switch (
-        ::InterlockedCompareExchange(&critical_section_init_phase_, 1L, 0L)) {
-      case 0:
-        // If critical_section_init_phase_ was 0 before the exchange, we
-        // are the first to test it and need to perform the initialization.
-        owner_thread_id_ = 0;
-        critical_section_ = new CRITICAL_SECTION;
-        ::InitializeCriticalSection(critical_section_);
-        // Updates the critical_section_init_phase_ to 2 to signal
-        // initialization complete.
-        GTEST_CHECK_(::InterlockedCompareExchange(
-                          &critical_section_init_phase_, 2L, 1L) ==
-                      1L);
-        break;
-      case 1:
-        // Somebody else is already initializing the mutex; spin until they
-        // are done.
-        while (::InterlockedCompareExchange(&critical_section_init_phase_,
-                                            2L,
-                                            2L) != 2L) {
-          // Possibly yields the rest of the thread's time slice to other
-          // threads.
-          ::Sleep(0);
-        }
-        break;
-
-      case 2:
-        break;  // The mutex is already initialized and ready for use.
-
-      default:
-        GTEST_CHECK_(false)
-            << "Unexpected value of critical_section_init_phase_ "
-            << "while initializing a static mutex.";
-    }
-  }
-}
-
-namespace {
-
-class ThreadWithParamSupport : public ThreadWithParamBase {
- public:
-  static HANDLE CreateThread(Runnable* runnable,
-                             Notification* thread_can_start) {
-    ThreadMainParam* param = new ThreadMainParam(runnable, thread_can_start);
-    DWORD thread_id;
-    // TODO(yukawa): Consider to use _beginthreadex instead.
-    HANDLE thread_handle = ::CreateThread(
-        NULL,    // Default security.
-        0,       // Default stack size.
-        &ThreadWithParamSupport::ThreadMain,
-        param,   // Parameter to ThreadMainStatic
-        0x0,     // Default creation flags.
-        &thread_id);  // Need a valid pointer for the call to work under Win98.
-    GTEST_CHECK_(thread_handle != NULL) << "CreateThread failed with error "
-                                        << ::GetLastError() << ".";
-    if (thread_handle == NULL) {
-      delete param;
-    }
-    return thread_handle;
-  }
-
- private:
-  struct ThreadMainParam {
-    ThreadMainParam(Runnable* runnable, Notification* thread_can_start)
-        : runnable_(runnable),
-          thread_can_start_(thread_can_start) {
-    }
-    scoped_ptr<Runnable> runnable_;
-    // Does not own.
-    Notification* thread_can_start_;
-  };
-
-  static DWORD WINAPI ThreadMain(void* ptr) {
-    // Transfers ownership.
-    scoped_ptr<ThreadMainParam> param(static_cast<ThreadMainParam*>(ptr));
-    if (param->thread_can_start_ != NULL)
-      param->thread_can_start_->WaitForNotification();
-    param->runnable_->Run();
-    return 0;
-  }
-
-  // Prohibit instantiation.
-  ThreadWithParamSupport();
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ThreadWithParamSupport);
-};
-
-}  // namespace
-
-ThreadWithParamBase::ThreadWithParamBase(Runnable *runnable,
-                                         Notification* thread_can_start)
-      : thread_(ThreadWithParamSupport::CreateThread(runnable,
-                                                     thread_can_start)) {
-}
-
-ThreadWithParamBase::~ThreadWithParamBase() {
-  Join();
-}
-
-void ThreadWithParamBase::Join() {
-  GTEST_CHECK_(::WaitForSingleObject(thread_.Get(), INFINITE) == WAIT_OBJECT_0)
-      << "Failed to join the thread with error " << ::GetLastError() << ".";
-}
-
-// Maps a thread to a set of ThreadIdToThreadLocals that have values
-// instantiated on that thread and notifies them when the thread exits.  A
-// ThreadLocal instance is expected to persist until all threads it has
-// values on have terminated.
-class ThreadLocalRegistryImpl {
- public:
-  // Registers thread_local_instance as having value on the current thread.
-  // Returns a value that can be used to identify the thread from other threads.
-  static ThreadLocalValueHolderBase* GetValueOnCurrentThread(
-      const ThreadLocalBase* thread_local_instance) {
-    DWORD current_thread = ::GetCurrentThreadId();
-    MutexLock lock(&mutex_);
-    ThreadIdToThreadLocals* const thread_to_thread_locals =
-        GetThreadLocalsMapLocked();
-    ThreadIdToThreadLocals::iterator thread_local_pos =
-        thread_to_thread_locals->find(current_thread);
-    if (thread_local_pos == thread_to_thread_locals->end()) {
-      thread_local_pos = thread_to_thread_locals->insert(
-          std::make_pair(current_thread, ThreadLocalValues())).first;
-      StartWatcherThreadFor(current_thread);
-    }
-    ThreadLocalValues& thread_local_values = thread_local_pos->second;
-    ThreadLocalValues::iterator value_pos =
-        thread_local_values.find(thread_local_instance);
-    if (value_pos == thread_local_values.end()) {
-      value_pos =
-          thread_local_values
-              .insert(std::make_pair(
-                  thread_local_instance,
-                  linked_ptr<ThreadLocalValueHolderBase>(
-                      thread_local_instance->NewValueForCurrentThread())))
-              .first;
-    }
-    return value_pos->second.get();
-  }
-
-  static void OnThreadLocalDestroyed(
-      const ThreadLocalBase* thread_local_instance) {
-    std::vector<linked_ptr<ThreadLocalValueHolderBase> > value_holders;
-    // Clean up the ThreadLocalValues data structure while holding the lock, but
-    // defer the destruction of the ThreadLocalValueHolderBases.
-    {
-      MutexLock lock(&mutex_);
-      ThreadIdToThreadLocals* const thread_to_thread_locals =
-          GetThreadLocalsMapLocked();
-      for (ThreadIdToThreadLocals::iterator it =
-          thread_to_thread_locals->begin();
-          it != thread_to_thread_locals->end();
-          ++it) {
-        ThreadLocalValues& thread_local_values = it->second;
-        ThreadLocalValues::iterator value_pos =
-            thread_local_values.find(thread_local_instance);
-        if (value_pos != thread_local_values.end()) {
-          value_holders.push_back(value_pos->second);
-          thread_local_values.erase(value_pos);
-          // This 'if' can only be successful at most once, so theoretically we
-          // could break out of the loop here, but we don't bother doing so.
-        }
-      }
-    }
-    // Outside the lock, let the destructor for 'value_holders' deallocate the
-    // ThreadLocalValueHolderBases.
-  }
-
-  static void OnThreadExit(DWORD thread_id) {
-    GTEST_CHECK_(thread_id != 0) << ::GetLastError();
-    std::vector<linked_ptr<ThreadLocalValueHolderBase> > value_holders;
-    // Clean up the ThreadIdToThreadLocals data structure while holding the
-    // lock, but defer the destruction of the ThreadLocalValueHolderBases.
-    {
-      MutexLock lock(&mutex_);
-      ThreadIdToThreadLocals* const thread_to_thread_locals =
-          GetThreadLocalsMapLocked();
-      ThreadIdToThreadLocals::iterator thread_local_pos =
-          thread_to_thread_locals->find(thread_id);
-      if (thread_local_pos != thread_to_thread_locals->end()) {
-        ThreadLocalValues& thread_local_values = thread_local_pos->second;
-        for (ThreadLocalValues::iterator value_pos =
-            thread_local_values.begin();
-            value_pos != thread_local_values.end();
-            ++value_pos) {
-          value_holders.push_back(value_pos->second);
-        }
-        thread_to_thread_locals->erase(thread_local_pos);
-      }
-    }
-    // Outside the lock, let the destructor for 'value_holders' deallocate the
-    // ThreadLocalValueHolderBases.
-  }
-
- private:
-  // In a particular thread, maps a ThreadLocal object to its value.
-  typedef std::map<const ThreadLocalBase*,
-                   linked_ptr<ThreadLocalValueHolderBase> > ThreadLocalValues;
-  // Stores all ThreadIdToThreadLocals having values in a thread, indexed by
-  // thread's ID.
-  typedef std::map<DWORD, ThreadLocalValues> ThreadIdToThreadLocals;
-
-  // Holds the thread id and thread handle that we pass from
-  // StartWatcherThreadFor to WatcherThreadFunc.
-  typedef std::pair<DWORD, HANDLE> ThreadIdAndHandle;
-
-  static void StartWatcherThreadFor(DWORD thread_id) {
-    // The returned handle will be kept in thread_map and closed by
-    // watcher_thread in WatcherThreadFunc.
-    HANDLE thread = ::OpenThread(SYNCHRONIZE | THREAD_QUERY_INFORMATION,
-                                 FALSE,
-                                 thread_id);
-    GTEST_CHECK_(thread != NULL);
-    // We need to to pass a valid thread ID pointer into CreateThread for it
-    // to work correctly under Win98.
-    DWORD watcher_thread_id;
-    HANDLE watcher_thread = ::CreateThread(
-        NULL,   // Default security.
-        0,      // Default stack size
-        &ThreadLocalRegistryImpl::WatcherThreadFunc,
-        reinterpret_cast<LPVOID>(new ThreadIdAndHandle(thread_id, thread)),
-        CREATE_SUSPENDED,
-        &watcher_thread_id);
-    GTEST_CHECK_(watcher_thread != NULL);
-    // Give the watcher thread the same priority as ours to avoid being
-    // blocked by it.
-    ::SetThreadPriority(watcher_thread,
-                        ::GetThreadPriority(::GetCurrentThread()));
-    ::ResumeThread(watcher_thread);
-    ::CloseHandle(watcher_thread);
-  }
-
-  // Monitors exit from a given thread and notifies those
-  // ThreadIdToThreadLocals about thread termination.
-  static DWORD WINAPI WatcherThreadFunc(LPVOID param) {
-    const ThreadIdAndHandle* tah =
-        reinterpret_cast<const ThreadIdAndHandle*>(param);
-    GTEST_CHECK_(
-        ::WaitForSingleObject(tah->second, INFINITE) == WAIT_OBJECT_0);
-    OnThreadExit(tah->first);
-    ::CloseHandle(tah->second);
-    delete tah;
-    return 0;
-  }
-
-  // Returns map of thread local instances.
-  static ThreadIdToThreadLocals* GetThreadLocalsMapLocked() {
-    mutex_.AssertHeld();
-    static ThreadIdToThreadLocals* map = new ThreadIdToThreadLocals;
-    return map;
-  }
-
-  // Protects access to GetThreadLocalsMapLocked() and its return value.
-  static Mutex mutex_;
-  // Protects access to GetThreadMapLocked() and its return value.
-  static Mutex thread_map_mutex_;
-};
-
-Mutex ThreadLocalRegistryImpl::mutex_(Mutex::kStaticMutex);
-Mutex ThreadLocalRegistryImpl::thread_map_mutex_(Mutex::kStaticMutex);
-
-ThreadLocalValueHolderBase* ThreadLocalRegistry::GetValueOnCurrentThread(
-      const ThreadLocalBase* thread_local_instance) {
-  return ThreadLocalRegistryImpl::GetValueOnCurrentThread(
-      thread_local_instance);
-}
-
-void ThreadLocalRegistry::OnThreadLocalDestroyed(
-      const ThreadLocalBase* thread_local_instance) {
-  ThreadLocalRegistryImpl::OnThreadLocalDestroyed(thread_local_instance);
-}
-
-#endif  // GTEST_IS_THREADSAFE && GTEST_OS_WINDOWS
-
-#if GTEST_USES_POSIX_RE
-
-// Implements RE.  Currently only needed for death tests.
-
-RE::~RE() {
-  if (is_valid_) {
-    // regfree'ing an invalid regex might crash because the content
-    // of the regex is undefined. Since the regex's are essentially
-    // the same, one cannot be valid (or invalid) without the other
-    // being so too.
-    regfree(&partial_regex_);
-    regfree(&full_regex_);
-  }
-  free(const_cast<char*>(pattern_));
-}
-
-// Returns true iff regular expression re matches the entire str.
-bool RE::FullMatch(const char* str, const RE& re) {
-  if (!re.is_valid_) return false;
-
-  regmatch_t match;
-  return regexec(&re.full_regex_, str, 1, &match, 0) == 0;
-}
-
-// Returns true iff regular expression re matches a substring of str
-// (including str itself).
-bool RE::PartialMatch(const char* str, const RE& re) {
-  if (!re.is_valid_) return false;
-
-  regmatch_t match;
-  return regexec(&re.partial_regex_, str, 1, &match, 0) == 0;
-}
-
-// Initializes an RE from its string representation.
-void RE::Init(const char* regex) {
-  pattern_ = posix::StrDup(regex);
-
-  // Reserves enough bytes to hold the regular expression used for a
-  // full match.
-  const size_t full_regex_len = strlen(regex) + 10;
-  char* const full_pattern = new char[full_regex_len];
-
-  snprintf(full_pattern, full_regex_len, "^(%s)$", regex);
-  is_valid_ = regcomp(&full_regex_, full_pattern, REG_EXTENDED) == 0;
-  // We want to call regcomp(&partial_regex_, ...) even if the
-  // previous expression returns false.  Otherwise partial_regex_ may
-  // not be properly initialized can may cause trouble when it's
-  // freed.
-  //
-  // Some implementation of POSIX regex (e.g. on at least some
-  // versions of Cygwin) doesn't accept the empty string as a valid
-  // regex.  We change it to an equivalent form "()" to be safe.
-  if (is_valid_) {
-    const char* const partial_regex = (*regex == '\0') ? "()" : regex;
-    is_valid_ = regcomp(&partial_regex_, partial_regex, REG_EXTENDED) == 0;
-  }
-  EXPECT_TRUE(is_valid_)
-      << "Regular expression \"" << regex
-      << "\" is not a valid POSIX Extended regular expression.";
-
-  delete[] full_pattern;
-}
-
-#elif GTEST_USES_SIMPLE_RE
-
-// Returns true iff ch appears anywhere in str (excluding the
-// terminating '\0' character).
-bool IsInSet(char ch, const char* str) {
-  return ch != '\0' && strchr(str, ch) != NULL;
-}
-
-// Returns true iff ch belongs to the given classification.  Unlike
-// similar functions in <ctype.h>, these aren't affected by the
-// current locale.
-bool IsAsciiDigit(char ch) { return '0' <= ch && ch <= '9'; }
-bool IsAsciiPunct(char ch) {
-  return IsInSet(ch, "^-!\"#$%&'()*+,./:;<=>?@[\\]_`{|}~");
-}
-bool IsRepeat(char ch) { return IsInSet(ch, "?*+"); }
-bool IsAsciiWhiteSpace(char ch) { return IsInSet(ch, " \f\n\r\t\v"); }
-bool IsAsciiWordChar(char ch) {
-  return ('a' <= ch && ch <= 'z') || ('A' <= ch && ch <= 'Z') ||
-      ('0' <= ch && ch <= '9') || ch == '_';
-}
-
-// Returns true iff "\\c" is a supported escape sequence.
-bool IsValidEscape(char c) {
-  return (IsAsciiPunct(c) || IsInSet(c, "dDfnrsStvwW"));
-}
-
-// Returns true iff the given atom (specified by escaped and pattern)
-// matches ch.  The result is undefined if the atom is invalid.
-bool AtomMatchesChar(bool escaped, char pattern_char, char ch) {
-  if (escaped) {  // "\\p" where p is pattern_char.
-    switch (pattern_char) {
-      case 'd': return IsAsciiDigit(ch);
-      case 'D': return !IsAsciiDigit(ch);
-      case 'f': return ch == '\f';
-      case 'n': return ch == '\n';
-      case 'r': return ch == '\r';
-      case 's': return IsAsciiWhiteSpace(ch);
-      case 'S': return !IsAsciiWhiteSpace(ch);
-      case 't': return ch == '\t';
-      case 'v': return ch == '\v';
-      case 'w': return IsAsciiWordChar(ch);
-      case 'W': return !IsAsciiWordChar(ch);
-    }
-    return IsAsciiPunct(pattern_char) && pattern_char == ch;
-  }
-
-  return (pattern_char == '.' && ch != '\n') || pattern_char == ch;
-}
-
-// Helper function used by ValidateRegex() to format error messages.
-std::string FormatRegexSyntaxError(const char* regex, int index) {
-  return (Message() << "Syntax error at index " << index
-          << " in simple regular expression \"" << regex << "\": ").GetString();
-}
-
-// Generates non-fatal failures and returns false if regex is invalid;
-// otherwise returns true.
-bool ValidateRegex(const char* regex) {
-  if (regex == NULL) {
-    // TODO(wan@google.com): fix the source file location in the
-    // assertion failures to match where the regex is used in user
-    // code.
-    ADD_FAILURE() << "NULL is not a valid simple regular expression.";
-    return false;
-  }
-
-  bool is_valid = true;
-
-  // True iff ?, *, or + can follow the previous atom.
-  bool prev_repeatable = false;
-  for (int i = 0; regex[i]; i++) {
-    if (regex[i] == '\\') {  // An escape sequence
-      i++;
-      if (regex[i] == '\0') {
-        ADD_FAILURE() << FormatRegexSyntaxError(regex, i - 1)
-                      << "'\\' cannot appear at the end.";
-        return false;
-      }
-
-      if (!IsValidEscape(regex[i])) {
-        ADD_FAILURE() << FormatRegexSyntaxError(regex, i - 1)
-                      << "invalid escape sequence \"\\" << regex[i] << "\".";
-        is_valid = false;
-      }
-      prev_repeatable = true;
-    } else {  // Not an escape sequence.
-      const char ch = regex[i];
-
-      if (ch == '^' && i > 0) {
-        ADD_FAILURE() << FormatRegexSyntaxError(regex, i)
-                      << "'^' can only appear at the beginning.";
-        is_valid = false;
-      } else if (ch == '$' && regex[i + 1] != '\0') {
-        ADD_FAILURE() << FormatRegexSyntaxError(regex, i)
-                      << "'$' can only appear at the end.";
-        is_valid = false;
-      } else if (IsInSet(ch, "()[]{}|")) {
-        ADD_FAILURE() << FormatRegexSyntaxError(regex, i)
-                      << "'" << ch << "' is unsupported.";
-        is_valid = false;
-      } else if (IsRepeat(ch) && !prev_repeatable) {
-        ADD_FAILURE() << FormatRegexSyntaxError(regex, i)
-                      << "'" << ch << "' can only follow a repeatable token.";
-        is_valid = false;
-      }
-
-      prev_repeatable = !IsInSet(ch, "^$?*+");
-    }
-  }
-
-  return is_valid;
-}
-
-// Matches a repeated regex atom followed by a valid simple regular
-// expression.  The regex atom is defined as c if escaped is false,
-// or \c otherwise.  repeat is the repetition meta character (?, *,
-// or +).  The behavior is undefined if str contains too many
-// characters to be indexable by size_t, in which case the test will
-// probably time out anyway.  We are fine with this limitation as
-// std::string has it too.
-bool MatchRepetitionAndRegexAtHead(
-    bool escaped, char c, char repeat, const char* regex,
-    const char* str) {
-  const size_t min_count = (repeat == '+') ? 1 : 0;
-  const size_t max_count = (repeat == '?') ? 1 :
-      static_cast<size_t>(-1) - 1;
-  // We cannot call numeric_limits::max() as it conflicts with the
-  // max() macro on Windows.
-
-  for (size_t i = 0; i <= max_count; ++i) {
-    // We know that the atom matches each of the first i characters in str.
-    if (i >= min_count && MatchRegexAtHead(regex, str + i)) {
-      // We have enough matches at the head, and the tail matches too.
-      // Since we only care about *whether* the pattern matches str
-      // (as opposed to *how* it matches), there is no need to find a
-      // greedy match.
-      return true;
-    }
-    if (str[i] == '\0' || !AtomMatchesChar(escaped, c, str[i]))
-      return false;
-  }
-  return false;
-}
-
-// Returns true iff regex matches a prefix of str.  regex must be a
-// valid simple regular expression and not start with "^", or the
-// result is undefined.
-bool MatchRegexAtHead(const char* regex, const char* str) {
-  if (*regex == '\0')  // An empty regex matches a prefix of anything.
-    return true;
-
-  // "$" only matches the end of a string.  Note that regex being
-  // valid guarantees that there's nothing after "$" in it.
-  if (*regex == '$')
-    return *str == '\0';
-
-  // Is the first thing in regex an escape sequence?
-  const bool escaped = *regex == '\\';
-  if (escaped)
-    ++regex;
-  if (IsRepeat(regex[1])) {
-    // MatchRepetitionAndRegexAtHead() calls MatchRegexAtHead(), so
-    // here's an indirect recursion.  It terminates as the regex gets
-    // shorter in each recursion.
-    return MatchRepetitionAndRegexAtHead(
-        escaped, regex[0], regex[1], regex + 2, str);
-  } else {
-    // regex isn't empty, isn't "$", and doesn't start with a
-    // repetition.  We match the first atom of regex with the first
-    // character of str and recurse.
-    return (*str != '\0') && AtomMatchesChar(escaped, *regex, *str) &&
-        MatchRegexAtHead(regex + 1, str + 1);
-  }
-}
-
-// Returns true iff regex matches any substring of str.  regex must be
-// a valid simple regular expression, or the result is undefined.
-//
-// The algorithm is recursive, but the recursion depth doesn't exceed
-// the regex length, so we won't need to worry about running out of
-// stack space normally.  In rare cases the time complexity can be
-// exponential with respect to the regex length + the string length,
-// but usually it's must faster (often close to linear).
-bool MatchRegexAnywhere(const char* regex, const char* str) {
-  if (regex == NULL || str == NULL)
-    return false;
-
-  if (*regex == '^')
-    return MatchRegexAtHead(regex + 1, str);
-
-  // A successful match can be anywhere in str.
-  do {
-    if (MatchRegexAtHead(regex, str))
-      return true;
-  } while (*str++ != '\0');
-  return false;
-}
-
-// Implements the RE class.
-
-RE::~RE() {
-  free(const_cast<char*>(pattern_));
-  free(const_cast<char*>(full_pattern_));
-}
-
-// Returns true iff regular expression re matches the entire str.
-bool RE::FullMatch(const char* str, const RE& re) {
-  return re.is_valid_ && MatchRegexAnywhere(re.full_pattern_, str);
-}
-
-// Returns true iff regular expression re matches a substring of str
-// (including str itself).
-bool RE::PartialMatch(const char* str, const RE& re) {
-  return re.is_valid_ && MatchRegexAnywhere(re.pattern_, str);
-}
-
-// Initializes an RE from its string representation.
-void RE::Init(const char* regex) {
-  pattern_ = full_pattern_ = NULL;
-  if (regex != NULL) {
-    pattern_ = posix::StrDup(regex);
-  }
-
-  is_valid_ = ValidateRegex(regex);
-  if (!is_valid_) {
-    // No need to calculate the full pattern when the regex is invalid.
-    return;
-  }
-
-  const size_t len = strlen(regex);
-  // Reserves enough bytes to hold the regular expression used for a
-  // full match: we need space to prepend a '^', append a '$', and
-  // terminate the string with '\0'.
-  char* buffer = static_cast<char*>(malloc(len + 3));
-  full_pattern_ = buffer;
-
-  if (*regex != '^')
-    *buffer++ = '^';  // Makes sure full_pattern_ starts with '^'.
-
-  // We don't use snprintf or strncpy, as they trigger a warning when
-  // compiled with VC++ 8.0.
-  memcpy(buffer, regex, len);
-  buffer += len;
-
-  if (len == 0 || regex[len - 1] != '$')
-    *buffer++ = '$';  // Makes sure full_pattern_ ends with '$'.
-
-  *buffer = '\0';
-}
-
-#endif  // GTEST_USES_POSIX_RE
-
-const char kUnknownFile[] = "unknown file";
-
-// Formats a source file path and a line number as they would appear
-// in an error message from the compiler used to compile this code.
-GTEST_API_ ::std::string FormatFileLocation(const char* file, int line) {
-  const std::string file_name(file == NULL ? kUnknownFile : file);
-
-  if (line < 0) {
-    return file_name + ":";
-  }
-#ifdef _MSC_VER
-  return file_name + "(" + StreamableToString(line) + "):";
-#else
-  return file_name + ":" + StreamableToString(line) + ":";
-#endif  // _MSC_VER
-}
-
-// Formats a file location for compiler-independent XML output.
-// Although this function is not platform dependent, we put it next to
-// FormatFileLocation in order to contrast the two functions.
-// Note that FormatCompilerIndependentFileLocation() does NOT append colon
-// to the file location it produces, unlike FormatFileLocation().
-GTEST_API_ ::std::string FormatCompilerIndependentFileLocation(
-    const char* file, int line) {
-  const std::string file_name(file == NULL ? kUnknownFile : file);
-
-  if (line < 0)
-    return file_name;
-  else
-    return file_name + ":" + StreamableToString(line);
-}
-
-GTestLog::GTestLog(GTestLogSeverity severity, const char* file, int line)
-    : severity_(severity) {
-  const char* const marker =
-      severity == GTEST_INFO ?    "[  INFO ]" :
-      severity == GTEST_WARNING ? "[WARNING]" :
-      severity == GTEST_ERROR ?   "[ ERROR ]" : "[ FATAL ]";
-  GetStream() << ::std::endl << marker << " "
-              << FormatFileLocation(file, line).c_str() << ": ";
-}
-
-// Flushes the buffers and, if severity is GTEST_FATAL, aborts the program.
-GTestLog::~GTestLog() {
-  GetStream() << ::std::endl;
-  if (severity_ == GTEST_FATAL) {
-    fflush(stderr);
-    posix::Abort();
-  }
-}
-// Disable Microsoft deprecation warnings for POSIX functions called from
-// this class (creat, dup, dup2, and close)
-GTEST_DISABLE_MSC_WARNINGS_PUSH_(4996)
-
-#if GTEST_HAS_STREAM_REDIRECTION
-
-// Object that captures an output stream (stdout/stderr).
-class CapturedStream {
- public:
-  // The ctor redirects the stream to a temporary file.
-  explicit CapturedStream(int fd) : fd_(fd), uncaptured_fd_(dup(fd)) {
-# if GTEST_OS_WINDOWS
-    char temp_dir_path[MAX_PATH + 1] = { '\0' };  // NOLINT
-    char temp_file_path[MAX_PATH + 1] = { '\0' };  // NOLINT
-
-    ::GetTempPathA(sizeof(temp_dir_path), temp_dir_path);
-    const UINT success = ::GetTempFileNameA(temp_dir_path,
-                                            "gtest_redir",
-                                            0,  // Generate unique file name.
-                                            temp_file_path);
-    GTEST_CHECK_(success != 0)
-        << "Unable to create a temporary file in " << temp_dir_path;
-    const int captured_fd = creat(temp_file_path, _S_IREAD | _S_IWRITE);
-    GTEST_CHECK_(captured_fd != -1) << "Unable to open temporary file "
-                                    << temp_file_path;
-    filename_ = temp_file_path;
-# else
-    // There's no guarantee that a test has write access to the current
-    // directory, so we create the temporary file in the /tmp directory
-    // instead. We use /tmp on most systems, and /sdcard on Android.
-    // That's because Android doesn't have /tmp.
-#  if GTEST_OS_LINUX_ANDROID
-    // Note: Android applications are expected to call the framework's
-    // Context.getExternalStorageDirectory() method through JNI to get
-    // the location of the world-writable SD Card directory. However,
-    // this requires a Context handle, which cannot be retrieved
-    // globally from native code. Doing so also precludes running the
-    // code as part of a regular standalone executable, which doesn't
-    // run in a Dalvik process (e.g. when running it through 'adb shell').
-    //
-    // The location /sdcard is directly accessible from native code
-    // and is the only location (unofficially) supported by the Android
-    // team. It's generally a symlink to the real SD Card mount point
-    // which can be /mnt/sdcard, /mnt/sdcard0, /system/media/sdcard, or
-    // other OEM-customized locations. Never rely on these, and always
-    // use /sdcard.
-    char name_template[] = "/sdcard/gtest_captured_stream.XXXXXX";
-#  else
-    char name_template[] = "/tmp/captured_stream.XXXXXX";
-#  endif  // GTEST_OS_LINUX_ANDROID
-    const int captured_fd = mkstemp(name_template);
-    filename_ = name_template;
-# endif  // GTEST_OS_WINDOWS
-    fflush(NULL);
-    dup2(captured_fd, fd_);
-    close(captured_fd);
-  }
-
-  ~CapturedStream() {
-    remove(filename_.c_str());
-  }
-
-  std::string GetCapturedString() {
-    if (uncaptured_fd_ != -1) {
-      // Restores the original stream.
-      fflush(NULL);
-      dup2(uncaptured_fd_, fd_);
-      close(uncaptured_fd_);
-      uncaptured_fd_ = -1;
-    }
-
-    FILE* const file = posix::FOpen(filename_.c_str(), "r");
-    const std::string content = ReadEntireFile(file);
-    posix::FClose(file);
-    return content;
-  }
-
- private:
-  const int fd_;  // A stream to capture.
-  int uncaptured_fd_;
-  // Name of the temporary file holding the stderr output.
-  ::std::string filename_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(CapturedStream);
-};
-
-GTEST_DISABLE_MSC_WARNINGS_POP_()
-
-static CapturedStream* g_captured_stderr = NULL;
-static CapturedStream* g_captured_stdout = NULL;
-
-// Starts capturing an output stream (stdout/stderr).
-void CaptureStream(int fd, const char* stream_name, CapturedStream** stream) {
-  if (*stream != NULL) {
-    GTEST_LOG_(FATAL) << "Only one " << stream_name
-                      << " capturer can exist at a time.";
-  }
-  *stream = new CapturedStream(fd);
-}
-
-// Stops capturing the output stream and returns the captured string.
-std::string GetCapturedStream(CapturedStream** captured_stream) {
-  const std::string content = (*captured_stream)->GetCapturedString();
-
-  delete *captured_stream;
-  *captured_stream = NULL;
-
-  return content;
-}
-
-// Starts capturing stdout.
-void CaptureStdout() {
-  CaptureStream(kStdOutFileno, "stdout", &g_captured_stdout);
-}
-
-// Starts capturing stderr.
-void CaptureStderr() {
-  CaptureStream(kStdErrFileno, "stderr", &g_captured_stderr);
-}
-
-// Stops capturing stdout and returns the captured string.
-std::string GetCapturedStdout() {
-  return GetCapturedStream(&g_captured_stdout);
-}
-
-// Stops capturing stderr and returns the captured string.
-std::string GetCapturedStderr() {
-  return GetCapturedStream(&g_captured_stderr);
-}
-
-#endif  // GTEST_HAS_STREAM_REDIRECTION
-
-std::string TempDir() {
-#if GTEST_OS_WINDOWS_MOBILE
-  return "\\temp\\";
-#elif GTEST_OS_WINDOWS
-  const char* temp_dir = posix::GetEnv("TEMP");
-  if (temp_dir == NULL || temp_dir[0] == '\0')
-    return "\\temp\\";
-  else if (temp_dir[strlen(temp_dir) - 1] == '\\')
-    return temp_dir;
-  else
-    return std::string(temp_dir) + "\\";
-#elif GTEST_OS_LINUX_ANDROID
-  return "/sdcard/";
-#else
-  return "/tmp/";
-#endif  // GTEST_OS_WINDOWS_MOBILE
-}
-
-size_t GetFileSize(FILE* file) {
-  fseek(file, 0, SEEK_END);
-  return static_cast<size_t>(ftell(file));
-}
-
-std::string ReadEntireFile(FILE* file) {
-  const size_t file_size = GetFileSize(file);
-  char* const buffer = new char[file_size];
-
-  size_t bytes_last_read = 0;  // # of bytes read in the last fread()
-  size_t bytes_read = 0;       // # of bytes read so far
-
-  fseek(file, 0, SEEK_SET);
-
-  // Keeps reading the file until we cannot read further or the
-  // pre-determined file size is reached.
-  do {
-    bytes_last_read = fread(buffer+bytes_read, 1, file_size-bytes_read, file);
-    bytes_read += bytes_last_read;
-  } while (bytes_last_read > 0 && bytes_read < file_size);
-
-  const std::string content(buffer, bytes_read);
-  delete[] buffer;
-
-  return content;
-}
-
-#if GTEST_HAS_DEATH_TEST
-
-static const ::std::vector<testing::internal::string>* g_injected_test_argvs =
-                                        NULL;  // Owned.
-
-void SetInjectableArgvs(const ::std::vector<testing::internal::string>* argvs) {
-  if (g_injected_test_argvs != argvs)
-    delete g_injected_test_argvs;
-  g_injected_test_argvs = argvs;
-}
-
-const ::std::vector<testing::internal::string>& GetInjectableArgvs() {
-  if (g_injected_test_argvs != NULL) {
-    return *g_injected_test_argvs;
-  }
-  return GetArgvs();
-}
-#endif  // GTEST_HAS_DEATH_TEST
-
-#if GTEST_OS_WINDOWS_MOBILE
-namespace posix {
-void Abort() {
-  DebugBreak();
-  TerminateProcess(GetCurrentProcess(), 1);
-}
-}  // namespace posix
-#endif  // GTEST_OS_WINDOWS_MOBILE
-
-// Returns the name of the environment variable corresponding to the
-// given flag.  For example, FlagToEnvVar("foo") will return
-// "GTEST_FOO" in the open-source version.
-static std::string FlagToEnvVar(const char* flag) {
-  const std::string full_flag =
-      (Message() << GTEST_FLAG_PREFIX_ << flag).GetString();
-
-  Message env_var;
-  for (size_t i = 0; i != full_flag.length(); i++) {
-    env_var << ToUpper(full_flag.c_str()[i]);
-  }
-
-  return env_var.GetString();
-}
-
-// Parses 'str' for a 32-bit signed integer.  If successful, writes
-// the result to *value and returns true; otherwise leaves *value
-// unchanged and returns false.
-bool ParseInt32(const Message& src_text, const char* str, Int32* value) {
-  // Parses the environment variable as a decimal integer.
-  char* end = NULL;
-  const long long_value = strtol(str, &end, 10);  // NOLINT
-
-  // Has strtol() consumed all characters in the string?
-  if (*end != '\0') {
-    // No - an invalid character was encountered.
-    Message msg;
-    msg << "WARNING: " << src_text
-        << " is expected to be a 32-bit integer, but actually"
-        << " has value \"" << str << "\".\n";
-    printf("%s", msg.GetString().c_str());
-    fflush(stdout);
-    return false;
-  }
-
-  // Is the parsed value in the range of an Int32?
-  const Int32 result = static_cast<Int32>(long_value);
-  if (long_value == LONG_MAX || long_value == LONG_MIN ||
-      // The parsed value overflows as a long.  (strtol() returns
-      // LONG_MAX or LONG_MIN when the input overflows.)
-      result != long_value
-      // The parsed value overflows as an Int32.
-      ) {
-    Message msg;
-    msg << "WARNING: " << src_text
-        << " is expected to be a 32-bit integer, but actually"
-        << " has value " << str << ", which overflows.\n";
-    printf("%s", msg.GetString().c_str());
-    fflush(stdout);
-    return false;
-  }
-
-  *value = result;
-  return true;
-}
-
-// Reads and returns the Boolean environment variable corresponding to
-// the given flag; if it's not set, returns default_value.
-//
-// The value is considered true iff it's not "0".
-bool BoolFromGTestEnv(const char* flag, bool default_value) {
-#if defined(GTEST_GET_BOOL_FROM_ENV_)
-  return GTEST_GET_BOOL_FROM_ENV_(flag, default_value);
-#endif  // defined(GTEST_GET_BOOL_FROM_ENV_)
-  const std::string env_var = FlagToEnvVar(flag);
-  const char* const string_value = posix::GetEnv(env_var.c_str());
-  return string_value == NULL ?
-      default_value : strcmp(string_value, "0") != 0;
-}
-
-// Reads and returns a 32-bit integer stored in the environment
-// variable corresponding to the given flag; if it isn't set or
-// doesn't represent a valid 32-bit integer, returns default_value.
-Int32 Int32FromGTestEnv(const char* flag, Int32 default_value) {
-#if defined(GTEST_GET_INT32_FROM_ENV_)
-  return GTEST_GET_INT32_FROM_ENV_(flag, default_value);
-#endif  // defined(GTEST_GET_INT32_FROM_ENV_)
-  const std::string env_var = FlagToEnvVar(flag);
-  const char* const string_value = posix::GetEnv(env_var.c_str());
-  if (string_value == NULL) {
-    // The environment variable is not set.
-    return default_value;
-  }
-
-  Int32 result = default_value;
-  if (!ParseInt32(Message() << "Environment variable " << env_var,
-                  string_value, &result)) {
-    printf("The default value %s is used.\n",
-           (Message() << default_value).GetString().c_str());
-    fflush(stdout);
-    return default_value;
-  }
-
-  return result;
-}
-
-// Reads and returns the string environment variable corresponding to
-// the given flag; if it's not set, returns default_value.
-std::string StringFromGTestEnv(const char* flag, const char* default_value) {
-#if defined(GTEST_GET_STRING_FROM_ENV_)
-  return GTEST_GET_STRING_FROM_ENV_(flag, default_value);
-#endif  // defined(GTEST_GET_STRING_FROM_ENV_)
-  const std::string env_var = FlagToEnvVar(flag);
-  const char* value = posix::GetEnv(env_var.c_str());
-  if (value != NULL) {
-    return value;
-  }
-
-  // As a special case for the 'output' flag, if GTEST_OUTPUT is not
-  // set, we look for XML_OUTPUT_FILE, which is set by the Bazel build
-  // system.  The value of XML_OUTPUT_FILE is a filename without the
-  // "xml:" prefix of GTEST_OUTPUT.
-  //
-  // The net priority order after flag processing is thus:
-  //   --gtest_output command line flag
-  //   GTEST_OUTPUT environment variable
-  //   XML_OUTPUT_FILE environment variable
-  //   'default_value'
-  if (strcmp(flag, "output") == 0) {
-    value = posix::GetEnv("XML_OUTPUT_FILE");
-    if (value != NULL) {
-      return std::string("xml:") + value;
-    }
-  }
-  return default_value;
-}
-
-}  // namespace internal
-}  // namespace testing
-// Copyright 2007, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-
-// Google Test - The Google C++ Testing Framework
-//
-// This file implements a universal value printer that can print a
-// value of any type T:
-//
-//   void ::testing::internal::UniversalPrinter<T>::Print(value, ostream_ptr);
-//
-// It uses the << operator when possible, and prints the bytes in the
-// object otherwise.  A user can override its behavior for a class
-// type Foo by defining either operator<<(::std::ostream&, const Foo&)
-// or void PrintTo(const Foo&, ::std::ostream*) in the namespace that
-// defines Foo.
-
-#include <ctype.h>
-#include <stdio.h>
-#include <cwchar>
-#include <ostream>  // NOLINT
-#include <string>
-
-namespace testing {
-
-namespace {
-
-using ::std::ostream;
-
-// Prints a segment of bytes in the given object.
-GTEST_ATTRIBUTE_NO_SANITIZE_MEMORY_
-GTEST_ATTRIBUTE_NO_SANITIZE_ADDRESS_
-GTEST_ATTRIBUTE_NO_SANITIZE_THREAD_
-void PrintByteSegmentInObjectTo(const unsigned char* obj_bytes, size_t start,
-                                size_t count, ostream* os) {
-  char text[5] = "";
-  for (size_t i = 0; i != count; i++) {
-    const size_t j = start + i;
-    if (i != 0) {
-      // Organizes the bytes into groups of 2 for easy parsing by
-      // human.
-      if ((j % 2) == 0)
-        *os << ' ';
-      else
-        *os << '-';
-    }
-    GTEST_SNPRINTF_(text, sizeof(text), "%02X", obj_bytes[j]);
-    *os << text;
-  }
-}
-
-// Prints the bytes in the given value to the given ostream.
-void PrintBytesInObjectToImpl(const unsigned char* obj_bytes, size_t count,
-                              ostream* os) {
-  // Tells the user how big the object is.
-  *os << count << "-byte object <";
-
-  const size_t kThreshold = 132;
-  const size_t kChunkSize = 64;
-  // If the object size is bigger than kThreshold, we'll have to omit
-  // some details by printing only the first and the last kChunkSize
-  // bytes.
-  // TODO(wan): let the user control the threshold using a flag.
-  if (count < kThreshold) {
-    PrintByteSegmentInObjectTo(obj_bytes, 0, count, os);
-  } else {
-    PrintByteSegmentInObjectTo(obj_bytes, 0, kChunkSize, os);
-    *os << " ... ";
-    // Rounds up to 2-byte boundary.
-    const size_t resume_pos = (count - kChunkSize + 1)/2*2;
-    PrintByteSegmentInObjectTo(obj_bytes, resume_pos, count - resume_pos, os);
-  }
-  *os << ">";
-}
-
-}  // namespace
-
-namespace internal2 {
-
-// Delegates to PrintBytesInObjectToImpl() to print the bytes in the
-// given object.  The delegation simplifies the implementation, which
-// uses the << operator and thus is easier done outside of the
-// ::testing::internal namespace, which contains a << operator that
-// sometimes conflicts with the one in STL.
-void PrintBytesInObjectTo(const unsigned char* obj_bytes, size_t count,
-                          ostream* os) {
-  PrintBytesInObjectToImpl(obj_bytes, count, os);
-}
-
-}  // namespace internal2
-
-namespace internal {
-
-// Depending on the value of a char (or wchar_t), we print it in one
-// of three formats:
-//   - as is if it's a printable ASCII (e.g. 'a', '2', ' '),
-//   - as a hexidecimal escape sequence (e.g. '\x7F'), or
-//   - as a special escape sequence (e.g. '\r', '\n').
-enum CharFormat {
-  kAsIs,
-  kHexEscape,
-  kSpecialEscape
-};
-
-// Returns true if c is a printable ASCII character.  We test the
-// value of c directly instead of calling isprint(), which is buggy on
-// Windows Mobile.
-inline bool IsPrintableAscii(wchar_t c) {
-  return 0x20 <= c && c <= 0x7E;
-}
-
-// Prints a wide or narrow char c as a character literal without the
-// quotes, escaping it when necessary; returns how c was formatted.
-// The template argument UnsignedChar is the unsigned version of Char,
-// which is the type of c.
-template <typename UnsignedChar, typename Char>
-static CharFormat PrintAsCharLiteralTo(Char c, ostream* os) {
-  switch (static_cast<wchar_t>(c)) {
-    case L'\0':
-      *os << "\\0";
-      break;
-    case L'\'':
-      *os << "\\'";
-      break;
-    case L'\\':
-      *os << "\\\\";
-      break;
-    case L'\a':
-      *os << "\\a";
-      break;
-    case L'\b':
-      *os << "\\b";
-      break;
-    case L'\f':
-      *os << "\\f";
-      break;
-    case L'\n':
-      *os << "\\n";
-      break;
-    case L'\r':
-      *os << "\\r";
-      break;
-    case L'\t':
-      *os << "\\t";
-      break;
-    case L'\v':
-      *os << "\\v";
-      break;
-    default:
-      if (IsPrintableAscii(c)) {
-        *os << static_cast<char>(c);
-        return kAsIs;
-      } else {
-        *os << "\\x" + String::FormatHexInt(static_cast<UnsignedChar>(c));
-        return kHexEscape;
-      }
-  }
-  return kSpecialEscape;
-}
-
-// Prints a wchar_t c as if it's part of a string literal, escaping it when
-// necessary; returns how c was formatted.
-static CharFormat PrintAsStringLiteralTo(wchar_t c, ostream* os) {
-  switch (c) {
-    case L'\'':
-      *os << "'";
-      return kAsIs;
-    case L'"':
-      *os << "\\\"";
-      return kSpecialEscape;
-    default:
-      return PrintAsCharLiteralTo<wchar_t>(c, os);
-  }
-}
-
-// Prints a char c as if it's part of a string literal, escaping it when
-// necessary; returns how c was formatted.
-static CharFormat PrintAsStringLiteralTo(char c, ostream* os) {
-  return PrintAsStringLiteralTo(
-      static_cast<wchar_t>(static_cast<unsigned char>(c)), os);
-}
-
-// Prints a wide or narrow character c and its code.  '\0' is printed
-// as "'\\0'", other unprintable characters are also properly escaped
-// using the standard C++ escape sequence.  The template argument
-// UnsignedChar is the unsigned version of Char, which is the type of c.
-template <typename UnsignedChar, typename Char>
-void PrintCharAndCodeTo(Char c, ostream* os) {
-  // First, print c as a literal in the most readable form we can find.
-  *os << ((sizeof(c) > 1) ? "L'" : "'");
-  const CharFormat format = PrintAsCharLiteralTo<UnsignedChar>(c, os);
-  *os << "'";
-
-  // To aid user debugging, we also print c's code in decimal, unless
-  // it's 0 (in which case c was printed as '\\0', making the code
-  // obvious).
-  if (c == 0)
-    return;
-  *os << " (" << static_cast<int>(c);
-
-  // For more convenience, we print c's code again in hexidecimal,
-  // unless c was already printed in the form '\x##' or the code is in
-  // [1, 9].
-  if (format == kHexEscape || (1 <= c && c <= 9)) {
-    // Do nothing.
-  } else {
-    *os << ", 0x" << String::FormatHexInt(static_cast<UnsignedChar>(c));
-  }
-  *os << ")";
-}
-
-void PrintTo(unsigned char c, ::std::ostream* os) {
-  PrintCharAndCodeTo<unsigned char>(c, os);
-}
-void PrintTo(signed char c, ::std::ostream* os) {
-  PrintCharAndCodeTo<unsigned char>(c, os);
-}
-
-// Prints a wchar_t as a symbol if it is printable or as its internal
-// code otherwise and also as its code.  L'\0' is printed as "L'\\0'".
-void PrintTo(wchar_t wc, ostream* os) {
-  PrintCharAndCodeTo<wchar_t>(wc, os);
-}
-
-// Prints the given array of characters to the ostream.  CharType must be either
-// char or wchar_t.
-// The array starts at begin, the length is len, it may include '\0' characters
-// and may not be NUL-terminated.
-template <typename CharType>
-GTEST_ATTRIBUTE_NO_SANITIZE_MEMORY_
-GTEST_ATTRIBUTE_NO_SANITIZE_ADDRESS_
-GTEST_ATTRIBUTE_NO_SANITIZE_THREAD_
-static void PrintCharsAsStringTo(
-    const CharType* begin, size_t len, ostream* os) {
-  const char* const kQuoteBegin = sizeof(CharType) == 1 ? "\"" : "L\"";
-  *os << kQuoteBegin;
-  bool is_previous_hex = false;
-  for (size_t index = 0; index < len; ++index) {
-    const CharType cur = begin[index];
-    if (is_previous_hex && IsXDigit(cur)) {
-      // Previous character is of '\x..' form and this character can be
-      // interpreted as another hexadecimal digit in its number. Break string to
-      // disambiguate.
-      *os << "\" " << kQuoteBegin;
-    }
-    is_previous_hex = PrintAsStringLiteralTo(cur, os) == kHexEscape;
-  }
-  *os << "\"";
-}
-
-// Prints a (const) char/wchar_t array of 'len' elements, starting at address
-// 'begin'.  CharType must be either char or wchar_t.
-template <typename CharType>
-GTEST_ATTRIBUTE_NO_SANITIZE_MEMORY_
-GTEST_ATTRIBUTE_NO_SANITIZE_ADDRESS_
-GTEST_ATTRIBUTE_NO_SANITIZE_THREAD_
-static void UniversalPrintCharArray(
-    const CharType* begin, size_t len, ostream* os) {
-  // The code
-  //   const char kFoo[] = "foo";
-  // generates an array of 4, not 3, elements, with the last one being '\0'.
-  //
-  // Therefore when printing a char array, we don't print the last element if
-  // it's '\0', such that the output matches the string literal as it's
-  // written in the source code.
-  if (len > 0 && begin[len - 1] == '\0') {
-    PrintCharsAsStringTo(begin, len - 1, os);
-    return;
-  }
-
-  // If, however, the last element in the array is not '\0', e.g.
-  //    const char kFoo[] = { 'f', 'o', 'o' };
-  // we must print the entire array.  We also print a message to indicate
-  // that the array is not NUL-terminated.
-  PrintCharsAsStringTo(begin, len, os);
-  *os << " (no terminating NUL)";
-}
-
-// Prints a (const) char array of 'len' elements, starting at address 'begin'.
-void UniversalPrintArray(const char* begin, size_t len, ostream* os) {
-  UniversalPrintCharArray(begin, len, os);
-}
-
-// Prints a (const) wchar_t array of 'len' elements, starting at address
-// 'begin'.
-void UniversalPrintArray(const wchar_t* begin, size_t len, ostream* os) {
-  UniversalPrintCharArray(begin, len, os);
-}
-
-// Prints the given C string to the ostream.
-void PrintTo(const char* s, ostream* os) {
-  if (s == NULL) {
-    *os << "NULL";
-  } else {
-    *os << ImplicitCast_<const void*>(s) << " pointing to ";
-    PrintCharsAsStringTo(s, strlen(s), os);
-  }
-}
-
-// MSVC compiler can be configured to define whar_t as a typedef
-// of unsigned short. Defining an overload for const wchar_t* in that case
-// would cause pointers to unsigned shorts be printed as wide strings,
-// possibly accessing more memory than intended and causing invalid
-// memory accesses. MSVC defines _NATIVE_WCHAR_T_DEFINED symbol when
-// wchar_t is implemented as a native type.
-#if !defined(_MSC_VER) || defined(_NATIVE_WCHAR_T_DEFINED)
-// Prints the given wide C string to the ostream.
-void PrintTo(const wchar_t* s, ostream* os) {
-  if (s == NULL) {
-    *os << "NULL";
-  } else {
-    *os << ImplicitCast_<const void*>(s) << " pointing to ";
-    PrintCharsAsStringTo(s, std::wcslen(s), os);
-  }
-}
-#endif  // wchar_t is native
-
-// Prints a ::string object.
-#if GTEST_HAS_GLOBAL_STRING
-void PrintStringTo(const ::string& s, ostream* os) {
-  PrintCharsAsStringTo(s.data(), s.size(), os);
-}
-#endif  // GTEST_HAS_GLOBAL_STRING
-
-void PrintStringTo(const ::std::string& s, ostream* os) {
-  PrintCharsAsStringTo(s.data(), s.size(), os);
-}
-
-// Prints a ::wstring object.
-#if GTEST_HAS_GLOBAL_WSTRING
-void PrintWideStringTo(const ::wstring& s, ostream* os) {
-  PrintCharsAsStringTo(s.data(), s.size(), os);
-}
-#endif  // GTEST_HAS_GLOBAL_WSTRING
-
-#if GTEST_HAS_STD_WSTRING
-void PrintWideStringTo(const ::std::wstring& s, ostream* os) {
-  PrintCharsAsStringTo(s.data(), s.size(), os);
-}
-#endif  // GTEST_HAS_STD_WSTRING
-
-}  // namespace internal
-
-}  // namespace testing
-// Copyright 2008, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: mheule@google.com (Markus Heule)
-//
-// The Google C++ Testing Framework (Google Test)
-
-
-// Indicates that this translation unit is part of Google Test's
-// implementation.  It must come before gtest-internal-inl.h is
-// included, or there will be a compiler error.  This trick exists to
-// prevent the accidental inclusion of gtest-internal-inl.h in the
-// user's code.
-#define GTEST_IMPLEMENTATION_ 1
-#undef GTEST_IMPLEMENTATION_
-
-namespace testing {
-
-using internal::GetUnitTestImpl;
-
-// Gets the summary of the failure message by omitting the stack trace
-// in it.
-std::string TestPartResult::ExtractSummary(const char* message) {
-  const char* const stack_trace = strstr(message, internal::kStackTraceMarker);
-  return stack_trace == NULL ? message :
-      std::string(message, stack_trace);
-}
-
-// Prints a TestPartResult object.
-std::ostream& operator<<(std::ostream& os, const TestPartResult& result) {
-  return os
-      << result.file_name() << ":" << result.line_number() << ": "
-      << (result.type() == TestPartResult::kSuccess ? "Success" :
-          result.type() == TestPartResult::kFatalFailure ? "Fatal failure" :
-          "Non-fatal failure") << ":\n"
-      << result.message() << std::endl;
-}
-
-// Appends a TestPartResult to the array.
-void TestPartResultArray::Append(const TestPartResult& result) {
-  array_.push_back(result);
-}
-
-// Returns the TestPartResult at the given index (0-based).
-const TestPartResult& TestPartResultArray::GetTestPartResult(int index) const {
-  if (index < 0 || index >= size()) {
-    printf("\nInvalid index (%d) into TestPartResultArray.\n", index);
-    internal::posix::Abort();
-  }
-
-  return array_[index];
-}
-
-// Returns the number of TestPartResult objects in the array.
-int TestPartResultArray::size() const {
-  return static_cast<int>(array_.size());
-}
-
-namespace internal {
-
-HasNewFatalFailureHelper::HasNewFatalFailureHelper()
-    : has_new_fatal_failure_(false),
-      original_reporter_(GetUnitTestImpl()->
-                         GetTestPartResultReporterForCurrentThread()) {
-  GetUnitTestImpl()->SetTestPartResultReporterForCurrentThread(this);
-}
-
-HasNewFatalFailureHelper::~HasNewFatalFailureHelper() {
-  GetUnitTestImpl()->SetTestPartResultReporterForCurrentThread(
-      original_reporter_);
-}
-
-void HasNewFatalFailureHelper::ReportTestPartResult(
-    const TestPartResult& result) {
-  if (result.fatally_failed())
-    has_new_fatal_failure_ = true;
-  original_reporter_->ReportTestPartResult(result);
-}
-
-}  // namespace internal
-
-}  // namespace testing
-// Copyright 2008 Google Inc.
-// All Rights Reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-
-
-namespace testing {
-namespace internal {
-
-#if GTEST_HAS_TYPED_TEST_P
-
-// Skips to the first non-space char in str. Returns an empty string if str
-// contains only whitespace characters.
-static const char* SkipSpaces(const char* str) {
-  while (IsSpace(*str))
-    str++;
-  return str;
-}
-
-static std::vector<std::string> SplitIntoTestNames(const char* src) {
-  std::vector<std::string> name_vec;
-  src = SkipSpaces(src);
-  for (; src != NULL; src = SkipComma(src)) {
-    name_vec.push_back(StripTrailingSpaces(GetPrefixUntilComma(src)));
-  }
-  return name_vec;
-}
-
-// Verifies that registered_tests match the test names in
-// registered_tests_; returns registered_tests if successful, or
-// aborts the program otherwise.
-const char* TypedTestCasePState::VerifyRegisteredTestNames(
-    const char* file, int line, const char* registered_tests) {
-  typedef RegisteredTestsMap::const_iterator RegisteredTestIter;
-  registered_ = true;
-
-  std::vector<std::string> name_vec = SplitIntoTestNames(registered_tests);
-
-  Message errors;
-
-  std::set<std::string> tests;
-  for (std::vector<std::string>::const_iterator name_it = name_vec.begin();
-       name_it != name_vec.end(); ++name_it) {
-    const std::string& name = *name_it;
-    if (tests.count(name) != 0) {
-      errors << "Test " << name << " is listed more than once.\n";
-      continue;
-    }
-
-    bool found = false;
-    for (RegisteredTestIter it = registered_tests_.begin();
-         it != registered_tests_.end();
-         ++it) {
-      if (name == it->first) {
-        found = true;
-        break;
-      }
-    }
-
-    if (found) {
-      tests.insert(name);
-    } else {
-      errors << "No test named " << name
-             << " can be found in this test case.\n";
-    }
-  }
-
-  for (RegisteredTestIter it = registered_tests_.begin();
-       it != registered_tests_.end();
-       ++it) {
-    if (tests.count(it->first) == 0) {
-      errors << "You forgot to list test " << it->first << ".\n";
-    }
-  }
-
-  const std::string& errors_str = errors.GetString();
-  if (errors_str != "") {
-    fprintf(stderr, "%s %s", FormatFileLocation(file, line).c_str(),
-            errors_str.c_str());
-    fflush(stderr);
-    posix::Abort();
-  }
-
-  return registered_tests;
-}
-
-#endif  // GTEST_HAS_TYPED_TEST_P
-
-}  // namespace internal
-}  // namespace testing
diff --git a/lib/clickhouse-cpp/contrib/gtest/gtest.h b/lib/clickhouse-cpp/contrib/gtest/gtest.h
deleted file mode 100644
index aea1f51..0000000
--- a/lib/clickhouse-cpp/contrib/gtest/gtest.h
+++ /dev/null
@@ -1,21202 +0,0 @@
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-//
-// The Google C++ Testing Framework (Google Test)
-//
-// This header file defines the public API for Google Test.  It should be
-// included by any test program that uses Google Test.
-//
-// IMPORTANT NOTE: Due to limitation of the C++ language, we have to
-// leave some internal implementation details in this header file.
-// They are clearly marked by comments like this:
-//
-//   // INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-//
-// Such code is NOT meant to be used by a user directly, and is subject
-// to CHANGE WITHOUT NOTICE.  Therefore DO NOT DEPEND ON IT in a user
-// program!
-//
-// Acknowledgment: Google Test borrowed the idea of automatic test
-// registration from Barthelemy Dagenais' (barthelemy@prologique.com)
-// easyUnit framework.
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_H_
-#define GTEST_INCLUDE_GTEST_GTEST_H_
-
-#include <limits>
-#include <ostream>
-#include <vector>
-
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Authors: wan@google.com (Zhanyong Wan), eefacm@gmail.com (Sean Mcafee)
-//
-// The Google C++ Testing Framework (Google Test)
-//
-// This header file declares functions and macros used internally by
-// Google Test.  They are subject to change without notice.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_INTERNAL_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_INTERNAL_H_
-
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Authors: wan@google.com (Zhanyong Wan)
-//
-// Low-level types and utilities for porting Google Test to various
-// platforms.  All macros ending with _ and symbols defined in an
-// internal namespace are subject to change without notice.  Code
-// outside Google Test MUST NOT USE THEM DIRECTLY.  Macros that don't
-// end with _ are part of Google Test's public API and can be used by
-// code outside Google Test.
-//
-// This file is fundamental to Google Test.  All other Google Test source
-// files are expected to #include this.  Therefore, it cannot #include
-// any other Google Test header.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PORT_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PORT_H_
-
-// Environment-describing macros
-// -----------------------------
-//
-// Google Test can be used in many different environments.  Macros in
-// this section tell Google Test what kind of environment it is being
-// used in, such that Google Test can provide environment-specific
-// features and implementations.
-//
-// Google Test tries to automatically detect the properties of its
-// environment, so users usually don't need to worry about these
-// macros.  However, the automatic detection is not perfect.
-// Sometimes it's necessary for a user to define some of the following
-// macros in the build script to override Google Test's decisions.
-//
-// If the user doesn't define a macro in the list, Google Test will
-// provide a default definition.  After this header is #included, all
-// macros in this list will be defined to either 1 or 0.
-//
-// Notes to maintainers:
-//   - Each macro here is a user-tweakable knob; do not grow the list
-//     lightly.
-//   - Use #if to key off these macros.  Don't use #ifdef or "#if
-//     defined(...)", which will not work as these macros are ALWAYS
-//     defined.
-//
-//   GTEST_HAS_CLONE          - Define it to 1/0 to indicate that clone(2)
-//                              is/isn't available.
-//   GTEST_HAS_EXCEPTIONS     - Define it to 1/0 to indicate that exceptions
-//                              are enabled.
-//   GTEST_HAS_GLOBAL_STRING  - Define it to 1/0 to indicate that ::string
-//                              is/isn't available (some systems define
-//                              ::string, which is different to std::string).
-//   GTEST_HAS_GLOBAL_WSTRING - Define it to 1/0 to indicate that ::string
-//                              is/isn't available (some systems define
-//                              ::wstring, which is different to std::wstring).
-//   GTEST_HAS_POSIX_RE       - Define it to 1/0 to indicate that POSIX regular
-//                              expressions are/aren't available.
-//   GTEST_HAS_PTHREAD        - Define it to 1/0 to indicate that <pthread.h>
-//                              is/isn't available.
-//   GTEST_HAS_RTTI           - Define it to 1/0 to indicate that RTTI is/isn't
-//                              enabled.
-//   GTEST_HAS_STD_WSTRING    - Define it to 1/0 to indicate that
-//                              std::wstring does/doesn't work (Google Test can
-//                              be used where std::wstring is unavailable).
-//   GTEST_HAS_TR1_TUPLE      - Define it to 1/0 to indicate tr1::tuple
-//                              is/isn't available.
-//   GTEST_HAS_SEH            - Define it to 1/0 to indicate whether the
-//                              compiler supports Microsoft's "Structured
-//                              Exception Handling".
-//   GTEST_HAS_STREAM_REDIRECTION
-//                            - Define it to 1/0 to indicate whether the
-//                              platform supports I/O stream redirection using
-//                              dup() and dup2().
-//   GTEST_USE_OWN_TR1_TUPLE  - Define it to 1/0 to indicate whether Google
-//                              Test's own tr1 tuple implementation should be
-//                              used.  Unused when the user sets
-//                              GTEST_HAS_TR1_TUPLE to 0.
-//   GTEST_LANG_CXX11         - Define it to 1/0 to indicate that Google Test
-//                              is building in C++11/C++98 mode.
-//   GTEST_LINKED_AS_SHARED_LIBRARY
-//                            - Define to 1 when compiling tests that use
-//                              Google Test as a shared library (known as
-//                              DLL on Windows).
-//   GTEST_CREATE_SHARED_LIBRARY
-//                            - Define to 1 when compiling Google Test itself
-//                              as a shared library.
-
-// Platform-indicating macros
-// --------------------------
-//
-// Macros indicating the platform on which Google Test is being used
-// (a macro is defined to 1 if compiled on the given platform;
-// otherwise UNDEFINED -- it's never defined to 0.).  Google Test
-// defines these macros automatically.  Code outside Google Test MUST
-// NOT define them.
-//
-//   GTEST_OS_AIX      - IBM AIX
-//   GTEST_OS_CYGWIN   - Cygwin
-//   GTEST_OS_FREEBSD  - FreeBSD
-//   GTEST_OS_HPUX     - HP-UX
-//   GTEST_OS_LINUX    - Linux
-//     GTEST_OS_LINUX_ANDROID - Google Android
-//   GTEST_OS_MAC      - Mac OS X
-//     GTEST_OS_IOS    - iOS
-//   GTEST_OS_NACL     - Google Native Client (NaCl)
-//   GTEST_OS_OPENBSD  - OpenBSD
-//   GTEST_OS_QNX      - QNX
-//   GTEST_OS_SOLARIS  - Sun Solaris
-//   GTEST_OS_SYMBIAN  - Symbian
-//   GTEST_OS_WINDOWS  - Windows (Desktop, MinGW, or Mobile)
-//     GTEST_OS_WINDOWS_DESKTOP  - Windows Desktop
-//     GTEST_OS_WINDOWS_MINGW    - MinGW
-//     GTEST_OS_WINDOWS_MOBILE   - Windows Mobile
-//     GTEST_OS_WINDOWS_PHONE    - Windows Phone
-//     GTEST_OS_WINDOWS_RT       - Windows Store App/WinRT
-//   GTEST_OS_ZOS      - z/OS
-//
-// Among the platforms, Cygwin, Linux, Max OS X, and Windows have the
-// most stable support.  Since core members of the Google Test project
-// don't have access to other platforms, support for them may be less
-// stable.  If you notice any problems on your platform, please notify
-// googletestframework@googlegroups.com (patches for fixing them are
-// even more welcome!).
-//
-// It is possible that none of the GTEST_OS_* macros are defined.
-
-// Feature-indicating macros
-// -------------------------
-//
-// Macros indicating which Google Test features are available (a macro
-// is defined to 1 if the corresponding feature is supported;
-// otherwise UNDEFINED -- it's never defined to 0.).  Google Test
-// defines these macros automatically.  Code outside Google Test MUST
-// NOT define them.
-//
-// These macros are public so that portable tests can be written.
-// Such tests typically surround code using a feature with an #if
-// which controls that code.  For example:
-//
-// #if GTEST_HAS_DEATH_TEST
-//   EXPECT_DEATH(DoSomethingDeadly());
-// #endif
-//
-//   GTEST_HAS_COMBINE      - the Combine() function (for value-parameterized
-//                            tests)
-//   GTEST_HAS_DEATH_TEST   - death tests
-//   GTEST_HAS_PARAM_TEST   - value-parameterized tests
-//   GTEST_HAS_TYPED_TEST   - typed tests
-//   GTEST_HAS_TYPED_TEST_P - type-parameterized tests
-//   GTEST_IS_THREADSAFE    - Google Test is thread-safe.
-//   GTEST_USES_POSIX_RE    - enhanced POSIX regex is used. Do not confuse with
-//                            GTEST_HAS_POSIX_RE (see above) which users can
-//                            define themselves.
-//   GTEST_USES_SIMPLE_RE   - our own simple regex is used;
-//                            the above two are mutually exclusive.
-//   GTEST_CAN_COMPARE_NULL - accepts untyped NULL in EXPECT_EQ().
-
-// Misc public macros
-// ------------------
-//
-//   GTEST_FLAG(flag_name)  - references the variable corresponding to
-//                            the given Google Test flag.
-
-// Internal utilities
-// ------------------
-//
-// The following macros and utilities are for Google Test's INTERNAL
-// use only.  Code outside Google Test MUST NOT USE THEM DIRECTLY.
-//
-// Macros for basic C++ coding:
-//   GTEST_AMBIGUOUS_ELSE_BLOCKER_ - for disabling a gcc warning.
-//   GTEST_ATTRIBUTE_UNUSED_  - declares that a class' instances or a
-//                              variable don't have to be used.
-//   GTEST_DISALLOW_ASSIGN_   - disables operator=.
-//   GTEST_DISALLOW_COPY_AND_ASSIGN_ - disables copy ctor and operator=.
-//   GTEST_MUST_USE_RESULT_   - declares that a function's result must be used.
-//   GTEST_INTENTIONAL_CONST_COND_PUSH_ - start code section where MSVC C4127 is
-//                                        suppressed (constant conditional).
-//   GTEST_INTENTIONAL_CONST_COND_POP_  - finish code section where MSVC C4127
-//                                        is suppressed.
-//
-// C++11 feature wrappers:
-//
-//   testing::internal::move  - portability wrapper for std::move.
-//
-// Synchronization:
-//   Mutex, MutexLock, ThreadLocal, GetThreadCount()
-//                            - synchronization primitives.
-//
-// Template meta programming:
-//   is_pointer     - as in TR1; needed on Symbian and IBM XL C/C++ only.
-//   IteratorTraits - partial implementation of std::iterator_traits, which
-//                    is not available in libCstd when compiled with Sun C++.
-//
-// Smart pointers:
-//   scoped_ptr     - as in TR2.
-//
-// Regular expressions:
-//   RE             - a simple regular expression class using the POSIX
-//                    Extended Regular Expression syntax on UNIX-like
-//                    platforms, or a reduced regular exception syntax on
-//                    other platforms, including Windows.
-//
-// Logging:
-//   GTEST_LOG_()   - logs messages at the specified severity level.
-//   LogToStderr()  - directs all log messages to stderr.
-//   FlushInfoLog() - flushes informational log messages.
-//
-// Stdout and stderr capturing:
-//   CaptureStdout()     - starts capturing stdout.
-//   GetCapturedStdout() - stops capturing stdout and returns the captured
-//                         string.
-//   CaptureStderr()     - starts capturing stderr.
-//   GetCapturedStderr() - stops capturing stderr and returns the captured
-//                         string.
-//
-// Integer types:
-//   TypeWithSize   - maps an integer to a int type.
-//   Int32, UInt32, Int64, UInt64, TimeInMillis
-//                  - integers of known sizes.
-//   BiggestInt     - the biggest signed integer type.
-//
-// Command-line utilities:
-//   GTEST_DECLARE_*()  - declares a flag.
-//   GTEST_DEFINE_*()   - defines a flag.
-//   GetInjectableArgvs() - returns the command line as a vector of strings.
-//
-// Environment variable utilities:
-//   GetEnv()             - gets the value of an environment variable.
-//   BoolFromGTestEnv()   - parses a bool environment variable.
-//   Int32FromGTestEnv()  - parses an Int32 environment variable.
-//   StringFromGTestEnv() - parses a string environment variable.
-
-#include <ctype.h>   // for isspace, etc
-#include <stddef.h>  // for ptrdiff_t
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#ifndef _WIN32_WCE
-# include <sys/types.h>
-# include <sys/stat.h>
-#endif  // !_WIN32_WCE
-
-#if defined __APPLE__
-# include <AvailabilityMacros.h>
-# include <TargetConditionals.h>
-#endif
-
-#include <algorithm>  // NOLINT
-#include <iostream>  // NOLINT
-#include <sstream>  // NOLINT
-#include <string>  // NOLINT
-#include <utility>
-#include <vector>  // NOLINT
-
-// Copyright 2015, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// The Google C++ Testing Framework (Google Test)
-//
-// This header file defines the GTEST_OS_* macro.
-// It is separate from gtest-port.h so that custom/gtest-port.h can include it.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PORT_ARCH_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PORT_ARCH_H_
-
-// Determines the platform on which Google Test is compiled.
-#ifdef __CYGWIN__
-# define GTEST_OS_CYGWIN 1
-#elif defined __SYMBIAN32__
-# define GTEST_OS_SYMBIAN 1
-#elif defined _WIN32
-# define GTEST_OS_WINDOWS 1
-# ifdef _WIN32_WCE
-#  define GTEST_OS_WINDOWS_MOBILE 1
-# elif defined(__MINGW__) || defined(__MINGW32__)
-#  define GTEST_OS_WINDOWS_MINGW 1
-# elif defined(WINAPI_FAMILY)
-#  include <winapifamily.h>
-#  if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
-#   define GTEST_OS_WINDOWS_DESKTOP 1
-#  elif WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_PHONE_APP)
-#   define GTEST_OS_WINDOWS_PHONE 1
-#  elif WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP)
-#   define GTEST_OS_WINDOWS_RT 1
-#  else
-    // WINAPI_FAMILY defined but no known partition matched.
-    // Default to desktop.
-#   define GTEST_OS_WINDOWS_DESKTOP 1
-#  endif
-# else
-#  define GTEST_OS_WINDOWS_DESKTOP 1
-# endif  // _WIN32_WCE
-#elif defined __APPLE__
-# define GTEST_OS_MAC 1
-# if TARGET_OS_IPHONE
-#  define GTEST_OS_IOS 1
-# endif
-#elif defined __FreeBSD__
-# define GTEST_OS_FREEBSD 1
-#elif defined __linux__
-# define GTEST_OS_LINUX 1
-# if defined __ANDROID__
-#  define GTEST_OS_LINUX_ANDROID 1
-# endif
-#elif defined __MVS__
-# define GTEST_OS_ZOS 1
-#elif defined(__sun) && defined(__SVR4)
-# define GTEST_OS_SOLARIS 1
-#elif defined(_AIX)
-# define GTEST_OS_AIX 1
-#elif defined(__hpux)
-# define GTEST_OS_HPUX 1
-#elif defined __native_client__
-# define GTEST_OS_NACL 1
-#elif defined __OpenBSD__
-# define GTEST_OS_OPENBSD 1
-#elif defined __QNX__
-# define GTEST_OS_QNX 1
-#endif  // __CYGWIN__
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PORT_ARCH_H_
-// Copyright 2015, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Injection point for custom user configurations.
-// The following macros can be defined:
-//
-//   Flag related macros:
-//     GTEST_FLAG(flag_name)
-//     GTEST_USE_OWN_FLAGFILE_FLAG_  - Define to 0 when the system provides its
-//                                     own flagfile flag parsing.
-//     GTEST_DECLARE_bool_(name)
-//     GTEST_DECLARE_int32_(name)
-//     GTEST_DECLARE_string_(name)
-//     GTEST_DEFINE_bool_(name, default_val, doc)
-//     GTEST_DEFINE_int32_(name, default_val, doc)
-//     GTEST_DEFINE_string_(name, default_val, doc)
-//
-//   Test filtering:
-//     GTEST_TEST_FILTER_ENV_VAR_ - The name of an environment variable that
-//                                  will be used if --GTEST_FLAG(test_filter)
-//                                  is not provided.
-//
-//   Logging:
-//     GTEST_LOG_(severity)
-//     GTEST_CHECK_(condition)
-//     Functions LogToStderr() and FlushInfoLog() have to be provided too.
-//
-//   Threading:
-//     GTEST_HAS_NOTIFICATION_ - Enabled if Notification is already provided.
-//     GTEST_HAS_MUTEX_AND_THREAD_LOCAL_ - Enabled if Mutex and ThreadLocal are
-//                                         already provided.
-//     Must also provide GTEST_DECLARE_STATIC_MUTEX_(mutex) and
-//     GTEST_DEFINE_STATIC_MUTEX_(mutex)
-//
-//     GTEST_EXCLUSIVE_LOCK_REQUIRED_(locks)
-//     GTEST_LOCK_EXCLUDED_(locks)
-//
-// ** Custom implementation starts here **
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_CUSTOM_GTEST_PORT_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_CUSTOM_GTEST_PORT_H_
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_CUSTOM_GTEST_PORT_H_
-
-#if !defined(GTEST_DEV_EMAIL_)
-# define GTEST_DEV_EMAIL_ "googletestframework@@googlegroups.com"
-# define GTEST_FLAG_PREFIX_ "gtest_"
-# define GTEST_FLAG_PREFIX_DASH_ "gtest-"
-# define GTEST_FLAG_PREFIX_UPPER_ "GTEST_"
-# define GTEST_NAME_ "Google Test"
-# define GTEST_PROJECT_URL_ "https://github.com/google/googletest/"
-#endif  // !defined(GTEST_DEV_EMAIL_)
-
-#if !defined(GTEST_INIT_GOOGLE_TEST_NAME_)
-# define GTEST_INIT_GOOGLE_TEST_NAME_ "testing::InitGoogleTest"
-#endif  // !defined(GTEST_INIT_GOOGLE_TEST_NAME_)
-
-// Determines the version of gcc that is used to compile this.
-#ifdef __GNUC__
-// 40302 means version 4.3.2.
-# define GTEST_GCC_VER_ \
-    (__GNUC__*10000 + __GNUC_MINOR__*100 + __GNUC_PATCHLEVEL__)
-#endif  // __GNUC__
-
-// Macros for disabling Microsoft Visual C++ warnings.
-//
-//   GTEST_DISABLE_MSC_WARNINGS_PUSH_(4800 4385)
-//   /* code that triggers warnings C4800 and C4385 */
-//   GTEST_DISABLE_MSC_WARNINGS_POP_()
-#if _MSC_VER >= 1500
-# define GTEST_DISABLE_MSC_WARNINGS_PUSH_(warnings) \
-    __pragma(warning(push))                        \
-    __pragma(warning(disable: warnings))
-# define GTEST_DISABLE_MSC_WARNINGS_POP_()          \
-    __pragma(warning(pop))
-#else
-// Older versions of MSVC don't have __pragma.
-# define GTEST_DISABLE_MSC_WARNINGS_PUSH_(warnings)
-# define GTEST_DISABLE_MSC_WARNINGS_POP_()
-#endif
-
-#ifndef GTEST_LANG_CXX11
-// gcc and clang define __GXX_EXPERIMENTAL_CXX0X__ when
-// -std={c,gnu}++{0x,11} is passed.  The C++11 standard specifies a
-// value for __cplusplus, and recent versions of clang, gcc, and
-// probably other compilers set that too in C++11 mode.
-# if __GXX_EXPERIMENTAL_CXX0X__ || __cplusplus >= 201103L
-// Compiling in at least C++11 mode.
-#  define GTEST_LANG_CXX11 1
-# else
-#  define GTEST_LANG_CXX11 0
-# endif
-#endif
-
-// Distinct from C++11 language support, some environments don't provide
-// proper C++11 library support. Notably, it's possible to build in
-// C++11 mode when targeting Mac OS X 10.6, which has an old libstdc++
-// with no C++11 support.
-//
-// libstdc++ has sufficient C++11 support as of GCC 4.6.0, __GLIBCXX__
-// 20110325, but maintenance releases in the 4.4 and 4.5 series followed
-// this date, so check for those versions by their date stamps.
-// https://gcc.gnu.org/onlinedocs/libstdc++/manual/abi.html#abi.versioning
-#if GTEST_LANG_CXX11 && \
-    (!defined(__GLIBCXX__) || ( \
-        __GLIBCXX__ >= 20110325ul &&  /* GCC >= 4.6.0 */ \
-        /* Blacklist of patch releases of older branches: */ \
-        __GLIBCXX__ != 20110416ul &&  /* GCC 4.4.6 */ \
-        __GLIBCXX__ != 20120313ul &&  /* GCC 4.4.7 */ \
-        __GLIBCXX__ != 20110428ul &&  /* GCC 4.5.3 */ \
-        __GLIBCXX__ != 20120702ul))   /* GCC 4.5.4 */
-# define GTEST_STDLIB_CXX11 1
-#endif
-
-// Only use C++11 library features if the library provides them.
-#if GTEST_STDLIB_CXX11
-# define GTEST_HAS_STD_BEGIN_AND_END_ 1
-# define GTEST_HAS_STD_FORWARD_LIST_ 1
-# define GTEST_HAS_STD_FUNCTION_ 1
-# define GTEST_HAS_STD_INITIALIZER_LIST_ 1
-# define GTEST_HAS_STD_MOVE_ 1
-# define GTEST_HAS_STD_SHARED_PTR_ 1
-# define GTEST_HAS_STD_TYPE_TRAITS_ 1
-# define GTEST_HAS_STD_UNIQUE_PTR_ 1
-#endif
-
-// C++11 specifies that <tuple> provides std::tuple.
-// Some platforms still might not have it, however.
-#if GTEST_LANG_CXX11
-# define GTEST_HAS_STD_TUPLE_ 1
-# if defined(__clang__)
-// Inspired by http://clang.llvm.org/docs/LanguageExtensions.html#__has_include
-#  if defined(__has_include) && !__has_include(<tuple>)
-#   undef GTEST_HAS_STD_TUPLE_
-#  endif
-# elif defined(_MSC_VER)
-// Inspired by boost/config/stdlib/dinkumware.hpp
-#  if defined(_CPPLIB_VER) && _CPPLIB_VER < 520
-#   undef GTEST_HAS_STD_TUPLE_
-#  endif
-# elif defined(__GLIBCXX__)
-// Inspired by boost/config/stdlib/libstdcpp3.hpp,
-// http://gcc.gnu.org/gcc-4.2/changes.html and
-// http://gcc.gnu.org/onlinedocs/libstdc++/manual/bk01pt01ch01.html#manual.intro.status.standard.200x
-#  if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 2)
-#   undef GTEST_HAS_STD_TUPLE_
-#  endif
-# endif
-#endif
-
-// Brings in definitions for functions used in the testing::internal::posix
-// namespace (read, write, close, chdir, isatty, stat). We do not currently
-// use them on Windows Mobile.
-#if GTEST_OS_WINDOWS
-# if !GTEST_OS_WINDOWS_MOBILE
-#  include <direct.h>
-#  include <io.h>
-# endif
-// In order to avoid having to include <windows.h>, use forward declaration
-#if GTEST_OS_WINDOWS_MINGW && !defined(__MINGW64_VERSION_MAJOR)
-// MinGW defined _CRITICAL_SECTION and _RTL_CRITICAL_SECTION as two
-// separate (equivalent) structs, instead of using typedef
-typedef struct _CRITICAL_SECTION GTEST_CRITICAL_SECTION;
-#else
-// Assume CRITICAL_SECTION is a typedef of _RTL_CRITICAL_SECTION.
-// This assumption is verified by
-// WindowsTypesTest.CRITICAL_SECTIONIs_RTL_CRITICAL_SECTION.
-typedef struct _RTL_CRITICAL_SECTION GTEST_CRITICAL_SECTION;
-#endif
-#else
-// This assumes that non-Windows OSes provide unistd.h. For OSes where this
-// is not the case, we need to include headers that provide the functions
-// mentioned above.
-# include <unistd.h>
-# include <strings.h>
-#endif  // GTEST_OS_WINDOWS
-
-#if GTEST_OS_LINUX_ANDROID
-// Used to define __ANDROID_API__ matching the target NDK API level.
-#  include <android/api-level.h>  // NOLINT
-#endif
-
-// Defines this to true iff Google Test can use POSIX regular expressions.
-#ifndef GTEST_HAS_POSIX_RE
-# if GTEST_OS_LINUX_ANDROID
-// On Android, <regex.h> is only available starting with Gingerbread.
-#  define GTEST_HAS_POSIX_RE (__ANDROID_API__ >= 9)
-# else
-#  define GTEST_HAS_POSIX_RE (!GTEST_OS_WINDOWS)
-# endif
-#endif
-
-#if GTEST_USES_PCRE
-// The appropriate headers have already been included.
-
-#elif GTEST_HAS_POSIX_RE
-
-// On some platforms, <regex.h> needs someone to define size_t, and
-// won't compile otherwise.  We can #include it here as we already
-// included <stdlib.h>, which is guaranteed to define size_t through
-// <stddef.h>.
-# include <regex.h>  // NOLINT
-
-# define GTEST_USES_POSIX_RE 1
-
-#elif GTEST_OS_WINDOWS
-
-// <regex.h> is not available on Windows.  Use our own simple regex
-// implementation instead.
-# define GTEST_USES_SIMPLE_RE 1
-
-#else
-
-// <regex.h> may not be available on this platform.  Use our own
-// simple regex implementation instead.
-# define GTEST_USES_SIMPLE_RE 1
-
-#endif  // GTEST_USES_PCRE
-
-#ifndef GTEST_HAS_EXCEPTIONS
-// The user didn't tell us whether exceptions are enabled, so we need
-// to figure it out.
-# if defined(_MSC_VER) || defined(__BORLANDC__)
-// MSVC's and C++Builder's implementations of the STL use the _HAS_EXCEPTIONS
-// macro to enable exceptions, so we'll do the same.
-// Assumes that exceptions are enabled by default.
-#  ifndef _HAS_EXCEPTIONS
-#   define _HAS_EXCEPTIONS 1
-#  endif  // _HAS_EXCEPTIONS
-#  define GTEST_HAS_EXCEPTIONS _HAS_EXCEPTIONS
-# elif defined(__clang__)
-// clang defines __EXCEPTIONS iff exceptions are enabled before clang 220714,
-// but iff cleanups are enabled after that. In Obj-C++ files, there can be
-// cleanups for ObjC exceptions which also need cleanups, even if C++ exceptions
-// are disabled. clang has __has_feature(cxx_exceptions) which checks for C++
-// exceptions starting at clang r206352, but which checked for cleanups prior to
-// that. To reliably check for C++ exception availability with clang, check for
-// __EXCEPTIONS && __has_feature(cxx_exceptions).
-#  define GTEST_HAS_EXCEPTIONS (__EXCEPTIONS && __has_feature(cxx_exceptions))
-# elif defined(__GNUC__) && __EXCEPTIONS
-// gcc defines __EXCEPTIONS to 1 iff exceptions are enabled.
-#  define GTEST_HAS_EXCEPTIONS 1
-# elif defined(__SUNPRO_CC)
-// Sun Pro CC supports exceptions.  However, there is no compile-time way of
-// detecting whether they are enabled or not.  Therefore, we assume that
-// they are enabled unless the user tells us otherwise.
-#  define GTEST_HAS_EXCEPTIONS 1
-# elif defined(__IBMCPP__) && __EXCEPTIONS
-// xlC defines __EXCEPTIONS to 1 iff exceptions are enabled.
-#  define GTEST_HAS_EXCEPTIONS 1
-# elif defined(__HP_aCC)
-// Exception handling is in effect by default in HP aCC compiler. It has to
-// be turned of by +noeh compiler option if desired.
-#  define GTEST_HAS_EXCEPTIONS 1
-# else
-// For other compilers, we assume exceptions are disabled to be
-// conservative.
-#  define GTEST_HAS_EXCEPTIONS 0
-# endif  // defined(_MSC_VER) || defined(__BORLANDC__)
-#endif  // GTEST_HAS_EXCEPTIONS
-
-#if !defined(GTEST_HAS_STD_STRING)
-// Even though we don't use this macro any longer, we keep it in case
-// some clients still depend on it.
-# define GTEST_HAS_STD_STRING 1
-#elif !GTEST_HAS_STD_STRING
-// The user told us that ::std::string isn't available.
-# error "Google Test cannot be used where ::std::string isn't available."
-#endif  // !defined(GTEST_HAS_STD_STRING)
-
-#ifndef GTEST_HAS_GLOBAL_STRING
-// The user didn't tell us whether ::string is available, so we need
-// to figure it out.
-
-# define GTEST_HAS_GLOBAL_STRING 0
-
-#endif  // GTEST_HAS_GLOBAL_STRING
-
-#ifndef GTEST_HAS_STD_WSTRING
-// The user didn't tell us whether ::std::wstring is available, so we need
-// to figure it out.
-// TODO(wan@google.com): uses autoconf to detect whether ::std::wstring
-//   is available.
-
-// Cygwin 1.7 and below doesn't support ::std::wstring.
-// Solaris' libc++ doesn't support it either.  Android has
-// no support for it at least as recent as Froyo (2.2).
-# define GTEST_HAS_STD_WSTRING \
-    (!(GTEST_OS_LINUX_ANDROID || GTEST_OS_CYGWIN || GTEST_OS_SOLARIS))
-
-#endif  // GTEST_HAS_STD_WSTRING
-
-#ifndef GTEST_HAS_GLOBAL_WSTRING
-// The user didn't tell us whether ::wstring is available, so we need
-// to figure it out.
-# define GTEST_HAS_GLOBAL_WSTRING \
-    (GTEST_HAS_STD_WSTRING && GTEST_HAS_GLOBAL_STRING)
-#endif  // GTEST_HAS_GLOBAL_WSTRING
-
-// Determines whether RTTI is available.
-#ifndef GTEST_HAS_RTTI
-// The user didn't tell us whether RTTI is enabled, so we need to
-// figure it out.
-
-# ifdef _MSC_VER
-
-#  ifdef _CPPRTTI  // MSVC defines this macro iff RTTI is enabled.
-#   define GTEST_HAS_RTTI 1
-#  else
-#   define GTEST_HAS_RTTI 0
-#  endif
-
-// Starting with version 4.3.2, gcc defines __GXX_RTTI iff RTTI is enabled.
-# elif defined(__GNUC__) && (GTEST_GCC_VER_ >= 40302)
-
-#  ifdef __GXX_RTTI
-// When building against STLport with the Android NDK and with
-// -frtti -fno-exceptions, the build fails at link time with undefined
-// references to __cxa_bad_typeid. Note sure if STL or toolchain bug,
-// so disable RTTI when detected.
-#   if GTEST_OS_LINUX_ANDROID && defined(_STLPORT_MAJOR) && \
-       !defined(__EXCEPTIONS)
-#    define GTEST_HAS_RTTI 0
-#   else
-#    define GTEST_HAS_RTTI 1
-#   endif  // GTEST_OS_LINUX_ANDROID && __STLPORT_MAJOR && !__EXCEPTIONS
-#  else
-#   define GTEST_HAS_RTTI 0
-#  endif  // __GXX_RTTI
-
-// Clang defines __GXX_RTTI starting with version 3.0, but its manual recommends
-// using has_feature instead. has_feature(cxx_rtti) is supported since 2.7, the
-// first version with C++ support.
-# elif defined(__clang__)
-
-#  define GTEST_HAS_RTTI __has_feature(cxx_rtti)
-
-// Starting with version 9.0 IBM Visual Age defines __RTTI_ALL__ to 1 if
-// both the typeid and dynamic_cast features are present.
-# elif defined(__IBMCPP__) && (__IBMCPP__ >= 900)
-
-#  ifdef __RTTI_ALL__
-#   define GTEST_HAS_RTTI 1
-#  else
-#   define GTEST_HAS_RTTI 0
-#  endif
-
-# else
-
-// For all other compilers, we assume RTTI is enabled.
-#  define GTEST_HAS_RTTI 1
-
-# endif  // _MSC_VER
-
-#endif  // GTEST_HAS_RTTI
-
-// It's this header's responsibility to #include <typeinfo> when RTTI
-// is enabled.
-#if GTEST_HAS_RTTI
-# include <typeinfo>
-#endif
-
-// Determines whether Google Test can use the pthreads library.
-#ifndef GTEST_HAS_PTHREAD
-// The user didn't tell us explicitly, so we make reasonable assumptions about
-// which platforms have pthreads support.
-//
-// To disable threading support in Google Test, add -DGTEST_HAS_PTHREAD=0
-// to your compiler flags.
-# define GTEST_HAS_PTHREAD (GTEST_OS_LINUX || GTEST_OS_MAC || GTEST_OS_HPUX \
-    || GTEST_OS_QNX || GTEST_OS_FREEBSD || GTEST_OS_NACL)
-#endif  // GTEST_HAS_PTHREAD
-
-#if GTEST_HAS_PTHREAD
-// gtest-port.h guarantees to #include <pthread.h> when GTEST_HAS_PTHREAD is
-// true.
-# include <pthread.h>  // NOLINT
-
-// For timespec and nanosleep, used below.
-# include <time.h>  // NOLINT
-#endif
-
-// Determines if hash_map/hash_set are available.
-// Only used for testing against those containers.
-#if !defined(GTEST_HAS_HASH_MAP_)
-# if _MSC_VER
-#  define GTEST_HAS_HASH_MAP_ 1  // Indicates that hash_map is available.
-#  define GTEST_HAS_HASH_SET_ 1  // Indicates that hash_set is available.
-# endif  // _MSC_VER
-#endif  // !defined(GTEST_HAS_HASH_MAP_)
-
-// Determines whether Google Test can use tr1/tuple.  You can define
-// this macro to 0 to prevent Google Test from using tuple (any
-// feature depending on tuple with be disabled in this mode).
-#ifndef GTEST_HAS_TR1_TUPLE
-# if GTEST_OS_LINUX_ANDROID && defined(_STLPORT_MAJOR)
-// STLport, provided with the Android NDK, has neither <tr1/tuple> or <tuple>.
-#  define GTEST_HAS_TR1_TUPLE 0
-# else
-// The user didn't tell us not to do it, so we assume it's OK.
-#  define GTEST_HAS_TR1_TUPLE 1
-# endif
-#endif  // GTEST_HAS_TR1_TUPLE
-
-// Determines whether Google Test's own tr1 tuple implementation
-// should be used.
-#ifndef GTEST_USE_OWN_TR1_TUPLE
-// The user didn't tell us, so we need to figure it out.
-
-// We use our own TR1 tuple if we aren't sure the user has an
-// implementation of it already.  At this time, libstdc++ 4.0.0+ and
-// MSVC 2010 are the only mainstream standard libraries that come
-// with a TR1 tuple implementation.  NVIDIA's CUDA NVCC compiler
-// pretends to be GCC by defining __GNUC__ and friends, but cannot
-// compile GCC's tuple implementation.  MSVC 2008 (9.0) provides TR1
-// tuple in a 323 MB Feature Pack download, which we cannot assume the
-// user has.  QNX's QCC compiler is a modified GCC but it doesn't
-// support TR1 tuple.  libc++ only provides std::tuple, in C++11 mode,
-// and it can be used with some compilers that define __GNUC__.
-# if (defined(__GNUC__) && !defined(__CUDACC__) && (GTEST_GCC_VER_ >= 40000) \
-      && !GTEST_OS_QNX && !defined(_LIBCPP_VERSION)) || _MSC_VER >= 1600
-#  define GTEST_ENV_HAS_TR1_TUPLE_ 1
-# endif
-
-// C++11 specifies that <tuple> provides std::tuple. Use that if gtest is used
-// in C++11 mode and libstdc++ isn't very old (binaries targeting OS X 10.6
-// can build with clang but need to use gcc4.2's libstdc++).
-# if GTEST_LANG_CXX11 && (!defined(__GLIBCXX__) || __GLIBCXX__ > 20110325)
-#  define GTEST_ENV_HAS_STD_TUPLE_ 1
-# endif
-
-# if GTEST_ENV_HAS_TR1_TUPLE_ || GTEST_ENV_HAS_STD_TUPLE_
-#  define GTEST_USE_OWN_TR1_TUPLE 0
-# else
-#  define GTEST_USE_OWN_TR1_TUPLE 1
-# endif
-
-#endif  // GTEST_USE_OWN_TR1_TUPLE
-
-// To avoid conditional compilation everywhere, we make it
-// gtest-port.h's responsibility to #include the header implementing
-// tuple.
-#if GTEST_HAS_STD_TUPLE_
-# include <tuple>  // IWYU pragma: export
-# define GTEST_TUPLE_NAMESPACE_ ::std
-#endif  // GTEST_HAS_STD_TUPLE_
-
-// We include tr1::tuple even if std::tuple is available to define printers for
-// them.
-#if GTEST_HAS_TR1_TUPLE
-# ifndef GTEST_TUPLE_NAMESPACE_
-#  define GTEST_TUPLE_NAMESPACE_ ::std::tr1
-# endif  // GTEST_TUPLE_NAMESPACE_
-
-# if GTEST_USE_OWN_TR1_TUPLE
-// This file was GENERATED by command:
-//     pump.py gtest-tuple.h.pump
-// DO NOT EDIT BY HAND!!!
-
-// Copyright 2009 Google Inc.
-// All Rights Reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-
-// Implements a subset of TR1 tuple needed by Google Test and Google Mock.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_TUPLE_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_TUPLE_H_
-
-#include <utility>  // For ::std::pair.
-
-// The compiler used in Symbian has a bug that prevents us from declaring the
-// tuple template as a friend (it complains that tuple is redefined).  This
-// hack bypasses the bug by declaring the members that should otherwise be
-// private as public.
-// Sun Studio versions < 12 also have the above bug.
-#if defined(__SYMBIAN32__) || (defined(__SUNPRO_CC) && __SUNPRO_CC < 0x590)
-# define GTEST_DECLARE_TUPLE_AS_FRIEND_ public:
-#else
-# define GTEST_DECLARE_TUPLE_AS_FRIEND_ \
-    template <GTEST_10_TYPENAMES_(U)> friend class tuple; \
-   private:
-#endif
-
-// Visual Studio 2010, 2012, and 2013 define symbols in std::tr1 that conflict
-// with our own definitions. Therefore using our own tuple does not work on
-// those compilers.
-#if defined(_MSC_VER) && _MSC_VER >= 1600  /* 1600 is Visual Studio 2010 */
-# error "gtest's tuple doesn't compile on Visual Studio 2010 or later. \
-GTEST_USE_OWN_TR1_TUPLE must be set to 0 on those compilers."
-#endif
-
-// GTEST_n_TUPLE_(T) is the type of an n-tuple.
-#define GTEST_0_TUPLE_(T) tuple<>
-#define GTEST_1_TUPLE_(T) tuple<T##0, void, void, void, void, void, void, \
-    void, void, void>
-#define GTEST_2_TUPLE_(T) tuple<T##0, T##1, void, void, void, void, void, \
-    void, void, void>
-#define GTEST_3_TUPLE_(T) tuple<T##0, T##1, T##2, void, void, void, void, \
-    void, void, void>
-#define GTEST_4_TUPLE_(T) tuple<T##0, T##1, T##2, T##3, void, void, void, \
-    void, void, void>
-#define GTEST_5_TUPLE_(T) tuple<T##0, T##1, T##2, T##3, T##4, void, void, \
-    void, void, void>
-#define GTEST_6_TUPLE_(T) tuple<T##0, T##1, T##2, T##3, T##4, T##5, void, \
-    void, void, void>
-#define GTEST_7_TUPLE_(T) tuple<T##0, T##1, T##2, T##3, T##4, T##5, T##6, \
-    void, void, void>
-#define GTEST_8_TUPLE_(T) tuple<T##0, T##1, T##2, T##3, T##4, T##5, T##6, \
-    T##7, void, void>
-#define GTEST_9_TUPLE_(T) tuple<T##0, T##1, T##2, T##3, T##4, T##5, T##6, \
-    T##7, T##8, void>
-#define GTEST_10_TUPLE_(T) tuple<T##0, T##1, T##2, T##3, T##4, T##5, T##6, \
-    T##7, T##8, T##9>
-
-// GTEST_n_TYPENAMES_(T) declares a list of n typenames.
-#define GTEST_0_TYPENAMES_(T)
-#define GTEST_1_TYPENAMES_(T) typename T##0
-#define GTEST_2_TYPENAMES_(T) typename T##0, typename T##1
-#define GTEST_3_TYPENAMES_(T) typename T##0, typename T##1, typename T##2
-#define GTEST_4_TYPENAMES_(T) typename T##0, typename T##1, typename T##2, \
-    typename T##3
-#define GTEST_5_TYPENAMES_(T) typename T##0, typename T##1, typename T##2, \
-    typename T##3, typename T##4
-#define GTEST_6_TYPENAMES_(T) typename T##0, typename T##1, typename T##2, \
-    typename T##3, typename T##4, typename T##5
-#define GTEST_7_TYPENAMES_(T) typename T##0, typename T##1, typename T##2, \
-    typename T##3, typename T##4, typename T##5, typename T##6
-#define GTEST_8_TYPENAMES_(T) typename T##0, typename T##1, typename T##2, \
-    typename T##3, typename T##4, typename T##5, typename T##6, typename T##7
-#define GTEST_9_TYPENAMES_(T) typename T##0, typename T##1, typename T##2, \
-    typename T##3, typename T##4, typename T##5, typename T##6, \
-    typename T##7, typename T##8
-#define GTEST_10_TYPENAMES_(T) typename T##0, typename T##1, typename T##2, \
-    typename T##3, typename T##4, typename T##5, typename T##6, \
-    typename T##7, typename T##8, typename T##9
-
-// In theory, defining stuff in the ::std namespace is undefined
-// behavior.  We can do this as we are playing the role of a standard
-// library vendor.
-namespace std {
-namespace tr1 {
-
-template <typename T0 = void, typename T1 = void, typename T2 = void,
-    typename T3 = void, typename T4 = void, typename T5 = void,
-    typename T6 = void, typename T7 = void, typename T8 = void,
-    typename T9 = void>
-class tuple;
-
-// Anything in namespace gtest_internal is Google Test's INTERNAL
-// IMPLEMENTATION DETAIL and MUST NOT BE USED DIRECTLY in user code.
-namespace gtest_internal {
-
-// ByRef<T>::type is T if T is a reference; otherwise it's const T&.
-template <typename T>
-struct ByRef { typedef const T& type; };  // NOLINT
-template <typename T>
-struct ByRef<T&> { typedef T& type; };  // NOLINT
-
-// A handy wrapper for ByRef.
-#define GTEST_BY_REF_(T) typename ::std::tr1::gtest_internal::ByRef<T>::type
-
-// AddRef<T>::type is T if T is a reference; otherwise it's T&.  This
-// is the same as tr1::add_reference<T>::type.
-template <typename T>
-struct AddRef { typedef T& type; };  // NOLINT
-template <typename T>
-struct AddRef<T&> { typedef T& type; };  // NOLINT
-
-// A handy wrapper for AddRef.
-#define GTEST_ADD_REF_(T) typename ::std::tr1::gtest_internal::AddRef<T>::type
-
-// A helper for implementing get<k>().
-template <int k> class Get;
-
-// A helper for implementing tuple_element<k, T>.  kIndexValid is true
-// iff k < the number of fields in tuple type T.
-template <bool kIndexValid, int kIndex, class Tuple>
-struct TupleElement;
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 0, GTEST_10_TUPLE_(T) > {
-  typedef T0 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 1, GTEST_10_TUPLE_(T) > {
-  typedef T1 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 2, GTEST_10_TUPLE_(T) > {
-  typedef T2 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 3, GTEST_10_TUPLE_(T) > {
-  typedef T3 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 4, GTEST_10_TUPLE_(T) > {
-  typedef T4 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 5, GTEST_10_TUPLE_(T) > {
-  typedef T5 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 6, GTEST_10_TUPLE_(T) > {
-  typedef T6 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 7, GTEST_10_TUPLE_(T) > {
-  typedef T7 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 8, GTEST_10_TUPLE_(T) > {
-  typedef T8 type;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct TupleElement<true, 9, GTEST_10_TUPLE_(T) > {
-  typedef T9 type;
-};
-
-}  // namespace gtest_internal
-
-template <>
-class tuple<> {
- public:
-  tuple() {}
-  tuple(const tuple& /* t */)  {}
-  tuple& operator=(const tuple& /* t */) { return *this; }
-};
-
-template <GTEST_1_TYPENAMES_(T)>
-class GTEST_1_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0) : f0_(f0) {}
-
-  tuple(const tuple& t) : f0_(t.f0_) {}
-
-  template <GTEST_1_TYPENAMES_(U)>
-  tuple(const GTEST_1_TUPLE_(U)& t) : f0_(t.f0_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_1_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_1_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_1_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_1_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    return *this;
-  }
-
-  T0 f0_;
-};
-
-template <GTEST_2_TYPENAMES_(T)>
-class GTEST_2_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1) : f0_(f0),
-      f1_(f1) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_) {}
-
-  template <GTEST_2_TYPENAMES_(U)>
-  tuple(const GTEST_2_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_) {}
-  template <typename U0, typename U1>
-  tuple(const ::std::pair<U0, U1>& p) : f0_(p.first), f1_(p.second) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_2_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_2_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-  template <typename U0, typename U1>
-  tuple& operator=(const ::std::pair<U0, U1>& p) {
-    f0_ = p.first;
-    f1_ = p.second;
-    return *this;
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_2_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_2_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-};
-
-template <GTEST_3_TYPENAMES_(T)>
-class GTEST_3_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2) : f0_(f0), f1_(f1), f2_(f2) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_) {}
-
-  template <GTEST_3_TYPENAMES_(U)>
-  tuple(const GTEST_3_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_3_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_3_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_3_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_3_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-};
-
-template <GTEST_4_TYPENAMES_(T)>
-class GTEST_4_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_(), f3_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2, GTEST_BY_REF_(T3) f3) : f0_(f0), f1_(f1), f2_(f2),
-      f3_(f3) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_), f3_(t.f3_) {}
-
-  template <GTEST_4_TYPENAMES_(U)>
-  tuple(const GTEST_4_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_),
-      f3_(t.f3_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_4_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_4_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_4_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_4_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    f3_ = t.f3_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-  T3 f3_;
-};
-
-template <GTEST_5_TYPENAMES_(T)>
-class GTEST_5_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_(), f3_(), f4_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2, GTEST_BY_REF_(T3) f3,
-      GTEST_BY_REF_(T4) f4) : f0_(f0), f1_(f1), f2_(f2), f3_(f3), f4_(f4) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_), f3_(t.f3_),
-      f4_(t.f4_) {}
-
-  template <GTEST_5_TYPENAMES_(U)>
-  tuple(const GTEST_5_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_),
-      f3_(t.f3_), f4_(t.f4_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_5_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_5_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_5_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_5_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    f3_ = t.f3_;
-    f4_ = t.f4_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-  T3 f3_;
-  T4 f4_;
-};
-
-template <GTEST_6_TYPENAMES_(T)>
-class GTEST_6_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_(), f3_(), f4_(), f5_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2, GTEST_BY_REF_(T3) f3, GTEST_BY_REF_(T4) f4,
-      GTEST_BY_REF_(T5) f5) : f0_(f0), f1_(f1), f2_(f2), f3_(f3), f4_(f4),
-      f5_(f5) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_), f3_(t.f3_),
-      f4_(t.f4_), f5_(t.f5_) {}
-
-  template <GTEST_6_TYPENAMES_(U)>
-  tuple(const GTEST_6_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_),
-      f3_(t.f3_), f4_(t.f4_), f5_(t.f5_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_6_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_6_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_6_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_6_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    f3_ = t.f3_;
-    f4_ = t.f4_;
-    f5_ = t.f5_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-  T3 f3_;
-  T4 f4_;
-  T5 f5_;
-};
-
-template <GTEST_7_TYPENAMES_(T)>
-class GTEST_7_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_(), f3_(), f4_(), f5_(), f6_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2, GTEST_BY_REF_(T3) f3, GTEST_BY_REF_(T4) f4,
-      GTEST_BY_REF_(T5) f5, GTEST_BY_REF_(T6) f6) : f0_(f0), f1_(f1), f2_(f2),
-      f3_(f3), f4_(f4), f5_(f5), f6_(f6) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_), f3_(t.f3_),
-      f4_(t.f4_), f5_(t.f5_), f6_(t.f6_) {}
-
-  template <GTEST_7_TYPENAMES_(U)>
-  tuple(const GTEST_7_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_),
-      f3_(t.f3_), f4_(t.f4_), f5_(t.f5_), f6_(t.f6_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_7_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_7_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_7_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_7_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    f3_ = t.f3_;
-    f4_ = t.f4_;
-    f5_ = t.f5_;
-    f6_ = t.f6_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-  T3 f3_;
-  T4 f4_;
-  T5 f5_;
-  T6 f6_;
-};
-
-template <GTEST_8_TYPENAMES_(T)>
-class GTEST_8_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_(), f3_(), f4_(), f5_(), f6_(), f7_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2, GTEST_BY_REF_(T3) f3, GTEST_BY_REF_(T4) f4,
-      GTEST_BY_REF_(T5) f5, GTEST_BY_REF_(T6) f6,
-      GTEST_BY_REF_(T7) f7) : f0_(f0), f1_(f1), f2_(f2), f3_(f3), f4_(f4),
-      f5_(f5), f6_(f6), f7_(f7) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_), f3_(t.f3_),
-      f4_(t.f4_), f5_(t.f5_), f6_(t.f6_), f7_(t.f7_) {}
-
-  template <GTEST_8_TYPENAMES_(U)>
-  tuple(const GTEST_8_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_),
-      f3_(t.f3_), f4_(t.f4_), f5_(t.f5_), f6_(t.f6_), f7_(t.f7_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_8_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_8_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_8_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_8_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    f3_ = t.f3_;
-    f4_ = t.f4_;
-    f5_ = t.f5_;
-    f6_ = t.f6_;
-    f7_ = t.f7_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-  T3 f3_;
-  T4 f4_;
-  T5 f5_;
-  T6 f6_;
-  T7 f7_;
-};
-
-template <GTEST_9_TYPENAMES_(T)>
-class GTEST_9_TUPLE_(T) {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_(), f3_(), f4_(), f5_(), f6_(), f7_(), f8_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2, GTEST_BY_REF_(T3) f3, GTEST_BY_REF_(T4) f4,
-      GTEST_BY_REF_(T5) f5, GTEST_BY_REF_(T6) f6, GTEST_BY_REF_(T7) f7,
-      GTEST_BY_REF_(T8) f8) : f0_(f0), f1_(f1), f2_(f2), f3_(f3), f4_(f4),
-      f5_(f5), f6_(f6), f7_(f7), f8_(f8) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_), f3_(t.f3_),
-      f4_(t.f4_), f5_(t.f5_), f6_(t.f6_), f7_(t.f7_), f8_(t.f8_) {}
-
-  template <GTEST_9_TYPENAMES_(U)>
-  tuple(const GTEST_9_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_),
-      f3_(t.f3_), f4_(t.f4_), f5_(t.f5_), f6_(t.f6_), f7_(t.f7_), f8_(t.f8_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_9_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_9_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_9_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_9_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    f3_ = t.f3_;
-    f4_ = t.f4_;
-    f5_ = t.f5_;
-    f6_ = t.f6_;
-    f7_ = t.f7_;
-    f8_ = t.f8_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-  T3 f3_;
-  T4 f4_;
-  T5 f5_;
-  T6 f6_;
-  T7 f7_;
-  T8 f8_;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-class tuple {
- public:
-  template <int k> friend class gtest_internal::Get;
-
-  tuple() : f0_(), f1_(), f2_(), f3_(), f4_(), f5_(), f6_(), f7_(), f8_(),
-      f9_() {}
-
-  explicit tuple(GTEST_BY_REF_(T0) f0, GTEST_BY_REF_(T1) f1,
-      GTEST_BY_REF_(T2) f2, GTEST_BY_REF_(T3) f3, GTEST_BY_REF_(T4) f4,
-      GTEST_BY_REF_(T5) f5, GTEST_BY_REF_(T6) f6, GTEST_BY_REF_(T7) f7,
-      GTEST_BY_REF_(T8) f8, GTEST_BY_REF_(T9) f9) : f0_(f0), f1_(f1), f2_(f2),
-      f3_(f3), f4_(f4), f5_(f5), f6_(f6), f7_(f7), f8_(f8), f9_(f9) {}
-
-  tuple(const tuple& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_), f3_(t.f3_),
-      f4_(t.f4_), f5_(t.f5_), f6_(t.f6_), f7_(t.f7_), f8_(t.f8_), f9_(t.f9_) {}
-
-  template <GTEST_10_TYPENAMES_(U)>
-  tuple(const GTEST_10_TUPLE_(U)& t) : f0_(t.f0_), f1_(t.f1_), f2_(t.f2_),
-      f3_(t.f3_), f4_(t.f4_), f5_(t.f5_), f6_(t.f6_), f7_(t.f7_), f8_(t.f8_),
-      f9_(t.f9_) {}
-
-  tuple& operator=(const tuple& t) { return CopyFrom(t); }
-
-  template <GTEST_10_TYPENAMES_(U)>
-  tuple& operator=(const GTEST_10_TUPLE_(U)& t) {
-    return CopyFrom(t);
-  }
-
-  GTEST_DECLARE_TUPLE_AS_FRIEND_
-
-  template <GTEST_10_TYPENAMES_(U)>
-  tuple& CopyFrom(const GTEST_10_TUPLE_(U)& t) {
-    f0_ = t.f0_;
-    f1_ = t.f1_;
-    f2_ = t.f2_;
-    f3_ = t.f3_;
-    f4_ = t.f4_;
-    f5_ = t.f5_;
-    f6_ = t.f6_;
-    f7_ = t.f7_;
-    f8_ = t.f8_;
-    f9_ = t.f9_;
-    return *this;
-  }
-
-  T0 f0_;
-  T1 f1_;
-  T2 f2_;
-  T3 f3_;
-  T4 f4_;
-  T5 f5_;
-  T6 f6_;
-  T7 f7_;
-  T8 f8_;
-  T9 f9_;
-};
-
-// 6.1.3.2 Tuple creation functions.
-
-// Known limitations: we don't support passing an
-// std::tr1::reference_wrapper<T> to make_tuple().  And we don't
-// implement tie().
-
-inline tuple<> make_tuple() { return tuple<>(); }
-
-template <GTEST_1_TYPENAMES_(T)>
-inline GTEST_1_TUPLE_(T) make_tuple(const T0& f0) {
-  return GTEST_1_TUPLE_(T)(f0);
-}
-
-template <GTEST_2_TYPENAMES_(T)>
-inline GTEST_2_TUPLE_(T) make_tuple(const T0& f0, const T1& f1) {
-  return GTEST_2_TUPLE_(T)(f0, f1);
-}
-
-template <GTEST_3_TYPENAMES_(T)>
-inline GTEST_3_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2) {
-  return GTEST_3_TUPLE_(T)(f0, f1, f2);
-}
-
-template <GTEST_4_TYPENAMES_(T)>
-inline GTEST_4_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2,
-    const T3& f3) {
-  return GTEST_4_TUPLE_(T)(f0, f1, f2, f3);
-}
-
-template <GTEST_5_TYPENAMES_(T)>
-inline GTEST_5_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2,
-    const T3& f3, const T4& f4) {
-  return GTEST_5_TUPLE_(T)(f0, f1, f2, f3, f4);
-}
-
-template <GTEST_6_TYPENAMES_(T)>
-inline GTEST_6_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2,
-    const T3& f3, const T4& f4, const T5& f5) {
-  return GTEST_6_TUPLE_(T)(f0, f1, f2, f3, f4, f5);
-}
-
-template <GTEST_7_TYPENAMES_(T)>
-inline GTEST_7_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2,
-    const T3& f3, const T4& f4, const T5& f5, const T6& f6) {
-  return GTEST_7_TUPLE_(T)(f0, f1, f2, f3, f4, f5, f6);
-}
-
-template <GTEST_8_TYPENAMES_(T)>
-inline GTEST_8_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2,
-    const T3& f3, const T4& f4, const T5& f5, const T6& f6, const T7& f7) {
-  return GTEST_8_TUPLE_(T)(f0, f1, f2, f3, f4, f5, f6, f7);
-}
-
-template <GTEST_9_TYPENAMES_(T)>
-inline GTEST_9_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2,
-    const T3& f3, const T4& f4, const T5& f5, const T6& f6, const T7& f7,
-    const T8& f8) {
-  return GTEST_9_TUPLE_(T)(f0, f1, f2, f3, f4, f5, f6, f7, f8);
-}
-
-template <GTEST_10_TYPENAMES_(T)>
-inline GTEST_10_TUPLE_(T) make_tuple(const T0& f0, const T1& f1, const T2& f2,
-    const T3& f3, const T4& f4, const T5& f5, const T6& f6, const T7& f7,
-    const T8& f8, const T9& f9) {
-  return GTEST_10_TUPLE_(T)(f0, f1, f2, f3, f4, f5, f6, f7, f8, f9);
-}
-
-// 6.1.3.3 Tuple helper classes.
-
-template <typename Tuple> struct tuple_size;
-
-template <GTEST_0_TYPENAMES_(T)>
-struct tuple_size<GTEST_0_TUPLE_(T) > {
-  static const int value = 0;
-};
-
-template <GTEST_1_TYPENAMES_(T)>
-struct tuple_size<GTEST_1_TUPLE_(T) > {
-  static const int value = 1;
-};
-
-template <GTEST_2_TYPENAMES_(T)>
-struct tuple_size<GTEST_2_TUPLE_(T) > {
-  static const int value = 2;
-};
-
-template <GTEST_3_TYPENAMES_(T)>
-struct tuple_size<GTEST_3_TUPLE_(T) > {
-  static const int value = 3;
-};
-
-template <GTEST_4_TYPENAMES_(T)>
-struct tuple_size<GTEST_4_TUPLE_(T) > {
-  static const int value = 4;
-};
-
-template <GTEST_5_TYPENAMES_(T)>
-struct tuple_size<GTEST_5_TUPLE_(T) > {
-  static const int value = 5;
-};
-
-template <GTEST_6_TYPENAMES_(T)>
-struct tuple_size<GTEST_6_TUPLE_(T) > {
-  static const int value = 6;
-};
-
-template <GTEST_7_TYPENAMES_(T)>
-struct tuple_size<GTEST_7_TUPLE_(T) > {
-  static const int value = 7;
-};
-
-template <GTEST_8_TYPENAMES_(T)>
-struct tuple_size<GTEST_8_TUPLE_(T) > {
-  static const int value = 8;
-};
-
-template <GTEST_9_TYPENAMES_(T)>
-struct tuple_size<GTEST_9_TUPLE_(T) > {
-  static const int value = 9;
-};
-
-template <GTEST_10_TYPENAMES_(T)>
-struct tuple_size<GTEST_10_TUPLE_(T) > {
-  static const int value = 10;
-};
-
-template <int k, class Tuple>
-struct tuple_element {
-  typedef typename gtest_internal::TupleElement<
-      k < (tuple_size<Tuple>::value), k, Tuple>::type type;
-};
-
-#define GTEST_TUPLE_ELEMENT_(k, Tuple) typename tuple_element<k, Tuple >::type
-
-// 6.1.3.4 Element access.
-
-namespace gtest_internal {
-
-template <>
-class Get<0> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(0, Tuple))
-  Field(Tuple& t) { return t.f0_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(0, Tuple))
-  ConstField(const Tuple& t) { return t.f0_; }
-};
-
-template <>
-class Get<1> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(1, Tuple))
-  Field(Tuple& t) { return t.f1_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(1, Tuple))
-  ConstField(const Tuple& t) { return t.f1_; }
-};
-
-template <>
-class Get<2> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(2, Tuple))
-  Field(Tuple& t) { return t.f2_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(2, Tuple))
-  ConstField(const Tuple& t) { return t.f2_; }
-};
-
-template <>
-class Get<3> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(3, Tuple))
-  Field(Tuple& t) { return t.f3_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(3, Tuple))
-  ConstField(const Tuple& t) { return t.f3_; }
-};
-
-template <>
-class Get<4> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(4, Tuple))
-  Field(Tuple& t) { return t.f4_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(4, Tuple))
-  ConstField(const Tuple& t) { return t.f4_; }
-};
-
-template <>
-class Get<5> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(5, Tuple))
-  Field(Tuple& t) { return t.f5_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(5, Tuple))
-  ConstField(const Tuple& t) { return t.f5_; }
-};
-
-template <>
-class Get<6> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(6, Tuple))
-  Field(Tuple& t) { return t.f6_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(6, Tuple))
-  ConstField(const Tuple& t) { return t.f6_; }
-};
-
-template <>
-class Get<7> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(7, Tuple))
-  Field(Tuple& t) { return t.f7_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(7, Tuple))
-  ConstField(const Tuple& t) { return t.f7_; }
-};
-
-template <>
-class Get<8> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(8, Tuple))
-  Field(Tuple& t) { return t.f8_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(8, Tuple))
-  ConstField(const Tuple& t) { return t.f8_; }
-};
-
-template <>
-class Get<9> {
- public:
-  template <class Tuple>
-  static GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(9, Tuple))
-  Field(Tuple& t) { return t.f9_; }  // NOLINT
-
-  template <class Tuple>
-  static GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(9, Tuple))
-  ConstField(const Tuple& t) { return t.f9_; }
-};
-
-}  // namespace gtest_internal
-
-template <int k, GTEST_10_TYPENAMES_(T)>
-GTEST_ADD_REF_(GTEST_TUPLE_ELEMENT_(k, GTEST_10_TUPLE_(T)))
-get(GTEST_10_TUPLE_(T)& t) {
-  return gtest_internal::Get<k>::Field(t);
-}
-
-template <int k, GTEST_10_TYPENAMES_(T)>
-GTEST_BY_REF_(GTEST_TUPLE_ELEMENT_(k,  GTEST_10_TUPLE_(T)))
-get(const GTEST_10_TUPLE_(T)& t) {
-  return gtest_internal::Get<k>::ConstField(t);
-}
-
-// 6.1.3.5 Relational operators
-
-// We only implement == and !=, as we don't have a need for the rest yet.
-
-namespace gtest_internal {
-
-// SameSizeTuplePrefixComparator<k, k>::Eq(t1, t2) returns true if the
-// first k fields of t1 equals the first k fields of t2.
-// SameSizeTuplePrefixComparator(k1, k2) would be a compiler error if
-// k1 != k2.
-template <int kSize1, int kSize2>
-struct SameSizeTuplePrefixComparator;
-
-template <>
-struct SameSizeTuplePrefixComparator<0, 0> {
-  template <class Tuple1, class Tuple2>
-  static bool Eq(const Tuple1& /* t1 */, const Tuple2& /* t2 */) {
-    return true;
-  }
-};
-
-template <int k>
-struct SameSizeTuplePrefixComparator<k, k> {
-  template <class Tuple1, class Tuple2>
-  static bool Eq(const Tuple1& t1, const Tuple2& t2) {
-    return SameSizeTuplePrefixComparator<k - 1, k - 1>::Eq(t1, t2) &&
-        ::std::tr1::get<k - 1>(t1) == ::std::tr1::get<k - 1>(t2);
-  }
-};
-
-}  // namespace gtest_internal
-
-template <GTEST_10_TYPENAMES_(T), GTEST_10_TYPENAMES_(U)>
-inline bool operator==(const GTEST_10_TUPLE_(T)& t,
-                       const GTEST_10_TUPLE_(U)& u) {
-  return gtest_internal::SameSizeTuplePrefixComparator<
-      tuple_size<GTEST_10_TUPLE_(T) >::value,
-      tuple_size<GTEST_10_TUPLE_(U) >::value>::Eq(t, u);
-}
-
-template <GTEST_10_TYPENAMES_(T), GTEST_10_TYPENAMES_(U)>
-inline bool operator!=(const GTEST_10_TUPLE_(T)& t,
-                       const GTEST_10_TUPLE_(U)& u) { return !(t == u); }
-
-// 6.1.4 Pairs.
-// Unimplemented.
-
-}  // namespace tr1
-}  // namespace std
-
-#undef GTEST_0_TUPLE_
-#undef GTEST_1_TUPLE_
-#undef GTEST_2_TUPLE_
-#undef GTEST_3_TUPLE_
-#undef GTEST_4_TUPLE_
-#undef GTEST_5_TUPLE_
-#undef GTEST_6_TUPLE_
-#undef GTEST_7_TUPLE_
-#undef GTEST_8_TUPLE_
-#undef GTEST_9_TUPLE_
-#undef GTEST_10_TUPLE_
-
-#undef GTEST_0_TYPENAMES_
-#undef GTEST_1_TYPENAMES_
-#undef GTEST_2_TYPENAMES_
-#undef GTEST_3_TYPENAMES_
-#undef GTEST_4_TYPENAMES_
-#undef GTEST_5_TYPENAMES_
-#undef GTEST_6_TYPENAMES_
-#undef GTEST_7_TYPENAMES_
-#undef GTEST_8_TYPENAMES_
-#undef GTEST_9_TYPENAMES_
-#undef GTEST_10_TYPENAMES_
-
-#undef GTEST_DECLARE_TUPLE_AS_FRIEND_
-#undef GTEST_BY_REF_
-#undef GTEST_ADD_REF_
-#undef GTEST_TUPLE_ELEMENT_
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_TUPLE_H_
-# elif GTEST_ENV_HAS_STD_TUPLE_
-#  include <tuple>
-// C++11 puts its tuple into the ::std namespace rather than
-// ::std::tr1.  gtest expects tuple to live in ::std::tr1, so put it there.
-// This causes undefined behavior, but supported compilers react in
-// the way we intend.
-namespace std {
-namespace tr1 {
-using ::std::get;
-using ::std::make_tuple;
-using ::std::tuple;
-using ::std::tuple_element;
-using ::std::tuple_size;
-}
-}
-
-# elif GTEST_OS_SYMBIAN
-
-// On Symbian, BOOST_HAS_TR1_TUPLE causes Boost's TR1 tuple library to
-// use STLport's tuple implementation, which unfortunately doesn't
-// work as the copy of STLport distributed with Symbian is incomplete.
-// By making sure BOOST_HAS_TR1_TUPLE is undefined, we force Boost to
-// use its own tuple implementation.
-#  ifdef BOOST_HAS_TR1_TUPLE
-#   undef BOOST_HAS_TR1_TUPLE
-#  endif  // BOOST_HAS_TR1_TUPLE
-
-// This prevents <boost/tr1/detail/config.hpp>, which defines
-// BOOST_HAS_TR1_TUPLE, from being #included by Boost's <tuple>.
-#  define BOOST_TR1_DETAIL_CONFIG_HPP_INCLUDED
-#  include <tuple>  // IWYU pragma: export  // NOLINT
-
-# elif defined(__GNUC__) && (GTEST_GCC_VER_ >= 40000)
-// GCC 4.0+ implements tr1/tuple in the <tr1/tuple> header.  This does
-// not conform to the TR1 spec, which requires the header to be <tuple>.
-
-#  if !GTEST_HAS_RTTI && GTEST_GCC_VER_ < 40302
-// Until version 4.3.2, gcc has a bug that causes <tr1/functional>,
-// which is #included by <tr1/tuple>, to not compile when RTTI is
-// disabled.  _TR1_FUNCTIONAL is the header guard for
-// <tr1/functional>.  Hence the following #define is a hack to prevent
-// <tr1/functional> from being included.
-#   define _TR1_FUNCTIONAL 1
-#   include <tr1/tuple>
-#   undef _TR1_FUNCTIONAL  // Allows the user to #include
-                        // <tr1/functional> if he chooses to.
-#  else
-#   include <tr1/tuple>  // NOLINT
-#  endif  // !GTEST_HAS_RTTI && GTEST_GCC_VER_ < 40302
-
-# else
-// If the compiler is not GCC 4.0+, we assume the user is using a
-// spec-conforming TR1 implementation.
-#  include <tuple>  // IWYU pragma: export  // NOLINT
-# endif  // GTEST_USE_OWN_TR1_TUPLE
-
-#endif  // GTEST_HAS_TR1_TUPLE
-
-// Determines whether clone(2) is supported.
-// Usually it will only be available on Linux, excluding
-// Linux on the Itanium architecture.
-// Also see http://linux.die.net/man/2/clone.
-#ifndef GTEST_HAS_CLONE
-// The user didn't tell us, so we need to figure it out.
-
-# if GTEST_OS_LINUX && !defined(__ia64__)
-#  if GTEST_OS_LINUX_ANDROID
-// On Android, clone() became available at different API levels for each 32-bit
-// architecture.
-#    if defined(__LP64__) || \
-        (defined(__arm__) && __ANDROID_API__ >= 9) || \
-        (defined(__mips__) && __ANDROID_API__ >= 12) || \
-        (defined(__i386__) && __ANDROID_API__ >= 17)
-#     define GTEST_HAS_CLONE 1
-#    else
-#     define GTEST_HAS_CLONE 0
-#    endif
-#  else
-#   define GTEST_HAS_CLONE 1
-#  endif
-# else
-#  define GTEST_HAS_CLONE 0
-# endif  // GTEST_OS_LINUX && !defined(__ia64__)
-
-#endif  // GTEST_HAS_CLONE
-
-// Determines whether to support stream redirection. This is used to test
-// output correctness and to implement death tests.
-#ifndef GTEST_HAS_STREAM_REDIRECTION
-// By default, we assume that stream redirection is supported on all
-// platforms except known mobile ones.
-# if GTEST_OS_WINDOWS_MOBILE || GTEST_OS_SYMBIAN || \
-    GTEST_OS_WINDOWS_PHONE || GTEST_OS_WINDOWS_RT
-#  define GTEST_HAS_STREAM_REDIRECTION 0
-# else
-#  define GTEST_HAS_STREAM_REDIRECTION 1
-# endif  // !GTEST_OS_WINDOWS_MOBILE && !GTEST_OS_SYMBIAN
-#endif  // GTEST_HAS_STREAM_REDIRECTION
-
-// Determines whether to support death tests.
-// Google Test does not support death tests for VC 7.1 and earlier as
-// abort() in a VC 7.1 application compiled as GUI in debug config
-// pops up a dialog window that cannot be suppressed programmatically.
-#if (GTEST_OS_LINUX || GTEST_OS_CYGWIN || GTEST_OS_SOLARIS || \
-     (GTEST_OS_MAC && !GTEST_OS_IOS) || \
-     (GTEST_OS_WINDOWS_DESKTOP && _MSC_VER >= 1400) || \
-     GTEST_OS_WINDOWS_MINGW || GTEST_OS_AIX || GTEST_OS_HPUX || \
-     GTEST_OS_OPENBSD || GTEST_OS_QNX || GTEST_OS_FREEBSD)
-# define GTEST_HAS_DEATH_TEST 1
-#endif
-
-// We don't support MSVC 7.1 with exceptions disabled now.  Therefore
-// all the compilers we care about are adequate for supporting
-// value-parameterized tests.
-#define GTEST_HAS_PARAM_TEST 1
-
-// Determines whether to support type-driven tests.
-
-// Typed tests need <typeinfo> and variadic macros, which GCC, VC++ 8.0,
-// Sun Pro CC, IBM Visual Age, and HP aCC support.
-#if defined(__GNUC__) || (_MSC_VER >= 1400) || defined(__SUNPRO_CC) || \
-    defined(__IBMCPP__) || defined(__HP_aCC)
-# define GTEST_HAS_TYPED_TEST 1
-# define GTEST_HAS_TYPED_TEST_P 1
-#endif
-
-// Determines whether to support Combine(). This only makes sense when
-// value-parameterized tests are enabled.  The implementation doesn't
-// work on Sun Studio since it doesn't understand templated conversion
-// operators.
-#if GTEST_HAS_PARAM_TEST && GTEST_HAS_TR1_TUPLE && !defined(__SUNPRO_CC)
-# define GTEST_HAS_COMBINE 1
-#endif
-
-// Determines whether the system compiler uses UTF-16 for encoding wide strings.
-#define GTEST_WIDE_STRING_USES_UTF16_ \
-    (GTEST_OS_WINDOWS || GTEST_OS_CYGWIN || GTEST_OS_SYMBIAN || GTEST_OS_AIX)
-
-// Determines whether test results can be streamed to a socket.
-#if GTEST_OS_LINUX
-# define GTEST_CAN_STREAM_RESULTS_ 1
-#endif
-
-// Defines some utility macros.
-
-// The GNU compiler emits a warning if nested "if" statements are followed by
-// an "else" statement and braces are not used to explicitly disambiguate the
-// "else" binding.  This leads to problems with code like:
-//
-//   if (gate)
-//     ASSERT_*(condition) << "Some message";
-//
-// The "switch (0) case 0:" idiom is used to suppress this.
-#ifdef __INTEL_COMPILER
-# define GTEST_AMBIGUOUS_ELSE_BLOCKER_
-#else
-# define GTEST_AMBIGUOUS_ELSE_BLOCKER_ switch (0) case 0: default:  // NOLINT
-#endif
-
-// Use this annotation at the end of a struct/class definition to
-// prevent the compiler from optimizing away instances that are never
-// used.  This is useful when all interesting logic happens inside the
-// c'tor and / or d'tor.  Example:
-//
-//   struct Foo {
-//     Foo() { ... }
-//   } GTEST_ATTRIBUTE_UNUSED_;
-//
-// Also use it after a variable or parameter declaration to tell the
-// compiler the variable/parameter does not have to be used.
-#if defined(__GNUC__) && !defined(COMPILER_ICC)
-# define GTEST_ATTRIBUTE_UNUSED_ __attribute__ ((unused))
-#elif defined(__clang__)
-# if __has_attribute(unused)
-#  define GTEST_ATTRIBUTE_UNUSED_ __attribute__ ((unused))
-# endif
-#endif
-#ifndef GTEST_ATTRIBUTE_UNUSED_
-# define GTEST_ATTRIBUTE_UNUSED_
-#endif
-
-// A macro to disallow operator=
-// This should be used in the private: declarations for a class.
-#define GTEST_DISALLOW_ASSIGN_(type)\
-  void operator=(type const &)
-
-// A macro to disallow copy constructor and operator=
-// This should be used in the private: declarations for a class.
-#define GTEST_DISALLOW_COPY_AND_ASSIGN_(type)\
-  type(type const &);\
-  GTEST_DISALLOW_ASSIGN_(type)
-
-// Tell the compiler to warn about unused return values for functions declared
-// with this macro.  The macro should be used on function declarations
-// following the argument list:
-//
-//   Sprocket* AllocateSprocket() GTEST_MUST_USE_RESULT_;
-#if defined(__GNUC__) && (GTEST_GCC_VER_ >= 30400) && !defined(COMPILER_ICC)
-# define GTEST_MUST_USE_RESULT_ __attribute__ ((warn_unused_result))
-#else
-# define GTEST_MUST_USE_RESULT_
-#endif  // __GNUC__ && (GTEST_GCC_VER_ >= 30400) && !COMPILER_ICC
-
-// MS C++ compiler emits warning when a conditional expression is compile time
-// constant. In some contexts this warning is false positive and needs to be
-// suppressed. Use the following two macros in such cases:
-//
-// GTEST_INTENTIONAL_CONST_COND_PUSH_()
-// while (true) {
-// GTEST_INTENTIONAL_CONST_COND_POP_()
-// }
-# define GTEST_INTENTIONAL_CONST_COND_PUSH_() \
-    GTEST_DISABLE_MSC_WARNINGS_PUSH_(4127)
-# define GTEST_INTENTIONAL_CONST_COND_POP_() \
-    GTEST_DISABLE_MSC_WARNINGS_POP_()
-
-// Determine whether the compiler supports Microsoft's Structured Exception
-// Handling.  This is supported by several Windows compilers but generally
-// does not exist on any other system.
-#ifndef GTEST_HAS_SEH
-// The user didn't tell us, so we need to figure it out.
-
-# if defined(_MSC_VER) || defined(__BORLANDC__)
-// These two compilers are known to support SEH.
-#  define GTEST_HAS_SEH 1
-# else
-// Assume no SEH.
-#  define GTEST_HAS_SEH 0
-# endif
-
-#define GTEST_IS_THREADSAFE \
-    (GTEST_HAS_MUTEX_AND_THREAD_LOCAL_ \
-     || (GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT) \
-     || GTEST_HAS_PTHREAD)
-
-#endif  // GTEST_HAS_SEH
-
-#ifdef _MSC_VER
-# if GTEST_LINKED_AS_SHARED_LIBRARY
-#  define GTEST_API_ __declspec(dllimport)
-# elif GTEST_CREATE_SHARED_LIBRARY
-#  define GTEST_API_ __declspec(dllexport)
-# endif
-#elif __GNUC__ >= 4 || defined(__clang__)
-# define GTEST_API_ __attribute__((visibility ("default")))
-#endif // _MSC_VER
-
-#ifndef GTEST_API_
-# define GTEST_API_
-#endif
-
-#ifdef __GNUC__
-// Ask the compiler to never inline a given function.
-# define GTEST_NO_INLINE_ __attribute__((noinline))
-#else
-# define GTEST_NO_INLINE_
-#endif
-
-// _LIBCPP_VERSION is defined by the libc++ library from the LLVM project.
-#if defined(__GLIBCXX__) || defined(_LIBCPP_VERSION)
-# define GTEST_HAS_CXXABI_H_ 1
-#else
-# define GTEST_HAS_CXXABI_H_ 0
-#endif
-
-// A function level attribute to disable checking for use of uninitialized
-// memory when built with MemorySanitizer.
-#if defined(__clang__)
-# if __has_feature(memory_sanitizer)
-#  define GTEST_ATTRIBUTE_NO_SANITIZE_MEMORY_ \
-       __attribute__((no_sanitize_memory))
-# else
-#  define GTEST_ATTRIBUTE_NO_SANITIZE_MEMORY_
-# endif  // __has_feature(memory_sanitizer)
-#else
-# define GTEST_ATTRIBUTE_NO_SANITIZE_MEMORY_
-#endif  // __clang__
-
-// A function level attribute to disable AddressSanitizer instrumentation.
-#if defined(__clang__)
-# if __has_feature(address_sanitizer)
-#  define GTEST_ATTRIBUTE_NO_SANITIZE_ADDRESS_ \
-       __attribute__((no_sanitize_address))
-# else
-#  define GTEST_ATTRIBUTE_NO_SANITIZE_ADDRESS_
-# endif  // __has_feature(address_sanitizer)
-#else
-# define GTEST_ATTRIBUTE_NO_SANITIZE_ADDRESS_
-#endif  // __clang__
-
-// A function level attribute to disable ThreadSanitizer instrumentation.
-#if defined(__clang__)
-# if __has_feature(thread_sanitizer)
-#  define GTEST_ATTRIBUTE_NO_SANITIZE_THREAD_ \
-       __attribute__((no_sanitize_thread))
-# else
-#  define GTEST_ATTRIBUTE_NO_SANITIZE_THREAD_
-# endif  // __has_feature(thread_sanitizer)
-#else
-# define GTEST_ATTRIBUTE_NO_SANITIZE_THREAD_
-#endif  // __clang__
-
-namespace testing {
-
-class Message;
-
-#if defined(GTEST_TUPLE_NAMESPACE_)
-// Import tuple and friends into the ::testing namespace.
-// It is part of our interface, having them in ::testing allows us to change
-// their types as needed.
-using GTEST_TUPLE_NAMESPACE_::get;
-using GTEST_TUPLE_NAMESPACE_::make_tuple;
-using GTEST_TUPLE_NAMESPACE_::tuple;
-using GTEST_TUPLE_NAMESPACE_::tuple_size;
-using GTEST_TUPLE_NAMESPACE_::tuple_element;
-#endif  // defined(GTEST_TUPLE_NAMESPACE_)
-
-namespace internal {
-
-// A secret type that Google Test users don't know about.  It has no
-// definition on purpose.  Therefore it's impossible to create a
-// Secret object, which is what we want.
-class Secret;
-
-// The GTEST_COMPILE_ASSERT_ macro can be used to verify that a compile time
-// expression is true. For example, you could use it to verify the
-// size of a static array:
-//
-//   GTEST_COMPILE_ASSERT_(GTEST_ARRAY_SIZE_(names) == NUM_NAMES,
-//                         names_incorrect_size);
-//
-// or to make sure a struct is smaller than a certain size:
-//
-//   GTEST_COMPILE_ASSERT_(sizeof(foo) < 128, foo_too_large);
-//
-// The second argument to the macro is the name of the variable. If
-// the expression is false, most compilers will issue a warning/error
-// containing the name of the variable.
-
-#if GTEST_LANG_CXX11
-# define GTEST_COMPILE_ASSERT_(expr, msg) static_assert(expr, #msg)
-#else  // !GTEST_LANG_CXX11
-template <bool>
-  struct CompileAssert {
-};
-
-# define GTEST_COMPILE_ASSERT_(expr, msg) \
-  typedef ::testing::internal::CompileAssert<(static_cast<bool>(expr))> \
-      msg[static_cast<bool>(expr) ? 1 : -1] GTEST_ATTRIBUTE_UNUSED_
-#endif  // !GTEST_LANG_CXX11
-
-// Implementation details of GTEST_COMPILE_ASSERT_:
-//
-// (In C++11, we simply use static_assert instead of the following)
-//
-// - GTEST_COMPILE_ASSERT_ works by defining an array type that has -1
-//   elements (and thus is invalid) when the expression is false.
-//
-// - The simpler definition
-//
-//    #define GTEST_COMPILE_ASSERT_(expr, msg) typedef char msg[(expr) ? 1 : -1]
-//
-//   does not work, as gcc supports variable-length arrays whose sizes
-//   are determined at run-time (this is gcc's extension and not part
-//   of the C++ standard).  As a result, gcc fails to reject the
-//   following code with the simple definition:
-//
-//     int foo;
-//     GTEST_COMPILE_ASSERT_(foo, msg); // not supposed to compile as foo is
-//                                      // not a compile-time constant.
-//
-// - By using the type CompileAssert<(bool(expr))>, we ensures that
-//   expr is a compile-time constant.  (Template arguments must be
-//   determined at compile-time.)
-//
-// - The outter parentheses in CompileAssert<(bool(expr))> are necessary
-//   to work around a bug in gcc 3.4.4 and 4.0.1.  If we had written
-//
-//     CompileAssert<bool(expr)>
-//
-//   instead, these compilers will refuse to compile
-//
-//     GTEST_COMPILE_ASSERT_(5 > 0, some_message);
-//
-//   (They seem to think the ">" in "5 > 0" marks the end of the
-//   template argument list.)
-//
-// - The array size is (bool(expr) ? 1 : -1), instead of simply
-//
-//     ((expr) ? 1 : -1).
-//
-//   This is to avoid running into a bug in MS VC 7.1, which
-//   causes ((0.0) ? 1 : -1) to incorrectly evaluate to 1.
-
-// StaticAssertTypeEqHelper is used by StaticAssertTypeEq defined in gtest.h.
-//
-// This template is declared, but intentionally undefined.
-template <typename T1, typename T2>
-struct StaticAssertTypeEqHelper;
-
-template <typename T>
-struct StaticAssertTypeEqHelper<T, T> {
-  enum { value = true };
-};
-
-// Evaluates to the number of elements in 'array'.
-#define GTEST_ARRAY_SIZE_(array) (sizeof(array) / sizeof(array[0]))
-
-#if GTEST_HAS_GLOBAL_STRING
-typedef ::string string;
-#else
-typedef ::std::string string;
-#endif  // GTEST_HAS_GLOBAL_STRING
-
-#if GTEST_HAS_GLOBAL_WSTRING
-typedef ::wstring wstring;
-#elif GTEST_HAS_STD_WSTRING
-typedef ::std::wstring wstring;
-#endif  // GTEST_HAS_GLOBAL_WSTRING
-
-// A helper for suppressing warnings on constant condition.  It just
-// returns 'condition'.
-GTEST_API_ bool IsTrue(bool condition);
-
-// Defines scoped_ptr.
-
-// This implementation of scoped_ptr is PARTIAL - it only contains
-// enough stuff to satisfy Google Test's need.
-template <typename T>
-class scoped_ptr {
- public:
-  typedef T element_type;
-
-  explicit scoped_ptr(T* p = NULL) : ptr_(p) {}
-  ~scoped_ptr() { reset(); }
-
-  T& operator*() const { return *ptr_; }
-  T* operator->() const { return ptr_; }
-  T* get() const { return ptr_; }
-
-  T* release() {
-    T* const ptr = ptr_;
-    ptr_ = NULL;
-    return ptr;
-  }
-
-  void reset(T* p = NULL) {
-    if (p != ptr_) {
-      if (IsTrue(sizeof(T) > 0)) {  // Makes sure T is a complete type.
-        delete ptr_;
-      }
-      ptr_ = p;
-    }
-  }
-
-  friend void swap(scoped_ptr& a, scoped_ptr& b) {
-    using std::swap;
-    swap(a.ptr_, b.ptr_);
-  }
-
- private:
-  T* ptr_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(scoped_ptr);
-};
-
-// Defines RE.
-
-// A simple C++ wrapper for <regex.h>.  It uses the POSIX Extended
-// Regular Expression syntax.
-class GTEST_API_ RE {
- public:
-  // A copy constructor is required by the Standard to initialize object
-  // references from r-values.
-  RE(const RE& other) { Init(other.pattern()); }
-
-  // Constructs an RE from a string.
-  RE(const ::std::string& regex) { Init(regex.c_str()); }  // NOLINT
-
-#if GTEST_HAS_GLOBAL_STRING
-
-  RE(const ::string& regex) { Init(regex.c_str()); }  // NOLINT
-
-#endif  // GTEST_HAS_GLOBAL_STRING
-
-  RE(const char* regex) { Init(regex); }  // NOLINT
-  ~RE();
-
-  // Returns the string representation of the regex.
-  const char* pattern() const { return pattern_; }
-
-  // FullMatch(str, re) returns true iff regular expression re matches
-  // the entire str.
-  // PartialMatch(str, re) returns true iff regular expression re
-  // matches a substring of str (including str itself).
-  //
-  // TODO(wan@google.com): make FullMatch() and PartialMatch() work
-  // when str contains NUL characters.
-  static bool FullMatch(const ::std::string& str, const RE& re) {
-    return FullMatch(str.c_str(), re);
-  }
-  static bool PartialMatch(const ::std::string& str, const RE& re) {
-    return PartialMatch(str.c_str(), re);
-  }
-
-#if GTEST_HAS_GLOBAL_STRING
-
-  static bool FullMatch(const ::string& str, const RE& re) {
-    return FullMatch(str.c_str(), re);
-  }
-  static bool PartialMatch(const ::string& str, const RE& re) {
-    return PartialMatch(str.c_str(), re);
-  }
-
-#endif  // GTEST_HAS_GLOBAL_STRING
-
-  static bool FullMatch(const char* str, const RE& re);
-  static bool PartialMatch(const char* str, const RE& re);
-
- private:
-  void Init(const char* regex);
-
-  // We use a const char* instead of an std::string, as Google Test used to be
-  // used where std::string is not available.  TODO(wan@google.com): change to
-  // std::string.
-  const char* pattern_;
-  bool is_valid_;
-
-#if GTEST_USES_POSIX_RE
-
-  regex_t full_regex_;     // For FullMatch().
-  regex_t partial_regex_;  // For PartialMatch().
-
-#else  // GTEST_USES_SIMPLE_RE
-
-  const char* full_pattern_;  // For FullMatch();
-
-#endif
-
-  GTEST_DISALLOW_ASSIGN_(RE);
-};
-
-// Formats a source file path and a line number as they would appear
-// in an error message from the compiler used to compile this code.
-GTEST_API_ ::std::string FormatFileLocation(const char* file, int line);
-
-// Formats a file location for compiler-independent XML output.
-// Although this function is not platform dependent, we put it next to
-// FormatFileLocation in order to contrast the two functions.
-GTEST_API_ ::std::string FormatCompilerIndependentFileLocation(const char* file,
-                                                               int line);
-
-// Defines logging utilities:
-//   GTEST_LOG_(severity) - logs messages at the specified severity level. The
-//                          message itself is streamed into the macro.
-//   LogToStderr()  - directs all log messages to stderr.
-//   FlushInfoLog() - flushes informational log messages.
-
-enum GTestLogSeverity {
-  GTEST_INFO,
-  GTEST_WARNING,
-  GTEST_ERROR,
-  GTEST_FATAL
-};
-
-// Formats log entry severity, provides a stream object for streaming the
-// log message, and terminates the message with a newline when going out of
-// scope.
-class GTEST_API_ GTestLog {
- public:
-  GTestLog(GTestLogSeverity severity, const char* file, int line);
-
-  // Flushes the buffers and, if severity is GTEST_FATAL, aborts the program.
-  ~GTestLog();
-
-  ::std::ostream& GetStream() { return ::std::cerr; }
-
- private:
-  const GTestLogSeverity severity_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(GTestLog);
-};
-
-#if !defined(GTEST_LOG_)
-
-# define GTEST_LOG_(severity) \
-    ::testing::internal::GTestLog(::testing::internal::GTEST_##severity, \
-                                  __FILE__, __LINE__).GetStream()
-
-inline void LogToStderr() {}
-inline void FlushInfoLog() { fflush(NULL); }
-
-#endif  // !defined(GTEST_LOG_)
-
-#if !defined(GTEST_CHECK_)
-// INTERNAL IMPLEMENTATION - DO NOT USE.
-//
-// GTEST_CHECK_ is an all-mode assert. It aborts the program if the condition
-// is not satisfied.
-//  Synopsys:
-//    GTEST_CHECK_(boolean_condition);
-//     or
-//    GTEST_CHECK_(boolean_condition) << "Additional message";
-//
-//    This checks the condition and if the condition is not satisfied
-//    it prints message about the condition violation, including the
-//    condition itself, plus additional message streamed into it, if any,
-//    and then it aborts the program. It aborts the program irrespective of
-//    whether it is built in the debug mode or not.
-# define GTEST_CHECK_(condition) \
-    GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-    if (::testing::internal::IsTrue(condition)) \
-      ; \
-    else \
-      GTEST_LOG_(FATAL) << "Condition " #condition " failed. "
-#endif  // !defined(GTEST_CHECK_)
-
-// An all-mode assert to verify that the given POSIX-style function
-// call returns 0 (indicating success).  Known limitation: this
-// doesn't expand to a balanced 'if' statement, so enclose the macro
-// in {} if you need to use it as the only statement in an 'if'
-// branch.
-#define GTEST_CHECK_POSIX_SUCCESS_(posix_call) \
-  if (const int gtest_error = (posix_call)) \
-    GTEST_LOG_(FATAL) << #posix_call << "failed with error " \
-                      << gtest_error
-
-#if GTEST_HAS_STD_MOVE_
-using std::move;
-#else  // GTEST_HAS_STD_MOVE_
-template <typename T>
-const T& move(const T& t) {
-  return t;
-}
-#endif  // GTEST_HAS_STD_MOVE_
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Use ImplicitCast_ as a safe version of static_cast for upcasting in
-// the type hierarchy (e.g. casting a Foo* to a SuperclassOfFoo* or a
-// const Foo*).  When you use ImplicitCast_, the compiler checks that
-// the cast is safe.  Such explicit ImplicitCast_s are necessary in
-// surprisingly many situations where C++ demands an exact type match
-// instead of an argument type convertable to a target type.
-//
-// The syntax for using ImplicitCast_ is the same as for static_cast:
-//
-//   ImplicitCast_<ToType>(expr)
-//
-// ImplicitCast_ would have been part of the C++ standard library,
-// but the proposal was submitted too late.  It will probably make
-// its way into the language in the future.
-//
-// This relatively ugly name is intentional. It prevents clashes with
-// similar functions users may have (e.g., implicit_cast). The internal
-// namespace alone is not enough because the function can be found by ADL.
-template<typename To>
-inline To ImplicitCast_(To x) { return x; }
-
-// When you upcast (that is, cast a pointer from type Foo to type
-// SuperclassOfFoo), it's fine to use ImplicitCast_<>, since upcasts
-// always succeed.  When you downcast (that is, cast a pointer from
-// type Foo to type SubclassOfFoo), static_cast<> isn't safe, because
-// how do you know the pointer is really of type SubclassOfFoo?  It
-// could be a bare Foo, or of type DifferentSubclassOfFoo.  Thus,
-// when you downcast, you should use this macro.  In debug mode, we
-// use dynamic_cast<> to double-check the downcast is legal (we die
-// if it's not).  In normal mode, we do the efficient static_cast<>
-// instead.  Thus, it's important to test in debug mode to make sure
-// the cast is legal!
-//    This is the only place in the code we should use dynamic_cast<>.
-// In particular, you SHOULDN'T be using dynamic_cast<> in order to
-// do RTTI (eg code like this:
-//    if (dynamic_cast<Subclass1>(foo)) HandleASubclass1Object(foo);
-//    if (dynamic_cast<Subclass2>(foo)) HandleASubclass2Object(foo);
-// You should design the code some other way not to need this.
-//
-// This relatively ugly name is intentional. It prevents clashes with
-// similar functions users may have (e.g., down_cast). The internal
-// namespace alone is not enough because the function can be found by ADL.
-template<typename To, typename From>  // use like this: DownCast_<T*>(foo);
-inline To DownCast_(From* f) {  // so we only accept pointers
-  // Ensures that To is a sub-type of From *.  This test is here only
-  // for compile-time type checking, and has no overhead in an
-  // optimized build at run-time, as it will be optimized away
-  // completely.
-  GTEST_INTENTIONAL_CONST_COND_PUSH_()
-  if (false) {
-  GTEST_INTENTIONAL_CONST_COND_POP_()
-    const To to = NULL;
-    ::testing::internal::ImplicitCast_<From*>(to);
-  }
-
-#if GTEST_HAS_RTTI
-  // RTTI: debug mode only!
-  GTEST_CHECK_(f == NULL || dynamic_cast<To>(f) != NULL);
-#endif
-  return static_cast<To>(f);
-}
-
-// Downcasts the pointer of type Base to Derived.
-// Derived must be a subclass of Base. The parameter MUST
-// point to a class of type Derived, not any subclass of it.
-// When RTTI is available, the function performs a runtime
-// check to enforce this.
-template <class Derived, class Base>
-Derived* CheckedDowncastToActualType(Base* base) {
-#if GTEST_HAS_RTTI
-  GTEST_CHECK_(typeid(*base) == typeid(Derived));
-#endif
-
-#if GTEST_HAS_DOWNCAST_
-  return ::down_cast<Derived*>(base);
-#elif GTEST_HAS_RTTI
-  return dynamic_cast<Derived*>(base);  // NOLINT
-#else
-  return static_cast<Derived*>(base);  // Poor man's downcast.
-#endif
-}
-
-#if GTEST_HAS_STREAM_REDIRECTION
-
-// Defines the stderr capturer:
-//   CaptureStdout     - starts capturing stdout.
-//   GetCapturedStdout - stops capturing stdout and returns the captured string.
-//   CaptureStderr     - starts capturing stderr.
-//   GetCapturedStderr - stops capturing stderr and returns the captured string.
-//
-GTEST_API_ void CaptureStdout();
-GTEST_API_ std::string GetCapturedStdout();
-GTEST_API_ void CaptureStderr();
-GTEST_API_ std::string GetCapturedStderr();
-
-#endif  // GTEST_HAS_STREAM_REDIRECTION
-
-// Returns a path to temporary directory.
-GTEST_API_ std::string TempDir();
-
-// Returns the size (in bytes) of a file.
-GTEST_API_ size_t GetFileSize(FILE* file);
-
-// Reads the entire content of a file as a string.
-GTEST_API_ std::string ReadEntireFile(FILE* file);
-
-// All command line arguments.
-GTEST_API_ const ::std::vector<testing::internal::string>& GetArgvs();
-
-#if GTEST_HAS_DEATH_TEST
-
-const ::std::vector<testing::internal::string>& GetInjectableArgvs();
-void SetInjectableArgvs(const ::std::vector<testing::internal::string>*
-                             new_argvs);
-
-
-#endif  // GTEST_HAS_DEATH_TEST
-
-// Defines synchronization primitives.
-#if GTEST_IS_THREADSAFE
-# if GTEST_HAS_PTHREAD
-// Sleeps for (roughly) n milliseconds.  This function is only for testing
-// Google Test's own constructs.  Don't use it in user tests, either
-// directly or indirectly.
-inline void SleepMilliseconds(int n) {
-  const timespec time = {
-    0,                  // 0 seconds.
-    n * 1000L * 1000L,  // And n ms.
-  };
-  nanosleep(&time, NULL);
-}
-# endif  // GTEST_HAS_PTHREAD
-
-# if GTEST_HAS_NOTIFICATION_
-// Notification has already been imported into the namespace.
-// Nothing to do here.
-
-# elif GTEST_HAS_PTHREAD
-// Allows a controller thread to pause execution of newly created
-// threads until notified.  Instances of this class must be created
-// and destroyed in the controller thread.
-//
-// This class is only for testing Google Test's own constructs. Do not
-// use it in user tests, either directly or indirectly.
-class Notification {
- public:
-  Notification() : notified_(false) {
-    GTEST_CHECK_POSIX_SUCCESS_(pthread_mutex_init(&mutex_, NULL));
-  }
-  ~Notification() {
-    pthread_mutex_destroy(&mutex_);
-  }
-
-  // Notifies all threads created with this notification to start. Must
-  // be called from the controller thread.
-  void Notify() {
-    pthread_mutex_lock(&mutex_);
-    notified_ = true;
-    pthread_mutex_unlock(&mutex_);
-  }
-
-  // Blocks until the controller thread notifies. Must be called from a test
-  // thread.
-  void WaitForNotification() {
-    for (;;) {
-      pthread_mutex_lock(&mutex_);
-      const bool notified = notified_;
-      pthread_mutex_unlock(&mutex_);
-      if (notified)
-        break;
-      SleepMilliseconds(10);
-    }
-  }
-
- private:
-  pthread_mutex_t mutex_;
-  bool notified_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(Notification);
-};
-
-# elif GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT
-
-GTEST_API_ void SleepMilliseconds(int n);
-
-// Provides leak-safe Windows kernel handle ownership.
-// Used in death tests and in threading support.
-class GTEST_API_ AutoHandle {
- public:
-  // Assume that Win32 HANDLE type is equivalent to void*. Doing so allows us to
-  // avoid including <windows.h> in this header file. Including <windows.h> is
-  // undesirable because it defines a lot of symbols and macros that tend to
-  // conflict with client code. This assumption is verified by
-  // WindowsTypesTest.HANDLEIsVoidStar.
-  typedef void* Handle;
-  AutoHandle();
-  explicit AutoHandle(Handle handle);
-
-  ~AutoHandle();
-
-  Handle Get() const;
-  void Reset();
-  void Reset(Handle handle);
-
- private:
-  // Returns true iff the handle is a valid handle object that can be closed.
-  bool IsCloseable() const;
-
-  Handle handle_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(AutoHandle);
-};
-
-// Allows a controller thread to pause execution of newly created
-// threads until notified.  Instances of this class must be created
-// and destroyed in the controller thread.
-//
-// This class is only for testing Google Test's own constructs. Do not
-// use it in user tests, either directly or indirectly.
-class GTEST_API_ Notification {
- public:
-  Notification();
-  void Notify();
-  void WaitForNotification();
-
- private:
-  AutoHandle event_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(Notification);
-};
-# endif  // GTEST_HAS_NOTIFICATION_
-
-// On MinGW, we can have both GTEST_OS_WINDOWS and GTEST_HAS_PTHREAD
-// defined, but we don't want to use MinGW's pthreads implementation, which
-// has conformance problems with some versions of the POSIX standard.
-# if GTEST_HAS_PTHREAD && !GTEST_OS_WINDOWS_MINGW
-
-// As a C-function, ThreadFuncWithCLinkage cannot be templated itself.
-// Consequently, it cannot select a correct instantiation of ThreadWithParam
-// in order to call its Run(). Introducing ThreadWithParamBase as a
-// non-templated base class for ThreadWithParam allows us to bypass this
-// problem.
-class ThreadWithParamBase {
- public:
-  virtual ~ThreadWithParamBase() {}
-  virtual void Run() = 0;
-};
-
-// pthread_create() accepts a pointer to a function type with the C linkage.
-// According to the Standard (7.5/1), function types with different linkages
-// are different even if they are otherwise identical.  Some compilers (for
-// example, SunStudio) treat them as different types.  Since class methods
-// cannot be defined with C-linkage we need to define a free C-function to
-// pass into pthread_create().
-extern "C" inline void* ThreadFuncWithCLinkage(void* thread) {
-  static_cast<ThreadWithParamBase*>(thread)->Run();
-  return NULL;
-}
-
-// Helper class for testing Google Test's multi-threading constructs.
-// To use it, write:
-//
-//   void ThreadFunc(int param) { /* Do things with param */ }
-//   Notification thread_can_start;
-//   ...
-//   // The thread_can_start parameter is optional; you can supply NULL.
-//   ThreadWithParam<int> thread(&ThreadFunc, 5, &thread_can_start);
-//   thread_can_start.Notify();
-//
-// These classes are only for testing Google Test's own constructs. Do
-// not use them in user tests, either directly or indirectly.
-template <typename T>
-class ThreadWithParam : public ThreadWithParamBase {
- public:
-  typedef void UserThreadFunc(T);
-
-  ThreadWithParam(UserThreadFunc* func, T param, Notification* thread_can_start)
-      : func_(func),
-        param_(param),
-        thread_can_start_(thread_can_start),
-        finished_(false) {
-    ThreadWithParamBase* const base = this;
-    // The thread can be created only after all fields except thread_
-    // have been initialized.
-    GTEST_CHECK_POSIX_SUCCESS_(
-        pthread_create(&thread_, 0, &ThreadFuncWithCLinkage, base));
-  }
-  ~ThreadWithParam() { Join(); }
-
-  void Join() {
-    if (!finished_) {
-      GTEST_CHECK_POSIX_SUCCESS_(pthread_join(thread_, 0));
-      finished_ = true;
-    }
-  }
-
-  virtual void Run() {
-    if (thread_can_start_ != NULL)
-      thread_can_start_->WaitForNotification();
-    func_(param_);
-  }
-
- private:
-  UserThreadFunc* const func_;  // User-supplied thread function.
-  const T param_;  // User-supplied parameter to the thread function.
-  // When non-NULL, used to block execution until the controller thread
-  // notifies.
-  Notification* const thread_can_start_;
-  bool finished_;  // true iff we know that the thread function has finished.
-  pthread_t thread_;  // The native thread object.
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ThreadWithParam);
-};
-# endif  // !GTEST_OS_WINDOWS && GTEST_HAS_PTHREAD ||
-         // GTEST_HAS_MUTEX_AND_THREAD_LOCAL_
-
-# if GTEST_HAS_MUTEX_AND_THREAD_LOCAL_
-// Mutex and ThreadLocal have already been imported into the namespace.
-// Nothing to do here.
-
-# elif GTEST_OS_WINDOWS && !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT
-
-// Mutex implements mutex on Windows platforms.  It is used in conjunction
-// with class MutexLock:
-//
-//   Mutex mutex;
-//   ...
-//   MutexLock lock(&mutex);  // Acquires the mutex and releases it at the
-//                            // end of the current scope.
-//
-// A static Mutex *must* be defined or declared using one of the following
-// macros:
-//   GTEST_DEFINE_STATIC_MUTEX_(g_some_mutex);
-//   GTEST_DECLARE_STATIC_MUTEX_(g_some_mutex);
-//
-// (A non-static Mutex is defined/declared in the usual way).
-class GTEST_API_ Mutex {
- public:
-  enum MutexType { kStatic = 0, kDynamic = 1 };
-  // We rely on kStaticMutex being 0 as it is to what the linker initializes
-  // type_ in static mutexes.  critical_section_ will be initialized lazily
-  // in ThreadSafeLazyInit().
-  enum StaticConstructorSelector { kStaticMutex = 0 };
-
-  // This constructor intentionally does nothing.  It relies on type_ being
-  // statically initialized to 0 (effectively setting it to kStatic) and on
-  // ThreadSafeLazyInit() to lazily initialize the rest of the members.
-  explicit Mutex(StaticConstructorSelector /*dummy*/) {}
-
-  Mutex();
-  ~Mutex();
-
-  void Lock();
-
-  void Unlock();
-
-  // Does nothing if the current thread holds the mutex. Otherwise, crashes
-  // with high probability.
-  void AssertHeld();
-
- private:
-  // Initializes owner_thread_id_ and critical_section_ in static mutexes.
-  void ThreadSafeLazyInit();
-
-  // Per http://blogs.msdn.com/b/oldnewthing/archive/2004/02/23/78395.aspx,
-  // we assume that 0 is an invalid value for thread IDs.
-  unsigned int owner_thread_id_;
-
-  // For static mutexes, we rely on these members being initialized to zeros
-  // by the linker.
-  MutexType type_;
-  long critical_section_init_phase_;  // NOLINT
-  GTEST_CRITICAL_SECTION* critical_section_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(Mutex);
-};
-
-# define GTEST_DECLARE_STATIC_MUTEX_(mutex) \
-    extern ::testing::internal::Mutex mutex
-
-# define GTEST_DEFINE_STATIC_MUTEX_(mutex) \
-    ::testing::internal::Mutex mutex(::testing::internal::Mutex::kStaticMutex)
-
-// We cannot name this class MutexLock because the ctor declaration would
-// conflict with a macro named MutexLock, which is defined on some
-// platforms. That macro is used as a defensive measure to prevent against
-// inadvertent misuses of MutexLock like "MutexLock(&mu)" rather than
-// "MutexLock l(&mu)".  Hence the typedef trick below.
-class GTestMutexLock {
- public:
-  explicit GTestMutexLock(Mutex* mutex)
-      : mutex_(mutex) { mutex_->Lock(); }
-
-  ~GTestMutexLock() { mutex_->Unlock(); }
-
- private:
-  Mutex* const mutex_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(GTestMutexLock);
-};
-
-typedef GTestMutexLock MutexLock;
-
-// Base class for ValueHolder<T>.  Allows a caller to hold and delete a value
-// without knowing its type.
-class ThreadLocalValueHolderBase {
- public:
-  virtual ~ThreadLocalValueHolderBase() {}
-};
-
-// Provides a way for a thread to send notifications to a ThreadLocal
-// regardless of its parameter type.
-class ThreadLocalBase {
- public:
-  // Creates a new ValueHolder<T> object holding a default value passed to
-  // this ThreadLocal<T>'s constructor and returns it.  It is the caller's
-  // responsibility not to call this when the ThreadLocal<T> instance already
-  // has a value on the current thread.
-  virtual ThreadLocalValueHolderBase* NewValueForCurrentThread() const = 0;
-
- protected:
-  ThreadLocalBase() {}
-  virtual ~ThreadLocalBase() {}
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ThreadLocalBase);
-};
-
-// Maps a thread to a set of ThreadLocals that have values instantiated on that
-// thread and notifies them when the thread exits.  A ThreadLocal instance is
-// expected to persist until all threads it has values on have terminated.
-class GTEST_API_ ThreadLocalRegistry {
- public:
-  // Registers thread_local_instance as having value on the current thread.
-  // Returns a value that can be used to identify the thread from other threads.
-  static ThreadLocalValueHolderBase* GetValueOnCurrentThread(
-      const ThreadLocalBase* thread_local_instance);
-
-  // Invoked when a ThreadLocal instance is destroyed.
-  static void OnThreadLocalDestroyed(
-      const ThreadLocalBase* thread_local_instance);
-};
-
-class GTEST_API_ ThreadWithParamBase {
- public:
-  void Join();
-
- protected:
-  class Runnable {
-   public:
-    virtual ~Runnable() {}
-    virtual void Run() = 0;
-  };
-
-  ThreadWithParamBase(Runnable *runnable, Notification* thread_can_start);
-  virtual ~ThreadWithParamBase();
-
- private:
-  AutoHandle thread_;
-};
-
-// Helper class for testing Google Test's multi-threading constructs.
-template <typename T>
-class ThreadWithParam : public ThreadWithParamBase {
- public:
-  typedef void UserThreadFunc(T);
-
-  ThreadWithParam(UserThreadFunc* func, T param, Notification* thread_can_start)
-      : ThreadWithParamBase(new RunnableImpl(func, param), thread_can_start) {
-  }
-  virtual ~ThreadWithParam() {}
-
- private:
-  class RunnableImpl : public Runnable {
-   public:
-    RunnableImpl(UserThreadFunc* func, T param)
-        : func_(func),
-          param_(param) {
-    }
-    virtual ~RunnableImpl() {}
-    virtual void Run() {
-      func_(param_);
-    }
-
-   private:
-    UserThreadFunc* const func_;
-    const T param_;
-
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(RunnableImpl);
-  };
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ThreadWithParam);
-};
-
-// Implements thread-local storage on Windows systems.
-//
-//   // Thread 1
-//   ThreadLocal<int> tl(100);  // 100 is the default value for each thread.
-//
-//   // Thread 2
-//   tl.set(150);  // Changes the value for thread 2 only.
-//   EXPECT_EQ(150, tl.get());
-//
-//   // Thread 1
-//   EXPECT_EQ(100, tl.get());  // In thread 1, tl has the original value.
-//   tl.set(200);
-//   EXPECT_EQ(200, tl.get());
-//
-// The template type argument T must have a public copy constructor.
-// In addition, the default ThreadLocal constructor requires T to have
-// a public default constructor.
-//
-// The users of a TheadLocal instance have to make sure that all but one
-// threads (including the main one) using that instance have exited before
-// destroying it. Otherwise, the per-thread objects managed for them by the
-// ThreadLocal instance are not guaranteed to be destroyed on all platforms.
-//
-// Google Test only uses global ThreadLocal objects.  That means they
-// will die after main() has returned.  Therefore, no per-thread
-// object managed by Google Test will be leaked as long as all threads
-// using Google Test have exited when main() returns.
-template <typename T>
-class ThreadLocal : public ThreadLocalBase {
- public:
-  ThreadLocal() : default_factory_(new DefaultValueHolderFactory()) {}
-  explicit ThreadLocal(const T& value)
-      : default_factory_(new InstanceValueHolderFactory(value)) {}
-
-  ~ThreadLocal() { ThreadLocalRegistry::OnThreadLocalDestroyed(this); }
-
-  T* pointer() { return GetOrCreateValue(); }
-  const T* pointer() const { return GetOrCreateValue(); }
-  const T& get() const { return *pointer(); }
-  void set(const T& value) { *pointer() = value; }
-
- private:
-  // Holds a value of T.  Can be deleted via its base class without the caller
-  // knowing the type of T.
-  class ValueHolder : public ThreadLocalValueHolderBase {
-   public:
-    ValueHolder() : value_() {}
-    explicit ValueHolder(const T& value) : value_(value) {}
-
-    T* pointer() { return &value_; }
-
-   private:
-    T value_;
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(ValueHolder);
-  };
-
-
-  T* GetOrCreateValue() const {
-    return static_cast<ValueHolder*>(
-        ThreadLocalRegistry::GetValueOnCurrentThread(this))->pointer();
-  }
-
-  virtual ThreadLocalValueHolderBase* NewValueForCurrentThread() const {
-    return default_factory_->MakeNewHolder();
-  }
-
-  class ValueHolderFactory {
-   public:
-    ValueHolderFactory() {}
-    virtual ~ValueHolderFactory() {}
-    virtual ValueHolder* MakeNewHolder() const = 0;
-
-   private:
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(ValueHolderFactory);
-  };
-
-  class DefaultValueHolderFactory : public ValueHolderFactory {
-   public:
-    DefaultValueHolderFactory() {}
-    virtual ValueHolder* MakeNewHolder() const { return new ValueHolder(); }
-
-   private:
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(DefaultValueHolderFactory);
-  };
-
-  class InstanceValueHolderFactory : public ValueHolderFactory {
-   public:
-    explicit InstanceValueHolderFactory(const T& value) : value_(value) {}
-    virtual ValueHolder* MakeNewHolder() const {
-      return new ValueHolder(value_);
-    }
-
-   private:
-    const T value_;  // The value for each thread.
-
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(InstanceValueHolderFactory);
-  };
-
-  scoped_ptr<ValueHolderFactory> default_factory_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ThreadLocal);
-};
-
-# elif GTEST_HAS_PTHREAD
-
-// MutexBase and Mutex implement mutex on pthreads-based platforms.
-class MutexBase {
- public:
-  // Acquires this mutex.
-  void Lock() {
-    GTEST_CHECK_POSIX_SUCCESS_(pthread_mutex_lock(&mutex_));
-    owner_ = pthread_self();
-    has_owner_ = true;
-  }
-
-  // Releases this mutex.
-  void Unlock() {
-    // Since the lock is being released the owner_ field should no longer be
-    // considered valid. We don't protect writing to has_owner_ here, as it's
-    // the caller's responsibility to ensure that the current thread holds the
-    // mutex when this is called.
-    has_owner_ = false;
-    GTEST_CHECK_POSIX_SUCCESS_(pthread_mutex_unlock(&mutex_));
-  }
-
-  // Does nothing if the current thread holds the mutex. Otherwise, crashes
-  // with high probability.
-  void AssertHeld() const {
-    GTEST_CHECK_(has_owner_ && pthread_equal(owner_, pthread_self()))
-        << "The current thread is not holding the mutex @" << this;
-  }
-
-  // A static mutex may be used before main() is entered.  It may even
-  // be used before the dynamic initialization stage.  Therefore we
-  // must be able to initialize a static mutex object at link time.
-  // This means MutexBase has to be a POD and its member variables
-  // have to be public.
- public:
-  pthread_mutex_t mutex_;  // The underlying pthread mutex.
-  // has_owner_ indicates whether the owner_ field below contains a valid thread
-  // ID and is therefore safe to inspect (e.g., to use in pthread_equal()). All
-  // accesses to the owner_ field should be protected by a check of this field.
-  // An alternative might be to memset() owner_ to all zeros, but there's no
-  // guarantee that a zero'd pthread_t is necessarily invalid or even different
-  // from pthread_self().
-  bool has_owner_;
-  pthread_t owner_;  // The thread holding the mutex.
-};
-
-// Forward-declares a static mutex.
-#  define GTEST_DECLARE_STATIC_MUTEX_(mutex) \
-     extern ::testing::internal::MutexBase mutex
-
-// Defines and statically (i.e. at link time) initializes a static mutex.
-#  define GTEST_DEFINE_STATIC_MUTEX_(mutex) \
-     ::testing::internal::MutexBase mutex = { PTHREAD_MUTEX_INITIALIZER, false, pthread_t() }
-
-// The Mutex class can only be used for mutexes created at runtime. It
-// shares its API with MutexBase otherwise.
-class Mutex : public MutexBase {
- public:
-  Mutex() {
-    GTEST_CHECK_POSIX_SUCCESS_(pthread_mutex_init(&mutex_, NULL));
-    has_owner_ = false;
-  }
-  ~Mutex() {
-    GTEST_CHECK_POSIX_SUCCESS_(pthread_mutex_destroy(&mutex_));
-  }
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(Mutex);
-};
-
-// We cannot name this class MutexLock because the ctor declaration would
-// conflict with a macro named MutexLock, which is defined on some
-// platforms. That macro is used as a defensive measure to prevent against
-// inadvertent misuses of MutexLock like "MutexLock(&mu)" rather than
-// "MutexLock l(&mu)".  Hence the typedef trick below.
-class GTestMutexLock {
- public:
-  explicit GTestMutexLock(MutexBase* mutex)
-      : mutex_(mutex) { mutex_->Lock(); }
-
-  ~GTestMutexLock() { mutex_->Unlock(); }
-
- private:
-  MutexBase* const mutex_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(GTestMutexLock);
-};
-
-typedef GTestMutexLock MutexLock;
-
-// Helpers for ThreadLocal.
-
-// pthread_key_create() requires DeleteThreadLocalValue() to have
-// C-linkage.  Therefore it cannot be templatized to access
-// ThreadLocal<T>.  Hence the need for class
-// ThreadLocalValueHolderBase.
-class ThreadLocalValueHolderBase {
- public:
-  virtual ~ThreadLocalValueHolderBase() {}
-};
-
-// Called by pthread to delete thread-local data stored by
-// pthread_setspecific().
-extern "C" inline void DeleteThreadLocalValue(void* value_holder) {
-  delete static_cast<ThreadLocalValueHolderBase*>(value_holder);
-}
-
-// Implements thread-local storage on pthreads-based systems.
-template <typename T>
-class ThreadLocal {
- public:
-  ThreadLocal()
-      : key_(CreateKey()), default_factory_(new DefaultValueHolderFactory()) {}
-  explicit ThreadLocal(const T& value)
-      : key_(CreateKey()),
-        default_factory_(new InstanceValueHolderFactory(value)) {}
-
-  ~ThreadLocal() {
-    // Destroys the managed object for the current thread, if any.
-    DeleteThreadLocalValue(pthread_getspecific(key_));
-
-    // Releases resources associated with the key.  This will *not*
-    // delete managed objects for other threads.
-    GTEST_CHECK_POSIX_SUCCESS_(pthread_key_delete(key_));
-  }
-
-  T* pointer() { return GetOrCreateValue(); }
-  const T* pointer() const { return GetOrCreateValue(); }
-  const T& get() const { return *pointer(); }
-  void set(const T& value) { *pointer() = value; }
-
- private:
-  // Holds a value of type T.
-  class ValueHolder : public ThreadLocalValueHolderBase {
-   public:
-    ValueHolder() : value_() {}
-    explicit ValueHolder(const T& value) : value_(value) {}
-
-    T* pointer() { return &value_; }
-
-   private:
-    T value_;
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(ValueHolder);
-  };
-
-  static pthread_key_t CreateKey() {
-    pthread_key_t key;
-    // When a thread exits, DeleteThreadLocalValue() will be called on
-    // the object managed for that thread.
-    GTEST_CHECK_POSIX_SUCCESS_(
-        pthread_key_create(&key, &DeleteThreadLocalValue));
-    return key;
-  }
-
-  T* GetOrCreateValue() const {
-    ThreadLocalValueHolderBase* const holder =
-        static_cast<ThreadLocalValueHolderBase*>(pthread_getspecific(key_));
-    if (holder != NULL) {
-      return CheckedDowncastToActualType<ValueHolder>(holder)->pointer();
-    }
-
-    ValueHolder* const new_holder = default_factory_->MakeNewHolder();
-    ThreadLocalValueHolderBase* const holder_base = new_holder;
-    GTEST_CHECK_POSIX_SUCCESS_(pthread_setspecific(key_, holder_base));
-    return new_holder->pointer();
-  }
-
-  class ValueHolderFactory {
-   public:
-    ValueHolderFactory() {}
-    virtual ~ValueHolderFactory() {}
-    virtual ValueHolder* MakeNewHolder() const = 0;
-
-   private:
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(ValueHolderFactory);
-  };
-
-  class DefaultValueHolderFactory : public ValueHolderFactory {
-   public:
-    DefaultValueHolderFactory() {}
-    virtual ValueHolder* MakeNewHolder() const { return new ValueHolder(); }
-
-   private:
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(DefaultValueHolderFactory);
-  };
-
-  class InstanceValueHolderFactory : public ValueHolderFactory {
-   public:
-    explicit InstanceValueHolderFactory(const T& value) : value_(value) {}
-    virtual ValueHolder* MakeNewHolder() const {
-      return new ValueHolder(value_);
-    }
-
-   private:
-    const T value_;  // The value for each thread.
-
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(InstanceValueHolderFactory);
-  };
-
-  // A key pthreads uses for looking up per-thread values.
-  const pthread_key_t key_;
-  scoped_ptr<ValueHolderFactory> default_factory_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ThreadLocal);
-};
-
-# endif  // GTEST_HAS_MUTEX_AND_THREAD_LOCAL_
-
-#else  // GTEST_IS_THREADSAFE
-
-// A dummy implementation of synchronization primitives (mutex, lock,
-// and thread-local variable).  Necessary for compiling Google Test where
-// mutex is not supported - using Google Test in multiple threads is not
-// supported on such platforms.
-
-class Mutex {
- public:
-  Mutex() {}
-  void Lock() {}
-  void Unlock() {}
-  void AssertHeld() const {}
-};
-
-# define GTEST_DECLARE_STATIC_MUTEX_(mutex) \
-  extern ::testing::internal::Mutex mutex
-
-# define GTEST_DEFINE_STATIC_MUTEX_(mutex) ::testing::internal::Mutex mutex
-
-// We cannot name this class MutexLock because the ctor declaration would
-// conflict with a macro named MutexLock, which is defined on some
-// platforms. That macro is used as a defensive measure to prevent against
-// inadvertent misuses of MutexLock like "MutexLock(&mu)" rather than
-// "MutexLock l(&mu)".  Hence the typedef trick below.
-class GTestMutexLock {
- public:
-  explicit GTestMutexLock(Mutex*) {}  // NOLINT
-};
-
-typedef GTestMutexLock MutexLock;
-
-template <typename T>
-class ThreadLocal {
- public:
-  ThreadLocal() : value_() {}
-  explicit ThreadLocal(const T& value) : value_(value) {}
-  T* pointer() { return &value_; }
-  const T* pointer() const { return &value_; }
-  const T& get() const { return value_; }
-  void set(const T& value) { value_ = value; }
- private:
-  T value_;
-};
-
-#endif  // GTEST_IS_THREADSAFE
-
-// Returns the number of threads running in the process, or 0 to indicate that
-// we cannot detect it.
-GTEST_API_ size_t GetThreadCount();
-
-// Passing non-POD classes through ellipsis (...) crashes the ARM
-// compiler and generates a warning in Sun Studio.  The Nokia Symbian
-// and the IBM XL C/C++ compiler try to instantiate a copy constructor
-// for objects passed through ellipsis (...), failing for uncopyable
-// objects.  We define this to ensure that only POD is passed through
-// ellipsis on these systems.
-#if defined(__SYMBIAN32__) || defined(__IBMCPP__) || defined(__SUNPRO_CC)
-// We lose support for NULL detection where the compiler doesn't like
-// passing non-POD classes through ellipsis (...).
-# define GTEST_ELLIPSIS_NEEDS_POD_ 1
-#else
-# define GTEST_CAN_COMPARE_NULL 1
-#endif
-
-// The Nokia Symbian and IBM XL C/C++ compilers cannot decide between
-// const T& and const T* in a function template.  These compilers
-// _can_ decide between class template specializations for T and T*,
-// so a tr1::type_traits-like is_pointer works.
-#if defined(__SYMBIAN32__) || defined(__IBMCPP__)
-# define GTEST_NEEDS_IS_POINTER_ 1
-#endif
-
-template <bool bool_value>
-struct bool_constant {
-  typedef bool_constant<bool_value> type;
-  static const bool value = bool_value;
-};
-template <bool bool_value> const bool bool_constant<bool_value>::value;
-
-typedef bool_constant<false> false_type;
-typedef bool_constant<true> true_type;
-
-template <typename T>
-struct is_pointer : public false_type {};
-
-template <typename T>
-struct is_pointer<T*> : public true_type {};
-
-template <typename Iterator>
-struct IteratorTraits {
-  typedef typename Iterator::value_type value_type;
-};
-
-template <typename T>
-struct IteratorTraits<T*> {
-  typedef T value_type;
-};
-
-template <typename T>
-struct IteratorTraits<const T*> {
-  typedef T value_type;
-};
-
-#if GTEST_OS_WINDOWS
-# define GTEST_PATH_SEP_ "\\"
-# define GTEST_HAS_ALT_PATH_SEP_ 1
-// The biggest signed integer type the compiler supports.
-typedef __int64 BiggestInt;
-#else
-# define GTEST_PATH_SEP_ "/"
-# define GTEST_HAS_ALT_PATH_SEP_ 0
-typedef long long BiggestInt;  // NOLINT
-#endif  // GTEST_OS_WINDOWS
-
-// Utilities for char.
-
-// isspace(int ch) and friends accept an unsigned char or EOF.  char
-// may be signed, depending on the compiler (or compiler flags).
-// Therefore we need to cast a char to unsigned char before calling
-// isspace(), etc.
-
-inline bool IsAlpha(char ch) {
-  return isalpha(static_cast<unsigned char>(ch)) != 0;
-}
-inline bool IsAlNum(char ch) {
-  return isalnum(static_cast<unsigned char>(ch)) != 0;
-}
-inline bool IsDigit(char ch) {
-  return isdigit(static_cast<unsigned char>(ch)) != 0;
-}
-inline bool IsLower(char ch) {
-  return islower(static_cast<unsigned char>(ch)) != 0;
-}
-inline bool IsSpace(char ch) {
-  return isspace(static_cast<unsigned char>(ch)) != 0;
-}
-inline bool IsUpper(char ch) {
-  return isupper(static_cast<unsigned char>(ch)) != 0;
-}
-inline bool IsXDigit(char ch) {
-  return isxdigit(static_cast<unsigned char>(ch)) != 0;
-}
-inline bool IsXDigit(wchar_t ch) {
-  const unsigned char low_byte = static_cast<unsigned char>(ch);
-  return ch == low_byte && isxdigit(low_byte) != 0;
-}
-
-inline char ToLower(char ch) {
-  return static_cast<char>(tolower(static_cast<unsigned char>(ch)));
-}
-inline char ToUpper(char ch) {
-  return static_cast<char>(toupper(static_cast<unsigned char>(ch)));
-}
-
-inline std::string StripTrailingSpaces(std::string str) {
-  std::string::iterator it = str.end();
-  while (it != str.begin() && IsSpace(*--it))
-    it = str.erase(it);
-  return str;
-}
-
-// The testing::internal::posix namespace holds wrappers for common
-// POSIX functions.  These wrappers hide the differences between
-// Windows/MSVC and POSIX systems.  Since some compilers define these
-// standard functions as macros, the wrapper cannot have the same name
-// as the wrapped function.
-
-namespace posix {
-
-// Functions with a different name on Windows.
-
-#if GTEST_OS_WINDOWS
-
-typedef struct _stat StatStruct;
-
-# ifdef __BORLANDC__
-inline int IsATTY(int fd) { return isatty(fd); }
-inline int StrCaseCmp(const char* s1, const char* s2) {
-  return stricmp(s1, s2);
-}
-inline char* StrDup(const char* src) { return strdup(src); }
-# else  // !__BORLANDC__
-#  if GTEST_OS_WINDOWS_MOBILE
-inline int IsATTY(int /* fd */) { return 0; }
-#  else
-inline int IsATTY(int fd) { return _isatty(fd); }
-#  endif  // GTEST_OS_WINDOWS_MOBILE
-inline int StrCaseCmp(const char* s1, const char* s2) {
-  return _stricmp(s1, s2);
-}
-inline char* StrDup(const char* src) { return _strdup(src); }
-# endif  // __BORLANDC__
-
-# if GTEST_OS_WINDOWS_MOBILE
-inline int FileNo(FILE* file) { return reinterpret_cast<int>(_fileno(file)); }
-// Stat(), RmDir(), and IsDir() are not needed on Windows CE at this
-// time and thus not defined there.
-# else
-inline int FileNo(FILE* file) { return _fileno(file); }
-inline int Stat(const char* path, StatStruct* buf) { return _stat(path, buf); }
-inline int RmDir(const char* dir) { return _rmdir(dir); }
-inline bool IsDir(const StatStruct& st) {
-  return (_S_IFDIR & st.st_mode) != 0;
-}
-# endif  // GTEST_OS_WINDOWS_MOBILE
-
-#else
-
-typedef struct stat StatStruct;
-
-inline int FileNo(FILE* file) { return fileno(file); }
-inline int IsATTY(int fd) { return isatty(fd); }
-inline int Stat(const char* path, StatStruct* buf) { return stat(path, buf); }
-inline int StrCaseCmp(const char* s1, const char* s2) {
-  return strcasecmp(s1, s2);
-}
-inline char* StrDup(const char* src) { return strdup(src); }
-inline int RmDir(const char* dir) { return rmdir(dir); }
-inline bool IsDir(const StatStruct& st) { return S_ISDIR(st.st_mode); }
-
-#endif  // GTEST_OS_WINDOWS
-
-// Functions deprecated by MSVC 8.0.
-
-GTEST_DISABLE_MSC_WARNINGS_PUSH_(4996 /* deprecated function */)
-
-inline const char* StrNCpy(char* dest, const char* src, size_t n) {
-  return strncpy(dest, src, n);
-}
-
-// ChDir(), FReopen(), FDOpen(), Read(), Write(), Close(), and
-// StrError() aren't needed on Windows CE at this time and thus not
-// defined there.
-
-#if !GTEST_OS_WINDOWS_MOBILE && !GTEST_OS_WINDOWS_PHONE && !GTEST_OS_WINDOWS_RT
-inline int ChDir(const char* dir) { return chdir(dir); }
-#endif
-inline FILE* FOpen(const char* path, const char* mode) {
-  return fopen(path, mode);
-}
-#if !GTEST_OS_WINDOWS_MOBILE
-inline FILE *FReopen(const char* path, const char* mode, FILE* stream) {
-  return freopen(path, mode, stream);
-}
-inline FILE* FDOpen(int fd, const char* mode) { return fdopen(fd, mode); }
-#endif
-inline int FClose(FILE* fp) { return fclose(fp); }
-#if !GTEST_OS_WINDOWS_MOBILE
-inline int Read(int fd, void* buf, unsigned int count) {
-  return static_cast<int>(read(fd, buf, count));
-}
-inline int Write(int fd, const void* buf, unsigned int count) {
-  return static_cast<int>(write(fd, buf, count));
-}
-inline int Close(int fd) { return close(fd); }
-inline const char* StrError(int errnum) { return strerror(errnum); }
-#endif
-inline const char* GetEnv(const char* name) {
-#if GTEST_OS_WINDOWS_MOBILE || GTEST_OS_WINDOWS_PHONE | GTEST_OS_WINDOWS_RT
-  // We are on Windows CE, which has no environment variables.
-  static_cast<void>(name);  // To prevent 'unused argument' warning.
-  return NULL;
-#elif defined(__BORLANDC__) || defined(__SunOS_5_8) || defined(__SunOS_5_9)
-  // Environment variables which we programmatically clear will be set to the
-  // empty string rather than unset (NULL).  Handle that case.
-  const char* const env = getenv(name);
-  return (env != NULL && env[0] != '\0') ? env : NULL;
-#else
-  return getenv(name);
-#endif
-}
-
-GTEST_DISABLE_MSC_WARNINGS_POP_()
-
-#if GTEST_OS_WINDOWS_MOBILE
-// Windows CE has no C library. The abort() function is used in
-// several places in Google Test. This implementation provides a reasonable
-// imitation of standard behaviour.
-void Abort();
-#else
-inline void Abort() { abort(); }
-#endif  // GTEST_OS_WINDOWS_MOBILE
-
-}  // namespace posix
-
-// MSVC "deprecates" snprintf and issues warnings wherever it is used.  In
-// order to avoid these warnings, we need to use _snprintf or _snprintf_s on
-// MSVC-based platforms.  We map the GTEST_SNPRINTF_ macro to the appropriate
-// function in order to achieve that.  We use macro definition here because
-// snprintf is a variadic function.
-#if _MSC_VER >= 1400 && !GTEST_OS_WINDOWS_MOBILE
-// MSVC 2005 and above support variadic macros.
-# define GTEST_SNPRINTF_(buffer, size, format, ...) \
-     _snprintf_s(buffer, size, size, format, __VA_ARGS__)
-#elif defined(_MSC_VER)
-// Windows CE does not define _snprintf_s and MSVC prior to 2005 doesn't
-// complain about _snprintf.
-# define GTEST_SNPRINTF_ _snprintf
-#else
-# define GTEST_SNPRINTF_ snprintf
-#endif
-
-// The maximum number a BiggestInt can represent.  This definition
-// works no matter BiggestInt is represented in one's complement or
-// two's complement.
-//
-// We cannot rely on numeric_limits in STL, as __int64 and long long
-// are not part of standard C++ and numeric_limits doesn't need to be
-// defined for them.
-const BiggestInt kMaxBiggestInt =
-    ~(static_cast<BiggestInt>(1) << (8*sizeof(BiggestInt) - 1));
-
-// This template class serves as a compile-time function from size to
-// type.  It maps a size in bytes to a primitive type with that
-// size. e.g.
-//
-//   TypeWithSize<4>::UInt
-//
-// is typedef-ed to be unsigned int (unsigned integer made up of 4
-// bytes).
-//
-// Such functionality should belong to STL, but I cannot find it
-// there.
-//
-// Google Test uses this class in the implementation of floating-point
-// comparison.
-//
-// For now it only handles UInt (unsigned int) as that's all Google Test
-// needs.  Other types can be easily added in the future if need
-// arises.
-template <size_t size>
-class TypeWithSize {
- public:
-  // This prevents the user from using TypeWithSize<N> with incorrect
-  // values of N.
-  typedef void UInt;
-};
-
-// The specialization for size 4.
-template <>
-class TypeWithSize<4> {
- public:
-  // unsigned int has size 4 in both gcc and MSVC.
-  //
-  // As base/basictypes.h doesn't compile on Windows, we cannot use
-  // uint32, uint64, and etc here.
-  typedef int Int;
-  typedef unsigned int UInt;
-};
-
-// The specialization for size 8.
-template <>
-class TypeWithSize<8> {
- public:
-#if GTEST_OS_WINDOWS
-  typedef __int64 Int;
-  typedef unsigned __int64 UInt;
-#else
-  typedef long long Int;  // NOLINT
-  typedef unsigned long long UInt;  // NOLINT
-#endif  // GTEST_OS_WINDOWS
-};
-
-// Integer types of known sizes.
-typedef TypeWithSize<4>::Int Int32;
-typedef TypeWithSize<4>::UInt UInt32;
-typedef TypeWithSize<8>::Int Int64;
-typedef TypeWithSize<8>::UInt UInt64;
-typedef TypeWithSize<8>::Int TimeInMillis;  // Represents time in milliseconds.
-
-// Utilities for command line flags and environment variables.
-
-// Macro for referencing flags.
-#if !defined(GTEST_FLAG)
-# define GTEST_FLAG(name) FLAGS_gtest_##name
-#endif  // !defined(GTEST_FLAG)
-
-#if !defined(GTEST_USE_OWN_FLAGFILE_FLAG_)
-# define GTEST_USE_OWN_FLAGFILE_FLAG_ 1
-#endif  // !defined(GTEST_USE_OWN_FLAGFILE_FLAG_)
-
-#if !defined(GTEST_DECLARE_bool_)
-# define GTEST_FLAG_SAVER_ ::testing::internal::GTestFlagSaver
-
-// Macros for declaring flags.
-# define GTEST_DECLARE_bool_(name) GTEST_API_ extern bool GTEST_FLAG(name)
-# define GTEST_DECLARE_int32_(name) \
-    GTEST_API_ extern ::testing::internal::Int32 GTEST_FLAG(name)
-#define GTEST_DECLARE_string_(name) \
-    GTEST_API_ extern ::std::string GTEST_FLAG(name)
-
-// Macros for defining flags.
-#define GTEST_DEFINE_bool_(name, default_val, doc) \
-    GTEST_API_ bool GTEST_FLAG(name) = (default_val)
-#define GTEST_DEFINE_int32_(name, default_val, doc) \
-    GTEST_API_ ::testing::internal::Int32 GTEST_FLAG(name) = (default_val)
-#define GTEST_DEFINE_string_(name, default_val, doc) \
-    GTEST_API_ ::std::string GTEST_FLAG(name) = (default_val)
-
-#endif  // !defined(GTEST_DECLARE_bool_)
-
-// Thread annotations
-#if !defined(GTEST_EXCLUSIVE_LOCK_REQUIRED_)
-# define GTEST_EXCLUSIVE_LOCK_REQUIRED_(locks)
-# define GTEST_LOCK_EXCLUDED_(locks)
-#endif  // !defined(GTEST_EXCLUSIVE_LOCK_REQUIRED_)
-
-// Parses 'str' for a 32-bit signed integer.  If successful, writes the result
-// to *value and returns true; otherwise leaves *value unchanged and returns
-// false.
-// TODO(chandlerc): Find a better way to refactor flag and environment parsing
-// out of both gtest-port.cc and gtest.cc to avoid exporting this utility
-// function.
-bool ParseInt32(const Message& src_text, const char* str, Int32* value);
-
-// Parses a bool/Int32/string from the environment variable
-// corresponding to the given Google Test flag.
-bool BoolFromGTestEnv(const char* flag, bool default_val);
-GTEST_API_ Int32 Int32FromGTestEnv(const char* flag, Int32 default_val);
-std::string StringFromGTestEnv(const char* flag, const char* default_val);
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PORT_H_
-
-#if GTEST_OS_LINUX
-# include <stdlib.h>
-# include <sys/types.h>
-# include <sys/wait.h>
-# include <unistd.h>
-#endif  // GTEST_OS_LINUX
-
-#if GTEST_HAS_EXCEPTIONS
-# include <stdexcept>
-#endif
-
-#include <ctype.h>
-#include <float.h>
-#include <string.h>
-#include <iomanip>
-#include <limits>
-#include <map>
-#include <set>
-#include <string>
-#include <vector>
-
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-//
-// The Google C++ Testing Framework (Google Test)
-//
-// This header file defines the Message class.
-//
-// IMPORTANT NOTE: Due to limitation of the C++ language, we have to
-// leave some internal implementation details in this header file.
-// They are clearly marked by comments like this:
-//
-//   // INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-//
-// Such code is NOT meant to be used by a user directly, and is subject
-// to CHANGE WITHOUT NOTICE.  Therefore DO NOT DEPEND ON IT in a user
-// program!
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_MESSAGE_H_
-#define GTEST_INCLUDE_GTEST_GTEST_MESSAGE_H_
-
-#include <limits>
-
-
-// Ensures that there is at least one operator<< in the global namespace.
-// See Message& operator<<(...) below for why.
-void operator<<(const testing::internal::Secret&, int);
-
-namespace testing {
-
-// The Message class works like an ostream repeater.
-//
-// Typical usage:
-//
-//   1. You stream a bunch of values to a Message object.
-//      It will remember the text in a stringstream.
-//   2. Then you stream the Message object to an ostream.
-//      This causes the text in the Message to be streamed
-//      to the ostream.
-//
-// For example;
-//
-//   testing::Message foo;
-//   foo << 1 << " != " << 2;
-//   std::cout << foo;
-//
-// will print "1 != 2".
-//
-// Message is not intended to be inherited from.  In particular, its
-// destructor is not virtual.
-//
-// Note that stringstream behaves differently in gcc and in MSVC.  You
-// can stream a NULL char pointer to it in the former, but not in the
-// latter (it causes an access violation if you do).  The Message
-// class hides this difference by treating a NULL char pointer as
-// "(null)".
-class GTEST_API_ Message {
- private:
-  // The type of basic IO manipulators (endl, ends, and flush) for
-  // narrow streams.
-  typedef std::ostream& (*BasicNarrowIoManip)(std::ostream&);
-
- public:
-  // Constructs an empty Message.
-  Message();
-
-  // Copy constructor.
-  Message(const Message& msg) : ss_(new ::std::stringstream) {  // NOLINT
-    *ss_ << msg.GetString();
-  }
-
-  // Constructs a Message from a C-string.
-  explicit Message(const char* str) : ss_(new ::std::stringstream) {
-    *ss_ << str;
-  }
-
-#if GTEST_OS_SYMBIAN
-  // Streams a value (either a pointer or not) to this object.
-  template <typename T>
-  inline Message& operator <<(const T& value) {
-    StreamHelper(typename internal::is_pointer<T>::type(), value);
-    return *this;
-  }
-#else
-  // Streams a non-pointer value to this object.
-  template <typename T>
-  inline Message& operator <<(const T& val) {
-    // Some libraries overload << for STL containers.  These
-    // overloads are defined in the global namespace instead of ::std.
-    //
-    // C++'s symbol lookup rule (i.e. Koenig lookup) says that these
-    // overloads are visible in either the std namespace or the global
-    // namespace, but not other namespaces, including the testing
-    // namespace which Google Test's Message class is in.
-    //
-    // To allow STL containers (and other types that has a << operator
-    // defined in the global namespace) to be used in Google Test
-    // assertions, testing::Message must access the custom << operator
-    // from the global namespace.  With this using declaration,
-    // overloads of << defined in the global namespace and those
-    // visible via Koenig lookup are both exposed in this function.
-    using ::operator <<;
-    *ss_ << val;
-    return *this;
-  }
-
-  // Streams a pointer value to this object.
-  //
-  // This function is an overload of the previous one.  When you
-  // stream a pointer to a Message, this definition will be used as it
-  // is more specialized.  (The C++ Standard, section
-  // [temp.func.order].)  If you stream a non-pointer, then the
-  // previous definition will be used.
-  //
-  // The reason for this overload is that streaming a NULL pointer to
-  // ostream is undefined behavior.  Depending on the compiler, you
-  // may get "0", "(nil)", "(null)", or an access violation.  To
-  // ensure consistent result across compilers, we always treat NULL
-  // as "(null)".
-  template <typename T>
-  inline Message& operator <<(T* const& pointer) {  // NOLINT
-    if (pointer == NULL) {
-      *ss_ << "(null)";
-    } else {
-      *ss_ << pointer;
-    }
-    return *this;
-  }
-#endif  // GTEST_OS_SYMBIAN
-
-  // Since the basic IO manipulators are overloaded for both narrow
-  // and wide streams, we have to provide this specialized definition
-  // of operator <<, even though its body is the same as the
-  // templatized version above.  Without this definition, streaming
-  // endl or other basic IO manipulators to Message will confuse the
-  // compiler.
-  Message& operator <<(BasicNarrowIoManip val) {
-    *ss_ << val;
-    return *this;
-  }
-
-  // Instead of 1/0, we want to see true/false for bool values.
-  Message& operator <<(bool b) {
-    return *this << (b ? "true" : "false");
-  }
-
-  // These two overloads allow streaming a wide C string to a Message
-  // using the UTF-8 encoding.
-  Message& operator <<(const wchar_t* wide_c_str);
-  Message& operator <<(wchar_t* wide_c_str);
-
-#if GTEST_HAS_STD_WSTRING
-  // Converts the given wide string to a narrow string using the UTF-8
-  // encoding, and streams the result to this Message object.
-  Message& operator <<(const ::std::wstring& wstr);
-#endif  // GTEST_HAS_STD_WSTRING
-
-#if GTEST_HAS_GLOBAL_WSTRING
-  // Converts the given wide string to a narrow string using the UTF-8
-  // encoding, and streams the result to this Message object.
-  Message& operator <<(const ::wstring& wstr);
-#endif  // GTEST_HAS_GLOBAL_WSTRING
-
-  // Gets the text streamed to this object so far as an std::string.
-  // Each '\0' character in the buffer is replaced with "\\0".
-  //
-  // INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-  std::string GetString() const;
-
- private:
-
-#if GTEST_OS_SYMBIAN
-  // These are needed as the Nokia Symbian Compiler cannot decide between
-  // const T& and const T* in a function template. The Nokia compiler _can_
-  // decide between class template specializations for T and T*, so a
-  // tr1::type_traits-like is_pointer works, and we can overload on that.
-  template <typename T>
-  inline void StreamHelper(internal::true_type /*is_pointer*/, T* pointer) {
-    if (pointer == NULL) {
-      *ss_ << "(null)";
-    } else {
-      *ss_ << pointer;
-    }
-  }
-  template <typename T>
-  inline void StreamHelper(internal::false_type /*is_pointer*/,
-                           const T& value) {
-    // See the comments in Message& operator <<(const T&) above for why
-    // we need this using statement.
-    using ::operator <<;
-    *ss_ << value;
-  }
-#endif  // GTEST_OS_SYMBIAN
-
-  // We'll hold the text streamed to this object here.
-  const internal::scoped_ptr< ::std::stringstream> ss_;
-
-  // We declare (but don't implement) this to prevent the compiler
-  // from implementing the assignment operator.
-  void operator=(const Message&);
-};
-
-// Streams a Message to an ostream.
-inline std::ostream& operator <<(std::ostream& os, const Message& sb) {
-  return os << sb.GetString();
-}
-
-namespace internal {
-
-// Converts a streamable value to an std::string.  A NULL pointer is
-// converted to "(null)".  When the input value is a ::string,
-// ::std::string, ::wstring, or ::std::wstring object, each NUL
-// character in it is replaced with "\\0".
-template <typename T>
-std::string StreamableToString(const T& streamable) {
-  return (Message() << streamable).GetString();
-}
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_MESSAGE_H_
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Authors: wan@google.com (Zhanyong Wan), eefacm@gmail.com (Sean Mcafee)
-//
-// The Google C++ Testing Framework (Google Test)
-//
-// This header file declares the String class and functions used internally by
-// Google Test.  They are subject to change without notice. They should not used
-// by code external to Google Test.
-//
-// This header file is #included by <gtest/internal/gtest-internal.h>.
-// It should not be #included by other files.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_STRING_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_STRING_H_
-
-#ifdef __BORLANDC__
-// string.h is not guaranteed to provide strcpy on C++ Builder.
-# include <mem.h>
-#endif
-
-#include <string.h>
-#include <string>
-
-
-namespace testing {
-namespace internal {
-
-// String - an abstract class holding static string utilities.
-class GTEST_API_ String {
- public:
-  // Static utility methods
-
-  // Clones a 0-terminated C string, allocating memory using new.  The
-  // caller is responsible for deleting the return value using
-  // delete[].  Returns the cloned string, or NULL if the input is
-  // NULL.
-  //
-  // This is different from strdup() in string.h, which allocates
-  // memory using malloc().
-  static const char* CloneCString(const char* c_str);
-
-#if GTEST_OS_WINDOWS_MOBILE
-  // Windows CE does not have the 'ANSI' versions of Win32 APIs. To be
-  // able to pass strings to Win32 APIs on CE we need to convert them
-  // to 'Unicode', UTF-16.
-
-  // Creates a UTF-16 wide string from the given ANSI string, allocating
-  // memory using new. The caller is responsible for deleting the return
-  // value using delete[]. Returns the wide string, or NULL if the
-  // input is NULL.
-  //
-  // The wide string is created using the ANSI codepage (CP_ACP) to
-  // match the behaviour of the ANSI versions of Win32 calls and the
-  // C runtime.
-  static LPCWSTR AnsiToUtf16(const char* c_str);
-
-  // Creates an ANSI string from the given wide string, allocating
-  // memory using new. The caller is responsible for deleting the return
-  // value using delete[]. Returns the ANSI string, or NULL if the
-  // input is NULL.
-  //
-  // The returned string is created using the ANSI codepage (CP_ACP) to
-  // match the behaviour of the ANSI versions of Win32 calls and the
-  // C runtime.
-  static const char* Utf16ToAnsi(LPCWSTR utf16_str);
-#endif
-
-  // Compares two C strings.  Returns true iff they have the same content.
-  //
-  // Unlike strcmp(), this function can handle NULL argument(s).  A
-  // NULL C string is considered different to any non-NULL C string,
-  // including the empty string.
-  static bool CStringEquals(const char* lhs, const char* rhs);
-
-  // Converts a wide C string to a String using the UTF-8 encoding.
-  // NULL will be converted to "(null)".  If an error occurred during
-  // the conversion, "(failed to convert from wide string)" is
-  // returned.
-  static std::string ShowWideCString(const wchar_t* wide_c_str);
-
-  // Compares two wide C strings.  Returns true iff they have the same
-  // content.
-  //
-  // Unlike wcscmp(), this function can handle NULL argument(s).  A
-  // NULL C string is considered different to any non-NULL C string,
-  // including the empty string.
-  static bool WideCStringEquals(const wchar_t* lhs, const wchar_t* rhs);
-
-  // Compares two C strings, ignoring case.  Returns true iff they
-  // have the same content.
-  //
-  // Unlike strcasecmp(), this function can handle NULL argument(s).
-  // A NULL C string is considered different to any non-NULL C string,
-  // including the empty string.
-  static bool CaseInsensitiveCStringEquals(const char* lhs,
-                                           const char* rhs);
-
-  // Compares two wide C strings, ignoring case.  Returns true iff they
-  // have the same content.
-  //
-  // Unlike wcscasecmp(), this function can handle NULL argument(s).
-  // A NULL C string is considered different to any non-NULL wide C string,
-  // including the empty string.
-  // NB: The implementations on different platforms slightly differ.
-  // On windows, this method uses _wcsicmp which compares according to LC_CTYPE
-  // environment variable. On GNU platform this method uses wcscasecmp
-  // which compares according to LC_CTYPE category of the current locale.
-  // On MacOS X, it uses towlower, which also uses LC_CTYPE category of the
-  // current locale.
-  static bool CaseInsensitiveWideCStringEquals(const wchar_t* lhs,
-                                               const wchar_t* rhs);
-
-  // Returns true iff the given string ends with the given suffix, ignoring
-  // case. Any string is considered to end with an empty suffix.
-  static bool EndsWithCaseInsensitive(
-      const std::string& str, const std::string& suffix);
-
-  // Formats an int value as "%02d".
-  static std::string FormatIntWidth2(int value);  // "%02d" for width == 2
-
-  // Formats an int value as "%X".
-  static std::string FormatHexInt(int value);
-
-  // Formats a byte as "%02X".
-  static std::string FormatByte(unsigned char value);
-
- private:
-  String();  // Not meant to be instantiated.
-};  // class String
-
-// Gets the content of the stringstream's buffer as an std::string.  Each '\0'
-// character in the buffer is replaced with "\\0".
-GTEST_API_ std::string StringStreamToString(::std::stringstream* stream);
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_STRING_H_
-// Copyright 2008, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: keith.ray@gmail.com (Keith Ray)
-//
-// Google Test filepath utilities
-//
-// This header file declares classes and functions used internally by
-// Google Test.  They are subject to change without notice.
-//
-// This file is #included in <gtest/internal/gtest-internal.h>.
-// Do not include this header file separately!
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_FILEPATH_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_FILEPATH_H_
-
-
-namespace testing {
-namespace internal {
-
-// FilePath - a class for file and directory pathname manipulation which
-// handles platform-specific conventions (like the pathname separator).
-// Used for helper functions for naming files in a directory for xml output.
-// Except for Set methods, all methods are const or static, which provides an
-// "immutable value object" -- useful for peace of mind.
-// A FilePath with a value ending in a path separator ("like/this/") represents
-// a directory, otherwise it is assumed to represent a file. In either case,
-// it may or may not represent an actual file or directory in the file system.
-// Names are NOT checked for syntax correctness -- no checking for illegal
-// characters, malformed paths, etc.
-
-class GTEST_API_ FilePath {
- public:
-  FilePath() : pathname_("") { }
-  FilePath(const FilePath& rhs) : pathname_(rhs.pathname_) { }
-
-  explicit FilePath(const std::string& pathname) : pathname_(pathname) {
-    Normalize();
-  }
-
-  FilePath& operator=(const FilePath& rhs) {
-    Set(rhs);
-    return *this;
-  }
-
-  void Set(const FilePath& rhs) {
-    pathname_ = rhs.pathname_;
-  }
-
-  const std::string& string() const { return pathname_; }
-  const char* c_str() const { return pathname_.c_str(); }
-
-  // Returns the current working directory, or "" if unsuccessful.
-  static FilePath GetCurrentDir();
-
-  // Given directory = "dir", base_name = "test", number = 0,
-  // extension = "xml", returns "dir/test.xml". If number is greater
-  // than zero (e.g., 12), returns "dir/test_12.xml".
-  // On Windows platform, uses \ as the separator rather than /.
-  static FilePath MakeFileName(const FilePath& directory,
-                               const FilePath& base_name,
-                               int number,
-                               const char* extension);
-
-  // Given directory = "dir", relative_path = "test.xml",
-  // returns "dir/test.xml".
-  // On Windows, uses \ as the separator rather than /.
-  static FilePath ConcatPaths(const FilePath& directory,
-                              const FilePath& relative_path);
-
-  // Returns a pathname for a file that does not currently exist. The pathname
-  // will be directory/base_name.extension or
-  // directory/base_name_<number>.extension if directory/base_name.extension
-  // already exists. The number will be incremented until a pathname is found
-  // that does not already exist.
-  // Examples: 'dir/foo_test.xml' or 'dir/foo_test_1.xml'.
-  // There could be a race condition if two or more processes are calling this
-  // function at the same time -- they could both pick the same filename.
-  static FilePath GenerateUniqueFileName(const FilePath& directory,
-                                         const FilePath& base_name,
-                                         const char* extension);
-
-  // Returns true iff the path is "".
-  bool IsEmpty() const { return pathname_.empty(); }
-
-  // If input name has a trailing separator character, removes it and returns
-  // the name, otherwise return the name string unmodified.
-  // On Windows platform, uses \ as the separator, other platforms use /.
-  FilePath RemoveTrailingPathSeparator() const;
-
-  // Returns a copy of the FilePath with the directory part removed.
-  // Example: FilePath("path/to/file").RemoveDirectoryName() returns
-  // FilePath("file"). If there is no directory part ("just_a_file"), it returns
-  // the FilePath unmodified. If there is no file part ("just_a_dir/") it
-  // returns an empty FilePath ("").
-  // On Windows platform, '\' is the path separator, otherwise it is '/'.
-  FilePath RemoveDirectoryName() const;
-
-  // RemoveFileName returns the directory path with the filename removed.
-  // Example: FilePath("path/to/file").RemoveFileName() returns "path/to/".
-  // If the FilePath is "a_file" or "/a_file", RemoveFileName returns
-  // FilePath("./") or, on Windows, FilePath(".\\"). If the filepath does
-  // not have a file, like "just/a/dir/", it returns the FilePath unmodified.
-  // On Windows platform, '\' is the path separator, otherwise it is '/'.
-  FilePath RemoveFileName() const;
-
-  // Returns a copy of the FilePath with the case-insensitive extension removed.
-  // Example: FilePath("dir/file.exe").RemoveExtension("EXE") returns
-  // FilePath("dir/file"). If a case-insensitive extension is not
-  // found, returns a copy of the original FilePath.
-  FilePath RemoveExtension(const char* extension) const;
-
-  // Creates directories so that path exists. Returns true if successful or if
-  // the directories already exist; returns false if unable to create
-  // directories for any reason. Will also return false if the FilePath does
-  // not represent a directory (that is, it doesn't end with a path separator).
-  bool CreateDirectoriesRecursively() const;
-
-  // Create the directory so that path exists. Returns true if successful or
-  // if the directory already exists; returns false if unable to create the
-  // directory for any reason, including if the parent directory does not
-  // exist. Not named "CreateDirectory" because that's a macro on Windows.
-  bool CreateFolder() const;
-
-  // Returns true if FilePath describes something in the file-system,
-  // either a file, directory, or whatever, and that something exists.
-  bool FileOrDirectoryExists() const;
-
-  // Returns true if pathname describes a directory in the file-system
-  // that exists.
-  bool DirectoryExists() const;
-
-  // Returns true if FilePath ends with a path separator, which indicates that
-  // it is intended to represent a directory. Returns false otherwise.
-  // This does NOT check that a directory (or file) actually exists.
-  bool IsDirectory() const;
-
-  // Returns true if pathname describes a root directory. (Windows has one
-  // root directory per disk drive.)
-  bool IsRootDirectory() const;
-
-  // Returns true if pathname describes an absolute path.
-  bool IsAbsolutePath() const;
-
- private:
-  // Replaces multiple consecutive separators with a single separator.
-  // For example, "bar///foo" becomes "bar/foo". Does not eliminate other
-  // redundancies that might be in a pathname involving "." or "..".
-  //
-  // A pathname with multiple consecutive separators may occur either through
-  // user error or as a result of some scripts or APIs that generate a pathname
-  // with a trailing separator. On other platforms the same API or script
-  // may NOT generate a pathname with a trailing "/". Then elsewhere that
-  // pathname may have another "/" and pathname components added to it,
-  // without checking for the separator already being there.
-  // The script language and operating system may allow paths like "foo//bar"
-  // but some of the functions in FilePath will not handle that correctly. In
-  // particular, RemoveTrailingPathSeparator() only removes one separator, and
-  // it is called in CreateDirectoriesRecursively() assuming that it will change
-  // a pathname from directory syntax (trailing separator) to filename syntax.
-  //
-  // On Windows this method also replaces the alternate path separator '/' with
-  // the primary path separator '\\', so that for example "bar\\/\\foo" becomes
-  // "bar\\foo".
-
-  void Normalize();
-
-  // Returns a pointer to the last occurence of a valid path separator in
-  // the FilePath. On Windows, for example, both '/' and '\' are valid path
-  // separators. Returns NULL if no path separator was found.
-  const char* FindLastPathSeparator() const;
-
-  std::string pathname_;
-};  // class FilePath
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_FILEPATH_H_
-// This file was GENERATED by command:
-//     pump.py gtest-type-util.h.pump
-// DO NOT EDIT BY HAND!!!
-
-// Copyright 2008 Google Inc.
-// All Rights Reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-
-// Type utilities needed for implementing typed and type-parameterized
-// tests.  This file is generated by a SCRIPT.  DO NOT EDIT BY HAND!
-//
-// Currently we support at most 50 types in a list, and at most 50
-// type-parameterized tests in one type-parameterized test case.
-// Please contact googletestframework@googlegroups.com if you need
-// more.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_TYPE_UTIL_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_TYPE_UTIL_H_
-
-
-// #ifdef __GNUC__ is too general here.  It is possible to use gcc without using
-// libstdc++ (which is where cxxabi.h comes from).
-# if GTEST_HAS_CXXABI_H_
-#  include <cxxabi.h>
-# elif defined(__HP_aCC)
-#  include <acxx_demangle.h>
-# endif  // GTEST_HASH_CXXABI_H_
-
-namespace testing {
-namespace internal {
-
-// GetTypeName<T>() returns a human-readable name of type T.
-// NB: This function is also used in Google Mock, so don't move it inside of
-// the typed-test-only section below.
-template <typename T>
-std::string GetTypeName() {
-# if GTEST_HAS_RTTI
-
-  const char* const name = typeid(T).name();
-#  if GTEST_HAS_CXXABI_H_ || defined(__HP_aCC)
-  int status = 0;
-  // gcc's implementation of typeid(T).name() mangles the type name,
-  // so we have to demangle it.
-#   if GTEST_HAS_CXXABI_H_
-  using abi::__cxa_demangle;
-#   endif  // GTEST_HAS_CXXABI_H_
-  char* const readable_name = __cxa_demangle(name, 0, 0, &status);
-  const std::string name_str(status == 0 ? readable_name : name);
-  free(readable_name);
-  return name_str;
-#  else
-  return name;
-#  endif  // GTEST_HAS_CXXABI_H_ || __HP_aCC
-
-# else
-
-  return "<type>";
-
-# endif  // GTEST_HAS_RTTI
-}
-
-#if GTEST_HAS_TYPED_TEST || GTEST_HAS_TYPED_TEST_P
-
-// AssertyTypeEq<T1, T2>::type is defined iff T1 and T2 are the same
-// type.  This can be used as a compile-time assertion to ensure that
-// two types are equal.
-
-template <typename T1, typename T2>
-struct AssertTypeEq;
-
-template <typename T>
-struct AssertTypeEq<T, T> {
-  typedef bool type;
-};
-
-// A unique type used as the default value for the arguments of class
-// template Types.  This allows us to simulate variadic templates
-// (e.g. Types<int>, Type<int, double>, and etc), which C++ doesn't
-// support directly.
-struct None {};
-
-// The following family of struct and struct templates are used to
-// represent type lists.  In particular, TypesN<T1, T2, ..., TN>
-// represents a type list with N types (T1, T2, ..., and TN) in it.
-// Except for Types0, every struct in the family has two member types:
-// Head for the first type in the list, and Tail for the rest of the
-// list.
-
-// The empty type list.
-struct Types0 {};
-
-// Type lists of length 1, 2, 3, and so on.
-
-template <typename T1>
-struct Types1 {
-  typedef T1 Head;
-  typedef Types0 Tail;
-};
-template <typename T1, typename T2>
-struct Types2 {
-  typedef T1 Head;
-  typedef Types1<T2> Tail;
-};
-
-template <typename T1, typename T2, typename T3>
-struct Types3 {
-  typedef T1 Head;
-  typedef Types2<T2, T3> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4>
-struct Types4 {
-  typedef T1 Head;
-  typedef Types3<T2, T3, T4> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5>
-struct Types5 {
-  typedef T1 Head;
-  typedef Types4<T2, T3, T4, T5> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6>
-struct Types6 {
-  typedef T1 Head;
-  typedef Types5<T2, T3, T4, T5, T6> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7>
-struct Types7 {
-  typedef T1 Head;
-  typedef Types6<T2, T3, T4, T5, T6, T7> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8>
-struct Types8 {
-  typedef T1 Head;
-  typedef Types7<T2, T3, T4, T5, T6, T7, T8> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9>
-struct Types9 {
-  typedef T1 Head;
-  typedef Types8<T2, T3, T4, T5, T6, T7, T8, T9> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10>
-struct Types10 {
-  typedef T1 Head;
-  typedef Types9<T2, T3, T4, T5, T6, T7, T8, T9, T10> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11>
-struct Types11 {
-  typedef T1 Head;
-  typedef Types10<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12>
-struct Types12 {
-  typedef T1 Head;
-  typedef Types11<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13>
-struct Types13 {
-  typedef T1 Head;
-  typedef Types12<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14>
-struct Types14 {
-  typedef T1 Head;
-  typedef Types13<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15>
-struct Types15 {
-  typedef T1 Head;
-  typedef Types14<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16>
-struct Types16 {
-  typedef T1 Head;
-  typedef Types15<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17>
-struct Types17 {
-  typedef T1 Head;
-  typedef Types16<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18>
-struct Types18 {
-  typedef T1 Head;
-  typedef Types17<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19>
-struct Types19 {
-  typedef T1 Head;
-  typedef Types18<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20>
-struct Types20 {
-  typedef T1 Head;
-  typedef Types19<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21>
-struct Types21 {
-  typedef T1 Head;
-  typedef Types20<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22>
-struct Types22 {
-  typedef T1 Head;
-  typedef Types21<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23>
-struct Types23 {
-  typedef T1 Head;
-  typedef Types22<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24>
-struct Types24 {
-  typedef T1 Head;
-  typedef Types23<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25>
-struct Types25 {
-  typedef T1 Head;
-  typedef Types24<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26>
-struct Types26 {
-  typedef T1 Head;
-  typedef Types25<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27>
-struct Types27 {
-  typedef T1 Head;
-  typedef Types26<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28>
-struct Types28 {
-  typedef T1 Head;
-  typedef Types27<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29>
-struct Types29 {
-  typedef T1 Head;
-  typedef Types28<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30>
-struct Types30 {
-  typedef T1 Head;
-  typedef Types29<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31>
-struct Types31 {
-  typedef T1 Head;
-  typedef Types30<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32>
-struct Types32 {
-  typedef T1 Head;
-  typedef Types31<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33>
-struct Types33 {
-  typedef T1 Head;
-  typedef Types32<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34>
-struct Types34 {
-  typedef T1 Head;
-  typedef Types33<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35>
-struct Types35 {
-  typedef T1 Head;
-  typedef Types34<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36>
-struct Types36 {
-  typedef T1 Head;
-  typedef Types35<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37>
-struct Types37 {
-  typedef T1 Head;
-  typedef Types36<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38>
-struct Types38 {
-  typedef T1 Head;
-  typedef Types37<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39>
-struct Types39 {
-  typedef T1 Head;
-  typedef Types38<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40>
-struct Types40 {
-  typedef T1 Head;
-  typedef Types39<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41>
-struct Types41 {
-  typedef T1 Head;
-  typedef Types40<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42>
-struct Types42 {
-  typedef T1 Head;
-  typedef Types41<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43>
-struct Types43 {
-  typedef T1 Head;
-  typedef Types42<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44>
-struct Types44 {
-  typedef T1 Head;
-  typedef Types43<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-      T44> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45>
-struct Types45 {
-  typedef T1 Head;
-  typedef Types44<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-      T44, T45> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46>
-struct Types46 {
-  typedef T1 Head;
-  typedef Types45<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-      T44, T45, T46> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47>
-struct Types47 {
-  typedef T1 Head;
-  typedef Types46<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-      T44, T45, T46, T47> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48>
-struct Types48 {
-  typedef T1 Head;
-  typedef Types47<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-      T44, T45, T46, T47, T48> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49>
-struct Types49 {
-  typedef T1 Head;
-  typedef Types48<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-      T44, T45, T46, T47, T48, T49> Tail;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49, typename T50>
-struct Types50 {
-  typedef T1 Head;
-  typedef Types49<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-      T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-      T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-      T44, T45, T46, T47, T48, T49, T50> Tail;
-};
-
-
-}  // namespace internal
-
-// We don't want to require the users to write TypesN<...> directly,
-// as that would require them to count the length.  Types<...> is much
-// easier to write, but generates horrible messages when there is a
-// compiler error, as gcc insists on printing out each template
-// argument, even if it has the default value (this means Types<int>
-// will appear as Types<int, None, None, ..., None> in the compiler
-// errors).
-//
-// Our solution is to combine the best part of the two approaches: a
-// user would write Types<T1, ..., TN>, and Google Test will translate
-// that to TypesN<T1, ..., TN> internally to make error messages
-// readable.  The translation is done by the 'type' member of the
-// Types template.
-template <typename T1 = internal::None, typename T2 = internal::None,
-    typename T3 = internal::None, typename T4 = internal::None,
-    typename T5 = internal::None, typename T6 = internal::None,
-    typename T7 = internal::None, typename T8 = internal::None,
-    typename T9 = internal::None, typename T10 = internal::None,
-    typename T11 = internal::None, typename T12 = internal::None,
-    typename T13 = internal::None, typename T14 = internal::None,
-    typename T15 = internal::None, typename T16 = internal::None,
-    typename T17 = internal::None, typename T18 = internal::None,
-    typename T19 = internal::None, typename T20 = internal::None,
-    typename T21 = internal::None, typename T22 = internal::None,
-    typename T23 = internal::None, typename T24 = internal::None,
-    typename T25 = internal::None, typename T26 = internal::None,
-    typename T27 = internal::None, typename T28 = internal::None,
-    typename T29 = internal::None, typename T30 = internal::None,
-    typename T31 = internal::None, typename T32 = internal::None,
-    typename T33 = internal::None, typename T34 = internal::None,
-    typename T35 = internal::None, typename T36 = internal::None,
-    typename T37 = internal::None, typename T38 = internal::None,
-    typename T39 = internal::None, typename T40 = internal::None,
-    typename T41 = internal::None, typename T42 = internal::None,
-    typename T43 = internal::None, typename T44 = internal::None,
-    typename T45 = internal::None, typename T46 = internal::None,
-    typename T47 = internal::None, typename T48 = internal::None,
-    typename T49 = internal::None, typename T50 = internal::None>
-struct Types {
-  typedef internal::Types50<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44, T45, T46, T47, T48, T49, T50> type;
-};
-
-template <>
-struct Types<internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types0 type;
-};
-template <typename T1>
-struct Types<T1, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types1<T1> type;
-};
-template <typename T1, typename T2>
-struct Types<T1, T2, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types2<T1, T2> type;
-};
-template <typename T1, typename T2, typename T3>
-struct Types<T1, T2, T3, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types3<T1, T2, T3> type;
-};
-template <typename T1, typename T2, typename T3, typename T4>
-struct Types<T1, T2, T3, T4, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types4<T1, T2, T3, T4> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5>
-struct Types<T1, T2, T3, T4, T5, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types5<T1, T2, T3, T4, T5> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6>
-struct Types<T1, T2, T3, T4, T5, T6, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types6<T1, T2, T3, T4, T5, T6> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7>
-struct Types<T1, T2, T3, T4, T5, T6, T7, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types7<T1, T2, T3, T4, T5, T6, T7> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types8<T1, T2, T3, T4, T5, T6, T7, T8> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types9<T1, T2, T3, T4, T5, T6, T7, T8, T9> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types10<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types11<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types12<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types13<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types14<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types15<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types16<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types17<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types18<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types19<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types20<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types21<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types22<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types23<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types24<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types25<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types26<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types27<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types28<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types29<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types30<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types31<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types32<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types33<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types34<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types35<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types36<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types37<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types38<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types39<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types40<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types41<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, internal::None,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types42<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None, internal::None> {
-  typedef internal::Types43<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None, internal::None> {
-  typedef internal::Types44<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44, T45,
-    internal::None, internal::None, internal::None, internal::None,
-    internal::None> {
-  typedef internal::Types45<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44, T45> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44, T45,
-    T46, internal::None, internal::None, internal::None, internal::None> {
-  typedef internal::Types46<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44, T45, T46> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44, T45,
-    T46, T47, internal::None, internal::None, internal::None> {
-  typedef internal::Types47<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44, T45, T46, T47> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44, T45,
-    T46, T47, T48, internal::None, internal::None> {
-  typedef internal::Types48<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44, T45, T46, T47, T48> type;
-};
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49>
-struct Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15,
-    T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29, T30,
-    T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44, T45,
-    T46, T47, T48, T49, internal::None> {
-  typedef internal::Types49<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44, T45, T46, T47, T48, T49> type;
-};
-
-namespace internal {
-
-# define GTEST_TEMPLATE_ template <typename T> class
-
-// The template "selector" struct TemplateSel<Tmpl> is used to
-// represent Tmpl, which must be a class template with one type
-// parameter, as a type.  TemplateSel<Tmpl>::Bind<T>::type is defined
-// as the type Tmpl<T>.  This allows us to actually instantiate the
-// template "selected" by TemplateSel<Tmpl>.
-//
-// This trick is necessary for simulating typedef for class templates,
-// which C++ doesn't support directly.
-template <GTEST_TEMPLATE_ Tmpl>
-struct TemplateSel {
-  template <typename T>
-  struct Bind {
-    typedef Tmpl<T> type;
-  };
-};
-
-# define GTEST_BIND_(TmplSel, T) \
-  TmplSel::template Bind<T>::type
-
-// A unique struct template used as the default value for the
-// arguments of class template Templates.  This allows us to simulate
-// variadic templates (e.g. Templates<int>, Templates<int, double>,
-// and etc), which C++ doesn't support directly.
-template <typename T>
-struct NoneT {};
-
-// The following family of struct and struct templates are used to
-// represent template lists.  In particular, TemplatesN<T1, T2, ...,
-// TN> represents a list of N templates (T1, T2, ..., and TN).  Except
-// for Templates0, every struct in the family has two member types:
-// Head for the selector of the first template in the list, and Tail
-// for the rest of the list.
-
-// The empty template list.
-struct Templates0 {};
-
-// Template lists of length 1, 2, 3, and so on.
-
-template <GTEST_TEMPLATE_ T1>
-struct Templates1 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates0 Tail;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2>
-struct Templates2 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates1<T2> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3>
-struct Templates3 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates2<T2, T3> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4>
-struct Templates4 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates3<T2, T3, T4> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5>
-struct Templates5 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates4<T2, T3, T4, T5> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6>
-struct Templates6 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates5<T2, T3, T4, T5, T6> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7>
-struct Templates7 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates6<T2, T3, T4, T5, T6, T7> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8>
-struct Templates8 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates7<T2, T3, T4, T5, T6, T7, T8> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9>
-struct Templates9 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates8<T2, T3, T4, T5, T6, T7, T8, T9> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10>
-struct Templates10 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates9<T2, T3, T4, T5, T6, T7, T8, T9, T10> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11>
-struct Templates11 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates10<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12>
-struct Templates12 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates11<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13>
-struct Templates13 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates12<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14>
-struct Templates14 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates13<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15>
-struct Templates15 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates14<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16>
-struct Templates16 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates15<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17>
-struct Templates17 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates16<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18>
-struct Templates18 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates17<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19>
-struct Templates19 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates18<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20>
-struct Templates20 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates19<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21>
-struct Templates21 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates20<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22>
-struct Templates22 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates21<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23>
-struct Templates23 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates22<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24>
-struct Templates24 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates23<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25>
-struct Templates25 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates24<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26>
-struct Templates26 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates25<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27>
-struct Templates27 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates26<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28>
-struct Templates28 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates27<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29>
-struct Templates29 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates28<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30>
-struct Templates30 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates29<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31>
-struct Templates31 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates30<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32>
-struct Templates32 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates31<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33>
-struct Templates33 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates32<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34>
-struct Templates34 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates33<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35>
-struct Templates35 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates34<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36>
-struct Templates36 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates35<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37>
-struct Templates37 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates36<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38>
-struct Templates38 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates37<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39>
-struct Templates39 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates38<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40>
-struct Templates40 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates39<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41>
-struct Templates41 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates40<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42>
-struct Templates42 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates41<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43>
-struct Templates43 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates42<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44>
-struct Templates44 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates43<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43, T44> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45>
-struct Templates45 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates44<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43, T44, T45> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46>
-struct Templates46 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates45<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43, T44, T45, T46> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46, GTEST_TEMPLATE_ T47>
-struct Templates47 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates46<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43, T44, T45, T46, T47> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46, GTEST_TEMPLATE_ T47, GTEST_TEMPLATE_ T48>
-struct Templates48 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates47<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43, T44, T45, T46, T47, T48> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46, GTEST_TEMPLATE_ T47, GTEST_TEMPLATE_ T48,
-    GTEST_TEMPLATE_ T49>
-struct Templates49 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates48<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43, T44, T45, T46, T47, T48, T49> Tail;
-};
-
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46, GTEST_TEMPLATE_ T47, GTEST_TEMPLATE_ T48,
-    GTEST_TEMPLATE_ T49, GTEST_TEMPLATE_ T50>
-struct Templates50 {
-  typedef TemplateSel<T1> Head;
-  typedef Templates49<T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-      T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-      T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-      T43, T44, T45, T46, T47, T48, T49, T50> Tail;
-};
-
-
-// We don't want to require the users to write TemplatesN<...> directly,
-// as that would require them to count the length.  Templates<...> is much
-// easier to write, but generates horrible messages when there is a
-// compiler error, as gcc insists on printing out each template
-// argument, even if it has the default value (this means Templates<list>
-// will appear as Templates<list, NoneT, NoneT, ..., NoneT> in the compiler
-// errors).
-//
-// Our solution is to combine the best part of the two approaches: a
-// user would write Templates<T1, ..., TN>, and Google Test will translate
-// that to TemplatesN<T1, ..., TN> internally to make error messages
-// readable.  The translation is done by the 'type' member of the
-// Templates template.
-template <GTEST_TEMPLATE_ T1 = NoneT, GTEST_TEMPLATE_ T2 = NoneT,
-    GTEST_TEMPLATE_ T3 = NoneT, GTEST_TEMPLATE_ T4 = NoneT,
-    GTEST_TEMPLATE_ T5 = NoneT, GTEST_TEMPLATE_ T6 = NoneT,
-    GTEST_TEMPLATE_ T7 = NoneT, GTEST_TEMPLATE_ T8 = NoneT,
-    GTEST_TEMPLATE_ T9 = NoneT, GTEST_TEMPLATE_ T10 = NoneT,
-    GTEST_TEMPLATE_ T11 = NoneT, GTEST_TEMPLATE_ T12 = NoneT,
-    GTEST_TEMPLATE_ T13 = NoneT, GTEST_TEMPLATE_ T14 = NoneT,
-    GTEST_TEMPLATE_ T15 = NoneT, GTEST_TEMPLATE_ T16 = NoneT,
-    GTEST_TEMPLATE_ T17 = NoneT, GTEST_TEMPLATE_ T18 = NoneT,
-    GTEST_TEMPLATE_ T19 = NoneT, GTEST_TEMPLATE_ T20 = NoneT,
-    GTEST_TEMPLATE_ T21 = NoneT, GTEST_TEMPLATE_ T22 = NoneT,
-    GTEST_TEMPLATE_ T23 = NoneT, GTEST_TEMPLATE_ T24 = NoneT,
-    GTEST_TEMPLATE_ T25 = NoneT, GTEST_TEMPLATE_ T26 = NoneT,
-    GTEST_TEMPLATE_ T27 = NoneT, GTEST_TEMPLATE_ T28 = NoneT,
-    GTEST_TEMPLATE_ T29 = NoneT, GTEST_TEMPLATE_ T30 = NoneT,
-    GTEST_TEMPLATE_ T31 = NoneT, GTEST_TEMPLATE_ T32 = NoneT,
-    GTEST_TEMPLATE_ T33 = NoneT, GTEST_TEMPLATE_ T34 = NoneT,
-    GTEST_TEMPLATE_ T35 = NoneT, GTEST_TEMPLATE_ T36 = NoneT,
-    GTEST_TEMPLATE_ T37 = NoneT, GTEST_TEMPLATE_ T38 = NoneT,
-    GTEST_TEMPLATE_ T39 = NoneT, GTEST_TEMPLATE_ T40 = NoneT,
-    GTEST_TEMPLATE_ T41 = NoneT, GTEST_TEMPLATE_ T42 = NoneT,
-    GTEST_TEMPLATE_ T43 = NoneT, GTEST_TEMPLATE_ T44 = NoneT,
-    GTEST_TEMPLATE_ T45 = NoneT, GTEST_TEMPLATE_ T46 = NoneT,
-    GTEST_TEMPLATE_ T47 = NoneT, GTEST_TEMPLATE_ T48 = NoneT,
-    GTEST_TEMPLATE_ T49 = NoneT, GTEST_TEMPLATE_ T50 = NoneT>
-struct Templates {
-  typedef Templates50<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43, T44, T45, T46, T47, T48, T49, T50> type;
-};
-
-template <>
-struct Templates<NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT> {
-  typedef Templates0 type;
-};
-template <GTEST_TEMPLATE_ T1>
-struct Templates<T1, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT> {
-  typedef Templates1<T1> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2>
-struct Templates<T1, T2, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT> {
-  typedef Templates2<T1, T2> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3>
-struct Templates<T1, T2, T3, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates3<T1, T2, T3> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4>
-struct Templates<T1, T2, T3, T4, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates4<T1, T2, T3, T4> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5>
-struct Templates<T1, T2, T3, T4, T5, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates5<T1, T2, T3, T4, T5> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6>
-struct Templates<T1, T2, T3, T4, T5, T6, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates6<T1, T2, T3, T4, T5, T6> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates7<T1, T2, T3, T4, T5, T6, T7> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates8<T1, T2, T3, T4, T5, T6, T7, T8> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates9<T1, T2, T3, T4, T5, T6, T7, T8, T9> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates10<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates11<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates12<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates13<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates14<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates15<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates16<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates17<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates18<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates19<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates20<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates21<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT> {
-  typedef Templates22<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT> {
-  typedef Templates23<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT> {
-  typedef Templates24<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT> {
-  typedef Templates25<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT> {
-  typedef Templates26<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT> {
-  typedef Templates27<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT> {
-  typedef Templates28<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT> {
-  typedef Templates29<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates30<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates31<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates32<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates33<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates34<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates35<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates36<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, NoneT, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates37<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, NoneT, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates38<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates39<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, NoneT, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates40<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, NoneT, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates41<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, NoneT,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates42<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates43<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44,
-    NoneT, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates44<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43, T44> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44,
-    T45, NoneT, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates45<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43, T44, T45> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44,
-    T45, T46, NoneT, NoneT, NoneT, NoneT> {
-  typedef Templates46<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43, T44, T45, T46> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46, GTEST_TEMPLATE_ T47>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44,
-    T45, T46, T47, NoneT, NoneT, NoneT> {
-  typedef Templates47<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43, T44, T45, T46, T47> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46, GTEST_TEMPLATE_ T47, GTEST_TEMPLATE_ T48>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44,
-    T45, T46, T47, T48, NoneT, NoneT> {
-  typedef Templates48<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43, T44, T45, T46, T47, T48> type;
-};
-template <GTEST_TEMPLATE_ T1, GTEST_TEMPLATE_ T2, GTEST_TEMPLATE_ T3,
-    GTEST_TEMPLATE_ T4, GTEST_TEMPLATE_ T5, GTEST_TEMPLATE_ T6,
-    GTEST_TEMPLATE_ T7, GTEST_TEMPLATE_ T8, GTEST_TEMPLATE_ T9,
-    GTEST_TEMPLATE_ T10, GTEST_TEMPLATE_ T11, GTEST_TEMPLATE_ T12,
-    GTEST_TEMPLATE_ T13, GTEST_TEMPLATE_ T14, GTEST_TEMPLATE_ T15,
-    GTEST_TEMPLATE_ T16, GTEST_TEMPLATE_ T17, GTEST_TEMPLATE_ T18,
-    GTEST_TEMPLATE_ T19, GTEST_TEMPLATE_ T20, GTEST_TEMPLATE_ T21,
-    GTEST_TEMPLATE_ T22, GTEST_TEMPLATE_ T23, GTEST_TEMPLATE_ T24,
-    GTEST_TEMPLATE_ T25, GTEST_TEMPLATE_ T26, GTEST_TEMPLATE_ T27,
-    GTEST_TEMPLATE_ T28, GTEST_TEMPLATE_ T29, GTEST_TEMPLATE_ T30,
-    GTEST_TEMPLATE_ T31, GTEST_TEMPLATE_ T32, GTEST_TEMPLATE_ T33,
-    GTEST_TEMPLATE_ T34, GTEST_TEMPLATE_ T35, GTEST_TEMPLATE_ T36,
-    GTEST_TEMPLATE_ T37, GTEST_TEMPLATE_ T38, GTEST_TEMPLATE_ T39,
-    GTEST_TEMPLATE_ T40, GTEST_TEMPLATE_ T41, GTEST_TEMPLATE_ T42,
-    GTEST_TEMPLATE_ T43, GTEST_TEMPLATE_ T44, GTEST_TEMPLATE_ T45,
-    GTEST_TEMPLATE_ T46, GTEST_TEMPLATE_ T47, GTEST_TEMPLATE_ T48,
-    GTEST_TEMPLATE_ T49>
-struct Templates<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14,
-    T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28, T29,
-    T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43, T44,
-    T45, T46, T47, T48, T49, NoneT> {
-  typedef Templates49<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-      T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-      T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-      T42, T43, T44, T45, T46, T47, T48, T49> type;
-};
-
-// The TypeList template makes it possible to use either a single type
-// or a Types<...> list in TYPED_TEST_CASE() and
-// INSTANTIATE_TYPED_TEST_CASE_P().
-
-template <typename T>
-struct TypeList {
-  typedef Types1<T> type;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49, typename T50>
-struct TypeList<Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44, T45, T46, T47, T48, T49, T50> > {
-  typedef typename Types<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-      T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-      T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-      T41, T42, T43, T44, T45, T46, T47, T48, T49, T50>::type type;
-};
-
-#endif  // GTEST_HAS_TYPED_TEST || GTEST_HAS_TYPED_TEST_P
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_TYPE_UTIL_H_
-
-// Due to C++ preprocessor weirdness, we need double indirection to
-// concatenate two tokens when one of them is __LINE__.  Writing
-//
-//   foo ## __LINE__
-//
-// will result in the token foo__LINE__, instead of foo followed by
-// the current line number.  For more details, see
-// http://www.parashift.com/c++-faq-lite/misc-technical-issues.html#faq-39.6
-#define GTEST_CONCAT_TOKEN_(foo, bar) GTEST_CONCAT_TOKEN_IMPL_(foo, bar)
-#define GTEST_CONCAT_TOKEN_IMPL_(foo, bar) foo ## bar
-
-class ProtocolMessage;
-namespace proto2 { class Message; }
-
-namespace testing {
-
-// Forward declarations.
-
-class AssertionResult;                 // Result of an assertion.
-class Message;                         // Represents a failure message.
-class Test;                            // Represents a test.
-class TestInfo;                        // Information about a test.
-class TestPartResult;                  // Result of a test part.
-class UnitTest;                        // A collection of test cases.
-
-template <typename T>
-::std::string PrintToString(const T& value);
-
-namespace internal {
-
-struct TraceInfo;                      // Information about a trace point.
-class ScopedTrace;                     // Implements scoped trace.
-class TestInfoImpl;                    // Opaque implementation of TestInfo
-class UnitTestImpl;                    // Opaque implementation of UnitTest
-
-// The text used in failure messages to indicate the start of the
-// stack trace.
-GTEST_API_ extern const char kStackTraceMarker[];
-
-// Two overloaded helpers for checking at compile time whether an
-// expression is a null pointer literal (i.e. NULL or any 0-valued
-// compile-time integral constant).  Their return values have
-// different sizes, so we can use sizeof() to test which version is
-// picked by the compiler.  These helpers have no implementations, as
-// we only need their signatures.
-//
-// Given IsNullLiteralHelper(x), the compiler will pick the first
-// version if x can be implicitly converted to Secret*, and pick the
-// second version otherwise.  Since Secret is a secret and incomplete
-// type, the only expression a user can write that has type Secret* is
-// a null pointer literal.  Therefore, we know that x is a null
-// pointer literal if and only if the first version is picked by the
-// compiler.
-char IsNullLiteralHelper(Secret* p);
-char (&IsNullLiteralHelper(...))[2];  // NOLINT
-
-// A compile-time bool constant that is true if and only if x is a
-// null pointer literal (i.e. NULL or any 0-valued compile-time
-// integral constant).
-#ifdef GTEST_ELLIPSIS_NEEDS_POD_
-// We lose support for NULL detection where the compiler doesn't like
-// passing non-POD classes through ellipsis (...).
-# define GTEST_IS_NULL_LITERAL_(x) false
-#else
-# define GTEST_IS_NULL_LITERAL_(x) \
-    (sizeof(::testing::internal::IsNullLiteralHelper(x)) == 1)
-#endif  // GTEST_ELLIPSIS_NEEDS_POD_
-
-// Appends the user-supplied message to the Google-Test-generated message.
-GTEST_API_ std::string AppendUserMessage(
-    const std::string& gtest_msg, const Message& user_msg);
-
-#if GTEST_HAS_EXCEPTIONS
-
-// This exception is thrown by (and only by) a failed Google Test
-// assertion when GTEST_FLAG(throw_on_failure) is true (if exceptions
-// are enabled).  We derive it from std::runtime_error, which is for
-// errors presumably detectable only at run time.  Since
-// std::runtime_error inherits from std::exception, many testing
-// frameworks know how to extract and print the message inside it.
-class GTEST_API_ GoogleTestFailureException : public ::std::runtime_error {
- public:
-  explicit GoogleTestFailureException(const TestPartResult& failure);
-};
-
-#endif  // GTEST_HAS_EXCEPTIONS
-
-// A helper class for creating scoped traces in user programs.
-class GTEST_API_ ScopedTrace {
- public:
-  // The c'tor pushes the given source file location and message onto
-  // a trace stack maintained by Google Test.
-  ScopedTrace(const char* file, int line, const Message& message);
-
-  // The d'tor pops the info pushed by the c'tor.
-  //
-  // Note that the d'tor is not virtual in order to be efficient.
-  // Don't inherit from ScopedTrace!
-  ~ScopedTrace();
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ScopedTrace);
-} GTEST_ATTRIBUTE_UNUSED_;  // A ScopedTrace object does its job in its
-                            // c'tor and d'tor.  Therefore it doesn't
-                            // need to be used otherwise.
-
-namespace edit_distance {
-// Returns the optimal edits to go from 'left' to 'right'.
-// All edits cost the same, with replace having lower priority than
-// add/remove.
-// Simple implementation of the Wagner–Fischer algorithm.
-// See http://en.wikipedia.org/wiki/Wagner-Fischer_algorithm
-enum EditType { kMatch, kAdd, kRemove, kReplace };
-GTEST_API_ std::vector<EditType> CalculateOptimalEdits(
-    const std::vector<size_t>& left, const std::vector<size_t>& right);
-
-// Same as above, but the input is represented as strings.
-GTEST_API_ std::vector<EditType> CalculateOptimalEdits(
-    const std::vector<std::string>& left,
-    const std::vector<std::string>& right);
-
-// Create a diff of the input strings in Unified diff format.
-GTEST_API_ std::string CreateUnifiedDiff(const std::vector<std::string>& left,
-                                         const std::vector<std::string>& right,
-                                         size_t context = 2);
-
-}  // namespace edit_distance
-
-// Calculate the diff between 'left' and 'right' and return it in unified diff
-// format.
-// If not null, stores in 'total_line_count' the total number of lines found
-// in left + right.
-GTEST_API_ std::string DiffStrings(const std::string& left,
-                                   const std::string& right,
-                                   size_t* total_line_count);
-
-// Constructs and returns the message for an equality assertion
-// (e.g. ASSERT_EQ, EXPECT_STREQ, etc) failure.
-//
-// The first four parameters are the expressions used in the assertion
-// and their values, as strings.  For example, for ASSERT_EQ(foo, bar)
-// where foo is 5 and bar is 6, we have:
-//
-//   expected_expression: "foo"
-//   actual_expression:   "bar"
-//   expected_value:      "5"
-//   actual_value:        "6"
-//
-// The ignoring_case parameter is true iff the assertion is a
-// *_STRCASEEQ*.  When it's true, the string " (ignoring case)" will
-// be inserted into the message.
-GTEST_API_ AssertionResult EqFailure(const char* expected_expression,
-                                     const char* actual_expression,
-                                     const std::string& expected_value,
-                                     const std::string& actual_value,
-                                     bool ignoring_case);
-
-// Constructs a failure message for Boolean assertions such as EXPECT_TRUE.
-GTEST_API_ std::string GetBoolAssertionFailureMessage(
-    const AssertionResult& assertion_result,
-    const char* expression_text,
-    const char* actual_predicate_value,
-    const char* expected_predicate_value);
-
-// This template class represents an IEEE floating-point number
-// (either single-precision or double-precision, depending on the
-// template parameters).
-//
-// The purpose of this class is to do more sophisticated number
-// comparison.  (Due to round-off error, etc, it's very unlikely that
-// two floating-points will be equal exactly.  Hence a naive
-// comparison by the == operation often doesn't work.)
-//
-// Format of IEEE floating-point:
-//
-//   The most-significant bit being the leftmost, an IEEE
-//   floating-point looks like
-//
-//     sign_bit exponent_bits fraction_bits
-//
-//   Here, sign_bit is a single bit that designates the sign of the
-//   number.
-//
-//   For float, there are 8 exponent bits and 23 fraction bits.
-//
-//   For double, there are 11 exponent bits and 52 fraction bits.
-//
-//   More details can be found at
-//   http://en.wikipedia.org/wiki/IEEE_floating-point_standard.
-//
-// Template parameter:
-//
-//   RawType: the raw floating-point type (either float or double)
-template <typename RawType>
-class FloatingPoint {
- public:
-  // Defines the unsigned integer type that has the same size as the
-  // floating point number.
-  typedef typename TypeWithSize<sizeof(RawType)>::UInt Bits;
-
-  // Constants.
-
-  // # of bits in a number.
-  static const size_t kBitCount = 8*sizeof(RawType);
-
-  // # of fraction bits in a number.
-  static const size_t kFractionBitCount =
-    std::numeric_limits<RawType>::digits - 1;
-
-  // # of exponent bits in a number.
-  static const size_t kExponentBitCount = kBitCount - 1 - kFractionBitCount;
-
-  // The mask for the sign bit.
-  static const Bits kSignBitMask = static_cast<Bits>(1) << (kBitCount - 1);
-
-  // The mask for the fraction bits.
-  static const Bits kFractionBitMask =
-    ~static_cast<Bits>(0) >> (kExponentBitCount + 1);
-
-  // The mask for the exponent bits.
-  static const Bits kExponentBitMask = ~(kSignBitMask | kFractionBitMask);
-
-  // How many ULP's (Units in the Last Place) we want to tolerate when
-  // comparing two numbers.  The larger the value, the more error we
-  // allow.  A 0 value means that two numbers must be exactly the same
-  // to be considered equal.
-  //
-  // The maximum error of a single floating-point operation is 0.5
-  // units in the last place.  On Intel CPU's, all floating-point
-  // calculations are done with 80-bit precision, while double has 64
-  // bits.  Therefore, 4 should be enough for ordinary use.
-  //
-  // See the following article for more details on ULP:
-  // http://randomascii.wordpress.com/2012/02/25/comparing-floating-point-numbers-2012-edition/
-  static const size_t kMaxUlps = 4;
-
-  // Constructs a FloatingPoint from a raw floating-point number.
-  //
-  // On an Intel CPU, passing a non-normalized NAN (Not a Number)
-  // around may change its bits, although the new value is guaranteed
-  // to be also a NAN.  Therefore, don't expect this constructor to
-  // preserve the bits in x when x is a NAN.
-  explicit FloatingPoint(const RawType& x) { u_.value_ = x; }
-
-  // Static methods
-
-  // Reinterprets a bit pattern as a floating-point number.
-  //
-  // This function is needed to test the AlmostEquals() method.
-  static RawType ReinterpretBits(const Bits bits) {
-    FloatingPoint fp(0);
-    fp.u_.bits_ = bits;
-    return fp.u_.value_;
-  }
-
-  // Returns the floating-point number that represent positive infinity.
-  static RawType Infinity() {
-    return ReinterpretBits(kExponentBitMask);
-  }
-
-  // Returns the maximum representable finite floating-point number.
-  static RawType Max();
-
-  // Non-static methods
-
-  // Returns the bits that represents this number.
-  const Bits &bits() const { return u_.bits_; }
-
-  // Returns the exponent bits of this number.
-  Bits exponent_bits() const { return kExponentBitMask & u_.bits_; }
-
-  // Returns the fraction bits of this number.
-  Bits fraction_bits() const { return kFractionBitMask & u_.bits_; }
-
-  // Returns the sign bit of this number.
-  Bits sign_bit() const { return kSignBitMask & u_.bits_; }
-
-  // Returns true iff this is NAN (not a number).
-  bool is_nan() const {
-    // It's a NAN if the exponent bits are all ones and the fraction
-    // bits are not entirely zeros.
-    return (exponent_bits() == kExponentBitMask) && (fraction_bits() != 0);
-  }
-
-  // Returns true iff this number is at most kMaxUlps ULP's away from
-  // rhs.  In particular, this function:
-  //
-  //   - returns false if either number is (or both are) NAN.
-  //   - treats really large numbers as almost equal to infinity.
-  //   - thinks +0.0 and -0.0 are 0 DLP's apart.
-  bool AlmostEquals(const FloatingPoint& rhs) const {
-    // The IEEE standard says that any comparison operation involving
-    // a NAN must return false.
-    if (is_nan() || rhs.is_nan()) return false;
-
-    return DistanceBetweenSignAndMagnitudeNumbers(u_.bits_, rhs.u_.bits_)
-        <= kMaxUlps;
-  }
-
- private:
-  // The data type used to store the actual floating-point number.
-  union FloatingPointUnion {
-    RawType value_;  // The raw floating-point number.
-    Bits bits_;      // The bits that represent the number.
-  };
-
-  // Converts an integer from the sign-and-magnitude representation to
-  // the biased representation.  More precisely, let N be 2 to the
-  // power of (kBitCount - 1), an integer x is represented by the
-  // unsigned number x + N.
-  //
-  // For instance,
-  //
-  //   -N + 1 (the most negative number representable using
-  //          sign-and-magnitude) is represented by 1;
-  //   0      is represented by N; and
-  //   N - 1  (the biggest number representable using
-  //          sign-and-magnitude) is represented by 2N - 1.
-  //
-  // Read http://en.wikipedia.org/wiki/Signed_number_representations
-  // for more details on signed number representations.
-  static Bits SignAndMagnitudeToBiased(const Bits &sam) {
-    if (kSignBitMask & sam) {
-      // sam represents a negative number.
-      return ~sam + 1;
-    } else {
-      // sam represents a positive number.
-      return kSignBitMask | sam;
-    }
-  }
-
-  // Given two numbers in the sign-and-magnitude representation,
-  // returns the distance between them as an unsigned number.
-  static Bits DistanceBetweenSignAndMagnitudeNumbers(const Bits &sam1,
-                                                     const Bits &sam2) {
-    const Bits biased1 = SignAndMagnitudeToBiased(sam1);
-    const Bits biased2 = SignAndMagnitudeToBiased(sam2);
-    return (biased1 >= biased2) ? (biased1 - biased2) : (biased2 - biased1);
-  }
-
-  FloatingPointUnion u_;
-};
-
-// We cannot use std::numeric_limits<T>::max() as it clashes with the max()
-// macro defined by <windows.h>.
-template <>
-inline float FloatingPoint<float>::Max() { return FLT_MAX; }
-template <>
-inline double FloatingPoint<double>::Max() { return DBL_MAX; }
-
-// Typedefs the instances of the FloatingPoint template class that we
-// care to use.
-typedef FloatingPoint<float> Float;
-typedef FloatingPoint<double> Double;
-
-// In order to catch the mistake of putting tests that use different
-// test fixture classes in the same test case, we need to assign
-// unique IDs to fixture classes and compare them.  The TypeId type is
-// used to hold such IDs.  The user should treat TypeId as an opaque
-// type: the only operation allowed on TypeId values is to compare
-// them for equality using the == operator.
-typedef const void* TypeId;
-
-template <typename T>
-class TypeIdHelper {
- public:
-  // dummy_ must not have a const type.  Otherwise an overly eager
-  // compiler (e.g. MSVC 7.1 & 8.0) may try to merge
-  // TypeIdHelper<T>::dummy_ for different Ts as an "optimization".
-  static bool dummy_;
-};
-
-template <typename T>
-bool TypeIdHelper<T>::dummy_ = false;
-
-// GetTypeId<T>() returns the ID of type T.  Different values will be
-// returned for different types.  Calling the function twice with the
-// same type argument is guaranteed to return the same ID.
-template <typename T>
-TypeId GetTypeId() {
-  // The compiler is required to allocate a different
-  // TypeIdHelper<T>::dummy_ variable for each T used to instantiate
-  // the template.  Therefore, the address of dummy_ is guaranteed to
-  // be unique.
-  return &(TypeIdHelper<T>::dummy_);
-}
-
-// Returns the type ID of ::testing::Test.  Always call this instead
-// of GetTypeId< ::testing::Test>() to get the type ID of
-// ::testing::Test, as the latter may give the wrong result due to a
-// suspected linker bug when compiling Google Test as a Mac OS X
-// framework.
-GTEST_API_ TypeId GetTestTypeId();
-
-// Defines the abstract factory interface that creates instances
-// of a Test object.
-class TestFactoryBase {
- public:
-  virtual ~TestFactoryBase() {}
-
-  // Creates a test instance to run. The instance is both created and destroyed
-  // within TestInfoImpl::Run()
-  virtual Test* CreateTest() = 0;
-
- protected:
-  TestFactoryBase() {}
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestFactoryBase);
-};
-
-// This class provides implementation of TeastFactoryBase interface.
-// It is used in TEST and TEST_F macros.
-template <class TestClass>
-class TestFactoryImpl : public TestFactoryBase {
- public:
-  virtual Test* CreateTest() { return new TestClass; }
-};
-
-#if GTEST_OS_WINDOWS
-
-// Predicate-formatters for implementing the HRESULT checking macros
-// {ASSERT|EXPECT}_HRESULT_{SUCCEEDED|FAILED}
-// We pass a long instead of HRESULT to avoid causing an
-// include dependency for the HRESULT type.
-GTEST_API_ AssertionResult IsHRESULTSuccess(const char* expr,
-                                            long hr);  // NOLINT
-GTEST_API_ AssertionResult IsHRESULTFailure(const char* expr,
-                                            long hr);  // NOLINT
-
-#endif  // GTEST_OS_WINDOWS
-
-// Types of SetUpTestCase() and TearDownTestCase() functions.
-typedef void (*SetUpTestCaseFunc)();
-typedef void (*TearDownTestCaseFunc)();
-
-struct CodeLocation {
-  CodeLocation(const string& a_file, int a_line) : file(a_file), line(a_line) {}
-
-  string file;
-  int line;
-};
-
-// Creates a new TestInfo object and registers it with Google Test;
-// returns the created object.
-//
-// Arguments:
-//
-//   test_case_name:   name of the test case
-//   name:             name of the test
-//   type_param        the name of the test's type parameter, or NULL if
-//                     this is not a typed or a type-parameterized test.
-//   value_param       text representation of the test's value parameter,
-//                     or NULL if this is not a type-parameterized test.
-//   code_location:    code location where the test is defined
-//   fixture_class_id: ID of the test fixture class
-//   set_up_tc:        pointer to the function that sets up the test case
-//   tear_down_tc:     pointer to the function that tears down the test case
-//   factory:          pointer to the factory that creates a test object.
-//                     The newly created TestInfo instance will assume
-//                     ownership of the factory object.
-GTEST_API_ TestInfo* MakeAndRegisterTestInfo(
-    const char* test_case_name,
-    const char* name,
-    const char* type_param,
-    const char* value_param,
-    CodeLocation code_location,
-    TypeId fixture_class_id,
-    SetUpTestCaseFunc set_up_tc,
-    TearDownTestCaseFunc tear_down_tc,
-    TestFactoryBase* factory);
-
-// If *pstr starts with the given prefix, modifies *pstr to be right
-// past the prefix and returns true; otherwise leaves *pstr unchanged
-// and returns false.  None of pstr, *pstr, and prefix can be NULL.
-GTEST_API_ bool SkipPrefix(const char* prefix, const char** pstr);
-
-#if GTEST_HAS_TYPED_TEST || GTEST_HAS_TYPED_TEST_P
-
-// State of the definition of a type-parameterized test case.
-class GTEST_API_ TypedTestCasePState {
- public:
-  TypedTestCasePState() : registered_(false) {}
-
-  // Adds the given test name to defined_test_names_ and return true
-  // if the test case hasn't been registered; otherwise aborts the
-  // program.
-  bool AddTestName(const char* file, int line, const char* case_name,
-                   const char* test_name) {
-    if (registered_) {
-      fprintf(stderr, "%s Test %s must be defined before "
-              "REGISTER_TYPED_TEST_CASE_P(%s, ...).\n",
-              FormatFileLocation(file, line).c_str(), test_name, case_name);
-      fflush(stderr);
-      posix::Abort();
-    }
-    registered_tests_.insert(
-        ::std::make_pair(test_name, CodeLocation(file, line)));
-    return true;
-  }
-
-  bool TestExists(const std::string& test_name) const {
-    return registered_tests_.count(test_name) > 0;
-  }
-
-  const CodeLocation& GetCodeLocation(const std::string& test_name) const {
-    RegisteredTestsMap::const_iterator it = registered_tests_.find(test_name);
-    GTEST_CHECK_(it != registered_tests_.end());
-    return it->second;
-  }
-
-  // Verifies that registered_tests match the test names in
-  // defined_test_names_; returns registered_tests if successful, or
-  // aborts the program otherwise.
-  const char* VerifyRegisteredTestNames(
-      const char* file, int line, const char* registered_tests);
-
- private:
-  typedef ::std::map<std::string, CodeLocation> RegisteredTestsMap;
-
-  bool registered_;
-  RegisteredTestsMap registered_tests_;
-};
-
-// Skips to the first non-space char after the first comma in 'str';
-// returns NULL if no comma is found in 'str'.
-inline const char* SkipComma(const char* str) {
-  const char* comma = strchr(str, ',');
-  if (comma == NULL) {
-    return NULL;
-  }
-  while (IsSpace(*(++comma))) {}
-  return comma;
-}
-
-// Returns the prefix of 'str' before the first comma in it; returns
-// the entire string if it contains no comma.
-inline std::string GetPrefixUntilComma(const char* str) {
-  const char* comma = strchr(str, ',');
-  return comma == NULL ? str : std::string(str, comma);
-}
-
-// Splits a given string on a given delimiter, populating a given
-// vector with the fields.
-void SplitString(const ::std::string& str, char delimiter,
-                 ::std::vector< ::std::string>* dest);
-
-// TypeParameterizedTest<Fixture, TestSel, Types>::Register()
-// registers a list of type-parameterized tests with Google Test.  The
-// return value is insignificant - we just need to return something
-// such that we can call this function in a namespace scope.
-//
-// Implementation note: The GTEST_TEMPLATE_ macro declares a template
-// template parameter.  It's defined in gtest-type-util.h.
-template <GTEST_TEMPLATE_ Fixture, class TestSel, typename Types>
-class TypeParameterizedTest {
- public:
-  // 'index' is the index of the test in the type list 'Types'
-  // specified in INSTANTIATE_TYPED_TEST_CASE_P(Prefix, TestCase,
-  // Types).  Valid values for 'index' are [0, N - 1] where N is the
-  // length of Types.
-  static bool Register(const char* prefix,
-                       CodeLocation code_location,
-                       const char* case_name, const char* test_names,
-                       int index) {
-    typedef typename Types::Head Type;
-    typedef Fixture<Type> FixtureClass;
-    typedef typename GTEST_BIND_(TestSel, Type) TestClass;
-
-    // First, registers the first type-parameterized test in the type
-    // list.
-    MakeAndRegisterTestInfo(
-        (std::string(prefix) + (prefix[0] == '\0' ? "" : "/") + case_name + "/"
-         + StreamableToString(index)).c_str(),
-        StripTrailingSpaces(GetPrefixUntilComma(test_names)).c_str(),
-        GetTypeName<Type>().c_str(),
-        NULL,  // No value parameter.
-        code_location,
-        GetTypeId<FixtureClass>(),
-        TestClass::SetUpTestCase,
-        TestClass::TearDownTestCase,
-        new TestFactoryImpl<TestClass>);
-
-    // Next, recurses (at compile time) with the tail of the type list.
-    return TypeParameterizedTest<Fixture, TestSel, typename Types::Tail>
-        ::Register(prefix, code_location, case_name, test_names, index + 1);
-  }
-};
-
-// The base case for the compile time recursion.
-template <GTEST_TEMPLATE_ Fixture, class TestSel>
-class TypeParameterizedTest<Fixture, TestSel, Types0> {
- public:
-  static bool Register(const char* /*prefix*/, CodeLocation,
-                       const char* /*case_name*/, const char* /*test_names*/,
-                       int /*index*/) {
-    return true;
-  }
-};
-
-// TypeParameterizedTestCase<Fixture, Tests, Types>::Register()
-// registers *all combinations* of 'Tests' and 'Types' with Google
-// Test.  The return value is insignificant - we just need to return
-// something such that we can call this function in a namespace scope.
-template <GTEST_TEMPLATE_ Fixture, typename Tests, typename Types>
-class TypeParameterizedTestCase {
- public:
-  static bool Register(const char* prefix, CodeLocation code_location,
-                       const TypedTestCasePState* state,
-                       const char* case_name, const char* test_names) {
-    std::string test_name = StripTrailingSpaces(
-        GetPrefixUntilComma(test_names));
-    if (!state->TestExists(test_name)) {
-      fprintf(stderr, "Failed to get code location for test %s.%s at %s.",
-              case_name, test_name.c_str(),
-              FormatFileLocation(code_location.file.c_str(),
-                                 code_location.line).c_str());
-      fflush(stderr);
-      posix::Abort();
-    }
-    const CodeLocation& test_location = state->GetCodeLocation(test_name);
-
-    typedef typename Tests::Head Head;
-
-    // First, register the first test in 'Test' for each type in 'Types'.
-    TypeParameterizedTest<Fixture, Head, Types>::Register(
-        prefix, test_location, case_name, test_names, 0);
-
-    // Next, recurses (at compile time) with the tail of the test list.
-    return TypeParameterizedTestCase<Fixture, typename Tests::Tail, Types>
-        ::Register(prefix, code_location, state,
-                   case_name, SkipComma(test_names));
-  }
-};
-
-// The base case for the compile time recursion.
-template <GTEST_TEMPLATE_ Fixture, typename Types>
-class TypeParameterizedTestCase<Fixture, Templates0, Types> {
- public:
-  static bool Register(const char* /*prefix*/, CodeLocation,
-                       const TypedTestCasePState* /*state*/,
-                       const char* /*case_name*/, const char* /*test_names*/) {
-    return true;
-  }
-};
-
-#endif  // GTEST_HAS_TYPED_TEST || GTEST_HAS_TYPED_TEST_P
-
-// Returns the current OS stack trace as an std::string.
-//
-// The maximum number of stack frames to be included is specified by
-// the gtest_stack_trace_depth flag.  The skip_count parameter
-// specifies the number of top frames to be skipped, which doesn't
-// count against the number of frames to be included.
-//
-// For example, if Foo() calls Bar(), which in turn calls
-// GetCurrentOsStackTraceExceptTop(..., 1), Foo() will be included in
-// the trace but Bar() and GetCurrentOsStackTraceExceptTop() won't.
-GTEST_API_ std::string GetCurrentOsStackTraceExceptTop(
-    UnitTest* unit_test, int skip_count);
-
-// Helpers for suppressing warnings on unreachable code or constant
-// condition.
-
-// Always returns true.
-GTEST_API_ bool AlwaysTrue();
-
-// Always returns false.
-inline bool AlwaysFalse() { return !AlwaysTrue(); }
-
-// Helper for suppressing false warning from Clang on a const char*
-// variable declared in a conditional expression always being NULL in
-// the else branch.
-struct GTEST_API_ ConstCharPtr {
-  ConstCharPtr(const char* str) : value(str) {}
-  operator bool() const { return true; }
-  const char* value;
-};
-
-// A simple Linear Congruential Generator for generating random
-// numbers with a uniform distribution.  Unlike rand() and srand(), it
-// doesn't use global state (and therefore can't interfere with user
-// code).  Unlike rand_r(), it's portable.  An LCG isn't very random,
-// but it's good enough for our purposes.
-class GTEST_API_ Random {
- public:
-  static const UInt32 kMaxRange = 1u << 31;
-
-  explicit Random(UInt32 seed) : state_(seed) {}
-
-  void Reseed(UInt32 seed) { state_ = seed; }
-
-  // Generates a random number from [0, range).  Crashes if 'range' is
-  // 0 or greater than kMaxRange.
-  UInt32 Generate(UInt32 range);
-
- private:
-  UInt32 state_;
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(Random);
-};
-
-// Defining a variable of type CompileAssertTypesEqual<T1, T2> will cause a
-// compiler error iff T1 and T2 are different types.
-template <typename T1, typename T2>
-struct CompileAssertTypesEqual;
-
-template <typename T>
-struct CompileAssertTypesEqual<T, T> {
-};
-
-// Removes the reference from a type if it is a reference type,
-// otherwise leaves it unchanged.  This is the same as
-// tr1::remove_reference, which is not widely available yet.
-template <typename T>
-struct RemoveReference { typedef T type; };  // NOLINT
-template <typename T>
-struct RemoveReference<T&> { typedef T type; };  // NOLINT
-
-// A handy wrapper around RemoveReference that works when the argument
-// T depends on template parameters.
-#define GTEST_REMOVE_REFERENCE_(T) \
-    typename ::testing::internal::RemoveReference<T>::type
-
-// Removes const from a type if it is a const type, otherwise leaves
-// it unchanged.  This is the same as tr1::remove_const, which is not
-// widely available yet.
-template <typename T>
-struct RemoveConst { typedef T type; };  // NOLINT
-template <typename T>
-struct RemoveConst<const T> { typedef T type; };  // NOLINT
-
-// MSVC 8.0, Sun C++, and IBM XL C++ have a bug which causes the above
-// definition to fail to remove the const in 'const int[3]' and 'const
-// char[3][4]'.  The following specialization works around the bug.
-template <typename T, size_t N>
-struct RemoveConst<const T[N]> {
-  typedef typename RemoveConst<T>::type type[N];
-};
-
-#if defined(_MSC_VER) && _MSC_VER < 1400
-// This is the only specialization that allows VC++ 7.1 to remove const in
-// 'const int[3] and 'const int[3][4]'.  However, it causes trouble with GCC
-// and thus needs to be conditionally compiled.
-template <typename T, size_t N>
-struct RemoveConst<T[N]> {
-  typedef typename RemoveConst<T>::type type[N];
-};
-#endif
-
-// A handy wrapper around RemoveConst that works when the argument
-// T depends on template parameters.
-#define GTEST_REMOVE_CONST_(T) \
-    typename ::testing::internal::RemoveConst<T>::type
-
-// Turns const U&, U&, const U, and U all into U.
-#define GTEST_REMOVE_REFERENCE_AND_CONST_(T) \
-    GTEST_REMOVE_CONST_(GTEST_REMOVE_REFERENCE_(T))
-
-// Adds reference to a type if it is not a reference type,
-// otherwise leaves it unchanged.  This is the same as
-// tr1::add_reference, which is not widely available yet.
-template <typename T>
-struct AddReference { typedef T& type; };  // NOLINT
-template <typename T>
-struct AddReference<T&> { typedef T& type; };  // NOLINT
-
-// A handy wrapper around AddReference that works when the argument T
-// depends on template parameters.
-#define GTEST_ADD_REFERENCE_(T) \
-    typename ::testing::internal::AddReference<T>::type
-
-// Adds a reference to const on top of T as necessary.  For example,
-// it transforms
-//
-//   char         ==> const char&
-//   const char   ==> const char&
-//   char&        ==> const char&
-//   const char&  ==> const char&
-//
-// The argument T must depend on some template parameters.
-#define GTEST_REFERENCE_TO_CONST_(T) \
-    GTEST_ADD_REFERENCE_(const GTEST_REMOVE_REFERENCE_(T))
-
-// ImplicitlyConvertible<From, To>::value is a compile-time bool
-// constant that's true iff type From can be implicitly converted to
-// type To.
-template <typename From, typename To>
-class ImplicitlyConvertible {
- private:
-  // We need the following helper functions only for their types.
-  // They have no implementations.
-
-  // MakeFrom() is an expression whose type is From.  We cannot simply
-  // use From(), as the type From may not have a public default
-  // constructor.
-  static typename AddReference<From>::type MakeFrom();
-
-  // These two functions are overloaded.  Given an expression
-  // Helper(x), the compiler will pick the first version if x can be
-  // implicitly converted to type To; otherwise it will pick the
-  // second version.
-  //
-  // The first version returns a value of size 1, and the second
-  // version returns a value of size 2.  Therefore, by checking the
-  // size of Helper(x), which can be done at compile time, we can tell
-  // which version of Helper() is used, and hence whether x can be
-  // implicitly converted to type To.
-  static char Helper(To);
-  static char (&Helper(...))[2];  // NOLINT
-
-  // We have to put the 'public' section after the 'private' section,
-  // or MSVC refuses to compile the code.
- public:
-#if defined(__BORLANDC__)
-  // C++Builder cannot use member overload resolution during template
-  // instantiation.  The simplest workaround is to use its C++0x type traits
-  // functions (C++Builder 2009 and above only).
-  static const bool value = __is_convertible(From, To);
-#else
-  // MSVC warns about implicitly converting from double to int for
-  // possible loss of data, so we need to temporarily disable the
-  // warning.
-  GTEST_DISABLE_MSC_WARNINGS_PUSH_(4244)
-  static const bool value =
-      sizeof(Helper(ImplicitlyConvertible::MakeFrom())) == 1;
-  GTEST_DISABLE_MSC_WARNINGS_POP_()
-#endif  // __BORLANDC__
-};
-template <typename From, typename To>
-const bool ImplicitlyConvertible<From, To>::value;
-
-// IsAProtocolMessage<T>::value is a compile-time bool constant that's
-// true iff T is type ProtocolMessage, proto2::Message, or a subclass
-// of those.
-template <typename T>
-struct IsAProtocolMessage
-    : public bool_constant<
-  ImplicitlyConvertible<const T*, const ::ProtocolMessage*>::value ||
-  ImplicitlyConvertible<const T*, const ::proto2::Message*>::value> {
-};
-
-// When the compiler sees expression IsContainerTest<C>(0), if C is an
-// STL-style container class, the first overload of IsContainerTest
-// will be viable (since both C::iterator* and C::const_iterator* are
-// valid types and NULL can be implicitly converted to them).  It will
-// be picked over the second overload as 'int' is a perfect match for
-// the type of argument 0.  If C::iterator or C::const_iterator is not
-// a valid type, the first overload is not viable, and the second
-// overload will be picked.  Therefore, we can determine whether C is
-// a container class by checking the type of IsContainerTest<C>(0).
-// The value of the expression is insignificant.
-//
-// Note that we look for both C::iterator and C::const_iterator.  The
-// reason is that C++ injects the name of a class as a member of the
-// class itself (e.g. you can refer to class iterator as either
-// 'iterator' or 'iterator::iterator').  If we look for C::iterator
-// only, for example, we would mistakenly think that a class named
-// iterator is an STL container.
-//
-// Also note that the simpler approach of overloading
-// IsContainerTest(typename C::const_iterator*) and
-// IsContainerTest(...) doesn't work with Visual Age C++ and Sun C++.
-typedef int IsContainer;
-template <class C>
-IsContainer IsContainerTest(int /* dummy */,
-                            typename C::iterator* /* it */ = NULL,
-                            typename C::const_iterator* /* const_it */ = NULL) {
-  return 0;
-}
-
-typedef char IsNotContainer;
-template <class C>
-IsNotContainer IsContainerTest(long /* dummy */) { return '\0'; }
-
-// EnableIf<condition>::type is void when 'Cond' is true, and
-// undefined when 'Cond' is false.  To use SFINAE to make a function
-// overload only apply when a particular expression is true, add
-// "typename EnableIf<expression>::type* = 0" as the last parameter.
-template<bool> struct EnableIf;
-template<> struct EnableIf<true> { typedef void type; };  // NOLINT
-
-// Utilities for native arrays.
-
-// ArrayEq() compares two k-dimensional native arrays using the
-// elements' operator==, where k can be any integer >= 0.  When k is
-// 0, ArrayEq() degenerates into comparing a single pair of values.
-
-template <typename T, typename U>
-bool ArrayEq(const T* lhs, size_t size, const U* rhs);
-
-// This generic version is used when k is 0.
-template <typename T, typename U>
-inline bool ArrayEq(const T& lhs, const U& rhs) { return lhs == rhs; }
-
-// This overload is used when k >= 1.
-template <typename T, typename U, size_t N>
-inline bool ArrayEq(const T(&lhs)[N], const U(&rhs)[N]) {
-  return internal::ArrayEq(lhs, N, rhs);
-}
-
-// This helper reduces code bloat.  If we instead put its logic inside
-// the previous ArrayEq() function, arrays with different sizes would
-// lead to different copies of the template code.
-template <typename T, typename U>
-bool ArrayEq(const T* lhs, size_t size, const U* rhs) {
-  for (size_t i = 0; i != size; i++) {
-    if (!internal::ArrayEq(lhs[i], rhs[i]))
-      return false;
-  }
-  return true;
-}
-
-// Finds the first element in the iterator range [begin, end) that
-// equals elem.  Element may be a native array type itself.
-template <typename Iter, typename Element>
-Iter ArrayAwareFind(Iter begin, Iter end, const Element& elem) {
-  for (Iter it = begin; it != end; ++it) {
-    if (internal::ArrayEq(*it, elem))
-      return it;
-  }
-  return end;
-}
-
-// CopyArray() copies a k-dimensional native array using the elements'
-// operator=, where k can be any integer >= 0.  When k is 0,
-// CopyArray() degenerates into copying a single value.
-
-template <typename T, typename U>
-void CopyArray(const T* from, size_t size, U* to);
-
-// This generic version is used when k is 0.
-template <typename T, typename U>
-inline void CopyArray(const T& from, U* to) { *to = from; }
-
-// This overload is used when k >= 1.
-template <typename T, typename U, size_t N>
-inline void CopyArray(const T(&from)[N], U(*to)[N]) {
-  internal::CopyArray(from, N, *to);
-}
-
-// This helper reduces code bloat.  If we instead put its logic inside
-// the previous CopyArray() function, arrays with different sizes
-// would lead to different copies of the template code.
-template <typename T, typename U>
-void CopyArray(const T* from, size_t size, U* to) {
-  for (size_t i = 0; i != size; i++) {
-    internal::CopyArray(from[i], to + i);
-  }
-}
-
-// The relation between an NativeArray object (see below) and the
-// native array it represents.
-// We use 2 different structs to allow non-copyable types to be used, as long
-// as RelationToSourceReference() is passed.
-struct RelationToSourceReference {};
-struct RelationToSourceCopy {};
-
-// Adapts a native array to a read-only STL-style container.  Instead
-// of the complete STL container concept, this adaptor only implements
-// members useful for Google Mock's container matchers.  New members
-// should be added as needed.  To simplify the implementation, we only
-// support Element being a raw type (i.e. having no top-level const or
-// reference modifier).  It's the client's responsibility to satisfy
-// this requirement.  Element can be an array type itself (hence
-// multi-dimensional arrays are supported).
-template <typename Element>
-class NativeArray {
- public:
-  // STL-style container typedefs.
-  typedef Element value_type;
-  typedef Element* iterator;
-  typedef const Element* const_iterator;
-
-  // Constructs from a native array. References the source.
-  NativeArray(const Element* array, size_t count, RelationToSourceReference) {
-    InitRef(array, count);
-  }
-
-  // Constructs from a native array. Copies the source.
-  NativeArray(const Element* array, size_t count, RelationToSourceCopy) {
-    InitCopy(array, count);
-  }
-
-  // Copy constructor.
-  NativeArray(const NativeArray& rhs) {
-    (this->*rhs.clone_)(rhs.array_, rhs.size_);
-  }
-
-  ~NativeArray() {
-    if (clone_ != &NativeArray::InitRef)
-      delete[] array_;
-  }
-
-  // STL-style container methods.
-  size_t size() const { return size_; }
-  const_iterator begin() const { return array_; }
-  const_iterator end() const { return array_ + size_; }
-  bool operator==(const NativeArray& rhs) const {
-    return size() == rhs.size() &&
-        ArrayEq(begin(), size(), rhs.begin());
-  }
-
- private:
-  enum {
-    kCheckTypeIsNotConstOrAReference = StaticAssertTypeEqHelper<
-        Element, GTEST_REMOVE_REFERENCE_AND_CONST_(Element)>::value,
-  };
-
-  // Initializes this object with a copy of the input.
-  void InitCopy(const Element* array, size_t a_size) {
-    Element* const copy = new Element[a_size];
-    CopyArray(array, a_size, copy);
-    array_ = copy;
-    size_ = a_size;
-    clone_ = &NativeArray::InitCopy;
-  }
-
-  // Initializes this object with a reference of the input.
-  void InitRef(const Element* array, size_t a_size) {
-    array_ = array;
-    size_ = a_size;
-    clone_ = &NativeArray::InitRef;
-  }
-
-  const Element* array_;
-  size_t size_;
-  void (NativeArray::*clone_)(const Element*, size_t);
-
-  GTEST_DISALLOW_ASSIGN_(NativeArray);
-};
-
-}  // namespace internal
-}  // namespace testing
-
-#define GTEST_MESSAGE_AT_(file, line, message, result_type) \
-  ::testing::internal::AssertHelper(result_type, file, line, message) \
-    = ::testing::Message()
-
-#define GTEST_MESSAGE_(message, result_type) \
-  GTEST_MESSAGE_AT_(__FILE__, __LINE__, message, result_type)
-
-#define GTEST_FATAL_FAILURE_(message) \
-  return GTEST_MESSAGE_(message, ::testing::TestPartResult::kFatalFailure)
-
-#define GTEST_NONFATAL_FAILURE_(message) \
-  GTEST_MESSAGE_(message, ::testing::TestPartResult::kNonFatalFailure)
-
-#define GTEST_SUCCESS_(message) \
-  GTEST_MESSAGE_(message, ::testing::TestPartResult::kSuccess)
-
-// Suppresses MSVC warnings 4072 (unreachable code) for the code following
-// statement if it returns or throws (or doesn't return or throw in some
-// situations).
-#define GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement) \
-  if (::testing::internal::AlwaysTrue()) { statement; }
-
-#define GTEST_TEST_THROW_(statement, expected_exception, fail) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (::testing::internal::ConstCharPtr gtest_msg = "") { \
-    bool gtest_caught_expected = false; \
-    try { \
-      GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement); \
-    } \
-    catch (expected_exception const&) { \
-      gtest_caught_expected = true; \
-    } \
-    catch (...) { \
-      gtest_msg.value = \
-          "Expected: " #statement " throws an exception of type " \
-          #expected_exception ".\n  Actual: it throws a different type."; \
-      goto GTEST_CONCAT_TOKEN_(gtest_label_testthrow_, __LINE__); \
-    } \
-    if (!gtest_caught_expected) { \
-      gtest_msg.value = \
-          "Expected: " #statement " throws an exception of type " \
-          #expected_exception ".\n  Actual: it throws nothing."; \
-      goto GTEST_CONCAT_TOKEN_(gtest_label_testthrow_, __LINE__); \
-    } \
-  } else \
-    GTEST_CONCAT_TOKEN_(gtest_label_testthrow_, __LINE__): \
-      fail(gtest_msg.value)
-
-#define GTEST_TEST_NO_THROW_(statement, fail) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (::testing::internal::AlwaysTrue()) { \
-    try { \
-      GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement); \
-    } \
-    catch (...) { \
-      goto GTEST_CONCAT_TOKEN_(gtest_label_testnothrow_, __LINE__); \
-    } \
-  } else \
-    GTEST_CONCAT_TOKEN_(gtest_label_testnothrow_, __LINE__): \
-      fail("Expected: " #statement " doesn't throw an exception.\n" \
-           "  Actual: it throws.")
-
-#define GTEST_TEST_ANY_THROW_(statement, fail) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (::testing::internal::AlwaysTrue()) { \
-    bool gtest_caught_any = false; \
-    try { \
-      GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement); \
-    } \
-    catch (...) { \
-      gtest_caught_any = true; \
-    } \
-    if (!gtest_caught_any) { \
-      goto GTEST_CONCAT_TOKEN_(gtest_label_testanythrow_, __LINE__); \
-    } \
-  } else \
-    GTEST_CONCAT_TOKEN_(gtest_label_testanythrow_, __LINE__): \
-      fail("Expected: " #statement " throws an exception.\n" \
-           "  Actual: it doesn't.")
-
-
-// Implements Boolean test assertions such as EXPECT_TRUE. expression can be
-// either a boolean expression or an AssertionResult. text is a textual
-// represenation of expression as it was passed into the EXPECT_TRUE.
-#define GTEST_TEST_BOOLEAN_(expression, text, actual, expected, fail) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (const ::testing::AssertionResult gtest_ar_ = \
-      ::testing::AssertionResult(expression)) \
-    ; \
-  else \
-    fail(::testing::internal::GetBoolAssertionFailureMessage(\
-        gtest_ar_, text, #actual, #expected).c_str())
-
-#define GTEST_TEST_NO_FATAL_FAILURE_(statement, fail) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (::testing::internal::AlwaysTrue()) { \
-    ::testing::internal::HasNewFatalFailureHelper gtest_fatal_failure_checker; \
-    GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement); \
-    if (gtest_fatal_failure_checker.has_new_fatal_failure()) { \
-      goto GTEST_CONCAT_TOKEN_(gtest_label_testnofatal_, __LINE__); \
-    } \
-  } else \
-    GTEST_CONCAT_TOKEN_(gtest_label_testnofatal_, __LINE__): \
-      fail("Expected: " #statement " doesn't generate new fatal " \
-           "failures in the current thread.\n" \
-           "  Actual: it does.")
-
-// Expands to the name of the class that implements the given test.
-#define GTEST_TEST_CLASS_NAME_(test_case_name, test_name) \
-  test_case_name##_##test_name##_Test
-
-// Helper macro for defining tests.
-#define GTEST_TEST_(test_case_name, test_name, parent_class, parent_id)\
-class GTEST_TEST_CLASS_NAME_(test_case_name, test_name) : public parent_class {\
- public:\
-  GTEST_TEST_CLASS_NAME_(test_case_name, test_name)() {}\
- private:\
-  virtual void TestBody();\
-  static ::testing::TestInfo* const test_info_ GTEST_ATTRIBUTE_UNUSED_;\
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(\
-      GTEST_TEST_CLASS_NAME_(test_case_name, test_name));\
-};\
-\
-::testing::TestInfo* const GTEST_TEST_CLASS_NAME_(test_case_name, test_name)\
-  ::test_info_ =\
-    ::testing::internal::MakeAndRegisterTestInfo(\
-        #test_case_name, #test_name, NULL, NULL, \
-        ::testing::internal::CodeLocation(__FILE__, __LINE__), \
-        (parent_id), \
-        parent_class::SetUpTestCase, \
-        parent_class::TearDownTestCase, \
-        new ::testing::internal::TestFactoryImpl<\
-            GTEST_TEST_CLASS_NAME_(test_case_name, test_name)>);\
-void GTEST_TEST_CLASS_NAME_(test_case_name, test_name)::TestBody()
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_INTERNAL_H_
-
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-//
-// The Google C++ Testing Framework (Google Test)
-//
-// This header file defines the public API for death tests.  It is
-// #included by gtest.h so a user doesn't need to include this
-// directly.
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_DEATH_TEST_H_
-#define GTEST_INCLUDE_GTEST_GTEST_DEATH_TEST_H_
-
-// Copyright 2005, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Authors: wan@google.com (Zhanyong Wan), eefacm@gmail.com (Sean Mcafee)
-//
-// The Google C++ Testing Framework (Google Test)
-//
-// This header file defines internal utilities needed for implementing
-// death tests.  They are subject to change without notice.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_DEATH_TEST_INTERNAL_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_DEATH_TEST_INTERNAL_H_
-
-
-#include <stdio.h>
-
-namespace testing {
-namespace internal {
-
-GTEST_DECLARE_string_(internal_run_death_test);
-
-// Names of the flags (needed for parsing Google Test flags).
-const char kDeathTestStyleFlag[] = "death_test_style";
-const char kDeathTestUseFork[] = "death_test_use_fork";
-const char kInternalRunDeathTestFlag[] = "internal_run_death_test";
-
-#if GTEST_HAS_DEATH_TEST
-
-// DeathTest is a class that hides much of the complexity of the
-// GTEST_DEATH_TEST_ macro.  It is abstract; its static Create method
-// returns a concrete class that depends on the prevailing death test
-// style, as defined by the --gtest_death_test_style and/or
-// --gtest_internal_run_death_test flags.
-
-// In describing the results of death tests, these terms are used with
-// the corresponding definitions:
-//
-// exit status:  The integer exit information in the format specified
-//               by wait(2)
-// exit code:    The integer code passed to exit(3), _exit(2), or
-//               returned from main()
-class GTEST_API_ DeathTest {
- public:
-  // Create returns false if there was an error determining the
-  // appropriate action to take for the current death test; for example,
-  // if the gtest_death_test_style flag is set to an invalid value.
-  // The LastMessage method will return a more detailed message in that
-  // case.  Otherwise, the DeathTest pointer pointed to by the "test"
-  // argument is set.  If the death test should be skipped, the pointer
-  // is set to NULL; otherwise, it is set to the address of a new concrete
-  // DeathTest object that controls the execution of the current test.
-  static bool Create(const char* statement, const RE* regex,
-                     const char* file, int line, DeathTest** test);
-  DeathTest();
-  virtual ~DeathTest() { }
-
-  // A helper class that aborts a death test when it's deleted.
-  class ReturnSentinel {
-   public:
-    explicit ReturnSentinel(DeathTest* test) : test_(test) { }
-    ~ReturnSentinel() { test_->Abort(TEST_ENCOUNTERED_RETURN_STATEMENT); }
-   private:
-    DeathTest* const test_;
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(ReturnSentinel);
-  } GTEST_ATTRIBUTE_UNUSED_;
-
-  // An enumeration of possible roles that may be taken when a death
-  // test is encountered.  EXECUTE means that the death test logic should
-  // be executed immediately.  OVERSEE means that the program should prepare
-  // the appropriate environment for a child process to execute the death
-  // test, then wait for it to complete.
-  enum TestRole { OVERSEE_TEST, EXECUTE_TEST };
-
-  // An enumeration of the three reasons that a test might be aborted.
-  enum AbortReason {
-    TEST_ENCOUNTERED_RETURN_STATEMENT,
-    TEST_THREW_EXCEPTION,
-    TEST_DID_NOT_DIE
-  };
-
-  // Assumes one of the above roles.
-  virtual TestRole AssumeRole() = 0;
-
-  // Waits for the death test to finish and returns its status.
-  virtual int Wait() = 0;
-
-  // Returns true if the death test passed; that is, the test process
-  // exited during the test, its exit status matches a user-supplied
-  // predicate, and its stderr output matches a user-supplied regular
-  // expression.
-  // The user-supplied predicate may be a macro expression rather
-  // than a function pointer or functor, or else Wait and Passed could
-  // be combined.
-  virtual bool Passed(bool exit_status_ok) = 0;
-
-  // Signals that the death test did not die as expected.
-  virtual void Abort(AbortReason reason) = 0;
-
-  // Returns a human-readable outcome message regarding the outcome of
-  // the last death test.
-  static const char* LastMessage();
-
-  static void set_last_death_test_message(const std::string& message);
-
- private:
-  // A string containing a description of the outcome of the last death test.
-  static std::string last_death_test_message_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(DeathTest);
-};
-
-// Factory interface for death tests.  May be mocked out for testing.
-class DeathTestFactory {
- public:
-  virtual ~DeathTestFactory() { }
-  virtual bool Create(const char* statement, const RE* regex,
-                      const char* file, int line, DeathTest** test) = 0;
-};
-
-// A concrete DeathTestFactory implementation for normal use.
-class DefaultDeathTestFactory : public DeathTestFactory {
- public:
-  virtual bool Create(const char* statement, const RE* regex,
-                      const char* file, int line, DeathTest** test);
-};
-
-// Returns true if exit_status describes a process that was terminated
-// by a signal, or exited normally with a nonzero exit code.
-GTEST_API_ bool ExitedUnsuccessfully(int exit_status);
-
-// Traps C++ exceptions escaping statement and reports them as test
-// failures. Note that trapping SEH exceptions is not implemented here.
-# if GTEST_HAS_EXCEPTIONS
-#  define GTEST_EXECUTE_DEATH_TEST_STATEMENT_(statement, death_test) \
-  try { \
-    GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement); \
-  } catch (const ::std::exception& gtest_exception) { \
-    fprintf(\
-        stderr, \
-        "\n%s: Caught std::exception-derived exception escaping the " \
-        "death test statement. Exception message: %s\n", \
-        ::testing::internal::FormatFileLocation(__FILE__, __LINE__).c_str(), \
-        gtest_exception.what()); \
-    fflush(stderr); \
-    death_test->Abort(::testing::internal::DeathTest::TEST_THREW_EXCEPTION); \
-  } catch (...) { \
-    death_test->Abort(::testing::internal::DeathTest::TEST_THREW_EXCEPTION); \
-  }
-
-# else
-#  define GTEST_EXECUTE_DEATH_TEST_STATEMENT_(statement, death_test) \
-  GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement)
-
-# endif
-
-// This macro is for implementing ASSERT_DEATH*, EXPECT_DEATH*,
-// ASSERT_EXIT*, and EXPECT_EXIT*.
-# define GTEST_DEATH_TEST_(statement, predicate, regex, fail) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (::testing::internal::AlwaysTrue()) { \
-    const ::testing::internal::RE& gtest_regex = (regex); \
-    ::testing::internal::DeathTest* gtest_dt; \
-    if (!::testing::internal::DeathTest::Create(#statement, &gtest_regex, \
-        __FILE__, __LINE__, &gtest_dt)) { \
-      goto GTEST_CONCAT_TOKEN_(gtest_label_, __LINE__); \
-    } \
-    if (gtest_dt != NULL) { \
-      ::testing::internal::scoped_ptr< ::testing::internal::DeathTest> \
-          gtest_dt_ptr(gtest_dt); \
-      switch (gtest_dt->AssumeRole()) { \
-        case ::testing::internal::DeathTest::OVERSEE_TEST: \
-          if (!gtest_dt->Passed(predicate(gtest_dt->Wait()))) { \
-            goto GTEST_CONCAT_TOKEN_(gtest_label_, __LINE__); \
-          } \
-          break; \
-        case ::testing::internal::DeathTest::EXECUTE_TEST: { \
-          ::testing::internal::DeathTest::ReturnSentinel \
-              gtest_sentinel(gtest_dt); \
-          GTEST_EXECUTE_DEATH_TEST_STATEMENT_(statement, gtest_dt); \
-          gtest_dt->Abort(::testing::internal::DeathTest::TEST_DID_NOT_DIE); \
-          break; \
-        } \
-        default: \
-          break; \
-      } \
-    } \
-  } else \
-    GTEST_CONCAT_TOKEN_(gtest_label_, __LINE__): \
-      fail(::testing::internal::DeathTest::LastMessage())
-// The symbol "fail" here expands to something into which a message
-// can be streamed.
-
-// This macro is for implementing ASSERT/EXPECT_DEBUG_DEATH when compiled in
-// NDEBUG mode. In this case we need the statements to be executed, the regex is
-// ignored, and the macro must accept a streamed message even though the message
-// is never printed.
-# define GTEST_EXECUTE_STATEMENT_(statement, regex) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (::testing::internal::AlwaysTrue()) { \
-     GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement); \
-  } else \
-    ::testing::Message()
-
-// A class representing the parsed contents of the
-// --gtest_internal_run_death_test flag, as it existed when
-// RUN_ALL_TESTS was called.
-class InternalRunDeathTestFlag {
- public:
-  InternalRunDeathTestFlag(const std::string& a_file,
-                           int a_line,
-                           int an_index,
-                           int a_write_fd)
-      : file_(a_file), line_(a_line), index_(an_index),
-        write_fd_(a_write_fd) {}
-
-  ~InternalRunDeathTestFlag() {
-    if (write_fd_ >= 0)
-      posix::Close(write_fd_);
-  }
-
-  const std::string& file() const { return file_; }
-  int line() const { return line_; }
-  int index() const { return index_; }
-  int write_fd() const { return write_fd_; }
-
- private:
-  std::string file_;
-  int line_;
-  int index_;
-  int write_fd_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(InternalRunDeathTestFlag);
-};
-
-// Returns a newly created InternalRunDeathTestFlag object with fields
-// initialized from the GTEST_FLAG(internal_run_death_test) flag if
-// the flag is specified; otherwise returns NULL.
-InternalRunDeathTestFlag* ParseInternalRunDeathTestFlag();
-
-#else  // GTEST_HAS_DEATH_TEST
-
-// This macro is used for implementing macros such as
-// EXPECT_DEATH_IF_SUPPORTED and ASSERT_DEATH_IF_SUPPORTED on systems where
-// death tests are not supported. Those macros must compile on such systems
-// iff EXPECT_DEATH and ASSERT_DEATH compile with the same parameters on
-// systems that support death tests. This allows one to write such a macro
-// on a system that does not support death tests and be sure that it will
-// compile on a death-test supporting system.
-//
-// Parameters:
-//   statement -  A statement that a macro such as EXPECT_DEATH would test
-//                for program termination. This macro has to make sure this
-//                statement is compiled but not executed, to ensure that
-//                EXPECT_DEATH_IF_SUPPORTED compiles with a certain
-//                parameter iff EXPECT_DEATH compiles with it.
-//   regex     -  A regex that a macro such as EXPECT_DEATH would use to test
-//                the output of statement.  This parameter has to be
-//                compiled but not evaluated by this macro, to ensure that
-//                this macro only accepts expressions that a macro such as
-//                EXPECT_DEATH would accept.
-//   terminator - Must be an empty statement for EXPECT_DEATH_IF_SUPPORTED
-//                and a return statement for ASSERT_DEATH_IF_SUPPORTED.
-//                This ensures that ASSERT_DEATH_IF_SUPPORTED will not
-//                compile inside functions where ASSERT_DEATH doesn't
-//                compile.
-//
-//  The branch that has an always false condition is used to ensure that
-//  statement and regex are compiled (and thus syntactically correct) but
-//  never executed. The unreachable code macro protects the terminator
-//  statement from generating an 'unreachable code' warning in case
-//  statement unconditionally returns or throws. The Message constructor at
-//  the end allows the syntax of streaming additional messages into the
-//  macro, for compilational compatibility with EXPECT_DEATH/ASSERT_DEATH.
-# define GTEST_UNSUPPORTED_DEATH_TEST_(statement, regex, terminator) \
-    GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-    if (::testing::internal::AlwaysTrue()) { \
-      GTEST_LOG_(WARNING) \
-          << "Death tests are not supported on this platform.\n" \
-          << "Statement '" #statement "' cannot be verified."; \
-    } else if (::testing::internal::AlwaysFalse()) { \
-      ::testing::internal::RE::PartialMatch(".*", (regex)); \
-      GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement); \
-      terminator; \
-    } else \
-      ::testing::Message()
-
-#endif  // GTEST_HAS_DEATH_TEST
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_DEATH_TEST_INTERNAL_H_
-
-namespace testing {
-
-// This flag controls the style of death tests.  Valid values are "threadsafe",
-// meaning that the death test child process will re-execute the test binary
-// from the start, running only a single death test, or "fast",
-// meaning that the child process will execute the test logic immediately
-// after forking.
-GTEST_DECLARE_string_(death_test_style);
-
-#if GTEST_HAS_DEATH_TEST
-
-namespace internal {
-
-// Returns a Boolean value indicating whether the caller is currently
-// executing in the context of the death test child process.  Tools such as
-// Valgrind heap checkers may need this to modify their behavior in death
-// tests.  IMPORTANT: This is an internal utility.  Using it may break the
-// implementation of death tests.  User code MUST NOT use it.
-GTEST_API_ bool InDeathTestChild();
-
-}  // namespace internal
-
-// The following macros are useful for writing death tests.
-
-// Here's what happens when an ASSERT_DEATH* or EXPECT_DEATH* is
-// executed:
-//
-//   1. It generates a warning if there is more than one active
-//   thread.  This is because it's safe to fork() or clone() only
-//   when there is a single thread.
-//
-//   2. The parent process clone()s a sub-process and runs the death
-//   test in it; the sub-process exits with code 0 at the end of the
-//   death test, if it hasn't exited already.
-//
-//   3. The parent process waits for the sub-process to terminate.
-//
-//   4. The parent process checks the exit code and error message of
-//   the sub-process.
-//
-// Examples:
-//
-//   ASSERT_DEATH(server.SendMessage(56, "Hello"), "Invalid port number");
-//   for (int i = 0; i < 5; i++) {
-//     EXPECT_DEATH(server.ProcessRequest(i),
-//                  "Invalid request .* in ProcessRequest()")
-//                  << "Failed to die on request " << i;
-//   }
-//
-//   ASSERT_EXIT(server.ExitNow(), ::testing::ExitedWithCode(0), "Exiting");
-//
-//   bool KilledBySIGHUP(int exit_code) {
-//     return WIFSIGNALED(exit_code) && WTERMSIG(exit_code) == SIGHUP;
-//   }
-//
-//   ASSERT_EXIT(client.HangUpServer(), KilledBySIGHUP, "Hanging up!");
-//
-// On the regular expressions used in death tests:
-//
-//   On POSIX-compliant systems (*nix), we use the <regex.h> library,
-//   which uses the POSIX extended regex syntax.
-//
-//   On other platforms (e.g. Windows), we only support a simple regex
-//   syntax implemented as part of Google Test.  This limited
-//   implementation should be enough most of the time when writing
-//   death tests; though it lacks many features you can find in PCRE
-//   or POSIX extended regex syntax.  For example, we don't support
-//   union ("x|y"), grouping ("(xy)"), brackets ("[xy]"), and
-//   repetition count ("x{5,7}"), among others.
-//
-//   Below is the syntax that we do support.  We chose it to be a
-//   subset of both PCRE and POSIX extended regex, so it's easy to
-//   learn wherever you come from.  In the following: 'A' denotes a
-//   literal character, period (.), or a single \\ escape sequence;
-//   'x' and 'y' denote regular expressions; 'm' and 'n' are for
-//   natural numbers.
-//
-//     c     matches any literal character c
-//     \\d   matches any decimal digit
-//     \\D   matches any character that's not a decimal digit
-//     \\f   matches \f
-//     \\n   matches \n
-//     \\r   matches \r
-//     \\s   matches any ASCII whitespace, including \n
-//     \\S   matches any character that's not a whitespace
-//     \\t   matches \t
-//     \\v   matches \v
-//     \\w   matches any letter, _, or decimal digit
-//     \\W   matches any character that \\w doesn't match
-//     \\c   matches any literal character c, which must be a punctuation
-//     .     matches any single character except \n
-//     A?    matches 0 or 1 occurrences of A
-//     A*    matches 0 or many occurrences of A
-//     A+    matches 1 or many occurrences of A
-//     ^     matches the beginning of a string (not that of each line)
-//     $     matches the end of a string (not that of each line)
-//     xy    matches x followed by y
-//
-//   If you accidentally use PCRE or POSIX extended regex features
-//   not implemented by us, you will get a run-time failure.  In that
-//   case, please try to rewrite your regular expression within the
-//   above syntax.
-//
-//   This implementation is *not* meant to be as highly tuned or robust
-//   as a compiled regex library, but should perform well enough for a
-//   death test, which already incurs significant overhead by launching
-//   a child process.
-//
-// Known caveats:
-//
-//   A "threadsafe" style death test obtains the path to the test
-//   program from argv[0] and re-executes it in the sub-process.  For
-//   simplicity, the current implementation doesn't search the PATH
-//   when launching the sub-process.  This means that the user must
-//   invoke the test program via a path that contains at least one
-//   path separator (e.g. path/to/foo_test and
-//   /absolute/path/to/bar_test are fine, but foo_test is not).  This
-//   is rarely a problem as people usually don't put the test binary
-//   directory in PATH.
-//
-// TODO(wan@google.com): make thread-safe death tests search the PATH.
-
-// Asserts that a given statement causes the program to exit, with an
-// integer exit status that satisfies predicate, and emitting error output
-// that matches regex.
-# define ASSERT_EXIT(statement, predicate, regex) \
-    GTEST_DEATH_TEST_(statement, predicate, regex, GTEST_FATAL_FAILURE_)
-
-// Like ASSERT_EXIT, but continues on to successive tests in the
-// test case, if any:
-# define EXPECT_EXIT(statement, predicate, regex) \
-    GTEST_DEATH_TEST_(statement, predicate, regex, GTEST_NONFATAL_FAILURE_)
-
-// Asserts that a given statement causes the program to exit, either by
-// explicitly exiting with a nonzero exit code or being killed by a
-// signal, and emitting error output that matches regex.
-# define ASSERT_DEATH(statement, regex) \
-    ASSERT_EXIT(statement, ::testing::internal::ExitedUnsuccessfully, regex)
-
-// Like ASSERT_DEATH, but continues on to successive tests in the
-// test case, if any:
-# define EXPECT_DEATH(statement, regex) \
-    EXPECT_EXIT(statement, ::testing::internal::ExitedUnsuccessfully, regex)
-
-// Two predicate classes that can be used in {ASSERT,EXPECT}_EXIT*:
-
-// Tests that an exit code describes a normal exit with a given exit code.
-class GTEST_API_ ExitedWithCode {
- public:
-  explicit ExitedWithCode(int exit_code);
-  bool operator()(int exit_status) const;
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ExitedWithCode& other);
-
-  const int exit_code_;
-};
-
-# if !GTEST_OS_WINDOWS
-// Tests that an exit code describes an exit due to termination by a
-// given signal.
-class GTEST_API_ KilledBySignal {
- public:
-  explicit KilledBySignal(int signum);
-  bool operator()(int exit_status) const;
- private:
-  const int signum_;
-};
-# endif  // !GTEST_OS_WINDOWS
-
-// EXPECT_DEBUG_DEATH asserts that the given statements die in debug mode.
-// The death testing framework causes this to have interesting semantics,
-// since the sideeffects of the call are only visible in opt mode, and not
-// in debug mode.
-//
-// In practice, this can be used to test functions that utilize the
-// LOG(DFATAL) macro using the following style:
-//
-// int DieInDebugOr12(int* sideeffect) {
-//   if (sideeffect) {
-//     *sideeffect = 12;
-//   }
-//   LOG(DFATAL) << "death";
-//   return 12;
-// }
-//
-// TEST(TestCase, TestDieOr12WorksInDgbAndOpt) {
-//   int sideeffect = 0;
-//   // Only asserts in dbg.
-//   EXPECT_DEBUG_DEATH(DieInDebugOr12(&sideeffect), "death");
-//
-// #ifdef NDEBUG
-//   // opt-mode has sideeffect visible.
-//   EXPECT_EQ(12, sideeffect);
-// #else
-//   // dbg-mode no visible sideeffect.
-//   EXPECT_EQ(0, sideeffect);
-// #endif
-// }
-//
-// This will assert that DieInDebugReturn12InOpt() crashes in debug
-// mode, usually due to a DCHECK or LOG(DFATAL), but returns the
-// appropriate fallback value (12 in this case) in opt mode. If you
-// need to test that a function has appropriate side-effects in opt
-// mode, include assertions against the side-effects.  A general
-// pattern for this is:
-//
-// EXPECT_DEBUG_DEATH({
-//   // Side-effects here will have an effect after this statement in
-//   // opt mode, but none in debug mode.
-//   EXPECT_EQ(12, DieInDebugOr12(&sideeffect));
-// }, "death");
-//
-# ifdef NDEBUG
-
-#  define EXPECT_DEBUG_DEATH(statement, regex) \
-  GTEST_EXECUTE_STATEMENT_(statement, regex)
-
-#  define ASSERT_DEBUG_DEATH(statement, regex) \
-  GTEST_EXECUTE_STATEMENT_(statement, regex)
-
-# else
-
-#  define EXPECT_DEBUG_DEATH(statement, regex) \
-  EXPECT_DEATH(statement, regex)
-
-#  define ASSERT_DEBUG_DEATH(statement, regex) \
-  ASSERT_DEATH(statement, regex)
-
-# endif  // NDEBUG for EXPECT_DEBUG_DEATH
-#endif  // GTEST_HAS_DEATH_TEST
-
-// EXPECT_DEATH_IF_SUPPORTED(statement, regex) and
-// ASSERT_DEATH_IF_SUPPORTED(statement, regex) expand to real death tests if
-// death tests are supported; otherwise they just issue a warning.  This is
-// useful when you are combining death test assertions with normal test
-// assertions in one test.
-#if GTEST_HAS_DEATH_TEST
-# define EXPECT_DEATH_IF_SUPPORTED(statement, regex) \
-    EXPECT_DEATH(statement, regex)
-# define ASSERT_DEATH_IF_SUPPORTED(statement, regex) \
-    ASSERT_DEATH(statement, regex)
-#else
-# define EXPECT_DEATH_IF_SUPPORTED(statement, regex) \
-    GTEST_UNSUPPORTED_DEATH_TEST_(statement, regex, )
-# define ASSERT_DEATH_IF_SUPPORTED(statement, regex) \
-    GTEST_UNSUPPORTED_DEATH_TEST_(statement, regex, return)
-#endif
-
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_DEATH_TEST_H_
-// This file was GENERATED by command:
-//     pump.py gtest-param-test.h.pump
-// DO NOT EDIT BY HAND!!!
-
-// Copyright 2008, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Authors: vladl@google.com (Vlad Losev)
-//
-// Macros and functions for implementing parameterized tests
-// in Google C++ Testing Framework (Google Test)
-//
-// This file is generated by a SCRIPT.  DO NOT EDIT BY HAND!
-//
-#ifndef GTEST_INCLUDE_GTEST_GTEST_PARAM_TEST_H_
-#define GTEST_INCLUDE_GTEST_GTEST_PARAM_TEST_H_
-
-
-// Value-parameterized tests allow you to test your code with different
-// parameters without writing multiple copies of the same test.
-//
-// Here is how you use value-parameterized tests:
-
-#if 0
-
-// To write value-parameterized tests, first you should define a fixture
-// class. It is usually derived from testing::TestWithParam<T> (see below for
-// another inheritance scheme that's sometimes useful in more complicated
-// class hierarchies), where the type of your parameter values.
-// TestWithParam<T> is itself derived from testing::Test. T can be any
-// copyable type. If it's a raw pointer, you are responsible for managing the
-// lifespan of the pointed values.
-
-class FooTest : public ::testing::TestWithParam<const char*> {
-  // You can implement all the usual class fixture members here.
-};
-
-// Then, use the TEST_P macro to define as many parameterized tests
-// for this fixture as you want. The _P suffix is for "parameterized"
-// or "pattern", whichever you prefer to think.
-
-TEST_P(FooTest, DoesBlah) {
-  // Inside a test, access the test parameter with the GetParam() method
-  // of the TestWithParam<T> class:
-  EXPECT_TRUE(foo.Blah(GetParam()));
-  ...
-}
-
-TEST_P(FooTest, HasBlahBlah) {
-  ...
-}
-
-// Finally, you can use INSTANTIATE_TEST_CASE_P to instantiate the test
-// case with any set of parameters you want. Google Test defines a number
-// of functions for generating test parameters. They return what we call
-// (surprise!) parameter generators. Here is a  summary of them, which
-// are all in the testing namespace:
-//
-//
-//  Range(begin, end [, step]) - Yields values {begin, begin+step,
-//                               begin+step+step, ...}. The values do not
-//                               include end. step defaults to 1.
-//  Values(v1, v2, ..., vN)    - Yields values {v1, v2, ..., vN}.
-//  ValuesIn(container)        - Yields values from a C-style array, an STL
-//  ValuesIn(begin,end)          container, or an iterator range [begin, end).
-//  Bool()                     - Yields sequence {false, true}.
-//  Combine(g1, g2, ..., gN)   - Yields all combinations (the Cartesian product
-//                               for the math savvy) of the values generated
-//                               by the N generators.
-//
-// For more details, see comments at the definitions of these functions below
-// in this file.
-//
-// The following statement will instantiate tests from the FooTest test case
-// each with parameter values "meeny", "miny", and "moe".
-
-INSTANTIATE_TEST_CASE_P(InstantiationName,
-                        FooTest,
-                        Values("meeny", "miny", "moe"));
-
-// To distinguish different instances of the pattern, (yes, you
-// can instantiate it more then once) the first argument to the
-// INSTANTIATE_TEST_CASE_P macro is a prefix that will be added to the
-// actual test case name. Remember to pick unique prefixes for different
-// instantiations. The tests from the instantiation above will have
-// these names:
-//
-//    * InstantiationName/FooTest.DoesBlah/0 for "meeny"
-//    * InstantiationName/FooTest.DoesBlah/1 for "miny"
-//    * InstantiationName/FooTest.DoesBlah/2 for "moe"
-//    * InstantiationName/FooTest.HasBlahBlah/0 for "meeny"
-//    * InstantiationName/FooTest.HasBlahBlah/1 for "miny"
-//    * InstantiationName/FooTest.HasBlahBlah/2 for "moe"
-//
-// You can use these names in --gtest_filter.
-//
-// This statement will instantiate all tests from FooTest again, each
-// with parameter values "cat" and "dog":
-
-const char* pets[] = {"cat", "dog"};
-INSTANTIATE_TEST_CASE_P(AnotherInstantiationName, FooTest, ValuesIn(pets));
-
-// The tests from the instantiation above will have these names:
-//
-//    * AnotherInstantiationName/FooTest.DoesBlah/0 for "cat"
-//    * AnotherInstantiationName/FooTest.DoesBlah/1 for "dog"
-//    * AnotherInstantiationName/FooTest.HasBlahBlah/0 for "cat"
-//    * AnotherInstantiationName/FooTest.HasBlahBlah/1 for "dog"
-//
-// Please note that INSTANTIATE_TEST_CASE_P will instantiate all tests
-// in the given test case, whether their definitions come before or
-// AFTER the INSTANTIATE_TEST_CASE_P statement.
-//
-// Please also note that generator expressions (including parameters to the
-// generators) are evaluated in InitGoogleTest(), after main() has started.
-// This allows the user on one hand, to adjust generator parameters in order
-// to dynamically determine a set of tests to run and on the other hand,
-// give the user a chance to inspect the generated tests with Google Test
-// reflection API before RUN_ALL_TESTS() is executed.
-//
-// You can see samples/sample7_unittest.cc and samples/sample8_unittest.cc
-// for more examples.
-//
-// In the future, we plan to publish the API for defining new parameter
-// generators. But for now this interface remains part of the internal
-// implementation and is subject to change.
-//
-//
-// A parameterized test fixture must be derived from testing::Test and from
-// testing::WithParamInterface<T>, where T is the type of the parameter
-// values. Inheriting from TestWithParam<T> satisfies that requirement because
-// TestWithParam<T> inherits from both Test and WithParamInterface. In more
-// complicated hierarchies, however, it is occasionally useful to inherit
-// separately from Test and WithParamInterface. For example:
-
-class BaseTest : public ::testing::Test {
-  // You can inherit all the usual members for a non-parameterized test
-  // fixture here.
-};
-
-class DerivedTest : public BaseTest, public ::testing::WithParamInterface<int> {
-  // The usual test fixture members go here too.
-};
-
-TEST_F(BaseTest, HasFoo) {
-  // This is an ordinary non-parameterized test.
-}
-
-TEST_P(DerivedTest, DoesBlah) {
-  // GetParam works just the same here as if you inherit from TestWithParam.
-  EXPECT_TRUE(foo.Blah(GetParam()));
-}
-
-#endif  // 0
-
-
-#if !GTEST_OS_SYMBIAN
-# include <utility>
-#endif
-
-// scripts/fuse_gtest.py depends on gtest's own header being #included
-// *unconditionally*.  Therefore these #includes cannot be moved
-// inside #if GTEST_HAS_PARAM_TEST.
-// Copyright 2008 Google Inc.
-// All Rights Reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: vladl@google.com (Vlad Losev)
-
-// Type and function utilities for implementing parameterized tests.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PARAM_UTIL_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PARAM_UTIL_H_
-
-#include <ctype.h>
-
-#include <iterator>
-#include <set>
-#include <utility>
-#include <vector>
-
-// scripts/fuse_gtest.py depends on gtest's own header being #included
-// *unconditionally*.  Therefore these #includes cannot be moved
-// inside #if GTEST_HAS_PARAM_TEST.
-// Copyright 2003 Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Authors: Dan Egnor (egnor@google.com)
-//
-// A "smart" pointer type with reference tracking.  Every pointer to a
-// particular object is kept on a circular linked list.  When the last pointer
-// to an object is destroyed or reassigned, the object is deleted.
-//
-// Used properly, this deletes the object when the last reference goes away.
-// There are several caveats:
-// - Like all reference counting schemes, cycles lead to leaks.
-// - Each smart pointer is actually two pointers (8 bytes instead of 4).
-// - Every time a pointer is assigned, the entire list of pointers to that
-//   object is traversed.  This class is therefore NOT SUITABLE when there
-//   will often be more than two or three pointers to a particular object.
-// - References are only tracked as long as linked_ptr<> objects are copied.
-//   If a linked_ptr<> is converted to a raw pointer and back, BAD THINGS
-//   will happen (double deletion).
-//
-// A good use of this class is storing object references in STL containers.
-// You can safely put linked_ptr<> in a vector<>.
-// Other uses may not be as good.
-//
-// Note: If you use an incomplete type with linked_ptr<>, the class
-// *containing* linked_ptr<> must have a constructor and destructor (even
-// if they do nothing!).
-//
-// Bill Gibbons suggested we use something like this.
-//
-// Thread Safety:
-//   Unlike other linked_ptr implementations, in this implementation
-//   a linked_ptr object is thread-safe in the sense that:
-//     - it's safe to copy linked_ptr objects concurrently,
-//     - it's safe to copy *from* a linked_ptr and read its underlying
-//       raw pointer (e.g. via get()) concurrently, and
-//     - it's safe to write to two linked_ptrs that point to the same
-//       shared object concurrently.
-// TODO(wan@google.com): rename this to safe_linked_ptr to avoid
-// confusion with normal linked_ptr.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_LINKED_PTR_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_LINKED_PTR_H_
-
-#include <stdlib.h>
-#include <assert.h>
-
-
-namespace testing {
-namespace internal {
-
-// Protects copying of all linked_ptr objects.
-GTEST_API_ GTEST_DECLARE_STATIC_MUTEX_(g_linked_ptr_mutex);
-
-// This is used internally by all instances of linked_ptr<>.  It needs to be
-// a non-template class because different types of linked_ptr<> can refer to
-// the same object (linked_ptr<Superclass>(obj) vs linked_ptr<Subclass>(obj)).
-// So, it needs to be possible for different types of linked_ptr to participate
-// in the same circular linked list, so we need a single class type here.
-//
-// DO NOT USE THIS CLASS DIRECTLY YOURSELF.  Use linked_ptr<T>.
-class linked_ptr_internal {
- public:
-  // Create a new circle that includes only this instance.
-  void join_new() {
-    next_ = this;
-  }
-
-  // Many linked_ptr operations may change p.link_ for some linked_ptr
-  // variable p in the same circle as this object.  Therefore we need
-  // to prevent two such operations from occurring concurrently.
-  //
-  // Note that different types of linked_ptr objects can coexist in a
-  // circle (e.g. linked_ptr<Base>, linked_ptr<Derived1>, and
-  // linked_ptr<Derived2>).  Therefore we must use a single mutex to
-  // protect all linked_ptr objects.  This can create serious
-  // contention in production code, but is acceptable in a testing
-  // framework.
-
-  // Join an existing circle.
-  void join(linked_ptr_internal const* ptr)
-      GTEST_LOCK_EXCLUDED_(g_linked_ptr_mutex) {
-    MutexLock lock(&g_linked_ptr_mutex);
-
-    linked_ptr_internal const* p = ptr;
-    while (p->next_ != ptr) {
-      assert(p->next_ != this &&
-             "Trying to join() a linked ring we are already in. "
-             "Is GMock thread safety enabled?");
-      p = p->next_;
-    }
-    p->next_ = this;
-    next_ = ptr;
-  }
-
-  // Leave whatever circle we're part of.  Returns true if we were the
-  // last member of the circle.  Once this is done, you can join() another.
-  bool depart()
-      GTEST_LOCK_EXCLUDED_(g_linked_ptr_mutex) {
-    MutexLock lock(&g_linked_ptr_mutex);
-
-    if (next_ == this) return true;
-    linked_ptr_internal const* p = next_;
-    while (p->next_ != this) {
-      assert(p->next_ != next_ &&
-             "Trying to depart() a linked ring we are not in. "
-             "Is GMock thread safety enabled?");
-      p = p->next_;
-    }
-    p->next_ = next_;
-    return false;
-  }
-
- private:
-  mutable linked_ptr_internal const* next_;
-};
-
-template <typename T>
-class linked_ptr {
- public:
-  typedef T element_type;
-
-  // Take over ownership of a raw pointer.  This should happen as soon as
-  // possible after the object is created.
-  explicit linked_ptr(T* ptr = NULL) { capture(ptr); }
-  ~linked_ptr() { depart(); }
-
-  // Copy an existing linked_ptr<>, adding ourselves to the list of references.
-  template <typename U> linked_ptr(linked_ptr<U> const& ptr) { copy(&ptr); }
-  linked_ptr(linked_ptr const& ptr) {  // NOLINT
-    assert(&ptr != this);
-    copy(&ptr);
-  }
-
-  // Assignment releases the old value and acquires the new.
-  template <typename U> linked_ptr& operator=(linked_ptr<U> const& ptr) {
-    depart();
-    copy(&ptr);
-    return *this;
-  }
-
-  linked_ptr& operator=(linked_ptr const& ptr) {
-    if (&ptr != this) {
-      depart();
-      copy(&ptr);
-    }
-    return *this;
-  }
-
-  // Smart pointer members.
-  void reset(T* ptr = NULL) {
-    depart();
-    capture(ptr);
-  }
-  T* get() const { return value_; }
-  T* operator->() const { return value_; }
-  T& operator*() const { return *value_; }
-
-  bool operator==(T* p) const { return value_ == p; }
-  bool operator!=(T* p) const { return value_ != p; }
-  template <typename U>
-  bool operator==(linked_ptr<U> const& ptr) const {
-    return value_ == ptr.get();
-  }
-  template <typename U>
-  bool operator!=(linked_ptr<U> const& ptr) const {
-    return value_ != ptr.get();
-  }
-
- private:
-  template <typename U>
-  friend class linked_ptr;
-
-  T* value_;
-  linked_ptr_internal link_;
-
-  void depart() {
-    if (link_.depart()) delete value_;
-  }
-
-  void capture(T* ptr) {
-    value_ = ptr;
-    link_.join_new();
-  }
-
-  template <typename U> void copy(linked_ptr<U> const* ptr) {
-    value_ = ptr->get();
-    if (value_)
-      link_.join(&ptr->link_);
-    else
-      link_.join_new();
-  }
-};
-
-template<typename T> inline
-bool operator==(T* ptr, const linked_ptr<T>& x) {
-  return ptr == x.get();
-}
-
-template<typename T> inline
-bool operator!=(T* ptr, const linked_ptr<T>& x) {
-  return ptr != x.get();
-}
-
-// A function to convert T* into linked_ptr<T>
-// Doing e.g. make_linked_ptr(new FooBarBaz<type>(arg)) is a shorter notation
-// for linked_ptr<FooBarBaz<type> >(new FooBarBaz<type>(arg))
-template <typename T>
-linked_ptr<T> make_linked_ptr(T* ptr) {
-  return linked_ptr<T>(ptr);
-}
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_LINKED_PTR_H_
-// Copyright 2007, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-
-// Google Test - The Google C++ Testing Framework
-//
-// This file implements a universal value printer that can print a
-// value of any type T:
-//
-//   void ::testing::internal::UniversalPrinter<T>::Print(value, ostream_ptr);
-//
-// A user can teach this function how to print a class type T by
-// defining either operator<<() or PrintTo() in the namespace that
-// defines T.  More specifically, the FIRST defined function in the
-// following list will be used (assuming T is defined in namespace
-// foo):
-//
-//   1. foo::PrintTo(const T&, ostream*)
-//   2. operator<<(ostream&, const T&) defined in either foo or the
-//      global namespace.
-//
-// If none of the above is defined, it will print the debug string of
-// the value if it is a protocol buffer, or print the raw bytes in the
-// value otherwise.
-//
-// To aid debugging: when T is a reference type, the address of the
-// value is also printed; when T is a (const) char pointer, both the
-// pointer value and the NUL-terminated string it points to are
-// printed.
-//
-// We also provide some convenient wrappers:
-//
-//   // Prints a value to a string.  For a (const or not) char
-//   // pointer, the NUL-terminated string (but not the pointer) is
-//   // printed.
-//   std::string ::testing::PrintToString(const T& value);
-//
-//   // Prints a value tersely: for a reference type, the referenced
-//   // value (but not the address) is printed; for a (const or not) char
-//   // pointer, the NUL-terminated string (but not the pointer) is
-//   // printed.
-//   void ::testing::internal::UniversalTersePrint(const T& value, ostream*);
-//
-//   // Prints value using the type inferred by the compiler.  The difference
-//   // from UniversalTersePrint() is that this function prints both the
-//   // pointer and the NUL-terminated string for a (const or not) char pointer.
-//   void ::testing::internal::UniversalPrint(const T& value, ostream*);
-//
-//   // Prints the fields of a tuple tersely to a string vector, one
-//   // element for each field. Tuple support must be enabled in
-//   // gtest-port.h.
-//   std::vector<string> UniversalTersePrintTupleFieldsToStrings(
-//       const Tuple& value);
-//
-// Known limitation:
-//
-// The print primitives print the elements of an STL-style container
-// using the compiler-inferred type of *iter where iter is a
-// const_iterator of the container.  When const_iterator is an input
-// iterator but not a forward iterator, this inferred type may not
-// match value_type, and the print output may be incorrect.  In
-// practice, this is rarely a problem as for most containers
-// const_iterator is a forward iterator.  We'll fix this if there's an
-// actual need for it.  Note that this fix cannot rely on value_type
-// being defined as many user-defined container types don't have
-// value_type.
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_PRINTERS_H_
-#define GTEST_INCLUDE_GTEST_GTEST_PRINTERS_H_
-
-#include <ostream>  // NOLINT
-#include <sstream>
-#include <string>
-#include <utility>
-#include <vector>
-
-#if GTEST_HAS_STD_TUPLE_
-# include <tuple>
-#endif
-
-namespace testing {
-
-// Definitions in the 'internal' and 'internal2' name spaces are
-// subject to change without notice.  DO NOT USE THEM IN USER CODE!
-namespace internal2 {
-
-// Prints the given number of bytes in the given object to the given
-// ostream.
-GTEST_API_ void PrintBytesInObjectTo(const unsigned char* obj_bytes,
-                                     size_t count,
-                                     ::std::ostream* os);
-
-// For selecting which printer to use when a given type has neither <<
-// nor PrintTo().
-enum TypeKind {
-  kProtobuf,              // a protobuf type
-  kConvertibleToInteger,  // a type implicitly convertible to BiggestInt
-                          // (e.g. a named or unnamed enum type)
-  kOtherType              // anything else
-};
-
-// TypeWithoutFormatter<T, kTypeKind>::PrintValue(value, os) is called
-// by the universal printer to print a value of type T when neither
-// operator<< nor PrintTo() is defined for T, where kTypeKind is the
-// "kind" of T as defined by enum TypeKind.
-template <typename T, TypeKind kTypeKind>
-class TypeWithoutFormatter {
- public:
-  // This default version is called when kTypeKind is kOtherType.
-  static void PrintValue(const T& value, ::std::ostream* os) {
-    PrintBytesInObjectTo(reinterpret_cast<const unsigned char*>(&value),
-                         sizeof(value), os);
-  }
-};
-
-// We print a protobuf using its ShortDebugString() when the string
-// doesn't exceed this many characters; otherwise we print it using
-// DebugString() for better readability.
-const size_t kProtobufOneLinerMaxLength = 50;
-
-template <typename T>
-class TypeWithoutFormatter<T, kProtobuf> {
- public:
-  static void PrintValue(const T& value, ::std::ostream* os) {
-    const ::testing::internal::string short_str = value.ShortDebugString();
-    const ::testing::internal::string pretty_str =
-        short_str.length() <= kProtobufOneLinerMaxLength ?
-        short_str : ("\n" + value.DebugString());
-    *os << ("<" + pretty_str + ">");
-  }
-};
-
-template <typename T>
-class TypeWithoutFormatter<T, kConvertibleToInteger> {
- public:
-  // Since T has no << operator or PrintTo() but can be implicitly
-  // converted to BiggestInt, we print it as a BiggestInt.
-  //
-  // Most likely T is an enum type (either named or unnamed), in which
-  // case printing it as an integer is the desired behavior.  In case
-  // T is not an enum, printing it as an integer is the best we can do
-  // given that it has no user-defined printer.
-  static void PrintValue(const T& value, ::std::ostream* os) {
-    const internal::BiggestInt kBigInt = value;
-    *os << kBigInt;
-  }
-};
-
-// Prints the given value to the given ostream.  If the value is a
-// protocol message, its debug string is printed; if it's an enum or
-// of a type implicitly convertible to BiggestInt, it's printed as an
-// integer; otherwise the bytes in the value are printed.  This is
-// what UniversalPrinter<T>::Print() does when it knows nothing about
-// type T and T has neither << operator nor PrintTo().
-//
-// A user can override this behavior for a class type Foo by defining
-// a << operator in the namespace where Foo is defined.
-//
-// We put this operator in namespace 'internal2' instead of 'internal'
-// to simplify the implementation, as much code in 'internal' needs to
-// use << in STL, which would conflict with our own << were it defined
-// in 'internal'.
-//
-// Note that this operator<< takes a generic std::basic_ostream<Char,
-// CharTraits> type instead of the more restricted std::ostream.  If
-// we define it to take an std::ostream instead, we'll get an
-// "ambiguous overloads" compiler error when trying to print a type
-// Foo that supports streaming to std::basic_ostream<Char,
-// CharTraits>, as the compiler cannot tell whether
-// operator<<(std::ostream&, const T&) or
-// operator<<(std::basic_stream<Char, CharTraits>, const Foo&) is more
-// specific.
-template <typename Char, typename CharTraits, typename T>
-::std::basic_ostream<Char, CharTraits>& operator<<(
-    ::std::basic_ostream<Char, CharTraits>& os, const T& x) {
-  TypeWithoutFormatter<T,
-      (internal::IsAProtocolMessage<T>::value ? kProtobuf :
-       internal::ImplicitlyConvertible<const T&, internal::BiggestInt>::value ?
-       kConvertibleToInteger : kOtherType)>::PrintValue(x, &os);
-  return os;
-}
-
-}  // namespace internal2
-}  // namespace testing
-
-// This namespace MUST NOT BE NESTED IN ::testing, or the name look-up
-// magic needed for implementing UniversalPrinter won't work.
-namespace testing_internal {
-
-// Used to print a value that is not an STL-style container when the
-// user doesn't define PrintTo() for it.
-template <typename T>
-void DefaultPrintNonContainerTo(const T& value, ::std::ostream* os) {
-  // With the following statement, during unqualified name lookup,
-  // testing::internal2::operator<< appears as if it was declared in
-  // the nearest enclosing namespace that contains both
-  // ::testing_internal and ::testing::internal2, i.e. the global
-  // namespace.  For more details, refer to the C++ Standard section
-  // 7.3.4-1 [namespace.udir].  This allows us to fall back onto
-  // testing::internal2::operator<< in case T doesn't come with a <<
-  // operator.
-  //
-  // We cannot write 'using ::testing::internal2::operator<<;', which
-  // gcc 3.3 fails to compile due to a compiler bug.
-  using namespace ::testing::internal2;  // NOLINT
-
-  // Assuming T is defined in namespace foo, in the next statement,
-  // the compiler will consider all of:
-  //
-  //   1. foo::operator<< (thanks to Koenig look-up),
-  //   2. ::operator<< (as the current namespace is enclosed in ::),
-  //   3. testing::internal2::operator<< (thanks to the using statement above).
-  //
-  // The operator<< whose type matches T best will be picked.
-  //
-  // We deliberately allow #2 to be a candidate, as sometimes it's
-  // impossible to define #1 (e.g. when foo is ::std, defining
-  // anything in it is undefined behavior unless you are a compiler
-  // vendor.).
-  *os << value;
-}
-
-}  // namespace testing_internal
-
-namespace testing {
-namespace internal {
-
-// FormatForComparison<ToPrint, OtherOperand>::Format(value) formats a
-// value of type ToPrint that is an operand of a comparison assertion
-// (e.g. ASSERT_EQ).  OtherOperand is the type of the other operand in
-// the comparison, and is used to help determine the best way to
-// format the value.  In particular, when the value is a C string
-// (char pointer) and the other operand is an STL string object, we
-// want to format the C string as a string, since we know it is
-// compared by value with the string object.  If the value is a char
-// pointer but the other operand is not an STL string object, we don't
-// know whether the pointer is supposed to point to a NUL-terminated
-// string, and thus want to print it as a pointer to be safe.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-
-// The default case.
-template <typename ToPrint, typename OtherOperand>
-class FormatForComparison {
- public:
-  static ::std::string Format(const ToPrint& value) {
-    return ::testing::PrintToString(value);
-  }
-};
-
-// Array.
-template <typename ToPrint, size_t N, typename OtherOperand>
-class FormatForComparison<ToPrint[N], OtherOperand> {
- public:
-  static ::std::string Format(const ToPrint* value) {
-    return FormatForComparison<const ToPrint*, OtherOperand>::Format(value);
-  }
-};
-
-// By default, print C string as pointers to be safe, as we don't know
-// whether they actually point to a NUL-terminated string.
-
-#define GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(CharType)                \
-  template <typename OtherOperand>                                      \
-  class FormatForComparison<CharType*, OtherOperand> {                  \
-   public:                                                              \
-    static ::std::string Format(CharType* value) {                      \
-      return ::testing::PrintToString(static_cast<const void*>(value)); \
-    }                                                                   \
-  }
-
-GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(char);
-GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(const char);
-GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(wchar_t);
-GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(const wchar_t);
-
-#undef GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_
-
-// If a C string is compared with an STL string object, we know it's meant
-// to point to a NUL-terminated string, and thus can print it as a string.
-
-#define GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(CharType, OtherStringType) \
-  template <>                                                           \
-  class FormatForComparison<CharType*, OtherStringType> {               \
-   public:                                                              \
-    static ::std::string Format(CharType* value) {                      \
-      return ::testing::PrintToString(value);                           \
-    }                                                                   \
-  }
-
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(char, ::std::string);
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const char, ::std::string);
-
-#if GTEST_HAS_GLOBAL_STRING
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(char, ::string);
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const char, ::string);
-#endif
-
-#if GTEST_HAS_GLOBAL_WSTRING
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(wchar_t, ::wstring);
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const wchar_t, ::wstring);
-#endif
-
-#if GTEST_HAS_STD_WSTRING
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(wchar_t, ::std::wstring);
-GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const wchar_t, ::std::wstring);
-#endif
-
-#undef GTEST_IMPL_FORMAT_C_STRING_AS_STRING_
-
-// Formats a comparison assertion (e.g. ASSERT_EQ, EXPECT_LT, and etc)
-// operand to be used in a failure message.  The type (but not value)
-// of the other operand may affect the format.  This allows us to
-// print a char* as a raw pointer when it is compared against another
-// char* or void*, and print it as a C string when it is compared
-// against an std::string object, for example.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-template <typename T1, typename T2>
-std::string FormatForComparisonFailureMessage(
-    const T1& value, const T2& /* other_operand */) {
-  return FormatForComparison<T1, T2>::Format(value);
-}
-
-// UniversalPrinter<T>::Print(value, ostream_ptr) prints the given
-// value to the given ostream.  The caller must ensure that
-// 'ostream_ptr' is not NULL, or the behavior is undefined.
-//
-// We define UniversalPrinter as a class template (as opposed to a
-// function template), as we need to partially specialize it for
-// reference types, which cannot be done with function templates.
-template <typename T>
-class UniversalPrinter;
-
-template <typename T>
-void UniversalPrint(const T& value, ::std::ostream* os);
-
-// Used to print an STL-style container when the user doesn't define
-// a PrintTo() for it.
-template <typename C>
-void DefaultPrintTo(IsContainer /* dummy */,
-                    false_type /* is not a pointer */,
-                    const C& container, ::std::ostream* os) {
-  const size_t kMaxCount = 32;  // The maximum number of elements to print.
-  *os << '{';
-  size_t count = 0;
-  for (typename C::const_iterator it = container.begin();
-       it != container.end(); ++it, ++count) {
-    if (count > 0) {
-      *os << ',';
-      if (count == kMaxCount) {  // Enough has been printed.
-        *os << " ...";
-        break;
-      }
-    }
-    *os << ' ';
-    // We cannot call PrintTo(*it, os) here as PrintTo() doesn't
-    // handle *it being a native array.
-    internal::UniversalPrint(*it, os);
-  }
-
-  if (count > 0) {
-    *os << ' ';
-  }
-  *os << '}';
-}
-
-// Used to print a pointer that is neither a char pointer nor a member
-// pointer, when the user doesn't define PrintTo() for it.  (A member
-// variable pointer or member function pointer doesn't really point to
-// a location in the address space.  Their representation is
-// implementation-defined.  Therefore they will be printed as raw
-// bytes.)
-template <typename T>
-void DefaultPrintTo(IsNotContainer /* dummy */,
-                    true_type /* is a pointer */,
-                    T* p, ::std::ostream* os) {
-  if (p == NULL) {
-    *os << "NULL";
-  } else {
-    // C++ doesn't allow casting from a function pointer to any object
-    // pointer.
-    //
-    // IsTrue() silences warnings: "Condition is always true",
-    // "unreachable code".
-    if (IsTrue(ImplicitlyConvertible<T*, const void*>::value)) {
-      // T is not a function type.  We just call << to print p,
-      // relying on ADL to pick up user-defined << for their pointer
-      // types, if any.
-      *os << p;
-    } else {
-      // T is a function type, so '*os << p' doesn't do what we want
-      // (it just prints p as bool).  We want to print p as a const
-      // void*.  However, we cannot cast it to const void* directly,
-      // even using reinterpret_cast, as earlier versions of gcc
-      // (e.g. 3.4.5) cannot compile the cast when p is a function
-      // pointer.  Casting to UInt64 first solves the problem.
-      *os << reinterpret_cast<const void*>(
-          reinterpret_cast<internal::UInt64>(p));
-    }
-  }
-}
-
-// Used to print a non-container, non-pointer value when the user
-// doesn't define PrintTo() for it.
-template <typename T>
-void DefaultPrintTo(IsNotContainer /* dummy */,
-                    false_type /* is not a pointer */,
-                    const T& value, ::std::ostream* os) {
-  ::testing_internal::DefaultPrintNonContainerTo(value, os);
-}
-
-// Prints the given value using the << operator if it has one;
-// otherwise prints the bytes in it.  This is what
-// UniversalPrinter<T>::Print() does when PrintTo() is not specialized
-// or overloaded for type T.
-//
-// A user can override this behavior for a class type Foo by defining
-// an overload of PrintTo() in the namespace where Foo is defined.  We
-// give the user this option as sometimes defining a << operator for
-// Foo is not desirable (e.g. the coding style may prevent doing it,
-// or there is already a << operator but it doesn't do what the user
-// wants).
-template <typename T>
-void PrintTo(const T& value, ::std::ostream* os) {
-  // DefaultPrintTo() is overloaded.  The type of its first two
-  // arguments determine which version will be picked.  If T is an
-  // STL-style container, the version for container will be called; if
-  // T is a pointer, the pointer version will be called; otherwise the
-  // generic version will be called.
-  //
-  // Note that we check for container types here, prior to we check
-  // for protocol message types in our operator<<.  The rationale is:
-  //
-  // For protocol messages, we want to give people a chance to
-  // override Google Mock's format by defining a PrintTo() or
-  // operator<<.  For STL containers, other formats can be
-  // incompatible with Google Mock's format for the container
-  // elements; therefore we check for container types here to ensure
-  // that our format is used.
-  //
-  // The second argument of DefaultPrintTo() is needed to bypass a bug
-  // in Symbian's C++ compiler that prevents it from picking the right
-  // overload between:
-  //
-  //   PrintTo(const T& x, ...);
-  //   PrintTo(T* x, ...);
-  DefaultPrintTo(IsContainerTest<T>(0), is_pointer<T>(), value, os);
-}
-
-// The following list of PrintTo() overloads tells
-// UniversalPrinter<T>::Print() how to print standard types (built-in
-// types, strings, plain arrays, and pointers).
-
-// Overloads for various char types.
-GTEST_API_ void PrintTo(unsigned char c, ::std::ostream* os);
-GTEST_API_ void PrintTo(signed char c, ::std::ostream* os);
-inline void PrintTo(char c, ::std::ostream* os) {
-  // When printing a plain char, we always treat it as unsigned.  This
-  // way, the output won't be affected by whether the compiler thinks
-  // char is signed or not.
-  PrintTo(static_cast<unsigned char>(c), os);
-}
-
-// Overloads for other simple built-in types.
-inline void PrintTo(bool x, ::std::ostream* os) {
-  *os << (x ? "true" : "false");
-}
-
-// Overload for wchar_t type.
-// Prints a wchar_t as a symbol if it is printable or as its internal
-// code otherwise and also as its decimal code (except for L'\0').
-// The L'\0' char is printed as "L'\\0'". The decimal code is printed
-// as signed integer when wchar_t is implemented by the compiler
-// as a signed type and is printed as an unsigned integer when wchar_t
-// is implemented as an unsigned type.
-GTEST_API_ void PrintTo(wchar_t wc, ::std::ostream* os);
-
-// Overloads for C strings.
-GTEST_API_ void PrintTo(const char* s, ::std::ostream* os);
-inline void PrintTo(char* s, ::std::ostream* os) {
-  PrintTo(ImplicitCast_<const char*>(s), os);
-}
-
-// signed/unsigned char is often used for representing binary data, so
-// we print pointers to it as void* to be safe.
-inline void PrintTo(const signed char* s, ::std::ostream* os) {
-  PrintTo(ImplicitCast_<const void*>(s), os);
-}
-inline void PrintTo(signed char* s, ::std::ostream* os) {
-  PrintTo(ImplicitCast_<const void*>(s), os);
-}
-inline void PrintTo(const unsigned char* s, ::std::ostream* os) {
-  PrintTo(ImplicitCast_<const void*>(s), os);
-}
-inline void PrintTo(unsigned char* s, ::std::ostream* os) {
-  PrintTo(ImplicitCast_<const void*>(s), os);
-}
-
-// MSVC can be configured to define wchar_t as a typedef of unsigned
-// short.  It defines _NATIVE_WCHAR_T_DEFINED when wchar_t is a native
-// type.  When wchar_t is a typedef, defining an overload for const
-// wchar_t* would cause unsigned short* be printed as a wide string,
-// possibly causing invalid memory accesses.
-#if !defined(_MSC_VER) || defined(_NATIVE_WCHAR_T_DEFINED)
-// Overloads for wide C strings
-GTEST_API_ void PrintTo(const wchar_t* s, ::std::ostream* os);
-inline void PrintTo(wchar_t* s, ::std::ostream* os) {
-  PrintTo(ImplicitCast_<const wchar_t*>(s), os);
-}
-#endif
-
-// Overload for C arrays.  Multi-dimensional arrays are printed
-// properly.
-
-// Prints the given number of elements in an array, without printing
-// the curly braces.
-template <typename T>
-void PrintRawArrayTo(const T a[], size_t count, ::std::ostream* os) {
-  UniversalPrint(a[0], os);
-  for (size_t i = 1; i != count; i++) {
-    *os << ", ";
-    UniversalPrint(a[i], os);
-  }
-}
-
-// Overloads for ::string and ::std::string.
-#if GTEST_HAS_GLOBAL_STRING
-GTEST_API_ void PrintStringTo(const ::string&s, ::std::ostream* os);
-inline void PrintTo(const ::string& s, ::std::ostream* os) {
-  PrintStringTo(s, os);
-}
-#endif  // GTEST_HAS_GLOBAL_STRING
-
-GTEST_API_ void PrintStringTo(const ::std::string&s, ::std::ostream* os);
-inline void PrintTo(const ::std::string& s, ::std::ostream* os) {
-  PrintStringTo(s, os);
-}
-
-// Overloads for ::wstring and ::std::wstring.
-#if GTEST_HAS_GLOBAL_WSTRING
-GTEST_API_ void PrintWideStringTo(const ::wstring&s, ::std::ostream* os);
-inline void PrintTo(const ::wstring& s, ::std::ostream* os) {
-  PrintWideStringTo(s, os);
-}
-#endif  // GTEST_HAS_GLOBAL_WSTRING
-
-#if GTEST_HAS_STD_WSTRING
-GTEST_API_ void PrintWideStringTo(const ::std::wstring&s, ::std::ostream* os);
-inline void PrintTo(const ::std::wstring& s, ::std::ostream* os) {
-  PrintWideStringTo(s, os);
-}
-#endif  // GTEST_HAS_STD_WSTRING
-
-#if GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
-// Helper function for printing a tuple.  T must be instantiated with
-// a tuple type.
-template <typename T>
-void PrintTupleTo(const T& t, ::std::ostream* os);
-#endif  // GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
-
-#if GTEST_HAS_TR1_TUPLE
-// Overload for ::std::tr1::tuple.  Needed for printing function arguments,
-// which are packed as tuples.
-
-// Overloaded PrintTo() for tuples of various arities.  We support
-// tuples of up-to 10 fields.  The following implementation works
-// regardless of whether tr1::tuple is implemented using the
-// non-standard variadic template feature or not.
-
-inline void PrintTo(const ::std::tr1::tuple<>& t, ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1>
-void PrintTo(const ::std::tr1::tuple<T1>& t, ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2>
-void PrintTo(const ::std::tr1::tuple<T1, T2>& t, ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3>
-void PrintTo(const ::std::tr1::tuple<T1, T2, T3>& t, ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3, typename T4>
-void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4>& t, ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5>
-void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5>& t,
-             ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-          typename T6>
-void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6>& t,
-             ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-          typename T6, typename T7>
-void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7>& t,
-             ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-          typename T6, typename T7, typename T8>
-void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7, T8>& t,
-             ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-          typename T6, typename T7, typename T8, typename T9>
-void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9>& t,
-             ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-          typename T6, typename T7, typename T8, typename T9, typename T10>
-void PrintTo(
-    const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10>& t,
-    ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-#endif  // GTEST_HAS_TR1_TUPLE
-
-#if GTEST_HAS_STD_TUPLE_
-template <typename... Types>
-void PrintTo(const ::std::tuple<Types...>& t, ::std::ostream* os) {
-  PrintTupleTo(t, os);
-}
-#endif  // GTEST_HAS_STD_TUPLE_
-
-// Overload for std::pair.
-template <typename T1, typename T2>
-void PrintTo(const ::std::pair<T1, T2>& value, ::std::ostream* os) {
-  *os << '(';
-  // We cannot use UniversalPrint(value.first, os) here, as T1 may be
-  // a reference type.  The same for printing value.second.
-  UniversalPrinter<T1>::Print(value.first, os);
-  *os << ", ";
-  UniversalPrinter<T2>::Print(value.second, os);
-  *os << ')';
-}
-
-// Implements printing a non-reference type T by letting the compiler
-// pick the right overload of PrintTo() for T.
-template <typename T>
-class UniversalPrinter {
- public:
-  // MSVC warns about adding const to a function type, so we want to
-  // disable the warning.
-  GTEST_DISABLE_MSC_WARNINGS_PUSH_(4180)
-
-  // Note: we deliberately don't call this PrintTo(), as that name
-  // conflicts with ::testing::internal::PrintTo in the body of the
-  // function.
-  static void Print(const T& value, ::std::ostream* os) {
-    // By default, ::testing::internal::PrintTo() is used for printing
-    // the value.
-    //
-    // Thanks to Koenig look-up, if T is a class and has its own
-    // PrintTo() function defined in its namespace, that function will
-    // be visible here.  Since it is more specific than the generic ones
-    // in ::testing::internal, it will be picked by the compiler in the
-    // following statement - exactly what we want.
-    PrintTo(value, os);
-  }
-
-  GTEST_DISABLE_MSC_WARNINGS_POP_()
-};
-
-// UniversalPrintArray(begin, len, os) prints an array of 'len'
-// elements, starting at address 'begin'.
-template <typename T>
-void UniversalPrintArray(const T* begin, size_t len, ::std::ostream* os) {
-  if (len == 0) {
-    *os << "{}";
-  } else {
-    *os << "{ ";
-    const size_t kThreshold = 18;
-    const size_t kChunkSize = 8;
-    // If the array has more than kThreshold elements, we'll have to
-    // omit some details by printing only the first and the last
-    // kChunkSize elements.
-    // TODO(wan@google.com): let the user control the threshold using a flag.
-    if (len <= kThreshold) {
-      PrintRawArrayTo(begin, len, os);
-    } else {
-      PrintRawArrayTo(begin, kChunkSize, os);
-      *os << ", ..., ";
-      PrintRawArrayTo(begin + len - kChunkSize, kChunkSize, os);
-    }
-    *os << " }";
-  }
-}
-// This overload prints a (const) char array compactly.
-GTEST_API_ void UniversalPrintArray(
-    const char* begin, size_t len, ::std::ostream* os);
-
-// This overload prints a (const) wchar_t array compactly.
-GTEST_API_ void UniversalPrintArray(
-    const wchar_t* begin, size_t len, ::std::ostream* os);
-
-// Implements printing an array type T[N].
-template <typename T, size_t N>
-class UniversalPrinter<T[N]> {
- public:
-  // Prints the given array, omitting some elements when there are too
-  // many.
-  static void Print(const T (&a)[N], ::std::ostream* os) {
-    UniversalPrintArray(a, N, os);
-  }
-};
-
-// Implements printing a reference type T&.
-template <typename T>
-class UniversalPrinter<T&> {
- public:
-  // MSVC warns about adding const to a function type, so we want to
-  // disable the warning.
-  GTEST_DISABLE_MSC_WARNINGS_PUSH_(4180)
-
-  static void Print(const T& value, ::std::ostream* os) {
-    // Prints the address of the value.  We use reinterpret_cast here
-    // as static_cast doesn't compile when T is a function type.
-    *os << "@" << reinterpret_cast<const void*>(&value) << " ";
-
-    // Then prints the value itself.
-    UniversalPrint(value, os);
-  }
-
-  GTEST_DISABLE_MSC_WARNINGS_POP_()
-};
-
-// Prints a value tersely: for a reference type, the referenced value
-// (but not the address) is printed; for a (const) char pointer, the
-// NUL-terminated string (but not the pointer) is printed.
-
-template <typename T>
-class UniversalTersePrinter {
- public:
-  static void Print(const T& value, ::std::ostream* os) {
-    UniversalPrint(value, os);
-  }
-};
-template <typename T>
-class UniversalTersePrinter<T&> {
- public:
-  static void Print(const T& value, ::std::ostream* os) {
-    UniversalPrint(value, os);
-  }
-};
-template <typename T, size_t N>
-class UniversalTersePrinter<T[N]> {
- public:
-  static void Print(const T (&value)[N], ::std::ostream* os) {
-    UniversalPrinter<T[N]>::Print(value, os);
-  }
-};
-template <>
-class UniversalTersePrinter<const char*> {
- public:
-  static void Print(const char* str, ::std::ostream* os) {
-    if (str == NULL) {
-      *os << "NULL";
-    } else {
-      UniversalPrint(string(str), os);
-    }
-  }
-};
-template <>
-class UniversalTersePrinter<char*> {
- public:
-  static void Print(char* str, ::std::ostream* os) {
-    UniversalTersePrinter<const char*>::Print(str, os);
-  }
-};
-
-#if GTEST_HAS_STD_WSTRING
-template <>
-class UniversalTersePrinter<const wchar_t*> {
- public:
-  static void Print(const wchar_t* str, ::std::ostream* os) {
-    if (str == NULL) {
-      *os << "NULL";
-    } else {
-      UniversalPrint(::std::wstring(str), os);
-    }
-  }
-};
-#endif
-
-template <>
-class UniversalTersePrinter<wchar_t*> {
- public:
-  static void Print(wchar_t* str, ::std::ostream* os) {
-    UniversalTersePrinter<const wchar_t*>::Print(str, os);
-  }
-};
-
-template <typename T>
-void UniversalTersePrint(const T& value, ::std::ostream* os) {
-  UniversalTersePrinter<T>::Print(value, os);
-}
-
-// Prints a value using the type inferred by the compiler.  The
-// difference between this and UniversalTersePrint() is that for a
-// (const) char pointer, this prints both the pointer and the
-// NUL-terminated string.
-template <typename T>
-void UniversalPrint(const T& value, ::std::ostream* os) {
-  // A workarond for the bug in VC++ 7.1 that prevents us from instantiating
-  // UniversalPrinter with T directly.
-  typedef T T1;
-  UniversalPrinter<T1>::Print(value, os);
-}
-
-typedef ::std::vector<string> Strings;
-
-// TuplePolicy<TupleT> must provide:
-// - tuple_size
-//     size of tuple TupleT.
-// - get<size_t I>(const TupleT& t)
-//     static function extracting element I of tuple TupleT.
-// - tuple_element<size_t I>::type
-//     type of element I of tuple TupleT.
-template <typename TupleT>
-struct TuplePolicy;
-
-#if GTEST_HAS_TR1_TUPLE
-template <typename TupleT>
-struct TuplePolicy {
-  typedef TupleT Tuple;
-  static const size_t tuple_size = ::std::tr1::tuple_size<Tuple>::value;
-
-  template <size_t I>
-  struct tuple_element : ::std::tr1::tuple_element<I, Tuple> {};
-
-  template <size_t I>
-  static typename AddReference<
-      const typename ::std::tr1::tuple_element<I, Tuple>::type>::type get(
-      const Tuple& tuple) {
-    return ::std::tr1::get<I>(tuple);
-  }
-};
-template <typename TupleT>
-const size_t TuplePolicy<TupleT>::tuple_size;
-#endif  // GTEST_HAS_TR1_TUPLE
-
-#if GTEST_HAS_STD_TUPLE_
-template <typename... Types>
-struct TuplePolicy< ::std::tuple<Types...> > {
-  typedef ::std::tuple<Types...> Tuple;
-  static const size_t tuple_size = ::std::tuple_size<Tuple>::value;
-
-  template <size_t I>
-  struct tuple_element : ::std::tuple_element<I, Tuple> {};
-
-  template <size_t I>
-  static const typename ::std::tuple_element<I, Tuple>::type& get(
-      const Tuple& tuple) {
-    return ::std::get<I>(tuple);
-  }
-};
-template <typename... Types>
-const size_t TuplePolicy< ::std::tuple<Types...> >::tuple_size;
-#endif  // GTEST_HAS_STD_TUPLE_
-
-#if GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
-// This helper template allows PrintTo() for tuples and
-// UniversalTersePrintTupleFieldsToStrings() to be defined by
-// induction on the number of tuple fields.  The idea is that
-// TuplePrefixPrinter<N>::PrintPrefixTo(t, os) prints the first N
-// fields in tuple t, and can be defined in terms of
-// TuplePrefixPrinter<N - 1>.
-//
-// The inductive case.
-template <size_t N>
-struct TuplePrefixPrinter {
-  // Prints the first N fields of a tuple.
-  template <typename Tuple>
-  static void PrintPrefixTo(const Tuple& t, ::std::ostream* os) {
-    TuplePrefixPrinter<N - 1>::PrintPrefixTo(t, os);
-    GTEST_INTENTIONAL_CONST_COND_PUSH_()
-    if (N > 1) {
-    GTEST_INTENTIONAL_CONST_COND_POP_()
-      *os << ", ";
-    }
-    UniversalPrinter<
-        typename TuplePolicy<Tuple>::template tuple_element<N - 1>::type>
-        ::Print(TuplePolicy<Tuple>::template get<N - 1>(t), os);
-  }
-
-  // Tersely prints the first N fields of a tuple to a string vector,
-  // one element for each field.
-  template <typename Tuple>
-  static void TersePrintPrefixToStrings(const Tuple& t, Strings* strings) {
-    TuplePrefixPrinter<N - 1>::TersePrintPrefixToStrings(t, strings);
-    ::std::stringstream ss;
-    UniversalTersePrint(TuplePolicy<Tuple>::template get<N - 1>(t), &ss);
-    strings->push_back(ss.str());
-  }
-};
-
-// Base case.
-template <>
-struct TuplePrefixPrinter<0> {
-  template <typename Tuple>
-  static void PrintPrefixTo(const Tuple&, ::std::ostream*) {}
-
-  template <typename Tuple>
-  static void TersePrintPrefixToStrings(const Tuple&, Strings*) {}
-};
-
-// Helper function for printing a tuple.
-// Tuple must be either std::tr1::tuple or std::tuple type.
-template <typename Tuple>
-void PrintTupleTo(const Tuple& t, ::std::ostream* os) {
-  *os << "(";
-  TuplePrefixPrinter<TuplePolicy<Tuple>::tuple_size>::PrintPrefixTo(t, os);
-  *os << ")";
-}
-
-// Prints the fields of a tuple tersely to a string vector, one
-// element for each field.  See the comment before
-// UniversalTersePrint() for how we define "tersely".
-template <typename Tuple>
-Strings UniversalTersePrintTupleFieldsToStrings(const Tuple& value) {
-  Strings result;
-  TuplePrefixPrinter<TuplePolicy<Tuple>::tuple_size>::
-      TersePrintPrefixToStrings(value, &result);
-  return result;
-}
-#endif  // GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
-
-}  // namespace internal
-
-template <typename T>
-::std::string PrintToString(const T& value) {
-  ::std::stringstream ss;
-  internal::UniversalTersePrinter<T>::Print(value, &ss);
-  return ss.str();
-}
-
-}  // namespace testing
-
-// Include any custom printer added by the local installation.
-// We must include this header at the end to make sure it can use the
-// declarations from this file.
-// Copyright 2015, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// This file provides an injection point for custom printers in a local
-// installation of gTest.
-// It will be included from gtest-printers.h and the overrides in this file
-// will be visible to everyone.
-// See documentation at gtest/gtest-printers.h for details on how to define a
-// custom printer.
-//
-// ** Custom implementation starts here **
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_CUSTOM_GTEST_PRINTERS_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_CUSTOM_GTEST_PRINTERS_H_
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_CUSTOM_GTEST_PRINTERS_H_
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_PRINTERS_H_
-
-#if GTEST_HAS_PARAM_TEST
-
-namespace testing {
-
-// Input to a parameterized test name generator, describing a test parameter.
-// Consists of the parameter value and the integer parameter index.
-template <class ParamType>
-struct TestParamInfo {
-  TestParamInfo(const ParamType& a_param, size_t an_index) :
-    param(a_param),
-    index(an_index) {}
-  ParamType param;
-  size_t index;
-};
-
-// A builtin parameterized test name generator which returns the result of
-// testing::PrintToString.
-struct PrintToStringParamName {
-  template <class ParamType>
-  std::string operator()(const TestParamInfo<ParamType>& info) const {
-    return PrintToString(info.param);
-  }
-};
-
-namespace internal {
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Outputs a message explaining invalid registration of different
-// fixture class for the same test case. This may happen when
-// TEST_P macro is used to define two tests with the same name
-// but in different namespaces.
-GTEST_API_ void ReportInvalidTestCaseType(const char* test_case_name,
-                                          CodeLocation code_location);
-
-template <typename> class ParamGeneratorInterface;
-template <typename> class ParamGenerator;
-
-// Interface for iterating over elements provided by an implementation
-// of ParamGeneratorInterface<T>.
-template <typename T>
-class ParamIteratorInterface {
- public:
-  virtual ~ParamIteratorInterface() {}
-  // A pointer to the base generator instance.
-  // Used only for the purposes of iterator comparison
-  // to make sure that two iterators belong to the same generator.
-  virtual const ParamGeneratorInterface<T>* BaseGenerator() const = 0;
-  // Advances iterator to point to the next element
-  // provided by the generator. The caller is responsible
-  // for not calling Advance() on an iterator equal to
-  // BaseGenerator()->End().
-  virtual void Advance() = 0;
-  // Clones the iterator object. Used for implementing copy semantics
-  // of ParamIterator<T>.
-  virtual ParamIteratorInterface* Clone() const = 0;
-  // Dereferences the current iterator and provides (read-only) access
-  // to the pointed value. It is the caller's responsibility not to call
-  // Current() on an iterator equal to BaseGenerator()->End().
-  // Used for implementing ParamGenerator<T>::operator*().
-  virtual const T* Current() const = 0;
-  // Determines whether the given iterator and other point to the same
-  // element in the sequence generated by the generator.
-  // Used for implementing ParamGenerator<T>::operator==().
-  virtual bool Equals(const ParamIteratorInterface& other) const = 0;
-};
-
-// Class iterating over elements provided by an implementation of
-// ParamGeneratorInterface<T>. It wraps ParamIteratorInterface<T>
-// and implements the const forward iterator concept.
-template <typename T>
-class ParamIterator {
- public:
-  typedef T value_type;
-  typedef const T& reference;
-  typedef ptrdiff_t difference_type;
-
-  // ParamIterator assumes ownership of the impl_ pointer.
-  ParamIterator(const ParamIterator& other) : impl_(other.impl_->Clone()) {}
-  ParamIterator& operator=(const ParamIterator& other) {
-    if (this != &other)
-      impl_.reset(other.impl_->Clone());
-    return *this;
-  }
-
-  const T& operator*() const { return *impl_->Current(); }
-  const T* operator->() const { return impl_->Current(); }
-  // Prefix version of operator++.
-  ParamIterator& operator++() {
-    impl_->Advance();
-    return *this;
-  }
-  // Postfix version of operator++.
-  ParamIterator operator++(int /*unused*/) {
-    ParamIteratorInterface<T>* clone = impl_->Clone();
-    impl_->Advance();
-    return ParamIterator(clone);
-  }
-  bool operator==(const ParamIterator& other) const {
-    return impl_.get() == other.impl_.get() || impl_->Equals(*other.impl_);
-  }
-  bool operator!=(const ParamIterator& other) const {
-    return !(*this == other);
-  }
-
- private:
-  friend class ParamGenerator<T>;
-  explicit ParamIterator(ParamIteratorInterface<T>* impl) : impl_(impl) {}
-  scoped_ptr<ParamIteratorInterface<T> > impl_;
-};
-
-// ParamGeneratorInterface<T> is the binary interface to access generators
-// defined in other translation units.
-template <typename T>
-class ParamGeneratorInterface {
- public:
-  typedef T ParamType;
-
-  virtual ~ParamGeneratorInterface() {}
-
-  // Generator interface definition
-  virtual ParamIteratorInterface<T>* Begin() const = 0;
-  virtual ParamIteratorInterface<T>* End() const = 0;
-};
-
-// Wraps ParamGeneratorInterface<T> and provides general generator syntax
-// compatible with the STL Container concept.
-// This class implements copy initialization semantics and the contained
-// ParamGeneratorInterface<T> instance is shared among all copies
-// of the original object. This is possible because that instance is immutable.
-template<typename T>
-class ParamGenerator {
- public:
-  typedef ParamIterator<T> iterator;
-
-  explicit ParamGenerator(ParamGeneratorInterface<T>* impl) : impl_(impl) {}
-  ParamGenerator(const ParamGenerator& other) : impl_(other.impl_) {}
-
-  ParamGenerator& operator=(const ParamGenerator& other) {
-    impl_ = other.impl_;
-    return *this;
-  }
-
-  iterator begin() const { return iterator(impl_->Begin()); }
-  iterator end() const { return iterator(impl_->End()); }
-
- private:
-  linked_ptr<const ParamGeneratorInterface<T> > impl_;
-};
-
-// Generates values from a range of two comparable values. Can be used to
-// generate sequences of user-defined types that implement operator+() and
-// operator<().
-// This class is used in the Range() function.
-template <typename T, typename IncrementT>
-class RangeGenerator : public ParamGeneratorInterface<T> {
- public:
-  RangeGenerator(T begin, T end, IncrementT step)
-      : begin_(begin), end_(end),
-        step_(step), end_index_(CalculateEndIndex(begin, end, step)) {}
-  virtual ~RangeGenerator() {}
-
-  virtual ParamIteratorInterface<T>* Begin() const {
-    return new Iterator(this, begin_, 0, step_);
-  }
-  virtual ParamIteratorInterface<T>* End() const {
-    return new Iterator(this, end_, end_index_, step_);
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<T> {
-   public:
-    Iterator(const ParamGeneratorInterface<T>* base, T value, int index,
-             IncrementT step)
-        : base_(base), value_(value), index_(index), step_(step) {}
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<T>* BaseGenerator() const {
-      return base_;
-    }
-    virtual void Advance() {
-      value_ = static_cast<T>(value_ + step_);
-      index_++;
-    }
-    virtual ParamIteratorInterface<T>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const T* Current() const { return &value_; }
-    virtual bool Equals(const ParamIteratorInterface<T>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const int other_index =
-          CheckedDowncastToActualType<const Iterator>(&other)->index_;
-      return index_ == other_index;
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : ParamIteratorInterface<T>(),
-          base_(other.base_), value_(other.value_), index_(other.index_),
-          step_(other.step_) {}
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<T>* const base_;
-    T value_;
-    int index_;
-    const IncrementT step_;
-  };  // class RangeGenerator::Iterator
-
-  static int CalculateEndIndex(const T& begin,
-                               const T& end,
-                               const IncrementT& step) {
-    int end_index = 0;
-    for (T i = begin; i < end; i = static_cast<T>(i + step))
-      end_index++;
-    return end_index;
-  }
-
-  // No implementation - assignment is unsupported.
-  void operator=(const RangeGenerator& other);
-
-  const T begin_;
-  const T end_;
-  const IncrementT step_;
-  // The index for the end() iterator. All the elements in the generated
-  // sequence are indexed (0-based) to aid iterator comparison.
-  const int end_index_;
-};  // class RangeGenerator
-
-
-// Generates values from a pair of STL-style iterators. Used in the
-// ValuesIn() function. The elements are copied from the source range
-// since the source can be located on the stack, and the generator
-// is likely to persist beyond that stack frame.
-template <typename T>
-class ValuesInIteratorRangeGenerator : public ParamGeneratorInterface<T> {
- public:
-  template <typename ForwardIterator>
-  ValuesInIteratorRangeGenerator(ForwardIterator begin, ForwardIterator end)
-      : container_(begin, end) {}
-  virtual ~ValuesInIteratorRangeGenerator() {}
-
-  virtual ParamIteratorInterface<T>* Begin() const {
-    return new Iterator(this, container_.begin());
-  }
-  virtual ParamIteratorInterface<T>* End() const {
-    return new Iterator(this, container_.end());
-  }
-
- private:
-  typedef typename ::std::vector<T> ContainerType;
-
-  class Iterator : public ParamIteratorInterface<T> {
-   public:
-    Iterator(const ParamGeneratorInterface<T>* base,
-             typename ContainerType::const_iterator iterator)
-        : base_(base), iterator_(iterator) {}
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<T>* BaseGenerator() const {
-      return base_;
-    }
-    virtual void Advance() {
-      ++iterator_;
-      value_.reset();
-    }
-    virtual ParamIteratorInterface<T>* Clone() const {
-      return new Iterator(*this);
-    }
-    // We need to use cached value referenced by iterator_ because *iterator_
-    // can return a temporary object (and of type other then T), so just
-    // having "return &*iterator_;" doesn't work.
-    // value_ is updated here and not in Advance() because Advance()
-    // can advance iterator_ beyond the end of the range, and we cannot
-    // detect that fact. The client code, on the other hand, is
-    // responsible for not calling Current() on an out-of-range iterator.
-    virtual const T* Current() const {
-      if (value_.get() == NULL)
-        value_.reset(new T(*iterator_));
-      return value_.get();
-    }
-    virtual bool Equals(const ParamIteratorInterface<T>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      return iterator_ ==
-          CheckedDowncastToActualType<const Iterator>(&other)->iterator_;
-    }
-
-   private:
-    Iterator(const Iterator& other)
-          // The explicit constructor call suppresses a false warning
-          // emitted by gcc when supplied with the -Wextra option.
-        : ParamIteratorInterface<T>(),
-          base_(other.base_),
-          iterator_(other.iterator_) {}
-
-    const ParamGeneratorInterface<T>* const base_;
-    typename ContainerType::const_iterator iterator_;
-    // A cached value of *iterator_. We keep it here to allow access by
-    // pointer in the wrapping iterator's operator->().
-    // value_ needs to be mutable to be accessed in Current().
-    // Use of scoped_ptr helps manage cached value's lifetime,
-    // which is bound by the lifespan of the iterator itself.
-    mutable scoped_ptr<const T> value_;
-  };  // class ValuesInIteratorRangeGenerator::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const ValuesInIteratorRangeGenerator& other);
-
-  const ContainerType container_;
-};  // class ValuesInIteratorRangeGenerator
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Default parameterized test name generator, returns a string containing the
-// integer test parameter index.
-template <class ParamType>
-std::string DefaultParamName(const TestParamInfo<ParamType>& info) {
-  Message name_stream;
-  name_stream << info.index;
-  return name_stream.GetString();
-}
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Parameterized test name overload helpers, which help the
-// INSTANTIATE_TEST_CASE_P macro choose between the default parameterized
-// test name generator and user param name generator.
-template <class ParamType, class ParamNameGenFunctor>
-ParamNameGenFunctor GetParamNameGen(ParamNameGenFunctor func) {
-  return func;
-}
-
-template <class ParamType>
-struct ParamNameGenFunc {
-  typedef std::string Type(const TestParamInfo<ParamType>&);
-};
-
-template <class ParamType>
-typename ParamNameGenFunc<ParamType>::Type *GetParamNameGen() {
-  return DefaultParamName;
-}
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Stores a parameter value and later creates tests parameterized with that
-// value.
-template <class TestClass>
-class ParameterizedTestFactory : public TestFactoryBase {
- public:
-  typedef typename TestClass::ParamType ParamType;
-  explicit ParameterizedTestFactory(ParamType parameter) :
-      parameter_(parameter) {}
-  virtual Test* CreateTest() {
-    TestClass::SetParam(&parameter_);
-    return new TestClass();
-  }
-
- private:
-  const ParamType parameter_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ParameterizedTestFactory);
-};
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// TestMetaFactoryBase is a base class for meta-factories that create
-// test factories for passing into MakeAndRegisterTestInfo function.
-template <class ParamType>
-class TestMetaFactoryBase {
- public:
-  virtual ~TestMetaFactoryBase() {}
-
-  virtual TestFactoryBase* CreateTestFactory(ParamType parameter) = 0;
-};
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// TestMetaFactory creates test factories for passing into
-// MakeAndRegisterTestInfo function. Since MakeAndRegisterTestInfo receives
-// ownership of test factory pointer, same factory object cannot be passed
-// into that method twice. But ParameterizedTestCaseInfo is going to call
-// it for each Test/Parameter value combination. Thus it needs meta factory
-// creator class.
-template <class TestCase>
-class TestMetaFactory
-    : public TestMetaFactoryBase<typename TestCase::ParamType> {
- public:
-  typedef typename TestCase::ParamType ParamType;
-
-  TestMetaFactory() {}
-
-  virtual TestFactoryBase* CreateTestFactory(ParamType parameter) {
-    return new ParameterizedTestFactory<TestCase>(parameter);
-  }
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestMetaFactory);
-};
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// ParameterizedTestCaseInfoBase is a generic interface
-// to ParameterizedTestCaseInfo classes. ParameterizedTestCaseInfoBase
-// accumulates test information provided by TEST_P macro invocations
-// and generators provided by INSTANTIATE_TEST_CASE_P macro invocations
-// and uses that information to register all resulting test instances
-// in RegisterTests method. The ParameterizeTestCaseRegistry class holds
-// a collection of pointers to the ParameterizedTestCaseInfo objects
-// and calls RegisterTests() on each of them when asked.
-class ParameterizedTestCaseInfoBase {
- public:
-  virtual ~ParameterizedTestCaseInfoBase() {}
-
-  // Base part of test case name for display purposes.
-  virtual const string& GetTestCaseName() const = 0;
-  // Test case id to verify identity.
-  virtual TypeId GetTestCaseTypeId() const = 0;
-  // UnitTest class invokes this method to register tests in this
-  // test case right before running them in RUN_ALL_TESTS macro.
-  // This method should not be called more then once on any single
-  // instance of a ParameterizedTestCaseInfoBase derived class.
-  virtual void RegisterTests() = 0;
-
- protected:
-  ParameterizedTestCaseInfoBase() {}
-
- private:
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ParameterizedTestCaseInfoBase);
-};
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// ParameterizedTestCaseInfo accumulates tests obtained from TEST_P
-// macro invocations for a particular test case and generators
-// obtained from INSTANTIATE_TEST_CASE_P macro invocations for that
-// test case. It registers tests with all values generated by all
-// generators when asked.
-template <class TestCase>
-class ParameterizedTestCaseInfo : public ParameterizedTestCaseInfoBase {
- public:
-  // ParamType and GeneratorCreationFunc are private types but are required
-  // for declarations of public methods AddTestPattern() and
-  // AddTestCaseInstantiation().
-  typedef typename TestCase::ParamType ParamType;
-  // A function that returns an instance of appropriate generator type.
-  typedef ParamGenerator<ParamType>(GeneratorCreationFunc)();
-  typedef typename ParamNameGenFunc<ParamType>::Type ParamNameGeneratorFunc;
-
-  explicit ParameterizedTestCaseInfo(
-      const char* name, CodeLocation code_location)
-      : test_case_name_(name), code_location_(code_location) {}
-
-  // Test case base name for display purposes.
-  virtual const string& GetTestCaseName() const { return test_case_name_; }
-  // Test case id to verify identity.
-  virtual TypeId GetTestCaseTypeId() const { return GetTypeId<TestCase>(); }
-  // TEST_P macro uses AddTestPattern() to record information
-  // about a single test in a LocalTestInfo structure.
-  // test_case_name is the base name of the test case (without invocation
-  // prefix). test_base_name is the name of an individual test without
-  // parameter index. For the test SequenceA/FooTest.DoBar/1 FooTest is
-  // test case base name and DoBar is test base name.
-  void AddTestPattern(const char* test_case_name,
-                      const char* test_base_name,
-                      TestMetaFactoryBase<ParamType>* meta_factory) {
-    tests_.push_back(linked_ptr<TestInfo>(new TestInfo(test_case_name,
-                                                       test_base_name,
-                                                       meta_factory)));
-  }
-  // INSTANTIATE_TEST_CASE_P macro uses AddGenerator() to record information
-  // about a generator.
-  int AddTestCaseInstantiation(const string& instantiation_name,
-                               GeneratorCreationFunc* func,
-                               ParamNameGeneratorFunc* name_func,
-                               const char* file,
-                               int line) {
-    instantiations_.push_back(
-        InstantiationInfo(instantiation_name, func, name_func, file, line));
-    return 0;  // Return value used only to run this method in namespace scope.
-  }
-  // UnitTest class invokes this method to register tests in this test case
-  // test cases right before running tests in RUN_ALL_TESTS macro.
-  // This method should not be called more then once on any single
-  // instance of a ParameterizedTestCaseInfoBase derived class.
-  // UnitTest has a guard to prevent from calling this method more then once.
-  virtual void RegisterTests() {
-    for (typename TestInfoContainer::iterator test_it = tests_.begin();
-         test_it != tests_.end(); ++test_it) {
-      linked_ptr<TestInfo> test_info = *test_it;
-      for (typename InstantiationContainer::iterator gen_it =
-               instantiations_.begin(); gen_it != instantiations_.end();
-               ++gen_it) {
-        const string& instantiation_name = gen_it->name;
-        ParamGenerator<ParamType> generator((*gen_it->generator)());
-        ParamNameGeneratorFunc* name_func = gen_it->name_func;
-        const char* file = gen_it->file;
-        int line = gen_it->line;
-
-        string test_case_name;
-        if ( !instantiation_name.empty() )
-          test_case_name = instantiation_name + "/";
-        test_case_name += test_info->test_case_base_name;
-
-        size_t i = 0;
-        std::set<std::string> test_param_names;
-        for (typename ParamGenerator<ParamType>::iterator param_it =
-                 generator.begin();
-             param_it != generator.end(); ++param_it, ++i) {
-          Message test_name_stream;
-
-          std::string param_name = name_func(
-              TestParamInfo<ParamType>(*param_it, i));
-
-          GTEST_CHECK_(IsValidParamName(param_name))
-              << "Parameterized test name '" << param_name
-              << "' is invalid, in " << file
-              << " line " << line << std::endl;
-
-          GTEST_CHECK_(test_param_names.count(param_name) == 0)
-              << "Duplicate parameterized test name '" << param_name
-              << "', in " << file << " line " << line << std::endl;
-
-          test_param_names.insert(param_name);
-
-          test_name_stream << test_info->test_base_name << "/" << param_name;
-          MakeAndRegisterTestInfo(
-              test_case_name.c_str(),
-              test_name_stream.GetString().c_str(),
-              NULL,  // No type parameter.
-              PrintToString(*param_it).c_str(),
-              code_location_,
-              GetTestCaseTypeId(),
-              TestCase::SetUpTestCase,
-              TestCase::TearDownTestCase,
-              test_info->test_meta_factory->CreateTestFactory(*param_it));
-        }  // for param_it
-      }  // for gen_it
-    }  // for test_it
-  }  // RegisterTests
-
- private:
-  // LocalTestInfo structure keeps information about a single test registered
-  // with TEST_P macro.
-  struct TestInfo {
-    TestInfo(const char* a_test_case_base_name,
-             const char* a_test_base_name,
-             TestMetaFactoryBase<ParamType>* a_test_meta_factory) :
-        test_case_base_name(a_test_case_base_name),
-        test_base_name(a_test_base_name),
-        test_meta_factory(a_test_meta_factory) {}
-
-    const string test_case_base_name;
-    const string test_base_name;
-    const scoped_ptr<TestMetaFactoryBase<ParamType> > test_meta_factory;
-  };
-  typedef ::std::vector<linked_ptr<TestInfo> > TestInfoContainer;
-  // Records data received from INSTANTIATE_TEST_CASE_P macros:
-  //  <Instantiation name, Sequence generator creation function,
-  //     Name generator function, Source file, Source line>
-  struct InstantiationInfo {
-      InstantiationInfo(const std::string &name_in,
-                        GeneratorCreationFunc* generator_in,
-                        ParamNameGeneratorFunc* name_func_in,
-                        const char* file_in,
-                        int line_in)
-          : name(name_in),
-            generator(generator_in),
-            name_func(name_func_in),
-            file(file_in),
-            line(line_in) {}
-
-      std::string name;
-      GeneratorCreationFunc* generator;
-      ParamNameGeneratorFunc* name_func;
-      const char* file;
-      int line;
-  };
-  typedef ::std::vector<InstantiationInfo> InstantiationContainer;
-
-  static bool IsValidParamName(const std::string& name) {
-    // Check for empty string
-    if (name.empty())
-      return false;
-
-    // Check for invalid characters
-    for (std::string::size_type index = 0; index < name.size(); ++index) {
-      if (!isalnum(name[index]) && name[index] != '_')
-        return false;
-    }
-
-    return true;
-  }
-
-  const string test_case_name_;
-  CodeLocation code_location_;
-  TestInfoContainer tests_;
-  InstantiationContainer instantiations_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ParameterizedTestCaseInfo);
-};  // class ParameterizedTestCaseInfo
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// ParameterizedTestCaseRegistry contains a map of ParameterizedTestCaseInfoBase
-// classes accessed by test case names. TEST_P and INSTANTIATE_TEST_CASE_P
-// macros use it to locate their corresponding ParameterizedTestCaseInfo
-// descriptors.
-class ParameterizedTestCaseRegistry {
- public:
-  ParameterizedTestCaseRegistry() {}
-  ~ParameterizedTestCaseRegistry() {
-    for (TestCaseInfoContainer::iterator it = test_case_infos_.begin();
-         it != test_case_infos_.end(); ++it) {
-      delete *it;
-    }
-  }
-
-  // Looks up or creates and returns a structure containing information about
-  // tests and instantiations of a particular test case.
-  template <class TestCase>
-  ParameterizedTestCaseInfo<TestCase>* GetTestCasePatternHolder(
-      const char* test_case_name,
-      CodeLocation code_location) {
-    ParameterizedTestCaseInfo<TestCase>* typed_test_info = NULL;
-    for (TestCaseInfoContainer::iterator it = test_case_infos_.begin();
-         it != test_case_infos_.end(); ++it) {
-      if ((*it)->GetTestCaseName() == test_case_name) {
-        if ((*it)->GetTestCaseTypeId() != GetTypeId<TestCase>()) {
-          // Complain about incorrect usage of Google Test facilities
-          // and terminate the program since we cannot guaranty correct
-          // test case setup and tear-down in this case.
-          ReportInvalidTestCaseType(test_case_name, code_location);
-          posix::Abort();
-        } else {
-          // At this point we are sure that the object we found is of the same
-          // type we are looking for, so we downcast it to that type
-          // without further checks.
-          typed_test_info = CheckedDowncastToActualType<
-              ParameterizedTestCaseInfo<TestCase> >(*it);
-        }
-        break;
-      }
-    }
-    if (typed_test_info == NULL) {
-      typed_test_info = new ParameterizedTestCaseInfo<TestCase>(
-          test_case_name, code_location);
-      test_case_infos_.push_back(typed_test_info);
-    }
-    return typed_test_info;
-  }
-  void RegisterTests() {
-    for (TestCaseInfoContainer::iterator it = test_case_infos_.begin();
-         it != test_case_infos_.end(); ++it) {
-      (*it)->RegisterTests();
-    }
-  }
-
- private:
-  typedef ::std::vector<ParameterizedTestCaseInfoBase*> TestCaseInfoContainer;
-
-  TestCaseInfoContainer test_case_infos_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(ParameterizedTestCaseRegistry);
-};
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  //  GTEST_HAS_PARAM_TEST
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PARAM_UTIL_H_
-// This file was GENERATED by command:
-//     pump.py gtest-param-util-generated.h.pump
-// DO NOT EDIT BY HAND!!!
-
-// Copyright 2008 Google Inc.
-// All Rights Reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: vladl@google.com (Vlad Losev)
-
-// Type and function utilities for implementing parameterized tests.
-// This file is generated by a SCRIPT.  DO NOT EDIT BY HAND!
-//
-// Currently Google Test supports at most 50 arguments in Values,
-// and at most 10 arguments in Combine. Please contact
-// googletestframework@googlegroups.com if you need more.
-// Please note that the number of arguments to Combine is limited
-// by the maximum arity of the implementation of tuple which is
-// currently set at 10.
-
-#ifndef GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PARAM_UTIL_GENERATED_H_
-#define GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PARAM_UTIL_GENERATED_H_
-
-// scripts/fuse_gtest.py depends on gtest's own header being #included
-// *unconditionally*.  Therefore these #includes cannot be moved
-// inside #if GTEST_HAS_PARAM_TEST.
-
-#if GTEST_HAS_PARAM_TEST
-
-namespace testing {
-
-// Forward declarations of ValuesIn(), which is implemented in
-// include/gtest/gtest-param-test.h.
-template <typename ForwardIterator>
-internal::ParamGenerator<
-  typename ::testing::internal::IteratorTraits<ForwardIterator>::value_type>
-ValuesIn(ForwardIterator begin, ForwardIterator end);
-
-template <typename T, size_t N>
-internal::ParamGenerator<T> ValuesIn(const T (&array)[N]);
-
-template <class Container>
-internal::ParamGenerator<typename Container::value_type> ValuesIn(
-    const Container& container);
-
-namespace internal {
-
-// Used in the Values() function to provide polymorphic capabilities.
-template <typename T1>
-class ValueArray1 {
- public:
-  explicit ValueArray1(T1 v1) : v1_(v1) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray1& other);
-
-  const T1 v1_;
-};
-
-template <typename T1, typename T2>
-class ValueArray2 {
- public:
-  ValueArray2(T1 v1, T2 v2) : v1_(v1), v2_(v2) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray2& other);
-
-  const T1 v1_;
-  const T2 v2_;
-};
-
-template <typename T1, typename T2, typename T3>
-class ValueArray3 {
- public:
-  ValueArray3(T1 v1, T2 v2, T3 v3) : v1_(v1), v2_(v2), v3_(v3) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray3& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4>
-class ValueArray4 {
- public:
-  ValueArray4(T1 v1, T2 v2, T3 v3, T4 v4) : v1_(v1), v2_(v2), v3_(v3),
-      v4_(v4) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray4& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5>
-class ValueArray5 {
- public:
-  ValueArray5(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5) : v1_(v1), v2_(v2), v3_(v3),
-      v4_(v4), v5_(v5) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray5& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6>
-class ValueArray6 {
- public:
-  ValueArray6(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6) : v1_(v1), v2_(v2),
-      v3_(v3), v4_(v4), v5_(v5), v6_(v6) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray6& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7>
-class ValueArray7 {
- public:
-  ValueArray7(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7) : v1_(v1),
-      v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray7& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8>
-class ValueArray8 {
- public:
-  ValueArray8(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7,
-      T8 v8) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray8& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9>
-class ValueArray9 {
- public:
-  ValueArray9(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8,
-      T9 v9) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray9& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10>
-class ValueArray10 {
- public:
-  ValueArray10(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray10& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11>
-class ValueArray11 {
- public:
-  ValueArray11(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6),
-      v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray11& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12>
-class ValueArray12 {
- public:
-  ValueArray12(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5),
-      v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray12& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13>
-class ValueArray13 {
- public:
-  ValueArray13(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13) : v1_(v1), v2_(v2), v3_(v3), v4_(v4),
-      v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11),
-      v12_(v12), v13_(v13) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray13& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14>
-class ValueArray14 {
- public:
-  ValueArray14(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14) : v1_(v1), v2_(v2), v3_(v3),
-      v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray14& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15>
-class ValueArray15 {
- public:
-  ValueArray15(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15) : v1_(v1), v2_(v2),
-      v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray15& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16>
-class ValueArray16 {
- public:
-  ValueArray16(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16) : v1_(v1),
-      v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9),
-      v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15),
-      v16_(v16) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray16& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17>
-class ValueArray17 {
- public:
-  ValueArray17(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16,
-      T17 v17) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray17& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18>
-class ValueArray18 {
- public:
-  ValueArray18(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray18& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19>
-class ValueArray19 {
- public:
-  ValueArray19(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6),
-      v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13),
-      v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray19& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20>
-class ValueArray20 {
- public:
-  ValueArray20(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5),
-      v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12),
-      v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18),
-      v19_(v19), v20_(v20) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray20& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21>
-class ValueArray21 {
- public:
-  ValueArray21(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21) : v1_(v1), v2_(v2), v3_(v3), v4_(v4),
-      v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11),
-      v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17),
-      v18_(v18), v19_(v19), v20_(v20), v21_(v21) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray21& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22>
-class ValueArray22 {
- public:
-  ValueArray22(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22) : v1_(v1), v2_(v2), v3_(v3),
-      v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray22& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23>
-class ValueArray23 {
- public:
-  ValueArray23(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23) : v1_(v1), v2_(v2),
-      v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22),
-      v23_(v23) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray23& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24>
-class ValueArray24 {
- public:
-  ValueArray24(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24) : v1_(v1),
-      v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9),
-      v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15),
-      v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21),
-      v22_(v22), v23_(v23), v24_(v24) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray24& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25>
-class ValueArray25 {
- public:
-  ValueArray25(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24,
-      T25 v25) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray25& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26>
-class ValueArray26 {
- public:
-  ValueArray26(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray26& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27>
-class ValueArray27 {
- public:
-  ValueArray27(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6),
-      v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13),
-      v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19),
-      v20_(v20), v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25),
-      v26_(v26), v27_(v27) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray27& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28>
-class ValueArray28 {
- public:
-  ValueArray28(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5),
-      v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12),
-      v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18),
-      v19_(v19), v20_(v20), v21_(v21), v22_(v22), v23_(v23), v24_(v24),
-      v25_(v25), v26_(v26), v27_(v27), v28_(v28) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray28& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29>
-class ValueArray29 {
- public:
-  ValueArray29(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29) : v1_(v1), v2_(v2), v3_(v3), v4_(v4),
-      v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11),
-      v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17),
-      v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22), v23_(v23),
-      v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28), v29_(v29) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray29& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30>
-class ValueArray30 {
- public:
-  ValueArray30(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30) : v1_(v1), v2_(v2), v3_(v3),
-      v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22),
-      v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28),
-      v29_(v29), v30_(v30) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray30& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31>
-class ValueArray31 {
- public:
-  ValueArray31(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31) : v1_(v1), v2_(v2),
-      v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22),
-      v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28),
-      v29_(v29), v30_(v30), v31_(v31) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray31& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32>
-class ValueArray32 {
- public:
-  ValueArray32(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32) : v1_(v1),
-      v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9),
-      v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15),
-      v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21),
-      v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27),
-      v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray32& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33>
-class ValueArray33 {
- public:
-  ValueArray33(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32,
-      T33 v33) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26),
-      v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32),
-      v33_(v33) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray33& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34>
-class ValueArray34 {
- public:
-  ValueArray34(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26),
-      v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32),
-      v33_(v33), v34_(v34) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray34& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35>
-class ValueArray35 {
- public:
-  ValueArray35(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6),
-      v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13),
-      v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19),
-      v20_(v20), v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25),
-      v26_(v26), v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31),
-      v32_(v32), v33_(v33), v34_(v34), v35_(v35) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray35& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36>
-class ValueArray36 {
- public:
-  ValueArray36(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5),
-      v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12),
-      v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18),
-      v19_(v19), v20_(v20), v21_(v21), v22_(v22), v23_(v23), v24_(v24),
-      v25_(v25), v26_(v26), v27_(v27), v28_(v28), v29_(v29), v30_(v30),
-      v31_(v31), v32_(v32), v33_(v33), v34_(v34), v35_(v35), v36_(v36) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray36& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37>
-class ValueArray37 {
- public:
-  ValueArray37(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37) : v1_(v1), v2_(v2), v3_(v3), v4_(v4),
-      v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11),
-      v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17),
-      v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22), v23_(v23),
-      v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28), v29_(v29),
-      v30_(v30), v31_(v31), v32_(v32), v33_(v33), v34_(v34), v35_(v35),
-      v36_(v36), v37_(v37) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray37& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38>
-class ValueArray38 {
- public:
-  ValueArray38(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38) : v1_(v1), v2_(v2), v3_(v3),
-      v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22),
-      v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28),
-      v29_(v29), v30_(v30), v31_(v31), v32_(v32), v33_(v33), v34_(v34),
-      v35_(v35), v36_(v36), v37_(v37), v38_(v38) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray38& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39>
-class ValueArray39 {
- public:
-  ValueArray39(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39) : v1_(v1), v2_(v2),
-      v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22),
-      v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28),
-      v29_(v29), v30_(v30), v31_(v31), v32_(v32), v33_(v33), v34_(v34),
-      v35_(v35), v36_(v36), v37_(v37), v38_(v38), v39_(v39) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray39& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40>
-class ValueArray40 {
- public:
-  ValueArray40(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40) : v1_(v1),
-      v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9),
-      v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15),
-      v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21),
-      v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27),
-      v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32), v33_(v33),
-      v34_(v34), v35_(v35), v36_(v36), v37_(v37), v38_(v38), v39_(v39),
-      v40_(v40) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray40& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41>
-class ValueArray41 {
- public:
-  ValueArray41(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40,
-      T41 v41) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26),
-      v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32),
-      v33_(v33), v34_(v34), v35_(v35), v36_(v36), v37_(v37), v38_(v38),
-      v39_(v39), v40_(v40), v41_(v41) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray41& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42>
-class ValueArray42 {
- public:
-  ValueArray42(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26),
-      v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32),
-      v33_(v33), v34_(v34), v35_(v35), v36_(v36), v37_(v37), v38_(v38),
-      v39_(v39), v40_(v40), v41_(v41), v42_(v42) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray42& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43>
-class ValueArray43 {
- public:
-  ValueArray43(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6),
-      v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13),
-      v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19),
-      v20_(v20), v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25),
-      v26_(v26), v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31),
-      v32_(v32), v33_(v33), v34_(v34), v35_(v35), v36_(v36), v37_(v37),
-      v38_(v38), v39_(v39), v40_(v40), v41_(v41), v42_(v42), v43_(v43) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray43& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44>
-class ValueArray44 {
- public:
-  ValueArray44(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43, T44 v44) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5),
-      v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12),
-      v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18),
-      v19_(v19), v20_(v20), v21_(v21), v22_(v22), v23_(v23), v24_(v24),
-      v25_(v25), v26_(v26), v27_(v27), v28_(v28), v29_(v29), v30_(v30),
-      v31_(v31), v32_(v32), v33_(v33), v34_(v34), v35_(v35), v36_(v36),
-      v37_(v37), v38_(v38), v39_(v39), v40_(v40), v41_(v41), v42_(v42),
-      v43_(v43), v44_(v44) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_), static_cast<T>(v44_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray44& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-  const T44 v44_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45>
-class ValueArray45 {
- public:
-  ValueArray45(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43, T44 v44, T45 v45) : v1_(v1), v2_(v2), v3_(v3), v4_(v4),
-      v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11),
-      v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16), v17_(v17),
-      v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22), v23_(v23),
-      v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28), v29_(v29),
-      v30_(v30), v31_(v31), v32_(v32), v33_(v33), v34_(v34), v35_(v35),
-      v36_(v36), v37_(v37), v38_(v38), v39_(v39), v40_(v40), v41_(v41),
-      v42_(v42), v43_(v43), v44_(v44), v45_(v45) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_), static_cast<T>(v44_),
-        static_cast<T>(v45_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray45& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-  const T44 v44_;
-  const T45 v45_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46>
-class ValueArray46 {
- public:
-  ValueArray46(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43, T44 v44, T45 v45, T46 v46) : v1_(v1), v2_(v2), v3_(v3),
-      v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22),
-      v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28),
-      v29_(v29), v30_(v30), v31_(v31), v32_(v32), v33_(v33), v34_(v34),
-      v35_(v35), v36_(v36), v37_(v37), v38_(v38), v39_(v39), v40_(v40),
-      v41_(v41), v42_(v42), v43_(v43), v44_(v44), v45_(v45), v46_(v46) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_), static_cast<T>(v44_),
-        static_cast<T>(v45_), static_cast<T>(v46_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray46& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-  const T44 v44_;
-  const T45 v45_;
-  const T46 v46_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47>
-class ValueArray47 {
- public:
-  ValueArray47(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43, T44 v44, T45 v45, T46 v46, T47 v47) : v1_(v1), v2_(v2),
-      v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9), v10_(v10),
-      v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15), v16_(v16),
-      v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21), v22_(v22),
-      v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27), v28_(v28),
-      v29_(v29), v30_(v30), v31_(v31), v32_(v32), v33_(v33), v34_(v34),
-      v35_(v35), v36_(v36), v37_(v37), v38_(v38), v39_(v39), v40_(v40),
-      v41_(v41), v42_(v42), v43_(v43), v44_(v44), v45_(v45), v46_(v46),
-      v47_(v47) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_), static_cast<T>(v44_),
-        static_cast<T>(v45_), static_cast<T>(v46_), static_cast<T>(v47_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray47& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-  const T44 v44_;
-  const T45 v45_;
-  const T46 v46_;
-  const T47 v47_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48>
-class ValueArray48 {
- public:
-  ValueArray48(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43, T44 v44, T45 v45, T46 v46, T47 v47, T48 v48) : v1_(v1),
-      v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7), v8_(v8), v9_(v9),
-      v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14), v15_(v15),
-      v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20), v21_(v21),
-      v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26), v27_(v27),
-      v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32), v33_(v33),
-      v34_(v34), v35_(v35), v36_(v36), v37_(v37), v38_(v38), v39_(v39),
-      v40_(v40), v41_(v41), v42_(v42), v43_(v43), v44_(v44), v45_(v45),
-      v46_(v46), v47_(v47), v48_(v48) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_), static_cast<T>(v44_),
-        static_cast<T>(v45_), static_cast<T>(v46_), static_cast<T>(v47_),
-        static_cast<T>(v48_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray48& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-  const T44 v44_;
-  const T45 v45_;
-  const T46 v46_;
-  const T47 v47_;
-  const T48 v48_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49>
-class ValueArray49 {
- public:
-  ValueArray49(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43, T44 v44, T45 v45, T46 v46, T47 v47, T48 v48,
-      T49 v49) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26),
-      v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32),
-      v33_(v33), v34_(v34), v35_(v35), v36_(v36), v37_(v37), v38_(v38),
-      v39_(v39), v40_(v40), v41_(v41), v42_(v42), v43_(v43), v44_(v44),
-      v45_(v45), v46_(v46), v47_(v47), v48_(v48), v49_(v49) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_), static_cast<T>(v44_),
-        static_cast<T>(v45_), static_cast<T>(v46_), static_cast<T>(v47_),
-        static_cast<T>(v48_), static_cast<T>(v49_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray49& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-  const T44 v44_;
-  const T45 v45_;
-  const T46 v46_;
-  const T47 v47_;
-  const T48 v48_;
-  const T49 v49_;
-};
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49, typename T50>
-class ValueArray50 {
- public:
-  ValueArray50(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-      T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-      T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-      T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-      T42 v42, T43 v43, T44 v44, T45 v45, T46 v46, T47 v47, T48 v48, T49 v49,
-      T50 v50) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6), v7_(v7),
-      v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13), v14_(v14),
-      v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19), v20_(v20),
-      v21_(v21), v22_(v22), v23_(v23), v24_(v24), v25_(v25), v26_(v26),
-      v27_(v27), v28_(v28), v29_(v29), v30_(v30), v31_(v31), v32_(v32),
-      v33_(v33), v34_(v34), v35_(v35), v36_(v36), v37_(v37), v38_(v38),
-      v39_(v39), v40_(v40), v41_(v41), v42_(v42), v43_(v43), v44_(v44),
-      v45_(v45), v46_(v46), v47_(v47), v48_(v48), v49_(v49), v50_(v50) {}
-
-  template <typename T>
-  operator ParamGenerator<T>() const {
-    const T array[] = {static_cast<T>(v1_), static_cast<T>(v2_),
-        static_cast<T>(v3_), static_cast<T>(v4_), static_cast<T>(v5_),
-        static_cast<T>(v6_), static_cast<T>(v7_), static_cast<T>(v8_),
-        static_cast<T>(v9_), static_cast<T>(v10_), static_cast<T>(v11_),
-        static_cast<T>(v12_), static_cast<T>(v13_), static_cast<T>(v14_),
-        static_cast<T>(v15_), static_cast<T>(v16_), static_cast<T>(v17_),
-        static_cast<T>(v18_), static_cast<T>(v19_), static_cast<T>(v20_),
-        static_cast<T>(v21_), static_cast<T>(v22_), static_cast<T>(v23_),
-        static_cast<T>(v24_), static_cast<T>(v25_), static_cast<T>(v26_),
-        static_cast<T>(v27_), static_cast<T>(v28_), static_cast<T>(v29_),
-        static_cast<T>(v30_), static_cast<T>(v31_), static_cast<T>(v32_),
-        static_cast<T>(v33_), static_cast<T>(v34_), static_cast<T>(v35_),
-        static_cast<T>(v36_), static_cast<T>(v37_), static_cast<T>(v38_),
-        static_cast<T>(v39_), static_cast<T>(v40_), static_cast<T>(v41_),
-        static_cast<T>(v42_), static_cast<T>(v43_), static_cast<T>(v44_),
-        static_cast<T>(v45_), static_cast<T>(v46_), static_cast<T>(v47_),
-        static_cast<T>(v48_), static_cast<T>(v49_), static_cast<T>(v50_)};
-    return ValuesIn(array);
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const ValueArray50& other);
-
-  const T1 v1_;
-  const T2 v2_;
-  const T3 v3_;
-  const T4 v4_;
-  const T5 v5_;
-  const T6 v6_;
-  const T7 v7_;
-  const T8 v8_;
-  const T9 v9_;
-  const T10 v10_;
-  const T11 v11_;
-  const T12 v12_;
-  const T13 v13_;
-  const T14 v14_;
-  const T15 v15_;
-  const T16 v16_;
-  const T17 v17_;
-  const T18 v18_;
-  const T19 v19_;
-  const T20 v20_;
-  const T21 v21_;
-  const T22 v22_;
-  const T23 v23_;
-  const T24 v24_;
-  const T25 v25_;
-  const T26 v26_;
-  const T27 v27_;
-  const T28 v28_;
-  const T29 v29_;
-  const T30 v30_;
-  const T31 v31_;
-  const T32 v32_;
-  const T33 v33_;
-  const T34 v34_;
-  const T35 v35_;
-  const T36 v36_;
-  const T37 v37_;
-  const T38 v38_;
-  const T39 v39_;
-  const T40 v40_;
-  const T41 v41_;
-  const T42 v42_;
-  const T43 v43_;
-  const T44 v44_;
-  const T45 v45_;
-  const T46 v46_;
-  const T47 v47_;
-  const T48 v48_;
-  const T49 v49_;
-  const T50 v50_;
-};
-
-# if GTEST_HAS_COMBINE
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Generates values from the Cartesian product of values produced
-// by the argument generators.
-//
-template <typename T1, typename T2>
-class CartesianProductGenerator2
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2> > {
- public:
-  typedef ::testing::tuple<T1, T2> ParamType;
-
-  CartesianProductGenerator2(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2)
-      : g1_(g1), g2_(g2) {}
-  virtual ~CartesianProductGenerator2() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current2_;
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator2::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator2& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-};  // class CartesianProductGenerator2
-
-
-template <typename T1, typename T2, typename T3>
-class CartesianProductGenerator3
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3> ParamType;
-
-  CartesianProductGenerator3(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3)
-      : g1_(g1), g2_(g2), g3_(g3) {}
-  virtual ~CartesianProductGenerator3() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current3_;
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator3::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator3& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-};  // class CartesianProductGenerator3
-
-
-template <typename T1, typename T2, typename T3, typename T4>
-class CartesianProductGenerator4
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3, T4> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3, T4> ParamType;
-
-  CartesianProductGenerator4(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3,
-      const ParamGenerator<T4>& g4)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4) {}
-  virtual ~CartesianProductGenerator4() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin(), g4_, g4_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end(),
-        g4_, g4_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3,
-      const ParamGenerator<T4>& g4,
-      const typename ParamGenerator<T4>::iterator& current4)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3),
-          begin4_(g4.begin()), end4_(g4.end()), current4_(current4)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current4_;
-      if (current4_ == end4_) {
-        current4_ = begin4_;
-        ++current3_;
-      }
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_ &&
-          current4_ == typed_other->current4_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_),
-        begin4_(other.begin4_),
-        end4_(other.end4_),
-        current4_(other.current4_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_,
-            *current4_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_ ||
-          current4_ == end4_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    const typename ParamGenerator<T4>::iterator begin4_;
-    const typename ParamGenerator<T4>::iterator end4_;
-    typename ParamGenerator<T4>::iterator current4_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator4::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator4& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-  const ParamGenerator<T4> g4_;
-};  // class CartesianProductGenerator4
-
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5>
-class CartesianProductGenerator5
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3, T4, T5> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3, T4, T5> ParamType;
-
-  CartesianProductGenerator5(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3,
-      const ParamGenerator<T4>& g4, const ParamGenerator<T5>& g5)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5) {}
-  virtual ~CartesianProductGenerator5() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin(), g4_, g4_.begin(), g5_, g5_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end(),
-        g4_, g4_.end(), g5_, g5_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3,
-      const ParamGenerator<T4>& g4,
-      const typename ParamGenerator<T4>::iterator& current4,
-      const ParamGenerator<T5>& g5,
-      const typename ParamGenerator<T5>::iterator& current5)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3),
-          begin4_(g4.begin()), end4_(g4.end()), current4_(current4),
-          begin5_(g5.begin()), end5_(g5.end()), current5_(current5)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current5_;
-      if (current5_ == end5_) {
-        current5_ = begin5_;
-        ++current4_;
-      }
-      if (current4_ == end4_) {
-        current4_ = begin4_;
-        ++current3_;
-      }
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_ &&
-          current4_ == typed_other->current4_ &&
-          current5_ == typed_other->current5_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_),
-        begin4_(other.begin4_),
-        end4_(other.end4_),
-        current4_(other.current4_),
-        begin5_(other.begin5_),
-        end5_(other.end5_),
-        current5_(other.current5_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_,
-            *current4_, *current5_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_ ||
-          current4_ == end4_ ||
-          current5_ == end5_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    const typename ParamGenerator<T4>::iterator begin4_;
-    const typename ParamGenerator<T4>::iterator end4_;
-    typename ParamGenerator<T4>::iterator current4_;
-    const typename ParamGenerator<T5>::iterator begin5_;
-    const typename ParamGenerator<T5>::iterator end5_;
-    typename ParamGenerator<T5>::iterator current5_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator5::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator5& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-  const ParamGenerator<T4> g4_;
-  const ParamGenerator<T5> g5_;
-};  // class CartesianProductGenerator5
-
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6>
-class CartesianProductGenerator6
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3, T4, T5,
-        T6> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3, T4, T5, T6> ParamType;
-
-  CartesianProductGenerator6(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3,
-      const ParamGenerator<T4>& g4, const ParamGenerator<T5>& g5,
-      const ParamGenerator<T6>& g6)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6) {}
-  virtual ~CartesianProductGenerator6() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin(), g4_, g4_.begin(), g5_, g5_.begin(), g6_, g6_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end(),
-        g4_, g4_.end(), g5_, g5_.end(), g6_, g6_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3,
-      const ParamGenerator<T4>& g4,
-      const typename ParamGenerator<T4>::iterator& current4,
-      const ParamGenerator<T5>& g5,
-      const typename ParamGenerator<T5>::iterator& current5,
-      const ParamGenerator<T6>& g6,
-      const typename ParamGenerator<T6>::iterator& current6)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3),
-          begin4_(g4.begin()), end4_(g4.end()), current4_(current4),
-          begin5_(g5.begin()), end5_(g5.end()), current5_(current5),
-          begin6_(g6.begin()), end6_(g6.end()), current6_(current6)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current6_;
-      if (current6_ == end6_) {
-        current6_ = begin6_;
-        ++current5_;
-      }
-      if (current5_ == end5_) {
-        current5_ = begin5_;
-        ++current4_;
-      }
-      if (current4_ == end4_) {
-        current4_ = begin4_;
-        ++current3_;
-      }
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_ &&
-          current4_ == typed_other->current4_ &&
-          current5_ == typed_other->current5_ &&
-          current6_ == typed_other->current6_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_),
-        begin4_(other.begin4_),
-        end4_(other.end4_),
-        current4_(other.current4_),
-        begin5_(other.begin5_),
-        end5_(other.end5_),
-        current5_(other.current5_),
-        begin6_(other.begin6_),
-        end6_(other.end6_),
-        current6_(other.current6_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_,
-            *current4_, *current5_, *current6_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_ ||
-          current4_ == end4_ ||
-          current5_ == end5_ ||
-          current6_ == end6_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    const typename ParamGenerator<T4>::iterator begin4_;
-    const typename ParamGenerator<T4>::iterator end4_;
-    typename ParamGenerator<T4>::iterator current4_;
-    const typename ParamGenerator<T5>::iterator begin5_;
-    const typename ParamGenerator<T5>::iterator end5_;
-    typename ParamGenerator<T5>::iterator current5_;
-    const typename ParamGenerator<T6>::iterator begin6_;
-    const typename ParamGenerator<T6>::iterator end6_;
-    typename ParamGenerator<T6>::iterator current6_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator6::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator6& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-  const ParamGenerator<T4> g4_;
-  const ParamGenerator<T5> g5_;
-  const ParamGenerator<T6> g6_;
-};  // class CartesianProductGenerator6
-
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7>
-class CartesianProductGenerator7
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3, T4, T5, T6,
-        T7> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3, T4, T5, T6, T7> ParamType;
-
-  CartesianProductGenerator7(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3,
-      const ParamGenerator<T4>& g4, const ParamGenerator<T5>& g5,
-      const ParamGenerator<T6>& g6, const ParamGenerator<T7>& g7)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7) {}
-  virtual ~CartesianProductGenerator7() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin(), g4_, g4_.begin(), g5_, g5_.begin(), g6_, g6_.begin(), g7_,
-        g7_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end(),
-        g4_, g4_.end(), g5_, g5_.end(), g6_, g6_.end(), g7_, g7_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3,
-      const ParamGenerator<T4>& g4,
-      const typename ParamGenerator<T4>::iterator& current4,
-      const ParamGenerator<T5>& g5,
-      const typename ParamGenerator<T5>::iterator& current5,
-      const ParamGenerator<T6>& g6,
-      const typename ParamGenerator<T6>::iterator& current6,
-      const ParamGenerator<T7>& g7,
-      const typename ParamGenerator<T7>::iterator& current7)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3),
-          begin4_(g4.begin()), end4_(g4.end()), current4_(current4),
-          begin5_(g5.begin()), end5_(g5.end()), current5_(current5),
-          begin6_(g6.begin()), end6_(g6.end()), current6_(current6),
-          begin7_(g7.begin()), end7_(g7.end()), current7_(current7)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current7_;
-      if (current7_ == end7_) {
-        current7_ = begin7_;
-        ++current6_;
-      }
-      if (current6_ == end6_) {
-        current6_ = begin6_;
-        ++current5_;
-      }
-      if (current5_ == end5_) {
-        current5_ = begin5_;
-        ++current4_;
-      }
-      if (current4_ == end4_) {
-        current4_ = begin4_;
-        ++current3_;
-      }
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_ &&
-          current4_ == typed_other->current4_ &&
-          current5_ == typed_other->current5_ &&
-          current6_ == typed_other->current6_ &&
-          current7_ == typed_other->current7_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_),
-        begin4_(other.begin4_),
-        end4_(other.end4_),
-        current4_(other.current4_),
-        begin5_(other.begin5_),
-        end5_(other.end5_),
-        current5_(other.current5_),
-        begin6_(other.begin6_),
-        end6_(other.end6_),
-        current6_(other.current6_),
-        begin7_(other.begin7_),
-        end7_(other.end7_),
-        current7_(other.current7_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_,
-            *current4_, *current5_, *current6_, *current7_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_ ||
-          current4_ == end4_ ||
-          current5_ == end5_ ||
-          current6_ == end6_ ||
-          current7_ == end7_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    const typename ParamGenerator<T4>::iterator begin4_;
-    const typename ParamGenerator<T4>::iterator end4_;
-    typename ParamGenerator<T4>::iterator current4_;
-    const typename ParamGenerator<T5>::iterator begin5_;
-    const typename ParamGenerator<T5>::iterator end5_;
-    typename ParamGenerator<T5>::iterator current5_;
-    const typename ParamGenerator<T6>::iterator begin6_;
-    const typename ParamGenerator<T6>::iterator end6_;
-    typename ParamGenerator<T6>::iterator current6_;
-    const typename ParamGenerator<T7>::iterator begin7_;
-    const typename ParamGenerator<T7>::iterator end7_;
-    typename ParamGenerator<T7>::iterator current7_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator7::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator7& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-  const ParamGenerator<T4> g4_;
-  const ParamGenerator<T5> g5_;
-  const ParamGenerator<T6> g6_;
-  const ParamGenerator<T7> g7_;
-};  // class CartesianProductGenerator7
-
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8>
-class CartesianProductGenerator8
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3, T4, T5, T6,
-        T7, T8> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8> ParamType;
-
-  CartesianProductGenerator8(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3,
-      const ParamGenerator<T4>& g4, const ParamGenerator<T5>& g5,
-      const ParamGenerator<T6>& g6, const ParamGenerator<T7>& g7,
-      const ParamGenerator<T8>& g8)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7),
-          g8_(g8) {}
-  virtual ~CartesianProductGenerator8() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin(), g4_, g4_.begin(), g5_, g5_.begin(), g6_, g6_.begin(), g7_,
-        g7_.begin(), g8_, g8_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end(),
-        g4_, g4_.end(), g5_, g5_.end(), g6_, g6_.end(), g7_, g7_.end(), g8_,
-        g8_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3,
-      const ParamGenerator<T4>& g4,
-      const typename ParamGenerator<T4>::iterator& current4,
-      const ParamGenerator<T5>& g5,
-      const typename ParamGenerator<T5>::iterator& current5,
-      const ParamGenerator<T6>& g6,
-      const typename ParamGenerator<T6>::iterator& current6,
-      const ParamGenerator<T7>& g7,
-      const typename ParamGenerator<T7>::iterator& current7,
-      const ParamGenerator<T8>& g8,
-      const typename ParamGenerator<T8>::iterator& current8)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3),
-          begin4_(g4.begin()), end4_(g4.end()), current4_(current4),
-          begin5_(g5.begin()), end5_(g5.end()), current5_(current5),
-          begin6_(g6.begin()), end6_(g6.end()), current6_(current6),
-          begin7_(g7.begin()), end7_(g7.end()), current7_(current7),
-          begin8_(g8.begin()), end8_(g8.end()), current8_(current8)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current8_;
-      if (current8_ == end8_) {
-        current8_ = begin8_;
-        ++current7_;
-      }
-      if (current7_ == end7_) {
-        current7_ = begin7_;
-        ++current6_;
-      }
-      if (current6_ == end6_) {
-        current6_ = begin6_;
-        ++current5_;
-      }
-      if (current5_ == end5_) {
-        current5_ = begin5_;
-        ++current4_;
-      }
-      if (current4_ == end4_) {
-        current4_ = begin4_;
-        ++current3_;
-      }
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_ &&
-          current4_ == typed_other->current4_ &&
-          current5_ == typed_other->current5_ &&
-          current6_ == typed_other->current6_ &&
-          current7_ == typed_other->current7_ &&
-          current8_ == typed_other->current8_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_),
-        begin4_(other.begin4_),
-        end4_(other.end4_),
-        current4_(other.current4_),
-        begin5_(other.begin5_),
-        end5_(other.end5_),
-        current5_(other.current5_),
-        begin6_(other.begin6_),
-        end6_(other.end6_),
-        current6_(other.current6_),
-        begin7_(other.begin7_),
-        end7_(other.end7_),
-        current7_(other.current7_),
-        begin8_(other.begin8_),
-        end8_(other.end8_),
-        current8_(other.current8_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_,
-            *current4_, *current5_, *current6_, *current7_, *current8_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_ ||
-          current4_ == end4_ ||
-          current5_ == end5_ ||
-          current6_ == end6_ ||
-          current7_ == end7_ ||
-          current8_ == end8_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    const typename ParamGenerator<T4>::iterator begin4_;
-    const typename ParamGenerator<T4>::iterator end4_;
-    typename ParamGenerator<T4>::iterator current4_;
-    const typename ParamGenerator<T5>::iterator begin5_;
-    const typename ParamGenerator<T5>::iterator end5_;
-    typename ParamGenerator<T5>::iterator current5_;
-    const typename ParamGenerator<T6>::iterator begin6_;
-    const typename ParamGenerator<T6>::iterator end6_;
-    typename ParamGenerator<T6>::iterator current6_;
-    const typename ParamGenerator<T7>::iterator begin7_;
-    const typename ParamGenerator<T7>::iterator end7_;
-    typename ParamGenerator<T7>::iterator current7_;
-    const typename ParamGenerator<T8>::iterator begin8_;
-    const typename ParamGenerator<T8>::iterator end8_;
-    typename ParamGenerator<T8>::iterator current8_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator8::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator8& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-  const ParamGenerator<T4> g4_;
-  const ParamGenerator<T5> g5_;
-  const ParamGenerator<T6> g6_;
-  const ParamGenerator<T7> g7_;
-  const ParamGenerator<T8> g8_;
-};  // class CartesianProductGenerator8
-
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9>
-class CartesianProductGenerator9
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3, T4, T5, T6,
-        T7, T8, T9> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9> ParamType;
-
-  CartesianProductGenerator9(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3,
-      const ParamGenerator<T4>& g4, const ParamGenerator<T5>& g5,
-      const ParamGenerator<T6>& g6, const ParamGenerator<T7>& g7,
-      const ParamGenerator<T8>& g8, const ParamGenerator<T9>& g9)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7), g8_(g8),
-          g9_(g9) {}
-  virtual ~CartesianProductGenerator9() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin(), g4_, g4_.begin(), g5_, g5_.begin(), g6_, g6_.begin(), g7_,
-        g7_.begin(), g8_, g8_.begin(), g9_, g9_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end(),
-        g4_, g4_.end(), g5_, g5_.end(), g6_, g6_.end(), g7_, g7_.end(), g8_,
-        g8_.end(), g9_, g9_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3,
-      const ParamGenerator<T4>& g4,
-      const typename ParamGenerator<T4>::iterator& current4,
-      const ParamGenerator<T5>& g5,
-      const typename ParamGenerator<T5>::iterator& current5,
-      const ParamGenerator<T6>& g6,
-      const typename ParamGenerator<T6>::iterator& current6,
-      const ParamGenerator<T7>& g7,
-      const typename ParamGenerator<T7>::iterator& current7,
-      const ParamGenerator<T8>& g8,
-      const typename ParamGenerator<T8>::iterator& current8,
-      const ParamGenerator<T9>& g9,
-      const typename ParamGenerator<T9>::iterator& current9)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3),
-          begin4_(g4.begin()), end4_(g4.end()), current4_(current4),
-          begin5_(g5.begin()), end5_(g5.end()), current5_(current5),
-          begin6_(g6.begin()), end6_(g6.end()), current6_(current6),
-          begin7_(g7.begin()), end7_(g7.end()), current7_(current7),
-          begin8_(g8.begin()), end8_(g8.end()), current8_(current8),
-          begin9_(g9.begin()), end9_(g9.end()), current9_(current9)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current9_;
-      if (current9_ == end9_) {
-        current9_ = begin9_;
-        ++current8_;
-      }
-      if (current8_ == end8_) {
-        current8_ = begin8_;
-        ++current7_;
-      }
-      if (current7_ == end7_) {
-        current7_ = begin7_;
-        ++current6_;
-      }
-      if (current6_ == end6_) {
-        current6_ = begin6_;
-        ++current5_;
-      }
-      if (current5_ == end5_) {
-        current5_ = begin5_;
-        ++current4_;
-      }
-      if (current4_ == end4_) {
-        current4_ = begin4_;
-        ++current3_;
-      }
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_ &&
-          current4_ == typed_other->current4_ &&
-          current5_ == typed_other->current5_ &&
-          current6_ == typed_other->current6_ &&
-          current7_ == typed_other->current7_ &&
-          current8_ == typed_other->current8_ &&
-          current9_ == typed_other->current9_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_),
-        begin4_(other.begin4_),
-        end4_(other.end4_),
-        current4_(other.current4_),
-        begin5_(other.begin5_),
-        end5_(other.end5_),
-        current5_(other.current5_),
-        begin6_(other.begin6_),
-        end6_(other.end6_),
-        current6_(other.current6_),
-        begin7_(other.begin7_),
-        end7_(other.end7_),
-        current7_(other.current7_),
-        begin8_(other.begin8_),
-        end8_(other.end8_),
-        current8_(other.current8_),
-        begin9_(other.begin9_),
-        end9_(other.end9_),
-        current9_(other.current9_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_,
-            *current4_, *current5_, *current6_, *current7_, *current8_,
-            *current9_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_ ||
-          current4_ == end4_ ||
-          current5_ == end5_ ||
-          current6_ == end6_ ||
-          current7_ == end7_ ||
-          current8_ == end8_ ||
-          current9_ == end9_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    const typename ParamGenerator<T4>::iterator begin4_;
-    const typename ParamGenerator<T4>::iterator end4_;
-    typename ParamGenerator<T4>::iterator current4_;
-    const typename ParamGenerator<T5>::iterator begin5_;
-    const typename ParamGenerator<T5>::iterator end5_;
-    typename ParamGenerator<T5>::iterator current5_;
-    const typename ParamGenerator<T6>::iterator begin6_;
-    const typename ParamGenerator<T6>::iterator end6_;
-    typename ParamGenerator<T6>::iterator current6_;
-    const typename ParamGenerator<T7>::iterator begin7_;
-    const typename ParamGenerator<T7>::iterator end7_;
-    typename ParamGenerator<T7>::iterator current7_;
-    const typename ParamGenerator<T8>::iterator begin8_;
-    const typename ParamGenerator<T8>::iterator end8_;
-    typename ParamGenerator<T8>::iterator current8_;
-    const typename ParamGenerator<T9>::iterator begin9_;
-    const typename ParamGenerator<T9>::iterator end9_;
-    typename ParamGenerator<T9>::iterator current9_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator9::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator9& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-  const ParamGenerator<T4> g4_;
-  const ParamGenerator<T5> g5_;
-  const ParamGenerator<T6> g6_;
-  const ParamGenerator<T7> g7_;
-  const ParamGenerator<T8> g8_;
-  const ParamGenerator<T9> g9_;
-};  // class CartesianProductGenerator9
-
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10>
-class CartesianProductGenerator10
-    : public ParamGeneratorInterface< ::testing::tuple<T1, T2, T3, T4, T5, T6,
-        T7, T8, T9, T10> > {
- public:
-  typedef ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10> ParamType;
-
-  CartesianProductGenerator10(const ParamGenerator<T1>& g1,
-      const ParamGenerator<T2>& g2, const ParamGenerator<T3>& g3,
-      const ParamGenerator<T4>& g4, const ParamGenerator<T5>& g5,
-      const ParamGenerator<T6>& g6, const ParamGenerator<T7>& g7,
-      const ParamGenerator<T8>& g8, const ParamGenerator<T9>& g9,
-      const ParamGenerator<T10>& g10)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7), g8_(g8),
-          g9_(g9), g10_(g10) {}
-  virtual ~CartesianProductGenerator10() {}
-
-  virtual ParamIteratorInterface<ParamType>* Begin() const {
-    return new Iterator(this, g1_, g1_.begin(), g2_, g2_.begin(), g3_,
-        g3_.begin(), g4_, g4_.begin(), g5_, g5_.begin(), g6_, g6_.begin(), g7_,
-        g7_.begin(), g8_, g8_.begin(), g9_, g9_.begin(), g10_, g10_.begin());
-  }
-  virtual ParamIteratorInterface<ParamType>* End() const {
-    return new Iterator(this, g1_, g1_.end(), g2_, g2_.end(), g3_, g3_.end(),
-        g4_, g4_.end(), g5_, g5_.end(), g6_, g6_.end(), g7_, g7_.end(), g8_,
-        g8_.end(), g9_, g9_.end(), g10_, g10_.end());
-  }
-
- private:
-  class Iterator : public ParamIteratorInterface<ParamType> {
-   public:
-    Iterator(const ParamGeneratorInterface<ParamType>* base,
-      const ParamGenerator<T1>& g1,
-      const typename ParamGenerator<T1>::iterator& current1,
-      const ParamGenerator<T2>& g2,
-      const typename ParamGenerator<T2>::iterator& current2,
-      const ParamGenerator<T3>& g3,
-      const typename ParamGenerator<T3>::iterator& current3,
-      const ParamGenerator<T4>& g4,
-      const typename ParamGenerator<T4>::iterator& current4,
-      const ParamGenerator<T5>& g5,
-      const typename ParamGenerator<T5>::iterator& current5,
-      const ParamGenerator<T6>& g6,
-      const typename ParamGenerator<T6>::iterator& current6,
-      const ParamGenerator<T7>& g7,
-      const typename ParamGenerator<T7>::iterator& current7,
-      const ParamGenerator<T8>& g8,
-      const typename ParamGenerator<T8>::iterator& current8,
-      const ParamGenerator<T9>& g9,
-      const typename ParamGenerator<T9>::iterator& current9,
-      const ParamGenerator<T10>& g10,
-      const typename ParamGenerator<T10>::iterator& current10)
-        : base_(base),
-          begin1_(g1.begin()), end1_(g1.end()), current1_(current1),
-          begin2_(g2.begin()), end2_(g2.end()), current2_(current2),
-          begin3_(g3.begin()), end3_(g3.end()), current3_(current3),
-          begin4_(g4.begin()), end4_(g4.end()), current4_(current4),
-          begin5_(g5.begin()), end5_(g5.end()), current5_(current5),
-          begin6_(g6.begin()), end6_(g6.end()), current6_(current6),
-          begin7_(g7.begin()), end7_(g7.end()), current7_(current7),
-          begin8_(g8.begin()), end8_(g8.end()), current8_(current8),
-          begin9_(g9.begin()), end9_(g9.end()), current9_(current9),
-          begin10_(g10.begin()), end10_(g10.end()), current10_(current10)    {
-      ComputeCurrentValue();
-    }
-    virtual ~Iterator() {}
-
-    virtual const ParamGeneratorInterface<ParamType>* BaseGenerator() const {
-      return base_;
-    }
-    // Advance should not be called on beyond-of-range iterators
-    // so no component iterators must be beyond end of range, either.
-    virtual void Advance() {
-      assert(!AtEnd());
-      ++current10_;
-      if (current10_ == end10_) {
-        current10_ = begin10_;
-        ++current9_;
-      }
-      if (current9_ == end9_) {
-        current9_ = begin9_;
-        ++current8_;
-      }
-      if (current8_ == end8_) {
-        current8_ = begin8_;
-        ++current7_;
-      }
-      if (current7_ == end7_) {
-        current7_ = begin7_;
-        ++current6_;
-      }
-      if (current6_ == end6_) {
-        current6_ = begin6_;
-        ++current5_;
-      }
-      if (current5_ == end5_) {
-        current5_ = begin5_;
-        ++current4_;
-      }
-      if (current4_ == end4_) {
-        current4_ = begin4_;
-        ++current3_;
-      }
-      if (current3_ == end3_) {
-        current3_ = begin3_;
-        ++current2_;
-      }
-      if (current2_ == end2_) {
-        current2_ = begin2_;
-        ++current1_;
-      }
-      ComputeCurrentValue();
-    }
-    virtual ParamIteratorInterface<ParamType>* Clone() const {
-      return new Iterator(*this);
-    }
-    virtual const ParamType* Current() const { return &current_value_; }
-    virtual bool Equals(const ParamIteratorInterface<ParamType>& other) const {
-      // Having the same base generator guarantees that the other
-      // iterator is of the same type and we can downcast.
-      GTEST_CHECK_(BaseGenerator() == other.BaseGenerator())
-          << "The program attempted to compare iterators "
-          << "from different generators." << std::endl;
-      const Iterator* typed_other =
-          CheckedDowncastToActualType<const Iterator>(&other);
-      // We must report iterators equal if they both point beyond their
-      // respective ranges. That can happen in a variety of fashions,
-      // so we have to consult AtEnd().
-      return (AtEnd() && typed_other->AtEnd()) ||
-         (
-          current1_ == typed_other->current1_ &&
-          current2_ == typed_other->current2_ &&
-          current3_ == typed_other->current3_ &&
-          current4_ == typed_other->current4_ &&
-          current5_ == typed_other->current5_ &&
-          current6_ == typed_other->current6_ &&
-          current7_ == typed_other->current7_ &&
-          current8_ == typed_other->current8_ &&
-          current9_ == typed_other->current9_ &&
-          current10_ == typed_other->current10_);
-    }
-
-   private:
-    Iterator(const Iterator& other)
-        : base_(other.base_),
-        begin1_(other.begin1_),
-        end1_(other.end1_),
-        current1_(other.current1_),
-        begin2_(other.begin2_),
-        end2_(other.end2_),
-        current2_(other.current2_),
-        begin3_(other.begin3_),
-        end3_(other.end3_),
-        current3_(other.current3_),
-        begin4_(other.begin4_),
-        end4_(other.end4_),
-        current4_(other.current4_),
-        begin5_(other.begin5_),
-        end5_(other.end5_),
-        current5_(other.current5_),
-        begin6_(other.begin6_),
-        end6_(other.end6_),
-        current6_(other.current6_),
-        begin7_(other.begin7_),
-        end7_(other.end7_),
-        current7_(other.current7_),
-        begin8_(other.begin8_),
-        end8_(other.end8_),
-        current8_(other.current8_),
-        begin9_(other.begin9_),
-        end9_(other.end9_),
-        current9_(other.current9_),
-        begin10_(other.begin10_),
-        end10_(other.end10_),
-        current10_(other.current10_) {
-      ComputeCurrentValue();
-    }
-
-    void ComputeCurrentValue() {
-      if (!AtEnd())
-        current_value_ = ParamType(*current1_, *current2_, *current3_,
-            *current4_, *current5_, *current6_, *current7_, *current8_,
-            *current9_, *current10_);
-    }
-    bool AtEnd() const {
-      // We must report iterator past the end of the range when either of the
-      // component iterators has reached the end of its range.
-      return
-          current1_ == end1_ ||
-          current2_ == end2_ ||
-          current3_ == end3_ ||
-          current4_ == end4_ ||
-          current5_ == end5_ ||
-          current6_ == end6_ ||
-          current7_ == end7_ ||
-          current8_ == end8_ ||
-          current9_ == end9_ ||
-          current10_ == end10_;
-    }
-
-    // No implementation - assignment is unsupported.
-    void operator=(const Iterator& other);
-
-    const ParamGeneratorInterface<ParamType>* const base_;
-    // begin[i]_ and end[i]_ define the i-th range that Iterator traverses.
-    // current[i]_ is the actual traversing iterator.
-    const typename ParamGenerator<T1>::iterator begin1_;
-    const typename ParamGenerator<T1>::iterator end1_;
-    typename ParamGenerator<T1>::iterator current1_;
-    const typename ParamGenerator<T2>::iterator begin2_;
-    const typename ParamGenerator<T2>::iterator end2_;
-    typename ParamGenerator<T2>::iterator current2_;
-    const typename ParamGenerator<T3>::iterator begin3_;
-    const typename ParamGenerator<T3>::iterator end3_;
-    typename ParamGenerator<T3>::iterator current3_;
-    const typename ParamGenerator<T4>::iterator begin4_;
-    const typename ParamGenerator<T4>::iterator end4_;
-    typename ParamGenerator<T4>::iterator current4_;
-    const typename ParamGenerator<T5>::iterator begin5_;
-    const typename ParamGenerator<T5>::iterator end5_;
-    typename ParamGenerator<T5>::iterator current5_;
-    const typename ParamGenerator<T6>::iterator begin6_;
-    const typename ParamGenerator<T6>::iterator end6_;
-    typename ParamGenerator<T6>::iterator current6_;
-    const typename ParamGenerator<T7>::iterator begin7_;
-    const typename ParamGenerator<T7>::iterator end7_;
-    typename ParamGenerator<T7>::iterator current7_;
-    const typename ParamGenerator<T8>::iterator begin8_;
-    const typename ParamGenerator<T8>::iterator end8_;
-    typename ParamGenerator<T8>::iterator current8_;
-    const typename ParamGenerator<T9>::iterator begin9_;
-    const typename ParamGenerator<T9>::iterator end9_;
-    typename ParamGenerator<T9>::iterator current9_;
-    const typename ParamGenerator<T10>::iterator begin10_;
-    const typename ParamGenerator<T10>::iterator end10_;
-    typename ParamGenerator<T10>::iterator current10_;
-    ParamType current_value_;
-  };  // class CartesianProductGenerator10::Iterator
-
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductGenerator10& other);
-
-  const ParamGenerator<T1> g1_;
-  const ParamGenerator<T2> g2_;
-  const ParamGenerator<T3> g3_;
-  const ParamGenerator<T4> g4_;
-  const ParamGenerator<T5> g5_;
-  const ParamGenerator<T6> g6_;
-  const ParamGenerator<T7> g7_;
-  const ParamGenerator<T8> g8_;
-  const ParamGenerator<T9> g9_;
-  const ParamGenerator<T10> g10_;
-};  // class CartesianProductGenerator10
-
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Helper classes providing Combine() with polymorphic features. They allow
-// casting CartesianProductGeneratorN<T> to ParamGenerator<U> if T is
-// convertible to U.
-//
-template <class Generator1, class Generator2>
-class CartesianProductHolder2 {
- public:
-CartesianProductHolder2(const Generator1& g1, const Generator2& g2)
-      : g1_(g1), g2_(g2) {}
-  template <typename T1, typename T2>
-  operator ParamGenerator< ::testing::tuple<T1, T2> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2> >(
-        new CartesianProductGenerator2<T1, T2>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder2& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-};  // class CartesianProductHolder2
-
-template <class Generator1, class Generator2, class Generator3>
-class CartesianProductHolder3 {
- public:
-CartesianProductHolder3(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3)
-      : g1_(g1), g2_(g2), g3_(g3) {}
-  template <typename T1, typename T2, typename T3>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3> >(
-        new CartesianProductGenerator3<T1, T2, T3>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder3& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-};  // class CartesianProductHolder3
-
-template <class Generator1, class Generator2, class Generator3,
-    class Generator4>
-class CartesianProductHolder4 {
- public:
-CartesianProductHolder4(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3, const Generator4& g4)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4) {}
-  template <typename T1, typename T2, typename T3, typename T4>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3, T4> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3, T4> >(
-        new CartesianProductGenerator4<T1, T2, T3, T4>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_),
-        static_cast<ParamGenerator<T4> >(g4_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder4& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-  const Generator4 g4_;
-};  // class CartesianProductHolder4
-
-template <class Generator1, class Generator2, class Generator3,
-    class Generator4, class Generator5>
-class CartesianProductHolder5 {
- public:
-CartesianProductHolder5(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3, const Generator4& g4, const Generator5& g5)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5) {}
-  template <typename T1, typename T2, typename T3, typename T4, typename T5>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5> >(
-        new CartesianProductGenerator5<T1, T2, T3, T4, T5>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_),
-        static_cast<ParamGenerator<T4> >(g4_),
-        static_cast<ParamGenerator<T5> >(g5_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder5& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-  const Generator4 g4_;
-  const Generator5 g5_;
-};  // class CartesianProductHolder5
-
-template <class Generator1, class Generator2, class Generator3,
-    class Generator4, class Generator5, class Generator6>
-class CartesianProductHolder6 {
- public:
-CartesianProductHolder6(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3, const Generator4& g4, const Generator5& g5,
-    const Generator6& g6)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6) {}
-  template <typename T1, typename T2, typename T3, typename T4, typename T5,
-      typename T6>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6> >(
-        new CartesianProductGenerator6<T1, T2, T3, T4, T5, T6>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_),
-        static_cast<ParamGenerator<T4> >(g4_),
-        static_cast<ParamGenerator<T5> >(g5_),
-        static_cast<ParamGenerator<T6> >(g6_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder6& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-  const Generator4 g4_;
-  const Generator5 g5_;
-  const Generator6 g6_;
-};  // class CartesianProductHolder6
-
-template <class Generator1, class Generator2, class Generator3,
-    class Generator4, class Generator5, class Generator6, class Generator7>
-class CartesianProductHolder7 {
- public:
-CartesianProductHolder7(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3, const Generator4& g4, const Generator5& g5,
-    const Generator6& g6, const Generator7& g7)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7) {}
-  template <typename T1, typename T2, typename T3, typename T4, typename T5,
-      typename T6, typename T7>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6,
-      T7> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6, T7> >(
-        new CartesianProductGenerator7<T1, T2, T3, T4, T5, T6, T7>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_),
-        static_cast<ParamGenerator<T4> >(g4_),
-        static_cast<ParamGenerator<T5> >(g5_),
-        static_cast<ParamGenerator<T6> >(g6_),
-        static_cast<ParamGenerator<T7> >(g7_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder7& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-  const Generator4 g4_;
-  const Generator5 g5_;
-  const Generator6 g6_;
-  const Generator7 g7_;
-};  // class CartesianProductHolder7
-
-template <class Generator1, class Generator2, class Generator3,
-    class Generator4, class Generator5, class Generator6, class Generator7,
-    class Generator8>
-class CartesianProductHolder8 {
- public:
-CartesianProductHolder8(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3, const Generator4& g4, const Generator5& g5,
-    const Generator6& g6, const Generator7& g7, const Generator8& g8)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7),
-          g8_(g8) {}
-  template <typename T1, typename T2, typename T3, typename T4, typename T5,
-      typename T6, typename T7, typename T8>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6, T7,
-      T8> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8> >(
-        new CartesianProductGenerator8<T1, T2, T3, T4, T5, T6, T7, T8>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_),
-        static_cast<ParamGenerator<T4> >(g4_),
-        static_cast<ParamGenerator<T5> >(g5_),
-        static_cast<ParamGenerator<T6> >(g6_),
-        static_cast<ParamGenerator<T7> >(g7_),
-        static_cast<ParamGenerator<T8> >(g8_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder8& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-  const Generator4 g4_;
-  const Generator5 g5_;
-  const Generator6 g6_;
-  const Generator7 g7_;
-  const Generator8 g8_;
-};  // class CartesianProductHolder8
-
-template <class Generator1, class Generator2, class Generator3,
-    class Generator4, class Generator5, class Generator6, class Generator7,
-    class Generator8, class Generator9>
-class CartesianProductHolder9 {
- public:
-CartesianProductHolder9(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3, const Generator4& g4, const Generator5& g5,
-    const Generator6& g6, const Generator7& g7, const Generator8& g8,
-    const Generator9& g9)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7), g8_(g8),
-          g9_(g9) {}
-  template <typename T1, typename T2, typename T3, typename T4, typename T5,
-      typename T6, typename T7, typename T8, typename T9>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8,
-      T9> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8,
-        T9> >(
-        new CartesianProductGenerator9<T1, T2, T3, T4, T5, T6, T7, T8, T9>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_),
-        static_cast<ParamGenerator<T4> >(g4_),
-        static_cast<ParamGenerator<T5> >(g5_),
-        static_cast<ParamGenerator<T6> >(g6_),
-        static_cast<ParamGenerator<T7> >(g7_),
-        static_cast<ParamGenerator<T8> >(g8_),
-        static_cast<ParamGenerator<T9> >(g9_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder9& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-  const Generator4 g4_;
-  const Generator5 g5_;
-  const Generator6 g6_;
-  const Generator7 g7_;
-  const Generator8 g8_;
-  const Generator9 g9_;
-};  // class CartesianProductHolder9
-
-template <class Generator1, class Generator2, class Generator3,
-    class Generator4, class Generator5, class Generator6, class Generator7,
-    class Generator8, class Generator9, class Generator10>
-class CartesianProductHolder10 {
- public:
-CartesianProductHolder10(const Generator1& g1, const Generator2& g2,
-    const Generator3& g3, const Generator4& g4, const Generator5& g5,
-    const Generator6& g6, const Generator7& g7, const Generator8& g8,
-    const Generator9& g9, const Generator10& g10)
-      : g1_(g1), g2_(g2), g3_(g3), g4_(g4), g5_(g5), g6_(g6), g7_(g7), g8_(g8),
-          g9_(g9), g10_(g10) {}
-  template <typename T1, typename T2, typename T3, typename T4, typename T5,
-      typename T6, typename T7, typename T8, typename T9, typename T10>
-  operator ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9,
-      T10> >() const {
-    return ParamGenerator< ::testing::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9,
-        T10> >(
-        new CartesianProductGenerator10<T1, T2, T3, T4, T5, T6, T7, T8, T9,
-            T10>(
-        static_cast<ParamGenerator<T1> >(g1_),
-        static_cast<ParamGenerator<T2> >(g2_),
-        static_cast<ParamGenerator<T3> >(g3_),
-        static_cast<ParamGenerator<T4> >(g4_),
-        static_cast<ParamGenerator<T5> >(g5_),
-        static_cast<ParamGenerator<T6> >(g6_),
-        static_cast<ParamGenerator<T7> >(g7_),
-        static_cast<ParamGenerator<T8> >(g8_),
-        static_cast<ParamGenerator<T9> >(g9_),
-        static_cast<ParamGenerator<T10> >(g10_)));
-  }
-
- private:
-  // No implementation - assignment is unsupported.
-  void operator=(const CartesianProductHolder10& other);
-
-  const Generator1 g1_;
-  const Generator2 g2_;
-  const Generator3 g3_;
-  const Generator4 g4_;
-  const Generator5 g5_;
-  const Generator6 g6_;
-  const Generator7 g7_;
-  const Generator8 g8_;
-  const Generator9 g9_;
-  const Generator10 g10_;
-};  // class CartesianProductHolder10
-
-# endif  // GTEST_HAS_COMBINE
-
-}  // namespace internal
-}  // namespace testing
-
-#endif  //  GTEST_HAS_PARAM_TEST
-
-#endif  // GTEST_INCLUDE_GTEST_INTERNAL_GTEST_PARAM_UTIL_GENERATED_H_
-
-#if GTEST_HAS_PARAM_TEST
-
-namespace testing {
-
-// Functions producing parameter generators.
-//
-// Google Test uses these generators to produce parameters for value-
-// parameterized tests. When a parameterized test case is instantiated
-// with a particular generator, Google Test creates and runs tests
-// for each element in the sequence produced by the generator.
-//
-// In the following sample, tests from test case FooTest are instantiated
-// each three times with parameter values 3, 5, and 8:
-//
-// class FooTest : public TestWithParam<int> { ... };
-//
-// TEST_P(FooTest, TestThis) {
-// }
-// TEST_P(FooTest, TestThat) {
-// }
-// INSTANTIATE_TEST_CASE_P(TestSequence, FooTest, Values(3, 5, 8));
-//
-
-// Range() returns generators providing sequences of values in a range.
-//
-// Synopsis:
-// Range(start, end)
-//   - returns a generator producing a sequence of values {start, start+1,
-//     start+2, ..., }.
-// Range(start, end, step)
-//   - returns a generator producing a sequence of values {start, start+step,
-//     start+step+step, ..., }.
-// Notes:
-//   * The generated sequences never include end. For example, Range(1, 5)
-//     returns a generator producing a sequence {1, 2, 3, 4}. Range(1, 9, 2)
-//     returns a generator producing {1, 3, 5, 7}.
-//   * start and end must have the same type. That type may be any integral or
-//     floating-point type or a user defined type satisfying these conditions:
-//     * It must be assignable (have operator=() defined).
-//     * It must have operator+() (operator+(int-compatible type) for
-//       two-operand version).
-//     * It must have operator<() defined.
-//     Elements in the resulting sequences will also have that type.
-//   * Condition start < end must be satisfied in order for resulting sequences
-//     to contain any elements.
-//
-template <typename T, typename IncrementT>
-internal::ParamGenerator<T> Range(T start, T end, IncrementT step) {
-  return internal::ParamGenerator<T>(
-      new internal::RangeGenerator<T, IncrementT>(start, end, step));
-}
-
-template <typename T>
-internal::ParamGenerator<T> Range(T start, T end) {
-  return Range(start, end, 1);
-}
-
-// ValuesIn() function allows generation of tests with parameters coming from
-// a container.
-//
-// Synopsis:
-// ValuesIn(const T (&array)[N])
-//   - returns a generator producing sequences with elements from
-//     a C-style array.
-// ValuesIn(const Container& container)
-//   - returns a generator producing sequences with elements from
-//     an STL-style container.
-// ValuesIn(Iterator begin, Iterator end)
-//   - returns a generator producing sequences with elements from
-//     a range [begin, end) defined by a pair of STL-style iterators. These
-//     iterators can also be plain C pointers.
-//
-// Please note that ValuesIn copies the values from the containers
-// passed in and keeps them to generate tests in RUN_ALL_TESTS().
-//
-// Examples:
-//
-// This instantiates tests from test case StringTest
-// each with C-string values of "foo", "bar", and "baz":
-//
-// const char* strings[] = {"foo", "bar", "baz"};
-// INSTANTIATE_TEST_CASE_P(StringSequence, SrtingTest, ValuesIn(strings));
-//
-// This instantiates tests from test case StlStringTest
-// each with STL strings with values "a" and "b":
-//
-// ::std::vector< ::std::string> GetParameterStrings() {
-//   ::std::vector< ::std::string> v;
-//   v.push_back("a");
-//   v.push_back("b");
-//   return v;
-// }
-//
-// INSTANTIATE_TEST_CASE_P(CharSequence,
-//                         StlStringTest,
-//                         ValuesIn(GetParameterStrings()));
-//
-//
-// This will also instantiate tests from CharTest
-// each with parameter values 'a' and 'b':
-//
-// ::std::list<char> GetParameterChars() {
-//   ::std::list<char> list;
-//   list.push_back('a');
-//   list.push_back('b');
-//   return list;
-// }
-// ::std::list<char> l = GetParameterChars();
-// INSTANTIATE_TEST_CASE_P(CharSequence2,
-//                         CharTest,
-//                         ValuesIn(l.begin(), l.end()));
-//
-template <typename ForwardIterator>
-internal::ParamGenerator<
-  typename ::testing::internal::IteratorTraits<ForwardIterator>::value_type>
-ValuesIn(ForwardIterator begin, ForwardIterator end) {
-  typedef typename ::testing::internal::IteratorTraits<ForwardIterator>
-      ::value_type ParamType;
-  return internal::ParamGenerator<ParamType>(
-      new internal::ValuesInIteratorRangeGenerator<ParamType>(begin, end));
-}
-
-template <typename T, size_t N>
-internal::ParamGenerator<T> ValuesIn(const T (&array)[N]) {
-  return ValuesIn(array, array + N);
-}
-
-template <class Container>
-internal::ParamGenerator<typename Container::value_type> ValuesIn(
-    const Container& container) {
-  return ValuesIn(container.begin(), container.end());
-}
-
-// Values() allows generating tests from explicitly specified list of
-// parameters.
-//
-// Synopsis:
-// Values(T v1, T v2, ..., T vN)
-//   - returns a generator producing sequences with elements v1, v2, ..., vN.
-//
-// For example, this instantiates tests from test case BarTest each
-// with values "one", "two", and "three":
-//
-// INSTANTIATE_TEST_CASE_P(NumSequence, BarTest, Values("one", "two", "three"));
-//
-// This instantiates tests from test case BazTest each with values 1, 2, 3.5.
-// The exact type of values will depend on the type of parameter in BazTest.
-//
-// INSTANTIATE_TEST_CASE_P(FloatingNumbers, BazTest, Values(1, 2, 3.5));
-//
-// Currently, Values() supports from 1 to 50 parameters.
-//
-template <typename T1>
-internal::ValueArray1<T1> Values(T1 v1) {
-  return internal::ValueArray1<T1>(v1);
-}
-
-template <typename T1, typename T2>
-internal::ValueArray2<T1, T2> Values(T1 v1, T2 v2) {
-  return internal::ValueArray2<T1, T2>(v1, v2);
-}
-
-template <typename T1, typename T2, typename T3>
-internal::ValueArray3<T1, T2, T3> Values(T1 v1, T2 v2, T3 v3) {
-  return internal::ValueArray3<T1, T2, T3>(v1, v2, v3);
-}
-
-template <typename T1, typename T2, typename T3, typename T4>
-internal::ValueArray4<T1, T2, T3, T4> Values(T1 v1, T2 v2, T3 v3, T4 v4) {
-  return internal::ValueArray4<T1, T2, T3, T4>(v1, v2, v3, v4);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5>
-internal::ValueArray5<T1, T2, T3, T4, T5> Values(T1 v1, T2 v2, T3 v3, T4 v4,
-    T5 v5) {
-  return internal::ValueArray5<T1, T2, T3, T4, T5>(v1, v2, v3, v4, v5);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6>
-internal::ValueArray6<T1, T2, T3, T4, T5, T6> Values(T1 v1, T2 v2, T3 v3,
-    T4 v4, T5 v5, T6 v6) {
-  return internal::ValueArray6<T1, T2, T3, T4, T5, T6>(v1, v2, v3, v4, v5, v6);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7>
-internal::ValueArray7<T1, T2, T3, T4, T5, T6, T7> Values(T1 v1, T2 v2, T3 v3,
-    T4 v4, T5 v5, T6 v6, T7 v7) {
-  return internal::ValueArray7<T1, T2, T3, T4, T5, T6, T7>(v1, v2, v3, v4, v5,
-      v6, v7);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8>
-internal::ValueArray8<T1, T2, T3, T4, T5, T6, T7, T8> Values(T1 v1, T2 v2,
-    T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8) {
-  return internal::ValueArray8<T1, T2, T3, T4, T5, T6, T7, T8>(v1, v2, v3, v4,
-      v5, v6, v7, v8);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9>
-internal::ValueArray9<T1, T2, T3, T4, T5, T6, T7, T8, T9> Values(T1 v1, T2 v2,
-    T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9) {
-  return internal::ValueArray9<T1, T2, T3, T4, T5, T6, T7, T8, T9>(v1, v2, v3,
-      v4, v5, v6, v7, v8, v9);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10>
-internal::ValueArray10<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10> Values(T1 v1,
-    T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10) {
-  return internal::ValueArray10<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10>(v1,
-      v2, v3, v4, v5, v6, v7, v8, v9, v10);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11>
-internal::ValueArray11<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10,
-    T11> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11) {
-  return internal::ValueArray11<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10,
-      T11>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12>
-internal::ValueArray12<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-    T12> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12) {
-  return internal::ValueArray12<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13>
-internal::ValueArray13<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12,
-    T13> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13) {
-  return internal::ValueArray13<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14>
-internal::ValueArray14<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14) {
-  return internal::ValueArray14<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13,
-      v14);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15>
-internal::ValueArray15<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8,
-    T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15) {
-  return internal::ValueArray15<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12,
-      v13, v14, v15);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16>
-internal::ValueArray16<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7,
-    T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16) {
-  return internal::ValueArray16<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11,
-      v12, v13, v14, v15, v16);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17>
-internal::ValueArray17<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7,
-    T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17) {
-  return internal::ValueArray17<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10,
-      v11, v12, v13, v14, v15, v16, v17);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18>
-internal::ValueArray18<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6,
-    T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17, T18 v18) {
-  return internal::ValueArray18<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18>(v1, v2, v3, v4, v5, v6, v7, v8, v9,
-      v10, v11, v12, v13, v14, v15, v16, v17, v18);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19>
-internal::ValueArray19<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5,
-    T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14,
-    T15 v15, T16 v16, T17 v17, T18 v18, T19 v19) {
-  return internal::ValueArray19<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19>(v1, v2, v3, v4, v5, v6, v7, v8,
-      v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20>
-internal::ValueArray20<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20> Values(T1 v1, T2 v2, T3 v3, T4 v4,
-    T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13,
-    T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20) {
-  return internal::ValueArray20<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20>(v1, v2, v3, v4, v5, v6, v7,
-      v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21>
-internal::ValueArray21<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21> Values(T1 v1, T2 v2, T3 v3, T4 v4,
-    T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13,
-    T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21) {
-  return internal::ValueArray21<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21>(v1, v2, v3, v4, v5, v6,
-      v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22>
-internal::ValueArray22<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22> Values(T1 v1, T2 v2, T3 v3,
-    T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12,
-    T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20,
-    T21 v21, T22 v22) {
-  return internal::ValueArray22<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22>(v1, v2, v3, v4,
-      v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19,
-      v20, v21, v22);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23>
-internal::ValueArray23<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23> Values(T1 v1, T2 v2,
-    T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12,
-    T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20,
-    T21 v21, T22 v22, T23 v23) {
-  return internal::ValueArray23<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23>(v1, v2, v3,
-      v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19,
-      v20, v21, v22, v23);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24>
-internal::ValueArray24<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24> Values(T1 v1, T2 v2,
-    T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12,
-    T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20,
-    T21 v21, T22 v22, T23 v23, T24 v24) {
-  return internal::ValueArray24<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24>(v1, v2,
-      v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18,
-      v19, v20, v21, v22, v23, v24);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25>
-internal::ValueArray25<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25> Values(T1 v1,
-    T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11,
-    T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19,
-    T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25) {
-  return internal::ValueArray25<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25>(v1,
-      v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17,
-      v18, v19, v20, v21, v22, v23, v24, v25);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26>
-internal::ValueArray26<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-    T26> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26) {
-  return internal::ValueArray26<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15,
-      v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27>
-internal::ValueArray27<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26,
-    T27> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26, T27 v27) {
-  return internal::ValueArray27<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14,
-      v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26, v27);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28>
-internal::ValueArray28<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27,
-    T28> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26, T27 v27, T28 v28) {
-  return internal::ValueArray28<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13,
-      v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26, v27,
-      v28);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29>
-internal::ValueArray29<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26, T27 v27, T28 v28, T29 v29) {
-  return internal::ValueArray29<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12,
-      v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26,
-      v27, v28, v29);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30>
-internal::ValueArray30<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8,
-    T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16,
-    T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24,
-    T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30) {
-  return internal::ValueArray30<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11,
-      v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25,
-      v26, v27, v28, v29, v30);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31>
-internal::ValueArray31<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7,
-    T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23,
-    T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31) {
-  return internal::ValueArray31<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10,
-      v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24,
-      v25, v26, v27, v28, v29, v30, v31);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32>
-internal::ValueArray32<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7,
-    T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23,
-    T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31,
-    T32 v32) {
-  return internal::ValueArray32<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32>(v1, v2, v3, v4, v5, v6, v7, v8, v9,
-      v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23,
-      v24, v25, v26, v27, v28, v29, v30, v31, v32);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33>
-internal::ValueArray33<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6,
-    T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23,
-    T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31,
-    T32 v32, T33 v33) {
-  return internal::ValueArray33<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33>(v1, v2, v3, v4, v5, v6, v7, v8,
-      v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23,
-      v24, v25, v26, v27, v28, v29, v30, v31, v32, v33);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34>
-internal::ValueArray34<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5,
-    T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14,
-    T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22,
-    T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30,
-    T31 v31, T32 v32, T33 v33, T34 v34) {
-  return internal::ValueArray34<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34>(v1, v2, v3, v4, v5, v6, v7,
-      v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22,
-      v23, v24, v25, v26, v27, v28, v29, v30, v31, v32, v33, v34);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35>
-internal::ValueArray35<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35> Values(T1 v1, T2 v2, T3 v3, T4 v4,
-    T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13,
-    T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21,
-    T22 v22, T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29,
-    T30 v30, T31 v31, T32 v32, T33 v33, T34 v34, T35 v35) {
-  return internal::ValueArray35<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35>(v1, v2, v3, v4, v5, v6,
-      v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21,
-      v22, v23, v24, v25, v26, v27, v28, v29, v30, v31, v32, v33, v34, v35);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36>
-internal::ValueArray36<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36> Values(T1 v1, T2 v2, T3 v3, T4 v4,
-    T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13,
-    T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21,
-    T22 v22, T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29,
-    T30 v30, T31 v31, T32 v32, T33 v33, T34 v34, T35 v35, T36 v36) {
-  return internal::ValueArray36<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36>(v1, v2, v3, v4,
-      v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19,
-      v20, v21, v22, v23, v24, v25, v26, v27, v28, v29, v30, v31, v32, v33,
-      v34, v35, v36);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37>
-internal::ValueArray37<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37> Values(T1 v1, T2 v2, T3 v3,
-    T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12,
-    T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20,
-    T21 v21, T22 v22, T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28,
-    T29 v29, T30 v30, T31 v31, T32 v32, T33 v33, T34 v34, T35 v35, T36 v36,
-    T37 v37) {
-  return internal::ValueArray37<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37>(v1, v2, v3,
-      v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19,
-      v20, v21, v22, v23, v24, v25, v26, v27, v28, v29, v30, v31, v32, v33,
-      v34, v35, v36, v37);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38>
-internal::ValueArray38<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38> Values(T1 v1, T2 v2,
-    T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12,
-    T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20,
-    T21 v21, T22 v22, T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28,
-    T29 v29, T30 v30, T31 v31, T32 v32, T33 v33, T34 v34, T35 v35, T36 v36,
-    T37 v37, T38 v38) {
-  return internal::ValueArray38<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38>(v1, v2,
-      v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18,
-      v19, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29, v30, v31, v32,
-      v33, v34, v35, v36, v37, v38);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39>
-internal::ValueArray39<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39> Values(T1 v1, T2 v2,
-    T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12,
-    T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20,
-    T21 v21, T22 v22, T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28,
-    T29 v29, T30 v30, T31 v31, T32 v32, T33 v33, T34 v34, T35 v35, T36 v36,
-    T37 v37, T38 v38, T39 v39) {
-  return internal::ValueArray39<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39>(v1,
-      v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17,
-      v18, v19, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29, v30, v31,
-      v32, v33, v34, v35, v36, v37, v38, v39);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40>
-internal::ValueArray40<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40> Values(T1 v1,
-    T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11,
-    T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19,
-    T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25, T26 v26, T27 v27,
-    T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33, T34 v34, T35 v35,
-    T36 v36, T37 v37, T38 v38, T39 v39, T40 v40) {
-  return internal::ValueArray40<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15,
-      v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29,
-      v30, v31, v32, v33, v34, v35, v36, v37, v38, v39, v40);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41>
-internal::ValueArray41<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40,
-    T41> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-    T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41) {
-  return internal::ValueArray41<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14,
-      v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26, v27, v28,
-      v29, v30, v31, v32, v33, v34, v35, v36, v37, v38, v39, v40, v41);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42>
-internal::ValueArray42<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41,
-    T42> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-    T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-    T42 v42) {
-  return internal::ValueArray42<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13,
-      v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26, v27,
-      v28, v29, v30, v31, v32, v33, v34, v35, v36, v37, v38, v39, v40, v41,
-      v42);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43>
-internal::ValueArray43<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42,
-    T43> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-    T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-    T42 v42, T43 v43) {
-  return internal::ValueArray43<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12,
-      v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26,
-      v27, v28, v29, v30, v31, v32, v33, v34, v35, v36, v37, v38, v39, v40,
-      v41, v42, v43);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44>
-internal::ValueArray44<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8, T9 v9,
-    T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
-    T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24, T25 v25,
-    T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32, T33 v33,
-    T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40, T41 v41,
-    T42 v42, T43 v43, T44 v44) {
-  return internal::ValueArray44<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43, T44>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11,
-      v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24, v25,
-      v26, v27, v28, v29, v30, v31, v32, v33, v34, v35, v36, v37, v38, v39,
-      v40, v41, v42, v43, v44);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45>
-internal::ValueArray45<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44, T45> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7, T8 v8,
-    T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16,
-    T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23, T24 v24,
-    T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31, T32 v32,
-    T33 v33, T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39, T40 v40,
-    T41 v41, T42 v42, T43 v43, T44 v44, T45 v45) {
-  return internal::ValueArray45<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43, T44, T45>(v1, v2, v3, v4, v5, v6, v7, v8, v9, v10,
-      v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23, v24,
-      v25, v26, v27, v28, v29, v30, v31, v32, v33, v34, v35, v36, v37, v38,
-      v39, v40, v41, v42, v43, v44, v45);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46>
-internal::ValueArray46<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44, T45, T46> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7,
-    T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23,
-    T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31,
-    T32 v32, T33 v33, T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39,
-    T40 v40, T41 v41, T42 v42, T43 v43, T44 v44, T45 v45, T46 v46) {
-  return internal::ValueArray46<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43, T44, T45, T46>(v1, v2, v3, v4, v5, v6, v7, v8, v9,
-      v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23,
-      v24, v25, v26, v27, v28, v29, v30, v31, v32, v33, v34, v35, v36, v37,
-      v38, v39, v40, v41, v42, v43, v44, v45, v46);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47>
-internal::ValueArray47<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44, T45, T46, T47> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6, T7 v7,
-    T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23,
-    T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31,
-    T32 v32, T33 v33, T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39,
-    T40 v40, T41 v41, T42 v42, T43 v43, T44 v44, T45 v45, T46 v46, T47 v47) {
-  return internal::ValueArray47<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43, T44, T45, T46, T47>(v1, v2, v3, v4, v5, v6, v7, v8,
-      v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22, v23,
-      v24, v25, v26, v27, v28, v29, v30, v31, v32, v33, v34, v35, v36, v37,
-      v38, v39, v40, v41, v42, v43, v44, v45, v46, v47);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48>
-internal::ValueArray48<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44, T45, T46, T47, T48> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5, T6 v6,
-    T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15,
-    T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22, T23 v23,
-    T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30, T31 v31,
-    T32 v32, T33 v33, T34 v34, T35 v35, T36 v36, T37 v37, T38 v38, T39 v39,
-    T40 v40, T41 v41, T42 v42, T43 v43, T44 v44, T45 v45, T46 v46, T47 v47,
-    T48 v48) {
-  return internal::ValueArray48<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43, T44, T45, T46, T47, T48>(v1, v2, v3, v4, v5, v6, v7,
-      v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, v22,
-      v23, v24, v25, v26, v27, v28, v29, v30, v31, v32, v33, v34, v35, v36,
-      v37, v38, v39, v40, v41, v42, v43, v44, v45, v46, v47, v48);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49>
-internal::ValueArray49<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44, T45, T46, T47, T48, T49> Values(T1 v1, T2 v2, T3 v3, T4 v4, T5 v5,
-    T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14,
-    T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21, T22 v22,
-    T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29, T30 v30,
-    T31 v31, T32 v32, T33 v33, T34 v34, T35 v35, T36 v36, T37 v37, T38 v38,
-    T39 v39, T40 v40, T41 v41, T42 v42, T43 v43, T44 v44, T45 v45, T46 v46,
-    T47 v47, T48 v48, T49 v49) {
-  return internal::ValueArray49<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43, T44, T45, T46, T47, T48, T49>(v1, v2, v3, v4, v5, v6,
-      v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19, v20, v21,
-      v22, v23, v24, v25, v26, v27, v28, v29, v30, v31, v32, v33, v34, v35,
-      v36, v37, v38, v39, v40, v41, v42, v43, v44, v45, v46, v47, v48, v49);
-}
-
-template <typename T1, typename T2, typename T3, typename T4, typename T5,
-    typename T6, typename T7, typename T8, typename T9, typename T10,
-    typename T11, typename T12, typename T13, typename T14, typename T15,
-    typename T16, typename T17, typename T18, typename T19, typename T20,
-    typename T21, typename T22, typename T23, typename T24, typename T25,
-    typename T26, typename T27, typename T28, typename T29, typename T30,
-    typename T31, typename T32, typename T33, typename T34, typename T35,
-    typename T36, typename T37, typename T38, typename T39, typename T40,
-    typename T41, typename T42, typename T43, typename T44, typename T45,
-    typename T46, typename T47, typename T48, typename T49, typename T50>
-internal::ValueArray50<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13,
-    T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25, T26, T27, T28,
-    T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39, T40, T41, T42, T43,
-    T44, T45, T46, T47, T48, T49, T50> Values(T1 v1, T2 v2, T3 v3, T4 v4,
-    T5 v5, T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13,
-    T14 v14, T15 v15, T16 v16, T17 v17, T18 v18, T19 v19, T20 v20, T21 v21,
-    T22 v22, T23 v23, T24 v24, T25 v25, T26 v26, T27 v27, T28 v28, T29 v29,
-    T30 v30, T31 v31, T32 v32, T33 v33, T34 v34, T35 v35, T36 v36, T37 v37,
-    T38 v38, T39 v39, T40 v40, T41 v41, T42 v42, T43 v43, T44 v44, T45 v45,
-    T46 v46, T47 v47, T48 v48, T49 v49, T50 v50) {
-  return internal::ValueArray50<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
-      T12, T13, T14, T15, T16, T17, T18, T19, T20, T21, T22, T23, T24, T25,
-      T26, T27, T28, T29, T30, T31, T32, T33, T34, T35, T36, T37, T38, T39,
-      T40, T41, T42, T43, T44, T45, T46, T47, T48, T49, T50>(v1, v2, v3, v4,
-      v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19,
-      v20, v21, v22, v23, v24, v25, v26, v27, v28, v29, v30, v31, v32, v33,
-      v34, v35, v36, v37, v38, v39, v40, v41, v42, v43, v44, v45, v46, v47,
-      v48, v49, v50);
-}
-
-// Bool() allows generating tests with parameters in a set of (false, true).
-//
-// Synopsis:
-// Bool()
-//   - returns a generator producing sequences with elements {false, true}.
-//
-// It is useful when testing code that depends on Boolean flags. Combinations
-// of multiple flags can be tested when several Bool()'s are combined using
-// Combine() function.
-//
-// In the following example all tests in the test case FlagDependentTest
-// will be instantiated twice with parameters false and true.
-//
-// class FlagDependentTest : public testing::TestWithParam<bool> {
-//   virtual void SetUp() {
-//     external_flag = GetParam();
-//   }
-// }
-// INSTANTIATE_TEST_CASE_P(BoolSequence, FlagDependentTest, Bool());
-//
-inline internal::ParamGenerator<bool> Bool() {
-  return Values(false, true);
-}
-
-# if GTEST_HAS_COMBINE
-// Combine() allows the user to combine two or more sequences to produce
-// values of a Cartesian product of those sequences' elements.
-//
-// Synopsis:
-// Combine(gen1, gen2, ..., genN)
-//   - returns a generator producing sequences with elements coming from
-//     the Cartesian product of elements from the sequences generated by
-//     gen1, gen2, ..., genN. The sequence elements will have a type of
-//     tuple<T1, T2, ..., TN> where T1, T2, ..., TN are the types
-//     of elements from sequences produces by gen1, gen2, ..., genN.
-//
-// Combine can have up to 10 arguments. This number is currently limited
-// by the maximum number of elements in the tuple implementation used by Google
-// Test.
-//
-// Example:
-//
-// This will instantiate tests in test case AnimalTest each one with
-// the parameter values tuple("cat", BLACK), tuple("cat", WHITE),
-// tuple("dog", BLACK), and tuple("dog", WHITE):
-//
-// enum Color { BLACK, GRAY, WHITE };
-// class AnimalTest
-//     : public testing::TestWithParam<tuple<const char*, Color> > {...};
-//
-// TEST_P(AnimalTest, AnimalLooksNice) {...}
-//
-// INSTANTIATE_TEST_CASE_P(AnimalVariations, AnimalTest,
-//                         Combine(Values("cat", "dog"),
-//                                 Values(BLACK, WHITE)));
-//
-// This will instantiate tests in FlagDependentTest with all variations of two
-// Boolean flags:
-//
-// class FlagDependentTest
-//     : public testing::TestWithParam<tuple<bool, bool> > {
-//   virtual void SetUp() {
-//     // Assigns external_flag_1 and external_flag_2 values from the tuple.
-//     tie(external_flag_1, external_flag_2) = GetParam();
-//   }
-// };
-//
-// TEST_P(FlagDependentTest, TestFeature1) {
-//   // Test your code using external_flag_1 and external_flag_2 here.
-// }
-// INSTANTIATE_TEST_CASE_P(TwoBoolSequence, FlagDependentTest,
-//                         Combine(Bool(), Bool()));
-//
-template <typename Generator1, typename Generator2>
-internal::CartesianProductHolder2<Generator1, Generator2> Combine(
-    const Generator1& g1, const Generator2& g2) {
-  return internal::CartesianProductHolder2<Generator1, Generator2>(
-      g1, g2);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3>
-internal::CartesianProductHolder3<Generator1, Generator2, Generator3> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3) {
-  return internal::CartesianProductHolder3<Generator1, Generator2, Generator3>(
-      g1, g2, g3);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3,
-    typename Generator4>
-internal::CartesianProductHolder4<Generator1, Generator2, Generator3,
-    Generator4> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3,
-        const Generator4& g4) {
-  return internal::CartesianProductHolder4<Generator1, Generator2, Generator3,
-      Generator4>(
-      g1, g2, g3, g4);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3,
-    typename Generator4, typename Generator5>
-internal::CartesianProductHolder5<Generator1, Generator2, Generator3,
-    Generator4, Generator5> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3,
-        const Generator4& g4, const Generator5& g5) {
-  return internal::CartesianProductHolder5<Generator1, Generator2, Generator3,
-      Generator4, Generator5>(
-      g1, g2, g3, g4, g5);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3,
-    typename Generator4, typename Generator5, typename Generator6>
-internal::CartesianProductHolder6<Generator1, Generator2, Generator3,
-    Generator4, Generator5, Generator6> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3,
-        const Generator4& g4, const Generator5& g5, const Generator6& g6) {
-  return internal::CartesianProductHolder6<Generator1, Generator2, Generator3,
-      Generator4, Generator5, Generator6>(
-      g1, g2, g3, g4, g5, g6);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3,
-    typename Generator4, typename Generator5, typename Generator6,
-    typename Generator7>
-internal::CartesianProductHolder7<Generator1, Generator2, Generator3,
-    Generator4, Generator5, Generator6, Generator7> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3,
-        const Generator4& g4, const Generator5& g5, const Generator6& g6,
-        const Generator7& g7) {
-  return internal::CartesianProductHolder7<Generator1, Generator2, Generator3,
-      Generator4, Generator5, Generator6, Generator7>(
-      g1, g2, g3, g4, g5, g6, g7);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3,
-    typename Generator4, typename Generator5, typename Generator6,
-    typename Generator7, typename Generator8>
-internal::CartesianProductHolder8<Generator1, Generator2, Generator3,
-    Generator4, Generator5, Generator6, Generator7, Generator8> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3,
-        const Generator4& g4, const Generator5& g5, const Generator6& g6,
-        const Generator7& g7, const Generator8& g8) {
-  return internal::CartesianProductHolder8<Generator1, Generator2, Generator3,
-      Generator4, Generator5, Generator6, Generator7, Generator8>(
-      g1, g2, g3, g4, g5, g6, g7, g8);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3,
-    typename Generator4, typename Generator5, typename Generator6,
-    typename Generator7, typename Generator8, typename Generator9>
-internal::CartesianProductHolder9<Generator1, Generator2, Generator3,
-    Generator4, Generator5, Generator6, Generator7, Generator8,
-    Generator9> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3,
-        const Generator4& g4, const Generator5& g5, const Generator6& g6,
-        const Generator7& g7, const Generator8& g8, const Generator9& g9) {
-  return internal::CartesianProductHolder9<Generator1, Generator2, Generator3,
-      Generator4, Generator5, Generator6, Generator7, Generator8, Generator9>(
-      g1, g2, g3, g4, g5, g6, g7, g8, g9);
-}
-
-template <typename Generator1, typename Generator2, typename Generator3,
-    typename Generator4, typename Generator5, typename Generator6,
-    typename Generator7, typename Generator8, typename Generator9,
-    typename Generator10>
-internal::CartesianProductHolder10<Generator1, Generator2, Generator3,
-    Generator4, Generator5, Generator6, Generator7, Generator8, Generator9,
-    Generator10> Combine(
-    const Generator1& g1, const Generator2& g2, const Generator3& g3,
-        const Generator4& g4, const Generator5& g5, const Generator6& g6,
-        const Generator7& g7, const Generator8& g8, const Generator9& g9,
-        const Generator10& g10) {
-  return internal::CartesianProductHolder10<Generator1, Generator2, Generator3,
-      Generator4, Generator5, Generator6, Generator7, Generator8, Generator9,
-      Generator10>(
-      g1, g2, g3, g4, g5, g6, g7, g8, g9, g10);
-}
-# endif  // GTEST_HAS_COMBINE
-
-
-
-# define TEST_P(test_case_name, test_name) \
-  class GTEST_TEST_CLASS_NAME_(test_case_name, test_name) \
-      : public test_case_name { \
-   public: \
-    GTEST_TEST_CLASS_NAME_(test_case_name, test_name)() {} \
-    virtual void TestBody(); \
-   private: \
-    static int AddToRegistry() { \
-      ::testing::UnitTest::GetInstance()->parameterized_test_registry(). \
-          GetTestCasePatternHolder<test_case_name>(\
-              #test_case_name, \
-              ::testing::internal::CodeLocation(\
-                  __FILE__, __LINE__))->AddTestPattern(\
-                      #test_case_name, \
-                      #test_name, \
-                      new ::testing::internal::TestMetaFactory< \
-                          GTEST_TEST_CLASS_NAME_(\
-                              test_case_name, test_name)>()); \
-      return 0; \
-    } \
-    static int gtest_registering_dummy_ GTEST_ATTRIBUTE_UNUSED_; \
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(\
-        GTEST_TEST_CLASS_NAME_(test_case_name, test_name)); \
-  }; \
-  int GTEST_TEST_CLASS_NAME_(test_case_name, \
-                             test_name)::gtest_registering_dummy_ = \
-      GTEST_TEST_CLASS_NAME_(test_case_name, test_name)::AddToRegistry(); \
-  void GTEST_TEST_CLASS_NAME_(test_case_name, test_name)::TestBody()
-
-// The optional last argument to INSTANTIATE_TEST_CASE_P allows the user
-// to specify a function or functor that generates custom test name suffixes
-// based on the test parameters. The function should accept one argument of
-// type testing::TestParamInfo<class ParamType>, and return std::string.
-//
-// testing::PrintToStringParamName is a builtin test suffix generator that
-// returns the value of testing::PrintToString(GetParam()). It does not work
-// for std::string or C strings.
-//
-// Note: test names must be non-empty, unique, and may only contain ASCII
-// alphanumeric characters or underscore.
-
-# define INSTANTIATE_TEST_CASE_P(prefix, test_case_name, generator, ...) \
-  ::testing::internal::ParamGenerator<test_case_name::ParamType> \
-      gtest_##prefix##test_case_name##_EvalGenerator_() { return generator; } \
-  ::std::string gtest_##prefix##test_case_name##_EvalGenerateName_( \
-      const ::testing::TestParamInfo<test_case_name::ParamType>& info) { \
-    return ::testing::internal::GetParamNameGen<test_case_name::ParamType> \
-        (__VA_ARGS__)(info); \
-  } \
-  int gtest_##prefix##test_case_name##_dummy_ GTEST_ATTRIBUTE_UNUSED_ = \
-      ::testing::UnitTest::GetInstance()->parameterized_test_registry(). \
-          GetTestCasePatternHolder<test_case_name>(\
-              #test_case_name, \
-              ::testing::internal::CodeLocation(\
-                  __FILE__, __LINE__))->AddTestCaseInstantiation(\
-                      #prefix, \
-                      &gtest_##prefix##test_case_name##_EvalGenerator_, \
-                      &gtest_##prefix##test_case_name##_EvalGenerateName_, \
-                      __FILE__, __LINE__)
-
-}  // namespace testing
-
-#endif  // GTEST_HAS_PARAM_TEST
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_PARAM_TEST_H_
-// Copyright 2006, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-//
-// Google C++ Testing Framework definitions useful in production code.
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_PROD_H_
-#define GTEST_INCLUDE_GTEST_GTEST_PROD_H_
-
-// When you need to test the private or protected members of a class,
-// use the FRIEND_TEST macro to declare your tests as friends of the
-// class.  For example:
-//
-// class MyClass {
-//  private:
-//   void MyMethod();
-//   FRIEND_TEST(MyClassTest, MyMethod);
-// };
-//
-// class MyClassTest : public testing::Test {
-//   // ...
-// };
-//
-// TEST_F(MyClassTest, MyMethod) {
-//   // Can call MyClass::MyMethod() here.
-// }
-
-#define FRIEND_TEST(test_case_name, test_name)\
-friend class test_case_name##_##test_name##_Test
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_PROD_H_
-// Copyright 2008, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: mheule@google.com (Markus Heule)
-//
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_TEST_PART_H_
-#define GTEST_INCLUDE_GTEST_GTEST_TEST_PART_H_
-
-#include <iosfwd>
-#include <vector>
-
-namespace testing {
-
-// A copyable object representing the result of a test part (i.e. an
-// assertion or an explicit FAIL(), ADD_FAILURE(), or SUCCESS()).
-//
-// Don't inherit from TestPartResult as its destructor is not virtual.
-class GTEST_API_ TestPartResult {
- public:
-  // The possible outcomes of a test part (i.e. an assertion or an
-  // explicit SUCCEED(), FAIL(), or ADD_FAILURE()).
-  enum Type {
-    kSuccess,          // Succeeded.
-    kNonFatalFailure,  // Failed but the test can continue.
-    kFatalFailure      // Failed and the test should be terminated.
-  };
-
-  // C'tor.  TestPartResult does NOT have a default constructor.
-  // Always use this constructor (with parameters) to create a
-  // TestPartResult object.
-  TestPartResult(Type a_type,
-                 const char* a_file_name,
-                 int a_line_number,
-                 const char* a_message)
-      : type_(a_type),
-        file_name_(a_file_name == NULL ? "" : a_file_name),
-        line_number_(a_line_number),
-        summary_(ExtractSummary(a_message)),
-        message_(a_message) {
-  }
-
-  // Gets the outcome of the test part.
-  Type type() const { return type_; }
-
-  // Gets the name of the source file where the test part took place, or
-  // NULL if it's unknown.
-  const char* file_name() const {
-    return file_name_.empty() ? NULL : file_name_.c_str();
-  }
-
-  // Gets the line in the source file where the test part took place,
-  // or -1 if it's unknown.
-  int line_number() const { return line_number_; }
-
-  // Gets the summary of the failure message.
-  const char* summary() const { return summary_.c_str(); }
-
-  // Gets the message associated with the test part.
-  const char* message() const { return message_.c_str(); }
-
-  // Returns true iff the test part passed.
-  bool passed() const { return type_ == kSuccess; }
-
-  // Returns true iff the test part failed.
-  bool failed() const { return type_ != kSuccess; }
-
-  // Returns true iff the test part non-fatally failed.
-  bool nonfatally_failed() const { return type_ == kNonFatalFailure; }
-
-  // Returns true iff the test part fatally failed.
-  bool fatally_failed() const { return type_ == kFatalFailure; }
-
- private:
-  Type type_;
-
-  // Gets the summary of the failure message by omitting the stack
-  // trace in it.
-  static std::string ExtractSummary(const char* message);
-
-  // The name of the source file where the test part took place, or
-  // "" if the source file is unknown.
-  std::string file_name_;
-  // The line in the source file where the test part took place, or -1
-  // if the line number is unknown.
-  int line_number_;
-  std::string summary_;  // The test failure summary.
-  std::string message_;  // The test failure message.
-};
-
-// Prints a TestPartResult object.
-std::ostream& operator<<(std::ostream& os, const TestPartResult& result);
-
-// An array of TestPartResult objects.
-//
-// Don't inherit from TestPartResultArray as its destructor is not
-// virtual.
-class GTEST_API_ TestPartResultArray {
- public:
-  TestPartResultArray() {}
-
-  // Appends the given TestPartResult to the array.
-  void Append(const TestPartResult& result);
-
-  // Returns the TestPartResult at the given index (0-based).
-  const TestPartResult& GetTestPartResult(int index) const;
-
-  // Returns the number of TestPartResult objects in the array.
-  int size() const;
-
- private:
-  std::vector<TestPartResult> array_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestPartResultArray);
-};
-
-// This interface knows how to report a test part result.
-class TestPartResultReporterInterface {
- public:
-  virtual ~TestPartResultReporterInterface() {}
-
-  virtual void ReportTestPartResult(const TestPartResult& result) = 0;
-};
-
-namespace internal {
-
-// This helper class is used by {ASSERT|EXPECT}_NO_FATAL_FAILURE to check if a
-// statement generates new fatal failures. To do so it registers itself as the
-// current test part result reporter. Besides checking if fatal failures were
-// reported, it only delegates the reporting to the former result reporter.
-// The original result reporter is restored in the destructor.
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-class GTEST_API_ HasNewFatalFailureHelper
-    : public TestPartResultReporterInterface {
- public:
-  HasNewFatalFailureHelper();
-  virtual ~HasNewFatalFailureHelper();
-  virtual void ReportTestPartResult(const TestPartResult& result);
-  bool has_new_fatal_failure() const { return has_new_fatal_failure_; }
- private:
-  bool has_new_fatal_failure_;
-  TestPartResultReporterInterface* original_reporter_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(HasNewFatalFailureHelper);
-};
-
-}  // namespace internal
-
-}  // namespace testing
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_TEST_PART_H_
-// Copyright 2008 Google Inc.
-// All Rights Reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-// Author: wan@google.com (Zhanyong Wan)
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_TYPED_TEST_H_
-#define GTEST_INCLUDE_GTEST_GTEST_TYPED_TEST_H_
-
-// This header implements typed tests and type-parameterized tests.
-
-// Typed (aka type-driven) tests repeat the same test for types in a
-// list.  You must know which types you want to test with when writing
-// typed tests. Here's how you do it:
-
-#if 0
-
-// First, define a fixture class template.  It should be parameterized
-// by a type.  Remember to derive it from testing::Test.
-template <typename T>
-class FooTest : public testing::Test {
- public:
-  ...
-  typedef std::list<T> List;
-  static T shared_;
-  T value_;
-};
-
-// Next, associate a list of types with the test case, which will be
-// repeated for each type in the list.  The typedef is necessary for
-// the macro to parse correctly.
-typedef testing::Types<char, int, unsigned int> MyTypes;
-TYPED_TEST_CASE(FooTest, MyTypes);
-
-// If the type list contains only one type, you can write that type
-// directly without Types<...>:
-//   TYPED_TEST_CASE(FooTest, int);
-
-// Then, use TYPED_TEST() instead of TEST_F() to define as many typed
-// tests for this test case as you want.
-TYPED_TEST(FooTest, DoesBlah) {
-  // Inside a test, refer to TypeParam to get the type parameter.
-  // Since we are inside a derived class template, C++ requires use to
-  // visit the members of FooTest via 'this'.
-  TypeParam n = this->value_;
-
-  // To visit static members of the fixture, add the TestFixture::
-  // prefix.
-  n += TestFixture::shared_;
-
-  // To refer to typedefs in the fixture, add the "typename
-  // TestFixture::" prefix.
-  typename TestFixture::List values;
-  values.push_back(n);
-  ...
-}
-
-TYPED_TEST(FooTest, HasPropertyA) { ... }
-
-#endif  // 0
-
-// Type-parameterized tests are abstract test patterns parameterized
-// by a type.  Compared with typed tests, type-parameterized tests
-// allow you to define the test pattern without knowing what the type
-// parameters are.  The defined pattern can be instantiated with
-// different types any number of times, in any number of translation
-// units.
-//
-// If you are designing an interface or concept, you can define a
-// suite of type-parameterized tests to verify properties that any
-// valid implementation of the interface/concept should have.  Then,
-// each implementation can easily instantiate the test suite to verify
-// that it conforms to the requirements, without having to write
-// similar tests repeatedly.  Here's an example:
-
-#if 0
-
-// First, define a fixture class template.  It should be parameterized
-// by a type.  Remember to derive it from testing::Test.
-template <typename T>
-class FooTest : public testing::Test {
-  ...
-};
-
-// Next, declare that you will define a type-parameterized test case
-// (the _P suffix is for "parameterized" or "pattern", whichever you
-// prefer):
-TYPED_TEST_CASE_P(FooTest);
-
-// Then, use TYPED_TEST_P() to define as many type-parameterized tests
-// for this type-parameterized test case as you want.
-TYPED_TEST_P(FooTest, DoesBlah) {
-  // Inside a test, refer to TypeParam to get the type parameter.
-  TypeParam n = 0;
-  ...
-}
-
-TYPED_TEST_P(FooTest, HasPropertyA) { ... }
-
-// Now the tricky part: you need to register all test patterns before
-// you can instantiate them.  The first argument of the macro is the
-// test case name; the rest are the names of the tests in this test
-// case.
-REGISTER_TYPED_TEST_CASE_P(FooTest,
-                           DoesBlah, HasPropertyA);
-
-// Finally, you are free to instantiate the pattern with the types you
-// want.  If you put the above code in a header file, you can #include
-// it in multiple C++ source files and instantiate it multiple times.
-//
-// To distinguish different instances of the pattern, the first
-// argument to the INSTANTIATE_* macro is a prefix that will be added
-// to the actual test case name.  Remember to pick unique prefixes for
-// different instances.
-typedef testing::Types<char, int, unsigned int> MyTypes;
-INSTANTIATE_TYPED_TEST_CASE_P(My, FooTest, MyTypes);
-
-// If the type list contains only one type, you can write that type
-// directly without Types<...>:
-//   INSTANTIATE_TYPED_TEST_CASE_P(My, FooTest, int);
-
-#endif  // 0
-
-
-// Implements typed tests.
-
-#if GTEST_HAS_TYPED_TEST
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Expands to the name of the typedef for the type parameters of the
-// given test case.
-# define GTEST_TYPE_PARAMS_(TestCaseName) gtest_type_params_##TestCaseName##_
-
-// The 'Types' template argument below must have spaces around it
-// since some compilers may choke on '>>' when passing a template
-// instance (e.g. Types<int>)
-# define TYPED_TEST_CASE(CaseName, Types) \
-  typedef ::testing::internal::TypeList< Types >::type \
-      GTEST_TYPE_PARAMS_(CaseName)
-
-# define TYPED_TEST(CaseName, TestName) \
-  template <typename gtest_TypeParam_> \
-  class GTEST_TEST_CLASS_NAME_(CaseName, TestName) \
-      : public CaseName<gtest_TypeParam_> { \
-   private: \
-    typedef CaseName<gtest_TypeParam_> TestFixture; \
-    typedef gtest_TypeParam_ TypeParam; \
-    virtual void TestBody(); \
-  }; \
-  bool gtest_##CaseName##_##TestName##_registered_ GTEST_ATTRIBUTE_UNUSED_ = \
-      ::testing::internal::TypeParameterizedTest< \
-          CaseName, \
-          ::testing::internal::TemplateSel< \
-              GTEST_TEST_CLASS_NAME_(CaseName, TestName)>, \
-          GTEST_TYPE_PARAMS_(CaseName)>::Register(\
-              "", ::testing::internal::CodeLocation(__FILE__, __LINE__), \
-              #CaseName, #TestName, 0); \
-  template <typename gtest_TypeParam_> \
-  void GTEST_TEST_CLASS_NAME_(CaseName, TestName)<gtest_TypeParam_>::TestBody()
-
-#endif  // GTEST_HAS_TYPED_TEST
-
-// Implements type-parameterized tests.
-
-#if GTEST_HAS_TYPED_TEST_P
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Expands to the namespace name that the type-parameterized tests for
-// the given type-parameterized test case are defined in.  The exact
-// name of the namespace is subject to change without notice.
-# define GTEST_CASE_NAMESPACE_(TestCaseName) \
-  gtest_case_##TestCaseName##_
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-//
-// Expands to the name of the variable used to remember the names of
-// the defined tests in the given test case.
-# define GTEST_TYPED_TEST_CASE_P_STATE_(TestCaseName) \
-  gtest_typed_test_case_p_state_##TestCaseName##_
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE DIRECTLY.
-//
-// Expands to the name of the variable used to remember the names of
-// the registered tests in the given test case.
-# define GTEST_REGISTERED_TEST_NAMES_(TestCaseName) \
-  gtest_registered_test_names_##TestCaseName##_
-
-// The variables defined in the type-parameterized test macros are
-// static as typically these macros are used in a .h file that can be
-// #included in multiple translation units linked together.
-# define TYPED_TEST_CASE_P(CaseName) \
-  static ::testing::internal::TypedTestCasePState \
-      GTEST_TYPED_TEST_CASE_P_STATE_(CaseName)
-
-# define TYPED_TEST_P(CaseName, TestName) \
-  namespace GTEST_CASE_NAMESPACE_(CaseName) { \
-  template <typename gtest_TypeParam_> \
-  class TestName : public CaseName<gtest_TypeParam_> { \
-   private: \
-    typedef CaseName<gtest_TypeParam_> TestFixture; \
-    typedef gtest_TypeParam_ TypeParam; \
-    virtual void TestBody(); \
-  }; \
-  static bool gtest_##TestName##_defined_ GTEST_ATTRIBUTE_UNUSED_ = \
-      GTEST_TYPED_TEST_CASE_P_STATE_(CaseName).AddTestName(\
-          __FILE__, __LINE__, #CaseName, #TestName); \
-  } \
-  template <typename gtest_TypeParam_> \
-  void GTEST_CASE_NAMESPACE_(CaseName)::TestName<gtest_TypeParam_>::TestBody()
-
-# define REGISTER_TYPED_TEST_CASE_P(CaseName, ...) \
-  namespace GTEST_CASE_NAMESPACE_(CaseName) { \
-  typedef ::testing::internal::Templates<__VA_ARGS__>::type gtest_AllTests_; \
-  } \
-  static const char* const GTEST_REGISTERED_TEST_NAMES_(CaseName) = \
-      GTEST_TYPED_TEST_CASE_P_STATE_(CaseName).VerifyRegisteredTestNames(\
-          __FILE__, __LINE__, #__VA_ARGS__)
-
-// The 'Types' template argument below must have spaces around it
-// since some compilers may choke on '>>' when passing a template
-// instance (e.g. Types<int>)
-# define INSTANTIATE_TYPED_TEST_CASE_P(Prefix, CaseName, Types) \
-  bool gtest_##Prefix##_##CaseName GTEST_ATTRIBUTE_UNUSED_ = \
-      ::testing::internal::TypeParameterizedTestCase<CaseName, \
-          GTEST_CASE_NAMESPACE_(CaseName)::gtest_AllTests_, \
-          ::testing::internal::TypeList< Types >::type>::Register(\
-              #Prefix, \
-              ::testing::internal::CodeLocation(__FILE__, __LINE__), \
-              &GTEST_TYPED_TEST_CASE_P_STATE_(CaseName), \
-              #CaseName, GTEST_REGISTERED_TEST_NAMES_(CaseName))
-
-#endif  // GTEST_HAS_TYPED_TEST_P
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_TYPED_TEST_H_
-
-// Depending on the platform, different string classes are available.
-// On Linux, in addition to ::std::string, Google also makes use of
-// class ::string, which has the same interface as ::std::string, but
-// has a different implementation.
-//
-// You can define GTEST_HAS_GLOBAL_STRING to 1 to indicate that
-// ::string is available AND is a distinct type to ::std::string, or
-// define it to 0 to indicate otherwise.
-//
-// If ::std::string and ::string are the same class on your platform
-// due to aliasing, you should define GTEST_HAS_GLOBAL_STRING to 0.
-//
-// If you do not define GTEST_HAS_GLOBAL_STRING, it is defined
-// heuristically.
-
-namespace testing {
-
-// Declares the flags.
-
-// This flag temporary enables the disabled tests.
-GTEST_DECLARE_bool_(also_run_disabled_tests);
-
-// This flag brings the debugger on an assertion failure.
-GTEST_DECLARE_bool_(break_on_failure);
-
-// This flag controls whether Google Test catches all test-thrown exceptions
-// and logs them as failures.
-GTEST_DECLARE_bool_(catch_exceptions);
-
-// This flag enables using colors in terminal output. Available values are
-// "yes" to enable colors, "no" (disable colors), or "auto" (the default)
-// to let Google Test decide.
-GTEST_DECLARE_string_(color);
-
-// This flag sets up the filter to select by name using a glob pattern
-// the tests to run. If the filter is not given all tests are executed.
-GTEST_DECLARE_string_(filter);
-
-// This flag causes the Google Test to list tests. None of the tests listed
-// are actually run if the flag is provided.
-GTEST_DECLARE_bool_(list_tests);
-
-// This flag controls whether Google Test emits a detailed XML report to a file
-// in addition to its normal textual output.
-GTEST_DECLARE_string_(output);
-
-// This flags control whether Google Test prints the elapsed time for each
-// test.
-GTEST_DECLARE_bool_(print_time);
-
-// This flag specifies the random number seed.
-GTEST_DECLARE_int32_(random_seed);
-
-// This flag sets how many times the tests are repeated. The default value
-// is 1. If the value is -1 the tests are repeating forever.
-GTEST_DECLARE_int32_(repeat);
-
-// This flag controls whether Google Test includes Google Test internal
-// stack frames in failure stack traces.
-GTEST_DECLARE_bool_(show_internal_stack_frames);
-
-// When this flag is specified, tests' order is randomized on every iteration.
-GTEST_DECLARE_bool_(shuffle);
-
-// This flag specifies the maximum number of stack frames to be
-// printed in a failure message.
-GTEST_DECLARE_int32_(stack_trace_depth);
-
-// When this flag is specified, a failed assertion will throw an
-// exception if exceptions are enabled, or exit the program with a
-// non-zero code otherwise.
-GTEST_DECLARE_bool_(throw_on_failure);
-
-// When this flag is set with a "host:port" string, on supported
-// platforms test results are streamed to the specified port on
-// the specified host machine.
-GTEST_DECLARE_string_(stream_result_to);
-
-// The upper limit for valid stack trace depths.
-const int kMaxStackTraceDepth = 100;
-
-namespace internal {
-
-class AssertHelper;
-class DefaultGlobalTestPartResultReporter;
-class ExecDeathTest;
-class NoExecDeathTest;
-class FinalSuccessChecker;
-class GTestFlagSaver;
-class StreamingListenerTest;
-class TestResultAccessor;
-class TestEventListenersAccessor;
-class TestEventRepeater;
-class UnitTestRecordPropertyTestHelper;
-class WindowsDeathTest;
-class UnitTestImpl* GetUnitTestImpl();
-void ReportFailureInUnknownLocation(TestPartResult::Type result_type,
-                                    const std::string& message);
-
-}  // namespace internal
-
-// The friend relationship of some of these classes is cyclic.
-// If we don't forward declare them the compiler might confuse the classes
-// in friendship clauses with same named classes on the scope.
-class Test;
-class TestCase;
-class TestInfo;
-class UnitTest;
-
-// A class for indicating whether an assertion was successful.  When
-// the assertion wasn't successful, the AssertionResult object
-// remembers a non-empty message that describes how it failed.
-//
-// To create an instance of this class, use one of the factory functions
-// (AssertionSuccess() and AssertionFailure()).
-//
-// This class is useful for two purposes:
-//   1. Defining predicate functions to be used with Boolean test assertions
-//      EXPECT_TRUE/EXPECT_FALSE and their ASSERT_ counterparts
-//   2. Defining predicate-format functions to be
-//      used with predicate assertions (ASSERT_PRED_FORMAT*, etc).
-//
-// For example, if you define IsEven predicate:
-//
-//   testing::AssertionResult IsEven(int n) {
-//     if ((n % 2) == 0)
-//       return testing::AssertionSuccess();
-//     else
-//       return testing::AssertionFailure() << n << " is odd";
-//   }
-//
-// Then the failed expectation EXPECT_TRUE(IsEven(Fib(5)))
-// will print the message
-//
-//   Value of: IsEven(Fib(5))
-//     Actual: false (5 is odd)
-//   Expected: true
-//
-// instead of a more opaque
-//
-//   Value of: IsEven(Fib(5))
-//     Actual: false
-//   Expected: true
-//
-// in case IsEven is a simple Boolean predicate.
-//
-// If you expect your predicate to be reused and want to support informative
-// messages in EXPECT_FALSE and ASSERT_FALSE (negative assertions show up
-// about half as often as positive ones in our tests), supply messages for
-// both success and failure cases:
-//
-//   testing::AssertionResult IsEven(int n) {
-//     if ((n % 2) == 0)
-//       return testing::AssertionSuccess() << n << " is even";
-//     else
-//       return testing::AssertionFailure() << n << " is odd";
-//   }
-//
-// Then a statement EXPECT_FALSE(IsEven(Fib(6))) will print
-//
-//   Value of: IsEven(Fib(6))
-//     Actual: true (8 is even)
-//   Expected: false
-//
-// NB: Predicates that support negative Boolean assertions have reduced
-// performance in positive ones so be careful not to use them in tests
-// that have lots (tens of thousands) of positive Boolean assertions.
-//
-// To use this class with EXPECT_PRED_FORMAT assertions such as:
-//
-//   // Verifies that Foo() returns an even number.
-//   EXPECT_PRED_FORMAT1(IsEven, Foo());
-//
-// you need to define:
-//
-//   testing::AssertionResult IsEven(const char* expr, int n) {
-//     if ((n % 2) == 0)
-//       return testing::AssertionSuccess();
-//     else
-//       return testing::AssertionFailure()
-//         << "Expected: " << expr << " is even\n  Actual: it's " << n;
-//   }
-//
-// If Foo() returns 5, you will see the following message:
-//
-//   Expected: Foo() is even
-//     Actual: it's 5
-//
-class GTEST_API_ AssertionResult {
- public:
-  // Copy constructor.
-  // Used in EXPECT_TRUE/FALSE(assertion_result).
-  AssertionResult(const AssertionResult& other);
-
-  GTEST_DISABLE_MSC_WARNINGS_PUSH_(4800 /* forcing value to bool */)
-
-  // Used in the EXPECT_TRUE/FALSE(bool_expression).
-  //
-  // T must be contextually convertible to bool.
-  //
-  // The second parameter prevents this overload from being considered if
-  // the argument is implicitly convertible to AssertionResult. In that case
-  // we want AssertionResult's copy constructor to be used.
-  template <typename T>
-  explicit AssertionResult(
-      const T& success,
-      typename internal::EnableIf<
-          !internal::ImplicitlyConvertible<T, AssertionResult>::value>::type*
-          /*enabler*/ = NULL)
-      : success_(success) {}
-
-  GTEST_DISABLE_MSC_WARNINGS_POP_()
-
-  // Assignment operator.
-  AssertionResult& operator=(AssertionResult other) {
-    swap(other);
-    return *this;
-  }
-
-  // Returns true iff the assertion succeeded.
-  operator bool() const { return success_; }  // NOLINT
-
-  // Returns the assertion's negation. Used with EXPECT/ASSERT_FALSE.
-  AssertionResult operator!() const;
-
-  // Returns the text streamed into this AssertionResult. Test assertions
-  // use it when they fail (i.e., the predicate's outcome doesn't match the
-  // assertion's expectation). When nothing has been streamed into the
-  // object, returns an empty string.
-  const char* message() const {
-    return message_.get() != NULL ?  message_->c_str() : "";
-  }
-  // TODO(vladl@google.com): Remove this after making sure no clients use it.
-  // Deprecated; please use message() instead.
-  const char* failure_message() const { return message(); }
-
-  // Streams a custom failure message into this object.
-  template <typename T> AssertionResult& operator<<(const T& value) {
-    AppendMessage(Message() << value);
-    return *this;
-  }
-
-  // Allows streaming basic output manipulators such as endl or flush into
-  // this object.
-  AssertionResult& operator<<(
-      ::std::ostream& (*basic_manipulator)(::std::ostream& stream)) {
-    AppendMessage(Message() << basic_manipulator);
-    return *this;
-  }
-
- private:
-  // Appends the contents of message to message_.
-  void AppendMessage(const Message& a_message) {
-    if (message_.get() == NULL)
-      message_.reset(new ::std::string);
-    message_->append(a_message.GetString().c_str());
-  }
-
-  // Swap the contents of this AssertionResult with other.
-  void swap(AssertionResult& other);
-
-  // Stores result of the assertion predicate.
-  bool success_;
-  // Stores the message describing the condition in case the expectation
-  // construct is not satisfied with the predicate's outcome.
-  // Referenced via a pointer to avoid taking too much stack frame space
-  // with test assertions.
-  internal::scoped_ptr< ::std::string> message_;
-};
-
-// Makes a successful assertion result.
-GTEST_API_ AssertionResult AssertionSuccess();
-
-// Makes a failed assertion result.
-GTEST_API_ AssertionResult AssertionFailure();
-
-// Makes a failed assertion result with the given failure message.
-// Deprecated; use AssertionFailure() << msg.
-GTEST_API_ AssertionResult AssertionFailure(const Message& msg);
-
-// The abstract class that all tests inherit from.
-//
-// In Google Test, a unit test program contains one or many TestCases, and
-// each TestCase contains one or many Tests.
-//
-// When you define a test using the TEST macro, you don't need to
-// explicitly derive from Test - the TEST macro automatically does
-// this for you.
-//
-// The only time you derive from Test is when defining a test fixture
-// to be used a TEST_F.  For example:
-//
-//   class FooTest : public testing::Test {
-//    protected:
-//     void SetUp() override { ... }
-//     void TearDown() override { ... }
-//     ...
-//   };
-//
-//   TEST_F(FooTest, Bar) { ... }
-//   TEST_F(FooTest, Baz) { ... }
-//
-// Test is not copyable.
-class GTEST_API_ Test {
- public:
-  friend class TestInfo;
-
-  // Defines types for pointers to functions that set up and tear down
-  // a test case.
-  typedef internal::SetUpTestCaseFunc SetUpTestCaseFunc;
-  typedef internal::TearDownTestCaseFunc TearDownTestCaseFunc;
-
-  // The d'tor is virtual as we intend to inherit from Test.
-  virtual ~Test();
-
-  // Sets up the stuff shared by all tests in this test case.
-  //
-  // Google Test will call Foo::SetUpTestCase() before running the first
-  // test in test case Foo.  Hence a sub-class can define its own
-  // SetUpTestCase() method to shadow the one defined in the super
-  // class.
-  static void SetUpTestCase() {}
-
-  // Tears down the stuff shared by all tests in this test case.
-  //
-  // Google Test will call Foo::TearDownTestCase() after running the last
-  // test in test case Foo.  Hence a sub-class can define its own
-  // TearDownTestCase() method to shadow the one defined in the super
-  // class.
-  static void TearDownTestCase() {}
-
-  // Returns true iff the current test has a fatal failure.
-  static bool HasFatalFailure();
-
-  // Returns true iff the current test has a non-fatal failure.
-  static bool HasNonfatalFailure();
-
-  // Returns true iff the current test has a (either fatal or
-  // non-fatal) failure.
-  static bool HasFailure() { return HasFatalFailure() || HasNonfatalFailure(); }
-
-  // Logs a property for the current test, test case, or for the entire
-  // invocation of the test program when used outside of the context of a
-  // test case.  Only the last value for a given key is remembered.  These
-  // are public static so they can be called from utility functions that are
-  // not members of the test fixture.  Calls to RecordProperty made during
-  // lifespan of the test (from the moment its constructor starts to the
-  // moment its destructor finishes) will be output in XML as attributes of
-  // the <testcase> element.  Properties recorded from fixture's
-  // SetUpTestCase or TearDownTestCase are logged as attributes of the
-  // corresponding <testsuite> element.  Calls to RecordProperty made in the
-  // global context (before or after invocation of RUN_ALL_TESTS and from
-  // SetUp/TearDown method of Environment objects registered with Google
-  // Test) will be output as attributes of the <testsuites> element.
-  static void RecordProperty(const std::string& key, const std::string& value);
-  static void RecordProperty(const std::string& key, int value);
-
- protected:
-  // Creates a Test object.
-  Test();
-
-  // Sets up the test fixture.
-  virtual void SetUp();
-
-  // Tears down the test fixture.
-  virtual void TearDown();
-
- private:
-  // Returns true iff the current test has the same fixture class as
-  // the first test in the current test case.
-  static bool HasSameFixtureClass();
-
-  // Runs the test after the test fixture has been set up.
-  //
-  // A sub-class must implement this to define the test logic.
-  //
-  // DO NOT OVERRIDE THIS FUNCTION DIRECTLY IN A USER PROGRAM.
-  // Instead, use the TEST or TEST_F macro.
-  virtual void TestBody() = 0;
-
-  // Sets up, executes, and tears down the test.
-  void Run();
-
-  // Deletes self.  We deliberately pick an unusual name for this
-  // internal method to avoid clashing with names used in user TESTs.
-  void DeleteSelf_() { delete this; }
-
-  const internal::scoped_ptr< GTEST_FLAG_SAVER_ > gtest_flag_saver_;
-
-  // Often a user misspells SetUp() as Setup() and spends a long time
-  // wondering why it is never called by Google Test.  The declaration of
-  // the following method is solely for catching such an error at
-  // compile time:
-  //
-  //   - The return type is deliberately chosen to be not void, so it
-  //   will be a conflict if void Setup() is declared in the user's
-  //   test fixture.
-  //
-  //   - This method is private, so it will be another compiler error
-  //   if the method is called from the user's test fixture.
-  //
-  // DO NOT OVERRIDE THIS FUNCTION.
-  //
-  // If you see an error about overriding the following function or
-  // about it being private, you have mis-spelled SetUp() as Setup().
-  struct Setup_should_be_spelled_SetUp {};
-  virtual Setup_should_be_spelled_SetUp* Setup() { return NULL; }
-
-  // We disallow copying Tests.
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(Test);
-};
-
-typedef internal::TimeInMillis TimeInMillis;
-
-// A copyable object representing a user specified test property which can be
-// output as a key/value string pair.
-//
-// Don't inherit from TestProperty as its destructor is not virtual.
-class TestProperty {
- public:
-  // C'tor.  TestProperty does NOT have a default constructor.
-  // Always use this constructor (with parameters) to create a
-  // TestProperty object.
-  TestProperty(const std::string& a_key, const std::string& a_value) :
-    key_(a_key), value_(a_value) {
-  }
-
-  // Gets the user supplied key.
-  const char* key() const {
-    return key_.c_str();
-  }
-
-  // Gets the user supplied value.
-  const char* value() const {
-    return value_.c_str();
-  }
-
-  // Sets a new value, overriding the one supplied in the constructor.
-  void SetValue(const std::string& new_value) {
-    value_ = new_value;
-  }
-
- private:
-  // The key supplied by the user.
-  std::string key_;
-  // The value supplied by the user.
-  std::string value_;
-};
-
-// The result of a single Test.  This includes a list of
-// TestPartResults, a list of TestProperties, a count of how many
-// death tests there are in the Test, and how much time it took to run
-// the Test.
-//
-// TestResult is not copyable.
-class GTEST_API_ TestResult {
- public:
-  // Creates an empty TestResult.
-  TestResult();
-
-  // D'tor.  Do not inherit from TestResult.
-  ~TestResult();
-
-  // Gets the number of all test parts.  This is the sum of the number
-  // of successful test parts and the number of failed test parts.
-  int total_part_count() const;
-
-  // Returns the number of the test properties.
-  int test_property_count() const;
-
-  // Returns true iff the test passed (i.e. no test part failed).
-  bool Passed() const { return !Failed(); }
-
-  // Returns true iff the test failed.
-  bool Failed() const;
-
-  // Returns true iff the test fatally failed.
-  bool HasFatalFailure() const;
-
-  // Returns true iff the test has a non-fatal failure.
-  bool HasNonfatalFailure() const;
-
-  // Returns the elapsed time, in milliseconds.
-  TimeInMillis elapsed_time() const { return elapsed_time_; }
-
-  // Returns the i-th test part result among all the results. i can range
-  // from 0 to test_property_count() - 1. If i is not in that range, aborts
-  // the program.
-  const TestPartResult& GetTestPartResult(int i) const;
-
-  // Returns the i-th test property. i can range from 0 to
-  // test_property_count() - 1. If i is not in that range, aborts the
-  // program.
-  const TestProperty& GetTestProperty(int i) const;
-
- private:
-  friend class TestInfo;
-  friend class TestCase;
-  friend class UnitTest;
-  friend class internal::DefaultGlobalTestPartResultReporter;
-  friend class internal::ExecDeathTest;
-  friend class internal::TestResultAccessor;
-  friend class internal::UnitTestImpl;
-  friend class internal::WindowsDeathTest;
-
-  // Gets the vector of TestPartResults.
-  const std::vector<TestPartResult>& test_part_results() const {
-    return test_part_results_;
-  }
-
-  // Gets the vector of TestProperties.
-  const std::vector<TestProperty>& test_properties() const {
-    return test_properties_;
-  }
-
-  // Sets the elapsed time.
-  void set_elapsed_time(TimeInMillis elapsed) { elapsed_time_ = elapsed; }
-
-  // Adds a test property to the list. The property is validated and may add
-  // a non-fatal failure if invalid (e.g., if it conflicts with reserved
-  // key names). If a property is already recorded for the same key, the
-  // value will be updated, rather than storing multiple values for the same
-  // key.  xml_element specifies the element for which the property is being
-  // recorded and is used for validation.
-  void RecordProperty(const std::string& xml_element,
-                      const TestProperty& test_property);
-
-  // Adds a failure if the key is a reserved attribute of Google Test
-  // testcase tags.  Returns true if the property is valid.
-  // TODO(russr): Validate attribute names are legal and human readable.
-  static bool ValidateTestProperty(const std::string& xml_element,
-                                   const TestProperty& test_property);
-
-  // Adds a test part result to the list.
-  void AddTestPartResult(const TestPartResult& test_part_result);
-
-  // Returns the death test count.
-  int death_test_count() const { return death_test_count_; }
-
-  // Increments the death test count, returning the new count.
-  int increment_death_test_count() { return ++death_test_count_; }
-
-  // Clears the test part results.
-  void ClearTestPartResults();
-
-  // Clears the object.
-  void Clear();
-
-  // Protects mutable state of the property vector and of owned
-  // properties, whose values may be updated.
-  internal::Mutex test_properites_mutex_;
-
-  // The vector of TestPartResults
-  std::vector<TestPartResult> test_part_results_;
-  // The vector of TestProperties
-  std::vector<TestProperty> test_properties_;
-  // Running count of death tests.
-  int death_test_count_;
-  // The elapsed time, in milliseconds.
-  TimeInMillis elapsed_time_;
-
-  // We disallow copying TestResult.
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestResult);
-};  // class TestResult
-
-// A TestInfo object stores the following information about a test:
-//
-//   Test case name
-//   Test name
-//   Whether the test should be run
-//   A function pointer that creates the test object when invoked
-//   Test result
-//
-// The constructor of TestInfo registers itself with the UnitTest
-// singleton such that the RUN_ALL_TESTS() macro knows which tests to
-// run.
-class GTEST_API_ TestInfo {
- public:
-  // Destructs a TestInfo object.  This function is not virtual, so
-  // don't inherit from TestInfo.
-  ~TestInfo();
-
-  // Returns the test case name.
-  const char* test_case_name() const { return test_case_name_.c_str(); }
-
-  // Returns the test name.
-  const char* name() const { return name_.c_str(); }
-
-  // Returns the name of the parameter type, or NULL if this is not a typed
-  // or a type-parameterized test.
-  const char* type_param() const {
-    if (type_param_.get() != NULL)
-      return type_param_->c_str();
-    return NULL;
-  }
-
-  // Returns the text representation of the value parameter, or NULL if this
-  // is not a value-parameterized test.
-  const char* value_param() const {
-    if (value_param_.get() != NULL)
-      return value_param_->c_str();
-    return NULL;
-  }
-
-  // Returns the file name where this test is defined.
-  const char* file() const { return location_.file.c_str(); }
-
-  // Returns the line where this test is defined.
-  int line() const { return location_.line; }
-
-  // Returns true if this test should run, that is if the test is not
-  // disabled (or it is disabled but the also_run_disabled_tests flag has
-  // been specified) and its full name matches the user-specified filter.
-  //
-  // Google Test allows the user to filter the tests by their full names.
-  // The full name of a test Bar in test case Foo is defined as
-  // "Foo.Bar".  Only the tests that match the filter will run.
-  //
-  // A filter is a colon-separated list of glob (not regex) patterns,
-  // optionally followed by a '-' and a colon-separated list of
-  // negative patterns (tests to exclude).  A test is run if it
-  // matches one of the positive patterns and does not match any of
-  // the negative patterns.
-  //
-  // For example, *A*:Foo.* is a filter that matches any string that
-  // contains the character 'A' or starts with "Foo.".
-  bool should_run() const { return should_run_; }
-
-  // Returns true iff this test will appear in the XML report.
-  bool is_reportable() const {
-    // For now, the XML report includes all tests matching the filter.
-    // In the future, we may trim tests that are excluded because of
-    // sharding.
-    return matches_filter_;
-  }
-
-  // Returns the result of the test.
-  const TestResult* result() const { return &result_; }
-
- private:
-#if GTEST_HAS_DEATH_TEST
-  friend class internal::DefaultDeathTestFactory;
-#endif  // GTEST_HAS_DEATH_TEST
-  friend class Test;
-  friend class TestCase;
-  friend class internal::UnitTestImpl;
-  friend class internal::StreamingListenerTest;
-  friend TestInfo* internal::MakeAndRegisterTestInfo(
-      const char* test_case_name,
-      const char* name,
-      const char* type_param,
-      const char* value_param,
-      internal::CodeLocation code_location,
-      internal::TypeId fixture_class_id,
-      Test::SetUpTestCaseFunc set_up_tc,
-      Test::TearDownTestCaseFunc tear_down_tc,
-      internal::TestFactoryBase* factory);
-
-  // Constructs a TestInfo object. The newly constructed instance assumes
-  // ownership of the factory object.
-  TestInfo(const std::string& test_case_name,
-           const std::string& name,
-           const char* a_type_param,   // NULL if not a type-parameterized test
-           const char* a_value_param,  // NULL if not a value-parameterized test
-           internal::CodeLocation a_code_location,
-           internal::TypeId fixture_class_id,
-           internal::TestFactoryBase* factory);
-
-  // Increments the number of death tests encountered in this test so
-  // far.
-  int increment_death_test_count() {
-    return result_.increment_death_test_count();
-  }
-
-  // Creates the test object, runs it, records its result, and then
-  // deletes it.
-  void Run();
-
-  static void ClearTestResult(TestInfo* test_info) {
-    test_info->result_.Clear();
-  }
-
-  // These fields are immutable properties of the test.
-  const std::string test_case_name_;     // Test case name
-  const std::string name_;               // Test name
-  // Name of the parameter type, or NULL if this is not a typed or a
-  // type-parameterized test.
-  const internal::scoped_ptr<const ::std::string> type_param_;
-  // Text representation of the value parameter, or NULL if this is not a
-  // value-parameterized test.
-  const internal::scoped_ptr<const ::std::string> value_param_;
-  internal::CodeLocation location_;
-  const internal::TypeId fixture_class_id_;   // ID of the test fixture class
-  bool should_run_;                 // True iff this test should run
-  bool is_disabled_;                // True iff this test is disabled
-  bool matches_filter_;             // True if this test matches the
-                                    // user-specified filter.
-  internal::TestFactoryBase* const factory_;  // The factory that creates
-                                              // the test object
-
-  // This field is mutable and needs to be reset before running the
-  // test for the second time.
-  TestResult result_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestInfo);
-};
-
-// A test case, which consists of a vector of TestInfos.
-//
-// TestCase is not copyable.
-class GTEST_API_ TestCase {
- public:
-  // Creates a TestCase with the given name.
-  //
-  // TestCase does NOT have a default constructor.  Always use this
-  // constructor to create a TestCase object.
-  //
-  // Arguments:
-  //
-  //   name:         name of the test case
-  //   a_type_param: the name of the test's type parameter, or NULL if
-  //                 this is not a type-parameterized test.
-  //   set_up_tc:    pointer to the function that sets up the test case
-  //   tear_down_tc: pointer to the function that tears down the test case
-  TestCase(const char* name, const char* a_type_param,
-           Test::SetUpTestCaseFunc set_up_tc,
-           Test::TearDownTestCaseFunc tear_down_tc);
-
-  // Destructor of TestCase.
-  virtual ~TestCase();
-
-  // Gets the name of the TestCase.
-  const char* name() const { return name_.c_str(); }
-
-  // Returns the name of the parameter type, or NULL if this is not a
-  // type-parameterized test case.
-  const char* type_param() const {
-    if (type_param_.get() != NULL)
-      return type_param_->c_str();
-    return NULL;
-  }
-
-  // Returns true if any test in this test case should run.
-  bool should_run() const { return should_run_; }
-
-  // Gets the number of successful tests in this test case.
-  int successful_test_count() const;
-
-  // Gets the number of failed tests in this test case.
-  int failed_test_count() const;
-
-  // Gets the number of disabled tests that will be reported in the XML report.
-  int reportable_disabled_test_count() const;
-
-  // Gets the number of disabled tests in this test case.
-  int disabled_test_count() const;
-
-  // Gets the number of tests to be printed in the XML report.
-  int reportable_test_count() const;
-
-  // Get the number of tests in this test case that should run.
-  int test_to_run_count() const;
-
-  // Gets the number of all tests in this test case.
-  int total_test_count() const;
-
-  // Returns true iff the test case passed.
-  bool Passed() const { return !Failed(); }
-
-  // Returns true iff the test case failed.
-  bool Failed() const { return failed_test_count() > 0; }
-
-  // Returns the elapsed time, in milliseconds.
-  TimeInMillis elapsed_time() const { return elapsed_time_; }
-
-  // Returns the i-th test among all the tests. i can range from 0 to
-  // total_test_count() - 1. If i is not in that range, returns NULL.
-  const TestInfo* GetTestInfo(int i) const;
-
-  // Returns the TestResult that holds test properties recorded during
-  // execution of SetUpTestCase and TearDownTestCase.
-  const TestResult& ad_hoc_test_result() const { return ad_hoc_test_result_; }
-
- private:
-  friend class Test;
-  friend class internal::UnitTestImpl;
-
-  // Gets the (mutable) vector of TestInfos in this TestCase.
-  std::vector<TestInfo*>& test_info_list() { return test_info_list_; }
-
-  // Gets the (immutable) vector of TestInfos in this TestCase.
-  const std::vector<TestInfo*>& test_info_list() const {
-    return test_info_list_;
-  }
-
-  // Returns the i-th test among all the tests. i can range from 0 to
-  // total_test_count() - 1. If i is not in that range, returns NULL.
-  TestInfo* GetMutableTestInfo(int i);
-
-  // Sets the should_run member.
-  void set_should_run(bool should) { should_run_ = should; }
-
-  // Adds a TestInfo to this test case.  Will delete the TestInfo upon
-  // destruction of the TestCase object.
-  void AddTestInfo(TestInfo * test_info);
-
-  // Clears the results of all tests in this test case.
-  void ClearResult();
-
-  // Clears the results of all tests in the given test case.
-  static void ClearTestCaseResult(TestCase* test_case) {
-    test_case->ClearResult();
-  }
-
-  // Runs every test in this TestCase.
-  void Run();
-
-  // Runs SetUpTestCase() for this TestCase.  This wrapper is needed
-  // for catching exceptions thrown from SetUpTestCase().
-  void RunSetUpTestCase() { (*set_up_tc_)(); }
-
-  // Runs TearDownTestCase() for this TestCase.  This wrapper is
-  // needed for catching exceptions thrown from TearDownTestCase().
-  void RunTearDownTestCase() { (*tear_down_tc_)(); }
-
-  // Returns true iff test passed.
-  static bool TestPassed(const TestInfo* test_info) {
-    return test_info->should_run() && test_info->result()->Passed();
-  }
-
-  // Returns true iff test failed.
-  static bool TestFailed(const TestInfo* test_info) {
-    return test_info->should_run() && test_info->result()->Failed();
-  }
-
-  // Returns true iff the test is disabled and will be reported in the XML
-  // report.
-  static bool TestReportableDisabled(const TestInfo* test_info) {
-    return test_info->is_reportable() && test_info->is_disabled_;
-  }
-
-  // Returns true iff test is disabled.
-  static bool TestDisabled(const TestInfo* test_info) {
-    return test_info->is_disabled_;
-  }
-
-  // Returns true iff this test will appear in the XML report.
-  static bool TestReportable(const TestInfo* test_info) {
-    return test_info->is_reportable();
-  }
-
-  // Returns true if the given test should run.
-  static bool ShouldRunTest(const TestInfo* test_info) {
-    return test_info->should_run();
-  }
-
-  // Shuffles the tests in this test case.
-  void ShuffleTests(internal::Random* random);
-
-  // Restores the test order to before the first shuffle.
-  void UnshuffleTests();
-
-  // Name of the test case.
-  std::string name_;
-  // Name of the parameter type, or NULL if this is not a typed or a
-  // type-parameterized test.
-  const internal::scoped_ptr<const ::std::string> type_param_;
-  // The vector of TestInfos in their original order.  It owns the
-  // elements in the vector.
-  std::vector<TestInfo*> test_info_list_;
-  // Provides a level of indirection for the test list to allow easy
-  // shuffling and restoring the test order.  The i-th element in this
-  // vector is the index of the i-th test in the shuffled test list.
-  std::vector<int> test_indices_;
-  // Pointer to the function that sets up the test case.
-  Test::SetUpTestCaseFunc set_up_tc_;
-  // Pointer to the function that tears down the test case.
-  Test::TearDownTestCaseFunc tear_down_tc_;
-  // True iff any test in this test case should run.
-  bool should_run_;
-  // Elapsed time, in milliseconds.
-  TimeInMillis elapsed_time_;
-  // Holds test properties recorded during execution of SetUpTestCase and
-  // TearDownTestCase.
-  TestResult ad_hoc_test_result_;
-
-  // We disallow copying TestCases.
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestCase);
-};
-
-// An Environment object is capable of setting up and tearing down an
-// environment.  You should subclass this to define your own
-// environment(s).
-//
-// An Environment object does the set-up and tear-down in virtual
-// methods SetUp() and TearDown() instead of the constructor and the
-// destructor, as:
-//
-//   1. You cannot safely throw from a destructor.  This is a problem
-//      as in some cases Google Test is used where exceptions are enabled, and
-//      we may want to implement ASSERT_* using exceptions where they are
-//      available.
-//   2. You cannot use ASSERT_* directly in a constructor or
-//      destructor.
-class Environment {
- public:
-  // The d'tor is virtual as we need to subclass Environment.
-  virtual ~Environment() {}
-
-  // Override this to define how to set up the environment.
-  virtual void SetUp() {}
-
-  // Override this to define how to tear down the environment.
-  virtual void TearDown() {}
- private:
-  // If you see an error about overriding the following function or
-  // about it being private, you have mis-spelled SetUp() as Setup().
-  struct Setup_should_be_spelled_SetUp {};
-  virtual Setup_should_be_spelled_SetUp* Setup() { return NULL; }
-};
-
-// The interface for tracing execution of tests. The methods are organized in
-// the order the corresponding events are fired.
-class TestEventListener {
- public:
-  virtual ~TestEventListener() {}
-
-  // Fired before any test activity starts.
-  virtual void OnTestProgramStart(const UnitTest& unit_test) = 0;
-
-  // Fired before each iteration of tests starts.  There may be more than
-  // one iteration if GTEST_FLAG(repeat) is set. iteration is the iteration
-  // index, starting from 0.
-  virtual void OnTestIterationStart(const UnitTest& unit_test,
-                                    int iteration) = 0;
-
-  // Fired before environment set-up for each iteration of tests starts.
-  virtual void OnEnvironmentsSetUpStart(const UnitTest& unit_test) = 0;
-
-  // Fired after environment set-up for each iteration of tests ends.
-  virtual void OnEnvironmentsSetUpEnd(const UnitTest& unit_test) = 0;
-
-  // Fired before the test case starts.
-  virtual void OnTestCaseStart(const TestCase& test_case) = 0;
-
-  // Fired before the test starts.
-  virtual void OnTestStart(const TestInfo& test_info) = 0;
-
-  // Fired after a failed assertion or a SUCCEED() invocation.
-  virtual void OnTestPartResult(const TestPartResult& test_part_result) = 0;
-
-  // Fired after the test ends.
-  virtual void OnTestEnd(const TestInfo& test_info) = 0;
-
-  // Fired after the test case ends.
-  virtual void OnTestCaseEnd(const TestCase& test_case) = 0;
-
-  // Fired before environment tear-down for each iteration of tests starts.
-  virtual void OnEnvironmentsTearDownStart(const UnitTest& unit_test) = 0;
-
-  // Fired after environment tear-down for each iteration of tests ends.
-  virtual void OnEnvironmentsTearDownEnd(const UnitTest& unit_test) = 0;
-
-  // Fired after each iteration of tests finishes.
-  virtual void OnTestIterationEnd(const UnitTest& unit_test,
-                                  int iteration) = 0;
-
-  // Fired after all test activities have ended.
-  virtual void OnTestProgramEnd(const UnitTest& unit_test) = 0;
-};
-
-// The convenience class for users who need to override just one or two
-// methods and are not concerned that a possible change to a signature of
-// the methods they override will not be caught during the build.  For
-// comments about each method please see the definition of TestEventListener
-// above.
-class EmptyTestEventListener : public TestEventListener {
- public:
-  virtual void OnTestProgramStart(const UnitTest& /*unit_test*/) {}
-  virtual void OnTestIterationStart(const UnitTest& /*unit_test*/,
-                                    int /*iteration*/) {}
-  virtual void OnEnvironmentsSetUpStart(const UnitTest& /*unit_test*/) {}
-  virtual void OnEnvironmentsSetUpEnd(const UnitTest& /*unit_test*/) {}
-  virtual void OnTestCaseStart(const TestCase& /*test_case*/) {}
-  virtual void OnTestStart(const TestInfo& /*test_info*/) {}
-  virtual void OnTestPartResult(const TestPartResult& /*test_part_result*/) {}
-  virtual void OnTestEnd(const TestInfo& /*test_info*/) {}
-  virtual void OnTestCaseEnd(const TestCase& /*test_case*/) {}
-  virtual void OnEnvironmentsTearDownStart(const UnitTest& /*unit_test*/) {}
-  virtual void OnEnvironmentsTearDownEnd(const UnitTest& /*unit_test*/) {}
-  virtual void OnTestIterationEnd(const UnitTest& /*unit_test*/,
-                                  int /*iteration*/) {}
-  virtual void OnTestProgramEnd(const UnitTest& /*unit_test*/) {}
-};
-
-// TestEventListeners lets users add listeners to track events in Google Test.
-class GTEST_API_ TestEventListeners {
- public:
-  TestEventListeners();
-  ~TestEventListeners();
-
-  // Appends an event listener to the end of the list. Google Test assumes
-  // the ownership of the listener (i.e. it will delete the listener when
-  // the test program finishes).
-  void Append(TestEventListener* listener);
-
-  // Removes the given event listener from the list and returns it.  It then
-  // becomes the caller's responsibility to delete the listener. Returns
-  // NULL if the listener is not found in the list.
-  TestEventListener* Release(TestEventListener* listener);
-
-  // Returns the standard listener responsible for the default console
-  // output.  Can be removed from the listeners list to shut down default
-  // console output.  Note that removing this object from the listener list
-  // with Release transfers its ownership to the caller and makes this
-  // function return NULL the next time.
-  TestEventListener* default_result_printer() const {
-    return default_result_printer_;
-  }
-
-  // Returns the standard listener responsible for the default XML output
-  // controlled by the --gtest_output=xml flag.  Can be removed from the
-  // listeners list by users who want to shut down the default XML output
-  // controlled by this flag and substitute it with custom one.  Note that
-  // removing this object from the listener list with Release transfers its
-  // ownership to the caller and makes this function return NULL the next
-  // time.
-  TestEventListener* default_xml_generator() const {
-    return default_xml_generator_;
-  }
-
- private:
-  friend class TestCase;
-  friend class TestInfo;
-  friend class internal::DefaultGlobalTestPartResultReporter;
-  friend class internal::NoExecDeathTest;
-  friend class internal::TestEventListenersAccessor;
-  friend class internal::UnitTestImpl;
-
-  // Returns repeater that broadcasts the TestEventListener events to all
-  // subscribers.
-  TestEventListener* repeater();
-
-  // Sets the default_result_printer attribute to the provided listener.
-  // The listener is also added to the listener list and previous
-  // default_result_printer is removed from it and deleted. The listener can
-  // also be NULL in which case it will not be added to the list. Does
-  // nothing if the previous and the current listener objects are the same.
-  void SetDefaultResultPrinter(TestEventListener* listener);
-
-  // Sets the default_xml_generator attribute to the provided listener.  The
-  // listener is also added to the listener list and previous
-  // default_xml_generator is removed from it and deleted. The listener can
-  // also be NULL in which case it will not be added to the list. Does
-  // nothing if the previous and the current listener objects are the same.
-  void SetDefaultXmlGenerator(TestEventListener* listener);
-
-  // Controls whether events will be forwarded by the repeater to the
-  // listeners in the list.
-  bool EventForwardingEnabled() const;
-  void SuppressEventForwarding();
-
-  // The actual list of listeners.
-  internal::TestEventRepeater* repeater_;
-  // Listener responsible for the standard result output.
-  TestEventListener* default_result_printer_;
-  // Listener responsible for the creation of the XML output file.
-  TestEventListener* default_xml_generator_;
-
-  // We disallow copying TestEventListeners.
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestEventListeners);
-};
-
-// A UnitTest consists of a vector of TestCases.
-//
-// This is a singleton class.  The only instance of UnitTest is
-// created when UnitTest::GetInstance() is first called.  This
-// instance is never deleted.
-//
-// UnitTest is not copyable.
-//
-// This class is thread-safe as long as the methods are called
-// according to their specification.
-class GTEST_API_ UnitTest {
- public:
-  // Gets the singleton UnitTest object.  The first time this method
-  // is called, a UnitTest object is constructed and returned.
-  // Consecutive calls will return the same object.
-  static UnitTest* GetInstance();
-
-  // Runs all tests in this UnitTest object and prints the result.
-  // Returns 0 if successful, or 1 otherwise.
-  //
-  // This method can only be called from the main thread.
-  //
-  // INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-  int Run() GTEST_MUST_USE_RESULT_;
-
-  // Returns the working directory when the first TEST() or TEST_F()
-  // was executed.  The UnitTest object owns the string.
-  const char* original_working_dir() const;
-
-  // Returns the TestCase object for the test that's currently running,
-  // or NULL if no test is running.
-  const TestCase* current_test_case() const
-      GTEST_LOCK_EXCLUDED_(mutex_);
-
-  // Returns the TestInfo object for the test that's currently running,
-  // or NULL if no test is running.
-  const TestInfo* current_test_info() const
-      GTEST_LOCK_EXCLUDED_(mutex_);
-
-  // Returns the random seed used at the start of the current test run.
-  int random_seed() const;
-
-#if GTEST_HAS_PARAM_TEST
-  // Returns the ParameterizedTestCaseRegistry object used to keep track of
-  // value-parameterized tests and instantiate and register them.
-  //
-  // INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-  internal::ParameterizedTestCaseRegistry& parameterized_test_registry()
-      GTEST_LOCK_EXCLUDED_(mutex_);
-#endif  // GTEST_HAS_PARAM_TEST
-
-  // Gets the number of successful test cases.
-  int successful_test_case_count() const;
-
-  // Gets the number of failed test cases.
-  int failed_test_case_count() const;
-
-  // Gets the number of all test cases.
-  int total_test_case_count() const;
-
-  // Gets the number of all test cases that contain at least one test
-  // that should run.
-  int test_case_to_run_count() const;
-
-  // Gets the number of successful tests.
-  int successful_test_count() const;
-
-  // Gets the number of failed tests.
-  int failed_test_count() const;
-
-  // Gets the number of disabled tests that will be reported in the XML report.
-  int reportable_disabled_test_count() const;
-
-  // Gets the number of disabled tests.
-  int disabled_test_count() const;
-
-  // Gets the number of tests to be printed in the XML report.
-  int reportable_test_count() const;
-
-  // Gets the number of all tests.
-  int total_test_count() const;
-
-  // Gets the number of tests that should run.
-  int test_to_run_count() const;
-
-  // Gets the time of the test program start, in ms from the start of the
-  // UNIX epoch.
-  TimeInMillis start_timestamp() const;
-
-  // Gets the elapsed time, in milliseconds.
-  TimeInMillis elapsed_time() const;
-
-  // Returns true iff the unit test passed (i.e. all test cases passed).
-  bool Passed() const;
-
-  // Returns true iff the unit test failed (i.e. some test case failed
-  // or something outside of all tests failed).
-  bool Failed() const;
-
-  // Gets the i-th test case among all the test cases. i can range from 0 to
-  // total_test_case_count() - 1. If i is not in that range, returns NULL.
-  const TestCase* GetTestCase(int i) const;
-
-  // Returns the TestResult containing information on test failures and
-  // properties logged outside of individual test cases.
-  const TestResult& ad_hoc_test_result() const;
-
-  // Returns the list of event listeners that can be used to track events
-  // inside Google Test.
-  TestEventListeners& listeners();
-
- private:
-  // Registers and returns a global test environment.  When a test
-  // program is run, all global test environments will be set-up in
-  // the order they were registered.  After all tests in the program
-  // have finished, all global test environments will be torn-down in
-  // the *reverse* order they were registered.
-  //
-  // The UnitTest object takes ownership of the given environment.
-  //
-  // This method can only be called from the main thread.
-  Environment* AddEnvironment(Environment* env);
-
-  // Adds a TestPartResult to the current TestResult object.  All
-  // Google Test assertion macros (e.g. ASSERT_TRUE, EXPECT_EQ, etc)
-  // eventually call this to report their results.  The user code
-  // should use the assertion macros instead of calling this directly.
-  void AddTestPartResult(TestPartResult::Type result_type,
-                         const char* file_name,
-                         int line_number,
-                         const std::string& message,
-                         const std::string& os_stack_trace)
-      GTEST_LOCK_EXCLUDED_(mutex_);
-
-  // Adds a TestProperty to the current TestResult object when invoked from
-  // inside a test, to current TestCase's ad_hoc_test_result_ when invoked
-  // from SetUpTestCase or TearDownTestCase, or to the global property set
-  // when invoked elsewhere.  If the result already contains a property with
-  // the same key, the value will be updated.
-  void RecordProperty(const std::string& key, const std::string& value);
-
-  // Gets the i-th test case among all the test cases. i can range from 0 to
-  // total_test_case_count() - 1. If i is not in that range, returns NULL.
-  TestCase* GetMutableTestCase(int i);
-
-  // Accessors for the implementation object.
-  internal::UnitTestImpl* impl() { return impl_; }
-  const internal::UnitTestImpl* impl() const { return impl_; }
-
-  // These classes and funcions are friends as they need to access private
-  // members of UnitTest.
-  friend class Test;
-  friend class internal::AssertHelper;
-  friend class internal::ScopedTrace;
-  friend class internal::StreamingListenerTest;
-  friend class internal::UnitTestRecordPropertyTestHelper;
-  friend Environment* AddGlobalTestEnvironment(Environment* env);
-  friend internal::UnitTestImpl* internal::GetUnitTestImpl();
-  friend void internal::ReportFailureInUnknownLocation(
-      TestPartResult::Type result_type,
-      const std::string& message);
-
-  // Creates an empty UnitTest.
-  UnitTest();
-
-  // D'tor
-  virtual ~UnitTest();
-
-  // Pushes a trace defined by SCOPED_TRACE() on to the per-thread
-  // Google Test trace stack.
-  void PushGTestTrace(const internal::TraceInfo& trace)
-      GTEST_LOCK_EXCLUDED_(mutex_);
-
-  // Pops a trace from the per-thread Google Test trace stack.
-  void PopGTestTrace()
-      GTEST_LOCK_EXCLUDED_(mutex_);
-
-  // Protects mutable state in *impl_.  This is mutable as some const
-  // methods need to lock it too.
-  mutable internal::Mutex mutex_;
-
-  // Opaque implementation object.  This field is never changed once
-  // the object is constructed.  We don't mark it as const here, as
-  // doing so will cause a warning in the constructor of UnitTest.
-  // Mutable state in *impl_ is protected by mutex_.
-  internal::UnitTestImpl* impl_;
-
-  // We disallow copying UnitTest.
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(UnitTest);
-};
-
-// A convenient wrapper for adding an environment for the test
-// program.
-//
-// You should call this before RUN_ALL_TESTS() is called, probably in
-// main().  If you use gtest_main, you need to call this before main()
-// starts for it to take effect.  For example, you can define a global
-// variable like this:
-//
-//   testing::Environment* const foo_env =
-//       testing::AddGlobalTestEnvironment(new FooEnvironment);
-//
-// However, we strongly recommend you to write your own main() and
-// call AddGlobalTestEnvironment() there, as relying on initialization
-// of global variables makes the code harder to read and may cause
-// problems when you register multiple environments from different
-// translation units and the environments have dependencies among them
-// (remember that the compiler doesn't guarantee the order in which
-// global variables from different translation units are initialized).
-inline Environment* AddGlobalTestEnvironment(Environment* env) {
-  return UnitTest::GetInstance()->AddEnvironment(env);
-}
-
-// Initializes Google Test.  This must be called before calling
-// RUN_ALL_TESTS().  In particular, it parses a command line for the
-// flags that Google Test recognizes.  Whenever a Google Test flag is
-// seen, it is removed from argv, and *argc is decremented.
-//
-// No value is returned.  Instead, the Google Test flag variables are
-// updated.
-//
-// Calling the function for the second time has no user-visible effect.
-GTEST_API_ void InitGoogleTest(int* argc, char** argv);
-
-// This overloaded version can be used in Windows programs compiled in
-// UNICODE mode.
-GTEST_API_ void InitGoogleTest(int* argc, wchar_t** argv);
-
-namespace internal {
-
-// Separate the error generating code from the code path to reduce the stack
-// frame size of CmpHelperEQ. This helps reduce the overhead of some sanitizers
-// when calling EXPECT_* in a tight loop.
-template <typename T1, typename T2>
-AssertionResult CmpHelperEQFailure(const char* lhs_expression,
-                                   const char* rhs_expression,
-                                   const T1& lhs, const T2& rhs) {
-  return EqFailure(lhs_expression,
-                   rhs_expression,
-                   FormatForComparisonFailureMessage(lhs, rhs),
-                   FormatForComparisonFailureMessage(rhs, lhs),
-                   false);
-}
-
-// The helper function for {ASSERT|EXPECT}_EQ.
-template <typename T1, typename T2>
-AssertionResult CmpHelperEQ(const char* lhs_expression,
-                            const char* rhs_expression,
-                            const T1& lhs,
-                            const T2& rhs) {
-GTEST_DISABLE_MSC_WARNINGS_PUSH_(4389 /* signed/unsigned mismatch */)
-  if (lhs == rhs) {
-    return AssertionSuccess();
-  }
-GTEST_DISABLE_MSC_WARNINGS_POP_()
-
-  return CmpHelperEQFailure(lhs_expression, rhs_expression, lhs, rhs);
-}
-
-// With this overloaded version, we allow anonymous enums to be used
-// in {ASSERT|EXPECT}_EQ when compiled with gcc 4, as anonymous enums
-// can be implicitly cast to BiggestInt.
-GTEST_API_ AssertionResult CmpHelperEQ(const char* lhs_expression,
-                                       const char* rhs_expression,
-                                       BiggestInt lhs,
-                                       BiggestInt rhs);
-
-// The helper class for {ASSERT|EXPECT}_EQ.  The template argument
-// lhs_is_null_literal is true iff the first argument to ASSERT_EQ()
-// is a null pointer literal.  The following default implementation is
-// for lhs_is_null_literal being false.
-template <bool lhs_is_null_literal>
-class EqHelper {
- public:
-  // This templatized version is for the general case.
-  template <typename T1, typename T2>
-  static AssertionResult Compare(const char* lhs_expression,
-                                 const char* rhs_expression,
-                                 const T1& lhs,
-                                 const T2& rhs) {
-    return CmpHelperEQ(lhs_expression, rhs_expression, lhs, rhs);
-  }
-
-  // With this overloaded version, we allow anonymous enums to be used
-  // in {ASSERT|EXPECT}_EQ when compiled with gcc 4, as anonymous
-  // enums can be implicitly cast to BiggestInt.
-  //
-  // Even though its body looks the same as the above version, we
-  // cannot merge the two, as it will make anonymous enums unhappy.
-  static AssertionResult Compare(const char* lhs_expression,
-                                 const char* rhs_expression,
-                                 BiggestInt lhs,
-                                 BiggestInt rhs) {
-    return CmpHelperEQ(lhs_expression, rhs_expression, lhs, rhs);
-  }
-};
-
-// This specialization is used when the first argument to ASSERT_EQ()
-// is a null pointer literal, like NULL, false, or 0.
-template <>
-class EqHelper<true> {
- public:
-  // We define two overloaded versions of Compare().  The first
-  // version will be picked when the second argument to ASSERT_EQ() is
-  // NOT a pointer, e.g. ASSERT_EQ(0, AnIntFunction()) or
-  // EXPECT_EQ(false, a_bool).
-  template <typename T1, typename T2>
-  static AssertionResult Compare(
-      const char* lhs_expression,
-      const char* rhs_expression,
-      const T1& lhs,
-      const T2& rhs,
-      // The following line prevents this overload from being considered if T2
-      // is not a pointer type.  We need this because ASSERT_EQ(NULL, my_ptr)
-      // expands to Compare("", "", NULL, my_ptr), which requires a conversion
-      // to match the Secret* in the other overload, which would otherwise make
-      // this template match better.
-      typename EnableIf<!is_pointer<T2>::value>::type* = 0) {
-    return CmpHelperEQ(lhs_expression, rhs_expression, lhs, rhs);
-  }
-
-  // This version will be picked when the second argument to ASSERT_EQ() is a
-  // pointer, e.g. ASSERT_EQ(NULL, a_pointer).
-  template <typename T>
-  static AssertionResult Compare(
-      const char* lhs_expression,
-      const char* rhs_expression,
-      // We used to have a second template parameter instead of Secret*.  That
-      // template parameter would deduce to 'long', making this a better match
-      // than the first overload even without the first overload's EnableIf.
-      // Unfortunately, gcc with -Wconversion-null warns when "passing NULL to
-      // non-pointer argument" (even a deduced integral argument), so the old
-      // implementation caused warnings in user code.
-      Secret* /* lhs (NULL) */,
-      T* rhs) {
-    // We already know that 'lhs' is a null pointer.
-    return CmpHelperEQ(lhs_expression, rhs_expression,
-                       static_cast<T*>(NULL), rhs);
-  }
-};
-
-// Separate the error generating code from the code path to reduce the stack
-// frame size of CmpHelperOP. This helps reduce the overhead of some sanitizers
-// when calling EXPECT_OP in a tight loop.
-template <typename T1, typename T2>
-AssertionResult CmpHelperOpFailure(const char* expr1, const char* expr2,
-                                   const T1& val1, const T2& val2,
-                                   const char* op) {
-  return AssertionFailure()
-         << "Expected: (" << expr1 << ") " << op << " (" << expr2
-         << "), actual: " << FormatForComparisonFailureMessage(val1, val2)
-         << " vs " << FormatForComparisonFailureMessage(val2, val1);
-}
-
-// A macro for implementing the helper functions needed to implement
-// ASSERT_?? and EXPECT_??.  It is here just to avoid copy-and-paste
-// of similar code.
-//
-// For each templatized helper function, we also define an overloaded
-// version for BiggestInt in order to reduce code bloat and allow
-// anonymous enums to be used with {ASSERT|EXPECT}_?? when compiled
-// with gcc 4.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-
-#define GTEST_IMPL_CMP_HELPER_(op_name, op)\
-template <typename T1, typename T2>\
-AssertionResult CmpHelper##op_name(const char* expr1, const char* expr2, \
-                                   const T1& val1, const T2& val2) {\
-  if (val1 op val2) {\
-    return AssertionSuccess();\
-  } else {\
-    return CmpHelperOpFailure(expr1, expr2, val1, val2, #op);\
-  }\
-}\
-GTEST_API_ AssertionResult CmpHelper##op_name(\
-    const char* expr1, const char* expr2, BiggestInt val1, BiggestInt val2)
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-
-// Implements the helper function for {ASSERT|EXPECT}_NE
-GTEST_IMPL_CMP_HELPER_(NE, !=);
-// Implements the helper function for {ASSERT|EXPECT}_LE
-GTEST_IMPL_CMP_HELPER_(LE, <=);
-// Implements the helper function for {ASSERT|EXPECT}_LT
-GTEST_IMPL_CMP_HELPER_(LT, <);
-// Implements the helper function for {ASSERT|EXPECT}_GE
-GTEST_IMPL_CMP_HELPER_(GE, >=);
-// Implements the helper function for {ASSERT|EXPECT}_GT
-GTEST_IMPL_CMP_HELPER_(GT, >);
-
-#undef GTEST_IMPL_CMP_HELPER_
-
-// The helper function for {ASSERT|EXPECT}_STREQ.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-GTEST_API_ AssertionResult CmpHelperSTREQ(const char* s1_expression,
-                                          const char* s2_expression,
-                                          const char* s1,
-                                          const char* s2);
-
-// The helper function for {ASSERT|EXPECT}_STRCASEEQ.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-GTEST_API_ AssertionResult CmpHelperSTRCASEEQ(const char* s1_expression,
-                                              const char* s2_expression,
-                                              const char* s1,
-                                              const char* s2);
-
-// The helper function for {ASSERT|EXPECT}_STRNE.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-GTEST_API_ AssertionResult CmpHelperSTRNE(const char* s1_expression,
-                                          const char* s2_expression,
-                                          const char* s1,
-                                          const char* s2);
-
-// The helper function for {ASSERT|EXPECT}_STRCASENE.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-GTEST_API_ AssertionResult CmpHelperSTRCASENE(const char* s1_expression,
-                                              const char* s2_expression,
-                                              const char* s1,
-                                              const char* s2);
-
-
-// Helper function for *_STREQ on wide strings.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-GTEST_API_ AssertionResult CmpHelperSTREQ(const char* s1_expression,
-                                          const char* s2_expression,
-                                          const wchar_t* s1,
-                                          const wchar_t* s2);
-
-// Helper function for *_STRNE on wide strings.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-GTEST_API_ AssertionResult CmpHelperSTRNE(const char* s1_expression,
-                                          const char* s2_expression,
-                                          const wchar_t* s1,
-                                          const wchar_t* s2);
-
-}  // namespace internal
-
-// IsSubstring() and IsNotSubstring() are intended to be used as the
-// first argument to {EXPECT,ASSERT}_PRED_FORMAT2(), not by
-// themselves.  They check whether needle is a substring of haystack
-// (NULL is considered a substring of itself only), and return an
-// appropriate error message when they fail.
-//
-// The {needle,haystack}_expr arguments are the stringified
-// expressions that generated the two real arguments.
-GTEST_API_ AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const char* needle, const char* haystack);
-GTEST_API_ AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const wchar_t* needle, const wchar_t* haystack);
-GTEST_API_ AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const char* needle, const char* haystack);
-GTEST_API_ AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const wchar_t* needle, const wchar_t* haystack);
-GTEST_API_ AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::string& needle, const ::std::string& haystack);
-GTEST_API_ AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::string& needle, const ::std::string& haystack);
-
-#if GTEST_HAS_STD_WSTRING
-GTEST_API_ AssertionResult IsSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::wstring& needle, const ::std::wstring& haystack);
-GTEST_API_ AssertionResult IsNotSubstring(
-    const char* needle_expr, const char* haystack_expr,
-    const ::std::wstring& needle, const ::std::wstring& haystack);
-#endif  // GTEST_HAS_STD_WSTRING
-
-namespace internal {
-
-// Helper template function for comparing floating-points.
-//
-// Template parameter:
-//
-//   RawType: the raw floating-point type (either float or double)
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-template <typename RawType>
-AssertionResult CmpHelperFloatingPointEQ(const char* lhs_expression,
-                                         const char* rhs_expression,
-                                         RawType lhs_value,
-                                         RawType rhs_value) {
-  const FloatingPoint<RawType> lhs(lhs_value), rhs(rhs_value);
-
-  if (lhs.AlmostEquals(rhs)) {
-    return AssertionSuccess();
-  }
-
-  ::std::stringstream lhs_ss;
-  lhs_ss << std::setprecision(std::numeric_limits<RawType>::digits10 + 2)
-         << lhs_value;
-
-  ::std::stringstream rhs_ss;
-  rhs_ss << std::setprecision(std::numeric_limits<RawType>::digits10 + 2)
-         << rhs_value;
-
-  return EqFailure(lhs_expression,
-                   rhs_expression,
-                   StringStreamToString(&lhs_ss),
-                   StringStreamToString(&rhs_ss),
-                   false);
-}
-
-// Helper function for implementing ASSERT_NEAR.
-//
-// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
-GTEST_API_ AssertionResult DoubleNearPredFormat(const char* expr1,
-                                                const char* expr2,
-                                                const char* abs_error_expr,
-                                                double val1,
-                                                double val2,
-                                                double abs_error);
-
-// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
-// A class that enables one to stream messages to assertion macros
-class GTEST_API_ AssertHelper {
- public:
-  // Constructor.
-  AssertHelper(TestPartResult::Type type,
-               const char* file,
-               int line,
-               const char* message);
-  ~AssertHelper();
-
-  // Message assignment is a semantic trick to enable assertion
-  // streaming; see the GTEST_MESSAGE_ macro below.
-  void operator=(const Message& message) const;
-
- private:
-  // We put our data in a struct so that the size of the AssertHelper class can
-  // be as small as possible.  This is important because gcc is incapable of
-  // re-using stack space even for temporary variables, so every EXPECT_EQ
-  // reserves stack space for another AssertHelper.
-  struct AssertHelperData {
-    AssertHelperData(TestPartResult::Type t,
-                     const char* srcfile,
-                     int line_num,
-                     const char* msg)
-        : type(t), file(srcfile), line(line_num), message(msg) { }
-
-    TestPartResult::Type const type;
-    const char* const file;
-    int const line;
-    std::string const message;
-
-   private:
-    GTEST_DISALLOW_COPY_AND_ASSIGN_(AssertHelperData);
-  };
-
-  AssertHelperData* const data_;
-
-  GTEST_DISALLOW_COPY_AND_ASSIGN_(AssertHelper);
-};
-
-}  // namespace internal
-
-#if GTEST_HAS_PARAM_TEST
-// The pure interface class that all value-parameterized tests inherit from.
-// A value-parameterized class must inherit from both ::testing::Test and
-// ::testing::WithParamInterface. In most cases that just means inheriting
-// from ::testing::TestWithParam, but more complicated test hierarchies
-// may need to inherit from Test and WithParamInterface at different levels.
-//
-// This interface has support for accessing the test parameter value via
-// the GetParam() method.
-//
-// Use it with one of the parameter generator defining functions, like Range(),
-// Values(), ValuesIn(), Bool(), and Combine().
-//
-// class FooTest : public ::testing::TestWithParam<int> {
-//  protected:
-//   FooTest() {
-//     // Can use GetParam() here.
-//   }
-//   virtual ~FooTest() {
-//     // Can use GetParam() here.
-//   }
-//   virtual void SetUp() {
-//     // Can use GetParam() here.
-//   }
-//   virtual void TearDown {
-//     // Can use GetParam() here.
-//   }
-// };
-// TEST_P(FooTest, DoesBar) {
-//   // Can use GetParam() method here.
-//   Foo foo;
-//   ASSERT_TRUE(foo.DoesBar(GetParam()));
-// }
-// INSTANTIATE_TEST_CASE_P(OneToTenRange, FooTest, ::testing::Range(1, 10));
-
-template <typename T>
-class WithParamInterface {
- public:
-  typedef T ParamType;
-  virtual ~WithParamInterface() {}
-
-  // The current parameter value. Is also available in the test fixture's
-  // constructor. This member function is non-static, even though it only
-  // references static data, to reduce the opportunity for incorrect uses
-  // like writing 'WithParamInterface<bool>::GetParam()' for a test that
-  // uses a fixture whose parameter type is int.
-  const ParamType& GetParam() const {
-    GTEST_CHECK_(parameter_ != NULL)
-        << "GetParam() can only be called inside a value-parameterized test "
-        << "-- did you intend to write TEST_P instead of TEST_F?";
-    return *parameter_;
-  }
-
- private:
-  // Sets parameter value. The caller is responsible for making sure the value
-  // remains alive and unchanged throughout the current test.
-  static void SetParam(const ParamType* parameter) {
-    parameter_ = parameter;
-  }
-
-  // Static value used for accessing parameter during a test lifetime.
-  static const ParamType* parameter_;
-
-  // TestClass must be a subclass of WithParamInterface<T> and Test.
-  template <class TestClass> friend class internal::ParameterizedTestFactory;
-};
-
-template <typename T>
-const T* WithParamInterface<T>::parameter_ = NULL;
-
-// Most value-parameterized classes can ignore the existence of
-// WithParamInterface, and can just inherit from ::testing::TestWithParam.
-
-template <typename T>
-class TestWithParam : public Test, public WithParamInterface<T> {
-};
-
-#endif  // GTEST_HAS_PARAM_TEST
-
-// Macros for indicating success/failure in test code.
-
-// ADD_FAILURE unconditionally adds a failure to the current test.
-// SUCCEED generates a success - it doesn't automatically make the
-// current test successful, as a test is only successful when it has
-// no failure.
-//
-// EXPECT_* verifies that a certain condition is satisfied.  If not,
-// it behaves like ADD_FAILURE.  In particular:
-//
-//   EXPECT_TRUE  verifies that a Boolean condition is true.
-//   EXPECT_FALSE verifies that a Boolean condition is false.
-//
-// FAIL and ASSERT_* are similar to ADD_FAILURE and EXPECT_*, except
-// that they will also abort the current function on failure.  People
-// usually want the fail-fast behavior of FAIL and ASSERT_*, but those
-// writing data-driven tests often find themselves using ADD_FAILURE
-// and EXPECT_* more.
-
-// Generates a nonfatal failure with a generic message.
-#define ADD_FAILURE() GTEST_NONFATAL_FAILURE_("Failed")
-
-// Generates a nonfatal failure at the given source file location with
-// a generic message.
-#define ADD_FAILURE_AT(file, line) \
-  GTEST_MESSAGE_AT_(file, line, "Failed", \
-                    ::testing::TestPartResult::kNonFatalFailure)
-
-// Generates a fatal failure with a generic message.
-#define GTEST_FAIL() GTEST_FATAL_FAILURE_("Failed")
-
-// Define this macro to 1 to omit the definition of FAIL(), which is a
-// generic name and clashes with some other libraries.
-#if !GTEST_DONT_DEFINE_FAIL
-# define FAIL() GTEST_FAIL()
-#endif
-
-// Generates a success with a generic message.
-#define GTEST_SUCCEED() GTEST_SUCCESS_("Succeeded")
-
-// Define this macro to 1 to omit the definition of SUCCEED(), which
-// is a generic name and clashes with some other libraries.
-#if !GTEST_DONT_DEFINE_SUCCEED
-# define SUCCEED() GTEST_SUCCEED()
-#endif
-
-// Macros for testing exceptions.
-//
-//    * {ASSERT|EXPECT}_THROW(statement, expected_exception):
-//         Tests that the statement throws the expected exception.
-//    * {ASSERT|EXPECT}_NO_THROW(statement):
-//         Tests that the statement doesn't throw any exception.
-//    * {ASSERT|EXPECT}_ANY_THROW(statement):
-//         Tests that the statement throws an exception.
-
-#define EXPECT_THROW(statement, expected_exception) \
-  GTEST_TEST_THROW_(statement, expected_exception, GTEST_NONFATAL_FAILURE_)
-#define EXPECT_NO_THROW(statement) \
-  GTEST_TEST_NO_THROW_(statement, GTEST_NONFATAL_FAILURE_)
-#define EXPECT_ANY_THROW(statement) \
-  GTEST_TEST_ANY_THROW_(statement, GTEST_NONFATAL_FAILURE_)
-#define ASSERT_THROW(statement, expected_exception) \
-  GTEST_TEST_THROW_(statement, expected_exception, GTEST_FATAL_FAILURE_)
-#define ASSERT_NO_THROW(statement) \
-  GTEST_TEST_NO_THROW_(statement, GTEST_FATAL_FAILURE_)
-#define ASSERT_ANY_THROW(statement) \
-  GTEST_TEST_ANY_THROW_(statement, GTEST_FATAL_FAILURE_)
-
-// Boolean assertions. Condition can be either a Boolean expression or an
-// AssertionResult. For more information on how to use AssertionResult with
-// these macros see comments on that class.
-#define EXPECT_TRUE(condition) \
-  GTEST_TEST_BOOLEAN_((condition), #condition, false, true, \
-                      GTEST_NONFATAL_FAILURE_)
-#define EXPECT_FALSE(condition) \
-  GTEST_TEST_BOOLEAN_(!(condition), #condition, true, false, \
-                      GTEST_NONFATAL_FAILURE_)
-#define ASSERT_TRUE(condition) \
-  GTEST_TEST_BOOLEAN_((condition), #condition, false, true, \
-                      GTEST_FATAL_FAILURE_)
-#define ASSERT_FALSE(condition) \
-  GTEST_TEST_BOOLEAN_(!(condition), #condition, true, false, \
-                      GTEST_FATAL_FAILURE_)
-
-// Includes the auto-generated header that implements a family of
-// generic predicate assertion macros.
-// Copyright 2006, Google Inc.
-// All rights reserved.
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-// This file is AUTOMATICALLY GENERATED on 10/31/2011 by command
-// 'gen_gtest_pred_impl.py 5'.  DO NOT EDIT BY HAND!
-//
-// Implements a family of generic predicate assertion macros.
-
-#ifndef GTEST_INCLUDE_GTEST_GTEST_PRED_IMPL_H_
-#define GTEST_INCLUDE_GTEST_GTEST_PRED_IMPL_H_
-
-// Makes sure this header is not included before gtest.h.
-#ifndef GTEST_INCLUDE_GTEST_GTEST_H_
-# error Do not include gtest_pred_impl.h directly.  Include gtest.h instead.
-#endif  // GTEST_INCLUDE_GTEST_GTEST_H_
-
-// This header implements a family of generic predicate assertion
-// macros:
-//
-//   ASSERT_PRED_FORMAT1(pred_format, v1)
-//   ASSERT_PRED_FORMAT2(pred_format, v1, v2)
-//   ...
-//
-// where pred_format is a function or functor that takes n (in the
-// case of ASSERT_PRED_FORMATn) values and their source expression
-// text, and returns a testing::AssertionResult.  See the definition
-// of ASSERT_EQ in gtest.h for an example.
-//
-// If you don't care about formatting, you can use the more
-// restrictive version:
-//
-//   ASSERT_PRED1(pred, v1)
-//   ASSERT_PRED2(pred, v1, v2)
-//   ...
-//
-// where pred is an n-ary function or functor that returns bool,
-// and the values v1, v2, ..., must support the << operator for
-// streaming to std::ostream.
-//
-// We also define the EXPECT_* variations.
-//
-// For now we only support predicates whose arity is at most 5.
-// Please email googletestframework@googlegroups.com if you need
-// support for higher arities.
-
-// GTEST_ASSERT_ is the basic statement to which all of the assertions
-// in this file reduce.  Don't use this in your code.
-
-#define GTEST_ASSERT_(expression, on_failure) \
-  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
-  if (const ::testing::AssertionResult gtest_ar = (expression)) \
-    ; \
-  else \
-    on_failure(gtest_ar.failure_message())
-
-
-// Helper function for implementing {EXPECT|ASSERT}_PRED1.  Don't use
-// this in your code.
-template <typename Pred,
-          typename T1>
-AssertionResult AssertPred1Helper(const char* pred_text,
-                                  const char* e1,
-                                  Pred pred,
-                                  const T1& v1) {
-  if (pred(v1)) return AssertionSuccess();
-
-  return AssertionFailure() << pred_text << "("
-                            << e1 << ") evaluates to false, where"
-                            << "\n" << e1 << " evaluates to " << v1;
-}
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT1.
-// Don't use this in your code.
-#define GTEST_PRED_FORMAT1_(pred_format, v1, on_failure)\
-  GTEST_ASSERT_(pred_format(#v1, v1), \
-                on_failure)
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED1.  Don't use
-// this in your code.
-#define GTEST_PRED1_(pred, v1, on_failure)\
-  GTEST_ASSERT_(::testing::AssertPred1Helper(#pred, \
-                                             #v1, \
-                                             pred, \
-                                             v1), on_failure)
-
-// Unary predicate assertion macros.
-#define EXPECT_PRED_FORMAT1(pred_format, v1) \
-  GTEST_PRED_FORMAT1_(pred_format, v1, GTEST_NONFATAL_FAILURE_)
-#define EXPECT_PRED1(pred, v1) \
-  GTEST_PRED1_(pred, v1, GTEST_NONFATAL_FAILURE_)
-#define ASSERT_PRED_FORMAT1(pred_format, v1) \
-  GTEST_PRED_FORMAT1_(pred_format, v1, GTEST_FATAL_FAILURE_)
-#define ASSERT_PRED1(pred, v1) \
-  GTEST_PRED1_(pred, v1, GTEST_FATAL_FAILURE_)
-
-
-
-// Helper function for implementing {EXPECT|ASSERT}_PRED2.  Don't use
-// this in your code.
-template <typename Pred,
-          typename T1,
-          typename T2>
-AssertionResult AssertPred2Helper(const char* pred_text,
-                                  const char* e1,
-                                  const char* e2,
-                                  Pred pred,
-                                  const T1& v1,
-                                  const T2& v2) {
-  if (pred(v1, v2)) return AssertionSuccess();
-
-  return AssertionFailure() << pred_text << "("
-                            << e1 << ", "
-                            << e2 << ") evaluates to false, where"
-                            << "\n" << e1 << " evaluates to " << v1
-                            << "\n" << e2 << " evaluates to " << v2;
-}
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT2.
-// Don't use this in your code.
-#define GTEST_PRED_FORMAT2_(pred_format, v1, v2, on_failure)\
-  GTEST_ASSERT_(pred_format(#v1, #v2, v1, v2), \
-                on_failure)
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED2.  Don't use
-// this in your code.
-#define GTEST_PRED2_(pred, v1, v2, on_failure)\
-  GTEST_ASSERT_(::testing::AssertPred2Helper(#pred, \
-                                             #v1, \
-                                             #v2, \
-                                             pred, \
-                                             v1, \
-                                             v2), on_failure)
-
-// Binary predicate assertion macros.
-#define EXPECT_PRED_FORMAT2(pred_format, v1, v2) \
-  GTEST_PRED_FORMAT2_(pred_format, v1, v2, GTEST_NONFATAL_FAILURE_)
-#define EXPECT_PRED2(pred, v1, v2) \
-  GTEST_PRED2_(pred, v1, v2, GTEST_NONFATAL_FAILURE_)
-#define ASSERT_PRED_FORMAT2(pred_format, v1, v2) \
-  GTEST_PRED_FORMAT2_(pred_format, v1, v2, GTEST_FATAL_FAILURE_)
-#define ASSERT_PRED2(pred, v1, v2) \
-  GTEST_PRED2_(pred, v1, v2, GTEST_FATAL_FAILURE_)
-
-
-
-// Helper function for implementing {EXPECT|ASSERT}_PRED3.  Don't use
-// this in your code.
-template <typename Pred,
-          typename T1,
-          typename T2,
-          typename T3>
-AssertionResult AssertPred3Helper(const char* pred_text,
-                                  const char* e1,
-                                  const char* e2,
-                                  const char* e3,
-                                  Pred pred,
-                                  const T1& v1,
-                                  const T2& v2,
-                                  const T3& v3) {
-  if (pred(v1, v2, v3)) return AssertionSuccess();
-
-  return AssertionFailure() << pred_text << "("
-                            << e1 << ", "
-                            << e2 << ", "
-                            << e3 << ") evaluates to false, where"
-                            << "\n" << e1 << " evaluates to " << v1
-                            << "\n" << e2 << " evaluates to " << v2
-                            << "\n" << e3 << " evaluates to " << v3;
-}
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT3.
-// Don't use this in your code.
-#define GTEST_PRED_FORMAT3_(pred_format, v1, v2, v3, on_failure)\
-  GTEST_ASSERT_(pred_format(#v1, #v2, #v3, v1, v2, v3), \
-                on_failure)
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED3.  Don't use
-// this in your code.
-#define GTEST_PRED3_(pred, v1, v2, v3, on_failure)\
-  GTEST_ASSERT_(::testing::AssertPred3Helper(#pred, \
-                                             #v1, \
-                                             #v2, \
-                                             #v3, \
-                                             pred, \
-                                             v1, \
-                                             v2, \
-                                             v3), on_failure)
-
-// Ternary predicate assertion macros.
-#define EXPECT_PRED_FORMAT3(pred_format, v1, v2, v3) \
-  GTEST_PRED_FORMAT3_(pred_format, v1, v2, v3, GTEST_NONFATAL_FAILURE_)
-#define EXPECT_PRED3(pred, v1, v2, v3) \
-  GTEST_PRED3_(pred, v1, v2, v3, GTEST_NONFATAL_FAILURE_)
-#define ASSERT_PRED_FORMAT3(pred_format, v1, v2, v3) \
-  GTEST_PRED_FORMAT3_(pred_format, v1, v2, v3, GTEST_FATAL_FAILURE_)
-#define ASSERT_PRED3(pred, v1, v2, v3) \
-  GTEST_PRED3_(pred, v1, v2, v3, GTEST_FATAL_FAILURE_)
-
-
-
-// Helper function for implementing {EXPECT|ASSERT}_PRED4.  Don't use
-// this in your code.
-template <typename Pred,
-          typename T1,
-          typename T2,
-          typename T3,
-          typename T4>
-AssertionResult AssertPred4Helper(const char* pred_text,
-                                  const char* e1,
-                                  const char* e2,
-                                  const char* e3,
-                                  const char* e4,
-                                  Pred pred,
-                                  const T1& v1,
-                                  const T2& v2,
-                                  const T3& v3,
-                                  const T4& v4) {
-  if (pred(v1, v2, v3, v4)) return AssertionSuccess();
-
-  return AssertionFailure() << pred_text << "("
-                            << e1 << ", "
-                            << e2 << ", "
-                            << e3 << ", "
-                            << e4 << ") evaluates to false, where"
-                            << "\n" << e1 << " evaluates to " << v1
-                            << "\n" << e2 << " evaluates to " << v2
-                            << "\n" << e3 << " evaluates to " << v3
-                            << "\n" << e4 << " evaluates to " << v4;
-}
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT4.
-// Don't use this in your code.
-#define GTEST_PRED_FORMAT4_(pred_format, v1, v2, v3, v4, on_failure)\
-  GTEST_ASSERT_(pred_format(#v1, #v2, #v3, #v4, v1, v2, v3, v4), \
-                on_failure)
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED4.  Don't use
-// this in your code.
-#define GTEST_PRED4_(pred, v1, v2, v3, v4, on_failure)\
-  GTEST_ASSERT_(::testing::AssertPred4Helper(#pred, \
-                                             #v1, \
-                                             #v2, \
-                                             #v3, \
-                                             #v4, \
-                                             pred, \
-                                             v1, \
-                                             v2, \
-                                             v3, \
-                                             v4), on_failure)
-
-// 4-ary predicate assertion macros.
-#define EXPECT_PRED_FORMAT4(pred_format, v1, v2, v3, v4) \
-  GTEST_PRED_FORMAT4_(pred_format, v1, v2, v3, v4, GTEST_NONFATAL_FAILURE_)
-#define EXPECT_PRED4(pred, v1, v2, v3, v4) \
-  GTEST_PRED4_(pred, v1, v2, v3, v4, GTEST_NONFATAL_FAILURE_)
-#define ASSERT_PRED_FORMAT4(pred_format, v1, v2, v3, v4) \
-  GTEST_PRED_FORMAT4_(pred_format, v1, v2, v3, v4, GTEST_FATAL_FAILURE_)
-#define ASSERT_PRED4(pred, v1, v2, v3, v4) \
-  GTEST_PRED4_(pred, v1, v2, v3, v4, GTEST_FATAL_FAILURE_)
-
-
-
-// Helper function for implementing {EXPECT|ASSERT}_PRED5.  Don't use
-// this in your code.
-template <typename Pred,
-          typename T1,
-          typename T2,
-          typename T3,
-          typename T4,
-          typename T5>
-AssertionResult AssertPred5Helper(const char* pred_text,
-                                  const char* e1,
-                                  const char* e2,
-                                  const char* e3,
-                                  const char* e4,
-                                  const char* e5,
-                                  Pred pred,
-                                  const T1& v1,
-                                  const T2& v2,
-                                  const T3& v3,
-                                  const T4& v4,
-                                  const T5& v5) {
-  if (pred(v1, v2, v3, v4, v5)) return AssertionSuccess();
-
-  return AssertionFailure() << pred_text << "("
-                            << e1 << ", "
-                            << e2 << ", "
-                            << e3 << ", "
-                            << e4 << ", "
-                            << e5 << ") evaluates to false, where"
-                            << "\n" << e1 << " evaluates to " << v1
-                            << "\n" << e2 << " evaluates to " << v2
-                            << "\n" << e3 << " evaluates to " << v3
-                            << "\n" << e4 << " evaluates to " << v4
-                            << "\n" << e5 << " evaluates to " << v5;
-}
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT5.
-// Don't use this in your code.
-#define GTEST_PRED_FORMAT5_(pred_format, v1, v2, v3, v4, v5, on_failure)\
-  GTEST_ASSERT_(pred_format(#v1, #v2, #v3, #v4, #v5, v1, v2, v3, v4, v5), \
-                on_failure)
-
-// Internal macro for implementing {EXPECT|ASSERT}_PRED5.  Don't use
-// this in your code.
-#define GTEST_PRED5_(pred, v1, v2, v3, v4, v5, on_failure)\
-  GTEST_ASSERT_(::testing::AssertPred5Helper(#pred, \
-                                             #v1, \
-                                             #v2, \
-                                             #v3, \
-                                             #v4, \
-                                             #v5, \
-                                             pred, \
-                                             v1, \
-                                             v2, \
-                                             v3, \
-                                             v4, \
-                                             v5), on_failure)
-
-// 5-ary predicate assertion macros.
-#define EXPECT_PRED_FORMAT5(pred_format, v1, v2, v3, v4, v5) \
-  GTEST_PRED_FORMAT5_(pred_format, v1, v2, v3, v4, v5, GTEST_NONFATAL_FAILURE_)
-#define EXPECT_PRED5(pred, v1, v2, v3, v4, v5) \
-  GTEST_PRED5_(pred, v1, v2, v3, v4, v5, GTEST_NONFATAL_FAILURE_)
-#define ASSERT_PRED_FORMAT5(pred_format, v1, v2, v3, v4, v5) \
-  GTEST_PRED_FORMAT5_(pred_format, v1, v2, v3, v4, v5, GTEST_FATAL_FAILURE_)
-#define ASSERT_PRED5(pred, v1, v2, v3, v4, v5) \
-  GTEST_PRED5_(pred, v1, v2, v3, v4, v5, GTEST_FATAL_FAILURE_)
-
-
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_PRED_IMPL_H_
-
-// Macros for testing equalities and inequalities.
-//
-//    * {ASSERT|EXPECT}_EQ(v1, v2): Tests that v1 == v2
-//    * {ASSERT|EXPECT}_NE(v1, v2): Tests that v1 != v2
-//    * {ASSERT|EXPECT}_LT(v1, v2): Tests that v1 < v2
-//    * {ASSERT|EXPECT}_LE(v1, v2): Tests that v1 <= v2
-//    * {ASSERT|EXPECT}_GT(v1, v2): Tests that v1 > v2
-//    * {ASSERT|EXPECT}_GE(v1, v2): Tests that v1 >= v2
-//
-// When they are not, Google Test prints both the tested expressions and
-// their actual values.  The values must be compatible built-in types,
-// or you will get a compiler error.  By "compatible" we mean that the
-// values can be compared by the respective operator.
-//
-// Note:
-//
-//   1. It is possible to make a user-defined type work with
-//   {ASSERT|EXPECT}_??(), but that requires overloading the
-//   comparison operators and is thus discouraged by the Google C++
-//   Usage Guide.  Therefore, you are advised to use the
-//   {ASSERT|EXPECT}_TRUE() macro to assert that two objects are
-//   equal.
-//
-//   2. The {ASSERT|EXPECT}_??() macros do pointer comparisons on
-//   pointers (in particular, C strings).  Therefore, if you use it
-//   with two C strings, you are testing how their locations in memory
-//   are related, not how their content is related.  To compare two C
-//   strings by content, use {ASSERT|EXPECT}_STR*().
-//
-//   3. {ASSERT|EXPECT}_EQ(v1, v2) is preferred to
-//   {ASSERT|EXPECT}_TRUE(v1 == v2), as the former tells you
-//   what the actual value is when it fails, and similarly for the
-//   other comparisons.
-//
-//   4. Do not depend on the order in which {ASSERT|EXPECT}_??()
-//   evaluate their arguments, which is undefined.
-//
-//   5. These macros evaluate their arguments exactly once.
-//
-// Examples:
-//
-//   EXPECT_NE(5, Foo());
-//   EXPECT_EQ(NULL, a_pointer);
-//   ASSERT_LT(i, array_size);
-//   ASSERT_GT(records.size(), 0) << "There is no record left.";
-
-#define EXPECT_EQ(val1, val2) \
-  EXPECT_PRED_FORMAT2(::testing::internal:: \
-                      EqHelper<GTEST_IS_NULL_LITERAL_(val1)>::Compare, \
-                      val1, val2)
-#define EXPECT_NE(val1, val2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperNE, val1, val2)
-#define EXPECT_LE(val1, val2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperLE, val1, val2)
-#define EXPECT_LT(val1, val2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperLT, val1, val2)
-#define EXPECT_GE(val1, val2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperGE, val1, val2)
-#define EXPECT_GT(val1, val2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperGT, val1, val2)
-
-#define GTEST_ASSERT_EQ(val1, val2) \
-  ASSERT_PRED_FORMAT2(::testing::internal:: \
-                      EqHelper<GTEST_IS_NULL_LITERAL_(val1)>::Compare, \
-                      val1, val2)
-#define GTEST_ASSERT_NE(val1, val2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperNE, val1, val2)
-#define GTEST_ASSERT_LE(val1, val2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperLE, val1, val2)
-#define GTEST_ASSERT_LT(val1, val2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperLT, val1, val2)
-#define GTEST_ASSERT_GE(val1, val2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperGE, val1, val2)
-#define GTEST_ASSERT_GT(val1, val2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperGT, val1, val2)
-
-// Define macro GTEST_DONT_DEFINE_ASSERT_XY to 1 to omit the definition of
-// ASSERT_XY(), which clashes with some users' own code.
-
-#if !GTEST_DONT_DEFINE_ASSERT_EQ
-# define ASSERT_EQ(val1, val2) GTEST_ASSERT_EQ(val1, val2)
-#endif
-
-#if !GTEST_DONT_DEFINE_ASSERT_NE
-# define ASSERT_NE(val1, val2) GTEST_ASSERT_NE(val1, val2)
-#endif
-
-#if !GTEST_DONT_DEFINE_ASSERT_LE
-# define ASSERT_LE(val1, val2) GTEST_ASSERT_LE(val1, val2)
-#endif
-
-#if !GTEST_DONT_DEFINE_ASSERT_LT
-# define ASSERT_LT(val1, val2) GTEST_ASSERT_LT(val1, val2)
-#endif
-
-#if !GTEST_DONT_DEFINE_ASSERT_GE
-# define ASSERT_GE(val1, val2) GTEST_ASSERT_GE(val1, val2)
-#endif
-
-#if !GTEST_DONT_DEFINE_ASSERT_GT
-# define ASSERT_GT(val1, val2) GTEST_ASSERT_GT(val1, val2)
-#endif
-
-// C-string Comparisons.  All tests treat NULL and any non-NULL string
-// as different.  Two NULLs are equal.
-//
-//    * {ASSERT|EXPECT}_STREQ(s1, s2):     Tests that s1 == s2
-//    * {ASSERT|EXPECT}_STRNE(s1, s2):     Tests that s1 != s2
-//    * {ASSERT|EXPECT}_STRCASEEQ(s1, s2): Tests that s1 == s2, ignoring case
-//    * {ASSERT|EXPECT}_STRCASENE(s1, s2): Tests that s1 != s2, ignoring case
-//
-// For wide or narrow string objects, you can use the
-// {ASSERT|EXPECT}_??() macros.
-//
-// Don't depend on the order in which the arguments are evaluated,
-// which is undefined.
-//
-// These macros evaluate their arguments exactly once.
-
-#define EXPECT_STREQ(s1, s2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperSTREQ, s1, s2)
-#define EXPECT_STRNE(s1, s2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperSTRNE, s1, s2)
-#define EXPECT_STRCASEEQ(s1, s2) \
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperSTRCASEEQ, s1, s2)
-#define EXPECT_STRCASENE(s1, s2)\
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperSTRCASENE, s1, s2)
-
-#define ASSERT_STREQ(s1, s2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperSTREQ, s1, s2)
-#define ASSERT_STRNE(s1, s2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperSTRNE, s1, s2)
-#define ASSERT_STRCASEEQ(s1, s2) \
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperSTRCASEEQ, s1, s2)
-#define ASSERT_STRCASENE(s1, s2)\
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperSTRCASENE, s1, s2)
-
-// Macros for comparing floating-point numbers.
-//
-//    * {ASSERT|EXPECT}_FLOAT_EQ(val1, val2):
-//         Tests that two float values are almost equal.
-//    * {ASSERT|EXPECT}_DOUBLE_EQ(val1, val2):
-//         Tests that two double values are almost equal.
-//    * {ASSERT|EXPECT}_NEAR(v1, v2, abs_error):
-//         Tests that v1 and v2 are within the given distance to each other.
-//
-// Google Test uses ULP-based comparison to automatically pick a default
-// error bound that is appropriate for the operands.  See the
-// FloatingPoint template class in gtest-internal.h if you are
-// interested in the implementation details.
-
-#define EXPECT_FLOAT_EQ(val1, val2)\
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperFloatingPointEQ<float>, \
-                      val1, val2)
-
-#define EXPECT_DOUBLE_EQ(val1, val2)\
-  EXPECT_PRED_FORMAT2(::testing::internal::CmpHelperFloatingPointEQ<double>, \
-                      val1, val2)
-
-#define ASSERT_FLOAT_EQ(val1, val2)\
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperFloatingPointEQ<float>, \
-                      val1, val2)
-
-#define ASSERT_DOUBLE_EQ(val1, val2)\
-  ASSERT_PRED_FORMAT2(::testing::internal::CmpHelperFloatingPointEQ<double>, \
-                      val1, val2)
-
-#define EXPECT_NEAR(val1, val2, abs_error)\
-  EXPECT_PRED_FORMAT3(::testing::internal::DoubleNearPredFormat, \
-                      val1, val2, abs_error)
-
-#define ASSERT_NEAR(val1, val2, abs_error)\
-  ASSERT_PRED_FORMAT3(::testing::internal::DoubleNearPredFormat, \
-                      val1, val2, abs_error)
-
-// These predicate format functions work on floating-point values, and
-// can be used in {ASSERT|EXPECT}_PRED_FORMAT2*(), e.g.
-//
-//   EXPECT_PRED_FORMAT2(testing::DoubleLE, Foo(), 5.0);
-
-// Asserts that val1 is less than, or almost equal to, val2.  Fails
-// otherwise.  In particular, it fails if either val1 or val2 is NaN.
-GTEST_API_ AssertionResult FloatLE(const char* expr1, const char* expr2,
-                                   float val1, float val2);
-GTEST_API_ AssertionResult DoubleLE(const char* expr1, const char* expr2,
-                                    double val1, double val2);
-
-
-#if GTEST_OS_WINDOWS
-
-// Macros that test for HRESULT failure and success, these are only useful
-// on Windows, and rely on Windows SDK macros and APIs to compile.
-//
-//    * {ASSERT|EXPECT}_HRESULT_{SUCCEEDED|FAILED}(expr)
-//
-// When expr unexpectedly fails or succeeds, Google Test prints the
-// expected result and the actual result with both a human-readable
-// string representation of the error, if available, as well as the
-// hex result code.
-# define EXPECT_HRESULT_SUCCEEDED(expr) \
-    EXPECT_PRED_FORMAT1(::testing::internal::IsHRESULTSuccess, (expr))
-
-# define ASSERT_HRESULT_SUCCEEDED(expr) \
-    ASSERT_PRED_FORMAT1(::testing::internal::IsHRESULTSuccess, (expr))
-
-# define EXPECT_HRESULT_FAILED(expr) \
-    EXPECT_PRED_FORMAT1(::testing::internal::IsHRESULTFailure, (expr))
-
-# define ASSERT_HRESULT_FAILED(expr) \
-    ASSERT_PRED_FORMAT1(::testing::internal::IsHRESULTFailure, (expr))
-
-#endif  // GTEST_OS_WINDOWS
-
-// Macros that execute statement and check that it doesn't generate new fatal
-// failures in the current thread.
-//
-//   * {ASSERT|EXPECT}_NO_FATAL_FAILURE(statement);
-//
-// Examples:
-//
-//   EXPECT_NO_FATAL_FAILURE(Process());
-//   ASSERT_NO_FATAL_FAILURE(Process()) << "Process() failed";
-//
-#define ASSERT_NO_FATAL_FAILURE(statement) \
-    GTEST_TEST_NO_FATAL_FAILURE_(statement, GTEST_FATAL_FAILURE_)
-#define EXPECT_NO_FATAL_FAILURE(statement) \
-    GTEST_TEST_NO_FATAL_FAILURE_(statement, GTEST_NONFATAL_FAILURE_)
-
-// Causes a trace (including the source file path, the current line
-// number, and the given message) to be included in every test failure
-// message generated by code in the current scope.  The effect is
-// undone when the control leaves the current scope.
-//
-// The message argument can be anything streamable to std::ostream.
-//
-// In the implementation, we include the current line number as part
-// of the dummy variable name, thus allowing multiple SCOPED_TRACE()s
-// to appear in the same block - as long as they are on different
-// lines.
-#define SCOPED_TRACE(message) \
-  ::testing::internal::ScopedTrace GTEST_CONCAT_TOKEN_(gtest_trace_, __LINE__)(\
-    __FILE__, __LINE__, ::testing::Message() << (message))
-
-// Compile-time assertion for type equality.
-// StaticAssertTypeEq<type1, type2>() compiles iff type1 and type2 are
-// the same type.  The value it returns is not interesting.
-//
-// Instead of making StaticAssertTypeEq a class template, we make it a
-// function template that invokes a helper class template.  This
-// prevents a user from misusing StaticAssertTypeEq<T1, T2> by
-// defining objects of that type.
-//
-// CAVEAT:
-//
-// When used inside a method of a class template,
-// StaticAssertTypeEq<T1, T2>() is effective ONLY IF the method is
-// instantiated.  For example, given:
-//
-//   template <typename T> class Foo {
-//    public:
-//     void Bar() { testing::StaticAssertTypeEq<int, T>(); }
-//   };
-//
-// the code:
-//
-//   void Test1() { Foo<bool> foo; }
-//
-// will NOT generate a compiler error, as Foo<bool>::Bar() is never
-// actually instantiated.  Instead, you need:
-//
-//   void Test2() { Foo<bool> foo; foo.Bar(); }
-//
-// to cause a compiler error.
-template <typename T1, typename T2>
-bool StaticAssertTypeEq() {
-  (void)internal::StaticAssertTypeEqHelper<T1, T2>();
-  return true;
-}
-
-// Defines a test.
-//
-// The first parameter is the name of the test case, and the second
-// parameter is the name of the test within the test case.
-//
-// The convention is to end the test case name with "Test".  For
-// example, a test case for the Foo class can be named FooTest.
-//
-// Test code should appear between braces after an invocation of
-// this macro.  Example:
-//
-//   TEST(FooTest, InitializesCorrectly) {
-//     Foo foo;
-//     EXPECT_TRUE(foo.StatusIsOK());
-//   }
-
-// Note that we call GetTestTypeId() instead of GetTypeId<
-// ::testing::Test>() here to get the type ID of testing::Test.  This
-// is to work around a suspected linker bug when using Google Test as
-// a framework on Mac OS X.  The bug causes GetTypeId<
-// ::testing::Test>() to return different values depending on whether
-// the call is from the Google Test framework itself or from user test
-// code.  GetTestTypeId() is guaranteed to always return the same
-// value, as it always calls GetTypeId<>() from the Google Test
-// framework.
-#define GTEST_TEST(test_case_name, test_name)\
-  GTEST_TEST_(test_case_name, test_name, \
-              ::testing::Test, ::testing::internal::GetTestTypeId())
-
-// Define this macro to 1 to omit the definition of TEST(), which
-// is a generic name and clashes with some other libraries.
-#if !GTEST_DONT_DEFINE_TEST
-# define TEST(test_case_name, test_name) GTEST_TEST(test_case_name, test_name)
-#endif
-
-// Defines a test that uses a test fixture.
-//
-// The first parameter is the name of the test fixture class, which
-// also doubles as the test case name.  The second parameter is the
-// name of the test within the test case.
-//
-// A test fixture class must be declared earlier.  The user should put
-// his test code between braces after using this macro.  Example:
-//
-//   class FooTest : public testing::Test {
-//    protected:
-//     virtual void SetUp() { b_.AddElement(3); }
-//
-//     Foo a_;
-//     Foo b_;
-//   };
-//
-//   TEST_F(FooTest, InitializesCorrectly) {
-//     EXPECT_TRUE(a_.StatusIsOK());
-//   }
-//
-//   TEST_F(FooTest, ReturnsElementCountCorrectly) {
-//     EXPECT_EQ(0, a_.size());
-//     EXPECT_EQ(1, b_.size());
-//   }
-
-#define TEST_F(test_fixture, test_name)\
-  GTEST_TEST_(test_fixture, test_name, test_fixture, \
-              ::testing::internal::GetTypeId<test_fixture>())
-
-}  // namespace testing
-
-// Use this function in main() to run all tests.  It returns 0 if all
-// tests are successful, or 1 otherwise.
-//
-// RUN_ALL_TESTS() should be invoked after the command line has been
-// parsed by InitGoogleTest().
-//
-// This function was formerly a macro; thus, it is in the global
-// namespace and has an all-caps name.
-int RUN_ALL_TESTS() GTEST_MUST_USE_RESULT_;
-
-inline int RUN_ALL_TESTS() {
-  return ::testing::UnitTest::GetInstance()->Run();
-}
-
-#endif  // GTEST_INCLUDE_GTEST_GTEST_H_
diff --git a/lib/clickhouse-cpp/contrib/lz4/CMakeLists.txt b/lib/clickhouse-cpp/contrib/lz4/CMakeLists.txt
deleted file mode 100644
index 7b471da..0000000
--- a/lib/clickhouse-cpp/contrib/lz4/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-ADD_LIBRARY (lz4-lib STATIC
-    lz4.c
-    lz4hc.c
-)
-
-set_property(TARGET lz4-lib PROPERTY POSITION_INDEPENDENT_CODE ON)
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4.c b/lib/clickhouse-cpp/contrib/lz4/lz4.c
deleted file mode 100644
index 08cf6b5..0000000
--- a/lib/clickhouse-cpp/contrib/lz4/lz4.c
+++ /dev/null
@@ -1,1516 +0,0 @@
-/*
-   LZ4 - Fast LZ compression algorithm
-   Copyright (C) 2011-2015, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-   You can contact the author at :
-   - LZ4 source repository : https://github.com/Cyan4973/lz4
-   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
-*/
-
-
-/**************************************
-*  Tuning parameters
-**************************************/
-/*
- * HEAPMODE :
- * Select how default compression functions will allocate memory for their hash table,
- * in memory stack (0:default, fastest), or in memory heap (1:requires malloc()).
- */
-#define HEAPMODE 0
-
-/*
- * ACCELERATION_DEFAULT :
- * Select "acceleration" for LZ4_compress_fast() when parameter value <= 0
- */
-#define ACCELERATION_DEFAULT 1
-
-
-/**************************************
-*  CPU Feature Detection
-**************************************/
-/*
- * LZ4_FORCE_SW_BITCOUNT
- * Define this parameter if your target system or compiler does not support hardware bit count
- */
-#if defined(_MSC_VER) && defined(_WIN32_WCE)   /* Visual Studio for Windows CE does not support Hardware bit count */
-#  define LZ4_FORCE_SW_BITCOUNT
-#endif
-
-
-/**************************************
-*  Includes
-**************************************/
-#include "lz4.h"
-
-
-/**************************************
-*  Compiler Options
-**************************************/
-#ifdef _MSC_VER    /* Visual Studio */
-#  define FORCE_INLINE static __forceinline
-#  include <intrin.h>
-#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
-#  pragma warning(disable : 4293)        /* disable: C4293: too large shift (32-bits) */
-#else
-#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)   /* C99 */
-#    if defined(__GNUC__) || defined(__clang__)
-#      define FORCE_INLINE static inline __attribute__((always_inline))
-#    else
-#      define FORCE_INLINE static inline
-#    endif
-#  else
-#    define FORCE_INLINE static
-#  endif   /* __STDC_VERSION__ */
-#endif  /* _MSC_VER */
-
-/* LZ4_GCC_VERSION is defined into lz4.h */
-#if (LZ4_GCC_VERSION >= 302) || (__INTEL_COMPILER >= 800) || defined(__clang__)
-#  define expect(expr,value)    (__builtin_expect ((expr),(value)) )
-#else
-#  define expect(expr,value)    (expr)
-#endif
-
-#define likely(expr)     expect((expr) != 0, 1)
-#define unlikely(expr)   expect((expr) != 0, 0)
-
-
-/**************************************
-*  Memory routines
-**************************************/
-#include <stdlib.h>   /* malloc, calloc, free */
-#define ALLOCATOR(n,s) calloc(n,s)
-#define FREEMEM        free
-#include <string.h>   /* memset, memcpy */
-#define MEM_INIT       memset
-
-
-/**************************************
-*  Basic Types
-**************************************/
-#if defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)   /* C99 */
-# include <stdint.h>
-  typedef  uint8_t BYTE;
-  typedef uint16_t U16;
-  typedef uint32_t U32;
-  typedef  int32_t S32;
-  typedef uint64_t U64;
-#else
-  typedef unsigned char       BYTE;
-  typedef unsigned short      U16;
-  typedef unsigned int        U32;
-  typedef   signed int        S32;
-  typedef unsigned long long  U64;
-#endif
-
-
-/**************************************
-*  Reading and writing into memory
-**************************************/
-#define STEPSIZE sizeof(size_t)
-
-static unsigned LZ4_64bits(void) { return sizeof(void*)==8; }
-
-static unsigned LZ4_isLittleEndian(void)
-{
-    const union { U32 i; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
-    return one.c[0];
-}
-
-
-static U16 LZ4_read16(const void* memPtr)
-{
-    U16 val16;
-    memcpy(&val16, memPtr, 2);
-    return val16;
-}
-
-static U16 LZ4_readLE16(const void* memPtr)
-{
-    if (LZ4_isLittleEndian())
-    {
-        return LZ4_read16(memPtr);
-    }
-    else
-    {
-        const BYTE* p = (const BYTE*)memPtr;
-        return (U16)((U16)p[0] + (p[1]<<8));
-    }
-}
-
-static void LZ4_writeLE16(void* memPtr, U16 value)
-{
-    if (LZ4_isLittleEndian())
-    {
-        memcpy(memPtr, &value, 2);
-    }
-    else
-    {
-        BYTE* p = (BYTE*)memPtr;
-        p[0] = (BYTE) value;
-        p[1] = (BYTE)(value>>8);
-    }
-}
-
-static U32 LZ4_read32(const void* memPtr)
-{
-    U32 val32;
-    memcpy(&val32, memPtr, 4);
-    return val32;
-}
-
-static U64 LZ4_read64(const void* memPtr)
-{
-    U64 val64;
-    memcpy(&val64, memPtr, 8);
-    return val64;
-}
-
-static size_t LZ4_read_ARCH(const void* p)
-{
-    if (LZ4_64bits())
-        return (size_t)LZ4_read64(p);
-    else
-        return (size_t)LZ4_read32(p);
-}
-
-
-static void LZ4_copy4(void* dstPtr, const void* srcPtr) { memcpy(dstPtr, srcPtr, 4); }
-
-static void LZ4_copy8(void* dstPtr, const void* srcPtr) { memcpy(dstPtr, srcPtr, 8); }
-
-/* customized version of memcpy, which may overwrite up to 7 bytes beyond dstEnd */
-static void LZ4_wildCopy(void* dstPtr, const void* srcPtr, void* dstEnd)
-{
-    BYTE* d = (BYTE*)dstPtr;
-    const BYTE* s = (const BYTE*)srcPtr;
-    BYTE* e = (BYTE*)dstEnd;
-    do { LZ4_copy8(d,s); d+=8; s+=8; } while (d<e);
-}
-
-
-/**************************************
-*  Common Constants
-**************************************/
-#define MINMATCH 4
-
-#define COPYLENGTH 8
-#define LASTLITERALS 5
-#define MFLIMIT (COPYLENGTH+MINMATCH)
-static const int LZ4_minLength = (MFLIMIT+1);
-
-#define KB *(1 <<10)
-#define MB *(1 <<20)
-#define GB *(1U<<30)
-
-#define MAXD_LOG 16
-#define MAX_DISTANCE ((1 << MAXD_LOG) - 1)
-
-#define ML_BITS  4
-#define ML_MASK  ((1U<<ML_BITS)-1)
-#define RUN_BITS (8-ML_BITS)
-#define RUN_MASK ((1U<<RUN_BITS)-1)
-
-
-/**************************************
-*  Common Utils
-**************************************/
-#define LZ4_STATIC_ASSERT(c)    { enum { LZ4_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
-
-
-/**************************************
-*  Common functions
-**************************************/
-static unsigned LZ4_NbCommonBytes (register size_t val)
-{
-    if (LZ4_isLittleEndian())
-    {
-        if (LZ4_64bits())
-        {
-#       if defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            unsigned long r = 0;
-            _BitScanForward64( &r, (U64)val );
-            return (int)(r>>3);
-#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            return (__builtin_ctzll((U64)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 };
-            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
-#       endif
-        }
-        else /* 32 bits */
-        {
-#       if defined(_MSC_VER) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            unsigned long r;
-            _BitScanForward( &r, (U32)val );
-            return (int)(r>>3);
-#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            return (__builtin_ctz((U32)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 };
-            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
-#       endif
-        }
-    }
-    else   /* Big Endian CPU */
-    {
-        if (LZ4_64bits())
-        {
-#       if defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            unsigned long r = 0;
-            _BitScanReverse64( &r, val );
-            return (unsigned)(r>>3);
-#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            return (__builtin_clzll((U64)val) >> 3);
-#       else
-            unsigned r;
-            if (!(val>>32)) { r=4; } else { r=0; val>>=32; }
-            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-        }
-        else /* 32 bits */
-        {
-#       if defined(_MSC_VER) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            unsigned long r = 0;
-            _BitScanReverse( &r, (unsigned long)val );
-            return (unsigned)(r>>3);
-#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
-            return (__builtin_clz((U32)val) >> 3);
-#       else
-            unsigned r;
-            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-        }
-    }
-}
-
-static unsigned LZ4_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* pInLimit)
-{
-    const BYTE* const pStart = pIn;
-
-    while (likely(pIn<pInLimit-(STEPSIZE-1)))
-    {
-        size_t diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
-        if (!diff) { pIn+=STEPSIZE; pMatch+=STEPSIZE; continue; }
-        pIn += LZ4_NbCommonBytes(diff);
-        return (unsigned)(pIn - pStart);
-    }
-
-    if (LZ4_64bits()) if ((pIn<(pInLimit-3)) && (LZ4_read32(pMatch) == LZ4_read32(pIn))) { pIn+=4; pMatch+=4; }
-    if ((pIn<(pInLimit-1)) && (LZ4_read16(pMatch) == LZ4_read16(pIn))) { pIn+=2; pMatch+=2; }
-    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
-    return (unsigned)(pIn - pStart);
-}
-
-
-#ifndef LZ4_COMMONDEFS_ONLY
-/**************************************
-*  Local Constants
-**************************************/
-#define LZ4_HASHLOG   (LZ4_MEMORY_USAGE-2)
-#define HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
-#define HASH_SIZE_U32 (1 << LZ4_HASHLOG)       /* required as macro for static allocation */
-
-static const int LZ4_64Klimit = ((64 KB) + (MFLIMIT-1));
-static const U32 LZ4_skipTrigger = 6;  /* Increase this value ==> compression run slower on incompressible data */
-
-
-/**************************************
-*  Local Structures and types
-**************************************/
-typedef struct {
-    U32 hashTable[HASH_SIZE_U32];
-    U32 currentOffset;
-    U32 initCheck;
-    const BYTE* dictionary;
-    BYTE* bufferStart;   /* obsolete, used for slideInputBuffer */
-    U32 dictSize;
-} LZ4_stream_t_internal;
-
-typedef enum { notLimited = 0, limitedOutput = 1 } limitedOutput_directive;
-typedef enum { byPtr, byU32, byU16 } tableType_t;
-
-typedef enum { noDict = 0, withPrefix64k, usingExtDict } dict_directive;
-typedef enum { noDictIssue = 0, dictSmall } dictIssue_directive;
-
-typedef enum { endOnOutputSize = 0, endOnInputSize = 1 } endCondition_directive;
-typedef enum { full = 0, partial = 1 } earlyEnd_directive;
-
-
-/**************************************
-*  Local Utils
-**************************************/
-int LZ4_versionNumber (void) { return LZ4_VERSION_NUMBER; }
-int LZ4_compressBound(int isize)  { return LZ4_COMPRESSBOUND(isize); }
-int LZ4_sizeofState() { return LZ4_STREAMSIZE; }
-
-
-
-/********************************
-*  Compression functions
-********************************/
-
-static U32 LZ4_hashSequence(U32 sequence, tableType_t const tableType)
-{
-    if (tableType == byU16)
-        return (((sequence) * 2654435761U) >> ((MINMATCH*8)-(LZ4_HASHLOG+1)));
-    else
-        return (((sequence) * 2654435761U) >> ((MINMATCH*8)-LZ4_HASHLOG));
-}
-
-static const U64 prime5bytes = 889523592379ULL;
-static U32 LZ4_hashSequence64(size_t sequence, tableType_t const tableType)
-{
-    const U32 hashLog = (tableType == byU16) ? LZ4_HASHLOG+1 : LZ4_HASHLOG;
-    const U32 hashMask = (1<<hashLog) - 1;
-    return ((sequence * prime5bytes) >> (40 - hashLog)) & hashMask;
-}
-
-static U32 LZ4_hashSequenceT(size_t sequence, tableType_t const tableType)
-{
-    if (LZ4_64bits())
-        return LZ4_hashSequence64(sequence, tableType);
-    return LZ4_hashSequence((U32)sequence, tableType);
-}
-
-static U32 LZ4_hashPosition(const void* p, tableType_t tableType) { return LZ4_hashSequenceT(LZ4_read_ARCH(p), tableType); }
-
-static void LZ4_putPositionOnHash(const BYTE* p, U32 h, void* tableBase, tableType_t const tableType, const BYTE* srcBase)
-{
-    switch (tableType)
-    {
-    case byPtr: { const BYTE** hashTable = (const BYTE**)tableBase; hashTable[h] = p; return; }
-    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = (U32)(p-srcBase); return; }
-    case byU16: { U16* hashTable = (U16*) tableBase; hashTable[h] = (U16)(p-srcBase); return; }
-    }
-}
-
-static void LZ4_putPosition(const BYTE* p, void* tableBase, tableType_t tableType, const BYTE* srcBase)
-{
-    U32 h = LZ4_hashPosition(p, tableType);
-    LZ4_putPositionOnHash(p, h, tableBase, tableType, srcBase);
-}
-
-static const BYTE* LZ4_getPositionOnHash(U32 h, void* tableBase, tableType_t tableType, const BYTE* srcBase)
-{
-    if (tableType == byPtr) { const BYTE** hashTable = (const BYTE**) tableBase; return hashTable[h]; }
-    if (tableType == byU32) { U32* hashTable = (U32*) tableBase; return hashTable[h] + srcBase; }
-    { U16* hashTable = (U16*) tableBase; return hashTable[h] + srcBase; }   /* default, to ensure a return */
-}
-
-static const BYTE* LZ4_getPosition(const BYTE* p, void* tableBase, tableType_t tableType, const BYTE* srcBase)
-{
-    U32 h = LZ4_hashPosition(p, tableType);
-    return LZ4_getPositionOnHash(h, tableBase, tableType, srcBase);
-}
-
-FORCE_INLINE int LZ4_compress_generic(
-                 void* const ctx,
-                 const char* const source,
-                 char* const dest,
-                 const int inputSize,
-                 const int maxOutputSize,
-                 const limitedOutput_directive outputLimited,
-                 const tableType_t tableType,
-                 const dict_directive dict,
-                 const dictIssue_directive dictIssue,
-                 const U32 acceleration)
-{
-    LZ4_stream_t_internal* const dictPtr = (LZ4_stream_t_internal*)ctx;
-
-    const BYTE* ip = (const BYTE*) source;
-    const BYTE* base;
-    const BYTE* lowLimit;
-    const BYTE* const lowRefLimit = ip - dictPtr->dictSize;
-    const BYTE* const dictionary = dictPtr->dictionary;
-    const BYTE* const dictEnd = dictionary + dictPtr->dictSize;
-    const size_t dictDelta = dictEnd - (const BYTE*)source;
-    const BYTE* anchor = (const BYTE*) source;
-    const BYTE* const iend = ip + inputSize;
-    const BYTE* const mflimit = iend - MFLIMIT;
-    const BYTE* const matchlimit = iend - LASTLITERALS;
-
-    BYTE* op = (BYTE*) dest;
-    BYTE* const olimit = op + maxOutputSize;
-
-    U32 forwardH;
-    size_t refDelta=0;
-
-    /* Init conditions */
-    if ((U32)inputSize > (U32)LZ4_MAX_INPUT_SIZE) return 0;   /* Unsupported input size, too large (or negative) */
-    switch(dict)
-    {
-    case noDict:
-    default:
-        base = (const BYTE*)source;
-        lowLimit = (const BYTE*)source;
-        break;
-    case withPrefix64k:
-        base = (const BYTE*)source - dictPtr->currentOffset;
-        lowLimit = (const BYTE*)source - dictPtr->dictSize;
-        break;
-    case usingExtDict:
-        base = (const BYTE*)source - dictPtr->currentOffset;
-        lowLimit = (const BYTE*)source;
-        break;
-    }
-    if ((tableType == byU16) && (inputSize>=LZ4_64Klimit)) return 0;   /* Size too large (not within 64K limit) */
-    if (inputSize<LZ4_minLength) goto _last_literals;                  /* Input too small, no compression (all literals) */
-
-    /* First Byte */
-    LZ4_putPosition(ip, ctx, tableType, base);
-    ip++; forwardH = LZ4_hashPosition(ip, tableType);
-
-    /* Main Loop */
-    for ( ; ; )
-    {
-        const BYTE* match;
-        BYTE* token;
-        {
-            const BYTE* forwardIp = ip;
-            unsigned step = 1;
-            unsigned searchMatchNb = acceleration << LZ4_skipTrigger;
-
-            /* Find a match */
-            do {
-                U32 h = forwardH;
-                ip = forwardIp;
-                forwardIp += step;
-                step = (searchMatchNb++ >> LZ4_skipTrigger);
-
-                if (unlikely(forwardIp > mflimit)) goto _last_literals;
-
-                match = LZ4_getPositionOnHash(h, ctx, tableType, base);
-                if (dict==usingExtDict)
-                {
-                    if (match<(const BYTE*)source)
-                    {
-                        refDelta = dictDelta;
-                        lowLimit = dictionary;
-                    }
-                    else
-                    {
-                        refDelta = 0;
-                        lowLimit = (const BYTE*)source;
-                    }
-                }
-                forwardH = LZ4_hashPosition(forwardIp, tableType);
-                LZ4_putPositionOnHash(ip, h, ctx, tableType, base);
-
-            } while ( ((dictIssue==dictSmall) ? (match < lowRefLimit) : 0)
-                || ((tableType==byU16) ? 0 : (match + MAX_DISTANCE < ip))
-                || (LZ4_read32(match+refDelta) != LZ4_read32(ip)) );
-        }
-
-        /* Catch up */
-        while ((ip>anchor) && (match+refDelta > lowLimit) && (unlikely(ip[-1]==match[refDelta-1]))) { ip--; match--; }
-
-        {
-            /* Encode Literal length */
-            unsigned litLength = (unsigned)(ip - anchor);
-            token = op++;
-            if ((outputLimited) && (unlikely(op + litLength + (2 + 1 + LASTLITERALS) + (litLength/255) > olimit)))
-                return 0;   /* Check output limit */
-            if (litLength>=RUN_MASK)
-            {
-                int len = (int)litLength-RUN_MASK;
-                *token=(RUN_MASK<<ML_BITS);
-                for(; len >= 255 ; len-=255) *op++ = 255;
-                *op++ = (BYTE)len;
-            }
-            else *token = (BYTE)(litLength<<ML_BITS);
-
-            /* Copy Literals */
-            LZ4_wildCopy(op, anchor, op+litLength);
-            op+=litLength;
-        }
-
-_next_match:
-        /* Encode Offset */
-        LZ4_writeLE16(op, (U16)(ip-match)); op+=2;
-
-        /* Encode MatchLength */
-        {
-            unsigned matchLength;
-
-            if ((dict==usingExtDict) && (lowLimit==dictionary))
-            {
-                const BYTE* limit;
-                match += refDelta;
-                limit = ip + (dictEnd-match);
-                if (limit > matchlimit) limit = matchlimit;
-                matchLength = LZ4_count(ip+MINMATCH, match+MINMATCH, limit);
-                ip += MINMATCH + matchLength;
-                if (ip==limit)
-                {
-                    unsigned more = LZ4_count(ip, (const BYTE*)source, matchlimit);
-                    matchLength += more;
-                    ip += more;
-                }
-            }
-            else
-            {
-                matchLength = LZ4_count(ip+MINMATCH, match+MINMATCH, matchlimit);
-                ip += MINMATCH + matchLength;
-            }
-
-            if ((outputLimited) && (unlikely(op + (1 + LASTLITERALS) + (matchLength>>8) > olimit)))
-                return 0;    /* Check output limit */
-            if (matchLength>=ML_MASK)
-            {
-                *token += ML_MASK;
-                matchLength -= ML_MASK;
-                for (; matchLength >= 510 ; matchLength-=510) { *op++ = 255; *op++ = 255; }
-                if (matchLength >= 255) { matchLength-=255; *op++ = 255; }
-                *op++ = (BYTE)matchLength;
-            }
-            else *token += (BYTE)(matchLength);
-        }
-
-        anchor = ip;
-
-        /* Test end of chunk */
-        if (ip > mflimit) break;
-
-        /* Fill table */
-        LZ4_putPosition(ip-2, ctx, tableType, base);
-
-        /* Test next position */
-        match = LZ4_getPosition(ip, ctx, tableType, base);
-        if (dict==usingExtDict)
-        {
-            if (match<(const BYTE*)source)
-            {
-                refDelta = dictDelta;
-                lowLimit = dictionary;
-            }
-            else
-            {
-                refDelta = 0;
-                lowLimit = (const BYTE*)source;
-            }
-        }
-        LZ4_putPosition(ip, ctx, tableType, base);
-        if ( ((dictIssue==dictSmall) ? (match>=lowRefLimit) : 1)
-            && (match+MAX_DISTANCE>=ip)
-            && (LZ4_read32(match+refDelta)==LZ4_read32(ip)) )
-        { token=op++; *token=0; goto _next_match; }
-
-        /* Prepare next loop */
-        forwardH = LZ4_hashPosition(++ip, tableType);
-    }
-
-_last_literals:
-    /* Encode Last Literals */
-    {
-        const size_t lastRun = (size_t)(iend - anchor);
-        if ((outputLimited) && ((op - (BYTE*)dest) + lastRun + 1 + ((lastRun+255-RUN_MASK)/255) > (U32)maxOutputSize))
-            return 0;   /* Check output limit */
-        if (lastRun >= RUN_MASK)
-        {
-            size_t accumulator = lastRun - RUN_MASK;
-            *op++ = RUN_MASK << ML_BITS;
-            for(; accumulator >= 255 ; accumulator-=255) *op++ = 255;
-            *op++ = (BYTE) accumulator;
-        }
-        else
-        {
-            *op++ = (BYTE)(lastRun<<ML_BITS);
-        }
-        memcpy(op, anchor, lastRun);
-        op += lastRun;
-    }
-
-    /* End */
-    return (int) (((char*)op)-dest);
-}
-
-
-int LZ4_compress_fast_extState(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
-{
-    LZ4_resetStream((LZ4_stream_t*)state);
-    if (acceleration < 1) acceleration = ACCELERATION_DEFAULT;
-
-    if (maxOutputSize >= LZ4_compressBound(inputSize))
-    {
-        if (inputSize < LZ4_64Klimit)
-            return LZ4_compress_generic(state, source, dest, inputSize, 0, notLimited, byU16,                        noDict, noDictIssue, acceleration);
-        else
-            return LZ4_compress_generic(state, source, dest, inputSize, 0, notLimited, LZ4_64bits() ? byU32 : byPtr, noDict, noDictIssue, acceleration);
-    }
-    else
-    {
-        if (inputSize < LZ4_64Klimit)
-            return LZ4_compress_generic(state, source, dest, inputSize, maxOutputSize, limitedOutput, byU16,                        noDict, noDictIssue, acceleration);
-        else
-            return LZ4_compress_generic(state, source, dest, inputSize, maxOutputSize, limitedOutput, LZ4_64bits() ? byU32 : byPtr, noDict, noDictIssue, acceleration);
-    }
-}
-
-
-int LZ4_compress_fast(const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
-{
-#if (HEAPMODE)
-    void* ctxPtr = ALLOCATOR(1, sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
-#else
-    LZ4_stream_t ctx;
-    void* ctxPtr = &ctx;
-#endif
-
-    int result = LZ4_compress_fast_extState(ctxPtr, source, dest, inputSize, maxOutputSize, acceleration);
-
-#if (HEAPMODE)
-    FREEMEM(ctxPtr);
-#endif
-    return result;
-}
-
-
-int LZ4_compress_default(const char* source, char* dest, int inputSize, int maxOutputSize)
-{
-    return LZ4_compress_fast(source, dest, inputSize, maxOutputSize, 1);
-}
-
-
-/* hidden debug function */
-/* strangely enough, gcc generates faster code when this function is uncommented, even if unused */
-int LZ4_compress_fast_force(const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
-{
-    LZ4_stream_t ctx;
-
-    LZ4_resetStream(&ctx);
-
-    if (inputSize < LZ4_64Klimit)
-        return LZ4_compress_generic(&ctx, source, dest, inputSize, maxOutputSize, limitedOutput, byU16,                        noDict, noDictIssue, acceleration);
-    else
-        return LZ4_compress_generic(&ctx, source, dest, inputSize, maxOutputSize, limitedOutput, LZ4_64bits() ? byU32 : byPtr, noDict, noDictIssue, acceleration);
-}
-
-
-/********************************
-*  destSize variant
-********************************/
-
-static int LZ4_compress_destSize_generic(
-                       void* const ctx,
-                 const char* const src,
-                       char* const dst,
-                       int*  const srcSizePtr,
-                 const int targetDstSize,
-                 const tableType_t tableType)
-{
-    const BYTE* ip = (const BYTE*) src;
-    const BYTE* base = (const BYTE*) src;
-    const BYTE* lowLimit = (const BYTE*) src;
-    const BYTE* anchor = ip;
-    const BYTE* const iend = ip + *srcSizePtr;
-    const BYTE* const mflimit = iend - MFLIMIT;
-    const BYTE* const matchlimit = iend - LASTLITERALS;
-
-    BYTE* op = (BYTE*) dst;
-    BYTE* const oend = op + targetDstSize;
-    BYTE* const oMaxLit = op + targetDstSize - 2 /* offset */ - 8 /* because 8+MINMATCH==MFLIMIT */ - 1 /* token */;
-    BYTE* const oMaxMatch = op + targetDstSize - (LASTLITERALS + 1 /* token */);
-    BYTE* const oMaxSeq = oMaxLit - 1 /* token */;
-
-    U32 forwardH;
-
-
-    /* Init conditions */
-    if (targetDstSize < 1) return 0;                                     /* Impossible to store anything */
-    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;            /* Unsupported input size, too large (or negative) */
-    if ((tableType == byU16) && (*srcSizePtr>=LZ4_64Klimit)) return 0;   /* Size too large (not within 64K limit) */
-    if (*srcSizePtr<LZ4_minLength) goto _last_literals;                  /* Input too small, no compression (all literals) */
-
-    /* First Byte */
-    *srcSizePtr = 0;
-    LZ4_putPosition(ip, ctx, tableType, base);
-    ip++; forwardH = LZ4_hashPosition(ip, tableType);
-
-    /* Main Loop */
-    for ( ; ; )
-    {
-        const BYTE* match;
-        BYTE* token;
-        {
-            const BYTE* forwardIp = ip;
-            unsigned step = 1;
-            unsigned searchMatchNb = 1 << LZ4_skipTrigger;
-
-            /* Find a match */
-            do {
-                U32 h = forwardH;
-                ip = forwardIp;
-                forwardIp += step;
-                step = (searchMatchNb++ >> LZ4_skipTrigger);
-
-                if (unlikely(forwardIp > mflimit))
-                    goto _last_literals;
-
-                match = LZ4_getPositionOnHash(h, ctx, tableType, base);
-                forwardH = LZ4_hashPosition(forwardIp, tableType);
-                LZ4_putPositionOnHash(ip, h, ctx, tableType, base);
-
-            } while ( ((tableType==byU16) ? 0 : (match + MAX_DISTANCE < ip))
-                || (LZ4_read32(match) != LZ4_read32(ip)) );
-        }
-
-        /* Catch up */
-        while ((ip>anchor) && (match > lowLimit) && (unlikely(ip[-1]==match[-1]))) { ip--; match--; }
-
-        {
-            /* Encode Literal length */
-            unsigned litLength = (unsigned)(ip - anchor);
-            token = op++;
-            if (op + ((litLength+240)/255) + litLength > oMaxLit)
-            {
-                /* Not enough space for a last match */
-                op--;
-                goto _last_literals;
-            }
-            if (litLength>=RUN_MASK)
-            {
-                unsigned len = litLength - RUN_MASK;
-                *token=(RUN_MASK<<ML_BITS);
-                for(; len >= 255 ; len-=255) *op++ = 255;
-                *op++ = (BYTE)len;
-            }
-            else *token = (BYTE)(litLength<<ML_BITS);
-
-            /* Copy Literals */
-            LZ4_wildCopy(op, anchor, op+litLength);
-            op += litLength;
-        }
-
-_next_match:
-        /* Encode Offset */
-        LZ4_writeLE16(op, (U16)(ip-match)); op+=2;
-
-        /* Encode MatchLength */
-        {
-            size_t matchLength;
-
-            matchLength = LZ4_count(ip+MINMATCH, match+MINMATCH, matchlimit);
-
-            if (op + ((matchLength+240)/255) > oMaxMatch)
-            {
-                /* Match description too long : reduce it */
-                matchLength = (15-1) + (oMaxMatch-op) * 255;
-            }
-            //printf("offset %5i, matchLength%5i \n", (int)(ip-match), matchLength + MINMATCH);
-            ip += MINMATCH + matchLength;
-
-            if (matchLength>=ML_MASK)
-            {
-                *token += ML_MASK;
-                matchLength -= ML_MASK;
-                while (matchLength >= 255) { matchLength-=255; *op++ = 255; }
-                *op++ = (BYTE)matchLength;
-            }
-            else *token += (BYTE)(matchLength);
-        }
-
-        anchor = ip;
-
-        /* Test end of block */
-        if (ip > mflimit) break;
-        if (op > oMaxSeq) break;
-
-        /* Fill table */
-        LZ4_putPosition(ip-2, ctx, tableType, base);
-
-        /* Test next position */
-        match = LZ4_getPosition(ip, ctx, tableType, base);
-        LZ4_putPosition(ip, ctx, tableType, base);
-        if ( (match+MAX_DISTANCE>=ip)
-            && (LZ4_read32(match)==LZ4_read32(ip)) )
-        { token=op++; *token=0; goto _next_match; }
-
-        /* Prepare next loop */
-        forwardH = LZ4_hashPosition(++ip, tableType);
-    }
-
-_last_literals:
-    /* Encode Last Literals */
-    {
-        size_t lastRunSize = (size_t)(iend - anchor);
-        if (op + 1 /* token */ + ((lastRunSize+240)/255) /* litLength */ + lastRunSize /* literals */ > oend)
-        {
-            /* adapt lastRunSize to fill 'dst' */
-            lastRunSize  = (oend-op) - 1;
-            lastRunSize -= (lastRunSize+240)/255;
-        }
-        ip = anchor + lastRunSize;
-
-        if (lastRunSize >= RUN_MASK)
-        {
-            size_t accumulator = lastRunSize - RUN_MASK;
-            *op++ = RUN_MASK << ML_BITS;
-            for(; accumulator >= 255 ; accumulator-=255) *op++ = 255;
-            *op++ = (BYTE) accumulator;
-        }
-        else
-        {
-            *op++ = (BYTE)(lastRunSize<<ML_BITS);
-        }
-        memcpy(op, anchor, lastRunSize);
-        op += lastRunSize;
-    }
-
-    /* End */
-    *srcSizePtr = (int) (((const char*)ip)-src);
-    return (int) (((char*)op)-dst);
-}
-
-
-static int LZ4_compress_destSize_extState (void* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize)
-{
-    LZ4_resetStream((LZ4_stream_t*)state);
-
-    if (targetDstSize >= LZ4_compressBound(*srcSizePtr))   /* compression success is guaranteed */
-    {
-        return LZ4_compress_fast_extState(state, src, dst, *srcSizePtr, targetDstSize, 1);
-    }
-    else
-    {
-        if (*srcSizePtr < LZ4_64Klimit)
-            return LZ4_compress_destSize_generic(state, src, dst, srcSizePtr, targetDstSize, byU16);
-        else
-            return LZ4_compress_destSize_generic(state, src, dst, srcSizePtr, targetDstSize, LZ4_64bits() ? byU32 : byPtr);
-    }
-}
-
-
-int LZ4_compress_destSize(const char* src, char* dst, int* srcSizePtr, int targetDstSize)
-{
-#if (HEAPMODE)
-    void* ctx = ALLOCATOR(1, sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
-#else
-    LZ4_stream_t ctxBody;
-    void* ctx = &ctxBody;
-#endif
-
-    int result = LZ4_compress_destSize_extState(ctx, src, dst, srcSizePtr, targetDstSize);
-
-#if (HEAPMODE)
-    FREEMEM(ctx);
-#endif
-    return result;
-}
-
-
-
-/********************************
-*  Streaming functions
-********************************/
-
-LZ4_stream_t* LZ4_createStream(void)
-{
-    LZ4_stream_t* lz4s = (LZ4_stream_t*)ALLOCATOR(8, LZ4_STREAMSIZE_U64);
-    LZ4_STATIC_ASSERT(LZ4_STREAMSIZE >= sizeof(LZ4_stream_t_internal));    /* A compilation error here means LZ4_STREAMSIZE is not large enough */
-    LZ4_resetStream(lz4s);
-    return lz4s;
-}
-
-void LZ4_resetStream (LZ4_stream_t* LZ4_stream)
-{
-    MEM_INIT(LZ4_stream, 0, sizeof(LZ4_stream_t));
-}
-
-int LZ4_freeStream (LZ4_stream_t* LZ4_stream)
-{
-    FREEMEM(LZ4_stream);
-    return (0);
-}
-
-
-#define HASH_UNIT sizeof(size_t)
-int LZ4_loadDict (LZ4_stream_t* LZ4_dict, const char* dictionary, int dictSize)
-{
-    LZ4_stream_t_internal* dict = (LZ4_stream_t_internal*) LZ4_dict;
-    const BYTE* p = (const BYTE*)dictionary;
-    const BYTE* const dictEnd = p + dictSize;
-    const BYTE* base;
-
-    if ((dict->initCheck) || (dict->currentOffset > 1 GB))  /* Uninitialized structure, or reuse overflow */
-        LZ4_resetStream(LZ4_dict);
-
-    if (dictSize < (int)HASH_UNIT)
-    {
-        dict->dictionary = NULL;
-        dict->dictSize = 0;
-        return 0;
-    }
-
-    if ((dictEnd - p) > 64 KB) p = dictEnd - 64 KB;
-    dict->currentOffset += 64 KB;
-    base = p - dict->currentOffset;
-    dict->dictionary = p;
-    dict->dictSize = (U32)(dictEnd - p);
-    dict->currentOffset += dict->dictSize;
-
-    while (p <= dictEnd-HASH_UNIT)
-    {
-        LZ4_putPosition(p, dict->hashTable, byU32, base);
-        p+=3;
-    }
-
-    return dict->dictSize;
-}
-
-
-static void LZ4_renormDictT(LZ4_stream_t_internal* LZ4_dict, const BYTE* src)
-{
-    if ((LZ4_dict->currentOffset > 0x80000000) ||
-        ((size_t)LZ4_dict->currentOffset > (size_t)src))   /* address space overflow */
-    {
-        /* rescale hash table */
-        U32 delta = LZ4_dict->currentOffset - 64 KB;
-        const BYTE* dictEnd = LZ4_dict->dictionary + LZ4_dict->dictSize;
-        int i;
-        for (i=0; i<HASH_SIZE_U32; i++)
-        {
-            if (LZ4_dict->hashTable[i] < delta) LZ4_dict->hashTable[i]=0;
-            else LZ4_dict->hashTable[i] -= delta;
-        }
-        LZ4_dict->currentOffset = 64 KB;
-        if (LZ4_dict->dictSize > 64 KB) LZ4_dict->dictSize = 64 KB;
-        LZ4_dict->dictionary = dictEnd - LZ4_dict->dictSize;
-    }
-}
-
-
-int LZ4_compress_fast_continue (LZ4_stream_t* LZ4_stream, const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
-{
-    LZ4_stream_t_internal* streamPtr = (LZ4_stream_t_internal*)LZ4_stream;
-    const BYTE* const dictEnd = streamPtr->dictionary + streamPtr->dictSize;
-
-    const BYTE* smallest = (const BYTE*) source;
-    if (streamPtr->initCheck) return 0;   /* Uninitialized structure detected */
-    if ((streamPtr->dictSize>0) && (smallest>dictEnd)) smallest = dictEnd;
-    LZ4_renormDictT(streamPtr, smallest);
-    if (acceleration < 1) acceleration = ACCELERATION_DEFAULT;
-
-    /* Check overlapping input/dictionary space */
-    {
-        const BYTE* sourceEnd = (const BYTE*) source + inputSize;
-        if ((sourceEnd > streamPtr->dictionary) && (sourceEnd < dictEnd))
-        {
-            streamPtr->dictSize = (U32)(dictEnd - sourceEnd);
-            if (streamPtr->dictSize > 64 KB) streamPtr->dictSize = 64 KB;
-            if (streamPtr->dictSize < 4) streamPtr->dictSize = 0;
-            streamPtr->dictionary = dictEnd - streamPtr->dictSize;
-        }
-    }
-
-    /* prefix mode : source data follows dictionary */
-    if (dictEnd == (const BYTE*)source)
-    {
-        int result;
-        if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset))
-            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, withPrefix64k, dictSmall, acceleration);
-        else
-            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, withPrefix64k, noDictIssue, acceleration);
-        streamPtr->dictSize += (U32)inputSize;
-        streamPtr->currentOffset += (U32)inputSize;
-        return result;
-    }
-
-    /* external dictionary mode */
-    {
-        int result;
-        if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset))
-            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, usingExtDict, dictSmall, acceleration);
-        else
-            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, usingExtDict, noDictIssue, acceleration);
-        streamPtr->dictionary = (const BYTE*)source;
-        streamPtr->dictSize = (U32)inputSize;
-        streamPtr->currentOffset += (U32)inputSize;
-        return result;
-    }
-}
-
-
-/* Hidden debug function, to force external dictionary mode */
-int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int inputSize)
-{
-    LZ4_stream_t_internal* streamPtr = (LZ4_stream_t_internal*)LZ4_dict;
-    int result;
-    const BYTE* const dictEnd = streamPtr->dictionary + streamPtr->dictSize;
-
-    const BYTE* smallest = dictEnd;
-    if (smallest > (const BYTE*) source) smallest = (const BYTE*) source;
-    LZ4_renormDictT((LZ4_stream_t_internal*)LZ4_dict, smallest);
-
-    result = LZ4_compress_generic(LZ4_dict, source, dest, inputSize, 0, notLimited, byU32, usingExtDict, noDictIssue, 1);
-
-    streamPtr->dictionary = (const BYTE*)source;
-    streamPtr->dictSize = (U32)inputSize;
-    streamPtr->currentOffset += (U32)inputSize;
-
-    return result;
-}
-
-
-int LZ4_saveDict (LZ4_stream_t* LZ4_dict, char* safeBuffer, int dictSize)
-{
-    LZ4_stream_t_internal* dict = (LZ4_stream_t_internal*) LZ4_dict;
-    const BYTE* previousDictEnd = dict->dictionary + dict->dictSize;
-
-    if ((U32)dictSize > 64 KB) dictSize = 64 KB;   /* useless to define a dictionary > 64 KB */
-    if ((U32)dictSize > dict->dictSize) dictSize = dict->dictSize;
-
-    memmove(safeBuffer, previousDictEnd - dictSize, dictSize);
-
-    dict->dictionary = (const BYTE*)safeBuffer;
-    dict->dictSize = (U32)dictSize;
-
-    return dictSize;
-}
-
-
-
-/*******************************
-*  Decompression functions
-*******************************/
-/*
- * This generic decompression function cover all use cases.
- * It shall be instantiated several times, using different sets of directives
- * Note that it is essential this generic function is really inlined,
- * in order to remove useless branches during compilation optimization.
- */
-FORCE_INLINE int LZ4_decompress_generic(
-                 const char* const source,
-                 char* const dest,
-                 int inputSize,
-                 int outputSize,         /* If endOnInput==endOnInputSize, this value is the max size of Output Buffer. */
-
-                 int endOnInput,         /* endOnOutputSize, endOnInputSize */
-                 int partialDecoding,    /* full, partial */
-                 int targetOutputSize,   /* only used if partialDecoding==partial */
-                 int dict,               /* noDict, withPrefix64k, usingExtDict */
-                 const BYTE* const lowPrefix,  /* == dest if dict == noDict */
-                 const BYTE* const dictStart,  /* only if dict==usingExtDict */
-                 const size_t dictSize         /* note : = 0 if noDict */
-                 )
-{
-    /* Local Variables */
-    const BYTE* ip = (const BYTE*) source;
-    const BYTE* const iend = ip + inputSize;
-
-    BYTE* op = (BYTE*) dest;
-    BYTE* const oend = op + outputSize;
-    BYTE* cpy;
-    BYTE* oexit = op + targetOutputSize;
-    const BYTE* const lowLimit = lowPrefix - dictSize;
-
-    const BYTE* const dictEnd = (const BYTE*)dictStart + dictSize;
-    const size_t dec32table[] = {4, 1, 2, 1, 4, 4, 4, 4};
-    const size_t dec64table[] = {0, 0, 0, (size_t)-1, 0, 1, 2, 3};
-
-    const int safeDecode = (endOnInput==endOnInputSize);
-    const int checkOffset = ((safeDecode) && (dictSize < (int)(64 KB)));
-
-
-    /* Special cases */
-    if ((partialDecoding) && (oexit> oend-MFLIMIT)) oexit = oend-MFLIMIT;                         /* targetOutputSize too high => decode everything */
-    if ((endOnInput) && (unlikely(outputSize==0))) return ((inputSize==1) && (*ip==0)) ? 0 : -1;  /* Empty output buffer */
-    if ((!endOnInput) && (unlikely(outputSize==0))) return (*ip==0?1:-1);
-
-
-    /* Main Loop */
-    while (1)
-    {
-        unsigned token;
-        size_t length;
-        const BYTE* match;
-
-        /* get literal length */
-        token = *ip++;
-        if ((length=(token>>ML_BITS)) == RUN_MASK)
-        {
-            unsigned s;
-            do
-            {
-                s = *ip++;
-                length += s;
-            }
-            while (likely((endOnInput)?ip<iend-RUN_MASK:1) && (s==255));
-            if ((safeDecode) && unlikely((size_t)(op+length)<(size_t)(op))) goto _output_error;   /* overflow detection */
-            if ((safeDecode) && unlikely((size_t)(ip+length)<(size_t)(ip))) goto _output_error;   /* overflow detection */
-        }
-
-        /* copy literals */
-        cpy = op+length;
-        if (((endOnInput) && ((cpy>(partialDecoding?oexit:oend-MFLIMIT)) || (ip+length>iend-(2+1+LASTLITERALS))) )
-            || ((!endOnInput) && (cpy>oend-COPYLENGTH)))
-        {
-            if (partialDecoding)
-            {
-                if (cpy > oend) goto _output_error;                           /* Error : write attempt beyond end of output buffer */
-                if ((endOnInput) && (ip+length > iend)) goto _output_error;   /* Error : read attempt beyond end of input buffer */
-            }
-            else
-            {
-                if ((!endOnInput) && (cpy != oend)) goto _output_error;       /* Error : block decoding must stop exactly there */
-                if ((endOnInput) && ((ip+length != iend) || (cpy > oend))) goto _output_error;   /* Error : input must be consumed */
-            }
-            memcpy(op, ip, length);
-            ip += length;
-            op += length;
-            break;     /* Necessarily EOF, due to parsing restrictions */
-        }
-        LZ4_wildCopy(op, ip, cpy);
-        ip += length; op = cpy;
-
-        /* get offset */
-        match = cpy - LZ4_readLE16(ip); ip+=2;
-        if ((checkOffset) && (unlikely(match < lowLimit))) goto _output_error;   /* Error : offset outside destination buffer */
-
-        /* get matchlength */
-        length = token & ML_MASK;
-        if (length == ML_MASK)
-        {
-            unsigned s;
-            do
-            {
-                if ((endOnInput) && (ip > iend-LASTLITERALS)) goto _output_error;
-                s = *ip++;
-                length += s;
-            } while (s==255);
-            if ((safeDecode) && unlikely((size_t)(op+length)<(size_t)op)) goto _output_error;   /* overflow detection */
-        }
-        length += MINMATCH;
-
-        /* check external dictionary */
-        if ((dict==usingExtDict) && (match < lowPrefix))
-        {
-            if (unlikely(op+length > oend-LASTLITERALS)) goto _output_error;   /* doesn't respect parsing restriction */
-
-            if (length <= (size_t)(lowPrefix-match))
-            {
-                /* match can be copied as a single segment from external dictionary */
-                match = dictEnd - (lowPrefix-match);
-                memmove(op, match, length); op += length;
-            }
-            else
-            {
-                /* match encompass external dictionary and current segment */
-                size_t copySize = (size_t)(lowPrefix-match);
-                memcpy(op, dictEnd - copySize, copySize);
-                op += copySize;
-                copySize = length - copySize;
-                if (copySize > (size_t)(op-lowPrefix))   /* overlap within current segment */
-                {
-                    BYTE* const endOfMatch = op + copySize;
-                    const BYTE* copyFrom = lowPrefix;
-                    while (op < endOfMatch) *op++ = *copyFrom++;
-                }
-                else
-                {
-                    memcpy(op, lowPrefix, copySize);
-                    op += copySize;
-                }
-            }
-            continue;
-        }
-
-        /* copy repeated sequence */
-        cpy = op + length;
-        if (unlikely((op-match)<8))
-        {
-            const size_t dec64 = dec64table[op-match];
-            op[0] = match[0];
-            op[1] = match[1];
-            op[2] = match[2];
-            op[3] = match[3];
-            match += dec32table[op-match];
-            LZ4_copy4(op+4, match);
-            op += 8; match -= dec64;
-        } else { LZ4_copy8(op, match); op+=8; match+=8; }
-
-        if (unlikely(cpy>oend-12))
-        {
-            if (cpy > oend-LASTLITERALS) goto _output_error;    /* Error : last LASTLITERALS bytes must be literals */
-            if (op < oend-8)
-            {
-                LZ4_wildCopy(op, match, oend-8);
-                match += (oend-8) - op;
-                op = oend-8;
-            }
-            while (op<cpy) *op++ = *match++;
-        }
-        else
-            LZ4_wildCopy(op, match, cpy);
-        op=cpy;   /* correction */
-    }
-
-    /* end of decoding */
-    if (endOnInput)
-       return (int) (((char*)op)-dest);     /* Nb of output bytes decoded */
-    else
-       return (int) (((const char*)ip)-source);   /* Nb of input bytes read */
-
-    /* Overflow error detected */
-_output_error:
-    return (int) (-(((const char*)ip)-source))-1;
-}
-
-
-int LZ4_decompress_safe(const char* source, char* dest, int compressedSize, int maxDecompressedSize)
-{
-    return LZ4_decompress_generic(source, dest, compressedSize, maxDecompressedSize, endOnInputSize, full, 0, noDict, (BYTE*)dest, NULL, 0);
-}
-
-int LZ4_decompress_safe_partial(const char* source, char* dest, int compressedSize, int targetOutputSize, int maxDecompressedSize)
-{
-    return LZ4_decompress_generic(source, dest, compressedSize, maxDecompressedSize, endOnInputSize, partial, targetOutputSize, noDict, (BYTE*)dest, NULL, 0);
-}
-
-int LZ4_decompress_fast(const char* source, char* dest, int originalSize)
-{
-    return LZ4_decompress_generic(source, dest, 0, originalSize, endOnOutputSize, full, 0, withPrefix64k, (BYTE*)(dest - 64 KB), NULL, 64 KB);
-}
-
-
-/* streaming decompression functions */
-
-typedef struct
-{
-    const BYTE* externalDict;
-    size_t extDictSize;
-    const BYTE* prefixEnd;
-    size_t prefixSize;
-} LZ4_streamDecode_t_internal;
-
-/*
- * If you prefer dynamic allocation methods,
- * LZ4_createStreamDecode()
- * provides a pointer (void*) towards an initialized LZ4_streamDecode_t structure.
- */
-LZ4_streamDecode_t* LZ4_createStreamDecode(void)
-{
-    LZ4_streamDecode_t* lz4s = (LZ4_streamDecode_t*) ALLOCATOR(1, sizeof(LZ4_streamDecode_t));
-    return lz4s;
-}
-
-int LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream)
-{
-    FREEMEM(LZ4_stream);
-    return 0;
-}
-
-/*
- * LZ4_setStreamDecode
- * Use this function to instruct where to find the dictionary
- * This function is not necessary if previous data is still available where it was decoded.
- * Loading a size of 0 is allowed (same effect as no dictionary).
- * Return : 1 if OK, 0 if error
- */
-int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize)
-{
-    LZ4_streamDecode_t_internal* lz4sd = (LZ4_streamDecode_t_internal*) LZ4_streamDecode;
-    lz4sd->prefixSize = (size_t) dictSize;
-    lz4sd->prefixEnd = (const BYTE*) dictionary + dictSize;
-    lz4sd->externalDict = NULL;
-    lz4sd->extDictSize  = 0;
-    return 1;
-}
-
-/*
-*_continue() :
-    These decoding functions allow decompression of multiple blocks in "streaming" mode.
-    Previously decoded blocks must still be available at the memory position where they were decoded.
-    If it's not possible, save the relevant part of decoded data into a safe buffer,
-    and indicate where it stands using LZ4_setStreamDecode()
-*/
-int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int compressedSize, int maxOutputSize)
-{
-    LZ4_streamDecode_t_internal* lz4sd = (LZ4_streamDecode_t_internal*) LZ4_streamDecode;
-    int result;
-
-    if (lz4sd->prefixEnd == (BYTE*)dest)
-    {
-        result = LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
-                                        endOnInputSize, full, 0,
-                                        usingExtDict, lz4sd->prefixEnd - lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
-        if (result <= 0) return result;
-        lz4sd->prefixSize += result;
-        lz4sd->prefixEnd  += result;
-    }
-    else
-    {
-        lz4sd->extDictSize = lz4sd->prefixSize;
-        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
-        result = LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
-                                        endOnInputSize, full, 0,
-                                        usingExtDict, (BYTE*)dest, lz4sd->externalDict, lz4sd->extDictSize);
-        if (result <= 0) return result;
-        lz4sd->prefixSize = result;
-        lz4sd->prefixEnd  = (BYTE*)dest + result;
-    }
-
-    return result;
-}
-
-int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int originalSize)
-{
-    LZ4_streamDecode_t_internal* lz4sd = (LZ4_streamDecode_t_internal*) LZ4_streamDecode;
-    int result;
-
-    if (lz4sd->prefixEnd == (BYTE*)dest)
-    {
-        result = LZ4_decompress_generic(source, dest, 0, originalSize,
-                                        endOnOutputSize, full, 0,
-                                        usingExtDict, lz4sd->prefixEnd - lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
-        if (result <= 0) return result;
-        lz4sd->prefixSize += originalSize;
-        lz4sd->prefixEnd  += originalSize;
-    }
-    else
-    {
-        lz4sd->extDictSize = lz4sd->prefixSize;
-        lz4sd->externalDict = (BYTE*)dest - lz4sd->extDictSize;
-        result = LZ4_decompress_generic(source, dest, 0, originalSize,
-                                        endOnOutputSize, full, 0,
-                                        usingExtDict, (BYTE*)dest, lz4sd->externalDict, lz4sd->extDictSize);
-        if (result <= 0) return result;
-        lz4sd->prefixSize = originalSize;
-        lz4sd->prefixEnd  = (BYTE*)dest + originalSize;
-    }
-
-    return result;
-}
-
-
-/*
-Advanced decoding functions :
-*_usingDict() :
-    These decoding functions work the same as "_continue" ones,
-    the dictionary must be explicitly provided within parameters
-*/
-
-FORCE_INLINE int LZ4_decompress_usingDict_generic(const char* source, char* dest, int compressedSize, int maxOutputSize, int safe, const char* dictStart, int dictSize)
-{
-    if (dictSize==0)
-        return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, noDict, (BYTE*)dest, NULL, 0);
-    if (dictStart+dictSize == dest)
-    {
-        if (dictSize >= (int)(64 KB - 1))
-            return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, withPrefix64k, (BYTE*)dest-64 KB, NULL, 0);
-        return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, noDict, (BYTE*)dest-dictSize, NULL, 0);
-    }
-    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, usingExtDict, (BYTE*)dest, (const BYTE*)dictStart, dictSize);
-}
-
-int LZ4_decompress_safe_usingDict(const char* source, char* dest, int compressedSize, int maxOutputSize, const char* dictStart, int dictSize)
-{
-    return LZ4_decompress_usingDict_generic(source, dest, compressedSize, maxOutputSize, 1, dictStart, dictSize);
-}
-
-int LZ4_decompress_fast_usingDict(const char* source, char* dest, int originalSize, const char* dictStart, int dictSize)
-{
-    return LZ4_decompress_usingDict_generic(source, dest, 0, originalSize, 0, dictStart, dictSize);
-}
-
-/* debug function */
-int LZ4_decompress_safe_forceExtDict(const char* source, char* dest, int compressedSize, int maxOutputSize, const char* dictStart, int dictSize)
-{
-    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, endOnInputSize, full, 0, usingExtDict, (BYTE*)dest, (const BYTE*)dictStart, dictSize);
-}
-
-
-/***************************************************
-*  Obsolete Functions
-***************************************************/
-/* obsolete compression functions */
-int LZ4_compress_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize) { return LZ4_compress_default(source, dest, inputSize, maxOutputSize); }
-int LZ4_compress(const char* source, char* dest, int inputSize) { return LZ4_compress_default(source, dest, inputSize, LZ4_compressBound(inputSize)); }
-int LZ4_compress_limitedOutput_withState (void* state, const char* src, char* dst, int srcSize, int dstSize) { return LZ4_compress_fast_extState(state, src, dst, srcSize, dstSize, 1); }
-int LZ4_compress_withState (void* state, const char* src, char* dst, int srcSize) { return LZ4_compress_fast_extState(state, src, dst, srcSize, LZ4_compressBound(srcSize), 1); }
-int LZ4_compress_limitedOutput_continue (LZ4_stream_t* LZ4_stream, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_fast_continue(LZ4_stream, src, dst, srcSize, maxDstSize, 1); }
-int LZ4_compress_continue (LZ4_stream_t* LZ4_stream, const char* source, char* dest, int inputSize) { return LZ4_compress_fast_continue(LZ4_stream, source, dest, inputSize, LZ4_compressBound(inputSize), 1); }
-
-/*
-These function names are deprecated and should no longer be used.
-They are only provided here for compatibility with older user programs.
-- LZ4_uncompress is totally equivalent to LZ4_decompress_fast
-- LZ4_uncompress_unknownOutputSize is totally equivalent to LZ4_decompress_safe
-*/
-int LZ4_uncompress (const char* source, char* dest, int outputSize) { return LZ4_decompress_fast(source, dest, outputSize); }
-int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize) { return LZ4_decompress_safe(source, dest, isize, maxOutputSize); }
-
-
-/* Obsolete Streaming functions */
-
-int LZ4_sizeofStreamState() { return LZ4_STREAMSIZE; }
-
-static void LZ4_init(LZ4_stream_t_internal* lz4ds, BYTE* base)
-{
-    MEM_INIT(lz4ds, 0, LZ4_STREAMSIZE);
-    lz4ds->bufferStart = base;
-}
-
-int LZ4_resetStreamState(void* state, char* inputBuffer)
-{
-    if ((((size_t)state) & 3) != 0) return 1;   /* Error : pointer is not aligned on 4-bytes boundary */
-    LZ4_init((LZ4_stream_t_internal*)state, (BYTE*)inputBuffer);
-    return 0;
-}
-
-void* LZ4_create (char* inputBuffer)
-{
-    void* lz4ds = ALLOCATOR(8, LZ4_STREAMSIZE_U64);
-    LZ4_init ((LZ4_stream_t_internal*)lz4ds, (BYTE*)inputBuffer);
-    return lz4ds;
-}
-
-char* LZ4_slideInputBuffer (void* LZ4_Data)
-{
-    LZ4_stream_t_internal* ctx = (LZ4_stream_t_internal*)LZ4_Data;
-    int dictSize = LZ4_saveDict((LZ4_stream_t*)LZ4_Data, (char*)ctx->bufferStart, 64 KB);
-    return (char*)(ctx->bufferStart + dictSize);
-}
-
-/* Obsolete streaming decompression functions */
-
-int LZ4_decompress_safe_withPrefix64k(const char* source, char* dest, int compressedSize, int maxOutputSize)
-{
-    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, endOnInputSize, full, 0, withPrefix64k, (BYTE*)dest - 64 KB, NULL, 64 KB);
-}
-
-int LZ4_decompress_fast_withPrefix64k(const char* source, char* dest, int originalSize)
-{
-    return LZ4_decompress_generic(source, dest, 0, originalSize, endOnOutputSize, full, 0, withPrefix64k, (BYTE*)dest - 64 KB, NULL, 64 KB);
-}
-
-#endif   /* LZ4_COMMONDEFS_ONLY */
-
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4.h b/lib/clickhouse-cpp/contrib/lz4/lz4.h
deleted file mode 100644
index 3e74002..0000000
--- a/lib/clickhouse-cpp/contrib/lz4/lz4.h
+++ /dev/null
@@ -1,360 +0,0 @@
-/*
-   LZ4 - Fast LZ compression algorithm
-   Header File
-   Copyright (C) 2011-2015, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-   You can contact the author at :
-   - LZ4 source repository : https://github.com/Cyan4973/lz4
-   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
-*/
-#pragma once
-
-#if defined (__cplusplus)
-extern "C" {
-#endif
-
-/*
- * lz4.h provides block compression functions, and gives full buffer control to programmer.
- * If you need to generate inter-operable compressed data (respecting LZ4 frame specification),
- * and can let the library handle its own memory, please use lz4frame.h instead.
-*/
-
-/**************************************
-*  Version
-**************************************/
-#define LZ4_VERSION_MAJOR    1    /* for breaking interface changes  */
-#define LZ4_VERSION_MINOR    7    /* for new (non-breaking) interface capabilities */
-#define LZ4_VERSION_RELEASE  1    /* for tweaks, bug-fixes, or development */
-#define LZ4_VERSION_NUMBER (LZ4_VERSION_MAJOR *100*100 + LZ4_VERSION_MINOR *100 + LZ4_VERSION_RELEASE)
-int LZ4_versionNumber (void);
-
-/**************************************
-*  Tuning parameter
-**************************************/
-/*
- * LZ4_MEMORY_USAGE :
- * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
- * Increasing memory usage improves compression ratio
- * Reduced memory usage can improve speed, due to cache effect
- * Default value is 14, for 16KB, which nicely fits into Intel x86 L1 cache
- */
-#define LZ4_MEMORY_USAGE 14
-
-
-/**************************************
-*  Simple Functions
-**************************************/
-
-int LZ4_compress_default(const char* source, char* dest, int sourceSize, int maxDestSize);
-int LZ4_decompress_safe (const char* source, char* dest, int compressedSize, int maxDecompressedSize);
-
-/*
-LZ4_compress_default() :
-    Compresses 'sourceSize' bytes from buffer 'source'
-    into already allocated 'dest' buffer of size 'maxDestSize'.
-    Compression is guaranteed to succeed if 'maxDestSize' >= LZ4_compressBound(sourceSize).
-    It also runs faster, so it's a recommended setting.
-    If the function cannot compress 'source' into a more limited 'dest' budget,
-    compression stops *immediately*, and the function result is zero.
-    As a consequence, 'dest' content is not valid.
-    This function never writes outside 'dest' buffer, nor read outside 'source' buffer.
-        sourceSize  : Max supported value is LZ4_MAX_INPUT_VALUE
-        maxDestSize : full or partial size of buffer 'dest' (which must be already allocated)
-        return : the number of bytes written into buffer 'dest' (necessarily <= maxOutputSize)
-              or 0 if compression fails
-
-LZ4_decompress_safe() :
-    compressedSize : is the precise full size of the compressed block.
-    maxDecompressedSize : is the size of destination buffer, which must be already allocated.
-    return : the number of bytes decompressed into destination buffer (necessarily <= maxDecompressedSize)
-             If destination buffer is not large enough, decoding will stop and output an error code (<0).
-             If the source stream is detected malformed, the function will stop decoding and return a negative result.
-             This function is protected against buffer overflow exploits, including malicious data packets.
-             It never writes outside output buffer, nor reads outside input buffer.
-*/
-
-
-/**************************************
-*  Advanced Functions
-**************************************/
-#define LZ4_MAX_INPUT_SIZE        0x7E000000   /* 2 113 929 216 bytes */
-#define LZ4_COMPRESSBOUND(isize)  ((unsigned)(isize) > (unsigned)LZ4_MAX_INPUT_SIZE ? 0 : (isize) + ((isize)/255) + 16)
-
-/*
-LZ4_compressBound() :
-    Provides the maximum size that LZ4 compression may output in a "worst case" scenario (input data not compressible)
-    This function is primarily useful for memory allocation purposes (destination buffer size).
-    Macro LZ4_COMPRESSBOUND() is also provided for compilation-time evaluation (stack memory allocation for example).
-    Note that LZ4_compress_default() compress faster when dest buffer size is >= LZ4_compressBound(srcSize)
-        inputSize  : max supported value is LZ4_MAX_INPUT_SIZE
-        return : maximum output size in a "worst case" scenario
-              or 0, if input size is too large ( > LZ4_MAX_INPUT_SIZE)
-*/
-int LZ4_compressBound(int inputSize);
-
-/*
-LZ4_compress_fast() :
-    Same as LZ4_compress_default(), but allows to select an "acceleration" factor.
-    The larger the acceleration value, the faster the algorithm, but also the lesser the compression.
-    It's a trade-off. It can be fine tuned, with each successive value providing roughly +~3% to speed.
-    An acceleration value of "1" is the same as regular LZ4_compress_default()
-    Values <= 0 will be replaced by ACCELERATION_DEFAULT (see lz4.c), which is 1.
-*/
-int LZ4_compress_fast (const char* source, char* dest, int sourceSize, int maxDestSize, int acceleration);
-
-
-/*
-LZ4_compress_fast_extState() :
-    Same compression function, just using an externally allocated memory space to store compression state.
-    Use LZ4_sizeofState() to know how much memory must be allocated,
-    and allocate it on 8-bytes boundaries (using malloc() typically).
-    Then, provide it as 'void* state' to compression function.
-*/
-int LZ4_sizeofState(void);
-int LZ4_compress_fast_extState (void* state, const char* source, char* dest, int inputSize, int maxDestSize, int acceleration);
-
-
-/*
-LZ4_compress_destSize() :
-    Reverse the logic, by compressing as much data as possible from 'source' buffer
-    into already allocated buffer 'dest' of size 'targetDestSize'.
-    This function either compresses the entire 'source' content into 'dest' if it's large enough,
-    or fill 'dest' buffer completely with as much data as possible from 'source'.
-        *sourceSizePtr : will be modified to indicate how many bytes where read from 'source' to fill 'dest'.
-                         New value is necessarily <= old value.
-        return : Nb bytes written into 'dest' (necessarily <= targetDestSize)
-              or 0 if compression fails
-*/
-int LZ4_compress_destSize (const char* source, char* dest, int* sourceSizePtr, int targetDestSize);
-
-
-/*
-LZ4_decompress_fast() :
-    originalSize : is the original and therefore uncompressed size
-    return : the number of bytes read from the source buffer (in other words, the compressed size)
-             If the source stream is detected malformed, the function will stop decoding and return a negative result.
-             Destination buffer must be already allocated. Its size must be a minimum of 'originalSize' bytes.
-    note : This function fully respect memory boundaries for properly formed compressed data.
-           It is a bit faster than LZ4_decompress_safe().
-           However, it does not provide any protection against intentionally modified data stream (malicious input).
-           Use this function in trusted environment only (data to decode comes from a trusted source).
-*/
-int LZ4_decompress_fast (const char* source, char* dest, int originalSize);
-
-/*
-LZ4_decompress_safe_partial() :
-    This function decompress a compressed block of size 'compressedSize' at position 'source'
-    into destination buffer 'dest' of size 'maxDecompressedSize'.
-    The function tries to stop decompressing operation as soon as 'targetOutputSize' has been reached,
-    reducing decompression time.
-    return : the number of bytes decoded in the destination buffer (necessarily <= maxDecompressedSize)
-       Note : this number can be < 'targetOutputSize' should the compressed block to decode be smaller.
-             Always control how many bytes were decoded.
-             If the source stream is detected malformed, the function will stop decoding and return a negative result.
-             This function never writes outside of output buffer, and never reads outside of input buffer. It is therefore protected against malicious data packets
-*/
-int LZ4_decompress_safe_partial (const char* source, char* dest, int compressedSize, int targetOutputSize, int maxDecompressedSize);
-
-
-/***********************************************
-*  Streaming Compression Functions
-***********************************************/
-#define LZ4_STREAMSIZE_U64 ((1 << (LZ4_MEMORY_USAGE-3)) + 4)
-#define LZ4_STREAMSIZE     (LZ4_STREAMSIZE_U64 * sizeof(long long))
-/*
- * LZ4_stream_t
- * information structure to track an LZ4 stream.
- * important : init this structure content before first use !
- * note : only allocated directly the structure if you are statically linking LZ4
- *        If you are using liblz4 as a DLL, please use below construction methods instead.
- */
-typedef struct { long long table[LZ4_STREAMSIZE_U64]; } LZ4_stream_t;
-
-/*
- * LZ4_resetStream
- * Use this function to init an allocated LZ4_stream_t structure
- */
-void LZ4_resetStream (LZ4_stream_t* streamPtr);
-
-/*
- * LZ4_createStream will allocate and initialize an LZ4_stream_t structure
- * LZ4_freeStream releases its memory.
- * In the context of a DLL (liblz4), please use these methods rather than the static struct.
- * They are more future proof, in case of a change of LZ4_stream_t size.
- */
-LZ4_stream_t* LZ4_createStream(void);
-int           LZ4_freeStream (LZ4_stream_t* streamPtr);
-
-/*
- * LZ4_loadDict
- * Use this function to load a static dictionary into LZ4_stream.
- * Any previous data will be forgotten, only 'dictionary' will remain in memory.
- * Loading a size of 0 is allowed.
- * Return : dictionary size, in bytes (necessarily <= 64 KB)
- */
-int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
-
-/*
- * LZ4_compress_fast_continue
- * Compress buffer content 'src', using data from previously compressed blocks as dictionary to improve compression ratio.
- * Important : Previous data blocks are assumed to still be present and unmodified !
- * 'dst' buffer must be already allocated.
- * If maxDstSize >= LZ4_compressBound(srcSize), compression is guaranteed to succeed, and runs faster.
- * If not, and if compressed data cannot fit into 'dst' buffer size, compression stops, and function returns a zero.
- */
-int LZ4_compress_fast_continue (LZ4_stream_t* streamPtr, const char* src, char* dst, int srcSize, int maxDstSize, int acceleration);
-
-/*
- * LZ4_saveDict
- * If previously compressed data block is not guaranteed to remain available at its memory location
- * save it into a safer place (char* safeBuffer)
- * Note : you don't need to call LZ4_loadDict() afterwards,
- *        dictionary is immediately usable, you can therefore call LZ4_compress_fast_continue()
- * Return : saved dictionary size in bytes (necessarily <= dictSize), or 0 if error
- */
-int LZ4_saveDict (LZ4_stream_t* streamPtr, char* safeBuffer, int dictSize);
-
-
-/************************************************
-*  Streaming Decompression Functions
-************************************************/
-
-#define LZ4_STREAMDECODESIZE_U64  4
-#define LZ4_STREAMDECODESIZE     (LZ4_STREAMDECODESIZE_U64 * sizeof(unsigned long long))
-typedef struct { unsigned long long table[LZ4_STREAMDECODESIZE_U64]; } LZ4_streamDecode_t;
-/*
- * LZ4_streamDecode_t
- * information structure to track an LZ4 stream.
- * init this structure content using LZ4_setStreamDecode or memset() before first use !
- *
- * In the context of a DLL (liblz4) please prefer usage of construction methods below.
- * They are more future proof, in case of a change of LZ4_streamDecode_t size in the future.
- * LZ4_createStreamDecode will allocate and initialize an LZ4_streamDecode_t structure
- * LZ4_freeStreamDecode releases its memory.
- */
-LZ4_streamDecode_t* LZ4_createStreamDecode(void);
-int                 LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream);
-
-/*
- * LZ4_setStreamDecode
- * Use this function to instruct where to find the dictionary.
- * Setting a size of 0 is allowed (same effect as reset).
- * Return : 1 if OK, 0 if error
- */
-int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize);
-
-/*
-*_continue() :
-    These decoding functions allow decompression of multiple blocks in "streaming" mode.
-    Previously decoded blocks *must* remain available at the memory position where they were decoded (up to 64 KB)
-    In the case of a ring buffers, decoding buffer must be either :
-    - Exactly same size as encoding buffer, with same update rule (block boundaries at same positions)
-      In which case, the decoding & encoding ring buffer can have any size, including very small ones ( < 64 KB).
-    - Larger than encoding buffer, by a minimum of maxBlockSize more bytes.
-      maxBlockSize is implementation dependent. It's the maximum size you intend to compress into a single block.
-      In which case, encoding and decoding buffers do not need to be synchronized,
-      and encoding ring buffer can have any size, including small ones ( < 64 KB).
-    - _At least_ 64 KB + 8 bytes + maxBlockSize.
-      In which case, encoding and decoding buffers do not need to be synchronized,
-      and encoding ring buffer can have any size, including larger than decoding buffer.
-    Whenever these conditions are not possible, save the last 64KB of decoded data into a safe buffer,
-    and indicate where it is saved using LZ4_setStreamDecode()
-*/
-int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int compressedSize, int maxDecompressedSize);
-int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int originalSize);
-
-
-/*
-Advanced decoding functions :
-*_usingDict() :
-    These decoding functions work the same as
-    a combination of LZ4_setStreamDecode() followed by LZ4_decompress_x_continue()
-    They are stand-alone. They don't need nor update an LZ4_streamDecode_t structure.
-*/
-int LZ4_decompress_safe_usingDict (const char* source, char* dest, int compressedSize, int maxDecompressedSize, const char* dictStart, int dictSize);
-int LZ4_decompress_fast_usingDict (const char* source, char* dest, int originalSize, const char* dictStart, int dictSize);
-
-
-
-/**************************************
-*  Obsolete Functions
-**************************************/
-/* Deprecate Warnings */
-/* Should these warnings messages be a problem,
-   it is generally possible to disable them,
-   with -Wno-deprecated-declarations for gcc
-   or _CRT_SECURE_NO_WARNINGS in Visual for example.
-   You can also define LZ4_DEPRECATE_WARNING_DEFBLOCK. */
-#ifndef LZ4_DEPRECATE_WARNING_DEFBLOCK
-#  define LZ4_DEPRECATE_WARNING_DEFBLOCK
-#  define LZ4_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
-#  if (LZ4_GCC_VERSION >= 405) || defined(__clang__)
-#    define LZ4_DEPRECATED(message) __attribute__((deprecated(message)))
-#  elif (LZ4_GCC_VERSION >= 301)
-#    define LZ4_DEPRECATED(message) __attribute__((deprecated))
-#  elif defined(_MSC_VER)
-#    define LZ4_DEPRECATED(message) __declspec(deprecated(message))
-#  else
-#    pragma message("WARNING: You need to implement LZ4_DEPRECATED for this compiler")
-#    define LZ4_DEPRECATED(message)
-#  endif
-#endif /* LZ4_DEPRECATE_WARNING_DEFBLOCK */
-
-/* Obsolete compression functions */
-/* These functions are planned to start generate warnings by r131 approximately */
-int LZ4_compress               (const char* source, char* dest, int sourceSize);
-int LZ4_compress_limitedOutput (const char* source, char* dest, int sourceSize, int maxOutputSize);
-int LZ4_compress_withState               (void* state, const char* source, char* dest, int inputSize);
-int LZ4_compress_limitedOutput_withState (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
-int LZ4_compress_continue                (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize);
-int LZ4_compress_limitedOutput_continue  (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
-
-/* Obsolete decompression functions */
-/* These function names are completely deprecated and must no longer be used.
-   They are only provided here for compatibility with older programs.
-    - LZ4_uncompress is the same as LZ4_decompress_fast
-    - LZ4_uncompress_unknownOutputSize is the same as LZ4_decompress_safe
-   These function prototypes are now disabled; uncomment them only if you really need them.
-   It is highly recommended to stop using these prototypes and migrate to maintained ones */
-/* int LZ4_uncompress (const char* source, char* dest, int outputSize); */
-/* int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize); */
-
-/* Obsolete streaming functions; use new streaming interface whenever possible */
-LZ4_DEPRECATED("use LZ4_createStream() instead") void* LZ4_create (char* inputBuffer);
-LZ4_DEPRECATED("use LZ4_createStream() instead") int   LZ4_sizeofStreamState(void);
-LZ4_DEPRECATED("use LZ4_resetStream() instead")  int   LZ4_resetStreamState(void* state, char* inputBuffer);
-LZ4_DEPRECATED("use LZ4_saveDict() instead")     char* LZ4_slideInputBuffer (void* state);
-
-/* Obsolete streaming decoding functions */
-LZ4_DEPRECATED("use LZ4_decompress_safe_usingDict() instead") int LZ4_decompress_safe_withPrefix64k (const char* src, char* dst, int compressedSize, int maxDstSize);
-LZ4_DEPRECATED("use LZ4_decompress_fast_usingDict() instead") int LZ4_decompress_fast_withPrefix64k (const char* src, char* dst, int originalSize);
-
-
-#if defined (__cplusplus)
-}
-#endif
diff --git a/lib/clickhouse-cpp/contrib/lz4/BUCK b/lib/clickhouse-cpp/contrib/lz4/lz4/BUCK
similarity index 100%
rename from lib/clickhouse-cpp/contrib/lz4/BUCK
rename to lib/clickhouse-cpp/contrib/lz4/lz4/BUCK
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4/CMakeLists.txt b/lib/clickhouse-cpp/contrib/lz4/lz4/CMakeLists.txt
new file mode 100644
index 0000000..1ec7fb0
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/lz4/lz4/CMakeLists.txt
@@ -0,0 +1,8 @@
+ADD_LIBRARY (lz4 STATIC
+    lz4.c
+    lz4hc.c
+)
+
+set_property(TARGET lz4 PROPERTY POSITION_INDEPENDENT_CODE ON)
+
+ADD_LIBRARY(lz4::lz4 ALIAS lz4)
diff --git a/lib/clickhouse-cpp/contrib/lz4/LICENSE b/lib/clickhouse-cpp/contrib/lz4/lz4/LICENSE
similarity index 97%
rename from lib/clickhouse-cpp/contrib/lz4/LICENSE
rename to lib/clickhouse-cpp/contrib/lz4/lz4/LICENSE
index 552349d..74c2cdd 100644
--- a/lib/clickhouse-cpp/contrib/lz4/LICENSE
+++ b/lib/clickhouse-cpp/contrib/lz4/lz4/LICENSE
@@ -1,5 +1,5 @@
 LZ4 Library
-Copyright (c) 2011-2014, Yann Collet
+Copyright (c) 2011-2016, Yann Collet
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without modification,
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4/lz4.c b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4.c
new file mode 100644
index 0000000..82ab490
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4.c
@@ -0,0 +1,2402 @@
+/*
+   LZ4 - Fast LZ compression algorithm
+   Copyright (C) 2011-present, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+    - LZ4 homepage : http://www.lz4.org
+    - LZ4 source repository : https://github.com/lz4/lz4
+*/
+
+/*-************************************
+*  Tuning parameters
+**************************************/
+/*
+ * LZ4_HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0:default, fastest), or in memory heap (1:requires malloc()).
+ */
+#ifndef LZ4_HEAPMODE
+#  define LZ4_HEAPMODE 0
+#endif
+
+/*
+ * ACCELERATION_DEFAULT :
+ * Select "acceleration" for LZ4_compress_fast() when parameter value <= 0
+ */
+#define ACCELERATION_DEFAULT 1
+
+
+/*-************************************
+*  CPU Feature Detection
+**************************************/
+/* LZ4_FORCE_MEMORY_ACCESS
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets which assembly generation depends on alignment.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See https://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef LZ4_FORCE_MEMORY_ACCESS   /* can be defined externally */
+#  if defined(__GNUC__) && \
+  ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) \
+  || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define LZ4_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(_WIN32)) || defined(__GNUC__)
+#    define LZ4_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+/*
+ * LZ4_FORCE_SW_BITCOUNT
+ * Define this parameter if your target system or compiler does not support hardware bit count
+ */
+#if defined(_MSC_VER) && defined(_WIN32_WCE)   /* Visual Studio for WinCE doesn't support Hardware bit count */
+#  define LZ4_FORCE_SW_BITCOUNT
+#endif
+
+
+
+/*-************************************
+*  Dependency
+**************************************/
+/*
+ * LZ4_SRC_INCLUDED:
+ * Amalgamation flag, whether lz4.c is included
+ */
+#ifndef LZ4_SRC_INCLUDED
+#  define LZ4_SRC_INCLUDED 1
+#endif
+
+#ifndef LZ4_STATIC_LINKING_ONLY
+#define LZ4_STATIC_LINKING_ONLY
+#endif
+
+#ifndef LZ4_DISABLE_DEPRECATE_WARNINGS
+#define LZ4_DISABLE_DEPRECATE_WARNINGS /* due to LZ4_decompress_safe_withPrefix64k */
+#endif
+
+#define LZ4_STATIC_LINKING_ONLY  /* LZ4_DISTANCE_MAX */
+#include "lz4.h"
+/* see also "memory routines" below */
+
+
+/*-************************************
+*  Compiler Options
+**************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4293)        /* disable: C4293: too large shift (32-bits) */
+#endif  /* _MSC_VER */
+
+#ifndef LZ4_FORCE_INLINE
+#  ifdef _MSC_VER    /* Visual Studio */
+#    define LZ4_FORCE_INLINE static __forceinline
+#  else
+#    if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#      ifdef __GNUC__
+#        define LZ4_FORCE_INLINE static inline __attribute__((always_inline))
+#      else
+#        define LZ4_FORCE_INLINE static inline
+#      endif
+#    else
+#      define LZ4_FORCE_INLINE static
+#    endif /* __STDC_VERSION__ */
+#  endif  /* _MSC_VER */
+#endif /* LZ4_FORCE_INLINE */
+
+/* LZ4_FORCE_O2_GCC_PPC64LE and LZ4_FORCE_O2_INLINE_GCC_PPC64LE
+ * gcc on ppc64le generates an unrolled SIMDized loop for LZ4_wildCopy8,
+ * together with a simple 8-byte copy loop as a fall-back path.
+ * However, this optimization hurts the decompression speed by >30%,
+ * because the execution does not go to the optimized loop
+ * for typical compressible data, and all of the preamble checks
+ * before going to the fall-back path become useless overhead.
+ * This optimization happens only with the -O3 flag, and -O2 generates
+ * a simple 8-byte copy loop.
+ * With gcc on ppc64le, all of the LZ4_decompress_* and LZ4_wildCopy8
+ * functions are annotated with __attribute__((optimize("O2"))),
+ * and also LZ4_wildCopy8 is forcibly inlined, so that the O2 attribute
+ * of LZ4_wildCopy8 does not affect the compression speed.
+ */
+#if defined(__PPC64__) && defined(__LITTLE_ENDIAN__) && defined(__GNUC__) && !defined(__clang__)
+#  define LZ4_FORCE_O2_GCC_PPC64LE __attribute__((optimize("O2")))
+#  define LZ4_FORCE_O2_INLINE_GCC_PPC64LE __attribute__((optimize("O2"))) LZ4_FORCE_INLINE
+#else
+#  define LZ4_FORCE_O2_GCC_PPC64LE
+#  define LZ4_FORCE_O2_INLINE_GCC_PPC64LE static
+#endif
+
+#if (defined(__GNUC__) && (__GNUC__ >= 3)) || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) || defined(__clang__)
+#  define expect(expr,value)    (__builtin_expect ((expr),(value)) )
+#else
+#  define expect(expr,value)    (expr)
+#endif
+
+#ifndef likely
+#define likely(expr)     expect((expr) != 0, 1)
+#endif
+#ifndef unlikely
+#define unlikely(expr)   expect((expr) != 0, 0)
+#endif
+
+
+/*-************************************
+*  Memory routines
+**************************************/
+#include <stdlib.h>   /* malloc, calloc, free */
+#define ALLOC(s)          malloc(s)
+#define ALLOC_AND_ZERO(s) calloc(1,s)
+#define FREEMEM(p)        free(p)
+#include <string.h>   /* memset, memcpy */
+#define MEM_INIT(p,v,s)   memset((p),(v),(s))
+
+
+/*-************************************
+*  Common Constants
+**************************************/
+#define MINMATCH 4
+
+#define WILDCOPYLENGTH 8
+#define LASTLITERALS   5   /* see ../doc/lz4_Block_format.md#parsing-restrictions */
+#define MFLIMIT       12   /* see ../doc/lz4_Block_format.md#parsing-restrictions */
+#define MATCH_SAFEGUARD_DISTANCE  ((2*WILDCOPYLENGTH) - MINMATCH)   /* ensure it's possible to write 2 x wildcopyLength without overflowing output buffer */
+#define FASTLOOP_SAFE_DISTANCE 64
+static const int LZ4_minLength = (MFLIMIT+1);
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define LZ4_DISTANCE_ABSOLUTE_MAX 65535
+#if (LZ4_DISTANCE_MAX > LZ4_DISTANCE_ABSOLUTE_MAX)   /* max supported by LZ4 format */
+#  error "LZ4_DISTANCE_MAX is too big : must be <= 65535"
+#endif
+
+#define ML_BITS  4
+#define ML_MASK  ((1U<<ML_BITS)-1)
+#define RUN_BITS (8-ML_BITS)
+#define RUN_MASK ((1U<<RUN_BITS)-1)
+
+
+/*-************************************
+*  Error detection
+**************************************/
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+#define LZ4_STATIC_ASSERT(c)   { enum { LZ4_static_assert = 1/(int)(!!(c)) }; }   /* use after variable declarations */
+
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG>=2)
+#  include <stdio.h>
+   static int g_debuglog_enable = 1;
+#  define DEBUGLOG(l, ...) {                          \
+        if ((g_debuglog_enable) && (l<=LZ4_DEBUG)) {  \
+            fprintf(stderr, __FILE__ ": ");           \
+            fprintf(stderr, __VA_ARGS__);             \
+            fprintf(stderr, " \n");                   \
+    }   }
+#else
+#  define DEBUGLOG(l, ...) {}    /* disabled */
+#endif
+
+
+/*-************************************
+*  Types
+**************************************/
+#if defined(__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef uintptr_t uptrval;
+#else
+# include <limits.h>
+# if UINT_MAX != 4294967295UL
+#   error "LZ4 code (when not C++ or C99) assumes that sizeof(int) == 4"
+# endif
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef size_t              uptrval;   /* generally true, except OpenVMS-64 */
+#endif
+
+#if defined(__x86_64__)
+  typedef U64    reg_t;   /* 64-bits in x32 mode */
+#else
+  typedef size_t reg_t;   /* 32-bits in x32 mode */
+#endif
+
+typedef enum {
+    notLimited = 0,
+    limitedOutput = 1,
+    fillOutput = 2
+} limitedOutput_directive;
+
+
+/*-************************************
+*  Reading and writing into memory
+**************************************/
+static unsigned LZ4_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental */
+    return one.c[0];
+}
+
+
+#if defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==2)
+/* lie to the compiler about data alignment; use with caution */
+
+static U16 LZ4_read16(const void* memPtr) { return *(const U16*) memPtr; }
+static U32 LZ4_read32(const void* memPtr) { return *(const U32*) memPtr; }
+static reg_t LZ4_read_ARCH(const void* memPtr) { return *(const reg_t*) memPtr; }
+
+static void LZ4_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+static void LZ4_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
+
+#elif defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; reg_t uArch; } __attribute__((packed)) unalign;
+
+static U16 LZ4_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+static U32 LZ4_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+static reg_t LZ4_read_ARCH(const void* ptr) { return ((const unalign*)ptr)->uArch; }
+
+static void LZ4_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
+static void LZ4_write32(void* memPtr, U32 value) { ((unalign*)memPtr)->u32 = value; }
+
+#else  /* safe and portable access using memcpy() */
+
+static U16 LZ4_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U32 LZ4_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static reg_t LZ4_read_ARCH(const void* memPtr)
+{
+    reg_t val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static void LZ4_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+static void LZ4_write32(void* memPtr, U32 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif /* LZ4_FORCE_MEMORY_ACCESS */
+
+
+static U16 LZ4_readLE16(const void* memPtr)
+{
+    if (LZ4_isLittleEndian()) {
+        return LZ4_read16(memPtr);
+    } else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)((U16)p[0] + (p[1]<<8));
+    }
+}
+
+static void LZ4_writeLE16(void* memPtr, U16 value)
+{
+    if (LZ4_isLittleEndian()) {
+        LZ4_write16(memPtr, value);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE) value;
+        p[1] = (BYTE)(value>>8);
+    }
+}
+
+/* customized variant of memcpy, which can overwrite up to 8 bytes beyond dstEnd */
+LZ4_FORCE_O2_INLINE_GCC_PPC64LE
+void LZ4_wildCopy8(void* dstPtr, const void* srcPtr, void* dstEnd)
+{
+    BYTE* d = (BYTE*)dstPtr;
+    const BYTE* s = (const BYTE*)srcPtr;
+    BYTE* const e = (BYTE*)dstEnd;
+
+    do { memcpy(d,s,8); d+=8; s+=8; } while (d<e);
+}
+
+static const unsigned inc32table[8] = {0, 1, 2,  1,  0,  4, 4, 4};
+static const int      dec64table[8] = {0, 0, 0, -1, -4,  1, 2, 3};
+
+
+#ifndef LZ4_FAST_DEC_LOOP
+#  if defined __i386__ || defined _M_IX86 || defined __x86_64__ || defined _M_X64
+#    define LZ4_FAST_DEC_LOOP 1
+#  elif defined(__aarch64__) && !defined(__clang__)
+     /* On aarch64, we disable this optimization for clang because on certain
+      * mobile chipsets, performance is reduced with clang. For information
+      * refer to https://github.com/lz4/lz4/pull/707 */
+#    define LZ4_FAST_DEC_LOOP 1
+#  else
+#    define LZ4_FAST_DEC_LOOP 0
+#  endif
+#endif
+
+#if LZ4_FAST_DEC_LOOP
+
+LZ4_FORCE_O2_INLINE_GCC_PPC64LE void
+LZ4_memcpy_using_offset_base(BYTE* dstPtr, const BYTE* srcPtr, BYTE* dstEnd, const size_t offset)
+{
+    if (offset < 8) {
+        dstPtr[0] = srcPtr[0];
+        dstPtr[1] = srcPtr[1];
+        dstPtr[2] = srcPtr[2];
+        dstPtr[3] = srcPtr[3];
+        srcPtr += inc32table[offset];
+        memcpy(dstPtr+4, srcPtr, 4);
+        srcPtr -= dec64table[offset];
+        dstPtr += 8;
+    } else {
+        memcpy(dstPtr, srcPtr, 8);
+        dstPtr += 8;
+        srcPtr += 8;
+    }
+
+    LZ4_wildCopy8(dstPtr, srcPtr, dstEnd);
+}
+
+/* customized variant of memcpy, which can overwrite up to 32 bytes beyond dstEnd
+ * this version copies two times 16 bytes (instead of one time 32 bytes)
+ * because it must be compatible with offsets >= 16. */
+LZ4_FORCE_O2_INLINE_GCC_PPC64LE void
+LZ4_wildCopy32(void* dstPtr, const void* srcPtr, void* dstEnd)
+{
+    BYTE* d = (BYTE*)dstPtr;
+    const BYTE* s = (const BYTE*)srcPtr;
+    BYTE* const e = (BYTE*)dstEnd;
+
+    do { memcpy(d,s,16); memcpy(d+16,s+16,16); d+=32; s+=32; } while (d<e);
+}
+
+/* LZ4_memcpy_using_offset()  presumes :
+ * - dstEnd >= dstPtr + MINMATCH
+ * - there is at least 8 bytes available to write after dstEnd */
+LZ4_FORCE_O2_INLINE_GCC_PPC64LE void
+LZ4_memcpy_using_offset(BYTE* dstPtr, const BYTE* srcPtr, BYTE* dstEnd, const size_t offset)
+{
+    BYTE v[8];
+
+    assert(dstEnd >= dstPtr + MINMATCH);
+    LZ4_write32(dstPtr, 0);   /* silence an msan warning when offset==0 */
+
+    switch(offset) {
+    case 1:
+        memset(v, *srcPtr, 8);
+        break;
+    case 2:
+        memcpy(v, srcPtr, 2);
+        memcpy(&v[2], srcPtr, 2);
+        memcpy(&v[4], &v[0], 4);
+        break;
+    case 4:
+        memcpy(v, srcPtr, 4);
+        memcpy(&v[4], srcPtr, 4);
+        break;
+    default:
+        LZ4_memcpy_using_offset_base(dstPtr, srcPtr, dstEnd, offset);
+        return;
+    }
+
+    memcpy(dstPtr, v, 8);
+    dstPtr += 8;
+    while (dstPtr < dstEnd) {
+        memcpy(dstPtr, v, 8);
+        dstPtr += 8;
+    }
+}
+#endif
+
+
+/*-************************************
+*  Common functions
+**************************************/
+static unsigned LZ4_NbCommonBytes (reg_t val)
+{
+    if (LZ4_isLittleEndian()) {
+        if (sizeof(val)==8) {
+#       if defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanForward64( &r, (U64)val );
+            return (int)(r>>3);
+#       elif (defined(__clang__) || (defined(__GNUC__) && (__GNUC__>=3))) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_ctzll((U64)val) >> 3;
+#       else
+            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
+                                                     0, 3, 1, 3, 1, 4, 2, 7,
+                                                     0, 2, 3, 6, 1, 5, 3, 5,
+                                                     1, 3, 4, 4, 2, 5, 6, 7,
+                                                     7, 0, 1, 2, 3, 3, 4, 6,
+                                                     2, 6, 5, 5, 3, 4, 5, 6,
+                                                     7, 1, 2, 4, 6, 4, 4, 5,
+                                                     7, 2, 6, 5, 7, 6, 7, 7 };
+            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
+#       endif
+        } else /* 32 bits */ {
+#       if defined(_MSC_VER) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r;
+            _BitScanForward( &r, (U32)val );
+            return (int)(r>>3);
+#       elif (defined(__clang__) || (defined(__GNUC__) && (__GNUC__>=3))) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_ctz((U32)val) >> 3;
+#       else
+            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
+                                                     3, 2, 2, 1, 3, 2, 0, 1,
+                                                     3, 3, 1, 2, 2, 2, 2, 0,
+                                                     3, 1, 2, 0, 1, 0, 1, 1 };
+            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
+#       endif
+        }
+    } else   /* Big Endian CPU */ {
+        if (sizeof(val)==8) {   /* 64-bits */
+#       if defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanReverse64( &r, val );
+            return (unsigned)(r>>3);
+#       elif (defined(__clang__) || (defined(__GNUC__) && (__GNUC__>=3))) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_clzll((U64)val) >> 3;
+#       else
+            static const U32 by32 = sizeof(val)*4;  /* 32 on 64 bits (goal), 16 on 32 bits.
+                Just to avoid some static analyzer complaining about shift by 32 on 32-bits target.
+                Note that this code path is never triggered in 32-bits mode. */
+            unsigned r;
+            if (!(val>>by32)) { r=4; } else { r=0; val>>=by32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        } else /* 32 bits */ {
+#       if defined(_MSC_VER) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanReverse( &r, (unsigned long)val );
+            return (unsigned)(r>>3);
+#       elif (defined(__clang__) || (defined(__GNUC__) && (__GNUC__>=3))) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_clz((U32)val) >> 3;
+#       else
+            unsigned r;
+            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        }
+    }
+}
+
+#define STEPSIZE sizeof(reg_t)
+LZ4_FORCE_INLINE
+unsigned LZ4_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* pInLimit)
+{
+    const BYTE* const pStart = pIn;
+
+    if (likely(pIn < pInLimit-(STEPSIZE-1))) {
+        reg_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
+        if (!diff) {
+            pIn+=STEPSIZE; pMatch+=STEPSIZE;
+        } else {
+            return LZ4_NbCommonBytes(diff);
+    }   }
+
+    while (likely(pIn < pInLimit-(STEPSIZE-1))) {
+        reg_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
+        if (!diff) { pIn+=STEPSIZE; pMatch+=STEPSIZE; continue; }
+        pIn += LZ4_NbCommonBytes(diff);
+        return (unsigned)(pIn - pStart);
+    }
+
+    if ((STEPSIZE==8) && (pIn<(pInLimit-3)) && (LZ4_read32(pMatch) == LZ4_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (LZ4_read16(pMatch) == LZ4_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (unsigned)(pIn - pStart);
+}
+
+
+#ifndef LZ4_COMMONDEFS_ONLY
+/*-************************************
+*  Local Constants
+**************************************/
+static const int LZ4_64Klimit = ((64 KB) + (MFLIMIT-1));
+static const U32 LZ4_skipTrigger = 6;  /* Increase this value ==> compression run slower on incompressible data */
+
+
+/*-************************************
+*  Local Structures and types
+**************************************/
+typedef enum { clearedTable = 0, byPtr, byU32, byU16 } tableType_t;
+
+/**
+ * This enum distinguishes several different modes of accessing previous
+ * content in the stream.
+ *
+ * - noDict        : There is no preceding content.
+ * - withPrefix64k : Table entries up to ctx->dictSize before the current blob
+ *                   blob being compressed are valid and refer to the preceding
+ *                   content (of length ctx->dictSize), which is available
+ *                   contiguously preceding in memory the content currently
+ *                   being compressed.
+ * - usingExtDict  : Like withPrefix64k, but the preceding content is somewhere
+ *                   else in memory, starting at ctx->dictionary with length
+ *                   ctx->dictSize.
+ * - usingDictCtx  : Like usingExtDict, but everything concerning the preceding
+ *                   content is in a separate context, pointed to by
+ *                   ctx->dictCtx. ctx->dictionary, ctx->dictSize, and table
+ *                   entries in the current context that refer to positions
+ *                   preceding the beginning of the current compression are
+ *                   ignored. Instead, ctx->dictCtx->dictionary and ctx->dictCtx
+ *                   ->dictSize describe the location and size of the preceding
+ *                   content, and matches are found by looking in the ctx
+ *                   ->dictCtx->hashTable.
+ */
+typedef enum { noDict = 0, withPrefix64k, usingExtDict, usingDictCtx } dict_directive;
+typedef enum { noDictIssue = 0, dictSmall } dictIssue_directive;
+
+
+/*-************************************
+*  Local Utils
+**************************************/
+int LZ4_versionNumber (void) { return LZ4_VERSION_NUMBER; }
+const char* LZ4_versionString(void) { return LZ4_VERSION_STRING; }
+int LZ4_compressBound(int isize)  { return LZ4_COMPRESSBOUND(isize); }
+int LZ4_sizeofState() { return LZ4_STREAMSIZE; }
+
+
+/*-************************************
+*  Internal Definitions used in Tests
+**************************************/
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int srcSize);
+
+int LZ4_decompress_safe_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int maxOutputSize,
+                                     const void* dictStart, size_t dictSize);
+
+#if defined (__cplusplus)
+}
+#endif
+
+/*-******************************
+*  Compression functions
+********************************/
+LZ4_FORCE_INLINE U32 LZ4_hash4(U32 sequence, tableType_t const tableType)
+{
+    if (tableType == byU16)
+        return ((sequence * 2654435761U) >> ((MINMATCH*8)-(LZ4_HASHLOG+1)));
+    else
+        return ((sequence * 2654435761U) >> ((MINMATCH*8)-LZ4_HASHLOG));
+}
+
+LZ4_FORCE_INLINE U32 LZ4_hash5(U64 sequence, tableType_t const tableType)
+{
+    const U32 hashLog = (tableType == byU16) ? LZ4_HASHLOG+1 : LZ4_HASHLOG;
+    if (LZ4_isLittleEndian()) {
+        const U64 prime5bytes = 889523592379ULL;
+        return (U32)(((sequence << 24) * prime5bytes) >> (64 - hashLog));
+    } else {
+        const U64 prime8bytes = 11400714785074694791ULL;
+        return (U32)(((sequence >> 24) * prime8bytes) >> (64 - hashLog));
+    }
+}
+
+LZ4_FORCE_INLINE U32 LZ4_hashPosition(const void* const p, tableType_t const tableType)
+{
+    if ((sizeof(reg_t)==8) && (tableType != byU16)) return LZ4_hash5(LZ4_read_ARCH(p), tableType);
+    return LZ4_hash4(LZ4_read32(p), tableType);
+}
+
+LZ4_FORCE_INLINE void LZ4_clearHash(U32 h, void* tableBase, tableType_t const tableType)
+{
+    switch (tableType)
+    {
+    default: /* fallthrough */
+    case clearedTable: { /* illegal! */ assert(0); return; }
+    case byPtr: { const BYTE** hashTable = (const BYTE**)tableBase; hashTable[h] = NULL; return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = 0; return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; hashTable[h] = 0; return; }
+    }
+}
+
+LZ4_FORCE_INLINE void LZ4_putIndexOnHash(U32 idx, U32 h, void* tableBase, tableType_t const tableType)
+{
+    switch (tableType)
+    {
+    default: /* fallthrough */
+    case clearedTable: /* fallthrough */
+    case byPtr: { /* illegal! */ assert(0); return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = idx; return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; assert(idx < 65536); hashTable[h] = (U16)idx; return; }
+    }
+}
+
+LZ4_FORCE_INLINE void LZ4_putPositionOnHash(const BYTE* p, U32 h,
+                                  void* tableBase, tableType_t const tableType,
+                            const BYTE* srcBase)
+{
+    switch (tableType)
+    {
+    case clearedTable: { /* illegal! */ assert(0); return; }
+    case byPtr: { const BYTE** hashTable = (const BYTE**)tableBase; hashTable[h] = p; return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = (U32)(p-srcBase); return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; hashTable[h] = (U16)(p-srcBase); return; }
+    }
+}
+
+LZ4_FORCE_INLINE void LZ4_putPosition(const BYTE* p, void* tableBase, tableType_t tableType, const BYTE* srcBase)
+{
+    U32 const h = LZ4_hashPosition(p, tableType);
+    LZ4_putPositionOnHash(p, h, tableBase, tableType, srcBase);
+}
+
+/* LZ4_getIndexOnHash() :
+ * Index of match position registered in hash table.
+ * hash position must be calculated by using base+index, or dictBase+index.
+ * Assumption 1 : only valid if tableType == byU32 or byU16.
+ * Assumption 2 : h is presumed valid (within limits of hash table)
+ */
+LZ4_FORCE_INLINE U32 LZ4_getIndexOnHash(U32 h, const void* tableBase, tableType_t tableType)
+{
+    LZ4_STATIC_ASSERT(LZ4_MEMORY_USAGE > 2);
+    if (tableType == byU32) {
+        const U32* const hashTable = (const U32*) tableBase;
+        assert(h < (1U << (LZ4_MEMORY_USAGE-2)));
+        return hashTable[h];
+    }
+    if (tableType == byU16) {
+        const U16* const hashTable = (const U16*) tableBase;
+        assert(h < (1U << (LZ4_MEMORY_USAGE-1)));
+        return hashTable[h];
+    }
+    assert(0); return 0;  /* forbidden case */
+}
+
+static const BYTE* LZ4_getPositionOnHash(U32 h, const void* tableBase, tableType_t tableType, const BYTE* srcBase)
+{
+    if (tableType == byPtr) { const BYTE* const* hashTable = (const BYTE* const*) tableBase; return hashTable[h]; }
+    if (tableType == byU32) { const U32* const hashTable = (const U32*) tableBase; return hashTable[h] + srcBase; }
+    { const U16* const hashTable = (const U16*) tableBase; return hashTable[h] + srcBase; }   /* default, to ensure a return */
+}
+
+LZ4_FORCE_INLINE const BYTE*
+LZ4_getPosition(const BYTE* p,
+                const void* tableBase, tableType_t tableType,
+                const BYTE* srcBase)
+{
+    U32 const h = LZ4_hashPosition(p, tableType);
+    return LZ4_getPositionOnHash(h, tableBase, tableType, srcBase);
+}
+
+LZ4_FORCE_INLINE void
+LZ4_prepareTable(LZ4_stream_t_internal* const cctx,
+           const int inputSize,
+           const tableType_t tableType) {
+    /* If compression failed during the previous step, then the context
+     * is marked as dirty, therefore, it has to be fully reset.
+     */
+    if (cctx->dirty) {
+        DEBUGLOG(5, "LZ4_prepareTable: Full reset for %p", cctx);
+        MEM_INIT(cctx, 0, sizeof(LZ4_stream_t_internal));
+        return;
+    }
+
+    /* If the table hasn't been used, it's guaranteed to be zeroed out, and is
+     * therefore safe to use no matter what mode we're in. Otherwise, we figure
+     * out if it's safe to leave as is or whether it needs to be reset.
+     */
+    if (cctx->tableType != clearedTable) {
+        assert(inputSize >= 0);
+        if (cctx->tableType != tableType
+          || ((tableType == byU16) && cctx->currentOffset + (unsigned)inputSize >= 0xFFFFU)
+          || ((tableType == byU32) && cctx->currentOffset > 1 GB)
+          || tableType == byPtr
+          || inputSize >= 4 KB)
+        {
+            DEBUGLOG(4, "LZ4_prepareTable: Resetting table in %p", cctx);
+            MEM_INIT(cctx->hashTable, 0, LZ4_HASHTABLESIZE);
+            cctx->currentOffset = 0;
+            cctx->tableType = clearedTable;
+        } else {
+            DEBUGLOG(4, "LZ4_prepareTable: Re-use hash table (no reset)");
+        }
+    }
+
+    /* Adding a gap, so all previous entries are > LZ4_DISTANCE_MAX back, is faster
+     * than compressing without a gap. However, compressing with
+     * currentOffset == 0 is faster still, so we preserve that case.
+     */
+    if (cctx->currentOffset != 0 && tableType == byU32) {
+        DEBUGLOG(5, "LZ4_prepareTable: adding 64KB to currentOffset");
+        cctx->currentOffset += 64 KB;
+    }
+
+    /* Finally, clear history */
+    cctx->dictCtx = NULL;
+    cctx->dictionary = NULL;
+    cctx->dictSize = 0;
+}
+
+/** LZ4_compress_generic() :
+    inlined, to ensure branches are decided at compilation time */
+LZ4_FORCE_INLINE int LZ4_compress_generic(
+                 LZ4_stream_t_internal* const cctx,
+                 const char* const source,
+                 char* const dest,
+                 const int inputSize,
+                 int *inputConsumed, /* only written when outputDirective == fillOutput */
+                 const int maxOutputSize,
+                 const limitedOutput_directive outputDirective,
+                 const tableType_t tableType,
+                 const dict_directive dictDirective,
+                 const dictIssue_directive dictIssue,
+                 const int acceleration)
+{
+    int result;
+    const BYTE* ip = (const BYTE*) source;
+
+    U32 const startIndex = cctx->currentOffset;
+    const BYTE* base = (const BYTE*) source - startIndex;
+    const BYTE* lowLimit;
+
+    const LZ4_stream_t_internal* dictCtx = (const LZ4_stream_t_internal*) cctx->dictCtx;
+    const BYTE* const dictionary =
+        dictDirective == usingDictCtx ? dictCtx->dictionary : cctx->dictionary;
+    const U32 dictSize =
+        dictDirective == usingDictCtx ? dictCtx->dictSize : cctx->dictSize;
+    const U32 dictDelta = (dictDirective == usingDictCtx) ? startIndex - dictCtx->currentOffset : 0;   /* make indexes in dictCtx comparable with index in current context */
+
+    int const maybe_extMem = (dictDirective == usingExtDict) || (dictDirective == usingDictCtx);
+    U32 const prefixIdxLimit = startIndex - dictSize;   /* used when dictDirective == dictSmall */
+    const BYTE* const dictEnd = dictionary + dictSize;
+    const BYTE* anchor = (const BYTE*) source;
+    const BYTE* const iend = ip + inputSize;
+    const BYTE* const mflimitPlusOne = iend - MFLIMIT + 1;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+
+    /* the dictCtx currentOffset is indexed on the start of the dictionary,
+     * while a dictionary in the current context precedes the currentOffset */
+    const BYTE* dictBase = (dictDirective == usingDictCtx) ?
+                            dictionary + dictSize - dictCtx->currentOffset :
+                            dictionary + dictSize - startIndex;
+
+    BYTE* op = (BYTE*) dest;
+    BYTE* const olimit = op + maxOutputSize;
+
+    U32 offset = 0;
+    U32 forwardH;
+
+    DEBUGLOG(5, "LZ4_compress_generic: srcSize=%i, tableType=%u", inputSize, tableType);
+    /* If init conditions are not met, we don't have to mark stream
+     * as having dirty context, since no action was taken yet */
+    if (outputDirective == fillOutput && maxOutputSize < 1) { return 0; } /* Impossible to store anything */
+    if ((U32)inputSize > (U32)LZ4_MAX_INPUT_SIZE) { return 0; }           /* Unsupported inputSize, too large (or negative) */
+    if ((tableType == byU16) && (inputSize>=LZ4_64Klimit)) { return 0; }  /* Size too large (not within 64K limit) */
+    if (tableType==byPtr) assert(dictDirective==noDict);      /* only supported use case with byPtr */
+    assert(acceleration >= 1);
+
+    lowLimit = (const BYTE*)source - (dictDirective == withPrefix64k ? dictSize : 0);
+
+    /* Update context state */
+    if (dictDirective == usingDictCtx) {
+        /* Subsequent linked blocks can't use the dictionary. */
+        /* Instead, they use the block we just compressed. */
+        cctx->dictCtx = NULL;
+        cctx->dictSize = (U32)inputSize;
+    } else {
+        cctx->dictSize += (U32)inputSize;
+    }
+    cctx->currentOffset += (U32)inputSize;
+    cctx->tableType = (U16)tableType;
+
+    if (inputSize<LZ4_minLength) goto _last_literals;        /* Input too small, no compression (all literals) */
+
+    /* First Byte */
+    LZ4_putPosition(ip, cctx->hashTable, tableType, base);
+    ip++; forwardH = LZ4_hashPosition(ip, tableType);
+
+    /* Main Loop */
+    for ( ; ; ) {
+        const BYTE* match;
+        BYTE* token;
+        const BYTE* filledIp;
+
+        /* Find a match */
+        if (tableType == byPtr) {
+            const BYTE* forwardIp = ip;
+            int step = 1;
+            int searchMatchNb = acceleration << LZ4_skipTrigger;
+            do {
+                U32 const h = forwardH;
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimitPlusOne)) goto _last_literals;
+                assert(ip < mflimitPlusOne);
+
+                match = LZ4_getPositionOnHash(h, cctx->hashTable, tableType, base);
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putPositionOnHash(ip, h, cctx->hashTable, tableType, base);
+
+            } while ( (match+LZ4_DISTANCE_MAX < ip)
+                   || (LZ4_read32(match) != LZ4_read32(ip)) );
+
+        } else {   /* byU32, byU16 */
+
+            const BYTE* forwardIp = ip;
+            int step = 1;
+            int searchMatchNb = acceleration << LZ4_skipTrigger;
+            do {
+                U32 const h = forwardH;
+                U32 const current = (U32)(forwardIp - base);
+                U32 matchIndex = LZ4_getIndexOnHash(h, cctx->hashTable, tableType);
+                assert(matchIndex <= current);
+                assert(forwardIp - base < (ptrdiff_t)(2 GB - 1));
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimitPlusOne)) goto _last_literals;
+                assert(ip < mflimitPlusOne);
+
+                if (dictDirective == usingDictCtx) {
+                    if (matchIndex < startIndex) {
+                        /* there was no match, try the dictionary */
+                        assert(tableType == byU32);
+                        matchIndex = LZ4_getIndexOnHash(h, dictCtx->hashTable, byU32);
+                        match = dictBase + matchIndex;
+                        matchIndex += dictDelta;   /* make dictCtx index comparable with current context */
+                        lowLimit = dictionary;
+                    } else {
+                        match = base + matchIndex;
+                        lowLimit = (const BYTE*)source;
+                    }
+                } else if (dictDirective==usingExtDict) {
+                    if (matchIndex < startIndex) {
+                        DEBUGLOG(7, "extDict candidate: matchIndex=%5u  <  startIndex=%5u", matchIndex, startIndex);
+                        assert(startIndex - matchIndex >= MINMATCH);
+                        match = dictBase + matchIndex;
+                        lowLimit = dictionary;
+                    } else {
+                        match = base + matchIndex;
+                        lowLimit = (const BYTE*)source;
+                    }
+                } else {   /* single continuous memory segment */
+                    match = base + matchIndex;
+                }
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putIndexOnHash(current, h, cctx->hashTable, tableType);
+
+                DEBUGLOG(7, "candidate at pos=%u  (offset=%u \n", matchIndex, current - matchIndex);
+                if ((dictIssue == dictSmall) && (matchIndex < prefixIdxLimit)) { continue; }    /* match outside of valid area */
+                assert(matchIndex < current);
+                if ( ((tableType != byU16) || (LZ4_DISTANCE_MAX < LZ4_DISTANCE_ABSOLUTE_MAX))
+                  && (matchIndex+LZ4_DISTANCE_MAX < current)) {
+                    continue;
+                } /* too far */
+                assert((current - matchIndex) <= LZ4_DISTANCE_MAX);  /* match now expected within distance */
+
+                if (LZ4_read32(match) == LZ4_read32(ip)) {
+                    if (maybe_extMem) offset = current - matchIndex;
+                    break;   /* match found */
+                }
+
+            } while(1);
+        }
+
+        /* Catch up */
+        filledIp = ip;
+        while (((ip>anchor) & (match > lowLimit)) && (unlikely(ip[-1]==match[-1]))) { ip--; match--; }
+
+        /* Encode Literals */
+        {   unsigned const litLength = (unsigned)(ip - anchor);
+            token = op++;
+            if ((outputDirective == limitedOutput) &&  /* Check output buffer overflow */
+                (unlikely(op + litLength + (2 + 1 + LASTLITERALS) + (litLength/255) > olimit)) ) {
+                return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+            }
+            if ((outputDirective == fillOutput) &&
+                (unlikely(op + (litLength+240)/255 /* litlen */ + litLength /* literals */ + 2 /* offset */ + 1 /* token */ + MFLIMIT - MINMATCH /* min last literals so last match is <= end - MFLIMIT */ > olimit))) {
+                op--;
+                goto _last_literals;
+            }
+            if (litLength >= RUN_MASK) {
+                int len = (int)(litLength - RUN_MASK);
+                *token = (RUN_MASK<<ML_BITS);
+                for(; len >= 255 ; len-=255) *op++ = 255;
+                *op++ = (BYTE)len;
+            }
+            else *token = (BYTE)(litLength<<ML_BITS);
+
+            /* Copy Literals */
+            LZ4_wildCopy8(op, anchor, op+litLength);
+            op+=litLength;
+            DEBUGLOG(6, "seq.start:%i, literals=%u, match.start:%i",
+                        (int)(anchor-(const BYTE*)source), litLength, (int)(ip-(const BYTE*)source));
+        }
+
+_next_match:
+        /* at this stage, the following variables must be correctly set :
+         * - ip : at start of LZ operation
+         * - match : at start of previous pattern occurence; can be within current prefix, or within extDict
+         * - offset : if maybe_ext_memSegment==1 (constant)
+         * - lowLimit : must be == dictionary to mean "match is within extDict"; must be == source otherwise
+         * - token and *token : position to write 4-bits for match length; higher 4-bits for literal length supposed already written
+         */
+
+        if ((outputDirective == fillOutput) &&
+            (op + 2 /* offset */ + 1 /* token */ + MFLIMIT - MINMATCH /* min last literals so last match is <= end - MFLIMIT */ > olimit)) {
+            /* the match was too close to the end, rewind and go to last literals */
+            op = token;
+            goto _last_literals;
+        }
+
+        /* Encode Offset */
+        if (maybe_extMem) {   /* static test */
+            DEBUGLOG(6, "             with offset=%u  (ext if > %i)", offset, (int)(ip - (const BYTE*)source));
+            assert(offset <= LZ4_DISTANCE_MAX && offset > 0);
+            LZ4_writeLE16(op, (U16)offset); op+=2;
+        } else  {
+            DEBUGLOG(6, "             with offset=%u  (same segment)", (U32)(ip - match));
+            assert(ip-match <= LZ4_DISTANCE_MAX);
+            LZ4_writeLE16(op, (U16)(ip - match)); op+=2;
+        }
+
+        /* Encode MatchLength */
+        {   unsigned matchCode;
+
+            if ( (dictDirective==usingExtDict || dictDirective==usingDictCtx)
+              && (lowLimit==dictionary) /* match within extDict */ ) {
+                const BYTE* limit = ip + (dictEnd-match);
+                assert(dictEnd > match);
+                if (limit > matchlimit) limit = matchlimit;
+                matchCode = LZ4_count(ip+MINMATCH, match+MINMATCH, limit);
+                ip += (size_t)matchCode + MINMATCH;
+                if (ip==limit) {
+                    unsigned const more = LZ4_count(limit, (const BYTE*)source, matchlimit);
+                    matchCode += more;
+                    ip += more;
+                }
+                DEBUGLOG(6, "             with matchLength=%u starting in extDict", matchCode+MINMATCH);
+            } else {
+                matchCode = LZ4_count(ip+MINMATCH, match+MINMATCH, matchlimit);
+                ip += (size_t)matchCode + MINMATCH;
+                DEBUGLOG(6, "             with matchLength=%u", matchCode+MINMATCH);
+            }
+
+            if ((outputDirective) &&    /* Check output buffer overflow */
+                (unlikely(op + (1 + LASTLITERALS) + (matchCode+240)/255 > olimit)) ) {
+                if (outputDirective == fillOutput) {
+                    /* Match description too long : reduce it */
+                    U32 newMatchCode = 15 /* in token */ - 1 /* to avoid needing a zero byte */ + ((U32)(olimit - op) - 1 - LASTLITERALS) * 255;
+                    ip -= matchCode - newMatchCode;
+                    assert(newMatchCode < matchCode);
+                    matchCode = newMatchCode;
+                    if (unlikely(ip <= filledIp)) {
+                        /* We have already filled up to filledIp so if ip ends up less than filledIp
+                         * we have positions in the hash table beyond the current position. This is
+                         * a problem if we reuse the hash table. So we have to remove these positions
+                         * from the hash table.
+                         */
+                        const BYTE* ptr;
+                        DEBUGLOG(5, "Clearing %u positions", (U32)(filledIp - ip));
+                        for (ptr = ip; ptr <= filledIp; ++ptr) {
+                            U32 const h = LZ4_hashPosition(ptr, tableType);
+                            LZ4_clearHash(h, cctx->hashTable, tableType);
+                        }
+                    }
+                } else {
+                    assert(outputDirective == limitedOutput);
+                    return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+                }
+            }
+            if (matchCode >= ML_MASK) {
+                *token += ML_MASK;
+                matchCode -= ML_MASK;
+                LZ4_write32(op, 0xFFFFFFFF);
+                while (matchCode >= 4*255) {
+                    op+=4;
+                    LZ4_write32(op, 0xFFFFFFFF);
+                    matchCode -= 4*255;
+                }
+                op += matchCode / 255;
+                *op++ = (BYTE)(matchCode % 255);
+            } else
+                *token += (BYTE)(matchCode);
+        }
+        /* Ensure we have enough space for the last literals. */
+        assert(!(outputDirective == fillOutput && op + 1 + LASTLITERALS > olimit));
+
+        anchor = ip;
+
+        /* Test end of chunk */
+        if (ip >= mflimitPlusOne) break;
+
+        /* Fill table */
+        LZ4_putPosition(ip-2, cctx->hashTable, tableType, base);
+
+        /* Test next position */
+        if (tableType == byPtr) {
+
+            match = LZ4_getPosition(ip, cctx->hashTable, tableType, base);
+            LZ4_putPosition(ip, cctx->hashTable, tableType, base);
+            if ( (match+LZ4_DISTANCE_MAX >= ip)
+              && (LZ4_read32(match) == LZ4_read32(ip)) )
+            { token=op++; *token=0; goto _next_match; }
+
+        } else {   /* byU32, byU16 */
+
+            U32 const h = LZ4_hashPosition(ip, tableType);
+            U32 const current = (U32)(ip-base);
+            U32 matchIndex = LZ4_getIndexOnHash(h, cctx->hashTable, tableType);
+            assert(matchIndex < current);
+            if (dictDirective == usingDictCtx) {
+                if (matchIndex < startIndex) {
+                    /* there was no match, try the dictionary */
+                    matchIndex = LZ4_getIndexOnHash(h, dictCtx->hashTable, byU32);
+                    match = dictBase + matchIndex;
+                    lowLimit = dictionary;   /* required for match length counter */
+                    matchIndex += dictDelta;
+                } else {
+                    match = base + matchIndex;
+                    lowLimit = (const BYTE*)source;  /* required for match length counter */
+                }
+            } else if (dictDirective==usingExtDict) {
+                if (matchIndex < startIndex) {
+                    match = dictBase + matchIndex;
+                    lowLimit = dictionary;   /* required for match length counter */
+                } else {
+                    match = base + matchIndex;
+                    lowLimit = (const BYTE*)source;   /* required for match length counter */
+                }
+            } else {   /* single memory segment */
+                match = base + matchIndex;
+            }
+            LZ4_putIndexOnHash(current, h, cctx->hashTable, tableType);
+            assert(matchIndex < current);
+            if ( ((dictIssue==dictSmall) ? (matchIndex >= prefixIdxLimit) : 1)
+              && (((tableType==byU16) && (LZ4_DISTANCE_MAX == LZ4_DISTANCE_ABSOLUTE_MAX)) ? 1 : (matchIndex+LZ4_DISTANCE_MAX >= current))
+              && (LZ4_read32(match) == LZ4_read32(ip)) ) {
+                token=op++;
+                *token=0;
+                if (maybe_extMem) offset = current - matchIndex;
+                DEBUGLOG(6, "seq.start:%i, literals=%u, match.start:%i",
+                            (int)(anchor-(const BYTE*)source), 0, (int)(ip-(const BYTE*)source));
+                goto _next_match;
+            }
+        }
+
+        /* Prepare next loop */
+        forwardH = LZ4_hashPosition(++ip, tableType);
+
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRun = (size_t)(iend - anchor);
+        if ( (outputDirective) &&  /* Check output buffer overflow */
+            (op + lastRun + 1 + ((lastRun+255-RUN_MASK)/255) > olimit)) {
+            if (outputDirective == fillOutput) {
+                /* adapt lastRun to fill 'dst' */
+                assert(olimit >= op);
+                lastRun  = (size_t)(olimit-op) - 1;
+                lastRun -= (lastRun+240)/255;
+            } else {
+                assert(outputDirective == limitedOutput);
+                return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+            }
+        }
+        if (lastRun >= RUN_MASK) {
+            size_t accumulator = lastRun - RUN_MASK;
+            *op++ = RUN_MASK << ML_BITS;
+            for(; accumulator >= 255 ; accumulator-=255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRun<<ML_BITS);
+        }
+        memcpy(op, anchor, lastRun);
+        ip = anchor + lastRun;
+        op += lastRun;
+    }
+
+    if (outputDirective == fillOutput) {
+        *inputConsumed = (int) (((const char*)ip)-source);
+    }
+    DEBUGLOG(5, "LZ4_compress_generic: compressed %i bytes into %i bytes", inputSize, (int)(((char*)op) - dest));
+    result = (int)(((char*)op) - dest);
+    assert(result > 0);
+    return result;
+}
+
+
+int LZ4_compress_fast_extState(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+    LZ4_stream_t_internal* const ctx = & LZ4_initStream(state, sizeof(LZ4_stream_t)) -> internal_donotuse;
+    assert(ctx != NULL);
+    if (acceleration < 1) acceleration = ACCELERATION_DEFAULT;
+    if (maxOutputSize >= LZ4_compressBound(inputSize)) {
+        if (inputSize < LZ4_64Klimit) {
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, 0, notLimited, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)source > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+        }
+    } else {
+        if (inputSize < LZ4_64Klimit) {
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)source > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+        }
+    }
+}
+
+/**
+ * LZ4_compress_fast_extState_fastReset() :
+ * A variant of LZ4_compress_fast_extState().
+ *
+ * Using this variant avoids an expensive initialization step. It is only safe
+ * to call if the state buffer is known to be correctly initialized already
+ * (see comment in lz4.h on LZ4_resetStream_fast() for a definition of
+ * "correctly initialized").
+ */
+int LZ4_compress_fast_extState_fastReset(void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration)
+{
+    LZ4_stream_t_internal* ctx = &((LZ4_stream_t*)state)->internal_donotuse;
+    if (acceleration < 1) acceleration = ACCELERATION_DEFAULT;
+
+    if (dstCapacity >= LZ4_compressBound(srcSize)) {
+        if (srcSize < LZ4_64Klimit) {
+            const tableType_t tableType = byU16;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            if (ctx->currentOffset) {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, dictSmall, acceleration);
+            } else {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+            }
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+        }
+    } else {
+        if (srcSize < LZ4_64Klimit) {
+            const tableType_t tableType = byU16;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            if (ctx->currentOffset) {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, dictSmall, acceleration);
+            } else {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+            }
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+        }
+    }
+}
+
+
+int LZ4_compress_fast(const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+    int result;
+#if (LZ4_HEAPMODE)
+    LZ4_stream_t* ctxPtr = ALLOC(sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+    if (ctxPtr == NULL) return 0;
+#else
+    LZ4_stream_t ctx;
+    LZ4_stream_t* const ctxPtr = &ctx;
+#endif
+    result = LZ4_compress_fast_extState(ctxPtr, source, dest, inputSize, maxOutputSize, acceleration);
+
+#if (LZ4_HEAPMODE)
+    FREEMEM(ctxPtr);
+#endif
+    return result;
+}
+
+
+int LZ4_compress_default(const char* src, char* dst, int srcSize, int maxOutputSize)
+{
+    return LZ4_compress_fast(src, dst, srcSize, maxOutputSize, 1);
+}
+
+
+/* hidden debug function */
+/* strangely enough, gcc generates faster code when this function is uncommented, even if unused */
+int LZ4_compress_fast_force(const char* src, char* dst, int srcSize, int dstCapacity, int acceleration)
+{
+    LZ4_stream_t ctx;
+    LZ4_initStream(&ctx, sizeof(ctx));
+
+    if (srcSize < LZ4_64Klimit) {
+        return LZ4_compress_generic(&ctx.internal_donotuse, src, dst, srcSize, NULL, dstCapacity, limitedOutput, byU16,    noDict, noDictIssue, acceleration);
+    } else {
+        tableType_t const addrMode = (sizeof(void*) > 4) ? byU32 : byPtr;
+        return LZ4_compress_generic(&ctx.internal_donotuse, src, dst, srcSize, NULL, dstCapacity, limitedOutput, addrMode, noDict, noDictIssue, acceleration);
+    }
+}
+
+
+/* Note!: This function leaves the stream in an unclean/broken state!
+ * It is not safe to subsequently use the same state with a _fastReset() or
+ * _continue() call without resetting it. */
+static int LZ4_compress_destSize_extState (LZ4_stream_t* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize)
+{
+    void* const s = LZ4_initStream(state, sizeof (*state));
+    assert(s != NULL); (void)s;
+
+    if (targetDstSize >= LZ4_compressBound(*srcSizePtr)) {  /* compression success is guaranteed */
+        return LZ4_compress_fast_extState(state, src, dst, *srcSizePtr, targetDstSize, 1);
+    } else {
+        if (*srcSizePtr < LZ4_64Klimit) {
+            return LZ4_compress_generic(&state->internal_donotuse, src, dst, *srcSizePtr, srcSizePtr, targetDstSize, fillOutput, byU16, noDict, noDictIssue, 1);
+        } else {
+            tableType_t const addrMode = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(&state->internal_donotuse, src, dst, *srcSizePtr, srcSizePtr, targetDstSize, fillOutput, addrMode, noDict, noDictIssue, 1);
+    }   }
+}
+
+
+int LZ4_compress_destSize(const char* src, char* dst, int* srcSizePtr, int targetDstSize)
+{
+#if (LZ4_HEAPMODE)
+    LZ4_stream_t* ctx = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+    if (ctx == NULL) return 0;
+#else
+    LZ4_stream_t ctxBody;
+    LZ4_stream_t* ctx = &ctxBody;
+#endif
+
+    int result = LZ4_compress_destSize_extState(ctx, src, dst, srcSizePtr, targetDstSize);
+
+#if (LZ4_HEAPMODE)
+    FREEMEM(ctx);
+#endif
+    return result;
+}
+
+
+
+/*-******************************
+*  Streaming functions
+********************************/
+
+LZ4_stream_t* LZ4_createStream(void)
+{
+    LZ4_stream_t* const lz4s = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));
+    LZ4_STATIC_ASSERT(LZ4_STREAMSIZE >= sizeof(LZ4_stream_t_internal));    /* A compilation error here means LZ4_STREAMSIZE is not large enough */
+    DEBUGLOG(4, "LZ4_createStream %p", lz4s);
+    if (lz4s == NULL) return NULL;
+    LZ4_initStream(lz4s, sizeof(*lz4s));
+    return lz4s;
+}
+
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                     it reports an aligment of 8-bytes,
+                     while actually aligning LZ4_stream_t on 4 bytes. */
+static size_t LZ4_stream_t_alignment(void)
+{
+    struct { char c; LZ4_stream_t t; } t_a;
+    return sizeof(t_a) - sizeof(t_a.t);
+}
+#endif
+
+LZ4_stream_t* LZ4_initStream (void* buffer, size_t size)
+{
+    DEBUGLOG(5, "LZ4_initStream");
+    if (buffer == NULL) { return NULL; }
+    if (size < sizeof(LZ4_stream_t)) { return NULL; }
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                     it reports an aligment of 8-bytes,
+                     while actually aligning LZ4_stream_t on 4 bytes. */
+    if (((size_t)buffer) & (LZ4_stream_t_alignment() - 1)) { return NULL; } /* alignment check */
+#endif
+    MEM_INIT(buffer, 0, sizeof(LZ4_stream_t));
+    return (LZ4_stream_t*)buffer;
+}
+
+/* resetStream is now deprecated,
+ * prefer initStream() which is more general */
+void LZ4_resetStream (LZ4_stream_t* LZ4_stream)
+{
+    DEBUGLOG(5, "LZ4_resetStream (ctx:%p)", LZ4_stream);
+    MEM_INIT(LZ4_stream, 0, sizeof(LZ4_stream_t));
+}
+
+void LZ4_resetStream_fast(LZ4_stream_t* ctx) {
+    LZ4_prepareTable(&(ctx->internal_donotuse), 0, byU32);
+}
+
+int LZ4_freeStream (LZ4_stream_t* LZ4_stream)
+{
+    if (!LZ4_stream) return 0;   /* support free on NULL */
+    DEBUGLOG(5, "LZ4_freeStream %p", LZ4_stream);
+    FREEMEM(LZ4_stream);
+    return (0);
+}
+
+
+#define HASH_UNIT sizeof(reg_t)
+int LZ4_loadDict (LZ4_stream_t* LZ4_dict, const char* dictionary, int dictSize)
+{
+    LZ4_stream_t_internal* dict = &LZ4_dict->internal_donotuse;
+    const tableType_t tableType = byU32;
+    const BYTE* p = (const BYTE*)dictionary;
+    const BYTE* const dictEnd = p + dictSize;
+    const BYTE* base;
+
+    DEBUGLOG(4, "LZ4_loadDict (%i bytes from %p into %p)", dictSize, dictionary, LZ4_dict);
+
+    /* It's necessary to reset the context,
+     * and not just continue it with prepareTable()
+     * to avoid any risk of generating overflowing matchIndex
+     * when compressing using this dictionary */
+    LZ4_resetStream(LZ4_dict);
+
+    /* We always increment the offset by 64 KB, since, if the dict is longer,
+     * we truncate it to the last 64k, and if it's shorter, we still want to
+     * advance by a whole window length so we can provide the guarantee that
+     * there are only valid offsets in the window, which allows an optimization
+     * in LZ4_compress_fast_continue() where it uses noDictIssue even when the
+     * dictionary isn't a full 64k. */
+    dict->currentOffset += 64 KB;
+
+    if (dictSize < (int)HASH_UNIT) {
+        return 0;
+    }
+
+    if ((dictEnd - p) > 64 KB) p = dictEnd - 64 KB;
+    base = dictEnd - dict->currentOffset;
+    dict->dictionary = p;
+    dict->dictSize = (U32)(dictEnd - p);
+    dict->tableType = tableType;
+
+    while (p <= dictEnd-HASH_UNIT) {
+        LZ4_putPosition(p, dict->hashTable, tableType, base);
+        p+=3;
+    }
+
+    return (int)dict->dictSize;
+}
+
+void LZ4_attach_dictionary(LZ4_stream_t* workingStream, const LZ4_stream_t* dictionaryStream) {
+    const LZ4_stream_t_internal* dictCtx = dictionaryStream == NULL ? NULL :
+        &(dictionaryStream->internal_donotuse);
+
+    DEBUGLOG(4, "LZ4_attach_dictionary (%p, %p, size %u)",
+             workingStream, dictionaryStream,
+             dictCtx != NULL ? dictCtx->dictSize : 0);
+
+    /* Calling LZ4_resetStream_fast() here makes sure that changes will not be
+     * erased by subsequent calls to LZ4_resetStream_fast() in case stream was
+     * marked as having dirty context, e.g. requiring full reset.
+     */
+    LZ4_resetStream_fast(workingStream);
+
+    if (dictCtx != NULL) {
+        /* If the current offset is zero, we will never look in the
+         * external dictionary context, since there is no value a table
+         * entry can take that indicate a miss. In that case, we need
+         * to bump the offset to something non-zero.
+         */
+        if (workingStream->internal_donotuse.currentOffset == 0) {
+            workingStream->internal_donotuse.currentOffset = 64 KB;
+        }
+
+        /* Don't actually attach an empty dictionary.
+         */
+        if (dictCtx->dictSize == 0) {
+            dictCtx = NULL;
+        }
+    }
+    workingStream->internal_donotuse.dictCtx = dictCtx;
+}
+
+
+static void LZ4_renormDictT(LZ4_stream_t_internal* LZ4_dict, int nextSize)
+{
+    assert(nextSize >= 0);
+    if (LZ4_dict->currentOffset + (unsigned)nextSize > 0x80000000) {   /* potential ptrdiff_t overflow (32-bits mode) */
+        /* rescale hash table */
+        U32 const delta = LZ4_dict->currentOffset - 64 KB;
+        const BYTE* dictEnd = LZ4_dict->dictionary + LZ4_dict->dictSize;
+        int i;
+        DEBUGLOG(4, "LZ4_renormDictT");
+        for (i=0; i<LZ4_HASH_SIZE_U32; i++) {
+            if (LZ4_dict->hashTable[i] < delta) LZ4_dict->hashTable[i]=0;
+            else LZ4_dict->hashTable[i] -= delta;
+        }
+        LZ4_dict->currentOffset = 64 KB;
+        if (LZ4_dict->dictSize > 64 KB) LZ4_dict->dictSize = 64 KB;
+        LZ4_dict->dictionary = dictEnd - LZ4_dict->dictSize;
+    }
+}
+
+
+int LZ4_compress_fast_continue (LZ4_stream_t* LZ4_stream,
+                                const char* source, char* dest,
+                                int inputSize, int maxOutputSize,
+                                int acceleration)
+{
+    const tableType_t tableType = byU32;
+    LZ4_stream_t_internal* streamPtr = &LZ4_stream->internal_donotuse;
+    const BYTE* dictEnd = streamPtr->dictionary + streamPtr->dictSize;
+
+    DEBUGLOG(5, "LZ4_compress_fast_continue (inputSize=%i)", inputSize);
+
+    if (streamPtr->dirty) { return 0; } /* Uninitialized structure detected */
+    LZ4_renormDictT(streamPtr, inputSize);   /* avoid index overflow */
+    if (acceleration < 1) acceleration = ACCELERATION_DEFAULT;
+
+    /* invalidate tiny dictionaries */
+    if ( (streamPtr->dictSize-1 < 4-1)   /* intentional underflow */
+      && (dictEnd != (const BYTE*)source) ) {
+        DEBUGLOG(5, "LZ4_compress_fast_continue: dictSize(%u) at addr:%p is too small", streamPtr->dictSize, streamPtr->dictionary);
+        streamPtr->dictSize = 0;
+        streamPtr->dictionary = (const BYTE*)source;
+        dictEnd = (const BYTE*)source;
+    }
+
+    /* Check overlapping input/dictionary space */
+    {   const BYTE* sourceEnd = (const BYTE*) source + inputSize;
+        if ((sourceEnd > streamPtr->dictionary) && (sourceEnd < dictEnd)) {
+            streamPtr->dictSize = (U32)(dictEnd - sourceEnd);
+            if (streamPtr->dictSize > 64 KB) streamPtr->dictSize = 64 KB;
+            if (streamPtr->dictSize < 4) streamPtr->dictSize = 0;
+            streamPtr->dictionary = dictEnd - streamPtr->dictSize;
+        }
+    }
+
+    /* prefix mode : source data follows dictionary */
+    if (dictEnd == (const BYTE*)source) {
+        if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset))
+            return LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, withPrefix64k, dictSmall, acceleration);
+        else
+            return LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, withPrefix64k, noDictIssue, acceleration);
+    }
+
+    /* external dictionary mode */
+    {   int result;
+        if (streamPtr->dictCtx) {
+            /* We depend here on the fact that dictCtx'es (produced by
+             * LZ4_loadDict) guarantee that their tables contain no references
+             * to offsets between dictCtx->currentOffset - 64 KB and
+             * dictCtx->currentOffset - dictCtx->dictSize. This makes it safe
+             * to use noDictIssue even when the dict isn't a full 64 KB.
+             */
+            if (inputSize > 4 KB) {
+                /* For compressing large blobs, it is faster to pay the setup
+                 * cost to copy the dictionary's tables into the active context,
+                 * so that the compression loop is only looking into one table.
+                 */
+                memcpy(streamPtr, streamPtr->dictCtx, sizeof(LZ4_stream_t));
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, noDictIssue, acceleration);
+            } else {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingDictCtx, noDictIssue, acceleration);
+            }
+        } else {
+            if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset)) {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, dictSmall, acceleration);
+            } else {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, noDictIssue, acceleration);
+            }
+        }
+        streamPtr->dictionary = (const BYTE*)source;
+        streamPtr->dictSize = (U32)inputSize;
+        return result;
+    }
+}
+
+
+/* Hidden debug function, to force-test external dictionary mode */
+int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int srcSize)
+{
+    LZ4_stream_t_internal* streamPtr = &LZ4_dict->internal_donotuse;
+    int result;
+
+    LZ4_renormDictT(streamPtr, srcSize);
+
+    if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset)) {
+        result = LZ4_compress_generic(streamPtr, source, dest, srcSize, NULL, 0, notLimited, byU32, usingExtDict, dictSmall, 1);
+    } else {
+        result = LZ4_compress_generic(streamPtr, source, dest, srcSize, NULL, 0, notLimited, byU32, usingExtDict, noDictIssue, 1);
+    }
+
+    streamPtr->dictionary = (const BYTE*)source;
+    streamPtr->dictSize = (U32)srcSize;
+
+    return result;
+}
+
+
+/*! LZ4_saveDict() :
+ *  If previously compressed data block is not guaranteed to remain available at its memory location,
+ *  save it into a safer place (char* safeBuffer).
+ *  Note : you don't need to call LZ4_loadDict() afterwards,
+ *         dictionary is immediately usable, you can therefore call LZ4_compress_fast_continue().
+ *  Return : saved dictionary size in bytes (necessarily <= dictSize), or 0 if error.
+ */
+int LZ4_saveDict (LZ4_stream_t* LZ4_dict, char* safeBuffer, int dictSize)
+{
+    LZ4_stream_t_internal* const dict = &LZ4_dict->internal_donotuse;
+    const BYTE* const previousDictEnd = dict->dictionary + dict->dictSize;
+
+    if ((U32)dictSize > 64 KB) { dictSize = 64 KB; } /* useless to define a dictionary > 64 KB */
+    if ((U32)dictSize > dict->dictSize) { dictSize = (int)dict->dictSize; }
+
+    memmove(safeBuffer, previousDictEnd - dictSize, dictSize);
+
+    dict->dictionary = (const BYTE*)safeBuffer;
+    dict->dictSize = (U32)dictSize;
+
+    return dictSize;
+}
+
+
+
+/*-*******************************
+ *  Decompression functions
+ ********************************/
+
+typedef enum { endOnOutputSize = 0, endOnInputSize = 1 } endCondition_directive;
+typedef enum { decode_full_block = 0, partial_decode = 1 } earlyEnd_directive;
+
+#undef MIN
+#define MIN(a,b)    ( (a) < (b) ? (a) : (b) )
+
+/* Read the variable-length literal or match length.
+ *
+ * ip - pointer to use as input.
+ * lencheck - end ip.  Return an error if ip advances >= lencheck.
+ * loop_check - check ip >= lencheck in body of loop.  Returns loop_error if so.
+ * initial_check - check ip >= lencheck before start of loop.  Returns initial_error if so.
+ * error (output) - error code.  Should be set to 0 before call.
+ */
+typedef enum { loop_error = -2, initial_error = -1, ok = 0 } variable_length_error;
+LZ4_FORCE_INLINE unsigned
+read_variable_length(const BYTE**ip, const BYTE* lencheck, int loop_check, int initial_check, variable_length_error* error)
+{
+  U32 length = 0;
+  U32 s;
+  if (initial_check && unlikely((*ip) >= lencheck)) {    /* overflow detection */
+    *error = initial_error;
+    return length;
+  }
+  do {
+    s = **ip;
+    (*ip)++;
+    length += s;
+    if (loop_check && unlikely((*ip) >= lencheck)) {    /* overflow detection */
+      *error = loop_error;
+      return length;
+    }
+  } while (s==255);
+
+  return length;
+}
+
+/*! LZ4_decompress_generic() :
+ *  This generic decompression function covers all use cases.
+ *  It shall be instantiated several times, using different sets of directives.
+ *  Note that it is important for performance that this function really get inlined,
+ *  in order to remove useless branches during compilation optimization.
+ */
+LZ4_FORCE_INLINE int
+LZ4_decompress_generic(
+                 const char* const src,
+                 char* const dst,
+                 int srcSize,
+                 int outputSize,         /* If endOnInput==endOnInputSize, this value is `dstCapacity` */
+
+                 endCondition_directive endOnInput,   /* endOnOutputSize, endOnInputSize */
+                 earlyEnd_directive partialDecoding,  /* full, partial */
+                 dict_directive dict,                 /* noDict, withPrefix64k, usingExtDict */
+                 const BYTE* const lowPrefix,  /* always <= dst, == dst when no prefix */
+                 const BYTE* const dictStart,  /* only if dict==usingExtDict */
+                 const size_t dictSize         /* note : = 0 if noDict */
+                 )
+{
+    if (src == NULL) { return -1; }
+
+    {   const BYTE* ip = (const BYTE*) src;
+        const BYTE* const iend = ip + srcSize;
+
+        BYTE* op = (BYTE*) dst;
+        BYTE* const oend = op + outputSize;
+        BYTE* cpy;
+
+        const BYTE* const dictEnd = (dictStart == NULL) ? NULL : dictStart + dictSize;
+
+        const int safeDecode = (endOnInput==endOnInputSize);
+        const int checkOffset = ((safeDecode) && (dictSize < (int)(64 KB)));
+
+
+        /* Set up the "end" pointers for the shortcut. */
+        const BYTE* const shortiend = iend - (endOnInput ? 14 : 8) /*maxLL*/ - 2 /*offset*/;
+        const BYTE* const shortoend = oend - (endOnInput ? 14 : 8) /*maxLL*/ - 18 /*maxML*/;
+
+        const BYTE* match;
+        size_t offset;
+        unsigned token;
+        size_t length;
+
+
+        DEBUGLOG(5, "LZ4_decompress_generic (srcSize:%i, dstSize:%i)", srcSize, outputSize);
+
+        /* Special cases */
+        assert(lowPrefix <= op);
+        if ((endOnInput) && (unlikely(outputSize==0))) {
+            /* Empty output buffer */
+            if (partialDecoding) return 0;
+            return ((srcSize==1) && (*ip==0)) ? 0 : -1;
+        }
+        if ((!endOnInput) && (unlikely(outputSize==0))) { return (*ip==0 ? 1 : -1); }
+        if ((endOnInput) && unlikely(srcSize==0)) { return -1; }
+
+	/* Currently the fast loop shows a regression on qualcomm arm chips. */
+#if LZ4_FAST_DEC_LOOP
+        if ((oend - op) < FASTLOOP_SAFE_DISTANCE) {
+            DEBUGLOG(6, "skip fast decode loop");
+            goto safe_decode;
+        }
+
+        /* Fast loop : decode sequences as long as output < iend-FASTLOOP_SAFE_DISTANCE */
+        while (1) {
+            /* Main fastloop assertion: We can always wildcopy FASTLOOP_SAFE_DISTANCE */
+            assert(oend - op >= FASTLOOP_SAFE_DISTANCE);
+            if (endOnInput) { assert(ip < iend); }
+            token = *ip++;
+            length = token >> ML_BITS;  /* literal length */
+
+            assert(!endOnInput || ip <= iend); /* ip < iend before the increment */
+
+            /* decode literal length */
+            if (length == RUN_MASK) {
+                variable_length_error error = ok;
+                length += read_variable_length(&ip, iend-RUN_MASK, endOnInput, endOnInput, &error);
+                if (error == initial_error) { goto _output_error; }
+                if ((safeDecode) && unlikely((uptrval)(op)+length<(uptrval)(op))) { goto _output_error; } /* overflow detection */
+                if ((safeDecode) && unlikely((uptrval)(ip)+length<(uptrval)(ip))) { goto _output_error; } /* overflow detection */
+
+                /* copy literals */
+                cpy = op+length;
+                LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
+                if (endOnInput) {  /* LZ4_decompress_safe() */
+                    if ((cpy>oend-32) || (ip+length>iend-32)) { goto safe_literal_copy; }
+                    LZ4_wildCopy32(op, ip, cpy);
+                } else {   /* LZ4_decompress_fast() */
+                    if (cpy>oend-8) { goto safe_literal_copy; }
+                    LZ4_wildCopy8(op, ip, cpy); /* LZ4_decompress_fast() cannot copy more than 8 bytes at a time :
+                                                 * it doesn't know input length, and only relies on end-of-block properties */
+                }
+                ip += length; op = cpy;
+            } else {
+                cpy = op+length;
+                if (endOnInput) {  /* LZ4_decompress_safe() */
+                    DEBUGLOG(7, "copy %u bytes in a 16-bytes stripe", (unsigned)length);
+                    /* We don't need to check oend, since we check it once for each loop below */
+                    if (ip > iend-(16 + 1/*max lit + offset + nextToken*/)) { goto safe_literal_copy; }
+                    /* Literals can only be 14, but hope compilers optimize if we copy by a register size */
+                    memcpy(op, ip, 16);
+                } else {  /* LZ4_decompress_fast() */
+                    /* LZ4_decompress_fast() cannot copy more than 8 bytes at a time :
+                     * it doesn't know input length, and relies on end-of-block properties */
+                    memcpy(op, ip, 8);
+                    if (length > 8) { memcpy(op+8, ip+8, 8); }
+                }
+                ip += length; op = cpy;
+            }
+
+            /* get offset */
+            offset = LZ4_readLE16(ip); ip+=2;
+            match = op - offset;
+            assert(match <= op);
+
+            /* get matchlength */
+            length = token & ML_MASK;
+
+            if (length == ML_MASK) {
+              variable_length_error error = ok;
+              if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) { goto _output_error; } /* Error : offset outside buffers */
+              length += read_variable_length(&ip, iend - LASTLITERALS + 1, endOnInput, 0, &error);
+              if (error != ok) { goto _output_error; }
+                if ((safeDecode) && unlikely((uptrval)(op)+length<(uptrval)op)) { goto _output_error; } /* overflow detection */
+                length += MINMATCH;
+                if (op + length >= oend - FASTLOOP_SAFE_DISTANCE) {
+                    goto safe_match_copy;
+                }
+            } else {
+                length += MINMATCH;
+                if (op + length >= oend - FASTLOOP_SAFE_DISTANCE) {
+                    goto safe_match_copy;
+                }
+
+                /* Fastpath check: Avoids a branch in LZ4_wildCopy32 if true */
+                if ((dict == withPrefix64k) || (match >= lowPrefix)) {
+                    if (offset >= 8) {
+                        assert(match >= lowPrefix);
+                        assert(match <= op);
+                        assert(op + 18 <= oend);
+
+                        memcpy(op, match, 8);
+                        memcpy(op+8, match+8, 8);
+                        memcpy(op+16, match+16, 2);
+                        op += length;
+                        continue;
+            }   }   }
+
+            if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) { goto _output_error; } /* Error : offset outside buffers */
+            /* match starting within external dictionary */
+            if ((dict==usingExtDict) && (match < lowPrefix)) {
+                if (unlikely(op+length > oend-LASTLITERALS)) {
+                    if (partialDecoding) {
+                        length = MIN(length, (size_t)(oend-op));  /* reach end of buffer */
+                    } else {
+                        goto _output_error;  /* end-of-block condition violated */
+                }   }
+
+                if (length <= (size_t)(lowPrefix-match)) {
+                    /* match fits entirely within external dictionary : just copy */
+                    memmove(op, dictEnd - (lowPrefix-match), length);
+                    op += length;
+                } else {
+                    /* match stretches into both external dictionary and current block */
+                    size_t const copySize = (size_t)(lowPrefix - match);
+                    size_t const restSize = length - copySize;
+                    memcpy(op, dictEnd - copySize, copySize);
+                    op += copySize;
+                    if (restSize > (size_t)(op - lowPrefix)) {  /* overlap copy */
+                        BYTE* const endOfMatch = op + restSize;
+                        const BYTE* copyFrom = lowPrefix;
+                        while (op < endOfMatch) { *op++ = *copyFrom++; }
+                    } else {
+                        memcpy(op, lowPrefix, restSize);
+                        op += restSize;
+                }   }
+                continue;
+            }
+
+            /* copy match within block */
+            cpy = op + length;
+
+            assert((op <= oend) && (oend-op >= 32));
+            if (unlikely(offset<16)) {
+                LZ4_memcpy_using_offset(op, match, cpy, offset);
+            } else {
+                LZ4_wildCopy32(op, match, cpy);
+            }
+
+            op = cpy;   /* wildcopy correction */
+        }
+    safe_decode:
+#endif
+
+        /* Main Loop : decode remaining sequences where output < FASTLOOP_SAFE_DISTANCE */
+        while (1) {
+            token = *ip++;
+            length = token >> ML_BITS;  /* literal length */
+
+            assert(!endOnInput || ip <= iend); /* ip < iend before the increment */
+
+            /* A two-stage shortcut for the most common case:
+             * 1) If the literal length is 0..14, and there is enough space,
+             * enter the shortcut and copy 16 bytes on behalf of the literals
+             * (in the fast mode, only 8 bytes can be safely copied this way).
+             * 2) Further if the match length is 4..18, copy 18 bytes in a similar
+             * manner; but we ensure that there's enough space in the output for
+             * those 18 bytes earlier, upon entering the shortcut (in other words,
+             * there is a combined check for both stages).
+             */
+            if ( (endOnInput ? length != RUN_MASK : length <= 8)
+                /* strictly "less than" on input, to re-enter the loop with at least one byte */
+              && likely((endOnInput ? ip < shortiend : 1) & (op <= shortoend)) ) {
+                /* Copy the literals */
+                memcpy(op, ip, endOnInput ? 16 : 8);
+                op += length; ip += length;
+
+                /* The second stage: prepare for match copying, decode full info.
+                 * If it doesn't work out, the info won't be wasted. */
+                length = token & ML_MASK; /* match length */
+                offset = LZ4_readLE16(ip); ip += 2;
+                match = op - offset;
+                assert(match <= op); /* check overflow */
+
+                /* Do not deal with overlapping matches. */
+                if ( (length != ML_MASK)
+                  && (offset >= 8)
+                  && (dict==withPrefix64k || match >= lowPrefix) ) {
+                    /* Copy the match. */
+                    memcpy(op + 0, match + 0, 8);
+                    memcpy(op + 8, match + 8, 8);
+                    memcpy(op +16, match +16, 2);
+                    op += length + MINMATCH;
+                    /* Both stages worked, load the next token. */
+                    continue;
+                }
+
+                /* The second stage didn't work out, but the info is ready.
+                 * Propel it right to the point of match copying. */
+                goto _copy_match;
+            }
+
+            /* decode literal length */
+            if (length == RUN_MASK) {
+                variable_length_error error = ok;
+                length += read_variable_length(&ip, iend-RUN_MASK, endOnInput, endOnInput, &error);
+                if (error == initial_error) { goto _output_error; }
+                if ((safeDecode) && unlikely((uptrval)(op)+length<(uptrval)(op))) { goto _output_error; } /* overflow detection */
+                if ((safeDecode) && unlikely((uptrval)(ip)+length<(uptrval)(ip))) { goto _output_error; } /* overflow detection */
+            }
+
+            /* copy literals */
+            cpy = op+length;
+#if LZ4_FAST_DEC_LOOP
+        safe_literal_copy:
+#endif
+            LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
+            if ( ((endOnInput) && ((cpy>oend-MFLIMIT) || (ip+length>iend-(2+1+LASTLITERALS))) )
+              || ((!endOnInput) && (cpy>oend-WILDCOPYLENGTH)) )
+            {
+                /* We've either hit the input parsing restriction or the output parsing restriction.
+                 * If we've hit the input parsing condition then this must be the last sequence.
+                 * If we've hit the output parsing condition then we are either using partialDecoding
+                 * or we've hit the output parsing condition.
+                 */
+                if (partialDecoding) {
+                    /* Since we are partial decoding we may be in this block because of the output parsing
+                     * restriction, which is not valid since the output buffer is allowed to be undersized.
+                     */
+                    assert(endOnInput);
+                    /* If we're in this block because of the input parsing condition, then we must be on the
+                     * last sequence (or invalid), so we must check that we exactly consume the input.
+                     */
+                    if ((ip+length>iend-(2+1+LASTLITERALS)) && (ip+length != iend)) { goto _output_error; }
+                    assert(ip+length <= iend);
+                    /* We are finishing in the middle of a literals segment.
+                     * Break after the copy.
+                     */
+                    if (cpy > oend) {
+                        cpy = oend;
+                        assert(op<=oend);
+                        length = (size_t)(oend-op);
+                    }
+                    assert(ip+length <= iend);
+                } else {
+                    /* We must be on the last sequence because of the parsing limitations so check
+                     * that we exactly regenerate the original size (must be exact when !endOnInput).
+                     */
+                    if ((!endOnInput) && (cpy != oend)) { goto _output_error; }
+                     /* We must be on the last sequence (or invalid) because of the parsing limitations
+                      * so check that we exactly consume the input and don't overrun the output buffer.
+                      */
+                    if ((endOnInput) && ((ip+length != iend) || (cpy > oend))) { goto _output_error; }
+                }
+                memmove(op, ip, length);  /* supports overlapping memory regions, which only matters for in-place decompression scenarios */
+                ip += length;
+                op += length;
+                /* Necessarily EOF when !partialDecoding. When partialDecoding
+                 * it is EOF if we've either filled the output buffer or hit
+                 * the input parsing restriction.
+                 */
+                if (!partialDecoding || (cpy == oend) || (ip == iend)) {
+                    break;
+                }
+            } else {
+                LZ4_wildCopy8(op, ip, cpy);   /* may overwrite up to WILDCOPYLENGTH beyond cpy */
+                ip += length; op = cpy;
+            }
+
+            /* get offset */
+            offset = LZ4_readLE16(ip); ip+=2;
+            match = op - offset;
+
+            /* get matchlength */
+            length = token & ML_MASK;
+
+    _copy_match:
+            if (length == ML_MASK) {
+              variable_length_error error = ok;
+              length += read_variable_length(&ip, iend - LASTLITERALS + 1, endOnInput, 0, &error);
+              if (error != ok) goto _output_error;
+                if ((safeDecode) && unlikely((uptrval)(op)+length<(uptrval)op)) goto _output_error;   /* overflow detection */
+            }
+            length += MINMATCH;
+
+#if LZ4_FAST_DEC_LOOP
+        safe_match_copy:
+#endif
+            if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) goto _output_error;   /* Error : offset outside buffers */
+            /* match starting within external dictionary */
+            if ((dict==usingExtDict) && (match < lowPrefix)) {
+                if (unlikely(op+length > oend-LASTLITERALS)) {
+                    if (partialDecoding) length = MIN(length, (size_t)(oend-op));
+                    else goto _output_error;   /* doesn't respect parsing restriction */
+                }
+
+                if (length <= (size_t)(lowPrefix-match)) {
+                    /* match fits entirely within external dictionary : just copy */
+                    memmove(op, dictEnd - (lowPrefix-match), length);
+                    op += length;
+                } else {
+                    /* match stretches into both external dictionary and current block */
+                    size_t const copySize = (size_t)(lowPrefix - match);
+                    size_t const restSize = length - copySize;
+                    memcpy(op, dictEnd - copySize, copySize);
+                    op += copySize;
+                    if (restSize > (size_t)(op - lowPrefix)) {  /* overlap copy */
+                        BYTE* const endOfMatch = op + restSize;
+                        const BYTE* copyFrom = lowPrefix;
+                        while (op < endOfMatch) *op++ = *copyFrom++;
+                    } else {
+                        memcpy(op, lowPrefix, restSize);
+                        op += restSize;
+                }   }
+                continue;
+            }
+            assert(match >= lowPrefix);
+
+            /* copy match within block */
+            cpy = op + length;
+
+            /* partialDecoding : may end anywhere within the block */
+            assert(op<=oend);
+            if (partialDecoding && (cpy > oend-MATCH_SAFEGUARD_DISTANCE)) {
+                size_t const mlen = MIN(length, (size_t)(oend-op));
+                const BYTE* const matchEnd = match + mlen;
+                BYTE* const copyEnd = op + mlen;
+                if (matchEnd > op) {   /* overlap copy */
+                    while (op < copyEnd) { *op++ = *match++; }
+                } else {
+                    memcpy(op, match, mlen);
+                }
+                op = copyEnd;
+                if (op == oend) { break; }
+                continue;
+            }
+
+            if (unlikely(offset<8)) {
+                LZ4_write32(op, 0);   /* silence msan warning when offset==0 */
+                op[0] = match[0];
+                op[1] = match[1];
+                op[2] = match[2];
+                op[3] = match[3];
+                match += inc32table[offset];
+                memcpy(op+4, match, 4);
+                match -= dec64table[offset];
+            } else {
+                memcpy(op, match, 8);
+                match += 8;
+            }
+            op += 8;
+
+            if (unlikely(cpy > oend-MATCH_SAFEGUARD_DISTANCE)) {
+                BYTE* const oCopyLimit = oend - (WILDCOPYLENGTH-1);
+                if (cpy > oend-LASTLITERALS) { goto _output_error; } /* Error : last LASTLITERALS bytes must be literals (uncompressed) */
+                if (op < oCopyLimit) {
+                    LZ4_wildCopy8(op, match, oCopyLimit);
+                    match += oCopyLimit - op;
+                    op = oCopyLimit;
+                }
+                while (op < cpy) { *op++ = *match++; }
+            } else {
+                memcpy(op, match, 8);
+                if (length > 16)  { LZ4_wildCopy8(op+8, match+8, cpy); }
+            }
+            op = cpy;   /* wildcopy correction */
+        }
+
+        /* end of decoding */
+        if (endOnInput) {
+           return (int) (((char*)op)-dst);     /* Nb of output bytes decoded */
+       } else {
+           return (int) (((const char*)ip)-src);   /* Nb of input bytes read */
+       }
+
+        /* Overflow error detected */
+    _output_error:
+        return (int) (-(((const char*)ip)-src))-1;
+    }
+}
+
+
+/*===== Instantiate the API decoding functions. =====*/
+
+LZ4_FORCE_O2_GCC_PPC64LE
+int LZ4_decompress_safe(const char* source, char* dest, int compressedSize, int maxDecompressedSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxDecompressedSize,
+                                  endOnInputSize, decode_full_block, noDict,
+                                  (BYTE*)dest, NULL, 0);
+}
+
+LZ4_FORCE_O2_GCC_PPC64LE
+int LZ4_decompress_safe_partial(const char* src, char* dst, int compressedSize, int targetOutputSize, int dstCapacity)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(src, dst, compressedSize, dstCapacity,
+                                  endOnInputSize, partial_decode,
+                                  noDict, (BYTE*)dst, NULL, 0);
+}
+
+LZ4_FORCE_O2_GCC_PPC64LE
+int LZ4_decompress_fast(const char* source, char* dest, int originalSize)
+{
+    return LZ4_decompress_generic(source, dest, 0, originalSize,
+                                  endOnOutputSize, decode_full_block, withPrefix64k,
+                                  (BYTE*)dest - 64 KB, NULL, 0);
+}
+
+/*===== Instantiate a few more decoding cases, used more than once. =====*/
+
+LZ4_FORCE_O2_GCC_PPC64LE /* Exported, an obsolete API function. */
+int LZ4_decompress_safe_withPrefix64k(const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  endOnInputSize, decode_full_block, withPrefix64k,
+                                  (BYTE*)dest - 64 KB, NULL, 0);
+}
+
+/* Another obsolete API function, paired with the previous one. */
+int LZ4_decompress_fast_withPrefix64k(const char* source, char* dest, int originalSize)
+{
+    /* LZ4_decompress_fast doesn't validate match offsets,
+     * and thus serves well with any prefixed dictionary. */
+    return LZ4_decompress_fast(source, dest, originalSize);
+}
+
+LZ4_FORCE_O2_GCC_PPC64LE
+static int LZ4_decompress_safe_withSmallPrefix(const char* source, char* dest, int compressedSize, int maxOutputSize,
+                                               size_t prefixSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  endOnInputSize, decode_full_block, noDict,
+                                  (BYTE*)dest-prefixSize, NULL, 0);
+}
+
+LZ4_FORCE_O2_GCC_PPC64LE
+int LZ4_decompress_safe_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int maxOutputSize,
+                                     const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  endOnInputSize, decode_full_block, usingExtDict,
+                                  (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+LZ4_FORCE_O2_GCC_PPC64LE
+static int LZ4_decompress_fast_extDict(const char* source, char* dest, int originalSize,
+                                       const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_generic(source, dest, 0, originalSize,
+                                  endOnOutputSize, decode_full_block, usingExtDict,
+                                  (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+/* The "double dictionary" mode, for use with e.g. ring buffers: the first part
+ * of the dictionary is passed as prefix, and the second via dictStart + dictSize.
+ * These routines are used only once, in LZ4_decompress_*_continue().
+ */
+LZ4_FORCE_INLINE
+int LZ4_decompress_safe_doubleDict(const char* source, char* dest, int compressedSize, int maxOutputSize,
+                                   size_t prefixSize, const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  endOnInputSize, decode_full_block, usingExtDict,
+                                  (BYTE*)dest-prefixSize, (const BYTE*)dictStart, dictSize);
+}
+
+LZ4_FORCE_INLINE
+int LZ4_decompress_fast_doubleDict(const char* source, char* dest, int originalSize,
+                                   size_t prefixSize, const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_generic(source, dest, 0, originalSize,
+                                  endOnOutputSize, decode_full_block, usingExtDict,
+                                  (BYTE*)dest-prefixSize, (const BYTE*)dictStart, dictSize);
+}
+
+/*===== streaming decompression functions =====*/
+
+LZ4_streamDecode_t* LZ4_createStreamDecode(void)
+{
+    LZ4_streamDecode_t* lz4s = (LZ4_streamDecode_t*) ALLOC_AND_ZERO(sizeof(LZ4_streamDecode_t));
+    LZ4_STATIC_ASSERT(LZ4_STREAMDECODESIZE >= sizeof(LZ4_streamDecode_t_internal));    /* A compilation error here means LZ4_STREAMDECODESIZE is not large enough */
+    return lz4s;
+}
+
+int LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream)
+{
+    if (LZ4_stream == NULL) { return 0; }  /* support free on NULL */
+    FREEMEM(LZ4_stream);
+    return 0;
+}
+
+/*! LZ4_setStreamDecode() :
+ *  Use this function to instruct where to find the dictionary.
+ *  This function is not necessary if previous data is still available where it was decoded.
+ *  Loading a size of 0 is allowed (same effect as no dictionary).
+ * @return : 1 if OK, 0 if error
+ */
+int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = &LZ4_streamDecode->internal_donotuse;
+    lz4sd->prefixSize = (size_t) dictSize;
+    lz4sd->prefixEnd = (const BYTE*) dictionary + dictSize;
+    lz4sd->externalDict = NULL;
+    lz4sd->extDictSize  = 0;
+    return 1;
+}
+
+/*! LZ4_decoderRingBufferSize() :
+ *  when setting a ring buffer for streaming decompression (optional scenario),
+ *  provides the minimum size of this ring buffer
+ *  to be compatible with any source respecting maxBlockSize condition.
+ *  Note : in a ring buffer scenario,
+ *  blocks are presumed decompressed next to each other.
+ *  When not enough space remains for next block (remainingSize < maxBlockSize),
+ *  decoding resumes from beginning of ring buffer.
+ * @return : minimum ring buffer size,
+ *           or 0 if there is an error (invalid maxBlockSize).
+ */
+int LZ4_decoderRingBufferSize(int maxBlockSize)
+{
+    if (maxBlockSize < 0) return 0;
+    if (maxBlockSize > LZ4_MAX_INPUT_SIZE) return 0;
+    if (maxBlockSize < 16) maxBlockSize = 16;
+    return LZ4_DECODER_RING_BUFFER_SIZE(maxBlockSize);
+}
+
+/*
+*_continue() :
+    These decoding functions allow decompression of multiple blocks in "streaming" mode.
+    Previously decoded blocks must still be available at the memory position where they were decoded.
+    If it's not possible, save the relevant part of decoded data into a safe buffer,
+    and indicate where it stands using LZ4_setStreamDecode()
+*/
+LZ4_FORCE_O2_GCC_PPC64LE
+int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = &LZ4_streamDecode->internal_donotuse;
+    int result;
+
+    if (lz4sd->prefixSize == 0) {
+        /* The first call, no dictionary yet. */
+        assert(lz4sd->extDictSize == 0);
+        result = LZ4_decompress_safe(source, dest, compressedSize, maxOutputSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)result;
+        lz4sd->prefixEnd = (BYTE*)dest + result;
+    } else if (lz4sd->prefixEnd == (BYTE*)dest) {
+        /* They're rolling the current segment. */
+        if (lz4sd->prefixSize >= 64 KB - 1)
+            result = LZ4_decompress_safe_withPrefix64k(source, dest, compressedSize, maxOutputSize);
+        else if (lz4sd->extDictSize == 0)
+            result = LZ4_decompress_safe_withSmallPrefix(source, dest, compressedSize, maxOutputSize,
+                                                         lz4sd->prefixSize);
+        else
+            result = LZ4_decompress_safe_doubleDict(source, dest, compressedSize, maxOutputSize,
+                                                    lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += (size_t)result;
+        lz4sd->prefixEnd  += result;
+    } else {
+        /* The buffer wraps around, or they're switching to another buffer. */
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
+        result = LZ4_decompress_safe_forceExtDict(source, dest, compressedSize, maxOutputSize,
+                                                  lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)result;
+        lz4sd->prefixEnd  = (BYTE*)dest + result;
+    }
+
+    return result;
+}
+
+LZ4_FORCE_O2_GCC_PPC64LE
+int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int originalSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = &LZ4_streamDecode->internal_donotuse;
+    int result;
+    assert(originalSize >= 0);
+
+    if (lz4sd->prefixSize == 0) {
+        assert(lz4sd->extDictSize == 0);
+        result = LZ4_decompress_fast(source, dest, originalSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)originalSize;
+        lz4sd->prefixEnd = (BYTE*)dest + originalSize;
+    } else if (lz4sd->prefixEnd == (BYTE*)dest) {
+        if (lz4sd->prefixSize >= 64 KB - 1 || lz4sd->extDictSize == 0)
+            result = LZ4_decompress_fast(source, dest, originalSize);
+        else
+            result = LZ4_decompress_fast_doubleDict(source, dest, originalSize,
+                                                    lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += (size_t)originalSize;
+        lz4sd->prefixEnd  += originalSize;
+    } else {
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
+        result = LZ4_decompress_fast_extDict(source, dest, originalSize,
+                                             lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)originalSize;
+        lz4sd->prefixEnd  = (BYTE*)dest + originalSize;
+    }
+
+    return result;
+}
+
+
+/*
+Advanced decoding functions :
+*_usingDict() :
+    These decoding functions work the same as "_continue" ones,
+    the dictionary must be explicitly provided within parameters
+*/
+
+int LZ4_decompress_safe_usingDict(const char* source, char* dest, int compressedSize, int maxOutputSize, const char* dictStart, int dictSize)
+{
+    if (dictSize==0)
+        return LZ4_decompress_safe(source, dest, compressedSize, maxOutputSize);
+    if (dictStart+dictSize == dest) {
+        if (dictSize >= 64 KB - 1) {
+            return LZ4_decompress_safe_withPrefix64k(source, dest, compressedSize, maxOutputSize);
+        }
+        assert(dictSize >= 0);
+        return LZ4_decompress_safe_withSmallPrefix(source, dest, compressedSize, maxOutputSize, (size_t)dictSize);
+    }
+    assert(dictSize >= 0);
+    return LZ4_decompress_safe_forceExtDict(source, dest, compressedSize, maxOutputSize, dictStart, (size_t)dictSize);
+}
+
+int LZ4_decompress_fast_usingDict(const char* source, char* dest, int originalSize, const char* dictStart, int dictSize)
+{
+    if (dictSize==0 || dictStart+dictSize == dest)
+        return LZ4_decompress_fast(source, dest, originalSize);
+    assert(dictSize >= 0);
+    return LZ4_decompress_fast_extDict(source, dest, originalSize, dictStart, (size_t)dictSize);
+}
+
+
+/*=*************************************************
+*  Obsolete Functions
+***************************************************/
+/* obsolete compression functions */
+int LZ4_compress_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize)
+{
+    return LZ4_compress_default(source, dest, inputSize, maxOutputSize);
+}
+int LZ4_compress(const char* src, char* dest, int srcSize)
+{
+    return LZ4_compress_default(src, dest, srcSize, LZ4_compressBound(srcSize));
+}
+int LZ4_compress_limitedOutput_withState (void* state, const char* src, char* dst, int srcSize, int dstSize)
+{
+    return LZ4_compress_fast_extState(state, src, dst, srcSize, dstSize, 1);
+}
+int LZ4_compress_withState (void* state, const char* src, char* dst, int srcSize)
+{
+    return LZ4_compress_fast_extState(state, src, dst, srcSize, LZ4_compressBound(srcSize), 1);
+}
+int LZ4_compress_limitedOutput_continue (LZ4_stream_t* LZ4_stream, const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    return LZ4_compress_fast_continue(LZ4_stream, src, dst, srcSize, dstCapacity, 1);
+}
+int LZ4_compress_continue (LZ4_stream_t* LZ4_stream, const char* source, char* dest, int inputSize)
+{
+    return LZ4_compress_fast_continue(LZ4_stream, source, dest, inputSize, LZ4_compressBound(inputSize), 1);
+}
+
+/*
+These decompression functions are deprecated and should no longer be used.
+They are only provided here for compatibility with older user programs.
+- LZ4_uncompress is totally equivalent to LZ4_decompress_fast
+- LZ4_uncompress_unknownOutputSize is totally equivalent to LZ4_decompress_safe
+*/
+int LZ4_uncompress (const char* source, char* dest, int outputSize)
+{
+    return LZ4_decompress_fast(source, dest, outputSize);
+}
+int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize)
+{
+    return LZ4_decompress_safe(source, dest, isize, maxOutputSize);
+}
+
+/* Obsolete Streaming functions */
+
+int LZ4_sizeofStreamState() { return LZ4_STREAMSIZE; }
+
+int LZ4_resetStreamState(void* state, char* inputBuffer)
+{
+    (void)inputBuffer;
+    LZ4_resetStream((LZ4_stream_t*)state);
+    return 0;
+}
+
+void* LZ4_create (char* inputBuffer)
+{
+    (void)inputBuffer;
+    return LZ4_createStream();
+}
+
+char* LZ4_slideInputBuffer (void* state)
+{
+    /* avoid const char * -> char * conversion warning */
+    return (char *)(uptrval)((LZ4_stream_t*)state)->internal_donotuse.dictionary;
+}
+
+#endif   /* LZ4_COMMONDEFS_ONLY */
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4/lz4.h b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4.h
new file mode 100644
index 0000000..32108e2
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4.h
@@ -0,0 +1,764 @@
+/*
+ *  LZ4 - Fast LZ compression algorithm
+ *  Header File
+ *  Copyright (C) 2011-present, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+    - LZ4 homepage : http://www.lz4.org
+    - LZ4 source repository : https://github.com/lz4/lz4
+*/
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef LZ4_H_2983827168210
+#define LZ4_H_2983827168210
+
+/* --- Dependency --- */
+#include <stddef.h>   /* size_t */
+
+
+/**
+  Introduction
+
+  LZ4 is lossless compression algorithm, providing compression speed >500 MB/s per core,
+  scalable with multi-cores CPU. It features an extremely fast decoder, with speed in
+  multiple GB/s per core, typically reaching RAM speed limits on multi-core systems.
+
+  The LZ4 compression library provides in-memory compression and decompression functions.
+  It gives full buffer control to user.
+  Compression can be done in:
+    - a single step (described as Simple Functions)
+    - a single step, reusing a context (described in Advanced Functions)
+    - unbounded multiple steps (described as Streaming compression)
+
+  lz4.h generates and decodes LZ4-compressed blocks (doc/lz4_Block_format.md).
+  Decompressing such a compressed block requires additional metadata.
+  Exact metadata depends on exact decompression function.
+  For the typical case of LZ4_decompress_safe(),
+  metadata includes block's compressed size, and maximum bound of decompressed size.
+  Each application is free to encode and pass such metadata in whichever way it wants.
+
+  lz4.h only handle blocks, it can not generate Frames.
+
+  Blocks are different from Frames (doc/lz4_Frame_format.md).
+  Frames bundle both blocks and metadata in a specified manner.
+  Embedding metadata is required for compressed data to be self-contained and portable.
+  Frame format is delivered through a companion API, declared in lz4frame.h.
+  The `lz4` CLI can only manage frames.
+*/
+
+/*^***************************************************************
+*  Export parameters
+*****************************************************************/
+/*
+*  LZ4_DLL_EXPORT :
+*  Enable exporting of functions when building a Windows DLL
+*  LZ4LIB_VISIBILITY :
+*  Control library symbols visibility.
+*/
+#ifndef LZ4LIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define LZ4LIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define LZ4LIB_VISIBILITY
+#  endif
+#endif
+#if defined(LZ4_DLL_EXPORT) && (LZ4_DLL_EXPORT==1)
+#  define LZ4LIB_API __declspec(dllexport) LZ4LIB_VISIBILITY
+#elif defined(LZ4_DLL_IMPORT) && (LZ4_DLL_IMPORT==1)
+#  define LZ4LIB_API __declspec(dllimport) LZ4LIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define LZ4LIB_API LZ4LIB_VISIBILITY
+#endif
+
+/*------   Version   ------*/
+#define LZ4_VERSION_MAJOR    1    /* for breaking interface changes  */
+#define LZ4_VERSION_MINOR    9    /* for new (non-breaking) interface capabilities */
+#define LZ4_VERSION_RELEASE  2    /* for tweaks, bug-fixes, or development */
+
+#define LZ4_VERSION_NUMBER (LZ4_VERSION_MAJOR *100*100 + LZ4_VERSION_MINOR *100 + LZ4_VERSION_RELEASE)
+
+#define LZ4_LIB_VERSION LZ4_VERSION_MAJOR.LZ4_VERSION_MINOR.LZ4_VERSION_RELEASE
+#define LZ4_QUOTE(str) #str
+#define LZ4_EXPAND_AND_QUOTE(str) LZ4_QUOTE(str)
+#define LZ4_VERSION_STRING LZ4_EXPAND_AND_QUOTE(LZ4_LIB_VERSION)
+
+LZ4LIB_API int LZ4_versionNumber (void);  /**< library version number; useful to check dll version */
+LZ4LIB_API const char* LZ4_versionString (void);   /**< library version string; useful to check dll version */
+
+
+/*-************************************
+*  Tuning parameter
+**************************************/
+/*!
+ * LZ4_MEMORY_USAGE :
+ * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+ * Increasing memory usage improves compression ratio.
+ * Reduced memory usage may improve speed, thanks to better cache locality.
+ * Default value is 14, for 16KB, which nicely fits into Intel x86 L1 cache
+ */
+#ifndef LZ4_MEMORY_USAGE
+# define LZ4_MEMORY_USAGE 14
+#endif
+
+
+/*-************************************
+*  Simple Functions
+**************************************/
+/*! LZ4_compress_default() :
+ *  Compresses 'srcSize' bytes from buffer 'src'
+ *  into already allocated 'dst' buffer of size 'dstCapacity'.
+ *  Compression is guaranteed to succeed if 'dstCapacity' >= LZ4_compressBound(srcSize).
+ *  It also runs faster, so it's a recommended setting.
+ *  If the function cannot compress 'src' into a more limited 'dst' budget,
+ *  compression stops *immediately*, and the function result is zero.
+ *  In which case, 'dst' content is undefined (invalid).
+ *      srcSize : max supported value is LZ4_MAX_INPUT_SIZE.
+ *      dstCapacity : size of buffer 'dst' (which must be already allocated)
+ *     @return  : the number of bytes written into buffer 'dst' (necessarily <= dstCapacity)
+ *                or 0 if compression fails
+ * Note : This function is protected against buffer overflow scenarios (never writes outside 'dst' buffer, nor read outside 'source' buffer).
+ */
+LZ4LIB_API int LZ4_compress_default(const char* src, char* dst, int srcSize, int dstCapacity);
+
+/*! LZ4_decompress_safe() :
+ *  compressedSize : is the exact complete size of the compressed block.
+ *  dstCapacity : is the size of destination buffer (which must be already allocated), presumed an upper bound of decompressed size.
+ * @return : the number of bytes decompressed into destination buffer (necessarily <= dstCapacity)
+ *           If destination buffer is not large enough, decoding will stop and output an error code (negative value).
+ *           If the source stream is detected malformed, the function will stop decoding and return a negative result.
+ * Note 1 : This function is protected against malicious data packets :
+ *          it will never writes outside 'dst' buffer, nor read outside 'source' buffer,
+ *          even if the compressed block is maliciously modified to order the decoder to do these actions.
+ *          In such case, the decoder stops immediately, and considers the compressed block malformed.
+ * Note 2 : compressedSize and dstCapacity must be provided to the function, the compressed block does not contain them.
+ *          The implementation is free to send / store / derive this information in whichever way is most beneficial.
+ *          If there is a need for a different format which bundles together both compressed data and its metadata, consider looking at lz4frame.h instead.
+ */
+LZ4LIB_API int LZ4_decompress_safe (const char* src, char* dst, int compressedSize, int dstCapacity);
+
+
+/*-************************************
+*  Advanced Functions
+**************************************/
+#define LZ4_MAX_INPUT_SIZE        0x7E000000   /* 2 113 929 216 bytes */
+#define LZ4_COMPRESSBOUND(isize)  ((unsigned)(isize) > (unsigned)LZ4_MAX_INPUT_SIZE ? 0 : (isize) + ((isize)/255) + 16)
+
+/*! LZ4_compressBound() :
+    Provides the maximum size that LZ4 compression may output in a "worst case" scenario (input data not compressible)
+    This function is primarily useful for memory allocation purposes (destination buffer size).
+    Macro LZ4_COMPRESSBOUND() is also provided for compilation-time evaluation (stack memory allocation for example).
+    Note that LZ4_compress_default() compresses faster when dstCapacity is >= LZ4_compressBound(srcSize)
+        inputSize  : max supported value is LZ4_MAX_INPUT_SIZE
+        return : maximum output size in a "worst case" scenario
+              or 0, if input size is incorrect (too large or negative)
+*/
+LZ4LIB_API int LZ4_compressBound(int inputSize);
+
+/*! LZ4_compress_fast() :
+    Same as LZ4_compress_default(), but allows selection of "acceleration" factor.
+    The larger the acceleration value, the faster the algorithm, but also the lesser the compression.
+    It's a trade-off. It can be fine tuned, with each successive value providing roughly +~3% to speed.
+    An acceleration value of "1" is the same as regular LZ4_compress_default()
+    Values <= 0 will be replaced by ACCELERATION_DEFAULT (currently == 1, see lz4.c).
+*/
+LZ4LIB_API int LZ4_compress_fast (const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+
+/*! LZ4_compress_fast_extState() :
+ *  Same as LZ4_compress_fast(), using an externally allocated memory space for its state.
+ *  Use LZ4_sizeofState() to know how much memory must be allocated,
+ *  and allocate it on 8-bytes boundaries (using `malloc()` typically).
+ *  Then, provide this buffer as `void* state` to compression function.
+ */
+LZ4LIB_API int LZ4_sizeofState(void);
+LZ4LIB_API int LZ4_compress_fast_extState (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+
+/*! LZ4_compress_destSize() :
+ *  Reverse the logic : compresses as much data as possible from 'src' buffer
+ *  into already allocated buffer 'dst', of size >= 'targetDestSize'.
+ *  This function either compresses the entire 'src' content into 'dst' if it's large enough,
+ *  or fill 'dst' buffer completely with as much data as possible from 'src'.
+ *  note: acceleration parameter is fixed to "default".
+ *
+ * *srcSizePtr : will be modified to indicate how many bytes where read from 'src' to fill 'dst'.
+ *               New value is necessarily <= input value.
+ * @return : Nb bytes written into 'dst' (necessarily <= targetDestSize)
+ *           or 0 if compression fails.
+*/
+LZ4LIB_API int LZ4_compress_destSize (const char* src, char* dst, int* srcSizePtr, int targetDstSize);
+
+
+/*! LZ4_decompress_safe_partial() :
+ *  Decompress an LZ4 compressed block, of size 'srcSize' at position 'src',
+ *  into destination buffer 'dst' of size 'dstCapacity'.
+ *  Up to 'targetOutputSize' bytes will be decoded.
+ *  The function stops decoding on reaching this objective,
+ *  which can boost performance when only the beginning of a block is required.
+ *
+ * @return : the number of bytes decoded in `dst` (necessarily <= dstCapacity)
+ *           If source stream is detected malformed, function returns a negative result.
+ *
+ *  Note : @return can be < targetOutputSize, if compressed block contains less data.
+ *
+ *  Note 2 : this function features 2 parameters, targetOutputSize and dstCapacity,
+ *           and expects targetOutputSize <= dstCapacity.
+ *           It effectively stops decoding on reaching targetOutputSize,
+ *           so dstCapacity is kind of redundant.
+ *           This is because in a previous version of this function,
+ *           decoding operation would not "break" a sequence in the middle.
+ *           As a consequence, there was no guarantee that decoding would stop at exactly targetOutputSize,
+ *           it could write more bytes, though only up to dstCapacity.
+ *           Some "margin" used to be required for this operation to work properly.
+ *           This is no longer necessary.
+ *           The function nonetheless keeps its signature, in an effort to not break API.
+ */
+LZ4LIB_API int LZ4_decompress_safe_partial (const char* src, char* dst, int srcSize, int targetOutputSize, int dstCapacity);
+
+
+/*-*********************************************
+*  Streaming Compression Functions
+***********************************************/
+typedef union LZ4_stream_u LZ4_stream_t;  /* incomplete type (defined later) */
+
+LZ4LIB_API LZ4_stream_t* LZ4_createStream(void);
+LZ4LIB_API int           LZ4_freeStream (LZ4_stream_t* streamPtr);
+
+/*! LZ4_resetStream_fast() : v1.9.0+
+ *  Use this to prepare an LZ4_stream_t for a new chain of dependent blocks
+ *  (e.g., LZ4_compress_fast_continue()).
+ *
+ *  An LZ4_stream_t must be initialized once before usage.
+ *  This is automatically done when created by LZ4_createStream().
+ *  However, should the LZ4_stream_t be simply declared on stack (for example),
+ *  it's necessary to initialize it first, using LZ4_initStream().
+ *
+ *  After init, start any new stream with LZ4_resetStream_fast().
+ *  A same LZ4_stream_t can be re-used multiple times consecutively
+ *  and compress multiple streams,
+ *  provided that it starts each new stream with LZ4_resetStream_fast().
+ *
+ *  LZ4_resetStream_fast() is much faster than LZ4_initStream(),
+ *  but is not compatible with memory regions containing garbage data.
+ *
+ *  Note: it's only useful to call LZ4_resetStream_fast()
+ *        in the context of streaming compression.
+ *        The *extState* functions perform their own resets.
+ *        Invoking LZ4_resetStream_fast() before is redundant, and even counterproductive.
+ */
+LZ4LIB_API void LZ4_resetStream_fast (LZ4_stream_t* streamPtr);
+
+/*! LZ4_loadDict() :
+ *  Use this function to reference a static dictionary into LZ4_stream_t.
+ *  The dictionary must remain available during compression.
+ *  LZ4_loadDict() triggers a reset, so any previous data will be forgotten.
+ *  The same dictionary will have to be loaded on decompression side for successful decoding.
+ *  Dictionary are useful for better compression of small data (KB range).
+ *  While LZ4 accept any input as dictionary,
+ *  results are generally better when using Zstandard's Dictionary Builder.
+ *  Loading a size of 0 is allowed, and is the same as reset.
+ * @return : loaded dictionary size, in bytes (necessarily <= 64 KB)
+ */
+LZ4LIB_API int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
+
+/*! LZ4_compress_fast_continue() :
+ *  Compress 'src' content using data from previously compressed blocks, for better compression ratio.
+ * 'dst' buffer must be already allocated.
+ *  If dstCapacity >= LZ4_compressBound(srcSize), compression is guaranteed to succeed, and runs faster.
+ *
+ * @return : size of compressed block
+ *           or 0 if there is an error (typically, cannot fit into 'dst').
+ *
+ *  Note 1 : Each invocation to LZ4_compress_fast_continue() generates a new block.
+ *           Each block has precise boundaries.
+ *           Each block must be decompressed separately, calling LZ4_decompress_*() with relevant metadata.
+ *           It's not possible to append blocks together and expect a single invocation of LZ4_decompress_*() to decompress them together.
+ *
+ *  Note 2 : The previous 64KB of source data is __assumed__ to remain present, unmodified, at same address in memory !
+ *
+ *  Note 3 : When input is structured as a double-buffer, each buffer can have any size, including < 64 KB.
+ *           Make sure that buffers are separated, by at least one byte.
+ *           This construction ensures that each block only depends on previous block.
+ *
+ *  Note 4 : If input buffer is a ring-buffer, it can have any size, including < 64 KB.
+ *
+ *  Note 5 : After an error, the stream status is undefined (invalid), it can only be reset or freed.
+ */
+LZ4LIB_API int LZ4_compress_fast_continue (LZ4_stream_t* streamPtr, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_saveDict() :
+ *  If last 64KB data cannot be guaranteed to remain available at its current memory location,
+ *  save it into a safer place (char* safeBuffer).
+ *  This is schematically equivalent to a memcpy() followed by LZ4_loadDict(),
+ *  but is much faster, because LZ4_saveDict() doesn't need to rebuild tables.
+ * @return : saved dictionary size in bytes (necessarily <= maxDictSize), or 0 if error.
+ */
+LZ4LIB_API int LZ4_saveDict (LZ4_stream_t* streamPtr, char* safeBuffer, int maxDictSize);
+
+
+/*-**********************************************
+*  Streaming Decompression Functions
+*  Bufferless synchronous API
+************************************************/
+typedef union LZ4_streamDecode_u LZ4_streamDecode_t;   /* tracking context */
+
+/*! LZ4_createStreamDecode() and LZ4_freeStreamDecode() :
+ *  creation / destruction of streaming decompression tracking context.
+ *  A tracking context can be re-used multiple times.
+ */
+LZ4LIB_API LZ4_streamDecode_t* LZ4_createStreamDecode(void);
+LZ4LIB_API int                 LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream);
+
+/*! LZ4_setStreamDecode() :
+ *  An LZ4_streamDecode_t context can be allocated once and re-used multiple times.
+ *  Use this function to start decompression of a new stream of blocks.
+ *  A dictionary can optionally be set. Use NULL or size 0 for a reset order.
+ *  Dictionary is presumed stable : it must remain accessible and unmodified during next decompression.
+ * @return : 1 if OK, 0 if error
+ */
+LZ4LIB_API int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize);
+
+/*! LZ4_decoderRingBufferSize() : v1.8.2+
+ *  Note : in a ring buffer scenario (optional),
+ *  blocks are presumed decompressed next to each other
+ *  up to the moment there is not enough remaining space for next block (remainingSize < maxBlockSize),
+ *  at which stage it resumes from beginning of ring buffer.
+ *  When setting such a ring buffer for streaming decompression,
+ *  provides the minimum size of this ring buffer
+ *  to be compatible with any source respecting maxBlockSize condition.
+ * @return : minimum ring buffer size,
+ *           or 0 if there is an error (invalid maxBlockSize).
+ */
+LZ4LIB_API int LZ4_decoderRingBufferSize(int maxBlockSize);
+#define LZ4_DECODER_RING_BUFFER_SIZE(maxBlockSize) (65536 + 14 + (maxBlockSize))  /* for static allocation; maxBlockSize presumed valid */
+
+/*! LZ4_decompress_*_continue() :
+ *  These decoding functions allow decompression of consecutive blocks in "streaming" mode.
+ *  A block is an unsplittable entity, it must be presented entirely to a decompression function.
+ *  Decompression functions only accepts one block at a time.
+ *  The last 64KB of previously decoded data *must* remain available and unmodified at the memory position where they were decoded.
+ *  If less than 64KB of data has been decoded, all the data must be present.
+ *
+ *  Special : if decompression side sets a ring buffer, it must respect one of the following conditions :
+ *  - Decompression buffer size is _at least_ LZ4_decoderRingBufferSize(maxBlockSize).
+ *    maxBlockSize is the maximum size of any single block. It can have any value > 16 bytes.
+ *    In which case, encoding and decoding buffers do not need to be synchronized.
+ *    Actually, data can be produced by any source compliant with LZ4 format specification, and respecting maxBlockSize.
+ *  - Synchronized mode :
+ *    Decompression buffer size is _exactly_ the same as compression buffer size,
+ *    and follows exactly same update rule (block boundaries at same positions),
+ *    and decoding function is provided with exact decompressed size of each block (exception for last block of the stream),
+ *    _then_ decoding & encoding ring buffer can have any size, including small ones ( < 64 KB).
+ *  - Decompression buffer is larger than encoding buffer, by a minimum of maxBlockSize more bytes.
+ *    In which case, encoding and decoding buffers do not need to be synchronized,
+ *    and encoding ring buffer can have any size, including small ones ( < 64 KB).
+ *
+ *  Whenever these conditions are not possible,
+ *  save the last 64KB of decoded data into a safe buffer where it can't be modified during decompression,
+ *  then indicate where this data is saved using LZ4_setStreamDecode(), before decompressing next block.
+*/
+LZ4LIB_API int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* src, char* dst, int srcSize, int dstCapacity);
+
+
+/*! LZ4_decompress_*_usingDict() :
+ *  These decoding functions work the same as
+ *  a combination of LZ4_setStreamDecode() followed by LZ4_decompress_*_continue()
+ *  They are stand-alone, and don't need an LZ4_streamDecode_t structure.
+ *  Dictionary is presumed stable : it must remain accessible and unmodified during decompression.
+ *  Performance tip : Decompression speed can be substantially increased
+ *                    when dst == dictStart + dictSize.
+ */
+LZ4LIB_API int LZ4_decompress_safe_usingDict (const char* src, char* dst, int srcSize, int dstCapcity, const char* dictStart, int dictSize);
+
+#endif /* LZ4_H_2983827168210 */
+
+
+/*^*************************************
+ * !!!!!!   STATIC LINKING ONLY   !!!!!!
+ ***************************************/
+
+/*-****************************************************************************
+ * Experimental section
+ *
+ * Symbols declared in this section must be considered unstable. Their
+ * signatures or semantics may change, or they may be removed altogether in the
+ * future. They are therefore only safe to depend on when the caller is
+ * statically linked against the library.
+ *
+ * To protect against unsafe usage, not only are the declarations guarded,
+ * the definitions are hidden by default
+ * when building LZ4 as a shared/dynamic library.
+ *
+ * In order to access these declarations,
+ * define LZ4_STATIC_LINKING_ONLY in your application
+ * before including LZ4's headers.
+ *
+ * In order to make their implementations accessible dynamically, you must
+ * define LZ4_PUBLISH_STATIC_FUNCTIONS when building the LZ4 library.
+ ******************************************************************************/
+
+#ifdef LZ4_STATIC_LINKING_ONLY
+
+#ifndef LZ4_STATIC_3504398509
+#define LZ4_STATIC_3504398509
+
+#ifdef LZ4_PUBLISH_STATIC_FUNCTIONS
+#define LZ4LIB_STATIC_API LZ4LIB_API
+#else
+#define LZ4LIB_STATIC_API
+#endif
+
+
+/*! LZ4_compress_fast_extState_fastReset() :
+ *  A variant of LZ4_compress_fast_extState().
+ *
+ *  Using this variant avoids an expensive initialization step.
+ *  It is only safe to call if the state buffer is known to be correctly initialized already
+ *  (see above comment on LZ4_resetStream_fast() for a definition of "correctly initialized").
+ *  From a high level, the difference is that
+ *  this function initializes the provided state with a call to something like LZ4_resetStream_fast()
+ *  while LZ4_compress_fast_extState() starts with a call to LZ4_resetStream().
+ */
+LZ4LIB_STATIC_API int LZ4_compress_fast_extState_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_attach_dictionary() :
+ *  This is an experimental API that allows
+ *  efficient use of a static dictionary many times.
+ *
+ *  Rather than re-loading the dictionary buffer into a working context before
+ *  each compression, or copying a pre-loaded dictionary's LZ4_stream_t into a
+ *  working LZ4_stream_t, this function introduces a no-copy setup mechanism,
+ *  in which the working stream references the dictionary stream in-place.
+ *
+ *  Several assumptions are made about the state of the dictionary stream.
+ *  Currently, only streams which have been prepared by LZ4_loadDict() should
+ *  be expected to work.
+ *
+ *  Alternatively, the provided dictionaryStream may be NULL,
+ *  in which case any existing dictionary stream is unset.
+ *
+ *  If a dictionary is provided, it replaces any pre-existing stream history.
+ *  The dictionary contents are the only history that can be referenced and
+ *  logically immediately precede the data compressed in the first subsequent
+ *  compression call.
+ *
+ *  The dictionary will only remain attached to the working stream through the
+ *  first compression call, at the end of which it is cleared. The dictionary
+ *  stream (and source buffer) must remain in-place / accessible / unchanged
+ *  through the completion of the first compression call on the stream.
+ */
+LZ4LIB_STATIC_API void LZ4_attach_dictionary(LZ4_stream_t* workingStream, const LZ4_stream_t* dictionaryStream);
+
+
+/*! In-place compression and decompression
+ *
+ * It's possible to have input and output sharing the same buffer,
+ * for highly contrained memory environments.
+ * In both cases, it requires input to lay at the end of the buffer,
+ * and decompression to start at beginning of the buffer.
+ * Buffer size must feature some margin, hence be larger than final size.
+ *
+ * |<------------------------buffer--------------------------------->|
+ *                             |<-----------compressed data--------->|
+ * |<-----------decompressed size------------------>|
+ *                                                  |<----margin---->|
+ *
+ * This technique is more useful for decompression,
+ * since decompressed size is typically larger,
+ * and margin is short.
+ *
+ * In-place decompression will work inside any buffer
+ * which size is >= LZ4_DECOMPRESS_INPLACE_BUFFER_SIZE(decompressedSize).
+ * This presumes that decompressedSize > compressedSize.
+ * Otherwise, it means compression actually expanded data,
+ * and it would be more efficient to store such data with a flag indicating it's not compressed.
+ * This can happen when data is not compressible (already compressed, or encrypted).
+ *
+ * For in-place compression, margin is larger, as it must be able to cope with both
+ * history preservation, requiring input data to remain unmodified up to LZ4_DISTANCE_MAX,
+ * and data expansion, which can happen when input is not compressible.
+ * As a consequence, buffer size requirements are much higher,
+ * and memory savings offered by in-place compression are more limited.
+ *
+ * There are ways to limit this cost for compression :
+ * - Reduce history size, by modifying LZ4_DISTANCE_MAX.
+ *   Note that it is a compile-time constant, so all compressions will apply this limit.
+ *   Lower values will reduce compression ratio, except when input_size < LZ4_DISTANCE_MAX,
+ *   so it's a reasonable trick when inputs are known to be small.
+ * - Require the compressor to deliver a "maximum compressed size".
+ *   This is the `dstCapacity` parameter in `LZ4_compress*()`.
+ *   When this size is < LZ4_COMPRESSBOUND(inputSize), then compression can fail,
+ *   in which case, the return code will be 0 (zero).
+ *   The caller must be ready for these cases to happen,
+ *   and typically design a backup scheme to send data uncompressed.
+ * The combination of both techniques can significantly reduce
+ * the amount of margin required for in-place compression.
+ *
+ * In-place compression can work in any buffer
+ * which size is >= (maxCompressedSize)
+ * with maxCompressedSize == LZ4_COMPRESSBOUND(srcSize) for guaranteed compression success.
+ * LZ4_COMPRESS_INPLACE_BUFFER_SIZE() depends on both maxCompressedSize and LZ4_DISTANCE_MAX,
+ * so it's possible to reduce memory requirements by playing with them.
+ */
+
+#define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)          (((compressedSize) >> 8) + 32)
+#define LZ4_DECOMPRESS_INPLACE_BUFFER_SIZE(decompressedSize)   ((decompressedSize) + LZ4_DECOMPRESS_INPLACE_MARGIN(decompressedSize))  /**< note: presumes that compressedSize < decompressedSize. note2: margin is overestimated a bit, since it could use compressedSize instead */
+
+#ifndef LZ4_DISTANCE_MAX   /* history window size; can be user-defined at compile time */
+#  define LZ4_DISTANCE_MAX 65535   /* set to maximum value by default */
+#endif
+
+#define LZ4_COMPRESS_INPLACE_MARGIN                           (LZ4_DISTANCE_MAX + 32)   /* LZ4_DISTANCE_MAX can be safely replaced by srcSize when it's smaller */
+#define LZ4_COMPRESS_INPLACE_BUFFER_SIZE(maxCompressedSize)   ((maxCompressedSize) + LZ4_COMPRESS_INPLACE_MARGIN)  /**< maxCompressedSize is generally LZ4_COMPRESSBOUND(inputSize), but can be set to any lower value, with the risk that compression can fail (return code 0(zero)) */
+
+#endif   /* LZ4_STATIC_3504398509 */
+#endif   /* LZ4_STATIC_LINKING_ONLY */
+
+
+
+#ifndef LZ4_H_98237428734687
+#define LZ4_H_98237428734687
+
+/*-************************************************************
+ *  PRIVATE DEFINITIONS
+ **************************************************************
+ * Do not use these definitions directly.
+ * They are only exposed to allow static allocation of `LZ4_stream_t` and `LZ4_streamDecode_t`.
+ * Accessing members will expose code to API and/or ABI break in future versions of the library.
+ **************************************************************/
+#define LZ4_HASHLOG   (LZ4_MEMORY_USAGE-2)
+#define LZ4_HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
+#define LZ4_HASH_SIZE_U32 (1 << LZ4_HASHLOG)       /* required as macro for static allocation */
+
+#if defined(__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#include <stdint.h>
+
+typedef struct LZ4_stream_t_internal LZ4_stream_t_internal;
+struct LZ4_stream_t_internal {
+    uint32_t hashTable[LZ4_HASH_SIZE_U32];
+    uint32_t currentOffset;
+    uint16_t dirty;
+    uint16_t tableType;
+    const uint8_t* dictionary;
+    const LZ4_stream_t_internal* dictCtx;
+    uint32_t dictSize;
+};
+
+typedef struct {
+    const uint8_t* externalDict;
+    size_t extDictSize;
+    const uint8_t* prefixEnd;
+    size_t prefixSize;
+} LZ4_streamDecode_t_internal;
+
+#else
+
+typedef struct LZ4_stream_t_internal LZ4_stream_t_internal;
+struct LZ4_stream_t_internal {
+    unsigned int hashTable[LZ4_HASH_SIZE_U32];
+    unsigned int currentOffset;
+    unsigned short dirty;
+    unsigned short tableType;
+    const unsigned char* dictionary;
+    const LZ4_stream_t_internal* dictCtx;
+    unsigned int dictSize;
+};
+
+typedef struct {
+    const unsigned char* externalDict;
+    const unsigned char* prefixEnd;
+    size_t extDictSize;
+    size_t prefixSize;
+} LZ4_streamDecode_t_internal;
+
+#endif
+
+/*! LZ4_stream_t :
+ *  information structure to track an LZ4 stream.
+ *  LZ4_stream_t can also be created using LZ4_createStream(), which is recommended.
+ *  The structure definition can be convenient for static allocation
+ *  (on stack, or as part of larger structure).
+ *  Init this structure with LZ4_initStream() before first use.
+ *  note : only use this definition in association with static linking !
+ *    this definition is not API/ABI safe, and may change in a future version.
+ */
+#define LZ4_STREAMSIZE_U64 ((1 << (LZ4_MEMORY_USAGE-3)) + 4 + ((sizeof(void*)==16) ? 4 : 0) /*AS-400*/ )
+#define LZ4_STREAMSIZE     (LZ4_STREAMSIZE_U64 * sizeof(unsigned long long))
+union LZ4_stream_u {
+    unsigned long long table[LZ4_STREAMSIZE_U64];
+    LZ4_stream_t_internal internal_donotuse;
+} ;  /* previously typedef'd to LZ4_stream_t */
+
+/*! LZ4_initStream() : v1.9.0+
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is automatically done when invoking LZ4_createStream(),
+ *  but it's not when the structure is simply declared on stack (for example).
+ *
+ *  Use LZ4_initStream() to properly initialize a newly declared LZ4_stream_t.
+ *  It can also initialize any arbitrary buffer of sufficient size,
+ *  and will @return a pointer of proper type upon initialization.
+ *
+ *  Note : initialization fails if size and alignment conditions are not respected.
+ *         In which case, the function will @return NULL.
+ *  Note2: An LZ4_stream_t structure guarantees correct alignment and size.
+ *  Note3: Before v1.9.0, use LZ4_resetStream() instead
+ */
+LZ4LIB_API LZ4_stream_t* LZ4_initStream (void* buffer, size_t size);
+
+
+/*! LZ4_streamDecode_t :
+ *  information structure to track an LZ4 stream during decompression.
+ *  init this structure  using LZ4_setStreamDecode() before first use.
+ *  note : only use in association with static linking !
+ *         this definition is not API/ABI safe,
+ *         and may change in a future version !
+ */
+#define LZ4_STREAMDECODESIZE_U64 (4 + ((sizeof(void*)==16) ? 2 : 0) /*AS-400*/ )
+#define LZ4_STREAMDECODESIZE     (LZ4_STREAMDECODESIZE_U64 * sizeof(unsigned long long))
+union LZ4_streamDecode_u {
+    unsigned long long table[LZ4_STREAMDECODESIZE_U64];
+    LZ4_streamDecode_t_internal internal_donotuse;
+} ;   /* previously typedef'd to LZ4_streamDecode_t */
+
+
+
+/*-************************************
+*  Obsolete Functions
+**************************************/
+
+/*! Deprecation warnings
+ *
+ *  Deprecated functions make the compiler generate a warning when invoked.
+ *  This is meant to invite users to update their source code.
+ *  Should deprecation warnings be a problem, it is generally possible to disable them,
+ *  typically with -Wno-deprecated-declarations for gcc
+ *  or _CRT_SECURE_NO_WARNINGS in Visual.
+ *
+ *  Another method is to define LZ4_DISABLE_DEPRECATE_WARNINGS
+ *  before including the header file.
+ */
+#ifdef LZ4_DISABLE_DEPRECATE_WARNINGS
+#  define LZ4_DEPRECATED(message)   /* disable deprecation warnings */
+#else
+#  define LZ4_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define LZ4_DEPRECATED(message) [[deprecated(message)]]
+#  elif (LZ4_GCC_VERSION >= 405) || defined(__clang__)
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif (LZ4_GCC_VERSION >= 301)
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define LZ4_DEPRECATED(message) __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement LZ4_DEPRECATED for this compiler")
+#    define LZ4_DEPRECATED(message)
+#  endif
+#endif /* LZ4_DISABLE_DEPRECATE_WARNINGS */
+
+/* Obsolete compression functions */
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress               (const char* src, char* dest, int srcSize);
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress_limitedOutput (const char* src, char* dest, int srcSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_withState               (void* state, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_limitedOutput_withState (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_continue() instead") LZ4LIB_API int LZ4_compress_continue                (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_continue() instead") LZ4LIB_API int LZ4_compress_limitedOutput_continue  (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
+
+/* Obsolete decompression functions */
+LZ4_DEPRECATED("use LZ4_decompress_fast() instead") LZ4LIB_API int LZ4_uncompress (const char* source, char* dest, int outputSize);
+LZ4_DEPRECATED("use LZ4_decompress_safe() instead") LZ4LIB_API int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize);
+
+/* Obsolete streaming functions; degraded functionality; do not use!
+ *
+ * In order to perform streaming compression, these functions depended on data
+ * that is no longer tracked in the state. They have been preserved as well as
+ * possible: using them will still produce a correct output. However, they don't
+ * actually retain any history between compression calls. The compression ratio
+ * achieved will therefore be no better than compressing each chunk
+ * independently.
+ */
+LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API void* LZ4_create (char* inputBuffer);
+LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API int   LZ4_sizeofStreamState(void);
+LZ4_DEPRECATED("Use LZ4_resetStream() instead")  LZ4LIB_API int   LZ4_resetStreamState(void* state, char* inputBuffer);
+LZ4_DEPRECATED("Use LZ4_saveDict() instead")     LZ4LIB_API char* LZ4_slideInputBuffer (void* state);
+
+/* Obsolete streaming decoding functions */
+LZ4_DEPRECATED("use LZ4_decompress_safe_usingDict() instead") LZ4LIB_API int LZ4_decompress_safe_withPrefix64k (const char* src, char* dst, int compressedSize, int maxDstSize);
+LZ4_DEPRECATED("use LZ4_decompress_fast_usingDict() instead") LZ4LIB_API int LZ4_decompress_fast_withPrefix64k (const char* src, char* dst, int originalSize);
+
+/*! LZ4_decompress_fast() : **unsafe!**
+ *  These functions used to be faster than LZ4_decompress_safe(),
+ *  but it has changed, and they are now slower than LZ4_decompress_safe().
+ *  This is because LZ4_decompress_fast() doesn't know the input size,
+ *  and therefore must progress more cautiously in the input buffer to not read beyond the end of block.
+ *  On top of that `LZ4_decompress_fast()` is not protected vs malformed or malicious inputs, making it a security liability.
+ *  As a consequence, LZ4_decompress_fast() is strongly discouraged, and deprecated.
+ *
+ *  The last remaining LZ4_decompress_fast() specificity is that
+ *  it can decompress a block without knowing its compressed size.
+ *  Such functionality could be achieved in a more secure manner,
+ *  by also providing the maximum size of input buffer,
+ *  but it would require new prototypes, and adaptation of the implementation to this new use case.
+ *
+ *  Parameters:
+ *  originalSize : is the uncompressed size to regenerate.
+ *                 `dst` must be already allocated, its size must be >= 'originalSize' bytes.
+ * @return : number of bytes read from source buffer (== compressed size).
+ *           The function expects to finish at block's end exactly.
+ *           If the source stream is detected malformed, the function stops decoding and returns a negative result.
+ *  note : LZ4_decompress_fast*() requires originalSize. Thanks to this information, it never writes past the output buffer.
+ *         However, since it doesn't know its 'src' size, it may read an unknown amount of input, past input buffer bounds.
+ *         Also, since match offsets are not validated, match reads from 'src' may underflow too.
+ *         These issues never happen if input (compressed) data is correct.
+ *         But they may happen if input data is invalid (error or intentional tampering).
+ *         As a consequence, use these functions in trusted environments with trusted data **only**.
+ */
+
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe() instead")
+LZ4LIB_API int LZ4_decompress_fast (const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_continue() instead")
+LZ4LIB_API int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_usingDict() instead")
+LZ4LIB_API int LZ4_decompress_fast_usingDict (const char* src, char* dst, int originalSize, const char* dictStart, int dictSize);
+
+/*! LZ4_resetStream() :
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is done with LZ4_initStream(), or LZ4_resetStream().
+ *  Consider switching to LZ4_initStream(),
+ *  invoking LZ4_resetStream() will trigger deprecation warnings in the future.
+ */
+LZ4LIB_API void LZ4_resetStream (LZ4_stream_t* streamPtr);
+
+
+#endif /* LZ4_H_98237428734687 */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4/lz4hc.c b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4hc.c
new file mode 100644
index 0000000..b75514f
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4hc.c
@@ -0,0 +1,1554 @@
+/*
+    LZ4 HC - High Compression Mode of LZ4
+    Copyright (C) 2011-2017, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+       - LZ4 source repository : https://github.com/lz4/lz4
+       - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+/* note : lz4hc is not an independent module, it requires lz4.h/lz4.c for proper compilation */
+
+
+/* *************************************
+*  Tuning Parameter
+***************************************/
+
+/*! HEAPMODE :
+ *  Select how default compression function will allocate workplace memory,
+ *  in stack (0:fastest), or in heap (1:requires malloc()).
+ *  Since workplace is rather large, heap mode is recommended.
+ */
+#ifndef LZ4HC_HEAPMODE
+#  define LZ4HC_HEAPMODE 1
+#endif
+
+
+/*===    Dependency    ===*/
+#define LZ4_HC_STATIC_LINKING_ONLY
+#include "lz4hc.h"
+
+
+/*===   Common LZ4 definitions   ===*/
+#if defined(__GNUC__)
+#  pragma GCC diagnostic ignored "-Wunused-function"
+#endif
+#if defined (__clang__)
+#  pragma clang diagnostic ignored "-Wunused-function"
+#endif
+
+/*===   Enums   ===*/
+typedef enum { noDictCtx, usingDictCtxHc } dictCtx_directive;
+
+
+#define LZ4_COMMONDEFS_ONLY
+#ifndef LZ4_SRC_INCLUDED
+#include "lz4.c"   /* LZ4_count, constants, mem */
+#endif
+
+/*===   Constants   ===*/
+#define OPTIMAL_ML (int)((ML_MASK-1)+MINMATCH)
+#define LZ4_OPT_NUM   (1<<12)
+
+
+/*===   Macros   ===*/
+#define MIN(a,b)   ( (a) < (b) ? (a) : (b) )
+#define MAX(a,b)   ( (a) > (b) ? (a) : (b) )
+#define HASH_FUNCTION(i)         (((i) * 2654435761U) >> ((MINMATCH*8)-LZ4HC_HASH_LOG))
+#define DELTANEXTMAXD(p)         chainTable[(p) & LZ4HC_MAXD_MASK]    /* flexible, LZ4HC_MAXD dependent */
+#define DELTANEXTU16(table, pos) table[(U16)(pos)]   /* faster */
+/* Make fields passed to, and updated by LZ4HC_encodeSequence explicit */
+#define UPDATABLE(ip, op, anchor) &ip, &op, &anchor
+
+static U32 LZ4HC_hashPtr(const void* ptr) { return HASH_FUNCTION(LZ4_read32(ptr)); }
+
+
+/**************************************
+*  HC Compression
+**************************************/
+static void LZ4HC_clearTables (LZ4HC_CCtx_internal* hc4)
+{
+    MEM_INIT((void*)hc4->hashTable, 0, sizeof(hc4->hashTable));
+    MEM_INIT(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
+}
+
+static void LZ4HC_init_internal (LZ4HC_CCtx_internal* hc4, const BYTE* start)
+{
+    uptrval startingOffset = (uptrval)(hc4->end - hc4->base);
+    if (startingOffset > 1 GB) {
+        LZ4HC_clearTables(hc4);
+        startingOffset = 0;
+    }
+    startingOffset += 64 KB;
+    hc4->nextToUpdate = (U32) startingOffset;
+    hc4->base = start - startingOffset;
+    hc4->end = start;
+    hc4->dictBase = start - startingOffset;
+    hc4->dictLimit = (U32) startingOffset;
+    hc4->lowLimit = (U32) startingOffset;
+}
+
+
+/* Update chains up to ip (excluded) */
+LZ4_FORCE_INLINE void LZ4HC_Insert (LZ4HC_CCtx_internal* hc4, const BYTE* ip)
+{
+    U16* const chainTable = hc4->chainTable;
+    U32* const hashTable  = hc4->hashTable;
+    const BYTE* const base = hc4->base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = hc4->nextToUpdate;
+
+    while (idx < target) {
+        U32 const h = LZ4HC_hashPtr(base+idx);
+        size_t delta = idx - hashTable[h];
+        if (delta>LZ4_DISTANCE_MAX) delta = LZ4_DISTANCE_MAX;
+        DELTANEXTU16(chainTable, idx) = (U16)delta;
+        hashTable[h] = idx;
+        idx++;
+    }
+
+    hc4->nextToUpdate = target;
+}
+
+/** LZ4HC_countBack() :
+ * @return : negative value, nb of common bytes before ip/match */
+LZ4_FORCE_INLINE
+int LZ4HC_countBack(const BYTE* const ip, const BYTE* const match,
+                    const BYTE* const iMin, const BYTE* const mMin)
+{
+    int back = 0;
+    int const min = (int)MAX(iMin - ip, mMin - match);
+    assert(min <= 0);
+    assert(ip >= iMin); assert((size_t)(ip-iMin) < (1U<<31));
+    assert(match >= mMin); assert((size_t)(match - mMin) < (1U<<31));
+    while ( (back > min)
+         && (ip[back-1] == match[back-1]) )
+            back--;
+    return back;
+}
+
+#if defined(_MSC_VER)
+#  define LZ4HC_rotl32(x,r) _rotl(x,r)
+#else
+#  define LZ4HC_rotl32(x,r) ((x << r) | (x >> (32 - r)))
+#endif
+
+
+static U32 LZ4HC_rotatePattern(size_t const rotate, U32 const pattern)
+{
+    size_t const bitsToRotate = (rotate & (sizeof(pattern) - 1)) << 3;
+    if (bitsToRotate == 0)
+        return pattern;
+    return LZ4HC_rotl32(pattern, (int)bitsToRotate);
+}
+
+/* LZ4HC_countPattern() :
+ * pattern32 must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!) */
+static unsigned
+LZ4HC_countPattern(const BYTE* ip, const BYTE* const iEnd, U32 const pattern32)
+{
+    const BYTE* const iStart = ip;
+    reg_t const pattern = (sizeof(pattern)==8) ? (reg_t)pattern32 + (((reg_t)pattern32) << 32) : pattern32;
+
+    while (likely(ip < iEnd-(sizeof(pattern)-1))) {
+        reg_t const diff = LZ4_read_ARCH(ip) ^ pattern;
+        if (!diff) { ip+=sizeof(pattern); continue; }
+        ip += LZ4_NbCommonBytes(diff);
+        return (unsigned)(ip - iStart);
+    }
+
+    if (LZ4_isLittleEndian()) {
+        reg_t patternByte = pattern;
+        while ((ip<iEnd) && (*ip == (BYTE)patternByte)) {
+            ip++; patternByte >>= 8;
+        }
+    } else {  /* big endian */
+        U32 bitOffset = (sizeof(pattern)*8) - 8;
+        while (ip < iEnd) {
+            BYTE const byte = (BYTE)(pattern >> bitOffset);
+            if (*ip != byte) break;
+            ip ++; bitOffset -= 8;
+        }
+    }
+
+    return (unsigned)(ip - iStart);
+}
+
+/* LZ4HC_reverseCountPattern() :
+ * pattern must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!)
+ * read using natural platform endianess */
+static unsigned
+LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow, U32 pattern)
+{
+    const BYTE* const iStart = ip;
+
+    while (likely(ip >= iLow+4)) {
+        if (LZ4_read32(ip-4) != pattern) break;
+        ip -= 4;
+    }
+    {   const BYTE* bytePtr = (const BYTE*)(&pattern) + 3; /* works for any endianess */
+        while (likely(ip>iLow)) {
+            if (ip[-1] != *bytePtr) break;
+            ip--; bytePtr--;
+    }   }
+    return (unsigned)(iStart - ip);
+}
+
+/* LZ4HC_protectDictEnd() :
+ * Checks if the match is in the last 3 bytes of the dictionary, so reading the
+ * 4 byte MINMATCH would overflow.
+ * @returns true if the match index is okay.
+ */
+static int LZ4HC_protectDictEnd(U32 const dictLimit, U32 const matchIndex)
+{
+    return ((U32)((dictLimit - 1) - matchIndex) >= 3);
+}
+
+typedef enum { rep_untested, rep_not, rep_confirmed } repeat_state_e;
+typedef enum { favorCompressionRatio=0, favorDecompressionSpeed } HCfavor_e;
+
+LZ4_FORCE_INLINE int
+LZ4HC_InsertAndGetWiderMatch (
+    LZ4HC_CCtx_internal* hc4,
+    const BYTE* const ip,
+    const BYTE* const iLowLimit,
+    const BYTE* const iHighLimit,
+    int longest,
+    const BYTE** matchpos,
+    const BYTE** startpos,
+    const int maxNbAttempts,
+    const int patternAnalysis,
+    const int chainSwap,
+    const dictCtx_directive dict,
+    const HCfavor_e favorDecSpeed)
+{
+    U16* const chainTable = hc4->chainTable;
+    U32* const HashTable = hc4->hashTable;
+    const LZ4HC_CCtx_internal * const dictCtx = hc4->dictCtx;
+    const BYTE* const base = hc4->base;
+    const U32 dictLimit = hc4->dictLimit;
+    const BYTE* const lowPrefixPtr = base + dictLimit;
+    const U32 ipIndex = (U32)(ip - base);
+    const U32 lowestMatchIndex = (hc4->lowLimit + (LZ4_DISTANCE_MAX + 1) > ipIndex) ? hc4->lowLimit : ipIndex - LZ4_DISTANCE_MAX;
+    const BYTE* const dictBase = hc4->dictBase;
+    int const lookBackLength = (int)(ip-iLowLimit);
+    int nbAttempts = maxNbAttempts;
+    U32 matchChainPos = 0;
+    U32 const pattern = LZ4_read32(ip);
+    U32 matchIndex;
+    repeat_state_e repeat = rep_untested;
+    size_t srcPatternLength = 0;
+
+    DEBUGLOG(7, "LZ4HC_InsertAndGetWiderMatch");
+    /* First Match */
+    LZ4HC_Insert(hc4, ip);
+    matchIndex = HashTable[LZ4HC_hashPtr(ip)];
+    DEBUGLOG(7, "First match at index %u / %u (lowestMatchIndex)",
+                matchIndex, lowestMatchIndex);
+
+    while ((matchIndex>=lowestMatchIndex) && (nbAttempts)) {
+        int matchLength=0;
+        nbAttempts--;
+        assert(matchIndex < ipIndex);
+        if (favorDecSpeed && (ipIndex - matchIndex < 8)) {
+            /* do nothing */
+        } else if (matchIndex >= dictLimit) {   /* within current Prefix */
+            const BYTE* const matchPtr = base + matchIndex;
+            assert(matchPtr >= lowPrefixPtr);
+            assert(matchPtr < ip);
+            assert(longest >= 1);
+            if (LZ4_read16(iLowLimit + longest - 1) == LZ4_read16(matchPtr - lookBackLength + longest - 1)) {
+                if (LZ4_read32(matchPtr) == pattern) {
+                    int const back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, lowPrefixPtr) : 0;
+                    matchLength = MINMATCH + (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
+                    matchLength -= back;
+                    if (matchLength > longest) {
+                        longest = matchLength;
+                        *matchpos = matchPtr + back;
+                        *startpos = ip + back;
+            }   }   }
+        } else {   /* lowestMatchIndex <= matchIndex < dictLimit */
+            const BYTE* const matchPtr = dictBase + matchIndex;
+            if (LZ4_read32(matchPtr) == pattern) {
+                const BYTE* const dictStart = dictBase + hc4->lowLimit;
+                int back = 0;
+                const BYTE* vLimit = ip + (dictLimit - matchIndex);
+                if (vLimit > iHighLimit) vLimit = iHighLimit;
+                matchLength = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                if ((ip+matchLength == vLimit) && (vLimit < iHighLimit))
+                    matchLength += LZ4_count(ip+matchLength, lowPrefixPtr, iHighLimit);
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictStart) : 0;
+                matchLength -= back;
+                if (matchLength > longest) {
+                    longest = matchLength;
+                    *matchpos = base + matchIndex + back;   /* virtual pos, relative to ip, to retrieve offset */
+                    *startpos = ip + back;
+        }   }   }
+
+        if (chainSwap && matchLength==longest) {    /* better match => select a better chain */
+            assert(lookBackLength==0);   /* search forward only */
+            if (matchIndex + (U32)longest <= ipIndex) {
+                int const kTrigger = 4;
+                U32 distanceToNextMatch = 1;
+                int const end = longest - MINMATCH + 1;
+                int step = 1;
+                int accel = 1 << kTrigger;
+                int pos;
+                for (pos = 0; pos < end; pos += step) {
+                    U32 const candidateDist = DELTANEXTU16(chainTable, matchIndex + (U32)pos);
+                    step = (accel++ >> kTrigger);
+                    if (candidateDist > distanceToNextMatch) {
+                        distanceToNextMatch = candidateDist;
+                        matchChainPos = (U32)pos;
+                        accel = 1 << kTrigger;
+                    }
+                }
+                if (distanceToNextMatch > 1) {
+                    if (distanceToNextMatch > matchIndex) break;   /* avoid overflow */
+                    matchIndex -= distanceToNextMatch;
+                    continue;
+        }   }   }
+
+        {   U32 const distNextMatch = DELTANEXTU16(chainTable, matchIndex);
+            if (patternAnalysis && distNextMatch==1 && matchChainPos==0) {
+                U32 const matchCandidateIdx = matchIndex-1;
+                /* may be a repeated pattern */
+                if (repeat == rep_untested) {
+                    if ( ((pattern & 0xFFFF) == (pattern >> 16))
+                      &  ((pattern & 0xFF)   == (pattern >> 24)) ) {
+                        repeat = rep_confirmed;
+                        srcPatternLength = LZ4HC_countPattern(ip+sizeof(pattern), iHighLimit, pattern) + sizeof(pattern);
+                    } else {
+                        repeat = rep_not;
+                }   }
+                if ( (repeat == rep_confirmed) && (matchCandidateIdx >= lowestMatchIndex)
+                  && LZ4HC_protectDictEnd(dictLimit, matchCandidateIdx) ) {
+                    const int extDict = matchCandidateIdx < dictLimit;
+                    const BYTE* const matchPtr = (extDict ? dictBase : base) + matchCandidateIdx;
+                    if (LZ4_read32(matchPtr) == pattern) {  /* good candidate */
+                        const BYTE* const dictStart = dictBase + hc4->lowLimit;
+                        const BYTE* const iLimit = extDict ? dictBase + dictLimit : iHighLimit;
+                        size_t forwardPatternLength = LZ4HC_countPattern(matchPtr+sizeof(pattern), iLimit, pattern) + sizeof(pattern);
+                        if (extDict && matchPtr + forwardPatternLength == iLimit) {
+                            U32 const rotatedPattern = LZ4HC_rotatePattern(forwardPatternLength, pattern);
+                            forwardPatternLength += LZ4HC_countPattern(lowPrefixPtr, iHighLimit, rotatedPattern);
+                        }
+                        {   const BYTE* const lowestMatchPtr = extDict ? dictStart : lowPrefixPtr;
+                            size_t backLength = LZ4HC_reverseCountPattern(matchPtr, lowestMatchPtr, pattern);
+                            size_t currentSegmentLength;
+                            if (!extDict && matchPtr - backLength == lowPrefixPtr && hc4->lowLimit < dictLimit) {
+                                U32 const rotatedPattern = LZ4HC_rotatePattern((U32)(-(int)backLength), pattern);
+                                backLength += LZ4HC_reverseCountPattern(dictBase + dictLimit, dictStart, rotatedPattern);
+                            }
+                            /* Limit backLength not go further than lowestMatchIndex */
+                            backLength = matchCandidateIdx - MAX(matchCandidateIdx - (U32)backLength, lowestMatchIndex);
+                            assert(matchCandidateIdx - backLength >= lowestMatchIndex);
+                            currentSegmentLength = backLength + forwardPatternLength;
+                            /* Adjust to end of pattern if the source pattern fits, otherwise the beginning of the pattern */
+                            if ( (currentSegmentLength >= srcPatternLength)   /* current pattern segment large enough to contain full srcPatternLength */
+                              && (forwardPatternLength <= srcPatternLength) ) { /* haven't reached this position yet */
+                                U32 const newMatchIndex = matchCandidateIdx + (U32)forwardPatternLength - (U32)srcPatternLength;  /* best position, full pattern, might be followed by more match */
+                                if (LZ4HC_protectDictEnd(dictLimit, newMatchIndex))
+                                    matchIndex = newMatchIndex;
+                                else {
+                                    /* Can only happen if started in the prefix */
+                                    assert(newMatchIndex >= dictLimit - 3 && newMatchIndex < dictLimit && !extDict);
+                                    matchIndex = dictLimit;
+                                }
+                            } else {
+                                U32 const newMatchIndex = matchCandidateIdx - (U32)backLength;   /* farthest position in current segment, will find a match of length currentSegmentLength + maybe some back */
+                                if (!LZ4HC_protectDictEnd(dictLimit, newMatchIndex)) {
+                                    assert(newMatchIndex >= dictLimit - 3 && newMatchIndex < dictLimit && !extDict);
+                                    matchIndex = dictLimit;
+                                } else {
+                                    matchIndex = newMatchIndex;
+                                    if (lookBackLength==0) {  /* no back possible */
+                                        size_t const maxML = MIN(currentSegmentLength, srcPatternLength);
+                                        if ((size_t)longest < maxML) {
+                                            assert(base + matchIndex != ip);
+                                            if ((size_t)(ip - base) - matchIndex > LZ4_DISTANCE_MAX) break;
+                                            assert(maxML < 2 GB);
+                                            longest = (int)maxML;
+                                            *matchpos = base + matchIndex;   /* virtual pos, relative to ip, to retrieve offset */
+                                            *startpos = ip;
+                                        }
+                                        {   U32 const distToNextPattern = DELTANEXTU16(chainTable, matchIndex);
+                                            if (distToNextPattern > matchIndex) break;  /* avoid overflow */
+                                            matchIndex -= distToNextPattern;
+                        }   }   }   }   }
+                        continue;
+                }   }
+        }   }   /* PA optimization */
+
+        /* follow current chain */
+        matchIndex -= DELTANEXTU16(chainTable, matchIndex + matchChainPos);
+
+    }  /* while ((matchIndex>=lowestMatchIndex) && (nbAttempts)) */
+
+    if ( dict == usingDictCtxHc
+      && nbAttempts
+      && ipIndex - lowestMatchIndex < LZ4_DISTANCE_MAX) {
+        size_t const dictEndOffset = (size_t)(dictCtx->end - dictCtx->base);
+        U32 dictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
+        assert(dictEndOffset <= 1 GB);
+        matchIndex = dictMatchIndex + lowestMatchIndex - (U32)dictEndOffset;
+        while (ipIndex - matchIndex <= LZ4_DISTANCE_MAX && nbAttempts--) {
+            const BYTE* const matchPtr = dictCtx->base + dictMatchIndex;
+
+            if (LZ4_read32(matchPtr) == pattern) {
+                int mlt;
+                int back = 0;
+                const BYTE* vLimit = ip + (dictEndOffset - dictMatchIndex);
+                if (vLimit > iHighLimit) vLimit = iHighLimit;
+                mlt = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictCtx->base + dictCtx->dictLimit) : 0;
+                mlt -= back;
+                if (mlt > longest) {
+                    longest = mlt;
+                    *matchpos = base + matchIndex + back;
+                    *startpos = ip + back;
+            }   }
+
+            {   U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, dictMatchIndex);
+                dictMatchIndex -= nextOffset;
+                matchIndex -= nextOffset;
+    }   }   }
+
+    return longest;
+}
+
+LZ4_FORCE_INLINE
+int LZ4HC_InsertAndFindBestMatch(LZ4HC_CCtx_internal* const hc4,   /* Index table will be updated */
+                                 const BYTE* const ip, const BYTE* const iLimit,
+                                 const BYTE** matchpos,
+                                 const int maxNbAttempts,
+                                 const int patternAnalysis,
+                                 const dictCtx_directive dict)
+{
+    const BYTE* uselessPtr = ip;
+    /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
+     * but this won't be the case here, as we define iLowLimit==ip,
+     * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
+    return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, matchpos, &uselessPtr, maxNbAttempts, patternAnalysis, 0 /*chainSwap*/, dict, favorCompressionRatio);
+}
+
+/* LZ4HC_encodeSequence() :
+ * @return : 0 if ok,
+ *           1 if buffer issue detected */
+LZ4_FORCE_INLINE int LZ4HC_encodeSequence (
+    const BYTE** ip,
+    BYTE** op,
+    const BYTE** anchor,
+    int matchLength,
+    const BYTE* const match,
+    limitedOutput_directive limit,
+    BYTE* oend)
+{
+    size_t length;
+    BYTE* const token = (*op)++;
+
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG >= 6)
+    static const BYTE* start = NULL;
+    static U32 totalCost = 0;
+    U32 const pos = (start==NULL) ? 0 : (U32)(*anchor - start);
+    U32 const ll = (U32)(*ip - *anchor);
+    U32 const llAdd = (ll>=15) ? ((ll-15) / 255) + 1 : 0;
+    U32 const mlAdd = (matchLength>=19) ? ((matchLength-19) / 255) + 1 : 0;
+    U32 const cost = 1 + llAdd + ll + 2 + mlAdd;
+    if (start==NULL) start = *anchor;  /* only works for single segment */
+    /* g_debuglog_enable = (pos >= 2228) & (pos <= 2262); */
+    DEBUGLOG(6, "pos:%7u -- literals:%3u, match:%4i, offset:%5u, cost:%3u + %u",
+                pos,
+                (U32)(*ip - *anchor), matchLength, (U32)(*ip-match),
+                cost, totalCost);
+    totalCost += cost;
+#endif
+
+    /* Encode Literal length */
+    length = (size_t)(*ip - *anchor);
+    if ((limit) && ((*op + (length / 255) + length + (2 + 1 + LASTLITERALS)) > oend)) return 1;   /* Check output limit */
+    if (length >= RUN_MASK) {
+        size_t len = length - RUN_MASK;
+        *token = (RUN_MASK << ML_BITS);
+        for(; len >= 255 ; len -= 255) *(*op)++ = 255;
+        *(*op)++ = (BYTE)len;
+    } else {
+        *token = (BYTE)(length << ML_BITS);
+    }
+
+    /* Copy Literals */
+    LZ4_wildCopy8(*op, *anchor, (*op) + length);
+    *op += length;
+
+    /* Encode Offset */
+    assert( (*ip - match) <= LZ4_DISTANCE_MAX );   /* note : consider providing offset as a value, rather than as a pointer difference */
+    LZ4_writeLE16(*op, (U16)(*ip-match)); *op += 2;
+
+    /* Encode MatchLength */
+    assert(matchLength >= MINMATCH);
+    length = (size_t)matchLength - MINMATCH;
+    if ((limit) && (*op + (length / 255) + (1 + LASTLITERALS) > oend)) return 1;   /* Check output limit */
+    if (length >= ML_MASK) {
+        *token += ML_MASK;
+        length -= ML_MASK;
+        for(; length >= 510 ; length -= 510) { *(*op)++ = 255; *(*op)++ = 255; }
+        if (length >= 255) { length -= 255; *(*op)++ = 255; }
+        *(*op)++ = (BYTE)length;
+    } else {
+        *token += (BYTE)(length);
+    }
+
+    /* Prepare next loop */
+    *ip += matchLength;
+    *anchor = *ip;
+
+    return 0;
+}
+
+LZ4_FORCE_INLINE int LZ4HC_compress_hashChain (
+    LZ4HC_CCtx_internal* const ctx,
+    const char* const source,
+    char* const dest,
+    int* srcSizePtr,
+    int const maxOutputSize,
+    unsigned maxNbAttempts,
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
+    )
+{
+    const int inputSize = *srcSizePtr;
+    const int patternAnalysis = (maxNbAttempts > 128);   /* levels 9+ */
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + inputSize;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = (iend - LASTLITERALS);
+
+    BYTE* optr = (BYTE*) dest;
+    BYTE* op = (BYTE*) dest;
+    BYTE* oend = op + maxOutputSize;
+
+    int   ml0, ml, ml2, ml3;
+    const BYTE* start0;
+    const BYTE* ref0;
+    const BYTE* ref = NULL;
+    const BYTE* start2 = NULL;
+    const BYTE* ref2 = NULL;
+    const BYTE* start3 = NULL;
+    const BYTE* ref3 = NULL;
+
+    /* init */
+    *srcSizePtr = 0;
+    if (limit == fillOutput) oend -= LASTLITERALS;                  /* Hack for support LZ4 format restriction */
+    if (inputSize < LZ4_minLength) goto _last_literals;                  /* Input too small, no compression (all literals) */
+
+    /* Main Loop */
+    while (ip <= mflimit) {
+        ml = LZ4HC_InsertAndFindBestMatch(ctx, ip, matchlimit, &ref, maxNbAttempts, patternAnalysis, dict);
+        if (ml<MINMATCH) { ip++; continue; }
+
+        /* saved, in case we would skip too much */
+        start0 = ip; ref0 = ref; ml0 = ml;
+
+_Search2:
+        if (ip+ml <= mflimit) {
+            ml2 = LZ4HC_InsertAndGetWiderMatch(ctx,
+                            ip + ml - 2, ip + 0, matchlimit, ml, &ref2, &start2,
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+        } else {
+            ml2 = ml;
+        }
+
+        if (ml2 == ml) { /* No better match => encode ML1 */
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
+            continue;
+        }
+
+        if (start0 < ip) {   /* first match was skipped at least once */
+            if (start2 < ip + ml0) {  /* squeezing ML1 between ML0(original ML1) and ML2 */
+                ip = start0; ref = ref0; ml = ml0;  /* restore initial ML1 */
+        }   }
+
+        /* Here, start0==ip */
+        if ((start2 - ip) < 3) {  /* First Match too small : removed */
+            ml = ml2;
+            ip = start2;
+            ref =ref2;
+            goto _Search2;
+        }
+
+_Search3:
+        /* At this stage, we have :
+        *  ml2 > ml1, and
+        *  ip1+3 <= ip2 (usually < ip1+ml1) */
+        if ((start2 - ip) < OPTIMAL_ML) {
+            int correction;
+            int new_ml = ml;
+            if (new_ml > OPTIMAL_ML) new_ml = OPTIMAL_ML;
+            if (ip+new_ml > start2 + ml2 - MINMATCH) new_ml = (int)(start2 - ip) + ml2 - MINMATCH;
+            correction = new_ml - (int)(start2 - ip);
+            if (correction > 0) {
+                start2 += correction;
+                ref2 += correction;
+                ml2 -= correction;
+            }
+        }
+        /* Now, we have start2 = ip+new_ml, with new_ml = min(ml, OPTIMAL_ML=18) */
+
+        if (start2 + ml2 <= mflimit) {
+            ml3 = LZ4HC_InsertAndGetWiderMatch(ctx,
+                            start2 + ml2 - 3, start2, matchlimit, ml2, &ref3, &start3,
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+        } else {
+            ml3 = ml2;
+        }
+
+        if (ml3 == ml2) {  /* No better match => encode ML1 and ML2 */
+            /* ip & ref are known; Now for ml */
+            if (start2 < ip+ml)  ml = (int)(start2 - ip);
+            /* Now, encode 2 sequences */
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
+            ip = start2;
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml2, ref2, limit, oend)) goto _dest_overflow;
+            continue;
+        }
+
+        if (start3 < ip+ml+3) {  /* Not enough space for match 2 : remove it */
+            if (start3 >= (ip+ml)) {  /* can write Seq1 immediately ==> Seq2 is removed, so Seq3 becomes Seq1 */
+                if (start2 < ip+ml) {
+                    int correction = (int)(ip+ml - start2);
+                    start2 += correction;
+                    ref2 += correction;
+                    ml2 -= correction;
+                    if (ml2 < MINMATCH) {
+                        start2 = start3;
+                        ref2 = ref3;
+                        ml2 = ml3;
+                    }
+                }
+
+                optr = op;
+                if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
+                ip  = start3;
+                ref = ref3;
+                ml  = ml3;
+
+                start0 = start2;
+                ref0 = ref2;
+                ml0 = ml2;
+                goto _Search2;
+            }
+
+            start2 = start3;
+            ref2 = ref3;
+            ml2 = ml3;
+            goto _Search3;
+        }
+
+        /*
+        * OK, now we have 3 ascending matches;
+        * let's write the first one ML1.
+        * ip & ref are known; Now decide ml.
+        */
+        if (start2 < ip+ml) {
+            if ((start2 - ip) < OPTIMAL_ML) {
+                int correction;
+                if (ml > OPTIMAL_ML) ml = OPTIMAL_ML;
+                if (ip + ml > start2 + ml2 - MINMATCH) ml = (int)(start2 - ip) + ml2 - MINMATCH;
+                correction = ml - (int)(start2 - ip);
+                if (correction > 0) {
+                    start2 += correction;
+                    ref2 += correction;
+                    ml2 -= correction;
+                }
+            } else {
+                ml = (int)(start2 - ip);
+            }
+        }
+        optr = op;
+        if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
+
+        /* ML2 becomes ML1 */
+        ip = start2; ref = ref2; ml = ml2;
+
+        /* ML3 becomes ML2 */
+        start2 = start3; ref2 = ref3; ml2 = ml3;
+
+        /* let's find a new ML3 */
+        goto _Search3;
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+        size_t litLength = (lastRunSize + 255 - RUN_MASK) / 255;
+        size_t const totalSize = 1 + litLength + lastRunSize;
+        if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+        if (limit && (op + totalSize > oend)) {
+            if (limit == limitedOutput) return 0;  /* Check output limit */
+            /* adapt lastRunSize to fill 'dest' */
+            lastRunSize  = (size_t)(oend - op) - 1;
+            litLength = (lastRunSize + 255 - RUN_MASK) / 255;
+            lastRunSize -= litLength;
+        }
+        ip = anchor + lastRunSize;
+
+        if (lastRunSize >= RUN_MASK) {
+            size_t accumulator = lastRunSize - RUN_MASK;
+            *op++ = (RUN_MASK << ML_BITS);
+            for(; accumulator >= 255 ; accumulator -= 255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRunSize << ML_BITS);
+        }
+        memcpy(op, anchor, lastRunSize);
+        op += lastRunSize;
+    }
+
+    /* End */
+    *srcSizePtr = (int) (((const char*)ip) - source);
+    return (int) (((char*)op)-dest);
+
+_dest_overflow:
+    if (limit == fillOutput) {
+        op = optr;  /* restore correct out pointer */
+        goto _last_literals;
+    }
+    return 0;
+}
+
+
+static int LZ4HC_compress_optimal( LZ4HC_CCtx_internal* ctx,
+    const char* const source, char* dst,
+    int* srcSizePtr, int dstCapacity,
+    int const nbSearches, size_t sufficient_len,
+    const limitedOutput_directive limit, int const fullUpdate,
+    const dictCtx_directive dict,
+    HCfavor_e favorDecSpeed);
+
+
+LZ4_FORCE_INLINE int LZ4HC_compress_generic_internal (
+    LZ4HC_CCtx_internal* const ctx,
+    const char* const src,
+    char* const dst,
+    int* const srcSizePtr,
+    int const dstCapacity,
+    int cLevel,
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
+    )
+{
+    typedef enum { lz4hc, lz4opt } lz4hc_strat_e;
+    typedef struct {
+        lz4hc_strat_e strat;
+        U32 nbSearches;
+        U32 targetLength;
+    } cParams_t;
+    static const cParams_t clTable[LZ4HC_CLEVEL_MAX+1] = {
+        { lz4hc,     2, 16 },  /* 0, unused */
+        { lz4hc,     2, 16 },  /* 1, unused */
+        { lz4hc,     2, 16 },  /* 2, unused */
+        { lz4hc,     4, 16 },  /* 3 */
+        { lz4hc,     8, 16 },  /* 4 */
+        { lz4hc,    16, 16 },  /* 5 */
+        { lz4hc,    32, 16 },  /* 6 */
+        { lz4hc,    64, 16 },  /* 7 */
+        { lz4hc,   128, 16 },  /* 8 */
+        { lz4hc,   256, 16 },  /* 9 */
+        { lz4opt,   96, 64 },  /*10==LZ4HC_CLEVEL_OPT_MIN*/
+        { lz4opt,  512,128 },  /*11 */
+        { lz4opt,16384,LZ4_OPT_NUM },  /* 12==LZ4HC_CLEVEL_MAX */
+    };
+
+    DEBUGLOG(4, "LZ4HC_compress_generic(ctx=%p, src=%p, srcSize=%d)", ctx, src, *srcSizePtr);
+
+    if (limit == fillOutput && dstCapacity < 1) return 0;   /* Impossible to store anything */
+    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;    /* Unsupported input size (too large or negative) */
+
+    ctx->end += *srcSizePtr;
+    if (cLevel < 1) cLevel = LZ4HC_CLEVEL_DEFAULT;   /* note : convention is different from lz4frame, maybe something to review */
+    cLevel = MIN(LZ4HC_CLEVEL_MAX, cLevel);
+    {   cParams_t const cParam = clTable[cLevel];
+        HCfavor_e const favor = ctx->favorDecSpeed ? favorDecompressionSpeed : favorCompressionRatio;
+        int result;
+
+        if (cParam.strat == lz4hc) {
+            result = LZ4HC_compress_hashChain(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                cParam.nbSearches, limit, dict);
+        } else {
+            assert(cParam.strat == lz4opt);
+            result = LZ4HC_compress_optimal(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                (int)cParam.nbSearches, cParam.targetLength, limit,
+                                cLevel == LZ4HC_CLEVEL_MAX,   /* ultra mode */
+                                dict, favor);
+        }
+        if (result <= 0) ctx->dirty = 1;
+        return result;
+    }
+}
+
+static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock);
+
+static int
+LZ4HC_compress_generic_noDictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    assert(ctx->dictCtx == NULL);
+    return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, noDictCtx);
+}
+
+static int
+LZ4HC_compress_generic_dictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    const size_t position = (size_t)(ctx->end - ctx->base) - ctx->lowLimit;
+    assert(ctx->dictCtx != NULL);
+    if (position >= 64 KB) {
+        ctx->dictCtx = NULL;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else if (position == 0 && *srcSizePtr > 4 KB) {
+        memcpy(ctx, ctx->dictCtx, sizeof(LZ4HC_CCtx_internal));
+        LZ4HC_setExternalDict(ctx, (const BYTE *)src);
+        ctx->compressionLevel = (short)cLevel;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, usingDictCtxHc);
+    }
+}
+
+static int
+LZ4HC_compress_generic (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    if (ctx->dictCtx == NULL) {
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_dictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    }
+}
+
+
+int LZ4_sizeofStateHC(void) { return (int)sizeof(LZ4_streamHC_t); }
+
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                   * it reports an aligment of 8-bytes,
+                   * while actually aligning LZ4_streamHC_t on 4 bytes. */
+static size_t LZ4_streamHC_t_alignment(void)
+{
+    struct { char c; LZ4_streamHC_t t; } t_a;
+    return sizeof(t_a) - sizeof(t_a.t);
+}
+#endif
+
+/* state is presumed correctly initialized,
+ * in which case its size and alignment have already been validate */
+int LZ4_compress_HC_extStateHC_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    LZ4HC_CCtx_internal* const ctx = &((LZ4_streamHC_t*)state)->internal_donotuse;
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                   * it reports an aligment of 8-bytes,
+                   * while actually aligning LZ4_streamHC_t on 4 bytes. */
+    assert(((size_t)state & (LZ4_streamHC_t_alignment() - 1)) == 0);  /* check alignment */
+#endif
+    if (((size_t)(state)&(sizeof(void*)-1)) != 0) return 0;   /* Error : state is not aligned for pointers (32 or 64 bits) */
+    LZ4_resetStreamHC_fast((LZ4_streamHC_t*)state, compressionLevel);
+    LZ4HC_init_internal (ctx, (const BYTE*)src);
+    if (dstCapacity < LZ4_compressBound(srcSize))
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, limitedOutput);
+    else
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, notLimited);
+}
+
+int LZ4_compress_HC_extStateHC (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    return LZ4_compress_HC_extStateHC_fastReset(state, src, dst, srcSize, dstCapacity, compressionLevel);
+}
+
+int LZ4_compress_HC(const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    LZ4_streamHC_t* const statePtr = (LZ4_streamHC_t*)ALLOC(sizeof(LZ4_streamHC_t));
+#else
+    LZ4_streamHC_t state;
+    LZ4_streamHC_t* const statePtr = &state;
+#endif
+    int const cSize = LZ4_compress_HC_extStateHC(statePtr, src, dst, srcSize, dstCapacity, compressionLevel);
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    FREEMEM(statePtr);
+#endif
+    return cSize;
+}
+
+/* state is presumed sized correctly (>= sizeof(LZ4_streamHC_t)) */
+int LZ4_compress_HC_destSize(void* state, const char* source, char* dest, int* sourceSizePtr, int targetDestSize, int cLevel)
+{
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    LZ4HC_init_internal(&ctx->internal_donotuse, (const BYTE*) source);
+    LZ4_setCompressionLevel(ctx, cLevel);
+    return LZ4HC_compress_generic(&ctx->internal_donotuse, source, dest, sourceSizePtr, targetDestSize, cLevel, fillOutput);
+}
+
+
+
+/**************************************
+*  Streaming Functions
+**************************************/
+/* allocation */
+LZ4_streamHC_t* LZ4_createStreamHC(void)
+{
+    LZ4_streamHC_t* const LZ4_streamHCPtr = (LZ4_streamHC_t*)ALLOC(sizeof(LZ4_streamHC_t));
+    if (LZ4_streamHCPtr==NULL) return NULL;
+    LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));  /* full initialization, malloc'ed buffer can be full of garbage */
+    return LZ4_streamHCPtr;
+}
+
+int LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr)
+{
+    DEBUGLOG(4, "LZ4_freeStreamHC(%p)", LZ4_streamHCPtr);
+    if (!LZ4_streamHCPtr) return 0;  /* support free on NULL */
+    FREEMEM(LZ4_streamHCPtr);
+    return 0;
+}
+
+
+LZ4_streamHC_t* LZ4_initStreamHC (void* buffer, size_t size)
+{
+    LZ4_streamHC_t* const LZ4_streamHCPtr = (LZ4_streamHC_t*)buffer;
+    if (buffer == NULL) return NULL;
+    if (size < sizeof(LZ4_streamHC_t)) return NULL;
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                   * it reports an aligment of 8-bytes,
+                   * while actually aligning LZ4_streamHC_t on 4 bytes. */
+    if (((size_t)buffer) & (LZ4_streamHC_t_alignment() - 1)) return NULL;  /* alignment check */
+#endif
+    /* if compilation fails here, LZ4_STREAMHCSIZE must be increased */
+    LZ4_STATIC_ASSERT(sizeof(LZ4HC_CCtx_internal) <= LZ4_STREAMHCSIZE);
+    DEBUGLOG(4, "LZ4_initStreamHC(%p, %u)", LZ4_streamHCPtr, (unsigned)size);
+    /* end-base will trigger a clearTable on starting compression */
+    LZ4_streamHCPtr->internal_donotuse.end = (const BYTE *)(ptrdiff_t)-1;
+    LZ4_streamHCPtr->internal_donotuse.base = NULL;
+    LZ4_streamHCPtr->internal_donotuse.dictCtx = NULL;
+    LZ4_streamHCPtr->internal_donotuse.favorDecSpeed = 0;
+    LZ4_streamHCPtr->internal_donotuse.dirty = 0;
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, LZ4HC_CLEVEL_DEFAULT);
+    return LZ4_streamHCPtr;
+}
+
+/* just a stub */
+void LZ4_resetStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
+}
+
+void LZ4_resetStreamHC_fast (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    DEBUGLOG(4, "LZ4_resetStreamHC_fast(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (LZ4_streamHCPtr->internal_donotuse.dirty) {
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    } else {
+        /* preserve end - base : can trigger clearTable's threshold */
+        LZ4_streamHCPtr->internal_donotuse.end -= (uptrval)LZ4_streamHCPtr->internal_donotuse.base;
+        LZ4_streamHCPtr->internal_donotuse.base = NULL;
+        LZ4_streamHCPtr->internal_donotuse.dictCtx = NULL;
+    }
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
+}
+
+void LZ4_setCompressionLevel(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    DEBUGLOG(5, "LZ4_setCompressionLevel(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (compressionLevel < 1) compressionLevel = LZ4HC_CLEVEL_DEFAULT;
+    if (compressionLevel > LZ4HC_CLEVEL_MAX) compressionLevel = LZ4HC_CLEVEL_MAX;
+    LZ4_streamHCPtr->internal_donotuse.compressionLevel = (short)compressionLevel;
+}
+
+void LZ4_favorDecompressionSpeed(LZ4_streamHC_t* LZ4_streamHCPtr, int favor)
+{
+    LZ4_streamHCPtr->internal_donotuse.favorDecSpeed = (favor!=0);
+}
+
+/* LZ4_loadDictHC() :
+ * LZ4_streamHCPtr is presumed properly initialized */
+int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr,
+              const char* dictionary, int dictSize)
+{
+    LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(4, "LZ4_loadDictHC(ctx:%p, dict:%p, dictSize:%d)", LZ4_streamHCPtr, dictionary, dictSize);
+    assert(LZ4_streamHCPtr != NULL);
+    if (dictSize > 64 KB) {
+        dictionary += (size_t)dictSize - 64 KB;
+        dictSize = 64 KB;
+    }
+    /* need a full initialization, there are bad side-effects when using resetFast() */
+    {   int const cLevel = ctxPtr->compressionLevel;
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+        LZ4_setCompressionLevel(LZ4_streamHCPtr, cLevel);
+    }
+    LZ4HC_init_internal (ctxPtr, (const BYTE*)dictionary);
+    ctxPtr->end = (const BYTE*)dictionary + dictSize;
+    if (dictSize >= 4) LZ4HC_Insert (ctxPtr, ctxPtr->end-3);
+    return dictSize;
+}
+
+void LZ4_attach_HC_dictionary(LZ4_streamHC_t *working_stream, const LZ4_streamHC_t *dictionary_stream) {
+    working_stream->internal_donotuse.dictCtx = dictionary_stream != NULL ? &(dictionary_stream->internal_donotuse) : NULL;
+}
+
+/* compression */
+
+static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock)
+{
+    DEBUGLOG(4, "LZ4HC_setExternalDict(%p, %p)", ctxPtr, newBlock);
+    if (ctxPtr->end >= ctxPtr->base + ctxPtr->dictLimit + 4)
+        LZ4HC_Insert (ctxPtr, ctxPtr->end-3);   /* Referencing remaining dictionary content */
+
+    /* Only one memory segment for extDict, so any previous extDict is lost at this stage */
+    ctxPtr->lowLimit  = ctxPtr->dictLimit;
+    ctxPtr->dictLimit = (U32)(ctxPtr->end - ctxPtr->base);
+    ctxPtr->dictBase  = ctxPtr->base;
+    ctxPtr->base = newBlock - ctxPtr->dictLimit;
+    ctxPtr->end  = newBlock;
+    ctxPtr->nextToUpdate = ctxPtr->dictLimit;   /* match referencing will resume from there */
+
+    /* cannot reference an extDict and a dictCtx at the same time */
+    ctxPtr->dictCtx = NULL;
+}
+
+static int LZ4_compressHC_continue_generic (LZ4_streamHC_t* LZ4_streamHCPtr,
+                                            const char* src, char* dst,
+                                            int* srcSizePtr, int dstCapacity,
+                                            limitedOutput_directive limit)
+{
+    LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(4, "LZ4_compressHC_continue_generic(ctx=%p, src=%p, srcSize=%d)",
+                LZ4_streamHCPtr, src, *srcSizePtr);
+    assert(ctxPtr != NULL);
+    /* auto-init if forgotten */
+    if (ctxPtr->base == NULL) LZ4HC_init_internal (ctxPtr, (const BYTE*) src);
+
+    /* Check overflow */
+    if ((size_t)(ctxPtr->end - ctxPtr->base) > 2 GB) {
+        size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->base) - ctxPtr->dictLimit;
+        if (dictSize > 64 KB) dictSize = 64 KB;
+        LZ4_loadDictHC(LZ4_streamHCPtr, (const char*)(ctxPtr->end) - dictSize, (int)dictSize);
+    }
+
+    /* Check if blocks follow each other */
+    if ((const BYTE*)src != ctxPtr->end)
+        LZ4HC_setExternalDict(ctxPtr, (const BYTE*)src);
+
+    /* Check overlapping input/dictionary space */
+    {   const BYTE* sourceEnd = (const BYTE*) src + *srcSizePtr;
+        const BYTE* const dictBegin = ctxPtr->dictBase + ctxPtr->lowLimit;
+        const BYTE* const dictEnd   = ctxPtr->dictBase + ctxPtr->dictLimit;
+        if ((sourceEnd > dictBegin) && ((const BYTE*)src < dictEnd)) {
+            if (sourceEnd > dictEnd) sourceEnd = dictEnd;
+            ctxPtr->lowLimit = (U32)(sourceEnd - ctxPtr->dictBase);
+            if (ctxPtr->dictLimit - ctxPtr->lowLimit < 4) ctxPtr->lowLimit = ctxPtr->dictLimit;
+        }
+    }
+
+    return LZ4HC_compress_generic (ctxPtr, src, dst, srcSizePtr, dstCapacity, ctxPtr->compressionLevel, limit);
+}
+
+int LZ4_compress_HC_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    if (dstCapacity < LZ4_compressBound(srcSize))
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, limitedOutput);
+    else
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, notLimited);
+}
+
+int LZ4_compress_HC_continue_destSize (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int* srcSizePtr, int targetDestSize)
+{
+    return LZ4_compressHC_continue_generic(LZ4_streamHCPtr, src, dst, srcSizePtr, targetDestSize, fillOutput);
+}
+
+
+
+/* dictionary saving */
+
+int LZ4_saveDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, char* safeBuffer, int dictSize)
+{
+    LZ4HC_CCtx_internal* const streamPtr = &LZ4_streamHCPtr->internal_donotuse;
+    int const prefixSize = (int)(streamPtr->end - (streamPtr->base + streamPtr->dictLimit));
+    DEBUGLOG(4, "LZ4_saveDictHC(%p, %p, %d)", LZ4_streamHCPtr, safeBuffer, dictSize);
+    if (dictSize > 64 KB) dictSize = 64 KB;
+    if (dictSize < 4) dictSize = 0;
+    if (dictSize > prefixSize) dictSize = prefixSize;
+    memmove(safeBuffer, streamPtr->end - dictSize, dictSize);
+    {   U32 const endIndex = (U32)(streamPtr->end - streamPtr->base);
+        streamPtr->end = (const BYTE*)safeBuffer + dictSize;
+        streamPtr->base = streamPtr->end - endIndex;
+        streamPtr->dictLimit = endIndex - (U32)dictSize;
+        streamPtr->lowLimit = endIndex - (U32)dictSize;
+        if (streamPtr->nextToUpdate < streamPtr->dictLimit) streamPtr->nextToUpdate = streamPtr->dictLimit;
+    }
+    return dictSize;
+}
+
+
+/***************************************************
+*  Deprecated Functions
+***************************************************/
+
+/* These functions currently generate deprecation warnings */
+
+/* Wrappers for deprecated compression functions */
+int LZ4_compressHC(const char* src, char* dst, int srcSize) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
+int LZ4_compressHC_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, 0); }
+int LZ4_compressHC2(const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
+int LZ4_compressHC2_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, cLevel); }
+int LZ4_compressHC_withStateHC (void* state, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
+int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, maxDstSize, 0); }
+int LZ4_compressHC2_withStateHC (void* state, const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
+int LZ4_compressHC2_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, maxDstSize, cLevel); }
+int LZ4_compressHC_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, LZ4_compressBound(srcSize)); }
+int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, maxDstSize); }
+
+
+/* Deprecated streaming functions */
+int LZ4_sizeofStreamStateHC(void) { return LZ4_STREAMHCSIZE; }
+
+/* state is presumed correctly sized, aka >= sizeof(LZ4_streamHC_t)
+ * @return : 0 on success, !=0 if error */
+int LZ4_resetStreamStateHC(void* state, char* inputBuffer)
+{
+    LZ4_streamHC_t* const hc4 = LZ4_initStreamHC(state, sizeof(*hc4));
+    if (hc4 == NULL) return 1;   /* init failed */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
+    return 0;
+}
+
+void* LZ4_createHC (const char* inputBuffer)
+{
+    LZ4_streamHC_t* const hc4 = LZ4_createStreamHC();
+    if (hc4 == NULL) return NULL;   /* not enough memory */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
+    return hc4;
+}
+
+int LZ4_freeHC (void* LZ4HC_Data)
+{
+    if (!LZ4HC_Data) return 0;  /* support free on NULL */
+    FREEMEM(LZ4HC_Data);
+    return 0;
+}
+
+int LZ4_compressHC2_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int cLevel)
+{
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, 0, cLevel, notLimited);
+}
+
+int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int dstCapacity, int cLevel)
+{
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, dstCapacity, cLevel, limitedOutput);
+}
+
+char* LZ4_slideInputBufferHC(void* LZ4HC_Data)
+{
+    LZ4_streamHC_t *ctx = (LZ4_streamHC_t*)LZ4HC_Data;
+    const BYTE *bufferStart = ctx->internal_donotuse.base + ctx->internal_donotuse.lowLimit;
+    LZ4_resetStreamHC_fast(ctx, ctx->internal_donotuse.compressionLevel);
+    /* avoid const char * -> char * conversion warning :( */
+    return (char *)(uptrval)bufferStart;
+}
+
+
+/* ================================================
+ *  LZ4 Optimal parser (levels [LZ4HC_CLEVEL_OPT_MIN - LZ4HC_CLEVEL_MAX])
+ * ===============================================*/
+typedef struct {
+    int price;
+    int off;
+    int mlen;
+    int litlen;
+} LZ4HC_optimal_t;
+
+/* price in bytes */
+LZ4_FORCE_INLINE int LZ4HC_literalsPrice(int const litlen)
+{
+    int price = litlen;
+    assert(litlen >= 0);
+    if (litlen >= (int)RUN_MASK)
+        price += 1 + ((litlen-(int)RUN_MASK) / 255);
+    return price;
+}
+
+
+/* requires mlen >= MINMATCH */
+LZ4_FORCE_INLINE int LZ4HC_sequencePrice(int litlen, int mlen)
+{
+    int price = 1 + 2 ; /* token + 16-bit offset */
+    assert(litlen >= 0);
+    assert(mlen >= MINMATCH);
+
+    price += LZ4HC_literalsPrice(litlen);
+
+    if (mlen >= (int)(ML_MASK+MINMATCH))
+        price += 1 + ((mlen-(int)(ML_MASK+MINMATCH)) / 255);
+
+    return price;
+}
+
+
+typedef struct {
+    int off;
+    int len;
+} LZ4HC_match_t;
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_FindLongerMatch(LZ4HC_CCtx_internal* const ctx,
+                      const BYTE* ip, const BYTE* const iHighLimit,
+                      int minLen, int nbSearches,
+                      const dictCtx_directive dict,
+                      const HCfavor_e favorDecSpeed)
+{
+    LZ4HC_match_t match = { 0 , 0 };
+    const BYTE* matchPtr = NULL;
+    /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
+     * but this won't be the case here, as we define iLowLimit==ip,
+     * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
+    int matchLength = LZ4HC_InsertAndGetWiderMatch(ctx, ip, ip, iHighLimit, minLen, &matchPtr, &ip, nbSearches, 1 /*patternAnalysis*/, 1 /*chainSwap*/, dict, favorDecSpeed);
+    if (matchLength <= minLen) return match;
+    if (favorDecSpeed) {
+        if ((matchLength>18) & (matchLength<=36)) matchLength=18;   /* favor shortcut */
+    }
+    match.len = matchLength;
+    match.off = (int)(ip-matchPtr);
+    return match;
+}
+
+
+static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
+                                    const char* const source,
+                                    char* dst,
+                                    int* srcSizePtr,
+                                    int dstCapacity,
+                                    int const nbSearches,
+                                    size_t sufficient_len,
+                                    const limitedOutput_directive limit,
+                                    int const fullUpdate,
+                                    const dictCtx_directive dict,
+                                    const HCfavor_e favorDecSpeed)
+{
+    int retval = 0;
+#define TRAILING_LITERALS 3
+#ifdef LZ4HC_HEAPMODE
+    LZ4HC_optimal_t* const opt = (LZ4HC_optimal_t*)malloc(sizeof(LZ4HC_optimal_t) * (LZ4_OPT_NUM + TRAILING_LITERALS));
+#else
+    LZ4HC_optimal_t opt[LZ4_OPT_NUM + TRAILING_LITERALS];   /* ~64 KB, which is a bit large for stack... */
+#endif
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + *srcSizePtr;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+    BYTE* op = (BYTE*) dst;
+    BYTE* opSaved = (BYTE*) dst;
+    BYTE* oend = op + dstCapacity;
+
+    /* init */
+#ifdef LZ4HC_HEAPMODE
+    if (opt == NULL) goto _return_label;
+#endif
+    DEBUGLOG(5, "LZ4HC_compress_optimal(dst=%p, dstCapa=%u)", dst, (unsigned)dstCapacity);
+    *srcSizePtr = 0;
+    if (limit == fillOutput) oend -= LASTLITERALS;   /* Hack for support LZ4 format restriction */
+    if (sufficient_len >= LZ4_OPT_NUM) sufficient_len = LZ4_OPT_NUM-1;
+
+    /* Main Loop */
+    assert(ip - anchor < LZ4_MAX_INPUT_SIZE);
+    while (ip <= mflimit) {
+         int const llen = (int)(ip - anchor);
+         int best_mlen, best_off;
+         int cur, last_match_pos = 0;
+
+         LZ4HC_match_t const firstMatch = LZ4HC_FindLongerMatch(ctx, ip, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+         if (firstMatch.len==0) { ip++; continue; }
+
+         if ((size_t)firstMatch.len > sufficient_len) {
+             /* good enough solution : immediate encoding */
+             int const firstML = firstMatch.len;
+             const BYTE* const matchPos = ip - firstMatch.off;
+             opSaved = op;
+             if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), firstML, matchPos, limit, oend) )   /* updates ip, op and anchor */
+                 goto _dest_overflow;
+             continue;
+         }
+
+         /* set prices for first positions (literals) */
+         {   int rPos;
+             for (rPos = 0 ; rPos < MINMATCH ; rPos++) {
+                 int const cost = LZ4HC_literalsPrice(llen + rPos);
+                 opt[rPos].mlen = 1;
+                 opt[rPos].off = 0;
+                 opt[rPos].litlen = llen + rPos;
+                 opt[rPos].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             rPos, cost, opt[rPos].litlen);
+         }   }
+         /* set prices using initial match */
+         {   int mlen = MINMATCH;
+             int const matchML = firstMatch.len;   /* necessarily < sufficient_len < LZ4_OPT_NUM */
+             int const offset = firstMatch.off;
+             assert(matchML < LZ4_OPT_NUM);
+             for ( ; mlen <= matchML ; mlen++) {
+                 int const cost = LZ4HC_sequencePrice(llen, mlen);
+                 opt[mlen].mlen = mlen;
+                 opt[mlen].off = offset;
+                 opt[mlen].litlen = llen;
+                 opt[mlen].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i) -- initial setup",
+                             mlen, cost, mlen);
+         }   }
+         last_match_pos = firstMatch.len;
+         {   int addLit;
+             for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                 opt[last_match_pos+addLit].mlen = 1; /* literal */
+                 opt[last_match_pos+addLit].off = 0;
+                 opt[last_match_pos+addLit].litlen = addLit;
+                 opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+         }   }
+
+         /* check further positions */
+         for (cur = 1; cur < last_match_pos; cur++) {
+             const BYTE* const curPtr = ip + cur;
+             LZ4HC_match_t newMatch;
+
+             if (curPtr > mflimit) break;
+             DEBUGLOG(7, "rPos:%u[%u] vs [%u]%u",
+                     cur, opt[cur].price, opt[cur+1].price, cur+1);
+             if (fullUpdate) {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if ( (opt[cur+1].price <= opt[cur].price)
+                   /* in some cases, next position has same cost, but cost rises sharply after, so a small match would still be beneficial */
+                   && (opt[cur+MINMATCH].price < opt[cur].price + 3/*min seq price*/) )
+                     continue;
+             } else {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if (opt[cur+1].price <= opt[cur].price) continue;
+             }
+
+             DEBUGLOG(7, "search at rPos:%u", cur);
+             if (fullUpdate)
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+             else
+                 /* only test matches of minimum length; slightly faster, but misses a few bytes */
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, last_match_pos - cur, nbSearches, dict, favorDecSpeed);
+             if (!newMatch.len) continue;
+
+             if ( ((size_t)newMatch.len > sufficient_len)
+               || (newMatch.len + cur >= LZ4_OPT_NUM) ) {
+                 /* immediate encoding */
+                 best_mlen = newMatch.len;
+                 best_off = newMatch.off;
+                 last_match_pos = cur + 1;
+                 goto encode;
+             }
+
+             /* before match : set price with literals at beginning */
+             {   int const baseLitlen = opt[cur].litlen;
+                 int litlen;
+                 for (litlen = 1; litlen < MINMATCH; litlen++) {
+                     int const price = opt[cur].price - LZ4HC_literalsPrice(baseLitlen) + LZ4HC_literalsPrice(baseLitlen+litlen);
+                     int const pos = cur + litlen;
+                     if (price < opt[pos].price) {
+                         opt[pos].mlen = 1; /* literal */
+                         opt[pos].off = 0;
+                         opt[pos].litlen = baseLitlen+litlen;
+                         opt[pos].price = price;
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)",
+                                     pos, price, opt[pos].litlen);
+             }   }   }
+
+             /* set prices using match at position = cur */
+             {   int const matchML = newMatch.len;
+                 int ml = MINMATCH;
+
+                 assert(cur + newMatch.len < LZ4_OPT_NUM);
+                 for ( ; ml <= matchML ; ml++) {
+                     int const pos = cur + ml;
+                     int const offset = newMatch.off;
+                     int price;
+                     int ll;
+                     DEBUGLOG(7, "testing price rPos %i (last_match_pos=%i)",
+                                 pos, last_match_pos);
+                     if (opt[cur].mlen == 1) {
+                         ll = opt[cur].litlen;
+                         price = ((cur > ll) ? opt[cur - ll].price : 0)
+                               + LZ4HC_sequencePrice(ll, ml);
+                     } else {
+                         ll = 0;
+                         price = opt[cur].price + LZ4HC_sequencePrice(0, ml);
+                     }
+
+                    assert((U32)favorDecSpeed <= 1);
+                     if (pos > last_match_pos+TRAILING_LITERALS
+                      || price <= opt[pos].price - (int)favorDecSpeed) {
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i)",
+                                     pos, price, ml);
+                         assert(pos < LZ4_OPT_NUM);
+                         if ( (ml == matchML)  /* last pos of last match */
+                           && (last_match_pos < pos) )
+                             last_match_pos = pos;
+                         opt[pos].mlen = ml;
+                         opt[pos].off = offset;
+                         opt[pos].litlen = ll;
+                         opt[pos].price = price;
+             }   }   }
+             /* complete following positions with literals */
+             {   int addLit;
+                 for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                     opt[last_match_pos+addLit].mlen = 1; /* literal */
+                     opt[last_match_pos+addLit].off = 0;
+                     opt[last_match_pos+addLit].litlen = addLit;
+                     opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                     DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)", last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+             }   }
+         }  /* for (cur = 1; cur <= last_match_pos; cur++) */
+
+         assert(last_match_pos < LZ4_OPT_NUM + TRAILING_LITERALS);
+         best_mlen = opt[last_match_pos].mlen;
+         best_off = opt[last_match_pos].off;
+         cur = last_match_pos - best_mlen;
+
+ encode: /* cur, last_match_pos, best_mlen, best_off must be set */
+         assert(cur < LZ4_OPT_NUM);
+         assert(last_match_pos >= 1);  /* == 1 when only one candidate */
+         DEBUGLOG(6, "reverse traversal, looking for shortest path (last_match_pos=%i)", last_match_pos);
+         {   int candidate_pos = cur;
+             int selected_matchLength = best_mlen;
+             int selected_offset = best_off;
+             while (1) {  /* from end to beginning */
+                 int const next_matchLength = opt[candidate_pos].mlen;  /* can be 1, means literal */
+                 int const next_offset = opt[candidate_pos].off;
+                 DEBUGLOG(7, "pos %i: sequence length %i", candidate_pos, selected_matchLength);
+                 opt[candidate_pos].mlen = selected_matchLength;
+                 opt[candidate_pos].off = selected_offset;
+                 selected_matchLength = next_matchLength;
+                 selected_offset = next_offset;
+                 if (next_matchLength > candidate_pos) break; /* last match elected, first match to encode */
+                 assert(next_matchLength > 0);  /* can be 1, means literal */
+                 candidate_pos -= next_matchLength;
+         }   }
+
+         /* encode all recorded sequences in order */
+         {   int rPos = 0;  /* relative position (to ip) */
+             while (rPos < last_match_pos) {
+                 int const ml = opt[rPos].mlen;
+                 int const offset = opt[rPos].off;
+                 if (ml == 1) { ip++; rPos++; continue; }  /* literal; note: can end up with several literals, in which case, skip them */
+                 rPos += ml;
+                 assert(ml >= MINMATCH);
+                 assert((offset >= 1) && (offset <= LZ4_DISTANCE_MAX));
+                 opSaved = op;
+                 if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ip - offset, limit, oend) )   /* updates ip, op and anchor */
+                     goto _dest_overflow;
+         }   }
+     }  /* while (ip <= mflimit) */
+
+ _last_literals:
+     /* Encode Last Literals */
+     {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+         size_t litLength = (lastRunSize + 255 - RUN_MASK) / 255;
+         size_t const totalSize = 1 + litLength + lastRunSize;
+         if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+         if (limit && (op + totalSize > oend)) {
+             if (limit == limitedOutput) { /* Check output limit */
+                retval = 0;
+                goto _return_label;
+             }
+             /* adapt lastRunSize to fill 'dst' */
+             lastRunSize  = (size_t)(oend - op) - 1;
+             litLength = (lastRunSize + 255 - RUN_MASK) / 255;
+             lastRunSize -= litLength;
+         }
+         ip = anchor + lastRunSize;
+
+         if (lastRunSize >= RUN_MASK) {
+             size_t accumulator = lastRunSize - RUN_MASK;
+             *op++ = (RUN_MASK << ML_BITS);
+             for(; accumulator >= 255 ; accumulator -= 255) *op++ = 255;
+             *op++ = (BYTE) accumulator;
+         } else {
+             *op++ = (BYTE)(lastRunSize << ML_BITS);
+         }
+         memcpy(op, anchor, lastRunSize);
+         op += lastRunSize;
+     }
+
+     /* End */
+     *srcSizePtr = (int) (((const char*)ip) - source);
+     retval = (int) ((char*)op-dst);
+     goto _return_label;
+
+ _dest_overflow:
+     if (limit == fillOutput) {
+         op = opSaved;  /* restore correct out pointer */
+         goto _last_literals;
+     }
+ _return_label:
+#ifdef LZ4HC_HEAPMODE
+     free(opt);
+#endif
+     return retval;
+ }
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4/lz4hc.h b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4hc.h
new file mode 100644
index 0000000..44e35bb
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/lz4/lz4/lz4hc.h
@@ -0,0 +1,438 @@
+/*
+   LZ4 HC - High Compression Mode of LZ4
+   Header File
+   Copyright (C) 2011-2017, Yann Collet.
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - LZ4 source repository : https://github.com/lz4/lz4
+   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+#ifndef LZ4_HC_H_19834876238432
+#define LZ4_HC_H_19834876238432
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* --- Dependency --- */
+/* note : lz4hc requires lz4.h/lz4.c for compilation */
+#include "lz4.h"   /* stddef, LZ4LIB_API, LZ4_DEPRECATED */
+
+
+/* --- Useful constants --- */
+#define LZ4HC_CLEVEL_MIN         3
+#define LZ4HC_CLEVEL_DEFAULT     9
+#define LZ4HC_CLEVEL_OPT_MIN    10
+#define LZ4HC_CLEVEL_MAX        12
+
+
+/*-************************************
+ *  Block Compression
+ **************************************/
+/*! LZ4_compress_HC() :
+ *  Compress data from `src` into `dst`, using the powerful but slower "HC" algorithm.
+ * `dst` must be already allocated.
+ *  Compression is guaranteed to succeed if `dstCapacity >= LZ4_compressBound(srcSize)` (see "lz4.h")
+ *  Max supported `srcSize` value is LZ4_MAX_INPUT_SIZE (see "lz4.h")
+ * `compressionLevel` : any value between 1 and LZ4HC_CLEVEL_MAX will work.
+ *                      Values > LZ4HC_CLEVEL_MAX behave the same as LZ4HC_CLEVEL_MAX.
+ * @return : the number of bytes written into 'dst'
+ *           or 0 if compression fails.
+ */
+LZ4LIB_API int LZ4_compress_HC (const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel);
+
+
+/* Note :
+ *   Decompression functions are provided within "lz4.h" (BSD license)
+ */
+
+
+/*! LZ4_compress_HC_extStateHC() :
+ *  Same as LZ4_compress_HC(), but using an externally allocated memory segment for `state`.
+ * `state` size is provided by LZ4_sizeofStateHC().
+ *  Memory segment must be aligned on 8-bytes boundaries (which a normal malloc() should do properly).
+ */
+LZ4LIB_API int LZ4_sizeofStateHC(void);
+LZ4LIB_API int LZ4_compress_HC_extStateHC(void* stateHC, const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel);
+
+
+/*! LZ4_compress_HC_destSize() : v1.9.0+
+ *  Will compress as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided in 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr is updated to indicate how much bytes were read from `src`
+ */
+LZ4LIB_API int LZ4_compress_HC_destSize(void* stateHC,
+                                  const char* src, char* dst,
+                                        int* srcSizePtr, int targetDstSize,
+                                        int compressionLevel);
+
+
+/*-************************************
+ *  Streaming Compression
+ *  Bufferless synchronous API
+ **************************************/
+ typedef union LZ4_streamHC_u LZ4_streamHC_t;   /* incomplete type (defined later) */
+
+/*! LZ4_createStreamHC() and LZ4_freeStreamHC() :
+ *  These functions create and release memory for LZ4 HC streaming state.
+ *  Newly created states are automatically initialized.
+ *  A same state can be used multiple times consecutively,
+ *  starting with LZ4_resetStreamHC_fast() to start a new stream of blocks.
+ */
+LZ4LIB_API LZ4_streamHC_t* LZ4_createStreamHC(void);
+LZ4LIB_API int             LZ4_freeStreamHC (LZ4_streamHC_t* streamHCPtr);
+
+/*
+  These functions compress data in successive blocks of any size,
+  using previous blocks as dictionary, to improve compression ratio.
+  One key assumption is that previous blocks (up to 64 KB) remain read-accessible while compressing next blocks.
+  There is an exception for ring buffers, which can be smaller than 64 KB.
+  Ring-buffer scenario is automatically detected and handled within LZ4_compress_HC_continue().
+
+  Before starting compression, state must be allocated and properly initialized.
+  LZ4_createStreamHC() does both, though compression level is set to LZ4HC_CLEVEL_DEFAULT.
+
+  Selecting the compression level can be done with LZ4_resetStreamHC_fast() (starts a new stream)
+  or LZ4_setCompressionLevel() (anytime, between blocks in the same stream) (experimental).
+  LZ4_resetStreamHC_fast() only works on states which have been properly initialized at least once,
+  which is automatically the case when state is created using LZ4_createStreamHC().
+
+  After reset, a first "fictional block" can be designated as initial dictionary,
+  using LZ4_loadDictHC() (Optional).
+
+  Invoke LZ4_compress_HC_continue() to compress each successive block.
+  The number of blocks is unlimited.
+  Previous input blocks, including initial dictionary when present,
+  must remain accessible and unmodified during compression.
+
+  It's allowed to update compression level anytime between blocks,
+  using LZ4_setCompressionLevel() (experimental).
+
+  'dst' buffer should be sized to handle worst case scenarios
+  (see LZ4_compressBound(), it ensures compression success).
+  In case of failure, the API does not guarantee recovery,
+  so the state _must_ be reset.
+  To ensure compression success
+  whenever `dst` buffer size cannot be made >= LZ4_compressBound(),
+  consider using LZ4_compress_HC_continue_destSize().
+
+  Whenever previous input blocks can't be preserved unmodified in-place during compression of next blocks,
+  it's possible to copy the last blocks into a more stable memory space, using LZ4_saveDictHC().
+  Return value of LZ4_saveDictHC() is the size of dictionary effectively saved into 'safeBuffer' (<= 64 KB)
+
+  After completing a streaming compression,
+  it's possible to start a new stream of blocks, using the same LZ4_streamHC_t state,
+  just by resetting it, using LZ4_resetStreamHC_fast().
+*/
+
+LZ4LIB_API void LZ4_resetStreamHC_fast(LZ4_streamHC_t* streamHCPtr, int compressionLevel);   /* v1.9.0+ */
+LZ4LIB_API int  LZ4_loadDictHC (LZ4_streamHC_t* streamHCPtr, const char* dictionary, int dictSize);
+
+LZ4LIB_API int LZ4_compress_HC_continue (LZ4_streamHC_t* streamHCPtr,
+                                   const char* src, char* dst,
+                                         int srcSize, int maxDstSize);
+
+/*! LZ4_compress_HC_continue_destSize() : v1.9.0+
+ *  Similar to LZ4_compress_HC_continue(),
+ *  but will read as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided into 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr will be updated to indicate how much bytes were read from `src`.
+ *           Note that this function may not consume the entire input.
+ */
+LZ4LIB_API int LZ4_compress_HC_continue_destSize(LZ4_streamHC_t* LZ4_streamHCPtr,
+                                           const char* src, char* dst,
+                                                 int* srcSizePtr, int targetDstSize);
+
+LZ4LIB_API int LZ4_saveDictHC (LZ4_streamHC_t* streamHCPtr, char* safeBuffer, int maxDictSize);
+
+
+
+/*^**********************************************
+ * !!!!!!   STATIC LINKING ONLY   !!!!!!
+ ***********************************************/
+
+/*-******************************************************************
+ * PRIVATE DEFINITIONS :
+ * Do not use these definitions directly.
+ * They are merely exposed to allow static allocation of `LZ4_streamHC_t`.
+ * Declare an `LZ4_streamHC_t` directly, rather than any type below.
+ * Even then, only do so in the context of static linking, as definitions may change between versions.
+ ********************************************************************/
+
+#define LZ4HC_DICTIONARY_LOGSIZE 16
+#define LZ4HC_MAXD (1<<LZ4HC_DICTIONARY_LOGSIZE)
+#define LZ4HC_MAXD_MASK (LZ4HC_MAXD - 1)
+
+#define LZ4HC_HASH_LOG 15
+#define LZ4HC_HASHTABLESIZE (1 << LZ4HC_HASH_LOG)
+#define LZ4HC_HASH_MASK (LZ4HC_HASHTABLESIZE - 1)
+
+
+#if defined(__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#include <stdint.h>
+
+typedef struct LZ4HC_CCtx_internal LZ4HC_CCtx_internal;
+struct LZ4HC_CCtx_internal
+{
+    uint32_t   hashTable[LZ4HC_HASHTABLESIZE];
+    uint16_t   chainTable[LZ4HC_MAXD];
+    const uint8_t* end;         /* next block here to continue on current prefix */
+    const uint8_t* base;        /* All index relative to this position */
+    const uint8_t* dictBase;    /* alternate base for extDict */
+    uint32_t   dictLimit;       /* below that point, need extDict */
+    uint32_t   lowLimit;        /* below that point, no more dict */
+    uint32_t   nextToUpdate;    /* index from which to continue dictionary update */
+    short      compressionLevel;
+    int8_t     favorDecSpeed;   /* favor decompression speed if this flag set,
+                                   otherwise, favor compression ratio */
+    int8_t     dirty;           /* stream has to be fully reset if this flag is set */
+    const LZ4HC_CCtx_internal* dictCtx;
+};
+
+#else
+
+typedef struct LZ4HC_CCtx_internal LZ4HC_CCtx_internal;
+struct LZ4HC_CCtx_internal
+{
+    unsigned int   hashTable[LZ4HC_HASHTABLESIZE];
+    unsigned short chainTable[LZ4HC_MAXD];
+    const unsigned char* end;        /* next block here to continue on current prefix */
+    const unsigned char* base;       /* All index relative to this position */
+    const unsigned char* dictBase;   /* alternate base for extDict */
+    unsigned int   dictLimit;        /* below that point, need extDict */
+    unsigned int   lowLimit;         /* below that point, no more dict */
+    unsigned int   nextToUpdate;     /* index from which to continue dictionary update */
+    short          compressionLevel;
+    char           favorDecSpeed;    /* favor decompression speed if this flag set,
+                                        otherwise, favor compression ratio */
+    char           dirty;            /* stream has to be fully reset if this flag is set */
+    const LZ4HC_CCtx_internal* dictCtx;
+};
+
+#endif
+
+
+/* Do not use these definitions directly !
+ * Declare or allocate an LZ4_streamHC_t instead.
+ */
+#define LZ4_STREAMHCSIZE       (4*LZ4HC_HASHTABLESIZE + 2*LZ4HC_MAXD + 56 + ((sizeof(void*)==16) ? 56 : 0) /* AS400*/ ) /* 262200 or 262256*/
+#define LZ4_STREAMHCSIZE_SIZET (LZ4_STREAMHCSIZE / sizeof(size_t))
+union LZ4_streamHC_u {
+    size_t table[LZ4_STREAMHCSIZE_SIZET];
+    LZ4HC_CCtx_internal internal_donotuse;
+}; /* previously typedef'd to LZ4_streamHC_t */
+
+/* LZ4_streamHC_t :
+ * This structure allows static allocation of LZ4 HC streaming state.
+ * This can be used to allocate statically, on state, or as part of a larger structure.
+ *
+ * Such state **must** be initialized using LZ4_initStreamHC() before first use.
+ *
+ * Note that invoking LZ4_initStreamHC() is not required when
+ * the state was created using LZ4_createStreamHC() (which is recommended).
+ * Using the normal builder, a newly created state is automatically initialized.
+ *
+ * Static allocation shall only be used in combination with static linking.
+ */
+
+/* LZ4_initStreamHC() : v1.9.0+
+ * Required before first use of a statically allocated LZ4_streamHC_t.
+ * Before v1.9.0 : use LZ4_resetStreamHC() instead
+ */
+LZ4LIB_API LZ4_streamHC_t* LZ4_initStreamHC (void* buffer, size_t size);
+
+
+/*-************************************
+*  Deprecated Functions
+**************************************/
+/* see lz4.h LZ4_DISABLE_DEPRECATE_WARNINGS to turn off deprecation warnings */
+
+/* deprecated compression functions */
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC               (const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput (const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2              (const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_withStateHC               (void* state, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_withStateHC              (void* state, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_withStateHC(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_continue               (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
+
+/* Obsolete streaming functions; degraded functionality; do not use!
+ *
+ * In order to perform streaming compression, these functions depended on data
+ * that is no longer tracked in the state. They have been preserved as well as
+ * possible: using them will still produce a correct output. However, use of
+ * LZ4_slideInputBufferHC() will truncate the history of the stream, rather
+ * than preserve a window-sized chunk of history.
+ */
+LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API void* LZ4_createHC (const char* inputBuffer);
+LZ4_DEPRECATED("use LZ4_saveDictHC() instead") LZ4LIB_API     char* LZ4_slideInputBufferHC (void* LZ4HC_Data);
+LZ4_DEPRECATED("use LZ4_freeStreamHC() instead") LZ4LIB_API   int   LZ4_freeHC (void* LZ4HC_Data);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_continue               (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API int   LZ4_sizeofStreamStateHC(void);
+LZ4_DEPRECATED("use LZ4_initStreamHC() instead") LZ4LIB_API  int   LZ4_resetStreamStateHC(void* state, char* inputBuffer);
+
+
+/* LZ4_resetStreamHC() is now replaced by LZ4_initStreamHC().
+ * The intention is to emphasize the difference with LZ4_resetStreamHC_fast(),
+ * which is now the recommended function to start a new stream of blocks,
+ * but cannot be used to initialize a memory segment containing arbitrary garbage data.
+ *
+ * It is recommended to switch to LZ4_initStreamHC().
+ * LZ4_resetStreamHC() will generate deprecation warnings in a future version.
+ */
+LZ4LIB_API void LZ4_resetStreamHC (LZ4_streamHC_t* streamHCPtr, int compressionLevel);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* LZ4_HC_H_19834876238432 */
+
+
+/*-**************************************************
+ * !!!!!     STATIC LINKING ONLY     !!!!!
+ * Following definitions are considered experimental.
+ * They should not be linked from DLL,
+ * as there is no guarantee of API stability yet.
+ * Prototypes will be promoted to "stable" status
+ * after successfull usage in real-life scenarios.
+ ***************************************************/
+#ifdef LZ4_HC_STATIC_LINKING_ONLY   /* protection macro */
+#ifndef LZ4_HC_SLO_098092834
+#define LZ4_HC_SLO_098092834
+
+#define LZ4_STATIC_LINKING_ONLY   /* LZ4LIB_STATIC_API */
+#include "lz4.h"
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*! LZ4_setCompressionLevel() : v1.8.0+ (experimental)
+ *  It's possible to change compression level
+ *  between successive invocations of LZ4_compress_HC_continue*()
+ *  for dynamic adaptation.
+ */
+LZ4LIB_STATIC_API void LZ4_setCompressionLevel(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
+
+/*! LZ4_favorDecompressionSpeed() : v1.8.2+ (experimental)
+ *  Opt. Parser will favor decompression speed over compression ratio.
+ *  Only applicable to levels >= LZ4HC_CLEVEL_OPT_MIN.
+ */
+LZ4LIB_STATIC_API void LZ4_favorDecompressionSpeed(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int favor);
+
+/*! LZ4_resetStreamHC_fast() : v1.9.0+
+ *  When an LZ4_streamHC_t is known to be in a internally coherent state,
+ *  it can often be prepared for a new compression with almost no work, only
+ *  sometimes falling back to the full, expensive reset that is always required
+ *  when the stream is in an indeterminate state (i.e., the reset performed by
+ *  LZ4_resetStreamHC()).
+ *
+ *  LZ4_streamHCs are guaranteed to be in a valid state when:
+ *  - returned from LZ4_createStreamHC()
+ *  - reset by LZ4_resetStreamHC()
+ *  - memset(stream, 0, sizeof(LZ4_streamHC_t))
+ *  - the stream was in a valid state and was reset by LZ4_resetStreamHC_fast()
+ *  - the stream was in a valid state and was then used in any compression call
+ *    that returned success
+ *  - the stream was in an indeterminate state and was used in a compression
+ *    call that fully reset the state (LZ4_compress_HC_extStateHC()) and that
+ *    returned success
+ *
+ *  Note:
+ *  A stream that was last used in a compression call that returned an error
+ *  may be passed to this function. However, it will be fully reset, which will
+ *  clear any existing history and settings from the context.
+ */
+LZ4LIB_STATIC_API void LZ4_resetStreamHC_fast(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
+
+/*! LZ4_compress_HC_extStateHC_fastReset() :
+ *  A variant of LZ4_compress_HC_extStateHC().
+ *
+ *  Using this variant avoids an expensive initialization step. It is only safe
+ *  to call if the state buffer is known to be correctly initialized already
+ *  (see above comment on LZ4_resetStreamHC_fast() for a definition of
+ *  "correctly initialized"). From a high level, the difference is that this
+ *  function initializes the provided state with a call to
+ *  LZ4_resetStreamHC_fast() while LZ4_compress_HC_extStateHC() starts with a
+ *  call to LZ4_resetStreamHC().
+ */
+LZ4LIB_STATIC_API int LZ4_compress_HC_extStateHC_fastReset (
+    void* state,
+    const char* src, char* dst,
+    int srcSize, int dstCapacity,
+    int compressionLevel);
+
+/*! LZ4_attach_HC_dictionary() :
+ *  This is an experimental API that allows for the efficient use of a
+ *  static dictionary many times.
+ *
+ *  Rather than re-loading the dictionary buffer into a working context before
+ *  each compression, or copying a pre-loaded dictionary's LZ4_streamHC_t into a
+ *  working LZ4_streamHC_t, this function introduces a no-copy setup mechanism,
+ *  in which the working stream references the dictionary stream in-place.
+ *
+ *  Several assumptions are made about the state of the dictionary stream.
+ *  Currently, only streams which have been prepared by LZ4_loadDictHC() should
+ *  be expected to work.
+ *
+ *  Alternatively, the provided dictionary stream pointer may be NULL, in which
+ *  case any existing dictionary stream is unset.
+ *
+ *  A dictionary should only be attached to a stream without any history (i.e.,
+ *  a stream that has just been reset).
+ *
+ *  The dictionary will remain attached to the working stream only for the
+ *  current stream session. Calls to LZ4_resetStreamHC(_fast) will remove the
+ *  dictionary context association from the working stream. The dictionary
+ *  stream (and source buffer) must remain in-place / accessible / unchanged
+ *  through the lifetime of the stream session.
+ */
+LZ4LIB_STATIC_API void LZ4_attach_HC_dictionary(
+          LZ4_streamHC_t *working_stream,
+    const LZ4_streamHC_t *dictionary_stream);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* LZ4_HC_SLO_098092834 */
+#endif   /* LZ4_HC_STATIC_LINKING_ONLY */
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4hc.c b/lib/clickhouse-cpp/contrib/lz4/lz4hc.c
deleted file mode 100644
index bbe7a9d..0000000
--- a/lib/clickhouse-cpp/contrib/lz4/lz4hc.c
+++ /dev/null
@@ -1,731 +0,0 @@
-/*
-    LZ4 HC - High Compression Mode of LZ4
-    Copyright (C) 2011-2015, Yann Collet.
-
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-    Redistribution and use in source and binary forms, with or without
-    modification, are permitted provided that the following conditions are
-    met:
-
-    * Redistributions of source code must retain the above copyright
-    notice, this list of conditions and the following disclaimer.
-    * Redistributions in binary form must reproduce the above
-    copyright notice, this list of conditions and the following disclaimer
-    in the documentation and/or other materials provided with the
-    distribution.
-
-    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-    You can contact the author at :
-       - LZ4 source repository : https://github.com/Cyan4973/lz4
-       - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
-*/
-
-
-
-/**************************************
-*  Tuning Parameter
-**************************************/
-static const int LZ4HC_compressionLevel_default = 9;
-
-
-/**************************************
-*  Includes
-**************************************/
-#include "lz4hc.h"
-
-
-/**************************************
-*  Local Compiler Options
-**************************************/
-#if defined(__GNUC__)
-#  pragma GCC diagnostic ignored "-Wunused-function"
-#endif
-
-#if defined (__clang__)
-#  pragma clang diagnostic ignored "-Wunused-function"
-#endif
-
-
-/**************************************
-*  Common LZ4 definition
-**************************************/
-#define LZ4_COMMONDEFS_ONLY
-#include "lz4.c"
-
-
-/**************************************
-*  Local Constants
-**************************************/
-#define DICTIONARY_LOGSIZE 16
-#define MAXD (1<<DICTIONARY_LOGSIZE)
-#define MAXD_MASK (MAXD - 1)
-
-#define HASH_LOG (DICTIONARY_LOGSIZE-1)
-#define HASHTABLESIZE (1 << HASH_LOG)
-#define HASH_MASK (HASHTABLESIZE - 1)
-
-#define OPTIMAL_ML (int)((ML_MASK-1)+MINMATCH)
-
-static const int g_maxCompressionLevel = 16;
-
-
-/**************************************
-*  Local Types
-**************************************/
-typedef struct
-{
-    U32   hashTable[HASHTABLESIZE];
-    U16   chainTable[MAXD];
-    const BYTE* end;        /* next block here to continue on current prefix */
-    const BYTE* base;       /* All index relative to this position */
-    const BYTE* dictBase;   /* alternate base for extDict */
-    BYTE* inputBuffer;      /* deprecated */
-    U32   dictLimit;        /* below that point, need extDict */
-    U32   lowLimit;         /* below that point, no more dict */
-    U32   nextToUpdate;     /* index from which to continue dictionary update */
-    U32   compressionLevel;
-} LZ4HC_Data_Structure;
-
-
-/**************************************
-*  Local Macros
-**************************************/
-#define HASH_FUNCTION(i)       (((i) * 2654435761U) >> ((MINMATCH*8)-HASH_LOG))
-//#define DELTANEXTU16(p)        chainTable[(p) & MAXD_MASK]   /* flexible, MAXD dependent */
-#define DELTANEXTU16(p)        chainTable[(U16)(p)]   /* faster */
-
-static U32 LZ4HC_hashPtr(const void* ptr) { return HASH_FUNCTION(LZ4_read32(ptr)); }
-
-
-
-/**************************************
-*  HC Compression
-**************************************/
-static void LZ4HC_init (LZ4HC_Data_Structure* hc4, const BYTE* start)
-{
-    MEM_INIT((void*)hc4->hashTable, 0, sizeof(hc4->hashTable));
-    MEM_INIT(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
-    hc4->nextToUpdate = 64 KB;
-    hc4->base = start - 64 KB;
-    hc4->end = start;
-    hc4->dictBase = start - 64 KB;
-    hc4->dictLimit = 64 KB;
-    hc4->lowLimit = 64 KB;
-}
-
-
-/* Update chains up to ip (excluded) */
-FORCE_INLINE void LZ4HC_Insert (LZ4HC_Data_Structure* hc4, const BYTE* ip)
-{
-    U16* chainTable = hc4->chainTable;
-    U32* HashTable  = hc4->hashTable;
-    const BYTE* const base = hc4->base;
-    const U32 target = (U32)(ip - base);
-    U32 idx = hc4->nextToUpdate;
-
-    while(idx < target)
-    {
-        U32 h = LZ4HC_hashPtr(base+idx);
-        size_t delta = idx - HashTable[h];
-        if (delta>MAX_DISTANCE) delta = MAX_DISTANCE;
-        DELTANEXTU16(idx) = (U16)delta;
-        HashTable[h] = idx;
-        idx++;
-    }
-
-    hc4->nextToUpdate = target;
-}
-
-
-FORCE_INLINE int LZ4HC_InsertAndFindBestMatch (LZ4HC_Data_Structure* hc4,   /* Index table will be updated */
-                                               const BYTE* ip, const BYTE* const iLimit,
-                                               const BYTE** matchpos,
-                                               const int maxNbAttempts)
-{
-    U16* const chainTable = hc4->chainTable;
-    U32* const HashTable = hc4->hashTable;
-    const BYTE* const base = hc4->base;
-    const BYTE* const dictBase = hc4->dictBase;
-    const U32 dictLimit = hc4->dictLimit;
-    const U32 lowLimit = (hc4->lowLimit + 64 KB > (U32)(ip-base)) ? hc4->lowLimit : (U32)(ip - base) - (64 KB - 1);
-    U32 matchIndex;
-    const BYTE* match;
-    int nbAttempts=maxNbAttempts;
-    size_t ml=0;
-
-    /* HC4 match finder */
-    LZ4HC_Insert(hc4, ip);
-    matchIndex = HashTable[LZ4HC_hashPtr(ip)];
-
-    while ((matchIndex>=lowLimit) && (nbAttempts))
-    {
-        nbAttempts--;
-        if (matchIndex >= dictLimit)
-        {
-            match = base + matchIndex;
-            if (*(match+ml) == *(ip+ml)
-                && (LZ4_read32(match) == LZ4_read32(ip)))
-            {
-                size_t mlt = LZ4_count(ip+MINMATCH, match+MINMATCH, iLimit) + MINMATCH;
-                if (mlt > ml) { ml = mlt; *matchpos = match; }
-            }
-        }
-        else
-        {
-            match = dictBase + matchIndex;
-            if (LZ4_read32(match) == LZ4_read32(ip))
-            {
-                size_t mlt;
-                const BYTE* vLimit = ip + (dictLimit - matchIndex);
-                if (vLimit > iLimit) vLimit = iLimit;
-                mlt = LZ4_count(ip+MINMATCH, match+MINMATCH, vLimit) + MINMATCH;
-                if ((ip+mlt == vLimit) && (vLimit < iLimit))
-                    mlt += LZ4_count(ip+mlt, base+dictLimit, iLimit);
-                if (mlt > ml) { ml = mlt; *matchpos = base + matchIndex; }   /* virtual matchpos */
-            }
-        }
-        matchIndex -= DELTANEXTU16(matchIndex);
-    }
-
-    return (int)ml;
-}
-
-
-FORCE_INLINE int LZ4HC_InsertAndGetWiderMatch (
-    LZ4HC_Data_Structure* hc4,
-    const BYTE* const ip,
-    const BYTE* const iLowLimit,
-    const BYTE* const iHighLimit,
-    int longest,
-    const BYTE** matchpos,
-    const BYTE** startpos,
-    const int maxNbAttempts)
-{
-    U16* const chainTable = hc4->chainTable;
-    U32* const HashTable = hc4->hashTable;
-    const BYTE* const base = hc4->base;
-    const U32 dictLimit = hc4->dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const U32 lowLimit = (hc4->lowLimit + 64 KB > (U32)(ip-base)) ? hc4->lowLimit : (U32)(ip - base) - (64 KB - 1);
-    const BYTE* const dictBase = hc4->dictBase;
-    U32   matchIndex;
-    int nbAttempts = maxNbAttempts;
-    int delta = (int)(ip-iLowLimit);
-
-
-    /* First Match */
-    LZ4HC_Insert(hc4, ip);
-    matchIndex = HashTable[LZ4HC_hashPtr(ip)];
-
-    while ((matchIndex>=lowLimit) && (nbAttempts))
-    {
-        nbAttempts--;
-        if (matchIndex >= dictLimit)
-        {
-            const BYTE* matchPtr = base + matchIndex;
-            if (*(iLowLimit + longest) == *(matchPtr - delta + longest))
-                if (LZ4_read32(matchPtr) == LZ4_read32(ip))
-                {
-                    int mlt = MINMATCH + LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
-                    int back = 0;
-
-                    while ((ip+back>iLowLimit)
-                           && (matchPtr+back > lowPrefixPtr)
-                           && (ip[back-1] == matchPtr[back-1]))
-                            back--;
-
-                    mlt -= back;
-
-                    if (mlt > longest)
-                    {
-                        longest = (int)mlt;
-                        *matchpos = matchPtr+back;
-                        *startpos = ip+back;
-                    }
-                }
-        }
-        else
-        {
-            const BYTE* matchPtr = dictBase + matchIndex;
-            if (LZ4_read32(matchPtr) == LZ4_read32(ip))
-            {
-                size_t mlt;
-                int back=0;
-                const BYTE* vLimit = ip + (dictLimit - matchIndex);
-                if (vLimit > iHighLimit) vLimit = iHighLimit;
-                mlt = LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
-                if ((ip+mlt == vLimit) && (vLimit < iHighLimit))
-                    mlt += LZ4_count(ip+mlt, base+dictLimit, iHighLimit);
-                while ((ip+back > iLowLimit) && (matchIndex+back > lowLimit) && (ip[back-1] == matchPtr[back-1])) back--;
-                mlt -= back;
-                if ((int)mlt > longest) { longest = (int)mlt; *matchpos = base + matchIndex + back; *startpos = ip+back; }
-            }
-        }
-        matchIndex -= DELTANEXTU16(matchIndex);
-    }
-
-    return longest;
-}
-
-
-typedef enum { noLimit = 0, limitedOutput = 1 } limitedOutput_directive;
-
-#define LZ4HC_DEBUG 0
-#if LZ4HC_DEBUG
-static unsigned debug = 0;
-#endif
-
-FORCE_INLINE int LZ4HC_encodeSequence (
-    const BYTE** ip,
-    BYTE** op,
-    const BYTE** anchor,
-    int matchLength,
-    const BYTE* const match,
-    limitedOutput_directive limitedOutputBuffer,
-    BYTE* oend)
-{
-    int length;
-    BYTE* token;
-
-#if LZ4HC_DEBUG
-    if (debug) printf("literal : %u  --  match : %u  --  offset : %u\n", (U32)(*ip - *anchor), (U32)matchLength, (U32)(*ip-match));
-#endif
-
-    /* Encode Literal length */
-    length = (int)(*ip - *anchor);
-    token = (*op)++;
-    if ((limitedOutputBuffer) && ((*op + (length>>8) + length + (2 + 1 + LASTLITERALS)) > oend)) return 1;   /* Check output limit */
-    if (length>=(int)RUN_MASK) { int len; *token=(RUN_MASK<<ML_BITS); len = length-RUN_MASK; for(; len > 254 ; len-=255) *(*op)++ = 255;  *(*op)++ = (BYTE)len; }
-    else *token = (BYTE)(length<<ML_BITS);
-
-    /* Copy Literals */
-    LZ4_wildCopy(*op, *anchor, (*op) + length);
-    *op += length;
-
-    /* Encode Offset */
-    LZ4_writeLE16(*op, (U16)(*ip-match)); *op += 2;
-
-    /* Encode MatchLength */
-    length = (int)(matchLength-MINMATCH);
-    if ((limitedOutputBuffer) && (*op + (length>>8) + (1 + LASTLITERALS) > oend)) return 1;   /* Check output limit */
-    if (length>=(int)ML_MASK) { *token+=ML_MASK; length-=ML_MASK; for(; length > 509 ; length-=510) { *(*op)++ = 255; *(*op)++ = 255; } if (length > 254) { length-=255; *(*op)++ = 255; } *(*op)++ = (BYTE)length; }
-    else *token += (BYTE)(length);
-
-    /* Prepare next loop */
-    *ip += matchLength;
-    *anchor = *ip;
-
-    return 0;
-}
-
-
-static int LZ4HC_compress_generic (
-    void* ctxvoid,
-    const char* source,
-    char* dest,
-    int inputSize,
-    int maxOutputSize,
-    int compressionLevel,
-    limitedOutput_directive limit
-    )
-{
-    LZ4HC_Data_Structure* ctx = (LZ4HC_Data_Structure*) ctxvoid;
-    const BYTE* ip = (const BYTE*) source;
-    const BYTE* anchor = ip;
-    const BYTE* const iend = ip + inputSize;
-    const BYTE* const mflimit = iend - MFLIMIT;
-    const BYTE* const matchlimit = (iend - LASTLITERALS);
-
-    BYTE* op = (BYTE*) dest;
-    BYTE* const oend = op + maxOutputSize;
-
-    unsigned maxNbAttempts;
-    int   ml, ml2, ml3, ml0;
-    const BYTE* ref=NULL;
-    const BYTE* start2=NULL;
-    const BYTE* ref2=NULL;
-    const BYTE* start3=NULL;
-    const BYTE* ref3=NULL;
-    const BYTE* start0;
-    const BYTE* ref0;
-
-
-    /* init */
-    if (compressionLevel > g_maxCompressionLevel) compressionLevel = g_maxCompressionLevel;
-    if (compressionLevel < 1) compressionLevel = LZ4HC_compressionLevel_default;
-    maxNbAttempts = 1 << (compressionLevel-1);
-    ctx->end += inputSize;
-
-    ip++;
-
-    /* Main Loop */
-    while (ip < mflimit)
-    {
-        ml = LZ4HC_InsertAndFindBestMatch (ctx, ip, matchlimit, (&ref), maxNbAttempts);
-        if (!ml) { ip++; continue; }
-
-        /* saved, in case we would skip too much */
-        start0 = ip;
-        ref0 = ref;
-        ml0 = ml;
-
-_Search2:
-        if (ip+ml < mflimit)
-            ml2 = LZ4HC_InsertAndGetWiderMatch(ctx, ip + ml - 2, ip + 1, matchlimit, ml, &ref2, &start2, maxNbAttempts);
-        else ml2 = ml;
-
-        if (ml2 == ml)  /* No better match */
-        {
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) return 0;
-            continue;
-        }
-
-        if (start0 < ip)
-        {
-            if (start2 < ip + ml0)   /* empirical */
-            {
-                ip = start0;
-                ref = ref0;
-                ml = ml0;
-            }
-        }
-
-        /* Here, start0==ip */
-        if ((start2 - ip) < 3)   /* First Match too small : removed */
-        {
-            ml = ml2;
-            ip = start2;
-            ref =ref2;
-            goto _Search2;
-        }
-
-_Search3:
-        /*
-        * Currently we have :
-        * ml2 > ml1, and
-        * ip1+3 <= ip2 (usually < ip1+ml1)
-        */
-        if ((start2 - ip) < OPTIMAL_ML)
-        {
-            int correction;
-            int new_ml = ml;
-            if (new_ml > OPTIMAL_ML) new_ml = OPTIMAL_ML;
-            if (ip+new_ml > start2 + ml2 - MINMATCH) new_ml = (int)(start2 - ip) + ml2 - MINMATCH;
-            correction = new_ml - (int)(start2 - ip);
-            if (correction > 0)
-            {
-                start2 += correction;
-                ref2 += correction;
-                ml2 -= correction;
-            }
-        }
-        /* Now, we have start2 = ip+new_ml, with new_ml = min(ml, OPTIMAL_ML=18) */
-
-        if (start2 + ml2 < mflimit)
-            ml3 = LZ4HC_InsertAndGetWiderMatch(ctx, start2 + ml2 - 3, start2, matchlimit, ml2, &ref3, &start3, maxNbAttempts);
-        else ml3 = ml2;
-
-        if (ml3 == ml2) /* No better match : 2 sequences to encode */
-        {
-            /* ip & ref are known; Now for ml */
-            if (start2 < ip+ml)  ml = (int)(start2 - ip);
-            /* Now, encode 2 sequences */
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) return 0;
-            ip = start2;
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml2, ref2, limit, oend)) return 0;
-            continue;
-        }
-
-        if (start3 < ip+ml+3) /* Not enough space for match 2 : remove it */
-        {
-            if (start3 >= (ip+ml)) /* can write Seq1 immediately ==> Seq2 is removed, so Seq3 becomes Seq1 */
-            {
-                if (start2 < ip+ml)
-                {
-                    int correction = (int)(ip+ml - start2);
-                    start2 += correction;
-                    ref2 += correction;
-                    ml2 -= correction;
-                    if (ml2 < MINMATCH)
-                    {
-                        start2 = start3;
-                        ref2 = ref3;
-                        ml2 = ml3;
-                    }
-                }
-
-                if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) return 0;
-                ip  = start3;
-                ref = ref3;
-                ml  = ml3;
-
-                start0 = start2;
-                ref0 = ref2;
-                ml0 = ml2;
-                goto _Search2;
-            }
-
-            start2 = start3;
-            ref2 = ref3;
-            ml2 = ml3;
-            goto _Search3;
-        }
-
-        /*
-        * OK, now we have 3 ascending matches; let's write at least the first one
-        * ip & ref are known; Now for ml
-        */
-        if (start2 < ip+ml)
-        {
-            if ((start2 - ip) < (int)ML_MASK)
-            {
-                int correction;
-                if (ml > OPTIMAL_ML) ml = OPTIMAL_ML;
-                if (ip + ml > start2 + ml2 - MINMATCH) ml = (int)(start2 - ip) + ml2 - MINMATCH;
-                correction = ml - (int)(start2 - ip);
-                if (correction > 0)
-                {
-                    start2 += correction;
-                    ref2 += correction;
-                    ml2 -= correction;
-                }
-            }
-            else
-            {
-                ml = (int)(start2 - ip);
-            }
-        }
-        if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) return 0;
-
-        ip = start2;
-        ref = ref2;
-        ml = ml2;
-
-        start2 = start3;
-        ref2 = ref3;
-        ml2 = ml3;
-
-        goto _Search3;
-    }
-
-    /* Encode Last Literals */
-    {
-        int lastRun = (int)(iend - anchor);
-        if ((limit) && (((char*)op - dest) + lastRun + 1 + ((lastRun+255-RUN_MASK)/255) > (U32)maxOutputSize)) return 0;  /* Check output limit */
-        if (lastRun>=(int)RUN_MASK) { *op++=(RUN_MASK<<ML_BITS); lastRun-=RUN_MASK; for(; lastRun > 254 ; lastRun-=255) *op++ = 255; *op++ = (BYTE) lastRun; }
-        else *op++ = (BYTE)(lastRun<<ML_BITS);
-        memcpy(op, anchor, iend - anchor);
-        op += iend-anchor;
-    }
-
-    /* End */
-    return (int) (((char*)op)-dest);
-}
-
-
-int LZ4_sizeofStateHC(void) { return sizeof(LZ4HC_Data_Structure); }
-
-int LZ4_compress_HC_extStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel)
-{
-    if (((size_t)(state)&(sizeof(void*)-1)) != 0) return 0;   /* Error : state is not aligned for pointers (32 or 64 bits) */
-    LZ4HC_init ((LZ4HC_Data_Structure*)state, (const BYTE*)src);
-    if (maxDstSize < LZ4_compressBound(srcSize))
-        return LZ4HC_compress_generic (state, src, dst, srcSize, maxDstSize, compressionLevel, limitedOutput);
-    else
-        return LZ4HC_compress_generic (state, src, dst, srcSize, maxDstSize, compressionLevel, noLimit);
-}
-
-int LZ4_compress_HC(const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel)
-{
-    LZ4HC_Data_Structure state;
-    return LZ4_compress_HC_extStateHC(&state, src, dst, srcSize, maxDstSize, compressionLevel);
-}
-
-
-
-/**************************************
-*  Streaming Functions
-**************************************/
-/* allocation */
-LZ4_streamHC_t* LZ4_createStreamHC(void) { return (LZ4_streamHC_t*)malloc(sizeof(LZ4_streamHC_t)); }
-int             LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr) { free(LZ4_streamHCPtr); return 0; }
-
-
-/* initialization */
-void LZ4_resetStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
-{
-    LZ4_STATIC_ASSERT(sizeof(LZ4HC_Data_Structure) <= sizeof(LZ4_streamHC_t));   /* if compilation fails here, LZ4_STREAMHCSIZE must be increased */
-    ((LZ4HC_Data_Structure*)LZ4_streamHCPtr)->base = NULL;
-    ((LZ4HC_Data_Structure*)LZ4_streamHCPtr)->compressionLevel = (unsigned)compressionLevel;
-}
-
-int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, const char* dictionary, int dictSize)
-{
-    LZ4HC_Data_Structure* ctxPtr = (LZ4HC_Data_Structure*) LZ4_streamHCPtr;
-    if (dictSize > 64 KB)
-    {
-        dictionary += dictSize - 64 KB;
-        dictSize = 64 KB;
-    }
-    LZ4HC_init (ctxPtr, (const BYTE*)dictionary);
-    if (dictSize >= 4) LZ4HC_Insert (ctxPtr, (const BYTE*)dictionary +(dictSize-3));
-    ctxPtr->end = (const BYTE*)dictionary + dictSize;
-    return dictSize;
-}
-
-
-/* compression */
-
-static void LZ4HC_setExternalDict(LZ4HC_Data_Structure* ctxPtr, const BYTE* newBlock)
-{
-    if (ctxPtr->end >= ctxPtr->base + 4)
-        LZ4HC_Insert (ctxPtr, ctxPtr->end-3);   /* Referencing remaining dictionary content */
-    /* Only one memory segment for extDict, so any previous extDict is lost at this stage */
-    ctxPtr->lowLimit  = ctxPtr->dictLimit;
-    ctxPtr->dictLimit = (U32)(ctxPtr->end - ctxPtr->base);
-    ctxPtr->dictBase  = ctxPtr->base;
-    ctxPtr->base = newBlock - ctxPtr->dictLimit;
-    ctxPtr->end  = newBlock;
-    ctxPtr->nextToUpdate = ctxPtr->dictLimit;   /* match referencing will resume from there */
-}
-
-static int LZ4_compressHC_continue_generic (LZ4HC_Data_Structure* ctxPtr,
-                                            const char* source, char* dest,
-                                            int inputSize, int maxOutputSize, limitedOutput_directive limit)
-{
-    /* auto-init if forgotten */
-    if (ctxPtr->base == NULL)
-        LZ4HC_init (ctxPtr, (const BYTE*) source);
-
-    /* Check overflow */
-    if ((size_t)(ctxPtr->end - ctxPtr->base) > 2 GB)
-    {
-        size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->base) - ctxPtr->dictLimit;
-        if (dictSize > 64 KB) dictSize = 64 KB;
-
-        LZ4_loadDictHC((LZ4_streamHC_t*)ctxPtr, (const char*)(ctxPtr->end) - dictSize, (int)dictSize);
-    }
-
-    /* Check if blocks follow each other */
-    if ((const BYTE*)source != ctxPtr->end)
-        LZ4HC_setExternalDict(ctxPtr, (const BYTE*)source);
-
-    /* Check overlapping input/dictionary space */
-    {
-        const BYTE* sourceEnd = (const BYTE*) source + inputSize;
-        const BYTE* dictBegin = ctxPtr->dictBase + ctxPtr->lowLimit;
-        const BYTE* dictEnd   = ctxPtr->dictBase + ctxPtr->dictLimit;
-        if ((sourceEnd > dictBegin) && ((const BYTE*)source < dictEnd))
-        {
-            if (sourceEnd > dictEnd) sourceEnd = dictEnd;
-            ctxPtr->lowLimit = (U32)(sourceEnd - ctxPtr->dictBase);
-            if (ctxPtr->dictLimit - ctxPtr->lowLimit < 4) ctxPtr->lowLimit = ctxPtr->dictLimit;
-        }
-    }
-
-    return LZ4HC_compress_generic (ctxPtr, source, dest, inputSize, maxOutputSize, ctxPtr->compressionLevel, limit);
-}
-
-int LZ4_compress_HC_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize, int maxOutputSize)
-{
-    if (maxOutputSize < LZ4_compressBound(inputSize))
-        return LZ4_compressHC_continue_generic ((LZ4HC_Data_Structure*)LZ4_streamHCPtr, source, dest, inputSize, maxOutputSize, limitedOutput);
-    else
-        return LZ4_compressHC_continue_generic ((LZ4HC_Data_Structure*)LZ4_streamHCPtr, source, dest, inputSize, maxOutputSize, noLimit);
-}
-
-
-/* dictionary saving */
-
-int LZ4_saveDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, char* safeBuffer, int dictSize)
-{
-    LZ4HC_Data_Structure* streamPtr = (LZ4HC_Data_Structure*)LZ4_streamHCPtr;
-    int prefixSize = (int)(streamPtr->end - (streamPtr->base + streamPtr->dictLimit));
-    if (dictSize > 64 KB) dictSize = 64 KB;
-    if (dictSize < 4) dictSize = 0;
-    if (dictSize > prefixSize) dictSize = prefixSize;
-    memmove(safeBuffer, streamPtr->end - dictSize, dictSize);
-    {
-        U32 endIndex = (U32)(streamPtr->end - streamPtr->base);
-        streamPtr->end = (const BYTE*)safeBuffer + dictSize;
-        streamPtr->base = streamPtr->end - endIndex;
-        streamPtr->dictLimit = endIndex - dictSize;
-        streamPtr->lowLimit = endIndex - dictSize;
-        if (streamPtr->nextToUpdate < streamPtr->dictLimit) streamPtr->nextToUpdate = streamPtr->dictLimit;
-    }
-    return dictSize;
-}
-
-
-/***********************************
-*  Deprecated Functions
-***********************************/
-/* Deprecated compression functions */
-/* These functions are planned to start generate warnings by r131 approximately */
-int LZ4_compressHC(const char* src, char* dst, int srcSize) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
-int LZ4_compressHC_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, 0); }
-int LZ4_compressHC2(const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
-int LZ4_compressHC2_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, cLevel); }
-int LZ4_compressHC_withStateHC (void* state, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
-int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, maxDstSize, 0); }
-int LZ4_compressHC2_withStateHC (void* state, const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
-int LZ4_compressHC2_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, maxDstSize, cLevel); }
-int LZ4_compressHC_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, LZ4_compressBound(srcSize)); }
-int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, maxDstSize); }
-
-
-/* Deprecated streaming functions */
-/* These functions currently generate deprecation warnings */
-int LZ4_sizeofStreamStateHC(void) { return LZ4_STREAMHCSIZE; }
-
-int LZ4_resetStreamStateHC(void* state, char* inputBuffer)
-{
-    if ((((size_t)state) & (sizeof(void*)-1)) != 0) return 1;   /* Error : pointer is not aligned for pointer (32 or 64 bits) */
-    LZ4HC_init((LZ4HC_Data_Structure*)state, (const BYTE*)inputBuffer);
-    ((LZ4HC_Data_Structure*)state)->inputBuffer = (BYTE*)inputBuffer;
-    return 0;
-}
-
-void* LZ4_createHC (char* inputBuffer)
-{
-    void* hc4 = ALLOCATOR(1, sizeof(LZ4HC_Data_Structure));
-    if (hc4 == NULL) return NULL;   /* not enough memory */
-    LZ4HC_init ((LZ4HC_Data_Structure*)hc4, (const BYTE*)inputBuffer);
-    ((LZ4HC_Data_Structure*)hc4)->inputBuffer = (BYTE*)inputBuffer;
-    return hc4;
-}
-
-int LZ4_freeHC (void* LZ4HC_Data)
-{
-    FREEMEM(LZ4HC_Data);
-    return (0);
-}
-
-int LZ4_compressHC2_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int compressionLevel)
-{
-    return LZ4HC_compress_generic (LZ4HC_Data, source, dest, inputSize, 0, compressionLevel, noLimit);
-}
-
-int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel)
-{
-    return LZ4HC_compress_generic (LZ4HC_Data, source, dest, inputSize, maxOutputSize, compressionLevel, limitedOutput);
-}
-
-char* LZ4_slideInputBufferHC(void* LZ4HC_Data)
-{
-    LZ4HC_Data_Structure* hc4 = (LZ4HC_Data_Structure*)LZ4HC_Data;
-    int dictSize = LZ4_saveDictHC((LZ4_streamHC_t*)LZ4HC_Data, (char*)(hc4->inputBuffer), 64 KB);
-    return (char*)(hc4->inputBuffer + dictSize);
-}
diff --git a/lib/clickhouse-cpp/contrib/lz4/lz4hc.h b/lib/clickhouse-cpp/contrib/lz4/lz4hc.h
deleted file mode 100644
index 431f7c8..0000000
--- a/lib/clickhouse-cpp/contrib/lz4/lz4hc.h
+++ /dev/null
@@ -1,189 +0,0 @@
-/*
-   LZ4 HC - High Compression Mode of LZ4
-   Header File
-   Copyright (C) 2011-2015, Yann Collet.
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-   You can contact the author at :
-   - LZ4 source repository : https://github.com/Cyan4973/lz4
-   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
-*/
-#pragma once
-
-
-#if defined (__cplusplus)
-extern "C" {
-#endif
-
-/*****************************
-*  Includes
-*****************************/
-#include <stddef.h>   /* size_t */
-
-
-/**************************************
-*  Block Compression
-**************************************/
-int LZ4_compress_HC (const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel);
-/*
-LZ4_compress_HC :
-    Destination buffer 'dst' must be already allocated.
-    Compression completion is guaranteed if 'dst' buffer is sized to handle worst circumstances (data not compressible)
-    Worst size evaluation is provided by function LZ4_compressBound() (see "lz4.h")
-      srcSize  : Max supported value is LZ4_MAX_INPUT_SIZE (see "lz4.h")
-      compressionLevel : Recommended values are between 4 and 9, although any value between 0 and 16 will work.
-                         0 means "use default value" (see lz4hc.c).
-                         Values >16 behave the same as 16.
-      return : the number of bytes written into buffer 'dst'
-            or 0 if compression fails.
-*/
-
-
-/* Note :
-   Decompression functions are provided within LZ4 source code (see "lz4.h") (BSD license)
-*/
-
-
-int LZ4_sizeofStateHC(void);
-int LZ4_compress_HC_extStateHC(void* state, const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel);
-/*
-LZ4_compress_HC_extStateHC() :
-   Use this function if you prefer to manually allocate memory for compression tables.
-   To know how much memory must be allocated for the compression tables, use :
-      int LZ4_sizeofStateHC();
-
-   Allocated memory must be aligned on 8-bytes boundaries (which a normal malloc() will do properly).
-
-   The allocated memory can then be provided to the compression functions using 'void* state' parameter.
-   LZ4_compress_HC_extStateHC() is equivalent to previously described function.
-   It just uses externally allocated memory for stateHC.
-*/
-
-
-/**************************************
-*  Streaming Compression
-**************************************/
-#define LZ4_STREAMHCSIZE        262192
-#define LZ4_STREAMHCSIZE_SIZET (LZ4_STREAMHCSIZE / sizeof(size_t))
-typedef struct { size_t table[LZ4_STREAMHCSIZE_SIZET]; } LZ4_streamHC_t;
-/*
-  LZ4_streamHC_t
-  This structure allows static allocation of LZ4 HC streaming state.
-  State must then be initialized using LZ4_resetStreamHC() before first use.
-
-  Static allocation should only be used in combination with static linking.
-  If you want to use LZ4 as a DLL, please use construction functions below, which are future-proof.
-*/
-
-
-LZ4_streamHC_t* LZ4_createStreamHC(void);
-int             LZ4_freeStreamHC (LZ4_streamHC_t* streamHCPtr);
-/*
-  These functions create and release memory for LZ4 HC streaming state.
-  Newly created states are already initialized.
-  Existing state space can be re-used anytime using LZ4_resetStreamHC().
-  If you use LZ4 as a DLL, use these functions instead of static structure allocation,
-  to avoid size mismatch between different versions.
-*/
-
-void LZ4_resetStreamHC (LZ4_streamHC_t* streamHCPtr, int compressionLevel);
-int  LZ4_loadDictHC (LZ4_streamHC_t* streamHCPtr, const char* dictionary, int dictSize);
-
-int LZ4_compress_HC_continue (LZ4_streamHC_t* streamHCPtr, const char* src, char* dst, int srcSize, int maxDstSize);
-
-int LZ4_saveDictHC (LZ4_streamHC_t* streamHCPtr, char* safeBuffer, int maxDictSize);
-
-/*
-  These functions compress data in successive blocks of any size, using previous blocks as dictionary.
-  One key assumption is that previous blocks (up to 64 KB) remain read-accessible while compressing next blocks.
-  There is an exception for ring buffers, which can be smaller 64 KB.
-  Such case is automatically detected and correctly handled by LZ4_compress_HC_continue().
-
-  Before starting compression, state must be properly initialized, using LZ4_resetStreamHC().
-  A first "fictional block" can then be designated as initial dictionary, using LZ4_loadDictHC() (Optional).
-
-  Then, use LZ4_compress_HC_continue() to compress each successive block.
-  It works like LZ4_compress_HC(), but use previous memory blocks as dictionary to improve compression.
-  Previous memory blocks (including initial dictionary when present) must remain accessible and unmodified during compression.
-  As a reminder, size 'dst' buffer to handle worst cases, using LZ4_compressBound(), to ensure success of compression operation.
-
-  If, for any reason, previous data blocks can't be preserved unmodified in memory during next compression block,
-  you must save it to a safer memory space, using LZ4_saveDictHC().
-  Return value of LZ4_saveDictHC() is the size of dictionary effectively saved into 'safeBuffer'.
-*/
-
-
-
-/**************************************
-*  Deprecated Functions
-**************************************/
-/* Deprecate Warnings */
-/* Should these warnings messages be a problem,
-   it is generally possible to disable them,
-   with -Wno-deprecated-declarations for gcc
-   or _CRT_SECURE_NO_WARNINGS in Visual for example.
-   You can also define LZ4_DEPRECATE_WARNING_DEFBLOCK. */
-#ifndef LZ4_DEPRECATE_WARNING_DEFBLOCK
-#  define LZ4_DEPRECATE_WARNING_DEFBLOCK
-#  define LZ4_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
-#  if (LZ4_GCC_VERSION >= 405) || defined(__clang__)
-#    define LZ4_DEPRECATED(message) __attribute__((deprecated(message)))
-#  elif (LZ4_GCC_VERSION >= 301)
-#    define LZ4_DEPRECATED(message) __attribute__((deprecated))
-#  elif defined(_MSC_VER)
-#    define LZ4_DEPRECATED(message) __declspec(deprecated(message))
-#  else
-#    pragma message("WARNING: You need to implement LZ4_DEPRECATED for this compiler")
-#    define LZ4_DEPRECATED(message)
-#  endif
-#endif // LZ4_DEPRECATE_WARNING_DEFBLOCK
-
-/* compression functions */
-/* these functions are planned to trigger warning messages by r131 approximately */
-int LZ4_compressHC                (const char* source, char* dest, int inputSize);
-int LZ4_compressHC_limitedOutput  (const char* source, char* dest, int inputSize, int maxOutputSize);
-int LZ4_compressHC2               (const char* source, char* dest, int inputSize, int compressionLevel);
-int LZ4_compressHC2_limitedOutput (const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
-int LZ4_compressHC_withStateHC               (void* state, const char* source, char* dest, int inputSize);
-int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
-int LZ4_compressHC2_withStateHC              (void* state, const char* source, char* dest, int inputSize, int compressionLevel);
-int LZ4_compressHC2_limitedOutput_withStateHC(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
-int LZ4_compressHC_continue               (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize);
-int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
-
-/* Streaming functions following the older model; should no longer be used */
-LZ4_DEPRECATED("use LZ4_createStreamHC() instead") void* LZ4_createHC (char* inputBuffer);
-LZ4_DEPRECATED("use LZ4_saveDictHC() instead")     char* LZ4_slideInputBufferHC (void* LZ4HC_Data);
-LZ4_DEPRECATED("use LZ4_freeStreamHC() instead")   int   LZ4_freeHC (void* LZ4HC_Data);
-LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") int   LZ4_compressHC2_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int compressionLevel);
-LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") int   LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
-LZ4_DEPRECATED("use LZ4_createStreamHC() instead") int   LZ4_sizeofStreamStateHC(void);
-LZ4_DEPRECATED("use LZ4_resetStreamHC() instead")  int   LZ4_resetStreamStateHC(void* state, char* inputBuffer);
-
-
-#if defined (__cplusplus)
-}
-#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/BUCK b/lib/clickhouse-cpp/contrib/zstd/zstd/BUCK
new file mode 100644
index 0000000..60c6bbb
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/BUCK
@@ -0,0 +1,232 @@
+cxx_library(
+    name='zstd',
+    header_namespace='',
+    exported_headers=['zstd.h'],
+    visibility=['PUBLIC'],
+    deps=[
+        ':common',
+        ':compress',
+        ':decompress',
+        ':deprecated',
+    ],
+)
+
+cxx_library(
+    name='compress',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('compress', 'zstd*.h'),
+    ]),
+    srcs=glob(['compress/zstd*.c', 'compress/hist.c']),
+    deps=[':common'],
+)
+
+cxx_library(
+    name='decompress',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    headers=subdir_glob([
+        ('decompress', '*_impl.h'),
+    ]),
+    srcs=glob(['decompress/zstd*.c']),
+    deps=[
+        ':common',
+        ':legacy',
+    ],
+)
+
+cxx_library(
+    name='deprecated',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('deprecated', '*.h'),
+    ]),
+    srcs=glob(['deprecated/*.c']),
+    deps=[':common'],
+)
+
+cxx_library(
+    name='legacy',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('legacy', '*.h'),
+    ]),
+    srcs=glob(['legacy/*.c']),
+    deps=[':common'],
+    exported_preprocessor_flags=[
+        '-DZSTD_LEGACY_SUPPORT=4',
+    ],
+)
+
+cxx_library(
+    name='zdict',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=['zdict.h'],
+    headers=subdir_glob([
+        ('dictBuilder', 'divsufsort.h'),
+        ('dictBuilder', 'cover.h'),
+    ]),
+    srcs=glob(['dictBuilder/*.c']),
+    deps=[':common'],
+)
+
+cxx_library(
+    name='compiler',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'compiler.h'),
+    ]),
+)
+
+cxx_library(
+    name='cpu',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'cpu.h'),
+    ]),
+)
+
+cxx_library(
+    name='bitstream',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'bitstream.h'),
+    ]),
+)
+
+cxx_library(
+    name='entropy',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'fse.h'),
+        ('common', 'huf.h'),
+    ]),
+    srcs=[
+        'common/entropy_common.c',
+        'common/fse_decompress.c',
+        'compress/fse_compress.c',
+        'compress/huf_compress.c',
+        'decompress/huf_decompress.c',
+    ],
+    deps=[
+        ':debug',
+        ':bitstream',
+        ':compiler',
+        ':errors',
+        ':mem',
+    ],
+)
+
+cxx_library(
+    name='errors',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=[
+        'zstd_errors.h',
+        'common/error_private.h',
+    ]
+    srcs=['common/error_private.c'],
+)
+
+cxx_library(
+    name='mem',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'mem.h'),
+    ]),
+)
+
+cxx_library(
+    name='pool',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'pool.h'),
+    ]),
+    srcs=['common/pool.c'],
+    deps=[
+        ':threading',
+        ':zstd_common',
+    ],
+)
+
+cxx_library(
+    name='threading',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'threading.h'),
+    ]),
+    srcs=['common/threading.c'],
+    exported_preprocessor_flags=[
+        '-DZSTD_MULTITHREAD',
+    ],
+    exported_linker_flags=[
+        '-pthread',
+    ],
+)
+
+cxx_library(
+    name='xxhash',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'xxhash.h'),
+    ]),
+    srcs=['common/xxhash.c'],
+    exported_preprocessor_flags=[
+        '-DXXH_NAMESPACE=ZSTD_',
+    ],
+)
+
+cxx_library(
+    name='zstd_common',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('', 'zstd.h'),
+        ('common', 'zstd_internal.h'),
+    ]),
+    srcs=['common/zstd_common.c'],
+    deps=[
+        ':compiler',
+        ':errors',
+        ':mem',
+    ],
+)
+
+cxx_library(
+    name='debug',
+    header_namespace='',
+    visibility=['PUBLIC'],
+    exported_headers=subdir_glob([
+        ('common', 'debug.h'),
+    ]),
+    srcs=['common/debug.c'],
+)
+
+cxx_library(
+    name='common',
+    deps=[
+        ':debug',
+        ':bitstream',
+        ':compiler',
+        ':cpu',
+        ':entropy',
+        ':errors',
+        ':mem',
+        ':pool',
+        ':threading',
+        ':xxhash',
+        ':zstd_common',
+    ]
+)
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/CMakeLists.txt b/lib/clickhouse-cpp/contrib/zstd/zstd/CMakeLists.txt
new file mode 100644
index 0000000..276f4d3
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/CMakeLists.txt
@@ -0,0 +1,115 @@
+enable_language(ASM)
+
+SET(Sources
+        common/debug.c
+        common/entropy_common.c
+        common/error_private.c
+        common/fse_decompress.c
+        common/pool.c
+        common/threading.c
+        common/xxhash.c
+        common/zstd_common.c
+        compress/fse_compress.c
+        compress/hist.c
+        compress/huf_compress.c
+        compress/zstd_compress.c
+        compress/zstd_compress_literals.c
+        compress/zstd_compress_sequences.c
+        compress/zstd_compress_superblock.c
+        compress/zstd_double_fast.c
+        compress/zstd_fast.c
+        compress/zstd_lazy.c
+        compress/zstd_ldm.c
+        compress/zstdmt_compress.c
+        compress/zstd_opt.c
+        decompress/huf_decompress_amd64.S
+        decompress/huf_decompress.c
+        decompress/zstd_ddict.c
+        decompress/zstd_decompress_block.c
+        decompress/zstd_decompress.c
+        dictBuilder/cover.c
+        dictBuilder/divsufsort.c
+        dictBuilder/fastcover.c
+        dictBuilder/zdict.c
+)
+
+SET(Headers
+        common/bits.h
+        common/bitstream.h
+        common/compiler.h
+        common/cpu.h
+        common/debug.h
+        common/error_private.h
+        common/fse.h
+        common/huf.h
+        common/mem.h
+        common/pool.h
+        common/portability_macros.h
+        common/threading.h
+        common/xxhash.h
+        common/zstd_deps.h
+        common/zstd_internal.h
+        common/zstd_trace.h
+        compress/clevels.h
+        compress/hist.h
+        compress/zstd_compress_internal.h
+        compress/zstd_compress_literals.h
+        compress/zstd_compress_sequences.h
+        compress/zstd_compress_superblock.h
+        compress/zstd_cwksp.h
+        compress/zstd_double_fast.h
+        compress/zstd_fast.h
+        compress/zstd_lazy.h
+        compress/zstd_ldm_geartab.h
+        compress/zstd_ldm.h
+        compress/zstdmt_compress.h
+        compress/zstd_opt.h
+        decompress/zstd_ddict.h
+        decompress/zstd_decompress_block.h
+        decompress/zstd_decompress_internal.h
+        dictBuilder/cover.h
+        dictBuilder/divsufsort.h
+        zdict.h
+        zstd_errors.h
+        zstd.h
+)
+
+SET(ZSTD_LEGACY_SUPPORT true)
+
+IF (ZSTD_LEGACY_SUPPORT)
+    SET(LIBRARY_LEGACY_DIR "${LIBRARY_DIR}/legacy")
+    INCLUDE_DIRECTORIES(BEFORE ${LIBRARY_LEGACY_DIR})
+    ADD_DEFINITIONS(-D ZSTD_LEGACY_SUPPORT=1)
+
+    SET(Sources ${Sources}
+            legacy/zstd_v01.c
+            legacy/zstd_v02.c
+            legacy/zstd_v03.c
+            legacy/zstd_v04.c
+            legacy/zstd_v05.c
+            legacy/zstd_v06.c
+            legacy/zstd_v07.c
+    )
+
+    SET(Headers ${Headers}
+            legacy/zstd_legacy.h
+            legacy/zstd_v01.h
+            legacy/zstd_v02.h
+            legacy/zstd_v03.h
+            legacy/zstd_v04.h
+            legacy/zstd_v05.h
+            legacy/zstd_v06.h
+            legacy/zstd_v07.h
+    )
+
+ENDIF (ZSTD_LEGACY_SUPPORT)
+
+ADD_LIBRARY (zstdstatic STATIC ${Sources} ${Headers})
+
+SET_PROPERTY(TARGET zstdstatic PROPERTY POSITION_INDEPENDENT_CODE ON)
+
+TARGET_INCLUDE_DIRECTORIES (zstdstatic
+    PUBLIC ${PROJECT_SOURCE_DIR}/contrib/zstd
+)
+
+ADD_LIBRARY(zstd::zstd ALIAS zstdstatic)
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/LICENSE b/lib/clickhouse-cpp/contrib/zstd/zstd/LICENSE
new file mode 100644
index 0000000..7580028
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/LICENSE
@@ -0,0 +1,30 @@
+BSD License
+
+For Zstandard software
+
+Copyright (c) Meta Platforms, Inc. and affiliates. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification,
+are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+ * Neither the name Facebook, nor Meta, nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/allocations.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/allocations.h
new file mode 100644
index 0000000..a3153c4
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/allocations.h
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* This file provides custom allocation primitives
+ */
+
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"   /* ZSTD_malloc, ZSTD_calloc, ZSTD_free, ZSTD_memset */
+
+#include "mem.h" /* MEM_STATIC */
+#define ZSTD_STATIC_LINKING_ONLY
+#include "../zstd.h" /* ZSTD_customMem */
+
+#ifndef ZSTD_ALLOCATIONS_H
+#define ZSTD_ALLOCATIONS_H
+
+/* custom memory allocation functions */
+
+MEM_STATIC void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem)
+{
+    if (customMem.customAlloc)
+        return customMem.customAlloc(customMem.opaque, size);
+    return ZSTD_malloc(size);
+}
+
+MEM_STATIC void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem)
+{
+    if (customMem.customAlloc) {
+        /* calloc implemented as malloc+memset;
+         * not as efficient as calloc, but next best guess for custom malloc */
+        void* const ptr = customMem.customAlloc(customMem.opaque, size);
+        ZSTD_memset(ptr, 0, size);
+        return ptr;
+    }
+    return ZSTD_calloc(1, size);
+}
+
+MEM_STATIC void ZSTD_customFree(void* ptr, ZSTD_customMem customMem)
+{
+    if (ptr!=NULL) {
+        if (customMem.customFree)
+            customMem.customFree(customMem.opaque, ptr);
+        else
+            ZSTD_free(ptr);
+    }
+}
+
+#endif /* ZSTD_ALLOCATIONS_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/bits.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/bits.h
new file mode 100644
index 0000000..def56c4
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/bits.h
@@ -0,0 +1,200 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_BITS_H
+#define ZSTD_BITS_H
+
+#include "mem.h"
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros32_fallback(U32 val)
+{
+    assert(val != 0);
+    {
+        static const U32 DeBruijnBytePos[32] = {0, 1, 28, 2, 29, 14, 24, 3,
+                                                30, 22, 20, 15, 25, 17, 4, 8,
+                                                31, 27, 13, 23, 21, 19, 16, 7,
+                                                26, 12, 18, 6, 11, 5, 10, 9};
+        return DeBruijnBytePos[((U32) ((val & -(S32) val) * 0x077CB531U)) >> 27];
+    }
+}
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros32(U32 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER)
+#       if STATIC_BMI2 == 1
+            return (unsigned)_tzcnt_u32(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanForward(&r, val);
+                return (unsigned)r;
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4)
+        return (unsigned)__builtin_ctz(val);
+#   else
+        return ZSTD_countTrailingZeros32_fallback(val);
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros32_fallback(U32 val) {
+    assert(val != 0);
+    {
+        static const U32 DeBruijnClz[32] = {0, 9, 1, 10, 13, 21, 2, 29,
+                                            11, 14, 16, 18, 22, 25, 3, 30,
+                                            8, 12, 20, 28, 15, 17, 24, 7,
+                                            19, 27, 23, 6, 26, 5, 4, 31};
+        val |= val >> 1;
+        val |= val >> 2;
+        val |= val >> 4;
+        val |= val >> 8;
+        val |= val >> 16;
+        return 31 - DeBruijnClz[(val * 0x07C4ACDDU) >> 27];
+    }
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros32(U32 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER)
+#       if STATIC_BMI2 == 1
+            return (unsigned)_lzcnt_u32(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanReverse(&r, val);
+                return (unsigned)(31 - r);
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4)
+        return (unsigned)__builtin_clz(val);
+#   else
+        return ZSTD_countLeadingZeros32_fallback(val);
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros64(U64 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER) && defined(_WIN64)
+#       if STATIC_BMI2 == 1
+            return (unsigned)_tzcnt_u64(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanForward64(&r, val);
+                return (unsigned)r;
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4) && defined(__LP64__)
+        return (unsigned)__builtin_ctzll(val);
+#   else
+        {
+            U32 mostSignificantWord = (U32)(val >> 32);
+            U32 leastSignificantWord = (U32)val;
+            if (leastSignificantWord == 0) {
+                return 32 + ZSTD_countTrailingZeros32(mostSignificantWord);
+            } else {
+                return ZSTD_countTrailingZeros32(leastSignificantWord);
+            }
+        }
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros64(U64 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER) && defined(_WIN64)
+#       if STATIC_BMI2 == 1
+            return (unsigned)_lzcnt_u64(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanReverse64(&r, val);
+                return (unsigned)(63 - r);
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4)
+        return (unsigned)(__builtin_clzll(val));
+#   else
+        {
+            U32 mostSignificantWord = (U32)(val >> 32);
+            U32 leastSignificantWord = (U32)val;
+            if (mostSignificantWord == 0) {
+                return 32 + ZSTD_countLeadingZeros32(leastSignificantWord);
+            } else {
+                return ZSTD_countLeadingZeros32(mostSignificantWord);
+            }
+        }
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_NbCommonBytes(size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+            return ZSTD_countTrailingZeros64((U64)val) >> 3;
+        } else {
+            return ZSTD_countTrailingZeros32((U32)val) >> 3;
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+            return ZSTD_countLeadingZeros64((U64)val) >> 3;
+        } else {
+            return ZSTD_countLeadingZeros32((U32)val) >> 3;
+        }
+    }
+}
+
+MEM_STATIC unsigned ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
+{
+    assert(val != 0);
+    return 31 - ZSTD_countLeadingZeros32(val);
+}
+
+/* ZSTD_rotateRight_*():
+ * Rotates a bitfield to the right by "count" bits.
+ * https://en.wikipedia.org/w/index.php?title=Circular_shift&oldid=991635599#Implementing_circular_shifts
+ */
+MEM_STATIC
+U64 ZSTD_rotateRight_U64(U64 const value, U32 count) {
+    assert(count < 64);
+    count &= 0x3F; /* for fickle pattern recognition */
+    return (value >> count) | (U64)(value << ((0U - count) & 0x3F));
+}
+
+MEM_STATIC
+U32 ZSTD_rotateRight_U32(U32 const value, U32 count) {
+    assert(count < 32);
+    count &= 0x1F; /* for fickle pattern recognition */
+    return (value >> count) | (U32)(value << ((0U - count) & 0x1F));
+}
+
+MEM_STATIC
+U16 ZSTD_rotateRight_U16(U16 const value, U32 count) {
+    assert(count < 16);
+    count &= 0x0F; /* for fickle pattern recognition */
+    return (value >> count) | (U16)(value << ((0U - count) & 0x0F));
+}
+
+#endif /* ZSTD_BITS_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/bitstream.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/bitstream.h
new file mode 100644
index 0000000..72b0b3d
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/bitstream.h
@@ -0,0 +1,437 @@
+/* ******************************************************************
+ * bitstream
+ * Part of FSE library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include "mem.h"            /* unaligned access routines */
+#include "compiler.h"       /* UNLIKELY() */
+#include "debug.h"          /* assert(), DEBUGLOG(), RAWLOG() */
+#include "error_private.h"  /* error codes and messages */
+#include "bits.h"           /* ZSTD_highbit32 */
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#ifndef ZSTD_NO_INTRINSICS
+#  if (defined(__BMI__) || defined(__BMI2__)) && defined(__GNUC__)
+#    include <immintrin.h>   /* support for bextr (experimental)/bzhi */
+#  elif defined(__ICCARM__)
+#    include <intrinsics.h>
+#  endif
+#endif
+
+#define STREAM_ACCUMULATOR_MIN_32  25
+#define STREAM_ACCUMULATOR_MIN_64  57
+#define STREAM_ACCUMULATOR_MIN    ((U32)(MEM_32bits() ? STREAM_ACCUMULATOR_MIN_32 : STREAM_ACCUMULATOR_MIN_64))
+
+
+/*-******************************************
+*  bitStream encoding API (write forward)
+********************************************/
+/* bitStream can mix input from multiple sources.
+ * A critical property of these streams is that they encode and decode in **reverse** direction.
+ * So the first bit sequence you add will be the last to be read, like a LIFO stack.
+ */
+typedef struct {
+    size_t bitContainer;
+    unsigned bitPos;
+    char*  startPtr;
+    char*  ptr;
+    char*  endPtr;
+} BIT_CStream_t;
+
+MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC, void* dstBuffer, size_t dstCapacity);
+MEM_STATIC void   BIT_addBits(BIT_CStream_t* bitC, size_t value, unsigned nbBits);
+MEM_STATIC void   BIT_flushBits(BIT_CStream_t* bitC);
+MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC);
+
+/* Start with initCStream, providing the size of buffer to write into.
+*  bitStream will never write outside of this buffer.
+*  `dstCapacity` must be >= sizeof(bitD->bitContainer), otherwise @return will be an error code.
+*
+*  bits are first added to a local register.
+*  Local register is size_t, hence 64-bits on 64-bits systems, or 32-bits on 32-bits systems.
+*  Writing data into memory is an explicit operation, performed by the flushBits function.
+*  Hence keep track how many bits are potentially stored into local register to avoid register overflow.
+*  After a flushBits, a maximum of 7 bits might still be stored into local register.
+*
+*  Avoid storing elements of more than 24 bits if you want compatibility with 32-bits bitstream readers.
+*
+*  Last operation is to close the bitStream.
+*  The function returns the final size of CStream in bytes.
+*  If data couldn't fit into `dstBuffer`, it will return a 0 ( == not storable)
+*/
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct {
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+    const char* limitPtr;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+/* Start by invoking BIT_initDStream().
+*  A chunk of the bitStream is then stored into a local register.
+*  Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).
+*  You can then retrieve bitFields stored into the local register, **in reverse order**.
+*  Local register is explicitly reloaded from memory by the BIT_reloadDStream() method.
+*  A reload guarantee a minimum of ((8*sizeof(bitD->bitContainer))-7) bits when its result is BIT_DStream_unfinished.
+*  Otherwise, it can be less than that, so proceed accordingly.
+*  Checking if DStream has reached its end can be performed with BIT_endOfDStream().
+*/
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC, size_t value, unsigned nbBits);
+/* faster, but works only if value is "clean", meaning all high bits above nbBits are 0 */
+
+MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC);
+/* unsafe version; does not check buffer overflow */
+
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+/*=====    Local Constants   =====*/
+static const unsigned BIT_mask[] = {
+    0,          1,         3,         7,         0xF,       0x1F,
+    0x3F,       0x7F,      0xFF,      0x1FF,     0x3FF,     0x7FF,
+    0xFFF,      0x1FFF,    0x3FFF,    0x7FFF,    0xFFFF,    0x1FFFF,
+    0x3FFFF,    0x7FFFF,   0xFFFFF,   0x1FFFFF,  0x3FFFFF,  0x7FFFFF,
+    0xFFFFFF,   0x1FFFFFF, 0x3FFFFFF, 0x7FFFFFF, 0xFFFFFFF, 0x1FFFFFFF,
+    0x3FFFFFFF, 0x7FFFFFFF}; /* up to 31 bits */
+#define BIT_MASK_SIZE (sizeof(BIT_mask) / sizeof(BIT_mask[0]))
+
+/*-**************************************************************
+*  bitStream encoding
+****************************************************************/
+/*! BIT_initCStream() :
+ *  `dstCapacity` must be > sizeof(size_t)
+ *  @return : 0 if success,
+ *            otherwise an error code (can be tested using ERR_isError()) */
+MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
+                                  void* startPtr, size_t dstCapacity)
+{
+    bitC->bitContainer = 0;
+    bitC->bitPos = 0;
+    bitC->startPtr = (char*)startPtr;
+    bitC->ptr = bitC->startPtr;
+    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer);
+    if (dstCapacity <= sizeof(bitC->bitContainer)) return ERROR(dstSize_tooSmall);
+    return 0;
+}
+
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
+{
+#if defined(STATIC_BMI2) && STATIC_BMI2 == 1 && !defined(ZSTD_NO_INTRINSICS)
+    return  _bzhi_u64(bitContainer, nbBits);
+#else
+    assert(nbBits < BIT_MASK_SIZE);
+    return bitContainer & BIT_mask[nbBits];
+#endif
+}
+
+/*! BIT_addBits() :
+ *  can add up to 31 bits into `bitC`.
+ *  Note : does not check for register overflow ! */
+MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
+                            size_t value, unsigned nbBits)
+{
+    DEBUG_STATIC_ASSERT(BIT_MASK_SIZE == 32);
+    assert(nbBits < BIT_MASK_SIZE);
+    assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    bitC->bitContainer |= BIT_getLowerBits(value, nbBits) << bitC->bitPos;
+    bitC->bitPos += nbBits;
+}
+
+/*! BIT_addBitsFast() :
+ *  works only if `value` is _clean_,
+ *  meaning all high bits above nbBits are 0 */
+MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC,
+                                size_t value, unsigned nbBits)
+{
+    assert((value>>nbBits) == 0);
+    assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    bitC->bitContainer |= value << bitC->bitPos;
+    bitC->bitPos += nbBits;
+}
+
+/*! BIT_flushBitsFast() :
+ *  assumption : bitContainer has not overflowed
+ *  unsafe version; does not check buffer overflow */
+MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC)
+{
+    size_t const nbBytes = bitC->bitPos >> 3;
+    assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    assert(bitC->ptr <= bitC->endPtr);
+    MEM_writeLEST(bitC->ptr, bitC->bitContainer);
+    bitC->ptr += nbBytes;
+    bitC->bitPos &= 7;
+    bitC->bitContainer >>= nbBytes*8;
+}
+
+/*! BIT_flushBits() :
+ *  assumption : bitContainer has not overflowed
+ *  safe version; check for buffer overflow, and prevents it.
+ *  note : does not signal buffer overflow.
+ *  overflow will be revealed later on using BIT_closeCStream() */
+MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC)
+{
+    size_t const nbBytes = bitC->bitPos >> 3;
+    assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    assert(bitC->ptr <= bitC->endPtr);
+    MEM_writeLEST(bitC->ptr, bitC->bitContainer);
+    bitC->ptr += nbBytes;
+    if (bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
+    bitC->bitPos &= 7;
+    bitC->bitContainer >>= nbBytes*8;
+}
+
+/*! BIT_closeCStream() :
+ *  @return : size of CStream, in bytes,
+ *            or 0 if it could not fit into dstBuffer */
+MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
+{
+    BIT_addBitsFast(bitC, 1, 1);   /* endMark */
+    BIT_flushBits(bitC);
+    if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
+    return (bitC->ptr - bitC->startPtr) + (bitC->bitPos > 0);
+}
+
+
+/*-********************************************************
+*  bitStream decoding
+**********************************************************/
+/*! BIT_initDStream() :
+ *  Initialize a BIT_DStream_t.
+ * `bitD` : a pointer to an already allocated BIT_DStream_t structure.
+ * `srcSize` must be the *exact* size of the bitStream, in bytes.
+ * @return : size of stream (== srcSize), or an errorCode if a problem is detected
+ */
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { ZSTD_memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    bitD->start = (const char*)srcBuffer;
+    bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+    } else {
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+        case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
+                ZSTD_FALLTHROUGH;
+
+        case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
+                ZSTD_FALLTHROUGH;
+
+        case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
+                ZSTD_FALLTHROUGH;
+
+        case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
+                ZSTD_FALLTHROUGH;
+
+        case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
+                ZSTD_FALLTHROUGH;
+
+        case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
+                ZSTD_FALLTHROUGH;
+
+        default: break;
+        }
+        {   BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+            bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
+            if (lastByte == 0) return ERROR(corruption_detected);  /* endMark not present */
+        }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
+{
+    return bitContainer >> start;
+}
+
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
+{
+    U32 const regMask = sizeof(bitContainer)*8 - 1;
+    /* if start > regMask, bitstream is corrupted, and result is undefined */
+    assert(nbBits < BIT_MASK_SIZE);
+    /* x86 transform & ((1 << nbBits) - 1) to bzhi instruction, it is better
+     * than accessing memory. When bmi2 instruction is not present, we consider
+     * such cpus old (pre-Haswell, 2013) and their performance is not of that
+     * importance.
+     */
+#if defined(__x86_64__) || defined(_M_X86)
+    return (bitContainer >> (start & regMask)) & ((((U64)1) << nbBits) - 1);
+#else
+    return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
+#endif
+}
+
+/*! BIT_lookBits() :
+ *  Provides next n bits from local register.
+ *  local register is not modified.
+ *  On 32-bits, maxNbBits==24.
+ *  On 64-bits, maxNbBits==56.
+ * @return : value extracted */
+MEM_STATIC  FORCE_INLINE_ATTR size_t BIT_lookBits(const BIT_DStream_t*  bitD, U32 nbBits)
+{
+    /* arbitrate between double-shift and shift+mask */
+#if 1
+    /* if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8,
+     * bitstream is likely corrupted, and result is undefined */
+    return BIT_getMiddleBits(bitD->bitContainer, (sizeof(bitD->bitContainer)*8) - bitD->bitsConsumed - nbBits, nbBits);
+#else
+    /* this code path is slower on my os-x laptop */
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> 1) >> ((regMask-nbBits) & regMask);
+#endif
+}
+
+/*! BIT_lookBitsFast() :
+ *  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
+{
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    assert(nbBits >= 1);
+    return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
+}
+
+MEM_STATIC FORCE_INLINE_ATTR void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+/*! BIT_readBits() :
+ *  Read (consume) next n bits from local register and update.
+ *  Pay attention to not read more than nbBits contained into local register.
+ * @return : extracted value. */
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
+{
+    size_t const value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BIT_readBitsFast() :
+ *  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits)
+{
+    size_t const value = BIT_lookBitsFast(bitD, nbBits);
+    assert(nbBits >= 1);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BIT_reloadDStreamFast() :
+ *  Similar to BIT_reloadDStream(), but with two differences:
+ *  1. bitsConsumed <= sizeof(bitD->bitContainer)*8 must hold!
+ *  2. Returns BIT_DStream_overflow when bitD->ptr < bitD->limitPtr, at this
+ *     point you must use BIT_reloadDStream() to reload.
+ */
+MEM_STATIC BIT_DStream_status BIT_reloadDStreamFast(BIT_DStream_t* bitD)
+{
+    if (UNLIKELY(bitD->ptr < bitD->limitPtr))
+        return BIT_DStream_overflow;
+    assert(bitD->bitsConsumed <= sizeof(bitD->bitContainer)*8);
+    bitD->ptr -= bitD->bitsConsumed >> 3;
+    bitD->bitsConsumed &= 7;
+    bitD->bitContainer = MEM_readLEST(bitD->ptr);
+    return BIT_DStream_unfinished;
+}
+
+/*! BIT_reloadDStream() :
+ *  Refill `bitD` from buffer previously set in BIT_initDStream() .
+ *  This function is safe, it guarantees it will not read beyond src buffer.
+ * @return : status of `BIT_DStream_t` internal register.
+ *           when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
+MEM_STATIC FORCE_INLINE_ATTR BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* overflow detected, like end of stream */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->limitPtr) {
+        return BIT_reloadDStreamFast(bitD);
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    /* start < ptr < limitPtr */
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD->bitContainer), otherwise bitD->ptr == bitD->start */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream() :
+ * @return : 1 if DStream has _exactly_ reached its end (all bits consumed).
+ */
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/compiler.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/compiler.h
new file mode 100644
index 0000000..73f8d01
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/compiler.h
@@ -0,0 +1,358 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPILER_H
+#define ZSTD_COMPILER_H
+
+#include "portability_macros.h"
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+/* force inlining */
+
+#if !defined(ZSTD_NO_INLINE)
+#if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#  define INLINE_KEYWORD inline
+#else
+#  define INLINE_KEYWORD
+#endif
+
+#if defined(__GNUC__) || defined(__ICCARM__)
+#  define FORCE_INLINE_ATTR __attribute__((always_inline))
+#elif defined(_MSC_VER)
+#  define FORCE_INLINE_ATTR __forceinline
+#else
+#  define FORCE_INLINE_ATTR
+#endif
+
+#else
+
+#define INLINE_KEYWORD
+#define FORCE_INLINE_ATTR
+
+#endif
+
+/**
+  On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
+  This explicitly marks such functions as __cdecl so that the code will still compile
+  if a CC other than __cdecl has been made the default.
+*/
+#if  defined(_MSC_VER)
+#  define WIN_CDECL __cdecl
+#else
+#  define WIN_CDECL
+#endif
+
+/**
+ * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
+ * parameters. They must be inlined for the compiler to eliminate the constant
+ * branches.
+ */
+#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
+/**
+ * HINT_INLINE is used to help the compiler generate better code. It is *not*
+ * used for "templates", so it can be tweaked based on the compilers
+ * performance.
+ *
+ * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the
+ * always_inline attribute.
+ *
+ * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline
+ * attribute.
+ */
+#if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5
+#  define HINT_INLINE static INLINE_KEYWORD
+#else
+#  define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR
+#endif
+
+/* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
+#if defined(__GNUC__)
+#  define UNUSED_ATTR __attribute__((unused))
+#else
+#  define UNUSED_ATTR
+#endif
+
+/* force no inlining */
+#ifdef _MSC_VER
+#  define FORCE_NOINLINE static __declspec(noinline)
+#else
+#  if defined(__GNUC__) || defined(__ICCARM__)
+#    define FORCE_NOINLINE static __attribute__((__noinline__))
+#  else
+#    define FORCE_NOINLINE static
+#  endif
+#endif
+
+
+/* target attribute */
+#if defined(__GNUC__) || defined(__ICCARM__)
+#  define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
+#else
+#  define TARGET_ATTRIBUTE(target)
+#endif
+
+/* Target attribute for BMI2 dynamic dispatch.
+ * Enable lzcnt, bmi, and bmi2.
+ * We test for bmi1 & bmi2. lzcnt is included in bmi1.
+ */
+#define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE("lzcnt,bmi,bmi2")
+
+/* prefetch
+ * can be disabled, by declaring NO_PREFETCH build macro */
+#if defined(NO_PREFETCH)
+#  define PREFETCH_L1(ptr)  (void)(ptr)  /* disabled */
+#  define PREFETCH_L2(ptr)  (void)(ptr)  /* disabled */
+#else
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
+#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#    define PREFETCH_L1(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
+#    define PREFETCH_L2(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
+#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
+#    define PREFETCH_L1(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
+#    define PREFETCH_L2(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
+#  elif defined(__aarch64__)
+#    define PREFETCH_L1(ptr)  __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
+#    define PREFETCH_L2(ptr)  __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
+#  else
+#    define PREFETCH_L1(ptr) (void)(ptr)  /* disabled */
+#    define PREFETCH_L2(ptr) (void)(ptr)  /* disabled */
+#  endif
+#endif  /* NO_PREFETCH */
+
+#define CACHELINE_SIZE 64
+
+#define PREFETCH_AREA(p, s)  {            \
+    const char* const _ptr = (const char*)(p);  \
+    size_t const _size = (size_t)(s);     \
+    size_t _pos;                          \
+    for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) {  \
+        PREFETCH_L2(_ptr + _pos);         \
+    }                                     \
+}
+
+/* vectorization
+ * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax,
+ * and some compilers, like Intel ICC and MCST LCC, do not support it at all. */
+#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) && !defined(__LCC__)
+#  if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5)
+#    define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
+#  else
+#    define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")")
+#  endif
+#else
+#  define DONT_VECTORIZE
+#endif
+
+/* Tell the compiler that a branch is likely or unlikely.
+ * Only use these macros if it causes the compiler to generate better code.
+ * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc
+ * and clang, please do.
+ */
+#if defined(__GNUC__)
+#define LIKELY(x) (__builtin_expect((x), 1))
+#define UNLIKELY(x) (__builtin_expect((x), 0))
+#else
+#define LIKELY(x) (x)
+#define UNLIKELY(x) (x)
+#endif
+
+#if __has_builtin(__builtin_unreachable) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5)))
+#  define ZSTD_UNREACHABLE { assert(0), __builtin_unreachable(); }
+#else
+#  define ZSTD_UNREACHABLE { assert(0); }
+#endif
+
+/* disable warnings */
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+/*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/
+#ifndef STATIC_BMI2
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))
+#    ifdef __AVX2__  //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2
+#       define STATIC_BMI2 1
+#    endif
+#  elif defined(__BMI2__) && defined(__x86_64__) && defined(__GNUC__)
+#    define STATIC_BMI2 1
+#  endif
+#endif
+
+#ifndef STATIC_BMI2
+    #define STATIC_BMI2 0
+#endif
+
+/* compile time determination of SIMD support */
+#if !defined(ZSTD_NO_INTRINSICS)
+#  if defined(__SSE2__) || defined(_M_AMD64) || (defined (_M_IX86) && defined(_M_IX86_FP) && (_M_IX86_FP >= 2))
+#    define ZSTD_ARCH_X86_SSE2
+#  endif
+#  if defined(__ARM_NEON) || defined(_M_ARM64)
+#    define ZSTD_ARCH_ARM_NEON
+#  endif
+#
+#  if defined(ZSTD_ARCH_X86_SSE2)
+#    include <emmintrin.h>
+#  elif defined(ZSTD_ARCH_ARM_NEON)
+#    include <arm_neon.h>
+#  endif
+#endif
+
+/* C-language Attributes are added in C23. */
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
+# define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
+#else
+# define ZSTD_HAS_C_ATTRIBUTE(x) 0
+#endif
+
+/* Only use C++ attributes in C++. Some compilers report support for C++
+ * attributes when compiling with C.
+ */
+#if defined(__cplusplus) && defined(__has_cpp_attribute)
+# define ZSTD_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+#else
+# define ZSTD_HAS_CPP_ATTRIBUTE(x) 0
+#endif
+
+/* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute.
+ * - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough
+ * - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough
+ * - Else: __attribute__((__fallthrough__))
+ */
+#ifndef ZSTD_FALLTHROUGH
+# if ZSTD_HAS_C_ATTRIBUTE(fallthrough)
+#  define ZSTD_FALLTHROUGH [[fallthrough]]
+# elif ZSTD_HAS_CPP_ATTRIBUTE(fallthrough)
+#  define ZSTD_FALLTHROUGH [[fallthrough]]
+# elif __has_attribute(__fallthrough__)
+/* Leading semicolon is to satisfy gcc-11 with -pedantic. Without the semicolon
+ * gcc complains about: a label can only be part of a statement and a declaration is not a statement.
+ */
+#  define ZSTD_FALLTHROUGH ; __attribute__((__fallthrough__))
+# else
+#  define ZSTD_FALLTHROUGH
+# endif
+#endif
+
+/*-**************************************************************
+*  Alignment check
+*****************************************************************/
+
+/* this test was initially positioned in mem.h,
+ * but this file is removed (or replaced) for linux kernel
+ * so it's now hosted in compiler.h,
+ * which remains valid for both user & kernel spaces.
+ */
+
+#ifndef ZSTD_ALIGNOF
+# if defined(__GNUC__) || defined(_MSC_VER)
+/* covers gcc, clang & MSVC */
+/* note : this section must come first, before C11,
+ * due to a limitation in the kernel source generator */
+#  define ZSTD_ALIGNOF(T) __alignof(T)
+
+# elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)
+/* C11 support */
+#  include <stdalign.h>
+#  define ZSTD_ALIGNOF(T) alignof(T)
+
+# else
+/* No known support for alignof() - imperfect backup */
+#  define ZSTD_ALIGNOF(T) (sizeof(void*) < sizeof(T) ? sizeof(void*) : sizeof(T))
+
+# endif
+#endif /* ZSTD_ALIGNOF */
+
+/*-**************************************************************
+*  Sanitizer
+*****************************************************************/
+
+/* Issue #3240 reports an ASAN failure on an llvm-mingw build. Out of an
+ * abundance of caution, disable our custom poisoning on mingw. */
+#ifdef __MINGW32__
+#ifndef ZSTD_ASAN_DONT_POISON_WORKSPACE
+#define ZSTD_ASAN_DONT_POISON_WORKSPACE 1
+#endif
+#ifndef ZSTD_MSAN_DONT_POISON_WORKSPACE
+#define ZSTD_MSAN_DONT_POISON_WORKSPACE 1
+#endif
+#endif
+
+#if ZSTD_MEMORY_SANITIZER && !defined(ZSTD_MSAN_DONT_POISON_WORKSPACE)
+/* Not all platforms that support msan provide sanitizers/msan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+#include <stddef.h>  /* size_t */
+#define ZSTD_DEPS_NEED_STDINT
+#include "zstd_deps.h"  /* intptr_t */
+
+/* Make memory region fully initialized (without changing its contents). */
+void __msan_unpoison(const volatile void *a, size_t size);
+
+/* Make memory region fully uninitialized (without changing its contents).
+   This is a legacy interface that does not update origin information. Use
+   __msan_allocated_memory() instead. */
+void __msan_poison(const volatile void *a, size_t size);
+
+/* Returns the offset of the first (at least partially) poisoned byte in the
+   memory range, or -1 if the whole range is good. */
+intptr_t __msan_test_shadow(const volatile void *x, size_t size);
+
+/* Print shadow and origin for the memory range to stderr in a human-readable
+   format. */
+void __msan_print_shadow(const volatile void *x, size_t size);
+#endif
+
+#if ZSTD_ADDRESS_SANITIZER && !defined(ZSTD_ASAN_DONT_POISON_WORKSPACE)
+/* Not all platforms that support asan provide sanitizers/asan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+#include <stddef.h>  /* size_t */
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
+ *
+ * This memory must be previously allocated by your program. Instrumented
+ * code is forbidden from accessing addresses in this region until it is
+ * unpoisoned. This function is not guaranteed to poison the entire region -
+ * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
+ * alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can poison or
+ * unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_poison_memory_region(void const volatile *addr, size_t size);
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
+ *
+ * This memory must be previously allocated by your program. Accessing
+ * addresses in this region is allowed until this region is poisoned again.
+ * This function could unpoison a super-region of <c>[addr, addr+size)</c> due
+ * to ASan alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can
+ * poison or unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
+#endif
+
+#endif /* ZSTD_COMPILER_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/cpu.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/cpu.h
new file mode 100644
index 0000000..8bc34a3
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/cpu.h
@@ -0,0 +1,213 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMMON_CPU_H
+#define ZSTD_COMMON_CPU_H
+
+/**
+ * Implementation taken from folly/CpuId.h
+ * https://github.com/facebook/folly/blob/master/folly/CpuId.h
+ */
+
+#include "mem.h"
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+
+typedef struct {
+    U32 f1c;
+    U32 f1d;
+    U32 f7b;
+    U32 f7c;
+} ZSTD_cpuid_t;
+
+MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
+    U32 f1c = 0;
+    U32 f1d = 0;
+    U32 f7b = 0;
+    U32 f7c = 0;
+#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
+    int reg[4];
+    __cpuid((int*)reg, 0);
+    {
+        int const n = reg[0];
+        if (n >= 1) {
+            __cpuid((int*)reg, 1);
+            f1c = (U32)reg[2];
+            f1d = (U32)reg[3];
+        }
+        if (n >= 7) {
+            __cpuidex((int*)reg, 7, 0);
+            f7b = (U32)reg[1];
+            f7c = (U32)reg[2];
+        }
+    }
+#elif defined(__i386__) && defined(__PIC__) && !defined(__clang__) && defined(__GNUC__)
+    /* The following block like the normal cpuid branch below, but gcc
+     * reserves ebx for use of its pic register so we must specially
+     * handle the save and restore to avoid clobbering the register
+     */
+    U32 n;
+    __asm__(
+        "pushl %%ebx\n\t"
+        "cpuid\n\t"
+        "popl %%ebx\n\t"
+        : "=a"(n)
+        : "a"(0)
+        : "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "popl %%ebx\n\t"
+          : "=a"(f1a), "=c"(f1c), "=d"(f1d)
+          : "a"(1));
+    }
+    if (n >= 7) {
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "movl %%ebx, %%eax\n\t"
+          "popl %%ebx"
+          : "=a"(f7b), "=c"(f7c)
+          : "a"(7), "c"(0)
+          : "edx");
+    }
+#elif defined(__x86_64__) || defined(_M_X64) || defined(__i386__)
+    U32 n;
+    __asm__("cpuid" : "=a"(n) : "a"(0) : "ebx", "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__("cpuid" : "=a"(f1a), "=c"(f1c), "=d"(f1d) : "a"(1) : "ebx");
+    }
+    if (n >= 7) {
+      U32 f7a;
+      __asm__("cpuid"
+              : "=a"(f7a), "=b"(f7b), "=c"(f7c)
+              : "a"(7), "c"(0)
+              : "edx");
+    }
+#endif
+    {
+        ZSTD_cpuid_t cpuid;
+        cpuid.f1c = f1c;
+        cpuid.f1d = f1d;
+        cpuid.f7b = f7b;
+        cpuid.f7c = f7c;
+        return cpuid;
+    }
+}
+
+#define X(name, r, bit)                                                        \
+  MEM_STATIC int ZSTD_cpuid_##name(ZSTD_cpuid_t const cpuid) {                 \
+    return ((cpuid.r) & (1U << bit)) != 0;                                     \
+  }
+
+/* cpuid(1): Processor Info and Feature Bits. */
+#define C(name, bit) X(name, f1c, bit)
+  C(sse3, 0)
+  C(pclmuldq, 1)
+  C(dtes64, 2)
+  C(monitor, 3)
+  C(dscpl, 4)
+  C(vmx, 5)
+  C(smx, 6)
+  C(eist, 7)
+  C(tm2, 8)
+  C(ssse3, 9)
+  C(cnxtid, 10)
+  C(fma, 12)
+  C(cx16, 13)
+  C(xtpr, 14)
+  C(pdcm, 15)
+  C(pcid, 17)
+  C(dca, 18)
+  C(sse41, 19)
+  C(sse42, 20)
+  C(x2apic, 21)
+  C(movbe, 22)
+  C(popcnt, 23)
+  C(tscdeadline, 24)
+  C(aes, 25)
+  C(xsave, 26)
+  C(osxsave, 27)
+  C(avx, 28)
+  C(f16c, 29)
+  C(rdrand, 30)
+#undef C
+#define D(name, bit) X(name, f1d, bit)
+  D(fpu, 0)
+  D(vme, 1)
+  D(de, 2)
+  D(pse, 3)
+  D(tsc, 4)
+  D(msr, 5)
+  D(pae, 6)
+  D(mce, 7)
+  D(cx8, 8)
+  D(apic, 9)
+  D(sep, 11)
+  D(mtrr, 12)
+  D(pge, 13)
+  D(mca, 14)
+  D(cmov, 15)
+  D(pat, 16)
+  D(pse36, 17)
+  D(psn, 18)
+  D(clfsh, 19)
+  D(ds, 21)
+  D(acpi, 22)
+  D(mmx, 23)
+  D(fxsr, 24)
+  D(sse, 25)
+  D(sse2, 26)
+  D(ss, 27)
+  D(htt, 28)
+  D(tm, 29)
+  D(pbe, 31)
+#undef D
+
+/* cpuid(7): Extended Features. */
+#define B(name, bit) X(name, f7b, bit)
+  B(bmi1, 3)
+  B(hle, 4)
+  B(avx2, 5)
+  B(smep, 7)
+  B(bmi2, 8)
+  B(erms, 9)
+  B(invpcid, 10)
+  B(rtm, 11)
+  B(mpx, 14)
+  B(avx512f, 16)
+  B(avx512dq, 17)
+  B(rdseed, 18)
+  B(adx, 19)
+  B(smap, 20)
+  B(avx512ifma, 21)
+  B(pcommit, 22)
+  B(clflushopt, 23)
+  B(clwb, 24)
+  B(avx512pf, 26)
+  B(avx512er, 27)
+  B(avx512cd, 28)
+  B(sha, 29)
+  B(avx512bw, 30)
+  B(avx512vl, 31)
+#undef B
+#define C(name, bit) X(name, f7c, bit)
+  C(prefetchwt1, 0)
+  C(avx512vbmi, 1)
+#undef C
+
+#undef X
+
+#endif /* ZSTD_COMMON_CPU_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/debug.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/debug.c
new file mode 100644
index 0000000..ebf7bfc
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/debug.c
@@ -0,0 +1,24 @@
+/* ******************************************************************
+ * debug
+ * Part of FSE library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+
+/*
+ * This module only hosts one global variable
+ * which can be used to dynamically influence the verbosity of traces,
+ * such as DEBUGLOG and RAWLOG
+ */
+
+#include "debug.h"
+
+int g_debuglevel = DEBUGLEVEL;
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/debug.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/debug.h
new file mode 100644
index 0000000..0e9817e
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/debug.h
@@ -0,0 +1,107 @@
+/* ******************************************************************
+ * debug
+ * Part of FSE library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+
+/*
+ * The purpose of this header is to enable debug functions.
+ * They regroup assert(), DEBUGLOG() and RAWLOG() for run-time,
+ * and DEBUG_STATIC_ASSERT() for compile-time.
+ *
+ * By default, DEBUGLEVEL==0, which means run-time debug is disabled.
+ *
+ * Level 1 enables assert() only.
+ * Starting level 2, traces can be generated and pushed to stderr.
+ * The higher the level, the more verbose the traces.
+ *
+ * It's possible to dynamically adjust level using variable g_debug_level,
+ * which is only declared if DEBUGLEVEL>=2,
+ * and is a global variable, not multi-thread protected (use with care)
+ */
+
+#ifndef DEBUG_H_12987983217
+#define DEBUG_H_12987983217
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* static assert is triggered at compile time, leaving no runtime artefact.
+ * static assert only works with compile-time constants.
+ * Also, this variant can only be used inside a function. */
+#define DEBUG_STATIC_ASSERT(c) (void)sizeof(char[(c) ? 1 : -1])
+
+
+/* DEBUGLEVEL is expected to be defined externally,
+ * typically through compiler command line.
+ * Value must be a number. */
+#ifndef DEBUGLEVEL
+#  define DEBUGLEVEL 0
+#endif
+
+
+/* recommended values for DEBUGLEVEL :
+ * 0 : release mode, no debug, all run-time checks disabled
+ * 1 : enables assert() only, no display
+ * 2 : reserved, for currently active debug path
+ * 3 : events once per object lifetime (CCtx, CDict, etc.)
+ * 4 : events once per frame
+ * 5 : events once per block
+ * 6 : events once per sequence (verbose)
+ * 7+: events at every position (*very* verbose)
+ *
+ * It's generally inconvenient to output traces > 5.
+ * In which case, it's possible to selectively trigger high verbosity levels
+ * by modifying g_debug_level.
+ */
+
+#if (DEBUGLEVEL>=1)
+#  define ZSTD_DEPS_NEED_ASSERT
+#  include "zstd_deps.h"
+#else
+#  ifndef assert   /* assert may be already defined, due to prior #include <assert.h> */
+#    define assert(condition) ((void)0)   /* disable assert (default) */
+#  endif
+#endif
+
+#if (DEBUGLEVEL>=2)
+#  define ZSTD_DEPS_NEED_IO
+#  include "zstd_deps.h"
+extern int g_debuglevel; /* the variable is only declared,
+                            it actually lives in debug.c,
+                            and is shared by the whole process.
+                            It's not thread-safe.
+                            It's useful when enabling very verbose levels
+                            on selective conditions (such as position in src) */
+
+#  define RAWLOG(l, ...) {                                       \
+                if (l<=g_debuglevel) {                           \
+                    ZSTD_DEBUG_PRINT(__VA_ARGS__);               \
+            }   }
+#  define DEBUGLOG(l, ...) {                                     \
+                if (l<=g_debuglevel) {                           \
+                    ZSTD_DEBUG_PRINT(__FILE__ ": " __VA_ARGS__); \
+                    ZSTD_DEBUG_PRINT(" \n");                     \
+            }   }
+#else
+#  define RAWLOG(l, ...)      {}    /* disabled */
+#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#endif
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* DEBUG_H_12987983217 */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/entropy_common.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/entropy_common.c
new file mode 100644
index 0000000..e2173af
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/entropy_common.c
@@ -0,0 +1,340 @@
+/* ******************************************************************
+ * Common functions of New Generation Entropy library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* *************************************
+*  Dependencies
+***************************************/
+#include "mem.h"
+#include "error_private.h"       /* ERR_*, ERROR */
+#define FSE_STATIC_LINKING_ONLY  /* FSE_MIN_TABLELOG */
+#include "fse.h"
+#include "huf.h"
+#include "bits.h"                /* ZSDT_highbit32, ZSTD_countTrailingZeros32 */
+
+
+/*===   Version   ===*/
+unsigned FSE_versionNumber(void) { return FSE_VERSION_NUMBER; }
+
+
+/*===   Error Management   ===*/
+unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+const char* FSE_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+FORCE_INLINE_TEMPLATE
+size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                           const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    unsigned const maxSV1 = *maxSVPtr + 1;
+    int previous0 = 0;
+
+    if (hbSize < 8) {
+        /* This function only works when hbSize >= 8 */
+        char buffer[8] = {0};
+        ZSTD_memcpy(buffer, headerBuffer, hbSize);
+        {   size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr,
+                                                    buffer, sizeof(buffer));
+            if (FSE_isError(countSize)) return countSize;
+            if (countSize > hbSize) return ERROR(corruption_detected);
+            return countSize;
+    }   }
+    assert(hbSize >= 8);
+
+    /* init */
+    ZSTD_memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0]));   /* all symbols not present in NCount have a frequency of 0 */
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    for (;;) {
+        if (previous0) {
+            /* Count the number of repeats. Each time the
+             * 2-bit repeat code is 0b11 there is another
+             * repeat.
+             * Avoid UB by setting the high bit to 1.
+             */
+            int repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
+            while (repeats >= 12) {
+                charnum += 3 * 12;
+                if (LIKELY(ip <= iend-7)) {
+                    ip += 3;
+                } else {
+                    bitCount -= (int)(8 * (iend - 7 - ip));
+                    bitCount &= 31;
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> bitCount;
+                repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
+            }
+            charnum += 3 * repeats;
+            bitStream >>= 2 * repeats;
+            bitCount += 2 * repeats;
+
+            /* Add the final repeat which isn't 0b11. */
+            assert((bitStream & 3) < 3);
+            charnum += bitStream & 3;
+            bitCount += 2;
+
+            /* This is an error, but break and return an error
+             * at the end, because returning out of a loop makes
+             * it harder for the compiler to optimize.
+             */
+            if (charnum >= maxSV1) break;
+
+            /* We don't need to set the normalized count to 0
+             * because we already memset the whole buffer to 0.
+             */
+
+            if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                assert((bitCount >> 3) <= 3); /* For first condition to work */
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                bitCount &= 31;
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> bitCount;
+        }
+        {
+            int const max = (2*threshold-1) - remaining;
+            int count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = bitStream & (threshold-1);
+                bitCount += nbBits-1;
+            } else {
+                count = bitStream & (2*threshold-1);
+                if (count >= threshold) count -= max;
+                bitCount += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            /* When it matters (small blocks), this is a
+             * predictable branch, because we don't use -1.
+             */
+            if (count >= 0) {
+                remaining -= count;
+            } else {
+                assert(count == -1);
+                remaining += count;
+            }
+            normalizedCounter[charnum++] = (short)count;
+            previous0 = !count;
+
+            assert(threshold > 1);
+            if (remaining < threshold) {
+                /* This branch can be folded into the
+                 * threshold update condition because we
+                 * know that threshold > 1.
+                 */
+                if (remaining <= 1) break;
+                nbBits = ZSTD_highbit32(remaining) + 1;
+                threshold = 1 << (nbBits - 1);
+            }
+            if (charnum >= maxSV1) break;
+
+            if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                bitCount &= 31;
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> bitCount;
+    }   }
+    if (remaining != 1) return ERROR(corruption_detected);
+    /* Only possible when there are too many zeros. */
+    if (charnum > maxSV1) return ERROR(maxSymbolValue_tooSmall);
+    if (bitCount > 32) return ERROR(corruption_detected);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    return ip-istart;
+}
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t FSE_readNCount_body_default(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+
+#if DYNAMIC_BMI2
+BMI2_TARGET_ATTRIBUTE static size_t FSE_readNCount_body_bmi2(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+#endif
+
+size_t FSE_readNCount_bmi2(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return FSE_readNCount_body_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+    }
+#endif
+    (void)bmi2;
+    return FSE_readNCount_body_default(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+
+size_t FSE_readNCount(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize, /* bmi2 */ 0);
+}
+
+
+/*! HUF_readStats() :
+    Read compact Huffman tree, saved by HUF_writeCTable().
+    `huffWeight` is destination buffer.
+    `rankStats` is assumed to be a table of at least HUF_TABLELOG_MAX U32.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUF_readCTable() and HUF_readDTableX?() .
+*/
+size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize)
+{
+    U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+    return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* flags */ 0);
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                   U32* nbSymbolsPtr, U32* tableLogPtr,
+                   const void* src, size_t srcSize,
+                   void* workSpace, size_t wkspSize,
+                   int bmi2)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* ZSTD_memset(huffWeight, 0, hwSize);   *//* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128) {  /* special header */
+        oSize = iSize - 127;
+        iSize = ((oSize+1)/2);
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        if (oSize >= hwSize) return ERROR(corruption_detected);
+        ip += 1;
+        {   U32 n;
+            for (n=0; n<oSize; n+=2) {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        /* max (hwSize-1) values decoded, as last one is implied */
+        oSize = FSE_decompress_wksp_bmi2(huffWeight, hwSize-1, ip+1, iSize, 6, workSpace, wkspSize, bmi2);
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = ZSTD_highbit32(weightTotal) + 1;
+        if (tableLog > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << ZSTD_highbit32(rest);
+            U32 const lastWeight = ZSTD_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize)
+{
+    return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 0);
+}
+
+#if DYNAMIC_BMI2
+static BMI2_TARGET_ATTRIBUTE size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize)
+{
+    return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 1);
+}
+#endif
+
+size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize,
+                     int flags)
+{
+#if DYNAMIC_BMI2
+    if (flags & HUF_flags_bmi2) {
+        return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
+    }
+#endif
+    (void)flags;
+    return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/error_private.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/error_private.c
new file mode 100644
index 0000000..075fc5e
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/error_private.c
@@ -0,0 +1,63 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* The purpose of this file is to have a single list of error strings embedded in binary */
+
+#include "error_private.h"
+
+const char* ERR_getErrorString(ERR_enum code)
+{
+#ifdef ZSTD_STRIP_ERROR_STRINGS
+    (void)code;
+    return "Error strings stripped";
+#else
+    static const char* const notErrorCode = "Unspecified error code";
+    switch( code )
+    {
+    case PREFIX(no_error): return "No error detected";
+    case PREFIX(GENERIC):  return "Error (generic)";
+    case PREFIX(prefix_unknown): return "Unknown frame descriptor";
+    case PREFIX(version_unsupported): return "Version not supported";
+    case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
+    case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
+    case PREFIX(corruption_detected): return "Data corruption detected";
+    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
+    case PREFIX(literals_headerWrong): return "Header of Literals' block doesn't respect format specification";
+    case PREFIX(parameter_unsupported): return "Unsupported parameter";
+    case PREFIX(parameter_combination_unsupported): return "Unsupported combination of parameters";
+    case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
+    case PREFIX(init_missing): return "Context should be init first";
+    case PREFIX(memory_allocation): return "Allocation error : not enough memory";
+    case PREFIX(workSpace_tooSmall): return "workSpace buffer is not large enough";
+    case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
+    case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
+    case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
+    case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
+    case PREFIX(stabilityCondition_notRespected): return "pledged buffer stability condition is not respected";
+    case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
+    case PREFIX(dictionary_wrong): return "Dictionary mismatch";
+    case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
+    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
+    case PREFIX(srcSize_wrong): return "Src size is incorrect";
+    case PREFIX(dstBuffer_null): return "Operation on NULL destination buffer";
+    case PREFIX(noForwardProgress_destFull): return "Operation made no progress over multiple calls, due to output buffer being full";
+    case PREFIX(noForwardProgress_inputEmpty): return "Operation made no progress over multiple calls, due to input being empty";
+        /* following error codes are not stable and may be removed or changed in a future version */
+    case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
+    case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
+    case PREFIX(dstBuffer_wrong): return "Destination buffer is wrong";
+    case PREFIX(srcBuffer_wrong): return "Source buffer is wrong";
+    case PREFIX(sequenceProducer_failed): return "Block-level external sequence producer returned an error code";
+    case PREFIX(externalSequences_invalid): return "External sequences are not valid";
+    case PREFIX(maxCode):
+    default: return notErrorCode;
+    }
+#endif
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/error_private.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/error_private.h
new file mode 100644
index 0000000..325daad
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/error_private.h
@@ -0,0 +1,159 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* Note : this module is expected to remain private, do not expose it */
+
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Dependencies
+******************************************/
+#include "../zstd_errors.h"  /* enum list */
+#include "compiler.h"
+#include "debug.h"
+#include "zstd_deps.h"       /* size_t */
+
+
+/* ****************************************
+*  Compiler-specific
+******************************************/
+#if defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-****************************************
+*  Customization (error_public.h)
+******************************************/
+typedef ZSTD_ErrorCode ERR_enum;
+#define PREFIX(name) ZSTD_error_##name
+
+
+/*-****************************************
+*  Error codes handling
+******************************************/
+#undef ERROR   /* already defined on Visual Studio */
+#define ERROR(name) ZSTD_ERROR(name)
+#define ZSTD_ERROR(name) ((size_t)-PREFIX(name))
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC ERR_enum ERR_getErrorCode(size_t code) { if (!ERR_isError(code)) return (ERR_enum)0; return (ERR_enum) (0-code); }
+
+/* check and forward error code */
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
+#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+
+
+/*-****************************************
+*  Error Strings
+******************************************/
+
+const char* ERR_getErrorString(ERR_enum code);   /* error_private.c */
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    return ERR_getErrorString(ERR_getErrorCode(code));
+}
+
+/**
+ * Ignore: this is an internal helper.
+ *
+ * This is a helper function to help force C99-correctness during compilation.
+ * Under strict compilation modes, variadic macro arguments can't be empty.
+ * However, variadic function arguments can be. Using a function therefore lets
+ * us statically check that at least one (string) argument was passed,
+ * independent of the compilation flags.
+ */
+static INLINE_KEYWORD UNUSED_ATTR
+void _force_has_format_string(const char *format, ...) {
+  (void)format;
+}
+
+/**
+ * Ignore: this is an internal helper.
+ *
+ * We want to force this function invocation to be syntactically correct, but
+ * we don't want to force runtime evaluation of its arguments.
+ */
+#define _FORCE_HAS_FORMAT_STRING(...) \
+  if (0) { \
+    _force_has_format_string(__VA_ARGS__); \
+  }
+
+#define ERR_QUOTE(str) #str
+
+/**
+ * Return the specified error if the condition evaluates to true.
+ *
+ * In debug modes, prints additional information.
+ * In order to do that (particularly, printing the conditional that failed),
+ * this can't just wrap RETURN_ERROR().
+ */
+#define RETURN_ERROR_IF(cond, err, ...) \
+  if (cond) { \
+    RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
+           __FILE__, __LINE__, ERR_QUOTE(cond), ERR_QUOTE(ERROR(err))); \
+    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+    RAWLOG(3, ": " __VA_ARGS__); \
+    RAWLOG(3, "\n"); \
+    return ERROR(err); \
+  }
+
+/**
+ * Unconditionally return the specified error.
+ *
+ * In debug modes, prints additional information.
+ */
+#define RETURN_ERROR(err, ...) \
+  do { \
+    RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
+           __FILE__, __LINE__, ERR_QUOTE(ERROR(err))); \
+    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+    RAWLOG(3, ": " __VA_ARGS__); \
+    RAWLOG(3, "\n"); \
+    return ERROR(err); \
+  } while(0);
+
+/**
+ * If the provided expression evaluates to an error code, returns that error code.
+ *
+ * In debug modes, prints additional information.
+ */
+#define FORWARD_IF_ERROR(err, ...) \
+  do { \
+    size_t const err_code = (err); \
+    if (ERR_isError(err_code)) { \
+      RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
+             __FILE__, __LINE__, ERR_QUOTE(err), ERR_getErrorName(err_code)); \
+      _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+      RAWLOG(3, ": " __VA_ARGS__); \
+      RAWLOG(3, "\n"); \
+      return err_code; \
+    } \
+  } while(0);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/fse.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/fse.h
new file mode 100644
index 0000000..02a1f0b
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/fse.h
@@ -0,0 +1,639 @@
+/* ******************************************************************
+ * FSE : Finite State Entropy codec
+ * Public Prototypes declaration
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef FSE_H
+#define FSE_H
+
+
+/*-*****************************************
+*  Dependencies
+******************************************/
+#include "zstd_deps.h"    /* size_t, ptrdiff_t */
+
+
+/*-*****************************************
+*  FSE_PUBLIC_API : control library symbols visibility
+******************************************/
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define FSE_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define FSE_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define FSE_PUBLIC_API __declspec(dllimport) /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define FSE_PUBLIC_API
+#endif
+
+/*------   Version   ------*/
+#define FSE_VERSION_MAJOR    0
+#define FSE_VERSION_MINOR    9
+#define FSE_VERSION_RELEASE  0
+
+#define FSE_LIB_VERSION FSE_VERSION_MAJOR.FSE_VERSION_MINOR.FSE_VERSION_RELEASE
+#define FSE_QUOTE(str) #str
+#define FSE_EXPAND_AND_QUOTE(str) FSE_QUOTE(str)
+#define FSE_VERSION_STRING FSE_EXPAND_AND_QUOTE(FSE_LIB_VERSION)
+
+#define FSE_VERSION_NUMBER  (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
+FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
+
+
+/*-*****************************************
+*  Tool functions
+******************************************/
+FSE_PUBLIC_API size_t FSE_compressBound(size_t size);       /* maximum compressed size */
+
+/* Error Management */
+FSE_PUBLIC_API unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+FSE_PUBLIC_API const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSE_compress() does the following:
+1. count symbol occurrence from source[] into table count[] (see hist.h)
+2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
+3. save normalized counters to memory buffer using writeNCount()
+4. build encoding table 'CTable' from normalized counters
+5. encode the data stream using encoding table 'CTable'
+
+FSE_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+/* *** COMPRESSION *** */
+
+/*! FSE_optimalTableLog():
+    dynamically downsize 'tableLog' when conditions are met.
+    It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
+    @return : recommended tableLog (necessarily <= 'maxTableLog') */
+FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
+
+/*! FSE_normalizeCount():
+    normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
+    'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
+    useLowProbCount is a boolean parameter which trades off compressed size for
+    faster header decoding. When it is set to 1, the compressed data will be slightly
+    smaller. And when it is set to 0, FSE_readNCount() and FSE_buildDTable() will be
+    faster. If you are compressing a small amount of data (< 2 KB) then useLowProbCount=0
+    is a good default, since header deserialization makes a big speed difference.
+    Otherwise, useLowProbCount=1 is a good default, since the speed difference is small.
+    @return : tableLog,
+              or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,
+                    const unsigned* count, size_t srcSize, unsigned maxSymbolValue, unsigned useLowProbCount);
+
+/*! FSE_NCountWriteBound():
+    Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
+    Typically useful for allocation purpose. */
+FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSE_writeNCount():
+    Compactly save 'normalizedCounter' into 'buffer'.
+    @return : size of the compressed table,
+              or an errorCode, which can be tested using FSE_isError(). */
+FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                                 const short* normalizedCounter,
+                                 unsigned maxSymbolValue, unsigned tableLog);
+
+/*! Constructor and Destructor of FSE_CTable.
+    Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
+
+/*! FSE_buildCTable():
+    Builds `ct`, which must be already allocated, using FSE_createCTable().
+    @return : 0, or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSE_compress_usingCTable():
+    Compress `src` using `ct` into `dst` which must be already allocated.
+    @return : size of compressed data (<= `dstCapacity`),
+              or 0 if compressed data could not fit into `dst`,
+              or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
+
+/*!
+Tutorial :
+----------
+The first step is to count all symbols. FSE_count() does this job very fast.
+Result will be saved into 'count', a table of unsigned int, which must be already allocated, and have 'maxSymbolValuePtr[0]+1' cells.
+'src' is a table of bytes of size 'srcSize'. All values within 'src' MUST be <= maxSymbolValuePtr[0]
+maxSymbolValuePtr[0] will be updated, with its real value (necessarily <= original value)
+FSE_count() will return the number of occurrence of the most frequent symbol.
+This can be used to know if there is a single symbol within 'src', and to quickly evaluate its compressibility.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
+
+The next step is to normalize the frequencies.
+FSE_normalizeCount() will ensure that sum of frequencies is == 2 ^'tableLog'.
+It also guarantees a minimum of 1 to any Symbol with frequency >= 1.
+You can use 'tableLog'==0 to mean "use default tableLog value".
+If you are unsure of which tableLog value to use, you can ask FSE_optimalTableLog(),
+which will provide the optimal valid tableLog given sourceSize, maxSymbolValue, and a user-defined maximum (0 means "default").
+
+The result of FSE_normalizeCount() will be saved into a table,
+called 'normalizedCounter', which is a table of signed short.
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValue+1' cells.
+The return value is tableLog if everything proceeded as expected.
+It is 0 if there is a single symbol within distribution.
+If there is an error (ex: invalid tableLog value), the function will return an ErrorCode (which can be tested using FSE_isError()).
+
+'normalizedCounter' can be saved in a compact manner to a memory area using FSE_writeNCount().
+'buffer' must be already allocated.
+For guaranteed success, buffer size must be at least FSE_headerBound().
+The result of the function is the number of bytes written into 'buffer'.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError(); ex : buffer size too small).
+
+'normalizedCounter' can then be used to create the compression table 'CTable'.
+The space required by 'CTable' must be already allocated, using FSE_createCTable().
+You can then use FSE_buildCTable() to fill 'CTable'.
+If there is an error, both functions will return an ErrorCode (which can be tested using FSE_isError()).
+
+'CTable' can then be used to compress 'src', with FSE_compress_usingCTable().
+Similar to FSE_count(), the convention is that 'src' is assumed to be a table of char of size 'srcSize'
+The function returns the size of compressed data (without header), necessarily <= `dstCapacity`.
+If it returns '0', compressed data could not fit into 'dst'.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSE_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSE_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,
+                           unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
+                           const void* rBuffer, size_t rBuffSize);
+
+/*! FSE_readNCount_bmi2():
+ * Same as FSE_readNCount() but pass bmi2=1 when your CPU supports BMI2 and 0 otherwise.
+ */
+FSE_PUBLIC_API size_t FSE_readNCount_bmi2(short* normalizedCounter,
+                           unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
+                           const void* rBuffer, size_t rBuffSize, int bmi2);
+
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.
+This is performed by the function FSE_buildDTable().
+The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+`FSE_DTable` can then be used to decompress `cSrc`, with FSE_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
+*/
+
+#endif  /* FSE_H */
+
+#if defined(FSE_STATIC_LINKING_ONLY) && !defined(FSE_H_FSE_STATIC_LINKING_ONLY)
+#define FSE_H_FSE_STATIC_LINKING_ONLY
+
+/* *** Dependency *** */
+#include "bitstream.h"
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) ((size) + ((size)>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<((maxTableLog)-1)) + (((maxSymbolValue)+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<(maxTableLog)))
+
+/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */
+#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue)   (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))
+#define FSE_DTABLE_SIZE(maxTableLog)                   (FSE_DTABLE_SIZE_U32(maxTableLog) * sizeof(FSE_DTable))
+
+
+/* *****************************************
+ *  FSE advanced API
+ ***************************************** */
+
+unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
+/**< same as FSE_optimalTableLog(), which used `minus==2` */
+
+size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
+/**< build a fake FSE_CTable, designed to compress always the same symbolValue */
+
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * `wkspSize` must be >= `FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog)` of `unsigned`.
+ * See FSE_buildCTable_wksp() for breakdown of workspace usage.
+ */
+#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (((maxSymbolValue + 2) + (1ull << (tableLog)))/2 + sizeof(U64)/sizeof(U32) /* additional 8 bytes for potential table overwrite */)
+#define FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) (sizeof(unsigned) * FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog))
+size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+
+#define FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) (sizeof(short) * (maxSymbolValue + 1) + (1ULL << maxTableLog) + 8)
+#define FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ((FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) + sizeof(unsigned) - 1) / sizeof(unsigned))
+FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+/**< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
+
+#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + 1 + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (FSE_MAX_SYMBOL_VALUE + 1) / 2 + 1)
+#define FSE_DECOMPRESS_WKSP_SIZE(maxTableLog, maxSymbolValue) (FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(unsigned))
+size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2);
+/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)`.
+ * Set bmi2 to 1 if your CPU supports BMI2 or 0 if it doesn't */
+
+typedef enum {
+   FSE_repeat_none,  /**< Cannot use the previous table */
+   FSE_repeat_check, /**< Can use the previous table but it must be checked */
+   FSE_repeat_valid  /**< Can use the previous table and it is assumed to be valid */
+ } FSE_repeat;
+
+/* *****************************************
+*  FSE symbol compression API
+*******************************************/
+/*!
+   This API consists of small unitary functions, which highly benefit from being inlined.
+   Hence their body are included in next section.
+*/
+typedef struct {
+    ptrdiff_t   value;
+    const void* stateTable;
+    const void* symbolTT;
+    unsigned    stateLog;
+} FSE_CState_t;
+
+static void FSE_initCState(FSE_CState_t* CStatePtr, const FSE_CTable* ct);
+
+static void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* CStatePtr, unsigned symbol);
+
+static void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* CStatePtr);
+
+/**<
+These functions are inner components of FSE_compress_usingCTable().
+They allow the creation of custom streams, mixing multiple tables and bit sources.
+
+A key property to keep in mind is that encoding and decoding are done **in reverse direction**.
+So the first symbol you will encode is the last you will decode, like a LIFO stack.
+
+You will need a few variables to track your CStream. They are :
+
+FSE_CTable    ct;         // Provided by FSE_buildCTable()
+BIT_CStream_t bitStream;  // bitStream tracking structure
+FSE_CState_t  state;      // State tracking structure (can have several)
+
+
+The first thing to do is to init bitStream and state.
+    size_t errorCode = BIT_initCStream(&bitStream, dstBuffer, maxDstSize);
+    FSE_initCState(&state, ct);
+
+Note that BIT_initCStream() can produce an error code, so its result should be tested, using FSE_isError();
+You can then encode your input data, byte after byte.
+FSE_encodeSymbol() outputs a maximum of 'tableLog' bits at a time.
+Remember decoding will be done in reverse direction.
+    FSE_encodeByte(&bitStream, &state, symbol);
+
+At any time, you can also add any bit sequence.
+Note : maximum allowed nbBits is 25, for compatibility with 32-bits decoders
+    BIT_addBits(&bitStream, bitField, nbBits);
+
+The above methods don't commit data to memory, they just store it into local register, for speed.
+Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).
+Writing data to memory is a manual operation, performed by the flushBits function.
+    BIT_flushBits(&bitStream);
+
+Your last FSE encoding operation shall be to flush your last state value(s).
+    FSE_flushState(&bitStream, &state);
+
+Finally, you must close the bitStream.
+The function returns the size of CStream in bytes.
+If data couldn't fit into dstBuffer, it will return a 0 ( == not compressible)
+If there is an error, it returns an errorCode (which can be tested using FSE_isError()).
+    size_t size = BIT_closeCStream(&bitStream);
+*/
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct {
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+/**<
+Let's now decompose FSE_decompress_usingDTable() into its unitary components.
+You will decode FSE-encoded symbols from the bitStream,
+and also any other bitFields you put in, **in reverse order**.
+
+You will need a few variables to track your bitStream. They are :
+
+BIT_DStream_t DStream;    // Stream context
+FSE_DState_t  DState;     // State context. Multiple ones are possible
+FSE_DTable*   DTablePtr;  // Decoding table, provided by FSE_buildDTable()
+
+The first thing to do is to init the bitStream.
+    errorCode = BIT_initDStream(&DStream, srcBuffer, srcSize);
+
+You should then retrieve your initial state(s)
+(in reverse flushing order if you have several ones) :
+    errorCode = FSE_initDState(&DState, &DStream, DTablePtr);
+
+You can then decode your data, symbol after symbol.
+For information the maximum number of bits read by FSE_decodeSymbol() is 'tableLog'.
+Keep in mind that symbols are decoded in reverse order, like a LIFO stack (last in, first out).
+    unsigned char symbol = FSE_decodeSymbol(&DState, &DStream);
+
+You can retrieve any bitfield you eventually stored into the bitStream (in reverse order)
+Note : maximum allowed nbBits is 25, for 32-bits compatibility
+    size_t bitField = BIT_readBits(&DStream, nbBits);
+
+All above operations only read from local register (which size depends on size_t).
+Refueling the register from memory is manually performed by the reload method.
+    endSignal = FSE_reloadDStream(&DStream);
+
+BIT_reloadDStream() result tells if there is still some more data to read from DStream.
+BIT_DStream_unfinished : there is still some data left into the DStream.
+BIT_DStream_endOfBuffer : Dstream reached end of buffer. Its container may no longer be completely filled.
+BIT_DStream_completed : Dstream reached its exact end, corresponding in general to decompression completed.
+BIT_DStream_tooFar : Dstream went too far. Decompression result is corrupted.
+
+When reaching end of buffer (BIT_DStream_endOfBuffer), progress slowly, notably if you decode multiple symbols per loop,
+to properly detect the exact end of stream.
+After each decoded symbol, check if DStream is fully consumed using this simple test :
+    BIT_reloadDStream(&DStream) >= BIT_DStream_completed
+
+When it's done, verify decompression is fully completed, by checking both DStream and the relevant states.
+Checking if DStream has reached its end is performed by :
+    BIT_endOfDStream(&DStream);
+Check also the states. There might be some symbols left there, if some high probability ones (>50%) are possible.
+    FSE_endOfDState(&DState);
+*/
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+typedef struct {
+    int deltaFindState;
+    U32 deltaNbBits;
+} FSE_symbolCompressionTransform; /* total 8 bytes */
+
+MEM_STATIC void FSE_initCState(FSE_CState_t* statePtr, const FSE_CTable* ct)
+{
+    const void* ptr = ct;
+    const U16* u16ptr = (const U16*) ptr;
+    const U32 tableLog = MEM_read16(ptr);
+    statePtr->value = (ptrdiff_t)1<<tableLog;
+    statePtr->stateTable = u16ptr+2;
+    statePtr->symbolTT = ct + 1 + (tableLog ? (1<<(tableLog-1)) : 1);
+    statePtr->stateLog = tableLog;
+}
+
+
+/*! FSE_initCState2() :
+*   Same as FSE_initCState(), but the first symbol to include (which will be the last to be read)
+*   uses the smallest state value possible, saving the cost of this symbol */
+MEM_STATIC void FSE_initCState2(FSE_CState_t* statePtr, const FSE_CTable* ct, U32 symbol)
+{
+    FSE_initCState(statePtr, ct);
+    {   const FSE_symbolCompressionTransform symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+        const U16* stateTable = (const U16*)(statePtr->stateTable);
+        U32 nbBitsOut  = (U32)((symbolTT.deltaNbBits + (1<<15)) >> 16);
+        statePtr->value = (nbBitsOut << 16) - symbolTT.deltaNbBits;
+        statePtr->value = stateTable[(statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
+    }
+}
+
+MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, unsigned symbol)
+{
+    FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+    const U16* const stateTable = (const U16*)(statePtr->stateTable);
+    U32 const nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
+    BIT_addBits(bitC, statePtr->value, nbBitsOut);
+    statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
+}
+
+MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePtr)
+{
+    BIT_addBits(bitC, statePtr->value, statePtr->stateLog);
+    BIT_flushBits(bitC);
+}
+
+
+/* FSE_getMaxNbBits() :
+ * Approximate maximum cost of a symbol, in bits.
+ * Fractional get rounded up (i.e. a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;
+}
+
+/* FSE_bitCost() :
+ * Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_bitCost(const void* symbolTTPtr, U32 tableLog, U32 symbolValue, U32 accuracyLog)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;
+    U32 const threshold = (minNbBits+1) << 16;
+    assert(tableLog < 16);
+    assert(accuracyLog < 31-tableLog);  /* ensure enough room for renormalization double shift */
+    {   U32 const tableSize = 1 << tableLog;
+        U32 const deltaFromThreshold = threshold - (symbolTT[symbolValue].deltaNbBits + tableSize);
+        U32 const normalizedDeltaFromThreshold = (deltaFromThreshold << accuracyLog) >> tableLog;   /* linear interpolation (very approximate) */
+        U32 const bitMultiplier = 1 << accuracyLog;
+        assert(symbolTT[symbolValue].deltaNbBits + tableSize <= threshold);
+        assert(normalizedDeltaFromThreshold <= bitMultiplier);
+        return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold;
+    }
+}
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSE_DTableHeader* const DTableH = (const FSE_DTableHeader*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_peekSymbol(const FSE_DState_t* DStatePtr)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSE_updateState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSE_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#ifndef FSE_MAX_MEMORY_USAGE
+#  define FSE_MAX_MEMORY_USAGE 14
+#endif
+#ifndef FSE_DEFAULT_MEMORY_USAGE
+#  define FSE_DEFAULT_MEMORY_USAGE 13
+#endif
+#if (FSE_DEFAULT_MEMORY_USAGE > FSE_MAX_MEMORY_USAGE)
+#  error "FSE_DEFAULT_MEMORY_USAGE must be <= FSE_MAX_MEMORY_USAGE"
+#endif
+
+/*!FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#ifndef FSE_MAX_SYMBOL_VALUE
+#  define FSE_MAX_SYMBOL_VALUE 255
+#endif
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#  error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSE_TABLESTEP(tableSize) (((tableSize)>>1) + ((tableSize)>>3) + 3)
+
+
+#endif /* FSE_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/fse_decompress.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/fse_decompress.c
new file mode 100644
index 0000000..1e1c9f9
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/fse_decompress.c
@@ -0,0 +1,311 @@
+/* ******************************************************************
+ * FSE : Finite State Entropy decoder
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include "debug.h"      /* assert */
+#include "bitstream.h"
+#include "compiler.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#include "error_private.h"
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"
+#include "bits.h"       /* ZSTD_highbit32 */
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_isError ERR_isError
+#define FSE_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
+    U16* symbolNext = (U16*)workSpace;
+    BYTE* spread = (BYTE*)(symbolNext + maxSymbolValue + 1);
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (FSE_BUILD_DTABLE_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(maxSymbolValue_tooLarge);
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSE_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    if (highThreshold == tableSize - 1) {
+        size_t const tableMask = tableSize-1;
+        size_t const step = FSE_TABLESTEP(tableSize);
+        /* First lay down the symbols in order.
+         * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
+         * misses since small blocks generally have small table logs, so nearly
+         * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
+         * our buffer to handle the over-write.
+         */
+        {
+            U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                pos += n;
+            }
+        }
+        /* Now we spread those positions across the table.
+         * The benefit of doing it in two stages is that we avoid the
+         * variable size inner loop, which caused lots of branch misses.
+         * Now we can run through all the positions without any branch misses.
+         * We unroll the loop twice, since that is what empirically worked best.
+         */
+        {
+            size_t position = 0;
+            size_t s;
+            size_t const unroll = 2;
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableDecode[uPosition].symbol = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);
+        }
+    } else {
+        U32 const tableMask = tableSize-1;
+        U32 const step = FSE_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+{
+    return FSE_buildDTable_internal(dt, normalizedCounter, maxSymbolValue, tableLog, workSpace, wkspSize);
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+
+FORCE_INLINE_TEMPLATE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+
+    /* Init */
+    CHECK_F(BIT_initDStream(&bitD, cSrc, cSrcSize));
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) & (op<olimit) ; op+=4) {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+        *op++ = FSE_GETSYMBOL(&state1);
+        if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
+            *op++ = FSE_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+        *op++ = FSE_GETSYMBOL(&state2);
+        if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
+            *op++ = FSE_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+typedef struct {
+    short ncount[FSE_MAX_SYMBOL_VALUE + 1];
+    FSE_DTable dtable[1]; /* Dynamically sized */
+} FSE_DecompressWksp;
+
+
+FORCE_INLINE_TEMPLATE size_t FSE_decompress_wksp_body(
+        void* dst, size_t dstCapacity,
+        const void* cSrc, size_t cSrcSize,
+        unsigned maxLog, void* workSpace, size_t wkspSize,
+        int bmi2)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    FSE_DecompressWksp* const wksp = (FSE_DecompressWksp*)workSpace;
+
+    DEBUG_STATIC_ASSERT((FSE_MAX_SYMBOL_VALUE + 1) % 2 == 0);
+    if (wkspSize < sizeof(*wksp)) return ERROR(GENERIC);
+
+    /* normal FSE decoding mode */
+    {
+        size_t const NCountLength = FSE_readNCount_bmi2(wksp->ncount, &maxSymbolValue, &tableLog, istart, cSrcSize, bmi2);
+        if (FSE_isError(NCountLength)) return NCountLength;
+        if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
+        assert(NCountLength <= cSrcSize);
+        ip += NCountLength;
+        cSrcSize -= NCountLength;
+    }
+
+    if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
+    assert(sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog) <= wkspSize);
+    workSpace = (BYTE*)workSpace + sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
+    wkspSize -= sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
+
+    CHECK_F( FSE_buildDTable_internal(wksp->dtable, wksp->ncount, maxSymbolValue, tableLog, workSpace, wkspSize) );
+
+    {
+        const void* ptr = wksp->dtable;
+        const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
+        const U32 fastMode = DTableH->fastMode;
+
+        /* select fast mode (static) */
+        if (fastMode) return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 1);
+        return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 0);
+    }
+}
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t FSE_decompress_wksp_body_default(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
+{
+    return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 0);
+}
+
+#if DYNAMIC_BMI2
+BMI2_TARGET_ATTRIBUTE static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
+{
+    return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
+}
+#endif
+
+size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return FSE_decompress_wksp_body_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
+    }
+#endif
+    (void)bmi2;
+    return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
+}
+
+#endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/huf.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/huf.h
new file mode 100644
index 0000000..73d1ee5
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/huf.h
@@ -0,0 +1,273 @@
+/* ******************************************************************
+ * huff0 huffman codec,
+ * part of Finite State Entropy library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef HUF_H_298734234
+#define HUF_H_298734234
+
+/* *** Dependencies *** */
+#include "zstd_deps.h"    /* size_t */
+#include "mem.h"          /* U32 */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+
+
+/* ***   Tool functions *** */
+#define HUF_BLOCKSIZE_MAX (128 * 1024)   /**< maximum input size for a single block compressed with HUF_compress */
+size_t HUF_compressBound(size_t size);   /**< maximum compressed size (worst case) */
+
+/* Error Management */
+unsigned    HUF_isError(size_t code);       /**< tells if a return value is an error code */
+const char* HUF_getErrorName(size_t code);  /**< provides error code string (useful for debugging) */
+
+
+#define HUF_WORKSPACE_SIZE ((8 << 10) + 512 /* sorting scratch space */)
+#define HUF_WORKSPACE_SIZE_U64 (HUF_WORKSPACE_SIZE / sizeof(U64))
+
+/* *** Constants *** */
+#define HUF_TABLELOG_MAX      12      /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_TABLELOG_ABSOLUTEMAX */
+#define HUF_TABLELOG_DEFAULT  11      /* default tableLog value when none specified */
+#define HUF_SYMBOLVALUE_MAX  255
+
+#define HUF_TABLELOG_ABSOLUTEMAX  12  /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
+#  error "HUF_TABLELOG_MAX is too large !"
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true when incompressible is pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of HUF's Compression Table */
+/* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
+typedef size_t HUF_CElt;   /* consider it an incomplete type */
+#define HUF_CTABLE_SIZE_ST(maxSymbolValue)   ((maxSymbolValue)+2)   /* Use tables of size_t, for proper alignment */
+#define HUF_CTABLE_SIZE(maxSymbolValue)       (HUF_CTABLE_SIZE_ST(maxSymbolValue) * sizeof(size_t))
+#define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
+    HUF_CElt name[HUF_CTABLE_SIZE_ST(maxSymbolValue)] /* no final ; */
+
+/* static allocation of HUF's DTable */
+typedef U32 HUF_DTable;
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
+#define HUF_CREATE_STATIC_DTABLEX1(DTable, maxTableLog) \
+        HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1) * 0x01000001) }
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog) * 0x01000001) }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+
+/**
+ * Huffman flags bitset.
+ * For all flags, 0 is the default value.
+ */
+typedef enum {
+    /**
+     * If compiled with DYNAMIC_BMI2: Set flag only if the CPU supports BMI2 at runtime.
+     * Otherwise: Ignored.
+     */
+    HUF_flags_bmi2 = (1 << 0),
+    /**
+     * If set: Test possible table depths to find the one that produces the smallest header + encoded size.
+     * If unset: Use heuristic to find the table depth.
+     */
+    HUF_flags_optimalDepth = (1 << 1),
+    /**
+     * If set: If the previous table can encode the input, always reuse the previous table.
+     * If unset: If the previous table can encode the input, reuse the previous table if it results in a smaller output.
+     */
+    HUF_flags_preferRepeat = (1 << 2),
+    /**
+     * If set: Sample the input and check if the sample is uncompressible, if it is then don't attempt to compress.
+     * If unset: Always histogram the entire input.
+     */
+    HUF_flags_suspectUncompressible = (1 << 3),
+    /**
+     * If set: Don't use assembly implementations
+     * If unset: Allow using assembly implementations
+     */
+    HUF_flags_disableAsm = (1 << 4),
+    /**
+     * If set: Don't use the fast decoding loop, always use the fallback decoding loop.
+     * If unset: Use the fast decoding loop when possible.
+     */
+    HUF_flags_disableFast = (1 << 5)
+} HUF_flags_e;
+
+
+/* ****************************************
+ *  HUF detailed API
+ * ****************************************/
+#define HUF_OPTIMAL_DEPTH_THRESHOLD ZSTD_btultra
+
+/*! HUF_compress() does the following:
+ *  1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
+ *  2. (optional) refine tableLog using HUF_optimalTableLog()
+ *  3. build Huffman table from count using HUF_buildCTable()
+ *  4. save Huffman table to memory buffer using HUF_writeCTable()
+ *  5. encode the data stream using HUF_compress4X_usingCTable()
+ *
+ *  The following API allows targeting specific sub-functions for advanced tasks.
+ *  For example, it's possible to compress several blocks using the same 'CTable',
+ *  or to save and regenerate 'CTable' using external methods.
+ */
+unsigned HUF_minTableLog(unsigned symbolCardinality);
+unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue);
+unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, void* workSpace,
+ size_t wkspSize, HUF_CElt* table, const unsigned* count, int flags); /* table is used as scratch space for building and testing tables, not a return value */
+size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog, void* workspace, size_t workspaceSize);
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
+size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
+int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
+
+typedef enum {
+   HUF_repeat_none,  /**< Cannot use the previous table */
+   HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
+   HUF_repeat_valid  /**< Can use the previous table and it is assumed to be valid */
+ } HUF_repeat;
+
+/** HUF_compress4X_repeat() :
+ *  Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid.
+ *  If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
+size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,    /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
+
+/** HUF_buildCTable_wksp() :
+ *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
+ * `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
+ */
+#define HUF_CTABLE_WORKSPACE_SIZE_U32 ((4 * (HUF_SYMBOLVALUE_MAX + 1)) + 192)
+#define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
+size_t HUF_buildCTable_wksp (HUF_CElt* tree,
+                       const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
+                             void* workSpace, size_t wkspSize);
+
+/*! HUF_readStats() :
+ *  Read compact Huffman tree, saved by HUF_writeCTable().
+ * `huffWeight` is destination buffer.
+ * @return : size read from `src` , or an error Code .
+ *  Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
+size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
+                     U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize);
+
+/*! HUF_readStats_wksp() :
+ * Same as HUF_readStats() but takes an external workspace which must be
+ * 4-byte aligned and its size must be >= HUF_READ_STATS_WORKSPACE_SIZE.
+ * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
+ */
+#define HUF_READ_STATS_WORKSPACE_SIZE_U32 FSE_DECOMPRESS_WKSP_SIZE_U32(6, HUF_TABLELOG_MAX-1)
+#define HUF_READ_STATS_WORKSPACE_SIZE (HUF_READ_STATS_WORKSPACE_SIZE_U32 * sizeof(unsigned))
+size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
+                          U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
+                          const void* src, size_t srcSize,
+                          void* workspace, size_t wkspSize,
+                          int flags);
+
+/** HUF_readCTable() :
+ *  Loading a CTable saved with HUF_writeCTable() */
+size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
+
+/** HUF_getNbBitsFromCTable() :
+ *  Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
+ *  Note 1 : is not inlined, as HUF_CElt definition is private */
+U32 HUF_getNbBitsFromCTable(const HUF_CElt* symbolTable, U32 symbolValue);
+
+/*
+ * HUF_decompress() does the following:
+ * 1. select the decompression algorithm (X1, X2) based on pre-computed heuristics
+ * 2. build Huffman table from save, using HUF_readDTableX?()
+ * 3. decode 1 or 4 segments in parallel using HUF_decompress?X?_usingDTable()
+ */
+
+/** HUF_selectDecoder() :
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
+ *  Assumption : 0 < dstSize <= 128 KB */
+U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
+
+/**
+ *  The minimum workspace size for the `workSpace` used in
+ *  HUF_readDTableX1_wksp() and HUF_readDTableX2_wksp().
+ *
+ *  The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
+ *  HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
+ *  Buffer overflow errors may potentially occur if code modifications result in
+ *  a required workspace size greater than that specified in the following
+ *  macro.
+ */
+#define HUF_DECOMPRESS_WORKSPACE_SIZE ((2 << 10) + (1 << 9))
+#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
+
+
+/* ====================== */
+/* single stream variants */
+/* ====================== */
+
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
+/** HUF_compress1X_repeat() :
+ *  Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid.
+ *  If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
+size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,   /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
+
+size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);   /**< double-symbols decoder */
+#endif
+
+/* BMI2 variants.
+ * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
+ */
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
+#endif
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
+#endif
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
+#endif
+
+#endif   /* HUF_H_298734234 */
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/mem.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/mem.h
new file mode 100644
index 0000000..98dd47a
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/mem.h
@@ -0,0 +1,435 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include <stddef.h>  /* size_t, ptrdiff_t */
+#include "compiler.h"  /* __has_builtin */
+#include "debug.h"  /* DEBUG_STATIC_ASSERT */
+#include "zstd_deps.h"  /* ZSTD_memcpy */
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __inline __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#  if defined(_AIX)
+#    include <inttypes.h>
+#  else
+#    include <stdint.h> /* intptr_t */
+#  endif
+  typedef   uint8_t BYTE;
+  typedef   uint8_t U8;
+  typedef    int8_t S8;
+  typedef  uint16_t U16;
+  typedef   int16_t S16;
+  typedef  uint32_t U32;
+  typedef   int32_t S32;
+  typedef  uint64_t U64;
+  typedef   int64_t S64;
+#else
+# include <limits.h>
+#if CHAR_BIT != 8
+#  error "this implementation requires char to be exactly 8-bit type"
+#endif
+  typedef unsigned char      BYTE;
+  typedef unsigned char      U8;
+  typedef   signed char      S8;
+#if USHRT_MAX != 65535
+#  error "this implementation requires short to be exactly 16-bit type"
+#endif
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+#if UINT_MAX != 4294967295
+#  error "this implementation requires int to be exactly 32-bit type"
+#endif
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+/* note : there are no limits defined for long long type in C90.
+ * limits exist in C99, however, in such case, <stdint.h> is preferred */
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O API
+*****************************************************************/
+/*=== Static platform detection ===*/
+MEM_STATIC unsigned MEM_32bits(void);
+MEM_STATIC unsigned MEM_64bits(void);
+MEM_STATIC unsigned MEM_isLittleEndian(void);
+
+/*=== Native unaligned read/write ===*/
+MEM_STATIC U16 MEM_read16(const void* memPtr);
+MEM_STATIC U32 MEM_read32(const void* memPtr);
+MEM_STATIC U64 MEM_read64(const void* memPtr);
+MEM_STATIC size_t MEM_readST(const void* memPtr);
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value);
+MEM_STATIC void MEM_write32(void* memPtr, U32 value);
+MEM_STATIC void MEM_write64(void* memPtr, U64 value);
+
+/*=== Little endian unaligned read/write ===*/
+MEM_STATIC U16 MEM_readLE16(const void* memPtr);
+MEM_STATIC U32 MEM_readLE24(const void* memPtr);
+MEM_STATIC U32 MEM_readLE32(const void* memPtr);
+MEM_STATIC U64 MEM_readLE64(const void* memPtr);
+MEM_STATIC size_t MEM_readLEST(const void* memPtr);
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val);
+MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val);
+MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32);
+MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64);
+MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val);
+
+/*=== Big endian unaligned read/write ===*/
+MEM_STATIC U32 MEM_readBE32(const void* memPtr);
+MEM_STATIC U64 MEM_readBE64(const void* memPtr);
+MEM_STATIC size_t MEM_readBEST(const void* memPtr);
+
+MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32);
+MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64);
+MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val);
+
+/*=== Byteswap ===*/
+MEM_STATIC U32 MEM_swap32(U32 in);
+MEM_STATIC U64 MEM_swap64(U64 in);
+MEM_STATIC size_t MEM_swapST(size_t in);
+
+
+/*-**************************************************************
+*  Memory I/O Implementation
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS : For accessing unaligned memory:
+ * Method 0 : always use `memcpy()`. Safe and portable.
+ * Method 1 : Use compiler extension to set unaligned access.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets depending on alignment.
+ * Default  : method 1 if supported, else method 0
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  ifdef __GNUC__
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+#if defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+    return 1;
+#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+    return 0;
+#elif defined(__clang__) && __LITTLE_ENDIAN__
+    return 1;
+#elif defined(__clang__) && __BIG_ENDIAN__
+    return 0;
+#elif defined(_MSC_VER) && (_M_AMD64 || _M_IX86)
+    return 1;
+#elif defined(__DMC__) && defined(_M_IX86)
+    return 1;
+#else
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+#endif
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard, by lying on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+MEM_STATIC size_t MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+typedef __attribute__((aligned(1))) U16 unalign16;
+typedef __attribute__((aligned(1))) U32 unalign32;
+typedef __attribute__((aligned(1))) U64 unalign64;
+typedef __attribute__((aligned(1))) size_t unalignArch;
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return *(const unalign16*)ptr; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return *(const unalign32*)ptr; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return *(const unalign64*)ptr; }
+MEM_STATIC size_t MEM_readST(const void* ptr) { return *(const unalignArch*)ptr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(unalign16*)memPtr = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(unalign32*)memPtr = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(unalign64*)memPtr = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC size_t MEM_readST(const void* memPtr)
+{
+    size_t val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    ZSTD_memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write32(void* memPtr, U32 value)
+{
+    ZSTD_memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write64(void* memPtr, U64 value)
+{
+    ZSTD_memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif /* MEM_FORCE_MEMORY_ACCESS */
+
+MEM_STATIC U32 MEM_swap32_fallback(U32 in)
+{
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+}
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap32))
+    return __builtin_bswap32(in);
+#else
+    return MEM_swap32_fallback(in);
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64_fallback(U64 in)
+{
+     return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap64))
+    return __builtin_bswap64(in);
+#else
+    return MEM_swap64_fallback(in);
+#endif
+}
+
+MEM_STATIC size_t MEM_swapST(size_t in)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_swap32((U32)in);
+    else
+        return (size_t)MEM_swap64((U64)in);
+}
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return (U32)MEM_readLE16(memPtr) + ((U32)(((const BYTE*)memPtr)[2]) << 16);
+}
+
+MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val)
+{
+    MEM_writeLE16(memPtr, (U16)val);
+    ((BYTE*)memPtr)[2] = (BYTE)(val>>16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32)
+{
+    if (MEM_isLittleEndian())
+        MEM_write32(memPtr, val32);
+    else
+        MEM_write32(memPtr, MEM_swap32(val32));
+}
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64)
+{
+    if (MEM_isLittleEndian())
+        MEM_write64(memPtr, val64);
+    else
+        MEM_write64(memPtr, MEM_swap64(val64));
+}
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val)
+{
+    if (MEM_32bits())
+        MEM_writeLE32(memPtr, (U32)val);
+    else
+        MEM_writeLE64(memPtr, (U64)val);
+}
+
+/*=== Big endian r/w ===*/
+
+MEM_STATIC U32 MEM_readBE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_swap32(MEM_read32(memPtr));
+    else
+        return MEM_read32(memPtr);
+}
+
+MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32)
+{
+    if (MEM_isLittleEndian())
+        MEM_write32(memPtr, MEM_swap32(val32));
+    else
+        MEM_write32(memPtr, val32);
+}
+
+MEM_STATIC U64 MEM_readBE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_swap64(MEM_read64(memPtr));
+    else
+        return MEM_read64(memPtr);
+}
+
+MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64)
+{
+    if (MEM_isLittleEndian())
+        MEM_write64(memPtr, MEM_swap64(val64));
+    else
+        MEM_write64(memPtr, val64);
+}
+
+MEM_STATIC size_t MEM_readBEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readBE32(memPtr);
+    else
+        return (size_t)MEM_readBE64(memPtr);
+}
+
+MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val)
+{
+    if (MEM_32bits())
+        MEM_writeBE32(memPtr, (U32)val);
+    else
+        MEM_writeBE64(memPtr, (U64)val);
+}
+
+/* code only tested on 32 and 64 bits systems */
+MEM_STATIC void MEM_check(void) { DEBUG_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/pool.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/pool.c
new file mode 100644
index 0000000..d5ca5a7
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/pool.c
@@ -0,0 +1,371 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ======   Dependencies   ======= */
+#include "../common/allocations.h"  /* ZSTD_customCalloc, ZSTD_customFree */
+#include "zstd_deps.h" /* size_t */
+#include "debug.h"     /* assert */
+#include "pool.h"
+
+/* ======   Compiler specifics   ====== */
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#endif
+
+
+#ifdef ZSTD_MULTITHREAD
+
+#include "threading.h"   /* pthread adaptation */
+
+/* A job is a function and an opaque argument */
+typedef struct POOL_job_s {
+    POOL_function function;
+    void *opaque;
+} POOL_job;
+
+struct POOL_ctx_s {
+    ZSTD_customMem customMem;
+    /* Keep track of the threads */
+    ZSTD_pthread_t* threads;
+    size_t threadCapacity;
+    size_t threadLimit;
+
+    /* The queue is a circular buffer */
+    POOL_job *queue;
+    size_t queueHead;
+    size_t queueTail;
+    size_t queueSize;
+
+    /* The number of threads working on jobs */
+    size_t numThreadsBusy;
+    /* Indicates if the queue is empty */
+    int queueEmpty;
+
+    /* The mutex protects the queue */
+    ZSTD_pthread_mutex_t queueMutex;
+    /* Condition variable for pushers to wait on when the queue is full */
+    ZSTD_pthread_cond_t queuePushCond;
+    /* Condition variables for poppers to wait on when the queue is empty */
+    ZSTD_pthread_cond_t queuePopCond;
+    /* Indicates if the queue is shutting down */
+    int shutdown;
+};
+
+/* POOL_thread() :
+ * Work thread for the thread pool.
+ * Waits for jobs and executes them.
+ * @returns : NULL on failure else non-null.
+ */
+static void* POOL_thread(void* opaque) {
+    POOL_ctx* const ctx = (POOL_ctx*)opaque;
+    if (!ctx) { return NULL; }
+    for (;;) {
+        /* Lock the mutex and wait for a non-empty queue or until shutdown */
+        ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+
+        while ( ctx->queueEmpty
+            || (ctx->numThreadsBusy >= ctx->threadLimit) ) {
+            if (ctx->shutdown) {
+                /* even if !queueEmpty, (possible if numThreadsBusy >= threadLimit),
+                 * a few threads will be shutdown while !queueEmpty,
+                 * but enough threads will remain active to finish the queue */
+                ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+                return opaque;
+            }
+            ZSTD_pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
+        }
+        /* Pop a job off the queue */
+        {   POOL_job const job = ctx->queue[ctx->queueHead];
+            ctx->queueHead = (ctx->queueHead + 1) % ctx->queueSize;
+            ctx->numThreadsBusy++;
+            ctx->queueEmpty = (ctx->queueHead == ctx->queueTail);
+            /* Unlock the mutex, signal a pusher, and run the job */
+            ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+
+            job.function(job.opaque);
+
+            /* If the intended queue size was 0, signal after finishing job */
+            ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+            ctx->numThreadsBusy--;
+            ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+        }
+    }  /* for (;;) */
+    assert(0);  /* Unreachable */
+}
+
+/* ZSTD_createThreadPool() : public access point */
+POOL_ctx* ZSTD_createThreadPool(size_t numThreads) {
+    return POOL_create (numThreads, 0);
+}
+
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
+    return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
+}
+
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
+                               ZSTD_customMem customMem)
+{
+    POOL_ctx* ctx;
+    /* Check parameters */
+    if (!numThreads) { return NULL; }
+    /* Allocate the context and zero initialize */
+    ctx = (POOL_ctx*)ZSTD_customCalloc(sizeof(POOL_ctx), customMem);
+    if (!ctx) { return NULL; }
+    /* Initialize the job queue.
+     * It needs one extra space since one space is wasted to differentiate
+     * empty and full queues.
+     */
+    ctx->queueSize = queueSize + 1;
+    ctx->queue = (POOL_job*)ZSTD_customCalloc(ctx->queueSize * sizeof(POOL_job), customMem);
+    ctx->queueHead = 0;
+    ctx->queueTail = 0;
+    ctx->numThreadsBusy = 0;
+    ctx->queueEmpty = 1;
+    {
+        int error = 0;
+        error |= ZSTD_pthread_mutex_init(&ctx->queueMutex, NULL);
+        error |= ZSTD_pthread_cond_init(&ctx->queuePushCond, NULL);
+        error |= ZSTD_pthread_cond_init(&ctx->queuePopCond, NULL);
+        if (error) { POOL_free(ctx); return NULL; }
+    }
+    ctx->shutdown = 0;
+    /* Allocate space for the thread handles */
+    ctx->threads = (ZSTD_pthread_t*)ZSTD_customCalloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
+    ctx->threadCapacity = 0;
+    ctx->customMem = customMem;
+    /* Check for errors */
+    if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; }
+    /* Initialize the threads */
+    {   size_t i;
+        for (i = 0; i < numThreads; ++i) {
+            if (ZSTD_pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
+                ctx->threadCapacity = i;
+                POOL_free(ctx);
+                return NULL;
+        }   }
+        ctx->threadCapacity = numThreads;
+        ctx->threadLimit = numThreads;
+    }
+    return ctx;
+}
+
+/*! POOL_join() :
+    Shutdown the queue, wake any sleeping threads, and join all of the threads.
+*/
+static void POOL_join(POOL_ctx* ctx) {
+    /* Shut down the queue */
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    ctx->shutdown = 1;
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    /* Wake up sleeping threads */
+    ZSTD_pthread_cond_broadcast(&ctx->queuePushCond);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
+    /* Join all of the threads */
+    {   size_t i;
+        for (i = 0; i < ctx->threadCapacity; ++i) {
+            ZSTD_pthread_join(ctx->threads[i]);  /* note : could fail */
+    }   }
+}
+
+void POOL_free(POOL_ctx *ctx) {
+    if (!ctx) { return; }
+    POOL_join(ctx);
+    ZSTD_pthread_mutex_destroy(&ctx->queueMutex);
+    ZSTD_pthread_cond_destroy(&ctx->queuePushCond);
+    ZSTD_pthread_cond_destroy(&ctx->queuePopCond);
+    ZSTD_customFree(ctx->queue, ctx->customMem);
+    ZSTD_customFree(ctx->threads, ctx->customMem);
+    ZSTD_customFree(ctx, ctx->customMem);
+}
+
+/*! POOL_joinJobs() :
+ *  Waits for all queued jobs to finish executing.
+ */
+void POOL_joinJobs(POOL_ctx* ctx) {
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    while(!ctx->queueEmpty || ctx->numThreadsBusy > 0) {
+        ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
+    }
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+}
+
+void ZSTD_freeThreadPool (ZSTD_threadPool* pool) {
+  POOL_free (pool);
+}
+
+size_t POOL_sizeof(const POOL_ctx* ctx) {
+    if (ctx==NULL) return 0;  /* supports sizeof NULL */
+    return sizeof(*ctx)
+        + ctx->queueSize * sizeof(POOL_job)
+        + ctx->threadCapacity * sizeof(ZSTD_pthread_t);
+}
+
+
+/* @return : 0 on success, 1 on error */
+static int POOL_resize_internal(POOL_ctx* ctx, size_t numThreads)
+{
+    if (numThreads <= ctx->threadCapacity) {
+        if (!numThreads) return 1;
+        ctx->threadLimit = numThreads;
+        return 0;
+    }
+    /* numThreads > threadCapacity */
+    {   ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_customCalloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
+        if (!threadPool) return 1;
+        /* replace existing thread pool */
+        ZSTD_memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
+        ZSTD_customFree(ctx->threads, ctx->customMem);
+        ctx->threads = threadPool;
+        /* Initialize additional threads */
+        {   size_t threadId;
+            for (threadId = ctx->threadCapacity; threadId < numThreads; ++threadId) {
+                if (ZSTD_pthread_create(&threadPool[threadId], NULL, &POOL_thread, ctx)) {
+                    ctx->threadCapacity = threadId;
+                    return 1;
+            }   }
+    }   }
+    /* successfully expanded */
+    ctx->threadCapacity = numThreads;
+    ctx->threadLimit = numThreads;
+    return 0;
+}
+
+/* @return : 0 on success, 1 on error */
+int POOL_resize(POOL_ctx* ctx, size_t numThreads)
+{
+    int result;
+    if (ctx==NULL) return 1;
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    result = POOL_resize_internal(ctx, numThreads);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    return result;
+}
+
+/**
+ * Returns 1 if the queue is full and 0 otherwise.
+ *
+ * When queueSize is 1 (pool was created with an intended queueSize of 0),
+ * then a queue is empty if there is a thread free _and_ no job is waiting.
+ */
+static int isQueueFull(POOL_ctx const* ctx) {
+    if (ctx->queueSize > 1) {
+        return ctx->queueHead == ((ctx->queueTail + 1) % ctx->queueSize);
+    } else {
+        return (ctx->numThreadsBusy == ctx->threadLimit) ||
+               !ctx->queueEmpty;
+    }
+}
+
+
+static void
+POOL_add_internal(POOL_ctx* ctx, POOL_function function, void *opaque)
+{
+    POOL_job job;
+    job.function = function;
+    job.opaque = opaque;
+    assert(ctx != NULL);
+    if (ctx->shutdown) return;
+
+    ctx->queueEmpty = 0;
+    ctx->queue[ctx->queueTail] = job;
+    ctx->queueTail = (ctx->queueTail + 1) % ctx->queueSize;
+    ZSTD_pthread_cond_signal(&ctx->queuePopCond);
+}
+
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    /* Wait until there is space in the queue for the new job */
+    while (isQueueFull(ctx) && (!ctx->shutdown)) {
+        ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
+    }
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+}
+
+
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    if (isQueueFull(ctx)) {
+        ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+        return 0;
+    }
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    return 1;
+}
+
+
+#else  /* ZSTD_MULTITHREAD  not defined */
+
+/* ========================== */
+/* No multi-threading support */
+/* ========================== */
+
+
+/* We don't need any data, but if it is empty, malloc() might return NULL. */
+struct POOL_ctx_s {
+    int dummy;
+};
+static POOL_ctx g_poolCtx;
+
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
+    return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
+}
+
+POOL_ctx*
+POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem)
+{
+    (void)numThreads;
+    (void)queueSize;
+    (void)customMem;
+    return &g_poolCtx;
+}
+
+void POOL_free(POOL_ctx* ctx) {
+    assert(!ctx || ctx == &g_poolCtx);
+    (void)ctx;
+}
+
+void POOL_joinJobs(POOL_ctx* ctx){
+    assert(!ctx || ctx == &g_poolCtx);
+    (void)ctx;
+}
+
+int POOL_resize(POOL_ctx* ctx, size_t numThreads) {
+    (void)ctx; (void)numThreads;
+    return 0;
+}
+
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque) {
+    (void)ctx;
+    function(opaque);
+}
+
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
+    (void)ctx;
+    function(opaque);
+    return 1;
+}
+
+size_t POOL_sizeof(const POOL_ctx* ctx) {
+    if (ctx==NULL) return 0;  /* supports sizeof NULL */
+    assert(ctx == &g_poolCtx);
+    return sizeof(*ctx);
+}
+
+#endif  /* ZSTD_MULTITHREAD */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/pool.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/pool.h
new file mode 100644
index 0000000..eb22ff5
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/pool.h
@@ -0,0 +1,90 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef POOL_H
+#define POOL_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+#include "zstd_deps.h"
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_customMem */
+#include "../zstd.h"
+
+typedef struct POOL_ctx_s POOL_ctx;
+
+/*! POOL_create() :
+ *  Create a thread pool with at most `numThreads` threads.
+ * `numThreads` must be at least 1.
+ *  The maximum number of queued jobs before blocking is `queueSize`.
+ * @return : POOL_ctx pointer on success, else NULL.
+*/
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize);
+
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
+                               ZSTD_customMem customMem);
+
+/*! POOL_free() :
+ *  Free a thread pool returned by POOL_create().
+ */
+void POOL_free(POOL_ctx* ctx);
+
+
+/*! POOL_joinJobs() :
+ *  Waits for all queued jobs to finish executing.
+ */
+void POOL_joinJobs(POOL_ctx* ctx);
+
+/*! POOL_resize() :
+ *  Expands or shrinks pool's number of threads.
+ *  This is more efficient than releasing + creating a new context,
+ *  since it tries to preserve and re-use existing threads.
+ * `numThreads` must be at least 1.
+ * @return : 0 when resize was successful,
+ *           !0 (typically 1) if there is an error.
+ *    note : only numThreads can be resized, queueSize remains unchanged.
+ */
+int POOL_resize(POOL_ctx* ctx, size_t numThreads);
+
+/*! POOL_sizeof() :
+ * @return threadpool memory usage
+ *  note : compatible with NULL (returns 0 in this case)
+ */
+size_t POOL_sizeof(const POOL_ctx* ctx);
+
+/*! POOL_function :
+ *  The function type that can be added to a thread pool.
+ */
+typedef void (*POOL_function)(void*);
+
+/*! POOL_add() :
+ *  Add the job `function(opaque)` to the thread pool. `ctx` must be valid.
+ *  Possibly blocks until there is room in the queue.
+ *  Note : The function may be executed asynchronously,
+ *         therefore, `opaque` must live until function has been completed.
+ */
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque);
+
+
+/*! POOL_tryAdd() :
+ *  Add the job `function(opaque)` to thread pool _if_ a queue slot is available.
+ *  Returns immediately even if not (does not block).
+ * @return : 1 if successful, 0 if not.
+ */
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/portability_macros.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/portability_macros.h
new file mode 100644
index 0000000..8fd6ea8
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/portability_macros.h
@@ -0,0 +1,156 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_PORTABILITY_MACROS_H
+#define ZSTD_PORTABILITY_MACROS_H
+
+/**
+ * This header file contains macro definitions to support portability.
+ * This header is shared between C and ASM code, so it MUST only
+ * contain macro definitions. It MUST not contain any C code.
+ *
+ * This header ONLY defines macros to detect platforms/feature support.
+ *
+ */
+
+
+/* compat. with non-clang compilers */
+#ifndef __has_attribute
+  #define __has_attribute(x) 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_builtin
+#  define __has_builtin(x) 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_feature
+#  define __has_feature(x) 0
+#endif
+
+/* detects whether we are being compiled under msan */
+#ifndef ZSTD_MEMORY_SANITIZER
+#  if __has_feature(memory_sanitizer)
+#    define ZSTD_MEMORY_SANITIZER 1
+#  else
+#    define ZSTD_MEMORY_SANITIZER 0
+#  endif
+#endif
+
+/* detects whether we are being compiled under asan */
+#ifndef ZSTD_ADDRESS_SANITIZER
+#  if __has_feature(address_sanitizer)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  elif defined(__SANITIZE_ADDRESS__)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  else
+#    define ZSTD_ADDRESS_SANITIZER 0
+#  endif
+#endif
+
+/* detects whether we are being compiled under dfsan */
+#ifndef ZSTD_DATAFLOW_SANITIZER
+# if __has_feature(dataflow_sanitizer)
+#  define ZSTD_DATAFLOW_SANITIZER 1
+# else
+#  define ZSTD_DATAFLOW_SANITIZER 0
+# endif
+#endif
+
+/* Mark the internal assembly functions as hidden  */
+#ifdef __ELF__
+# define ZSTD_HIDE_ASM_FUNCTION(func) .hidden func
+#else
+# define ZSTD_HIDE_ASM_FUNCTION(func)
+#endif
+
+/* Enable runtime BMI2 dispatch based on the CPU.
+ * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
+ */
+#ifndef DYNAMIC_BMI2
+  #if ((defined(__clang__) && __has_attribute(__target__)) \
+      || (defined(__GNUC__) \
+          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
+      && (defined(__x86_64__) || defined(_M_X64)) \
+      && !defined(__BMI2__)
+  #  define DYNAMIC_BMI2 1
+  #else
+  #  define DYNAMIC_BMI2 0
+  #endif
+#endif
+
+/**
+ * Only enable assembly for GNUC compatible compilers,
+ * because other platforms may not support GAS assembly syntax.
+ *
+ * Only enable assembly for Linux / MacOS, other platforms may
+ * work, but they haven't been tested. This could likely be
+ * extended to BSD systems.
+ *
+ * Disable assembly when MSAN is enabled, because MSAN requires
+ * 100% of code to be instrumented to work.
+ */
+#if defined(__GNUC__)
+#  if defined(__linux__) || defined(__linux) || defined(__APPLE__)
+#    if ZSTD_MEMORY_SANITIZER
+#      define ZSTD_ASM_SUPPORTED 0
+#    elif ZSTD_DATAFLOW_SANITIZER
+#      define ZSTD_ASM_SUPPORTED 0
+#    else
+#      define ZSTD_ASM_SUPPORTED 1
+#    endif
+#  else
+#    define ZSTD_ASM_SUPPORTED 0
+#  endif
+#else
+#  define ZSTD_ASM_SUPPORTED 0
+#endif
+
+/**
+ * Determines whether we should enable assembly for x86-64
+ * with BMI2.
+ *
+ * Enable if all of the following conditions hold:
+ * - ASM hasn't been explicitly disabled by defining ZSTD_DISABLE_ASM
+ * - Assembly is supported
+ * - We are compiling for x86-64 and either:
+ *   - DYNAMIC_BMI2 is enabled
+ *   - BMI2 is supported at compile time
+ */
+#if !defined(ZSTD_DISABLE_ASM) &&                                 \
+    ZSTD_ASM_SUPPORTED &&                                         \
+    defined(__x86_64__) &&                                        \
+    (DYNAMIC_BMI2 || defined(__BMI2__))
+# define ZSTD_ENABLE_ASM_X86_64_BMI2 1
+#else
+# define ZSTD_ENABLE_ASM_X86_64_BMI2 0
+#endif
+
+/*
+ * For x86 ELF targets, add .note.gnu.property section for Intel CET in
+ * assembly sources when CET is enabled.
+ *
+ * Additionally, any function that may be called indirectly must begin
+ * with ZSTD_CET_ENDBRANCH.
+ */
+#if defined(__ELF__) && (defined(__x86_64__) || defined(__i386__)) \
+    && defined(__has_include)
+# if __has_include(<cet.h>)
+#  include <cet.h>
+#  define ZSTD_CET_ENDBRANCH _CET_ENDBR
+# endif
+#endif
+
+#ifndef ZSTD_CET_ENDBRANCH
+# define ZSTD_CET_ENDBRANCH
+#endif
+
+#endif /* ZSTD_PORTABILITY_MACROS_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/threading.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/threading.c
new file mode 100644
index 0000000..ca155b9
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/threading.c
@@ -0,0 +1,176 @@
+/**
+ * Copyright (c) 2016 Tino Reichardt
+ * All rights reserved.
+ *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/**
+ * This file will hold wrapper for systems, which do not support pthreads
+ */
+
+#include "threading.h"
+
+/* create fake symbol to avoid empty translation unit warning */
+int g_ZSTD_threading_useless_symbol;
+
+#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
+
+/**
+ * Windows minimalist Pthread Wrapper
+ */
+
+
+/* ===  Dependencies  === */
+#include <process.h>
+#include <errno.h>
+
+
+/* ===  Implementation  === */
+
+typedef struct {
+    void* (*start_routine)(void*);
+    void* arg;
+    int initialized;
+    ZSTD_pthread_cond_t initialized_cond;
+    ZSTD_pthread_mutex_t initialized_mutex;
+} ZSTD_thread_params_t;
+
+static unsigned __stdcall worker(void *arg)
+{
+    void* (*start_routine)(void*);
+    void* thread_arg;
+
+    /* Initialized thread_arg and start_routine and signal main thread that we don't need it
+     * to wait any longer.
+     */
+    {
+        ZSTD_thread_params_t*  thread_param = (ZSTD_thread_params_t*)arg;
+        thread_arg = thread_param->arg;
+        start_routine = thread_param->start_routine;
+
+        /* Signal main thread that we are running and do not depend on its memory anymore */
+        ZSTD_pthread_mutex_lock(&thread_param->initialized_mutex);
+        thread_param->initialized = 1;
+        ZSTD_pthread_cond_signal(&thread_param->initialized_cond);
+        ZSTD_pthread_mutex_unlock(&thread_param->initialized_mutex);
+    }
+
+    start_routine(thread_arg);
+
+    return 0;
+}
+
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
+            void* (*start_routine) (void*), void* arg)
+{
+    ZSTD_thread_params_t thread_param;
+    (void)unused;
+
+    thread_param.start_routine = start_routine;
+    thread_param.arg = arg;
+    thread_param.initialized = 0;
+    *thread = NULL;
+
+    /* Setup thread initialization synchronization */
+    if(ZSTD_pthread_cond_init(&thread_param.initialized_cond, NULL)) {
+        /* Should never happen on Windows */
+        return -1;
+    }
+    if(ZSTD_pthread_mutex_init(&thread_param.initialized_mutex, NULL)) {
+        /* Should never happen on Windows */
+        ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+        return -1;
+    }
+
+    /* Spawn thread */
+    *thread = (HANDLE)_beginthreadex(NULL, 0, worker, &thread_param, 0, NULL);
+    if (!thread) {
+        ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
+        ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+        return errno;
+    }
+
+    /* Wait for thread to be initialized */
+    ZSTD_pthread_mutex_lock(&thread_param.initialized_mutex);
+    while(!thread_param.initialized) {
+        ZSTD_pthread_cond_wait(&thread_param.initialized_cond, &thread_param.initialized_mutex);
+    }
+    ZSTD_pthread_mutex_unlock(&thread_param.initialized_mutex);
+    ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
+    ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+
+    return 0;
+}
+
+int ZSTD_pthread_join(ZSTD_pthread_t thread)
+{
+    DWORD result;
+
+    if (!thread) return 0;
+
+    result = WaitForSingleObject(thread, INFINITE);
+    CloseHandle(thread);
+
+    switch (result) {
+    case WAIT_OBJECT_0:
+        return 0;
+    case WAIT_ABANDONED:
+        return EINVAL;
+    default:
+        return GetLastError();
+    }
+}
+
+#endif   /* ZSTD_MULTITHREAD */
+
+#if defined(ZSTD_MULTITHREAD) && DEBUGLEVEL >= 1 && !defined(_WIN32)
+
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"
+
+int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr)
+{
+    *mutex = (pthread_mutex_t*)ZSTD_malloc(sizeof(pthread_mutex_t));
+    if (!*mutex)
+        return 1;
+    return pthread_mutex_init(*mutex, attr);
+}
+
+int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex)
+{
+    if (!*mutex)
+        return 0;
+    {
+        int const ret = pthread_mutex_destroy(*mutex);
+        ZSTD_free(*mutex);
+        return ret;
+    }
+}
+
+int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr)
+{
+    *cond = (pthread_cond_t*)ZSTD_malloc(sizeof(pthread_cond_t));
+    if (!*cond)
+        return 1;
+    return pthread_cond_init(*cond, attr);
+}
+
+int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond)
+{
+    if (!*cond)
+        return 0;
+    {
+        int const ret = pthread_cond_destroy(*cond);
+        ZSTD_free(*cond);
+        return ret;
+    }
+}
+
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/threading.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/threading.h
new file mode 100644
index 0000000..fb5c1c8
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/threading.h
@@ -0,0 +1,150 @@
+/**
+ * Copyright (c) 2016 Tino Reichardt
+ * All rights reserved.
+ *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef THREADING_H_938743
+#define THREADING_H_938743
+
+#include "debug.h"
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
+
+/**
+ * Windows minimalist Pthread Wrapper
+ */
+#ifdef WINVER
+#  undef WINVER
+#endif
+#define WINVER       0x0600
+
+#ifdef _WIN32_WINNT
+#  undef _WIN32_WINNT
+#endif
+#define _WIN32_WINNT 0x0600
+
+#ifndef WIN32_LEAN_AND_MEAN
+#  define WIN32_LEAN_AND_MEAN
+#endif
+
+#undef ERROR   /* reported already defined on VS 2015 (Rich Geldreich) */
+#include <windows.h>
+#undef ERROR
+#define ERROR(name) ZSTD_ERROR(name)
+
+
+/* mutex */
+#define ZSTD_pthread_mutex_t           CRITICAL_SECTION
+#define ZSTD_pthread_mutex_init(a, b)  ((void)(b), InitializeCriticalSection((a)), 0)
+#define ZSTD_pthread_mutex_destroy(a)  DeleteCriticalSection((a))
+#define ZSTD_pthread_mutex_lock(a)     EnterCriticalSection((a))
+#define ZSTD_pthread_mutex_unlock(a)   LeaveCriticalSection((a))
+
+/* condition variable */
+#define ZSTD_pthread_cond_t             CONDITION_VARIABLE
+#define ZSTD_pthread_cond_init(a, b)    ((void)(b), InitializeConditionVariable((a)), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
+#define ZSTD_pthread_cond_signal(a)     WakeConditionVariable((a))
+#define ZSTD_pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
+
+/* ZSTD_pthread_create() and ZSTD_pthread_join() */
+typedef HANDLE ZSTD_pthread_t;
+
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
+                   void* (*start_routine) (void*), void* arg);
+
+int ZSTD_pthread_join(ZSTD_pthread_t thread);
+
+/**
+ * add here more wrappers as required
+ */
+
+
+#elif defined(ZSTD_MULTITHREAD)    /* posix assumed ; need a better detection method */
+/* ===   POSIX Systems   === */
+#  include <pthread.h>
+
+#if DEBUGLEVEL < 1
+
+#define ZSTD_pthread_mutex_t            pthread_mutex_t
+#define ZSTD_pthread_mutex_init(a, b)   pthread_mutex_init((a), (b))
+#define ZSTD_pthread_mutex_destroy(a)   pthread_mutex_destroy((a))
+#define ZSTD_pthread_mutex_lock(a)      pthread_mutex_lock((a))
+#define ZSTD_pthread_mutex_unlock(a)    pthread_mutex_unlock((a))
+
+#define ZSTD_pthread_cond_t             pthread_cond_t
+#define ZSTD_pthread_cond_init(a, b)    pthread_cond_init((a), (b))
+#define ZSTD_pthread_cond_destroy(a)    pthread_cond_destroy((a))
+#define ZSTD_pthread_cond_wait(a, b)    pthread_cond_wait((a), (b))
+#define ZSTD_pthread_cond_signal(a)     pthread_cond_signal((a))
+#define ZSTD_pthread_cond_broadcast(a)  pthread_cond_broadcast((a))
+
+#define ZSTD_pthread_t                  pthread_t
+#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
+#define ZSTD_pthread_join(a)         pthread_join((a),NULL)
+
+#else /* DEBUGLEVEL >= 1 */
+
+/* Debug implementation of threading.
+ * In this implementation we use pointers for mutexes and condition variables.
+ * This way, if we forget to init/destroy them the program will crash or ASAN
+ * will report leaks.
+ */
+
+#define ZSTD_pthread_mutex_t            pthread_mutex_t*
+int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr);
+int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex);
+#define ZSTD_pthread_mutex_lock(a)      pthread_mutex_lock(*(a))
+#define ZSTD_pthread_mutex_unlock(a)    pthread_mutex_unlock(*(a))
+
+#define ZSTD_pthread_cond_t             pthread_cond_t*
+int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr);
+int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond);
+#define ZSTD_pthread_cond_wait(a, b)    pthread_cond_wait(*(a), *(b))
+#define ZSTD_pthread_cond_signal(a)     pthread_cond_signal(*(a))
+#define ZSTD_pthread_cond_broadcast(a)  pthread_cond_broadcast(*(a))
+
+#define ZSTD_pthread_t                  pthread_t
+#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
+#define ZSTD_pthread_join(a)         pthread_join((a),NULL)
+
+#endif
+
+#else  /* ZSTD_MULTITHREAD not defined */
+/* No multithreading support */
+
+typedef int ZSTD_pthread_mutex_t;
+#define ZSTD_pthread_mutex_init(a, b)   ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_mutex_destroy(a)   ((void)(a))
+#define ZSTD_pthread_mutex_lock(a)      ((void)(a))
+#define ZSTD_pthread_mutex_unlock(a)    ((void)(a))
+
+typedef int ZSTD_pthread_cond_t;
+#define ZSTD_pthread_cond_init(a, b)    ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    ((void)(a), (void)(b))
+#define ZSTD_pthread_cond_signal(a)     ((void)(a))
+#define ZSTD_pthread_cond_broadcast(a)  ((void)(a))
+
+/* do not use ZSTD_pthread_t */
+
+#endif /* ZSTD_MULTITHREAD */
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* THREADING_H_938743 */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/xxhash.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/xxhash.c
new file mode 100644
index 0000000..fd237c9
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/xxhash.c
@@ -0,0 +1,24 @@
+/*
+ *  xxHash - Fast Hash algorithm
+ *  Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - xxHash homepage: https://cyan4973.github.io/xxHash/
+ *  - xxHash source repository : https://github.com/Cyan4973/xxHash
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+*/
+
+
+
+/*
+ * xxhash.c instantiates functions defined in xxhash.h
+ */
+
+#define XXH_STATIC_LINKING_ONLY   /* access advanced declarations */
+#define XXH_IMPLEMENTATION   /* access definitions */
+
+#include "xxhash.h"
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/xxhash.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/xxhash.h
new file mode 100644
index 0000000..b8b7329
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/xxhash.h
@@ -0,0 +1,5686 @@
+/*
+ *  xxHash - Fast Hash algorithm
+ *  Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - xxHash homepage: https://cyan4973.github.io/xxHash/
+ *  - xxHash source repository : https://github.com/Cyan4973/xxHash
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+*/
+
+
+#ifndef XXH_NO_XXH3
+# define XXH_NO_XXH3
+#endif
+
+#ifndef XXH_NAMESPACE
+# define XXH_NAMESPACE ZSTD_
+#endif
+
+/*!
+ * @mainpage xxHash
+ *
+ * @file xxhash.h
+ * xxHash prototypes and implementation
+ */
+/* TODO: update */
+/* Notice extracted from xxHash homepage:
+
+xxHash is an extremely fast hash algorithm, running at RAM speed limits.
+It also successfully passes all tests from the SMHasher suite.
+
+Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz)
+
+Name            Speed       Q.Score   Author
+xxHash          5.4 GB/s     10
+CrapWow         3.2 GB/s      2       Andrew
+MurmurHash 3a   2.7 GB/s     10       Austin Appleby
+SpookyHash      2.0 GB/s     10       Bob Jenkins
+SBox            1.4 GB/s      9       Bret Mulvey
+Lookup3         1.2 GB/s      9       Bob Jenkins
+SuperFastHash   1.2 GB/s      1       Paul Hsieh
+CityHash64      1.05 GB/s    10       Pike & Alakuijala
+FNV             0.55 GB/s     5       Fowler, Noll, Vo
+CRC32           0.43 GB/s     9
+MD5-32          0.33 GB/s    10       Ronald L. Rivest
+SHA1-32         0.28 GB/s    10
+
+Q.Score is a measure of quality of the hash function.
+It depends on successfully passing SMHasher test set.
+10 is a perfect score.
+
+Note: SMHasher's CRC32 implementation is not the fastest one.
+Other speed-oriented implementations can be faster,
+especially in combination with PCLMUL instruction:
+https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html?showComment=1552696407071#c3490092340461170735
+
+A 64-bit version, named XXH64, is available since r35.
+It offers much better speed, but for 64-bit applications only.
+Name     Speed on 64 bits    Speed on 32 bits
+XXH64       13.8 GB/s            1.9 GB/s
+XXH32        6.8 GB/s            6.0 GB/s
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* ****************************
+ *  INLINE mode
+ ******************************/
+/*!
+ * XXH_INLINE_ALL (and XXH_PRIVATE_API)
+ * Use these build macros to inline xxhash into the target unit.
+ * Inlining improves performance on small inputs, especially when the length is
+ * expressed as a compile-time constant:
+ *
+ *      https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
+ *
+ * It also keeps xxHash symbols private to the unit, so they are not exported.
+ *
+ * Usage:
+ *     #define XXH_INLINE_ALL
+ *     #include "xxhash.h"
+ *
+ * Do not compile and link xxhash.o as a separate object, as it is not useful.
+ */
+#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
+    && !defined(XXH_INLINE_ALL_31684351384)
+   /* this section should be traversed only once */
+#  define XXH_INLINE_ALL_31684351384
+   /* give access to the advanced API, required to compile implementations */
+#  undef XXH_STATIC_LINKING_ONLY   /* avoid macro redef */
+#  define XXH_STATIC_LINKING_ONLY
+   /* make all functions private */
+#  undef XXH_PUBLIC_API
+#  if defined(__GNUC__)
+#    define XXH_PUBLIC_API static __inline __attribute__((unused))
+#  elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#    define XXH_PUBLIC_API static inline
+#  elif defined(_MSC_VER)
+#    define XXH_PUBLIC_API static __inline
+#  else
+     /* note: this version may generate warnings for unused static functions */
+#    define XXH_PUBLIC_API static
+#  endif
+
+   /*
+    * This part deals with the special case where a unit wants to inline xxHash,
+    * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
+    * such as part of some previously included *.h header file.
+    * Without further action, the new include would just be ignored,
+    * and functions would effectively _not_ be inlined (silent failure).
+    * The following macros solve this situation by prefixing all inlined names,
+    * avoiding naming collision with previous inclusions.
+    */
+   /* Before that, we unconditionally #undef all symbols,
+    * in case they were already defined with XXH_NAMESPACE.
+    * They will then be redefined for XXH_INLINE_ALL
+    */
+#  undef XXH_versionNumber
+    /* XXH32 */
+#  undef XXH32
+#  undef XXH32_createState
+#  undef XXH32_freeState
+#  undef XXH32_reset
+#  undef XXH32_update
+#  undef XXH32_digest
+#  undef XXH32_copyState
+#  undef XXH32_canonicalFromHash
+#  undef XXH32_hashFromCanonical
+    /* XXH64 */
+#  undef XXH64
+#  undef XXH64_createState
+#  undef XXH64_freeState
+#  undef XXH64_reset
+#  undef XXH64_update
+#  undef XXH64_digest
+#  undef XXH64_copyState
+#  undef XXH64_canonicalFromHash
+#  undef XXH64_hashFromCanonical
+    /* XXH3_64bits */
+#  undef XXH3_64bits
+#  undef XXH3_64bits_withSecret
+#  undef XXH3_64bits_withSeed
+#  undef XXH3_64bits_withSecretandSeed
+#  undef XXH3_createState
+#  undef XXH3_freeState
+#  undef XXH3_copyState
+#  undef XXH3_64bits_reset
+#  undef XXH3_64bits_reset_withSeed
+#  undef XXH3_64bits_reset_withSecret
+#  undef XXH3_64bits_update
+#  undef XXH3_64bits_digest
+#  undef XXH3_generateSecret
+    /* XXH3_128bits */
+#  undef XXH128
+#  undef XXH3_128bits
+#  undef XXH3_128bits_withSeed
+#  undef XXH3_128bits_withSecret
+#  undef XXH3_128bits_reset
+#  undef XXH3_128bits_reset_withSeed
+#  undef XXH3_128bits_reset_withSecret
+#  undef XXH3_128bits_reset_withSecretandSeed
+#  undef XXH3_128bits_update
+#  undef XXH3_128bits_digest
+#  undef XXH128_isEqual
+#  undef XXH128_cmp
+#  undef XXH128_canonicalFromHash
+#  undef XXH128_hashFromCanonical
+    /* Finally, free the namespace itself */
+#  undef XXH_NAMESPACE
+
+    /* employ the namespace for XXH_INLINE_ALL */
+#  define XXH_NAMESPACE XXH_INLINE_
+   /*
+    * Some identifiers (enums, type names) are not symbols,
+    * but they must nonetheless be renamed to avoid redeclaration.
+    * Alternative solution: do not redeclare them.
+    * However, this requires some #ifdefs, and has a more dispersed impact.
+    * Meanwhile, renaming can be achieved in a single place.
+    */
+#  define XXH_IPREF(Id)   XXH_NAMESPACE ## Id
+#  define XXH_OK XXH_IPREF(XXH_OK)
+#  define XXH_ERROR XXH_IPREF(XXH_ERROR)
+#  define XXH_errorcode XXH_IPREF(XXH_errorcode)
+#  define XXH32_canonical_t  XXH_IPREF(XXH32_canonical_t)
+#  define XXH64_canonical_t  XXH_IPREF(XXH64_canonical_t)
+#  define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
+#  define XXH32_state_s XXH_IPREF(XXH32_state_s)
+#  define XXH32_state_t XXH_IPREF(XXH32_state_t)
+#  define XXH64_state_s XXH_IPREF(XXH64_state_s)
+#  define XXH64_state_t XXH_IPREF(XXH64_state_t)
+#  define XXH3_state_s  XXH_IPREF(XXH3_state_s)
+#  define XXH3_state_t  XXH_IPREF(XXH3_state_t)
+#  define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
+   /* Ensure the header is parsed again, even if it was previously included */
+#  undef XXHASH_H_5627135585666179
+#  undef XXHASH_H_STATIC_13879238742
+#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */
+
+
+
+/* ****************************************************************
+ *  Stable API
+ *****************************************************************/
+#ifndef XXHASH_H_5627135585666179
+#define XXHASH_H_5627135585666179 1
+
+
+/*!
+ * @defgroup public Public API
+ * Contains details on the public xxHash functions.
+ * @{
+ */
+/* specific declaration modes for Windows */
+#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+#    ifdef XXH_EXPORT
+#      define XXH_PUBLIC_API __declspec(dllexport)
+#    elif XXH_IMPORT
+#      define XXH_PUBLIC_API __declspec(dllimport)
+#    endif
+#  else
+#    define XXH_PUBLIC_API   /* do nothing */
+#  endif
+#endif
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Emulate a namespace by transparently prefixing all symbols.
+ *
+ * If you want to include _and expose_ xxHash functions from within your own
+ * library, but also want to avoid symbol collisions with other libraries which
+ * may also include xxHash, you can use XXH_NAMESPACE to automatically prefix
+ * any public symbol from xxhash library with the value of XXH_NAMESPACE
+ * (therefore, avoid empty or numeric values).
+ *
+ * Note that no change is required within the calling program as long as it
+ * includes `xxhash.h`: Regular symbol names will be automatically translated
+ * by this header.
+ */
+#  define XXH_NAMESPACE /* YOUR NAME HERE */
+#  undef XXH_NAMESPACE
+#endif
+
+#ifdef XXH_NAMESPACE
+#  define XXH_CAT(A,B) A##B
+#  define XXH_NAME2(A,B) XXH_CAT(A,B)
+#  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+/* XXH32 */
+#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
+#  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
+#  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
+#  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
+#  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
+#  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
+#  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
+#  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
+#  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+/* XXH64 */
+#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
+#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
+#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
+#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
+#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
+#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
+#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
+#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
+#  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
+/* XXH3_64bits */
+#  define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
+#  define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
+#  define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
+#  define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
+#  define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
+#  define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
+#  define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
+#  define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
+#  define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
+#  define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
+#  define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
+#  define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
+#  define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
+#  define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
+#  define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
+/* XXH3_128bits */
+#  define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
+#  define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
+#  define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
+#  define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
+#  define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
+#  define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
+#  define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
+#  define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
+#  define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
+#  define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
+#  define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
+#  define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
+#  define XXH128_cmp     XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
+#  define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
+#  define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
+#endif
+
+
+/* *************************************
+*  Version
+***************************************/
+#define XXH_VERSION_MAJOR    0
+#define XXH_VERSION_MINOR    8
+#define XXH_VERSION_RELEASE  1
+#define XXH_VERSION_NUMBER  (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
+
+/*!
+ * @brief Obtains the xxHash version.
+ *
+ * This is mostly useful when xxHash is compiled as a shared library,
+ * since the returned value comes from the library, as opposed to header file.
+ *
+ * @return `XXH_VERSION_NUMBER` of the invoked library.
+ */
+XXH_PUBLIC_API unsigned XXH_versionNumber (void);
+
+
+/* ****************************
+*  Common basic types
+******************************/
+#include <stddef.h>   /* size_t */
+typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
+
+
+/*-**********************************************************************
+*  32-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* Don't show <stdint.h> include */
+/*!
+ * @brief An unsigned 32-bit integer.
+ *
+ * Not necessarily defined to `uint32_t` but functionally equivalent.
+ */
+typedef uint32_t XXH32_hash_t;
+
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#   include <stdint.h>
+    typedef uint32_t XXH32_hash_t;
+
+#else
+#   include <limits.h>
+#   if UINT_MAX == 0xFFFFFFFFUL
+      typedef unsigned int XXH32_hash_t;
+#   else
+#     if ULONG_MAX == 0xFFFFFFFFUL
+        typedef unsigned long XXH32_hash_t;
+#     else
+#       error "unsupported platform: need a 32-bit type"
+#     endif
+#   endif
+#endif
+
+/*!
+ * @}
+ *
+ * @defgroup xxh32_family XXH32 family
+ * @ingroup public
+ * Contains functions used in the classic 32-bit xxHash algorithm.
+ *
+ * @note
+ *   XXH32 is useful for older platforms, with no or poor 64-bit performance.
+ *   Note that @ref xxh3_family provides competitive speed
+ *   for both 32-bit and 64-bit systems, and offers true 64/128 bit hash results.
+ *
+ * @see @ref xxh64_family, @ref xxh3_family : Other xxHash families
+ * @see @ref xxh32_impl for implementation details
+ * @{
+ */
+
+/*!
+ * @brief Calculates the 32-bit hash of @p input using xxHash32.
+ *
+ * Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark): 5.4 GB/s
+ *
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ * @param seed The 32-bit seed to alter the hash's output predictably.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 32-bit hash value.
+ *
+ * @see
+ *    XXH64(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
+ *    Direct equivalents for the other variants of xxHash.
+ * @see
+ *    XXH32_createState(), XXH32_update(), XXH32_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
+
+/*!
+ * Streaming functions generate the xxHash value from an incremental input.
+ * This method is slower than single-call functions, due to state management.
+ * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized.
+ *
+ * An XXH state must first be allocated using `XXH*_createState()`.
+ *
+ * Start a new hash by initializing the state with a seed using `XXH*_reset()`.
+ *
+ * Then, feed the hash state by calling `XXH*_update()` as many times as necessary.
+ *
+ * The function returns an error code, with 0 meaning OK, and any other value
+ * meaning there is an error.
+ *
+ * Finally, a hash value can be produced anytime, by using `XXH*_digest()`.
+ * This function returns the nn-bits hash as an int or long long.
+ *
+ * It's still possible to continue inserting input into the hash state after a
+ * digest, and generate new hash values later on by invoking `XXH*_digest()`.
+ *
+ * When done, release the state using `XXH*_freeState()`.
+ *
+ * Example code for incrementally hashing a file:
+ * @code{.c}
+ *    #include <stdio.h>
+ *    #include <xxhash.h>
+ *    #define BUFFER_SIZE 256
+ *
+ *    // Note: XXH64 and XXH3 use the same interface.
+ *    XXH32_hash_t
+ *    hashFile(FILE* stream)
+ *    {
+ *        XXH32_state_t* state;
+ *        unsigned char buf[BUFFER_SIZE];
+ *        size_t amt;
+ *        XXH32_hash_t hash;
+ *
+ *        state = XXH32_createState();       // Create a state
+ *        assert(state != NULL);             // Error check here
+ *        XXH32_reset(state, 0xbaad5eed);    // Reset state with our seed
+ *        while ((amt = fread(buf, 1, sizeof(buf), stream)) != 0) {
+ *            XXH32_update(state, buf, amt); // Hash the file in chunks
+ *        }
+ *        hash = XXH32_digest(state);        // Finalize the hash
+ *        XXH32_freeState(state);            // Clean up
+ *        return hash;
+ *    }
+ * @endcode
+ */
+
+/*!
+ * @typedef struct XXH32_state_s XXH32_state_t
+ * @brief The opaque state struct for the XXH32 streaming API.
+ *
+ * @see XXH32_state_s for details.
+ */
+typedef struct XXH32_state_s XXH32_state_t;
+
+/*!
+ * @brief Allocates an @ref XXH32_state_t.
+ *
+ * Must be freed with XXH32_freeState().
+ * @return An allocated XXH32_state_t on success, `NULL` on failure.
+ */
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
+/*!
+ * @brief Frees an @ref XXH32_state_t.
+ *
+ * Must be allocated with XXH32_createState().
+ * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState().
+ * @return XXH_OK.
+ */
+XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
+/*!
+ * @brief Copies one @ref XXH32_state_t to another.
+ *
+ * @param dst_state The state to copy to.
+ * @param src_state The state to copy from.
+ * @pre
+ *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
+ */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state);
+
+/*!
+ * @brief Resets an @ref XXH32_state_t to begin a new hash.
+ *
+ * This function resets and seeds a state. Call it before @ref XXH32_update().
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed The 32-bit seed to alter the hash result predictably.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, XXH32_hash_t seed);
+
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH32_state_t.
+ *
+ * Call this to incrementally consume blocks of data.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated hash value from an @ref XXH32_state_t.
+ *
+ * @note
+ *   Calling XXH32_digest() will not affect @p statePtr, so you can update,
+ *   digest, and update again.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ *  @p statePtr must not be `NULL`.
+ *
+ * @return The calculated xxHash32 value from that state.
+ */
+XXH_PUBLIC_API XXH32_hash_t  XXH32_digest (const XXH32_state_t* statePtr);
+
+/*******   Canonical representation   *******/
+
+/*
+ * The default return values from XXH functions are unsigned 32 and 64 bit
+ * integers.
+ * This the simplest and fastest format for further post-processing.
+ *
+ * However, this leaves open the question of what is the order on the byte level,
+ * since little and big endian conventions will store the same number differently.
+ *
+ * The canonical representation settles this issue by mandating big-endian
+ * convention, the same convention as human-readable numbers (large digits first).
+ *
+ * When writing hash values to storage, sending them over a network, or printing
+ * them, it's highly recommended to use the canonical representation to ensure
+ * portability across a wider range of systems, present and future.
+ *
+ * The following functions allow transformation of hash values to and from
+ * canonical format.
+ */
+
+/*!
+ * @brief Canonical (big endian) representation of @ref XXH32_hash_t.
+ */
+typedef struct {
+    unsigned char digest[4]; /*!< Hash bytes, big endian */
+} XXH32_canonical_t;
+
+/*!
+ * @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t.
+ *
+ * @param dst The @ref XXH32_canonical_t pointer to be stored to.
+ * @param hash The @ref XXH32_hash_t to be converted.
+ *
+ * @pre
+ *   @p dst must not be `NULL`.
+ */
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
+
+/*!
+ * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t.
+ *
+ * @param src The @ref XXH32_canonical_t to convert.
+ *
+ * @pre
+ *   @p src must not be `NULL`.
+ *
+ * @return The converted hash.
+ */
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
+
+
+#ifdef __has_attribute
+# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
+#else
+# define XXH_HAS_ATTRIBUTE(x) 0
+#endif
+
+/* C-language Attributes are added in C23. */
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
+# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
+#else
+# define XXH_HAS_C_ATTRIBUTE(x) 0
+#endif
+
+#if defined(__cplusplus) && defined(__has_cpp_attribute)
+# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+#else
+# define XXH_HAS_CPP_ATTRIBUTE(x) 0
+#endif
+
+/*
+Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
+introduced in CPP17 and C23.
+CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
+C23   : https://en.cppreference.com/w/c/language/attributes/fallthrough
+*/
+#if XXH_HAS_C_ATTRIBUTE(x)
+# define XXH_FALLTHROUGH [[fallthrough]]
+#elif XXH_HAS_CPP_ATTRIBUTE(x)
+# define XXH_FALLTHROUGH [[fallthrough]]
+#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
+# define XXH_FALLTHROUGH __attribute__ ((fallthrough))
+#else
+# define XXH_FALLTHROUGH
+#endif
+
+/*!
+ * @}
+ * @ingroup public
+ * @{
+ */
+
+#ifndef XXH_NO_LONG_LONG
+/*-**********************************************************************
+*  64-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* don't include <stdint.h> */
+/*!
+ * @brief An unsigned 64-bit integer.
+ *
+ * Not necessarily defined to `uint64_t` but functionally equivalent.
+ */
+typedef uint64_t XXH64_hash_t;
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#  include <stdint.h>
+   typedef uint64_t XXH64_hash_t;
+#else
+#  include <limits.h>
+#  if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
+     /* LP64 ABI says uint64_t is unsigned long */
+     typedef unsigned long XXH64_hash_t;
+#  else
+     /* the following type must have a width of 64-bit */
+     typedef unsigned long long XXH64_hash_t;
+#  endif
+#endif
+
+/*!
+ * @}
+ *
+ * @defgroup xxh64_family XXH64 family
+ * @ingroup public
+ * @{
+ * Contains functions used in the classic 64-bit xxHash algorithm.
+ *
+ * @note
+ *   XXH3 provides competitive speed for both 32-bit and 64-bit systems,
+ *   and offers true 64/128 bit hash results.
+ *   It provides better speed for systems with vector processing capabilities.
+ */
+
+
+/*!
+ * @brief Calculates the 64-bit hash of @p input using xxHash64.
+ *
+ * This function usually runs faster on 64-bit systems, but slower on 32-bit
+ * systems (see benchmark).
+ *
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ * @param seed The 64-bit seed to alter the hash's output predictably.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 64-bit hash.
+ *
+ * @see
+ *    XXH32(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
+ *    Direct equivalents for the other variants of xxHash.
+ * @see
+ *    XXH64_createState(), XXH64_update(), XXH64_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH64(const void* input, size_t length, XXH64_hash_t seed);
+
+/*******   Streaming   *******/
+/*!
+ * @brief The opaque state struct for the XXH64 streaming API.
+ *
+ * @see XXH64_state_s for details.
+ */
+typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
+XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dst_state, const XXH64_state_t* src_state);
+
+XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH64_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH64_digest (const XXH64_state_t* statePtr);
+
+/*******   Canonical representation   *******/
+typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
+
+#ifndef XXH_NO_XXH3
+/*!
+ * @}
+ * ************************************************************************
+ * @defgroup xxh3_family XXH3 family
+ * @ingroup public
+ * @{
+ *
+ * XXH3 is a more recent hash algorithm featuring:
+ *  - Improved speed for both small and large inputs
+ *  - True 64-bit and 128-bit outputs
+ *  - SIMD acceleration
+ *  - Improved 32-bit viability
+ *
+ * Speed analysis methodology is explained here:
+ *
+ *    https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html
+ *
+ * Compared to XXH64, expect XXH3 to run approximately
+ * ~2x faster on large inputs and >3x faster on small ones,
+ * exact differences vary depending on platform.
+ *
+ * XXH3's speed benefits greatly from SIMD and 64-bit arithmetic,
+ * but does not require it.
+ * Any 32-bit and 64-bit targets that can run XXH32 smoothly
+ * can run XXH3 at competitive speeds, even without vector support.
+ * Further details are explained in the implementation.
+ *
+ * Optimized implementations are provided for AVX512, AVX2, SSE2, NEON, POWER8,
+ * ZVector and scalar targets. This can be controlled via the XXH_VECTOR macro.
+ *
+ * XXH3 implementation is portable:
+ * it has a generic C90 formulation that can be compiled on any platform,
+ * all implementations generage exactly the same hash value on all platforms.
+ * Starting from v0.8.0, it's also labelled "stable", meaning that
+ * any future version will also generate the same hash value.
+ *
+ * XXH3 offers 2 variants, _64bits and _128bits.
+ *
+ * When only 64 bits are needed, prefer invoking the _64bits variant, as it
+ * reduces the amount of mixing, resulting in faster speed on small inputs.
+ * It's also generally simpler to manipulate a scalar return type than a struct.
+ *
+ * The API supports one-shot hashing, streaming mode, and custom secrets.
+ */
+
+/*-**********************************************************************
+*  XXH3 64-bit variant
+************************************************************************/
+
+/* XXH3_64bits():
+ * default 64-bit variant, using default secret and default seed of 0.
+ * It's the fastest variant. */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* data, size_t len);
+
+/*
+ * XXH3_64bits_withSeed():
+ * This variant generates a custom secret on the fly
+ * based on default secret altered using the `seed` value.
+ * While this operation is decently fast, note that it's not completely free.
+ * Note: seed==0 produces the same results as XXH3_64bits().
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
+
+/*!
+ * The bare minimum size for a custom secret.
+ *
+ * @see
+ *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
+ *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
+ */
+#define XXH3_SECRET_SIZE_MIN 136
+
+/*
+ * XXH3_64bits_withSecret():
+ * It's possible to provide any blob of bytes as a "secret" to generate the hash.
+ * This makes it more difficult for an external actor to prepare an intentional collision.
+ * The main condition is that secretSize *must* be large enough (>= XXH3_SECRET_SIZE_MIN).
+ * However, the quality of the secret impacts the dispersion of the hash algorithm.
+ * Therefore, the secret _must_ look like a bunch of random bytes.
+ * Avoid "trivial" or structured data such as repeated sequences or a text document.
+ * Whenever in doubt about the "randomness" of the blob of bytes,
+ * consider employing "XXH3_generateSecret()" instead (see below).
+ * It will generate a proper high entropy secret derived from the blob of bytes.
+ * Another advantage of using XXH3_generateSecret() is that
+ * it guarantees that all bits within the initial blob of bytes
+ * will impact every bit of the output.
+ * This is not necessarily the case when using the blob of bytes directly
+ * because, when hashing _small_ inputs, only a portion of the secret is employed.
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+
+
+/*******   Streaming   *******/
+/*
+ * Streaming requires state maintenance.
+ * This operation costs memory and CPU.
+ * As a consequence, streaming is slower than one-shot hashing.
+ * For better performance, prefer one-shot functions whenever applicable.
+ */
+
+/*!
+ * @brief The state struct for the XXH3 streaming API.
+ *
+ * @see XXH3_state_s for details.
+ */
+typedef struct XXH3_state_s XXH3_state_t;
+XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void);
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
+XXH_PUBLIC_API void XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state);
+
+/*
+ * XXH3_64bits_reset():
+ * Initialize with default parameters.
+ * digest will be equivalent to `XXH3_64bits()`.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t* statePtr);
+/*
+ * XXH3_64bits_reset_withSeed():
+ * Generate a custom secret from `seed`, and store it into `statePtr`.
+ * digest will be equivalent to `XXH3_64bits_withSeed()`.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
+/*
+ * XXH3_64bits_reset_withSecret():
+ * `secret` is referenced, it _must outlive_ the hash streaming session.
+ * Similar to one-shot API, `secretSize` must be >= `XXH3_SECRET_SIZE_MIN`,
+ * and the quality of produced hash values depends on secret's entropy
+ * (secret's content should look like a bunch of random bytes).
+ * When in doubt about the randomness of a candidate `secret`,
+ * consider employing `XXH3_generateSecret()` instead (see below).
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_digest (const XXH3_state_t* statePtr);
+
+/* note : canonical representation of XXH3 is the same as XXH64
+ * since they both produce XXH64_hash_t values */
+
+
+/*-**********************************************************************
+*  XXH3 128-bit variant
+************************************************************************/
+
+/*!
+ * @brief The return value from 128-bit hashes.
+ *
+ * Stored in little endian order, although the fields themselves are in native
+ * endianness.
+ */
+typedef struct {
+    XXH64_hash_t low64;   /*!< `value & 0xFFFFFFFFFFFFFFFF` */
+    XXH64_hash_t high64;  /*!< `value >> 64` */
+} XXH128_hash_t;
+
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* data, size_t len);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+
+/*******   Streaming   *******/
+/*
+ * Streaming requires state maintenance.
+ * This operation costs memory and CPU.
+ * As a consequence, streaming is slower than one-shot hashing.
+ * For better performance, prefer one-shot functions whenever applicable.
+ *
+ * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits().
+ * Use already declared XXH3_createState() and XXH3_freeState().
+ *
+ * All reset and streaming functions have same meaning as their 64-bit counterpart.
+ */
+
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t* statePtr);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* statePtr);
+
+/* Following helper functions make it possible to compare XXH128_hast_t values.
+ * Since XXH128_hash_t is a structure, this capability is not offered by the language.
+ * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */
+
+/*!
+ * XXH128_isEqual():
+ * Return: 1 if `h1` and `h2` are equal, 0 if they are not.
+ */
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
+
+/*!
+ * XXH128_cmp():
+ *
+ * This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
+ *
+ * return: >0 if *h128_1  > *h128_2
+ *         =0 if *h128_1 == *h128_2
+ *         <0 if *h128_1  < *h128_2
+ */
+XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2);
+
+
+/*******   Canonical representation   *******/
+typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;
+XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash);
+XXH_PUBLIC_API XXH128_hash_t XXH128_hashFromCanonical(const XXH128_canonical_t* src);
+
+
+#endif  /* !XXH_NO_XXH3 */
+#endif  /* XXH_NO_LONG_LONG */
+
+/*!
+ * @}
+ */
+#endif /* XXHASH_H_5627135585666179 */
+
+
+
+#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742)
+#define XXHASH_H_STATIC_13879238742
+/* ****************************************************************************
+ * This section contains declarations which are not guaranteed to remain stable.
+ * They may change in future versions, becoming incompatible with a different
+ * version of the library.
+ * These declarations should only be used with static linking.
+ * Never use them in association with dynamic linking!
+ ***************************************************************************** */
+
+/*
+ * These definitions are only present to allow static allocation
+ * of XXH states, on stack or in a struct, for example.
+ * Never **ever** access their members directly.
+ */
+
+/*!
+ * @internal
+ * @brief Structure for XXH32 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
+ * an opaque type. This allows fields to safely be changed.
+ *
+ * Typedef'd to @ref XXH32_state_t.
+ * Do not access the members of this struct directly.
+ * @see XXH64_state_s, XXH3_state_s
+ */
+struct XXH32_state_s {
+   XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */
+   XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
+   XXH32_hash_t v[4];         /*!< Accumulator lanes */
+   XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem32 */
+   XXH32_hash_t reserved;     /*!< Reserved field. Do not read nor write to it. */
+};   /* typedef'd to XXH32_state_t */
+
+
+#ifndef XXH_NO_LONG_LONG  /* defined when there is no 64-bit support */
+
+/*!
+ * @internal
+ * @brief Structure for XXH64 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
+ * an opaque type. This allows fields to safely be changed.
+ *
+ * Typedef'd to @ref XXH64_state_t.
+ * Do not access the members of this struct directly.
+ * @see XXH32_state_s, XXH3_state_s
+ */
+struct XXH64_state_s {
+   XXH64_hash_t total_len;    /*!< Total length hashed. This is always 64-bit. */
+   XXH64_hash_t v[4];         /*!< Accumulator lanes */
+   XXH64_hash_t mem64[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem64 */
+   XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
+   XXH64_hash_t reserved64;   /*!< Reserved field. Do not read or write to it. */
+};   /* typedef'd to XXH64_state_t */
+
+
+#ifndef XXH_NO_XXH3
+
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
+#  include <stdalign.h>
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */
+/* In C++ alignas() is a keyword */
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__GNUC__)
+#  define XXH_ALIGN(n)      __attribute__ ((aligned(n)))
+#elif defined(_MSC_VER)
+#  define XXH_ALIGN(n)      __declspec(align(n))
+#else
+#  define XXH_ALIGN(n)   /* disabled */
+#endif
+
+/* Old GCC versions only accept the attribute after the type in structures. */
+#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L))   /* C11+ */ \
+    && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
+    && defined(__GNUC__)
+#   define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
+#else
+#   define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
+#endif
+
+/*!
+ * @brief The size of the internal XXH3 buffer.
+ *
+ * This is the optimal update size for incremental hashing.
+ *
+ * @see XXH3_64b_update(), XXH3_128b_update().
+ */
+#define XXH3_INTERNALBUFFER_SIZE 256
+
+/*!
+ * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
+ *
+ * This is the size used in @ref XXH3_kSecret and the seeded functions.
+ *
+ * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
+ */
+#define XXH3_SECRET_DEFAULT_SIZE 192
+
+/*!
+ * @internal
+ * @brief Structure for XXH3 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined.
+ * Otherwise it is an opaque type.
+ * Never use this definition in combination with dynamic library.
+ * This allows fields to safely be changed in the future.
+ *
+ * @note ** This structure has a strict alignment requirement of 64 bytes!! **
+ * Do not allocate this with `malloc()` or `new`,
+ * it will not be sufficiently aligned.
+ * Use @ref XXH3_createState() and @ref XXH3_freeState(), or stack allocation.
+ *
+ * Typedef'd to @ref XXH3_state_t.
+ * Do never access the members of this struct directly.
+ *
+ * @see XXH3_INITSTATE() for stack initialization.
+ * @see XXH3_createState(), XXH3_freeState().
+ * @see XXH32_state_s, XXH64_state_s
+ */
+struct XXH3_state_s {
+   XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
+       /*!< The 8 accumulators. Similar to `vN` in @ref XXH32_state_s::v1 and @ref XXH64_state_s */
+   XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
+       /*!< Used to store a custom secret generated from a seed. */
+   XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
+       /*!< The internal buffer. @see XXH32_state_s::mem32 */
+   XXH32_hash_t bufferedSize;
+       /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
+   XXH32_hash_t useSeed;
+       /*!< Reserved field. Needed for padding on 64-bit. */
+   size_t nbStripesSoFar;
+       /*!< Number or stripes processed. */
+   XXH64_hash_t totalLen;
+       /*!< Total length hashed. 64-bit even on 32-bit targets. */
+   size_t nbStripesPerBlock;
+       /*!< Number of stripes per block. */
+   size_t secretLimit;
+       /*!< Size of @ref customSecret or @ref extSecret */
+   XXH64_hash_t seed;
+       /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */
+   XXH64_hash_t reserved64;
+       /*!< Reserved field. */
+   const unsigned char* extSecret;
+       /*!< Reference to an external secret for the _withSecret variants, NULL
+        *   for other variants. */
+   /* note: there may be some padding at the end due to alignment on 64 bytes */
+}; /* typedef'd to XXH3_state_t */
+
+#undef XXH_ALIGN_MEMBER
+
+/*!
+ * @brief Initializes a stack-allocated `XXH3_state_s`.
+ *
+ * When the @ref XXH3_state_t structure is merely emplaced on stack,
+ * it should be initialized with XXH3_INITSTATE() or a memset()
+ * in case its first reset uses XXH3_NNbits_reset_withSeed().
+ * This init can be omitted if the first reset uses default or _withSecret mode.
+ * This operation isn't necessary when the state is created with XXH3_createState().
+ * Note that this doesn't prepare the state for a streaming operation,
+ * it's still necessary to use XXH3_NNbits_reset*() afterwards.
+ */
+#define XXH3_INITSTATE(XXH3_state_ptr)   { (XXH3_state_ptr)->seed = 0; }
+
+
+/* XXH128() :
+ * simple alias to pre-selected XXH3_128bits variant
+ */
+XXH_PUBLIC_API XXH128_hash_t XXH128(const void* data, size_t len, XXH64_hash_t seed);
+
+
+/* ===   Experimental API   === */
+/* Symbols defined below must be considered tied to a specific library version. */
+
+/*
+ * XXH3_generateSecret():
+ *
+ * Derive a high-entropy secret from any user-defined content, named customSeed.
+ * The generated secret can be used in combination with `*_withSecret()` functions.
+ * The `_withSecret()` variants are useful to provide a higher level of protection than 64-bit seed,
+ * as it becomes much more difficult for an external actor to guess how to impact the calculation logic.
+ *
+ * The function accepts as input a custom seed of any length and any content,
+ * and derives from it a high-entropy secret of length @secretSize
+ * into an already allocated buffer @secretBuffer.
+ * @secretSize must be >= XXH3_SECRET_SIZE_MIN
+ *
+ * The generated secret can then be used with any `*_withSecret()` variant.
+ * Functions `XXH3_128bits_withSecret()`, `XXH3_64bits_withSecret()`,
+ * `XXH3_128bits_reset_withSecret()` and `XXH3_64bits_reset_withSecret()`
+ * are part of this list. They all accept a `secret` parameter
+ * which must be large enough for implementation reasons (>= XXH3_SECRET_SIZE_MIN)
+ * _and_ feature very high entropy (consist of random-looking bytes).
+ * These conditions can be a high bar to meet, so
+ * XXH3_generateSecret() can be employed to ensure proper quality.
+ *
+ * customSeed can be anything. It can have any size, even small ones,
+ * and its content can be anything, even "poor entropy" sources such as a bunch of zeroes.
+ * The resulting `secret` will nonetheless provide all required qualities.
+ *
+ * When customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize);
+
+
+/*
+ * XXH3_generateSecret_fromSeed():
+ *
+ * Generate the same secret as the _withSeed() variants.
+ *
+ * The resulting secret has a length of XXH3_SECRET_DEFAULT_SIZE (necessarily).
+ * @secretBuffer must be already allocated, of size at least XXH3_SECRET_DEFAULT_SIZE bytes.
+ *
+ * The generated secret can be used in combination with
+ *`*_withSecret()` and `_withSecretandSeed()` variants.
+ * This generator is notably useful in combination with `_withSecretandSeed()`,
+ * as a way to emulate a faster `_withSeed()` variant.
+ */
+XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed);
+
+/*
+ * *_withSecretandSeed() :
+ * These variants generate hash values using either
+ * @seed for "short" keys (< XXH3_MIDSIZE_MAX = 240 bytes)
+ * or @secret for "large" keys (>= XXH3_MIDSIZE_MAX).
+ *
+ * This generally benefits speed, compared to `_withSeed()` or `_withSecret()`.
+ * `_withSeed()` has to generate the secret on the fly for "large" keys.
+ * It's fast, but can be perceptible for "not so large" keys (< 1 KB).
+ * `_withSecret()` has to generate the masks on the fly for "small" keys,
+ * which requires more instructions than _withSeed() variants.
+ * Therefore, _withSecretandSeed variant combines the best of both worlds.
+ *
+ * When @secret has been generated by XXH3_generateSecret_fromSeed(),
+ * this variant produces *exactly* the same results as `_withSeed()` variant,
+ * hence offering only a pure speed benefit on "large" input,
+ * by skipping the need to regenerate the secret for every large input.
+ *
+ * Another usage scenario is to hash the secret to a 64-bit hash value,
+ * for example with XXH3_64bits(), which then becomes the seed,
+ * and then employ both the seed and the secret in _withSecretandSeed().
+ * On top of speed, an added benefit is that each bit in the secret
+ * has a 50% chance to swap each bit in the output,
+ * via its impact to the seed.
+ * This is not guaranteed when using the secret directly in "small data" scenarios,
+ * because only portions of the secret are employed for small data.
+ */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecretandSeed(const void* data, size_t len,
+                              const void* secret, size_t secretSize,
+                              XXH64_hash_t seed);
+
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecretandSeed(const void* data, size_t len,
+                               const void* secret, size_t secretSize,
+                               XXH64_hash_t seed64);
+
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
+                                    const void* secret, size_t secretSize,
+                                    XXH64_hash_t seed64);
+
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
+                                     const void* secret, size_t secretSize,
+                                     XXH64_hash_t seed64);
+
+
+#endif  /* XXH_NO_XXH3 */
+#endif  /* XXH_NO_LONG_LONG */
+#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
+#  define XXH_IMPLEMENTATION
+#endif
+
+#endif  /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */
+
+
+/* ======================================================================== */
+/* ======================================================================== */
+/* ======================================================================== */
+
+
+/*-**********************************************************************
+ * xxHash implementation
+ *-**********************************************************************
+ * xxHash's implementation used to be hosted inside xxhash.c.
+ *
+ * However, inlining requires implementation to be visible to the compiler,
+ * hence be included alongside the header.
+ * Previously, implementation was hosted inside xxhash.c,
+ * which was then #included when inlining was activated.
+ * This construction created issues with a few build and install systems,
+ * as it required xxhash.c to be stored in /include directory.
+ *
+ * xxHash implementation is now directly integrated within xxhash.h.
+ * As a consequence, xxhash.c is no longer needed in /include.
+ *
+ * xxhash.c is still available and is still useful.
+ * In a "normal" setup, when xxhash is not inlined,
+ * xxhash.h only exposes the prototypes and public symbols,
+ * while xxhash.c can be built into an object file xxhash.o
+ * which can then be linked into the final binary.
+ ************************************************************************/
+
+#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \
+   || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387)
+#  define XXH_IMPLEM_13a8737387
+
+/* *************************************
+*  Tuning parameters
+***************************************/
+
+/*!
+ * @defgroup tuning Tuning parameters
+ * @{
+ *
+ * Various macros to control xxHash's behavior.
+ */
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Define this to disable 64-bit code.
+ *
+ * Useful if only using the @ref xxh32_family and you have a strict C90 compiler.
+ */
+#  define XXH_NO_LONG_LONG
+#  undef XXH_NO_LONG_LONG /* don't actually */
+/*!
+ * @brief Controls how unaligned memory is accessed.
+ *
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is
+ * safe and portable.
+ *
+ * Unfortunately, on some target/compiler combinations, the generated assembly
+ * is sub-optimal.
+ *
+ * The below switch allow selection of a different access method
+ * in the search for improved performance.
+ *
+ * @par Possible options:
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
+ *   @par
+ *     Use `memcpy()`. Safe and portable. Note that most modern compilers will
+ *     eliminate the function call and treat it as an unaligned access.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((packed))`
+ *   @par
+ *     Depends on compiler extensions and is therefore not portable.
+ *     This method is safe _if_ your compiler supports it,
+ *     and *generally* as fast or faster than `memcpy`.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
+ *  @par
+ *     Casts directly and dereferences. This method doesn't depend on the
+ *     compiler, but it violates the C standard as it directly dereferences an
+ *     unaligned pointer. It can generate buggy code on targets which do not
+ *     support unaligned memory accesses, but in some circumstances, it's the
+ *     only known way to get the most performance.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
+ *  @par
+ *     Also portable. This can generate the best code on old compilers which don't
+ *     inline small `memcpy()` calls, and it might also be faster on big-endian
+ *     systems which lack a native byteswap instruction. However, some compilers
+ *     will emit literal byteshifts even if the target supports unaligned access.
+ *  .
+ *
+ * @warning
+ *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
+ *   care, as what works on one compiler/platform/optimization level may cause
+ *   another to read garbage data or even crash.
+ *
+ * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
+ *
+ * Prefer these methods in priority order (0 > 3 > 1 > 2)
+ */
+#  define XXH_FORCE_MEMORY_ACCESS 0
+
+/*!
+ * @def XXH_FORCE_ALIGN_CHECK
+ * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
+ * and XXH64() only).
+ *
+ * This is an important performance trick for architectures without decent
+ * unaligned memory access performance.
+ *
+ * It checks for input alignment, and when conditions are met, uses a "fast
+ * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
+ * faster_ read speed.
+ *
+ * The check costs one initial branch per hash, which is generally negligible,
+ * but not zero.
+ *
+ * Moreover, it's not useful to generate an additional code path if memory
+ * access uses the same instruction for both aligned and unaligned
+ * addresses (e.g. x86 and aarch64).
+ *
+ * In these cases, the alignment check can be removed by setting this macro to 0.
+ * Then the code will always use unaligned memory access.
+ * Align check is automatically disabled on x86, x64 & arm64,
+ * which are platforms known to offer good unaligned memory accesses performance.
+ *
+ * This option does not affect XXH3 (only XXH32 and XXH64).
+ */
+#  define XXH_FORCE_ALIGN_CHECK 0
+
+/*!
+ * @def XXH_NO_INLINE_HINTS
+ * @brief When non-zero, sets all functions to `static`.
+ *
+ * By default, xxHash tries to force the compiler to inline almost all internal
+ * functions.
+ *
+ * This can usually improve performance due to reduced jumping and improved
+ * constant folding, but significantly increases the size of the binary which
+ * might not be favorable.
+ *
+ * Additionally, sometimes the forced inlining can be detrimental to performance,
+ * depending on the architecture.
+ *
+ * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
+ * compiler full control on whether to inline or not.
+ *
+ * When not optimizing (-O0), optimizing for size (-Os, -Oz), or using
+ * -fno-inline with GCC or Clang, this will automatically be defined.
+ */
+#  define XXH_NO_INLINE_HINTS 0
+
+/*!
+ * @def XXH32_ENDJMP
+ * @brief Whether to use a jump for `XXH32_finalize`.
+ *
+ * For performance, `XXH32_finalize` uses multiple branches in the finalizer.
+ * This is generally preferable for performance,
+ * but depending on exact architecture, a jmp may be preferable.
+ *
+ * This setting is only possibly making a difference for very small inputs.
+ */
+#  define XXH32_ENDJMP 0
+
+/*!
+ * @internal
+ * @brief Redefines old internal names.
+ *
+ * For compatibility with code that uses xxHash's internals before the names
+ * were changed to improve namespacing. There is no other reason to use this.
+ */
+#  define XXH_OLD_NAMES
+#  undef XXH_OLD_NAMES /* don't actually use, it is ugly. */
+#endif /* XXH_DOXYGEN */
+/*!
+ * @}
+ */
+
+#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+   /* prefer __packed__ structures (method 1) for gcc on armv7+ and mips */
+#  if !defined(__clang__) && \
+( \
+    (defined(__INTEL_COMPILER) && !defined(_WIN32)) || \
+    ( \
+        defined(__GNUC__) && ( \
+            (defined(__ARM_ARCH) && __ARM_ARCH >= 7) || \
+            ( \
+                defined(__mips__) && \
+                (__mips <= 5 || __mips_isa_rev < 6) && \
+                (!defined(__mips16) || defined(__mips_mips16e2)) \
+            ) \
+        ) \
+    ) \
+)
+#    define XXH_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+#ifndef XXH_FORCE_ALIGN_CHECK  /* can be defined externally */
+#  if defined(__i386)  || defined(__x86_64__) || defined(__aarch64__) \
+   || defined(_M_IX86) || defined(_M_X64)     || defined(_M_ARM64) /* visual */
+#    define XXH_FORCE_ALIGN_CHECK 0
+#  else
+#    define XXH_FORCE_ALIGN_CHECK 1
+#  endif
+#endif
+
+#ifndef XXH_NO_INLINE_HINTS
+#  if defined(__OPTIMIZE_SIZE__) /* -Os, -Oz */ \
+   || defined(__NO_INLINE__)     /* -O0, -fno-inline */
+#    define XXH_NO_INLINE_HINTS 1
+#  else
+#    define XXH_NO_INLINE_HINTS 0
+#  endif
+#endif
+
+#ifndef XXH32_ENDJMP
+/* generally preferable for performance */
+#  define XXH32_ENDJMP 0
+#endif
+
+/*!
+ * @defgroup impl Implementation
+ * @{
+ */
+
+
+/* *************************************
+*  Includes & Memory related functions
+***************************************/
+/* Modify the local functions below should you wish to use some other memory routines */
+/* for ZSTD_malloc(), ZSTD_free() */
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"  /* size_t, ZSTD_malloc, ZSTD_free, ZSTD_memcpy */
+static void* XXH_malloc(size_t s) { return ZSTD_malloc(s); }
+static void  XXH_free  (void* p)  { ZSTD_free(p); }
+static void* XXH_memcpy(void* dest, const void* src, size_t size) { return ZSTD_memcpy(dest,src,size); }
+
+
+/* *************************************
+*  Compiler Specific Options
+***************************************/
+#ifdef _MSC_VER /* Visual Studio warning fix */
+#  pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
+#endif
+
+#if XXH_NO_INLINE_HINTS  /* disable inlining hints */
+#  if defined(__GNUC__) || defined(__clang__)
+#    define XXH_FORCE_INLINE static __attribute__((unused))
+#  else
+#    define XXH_FORCE_INLINE static
+#  endif
+#  define XXH_NO_INLINE static
+/* enable inlining hints */
+#elif defined(__GNUC__) || defined(__clang__)
+#  define XXH_FORCE_INLINE static __inline__ __attribute__((always_inline, unused))
+#  define XXH_NO_INLINE static __attribute__((noinline))
+#elif defined(_MSC_VER)  /* Visual Studio */
+#  define XXH_FORCE_INLINE static __forceinline
+#  define XXH_NO_INLINE static __declspec(noinline)
+#elif defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
+#  define XXH_FORCE_INLINE static inline
+#  define XXH_NO_INLINE static
+#else
+#  define XXH_FORCE_INLINE static
+#  define XXH_NO_INLINE static
+#endif
+
+
+
+/* *************************************
+*  Debug
+***************************************/
+/*!
+ * @ingroup tuning
+ * @def XXH_DEBUGLEVEL
+ * @brief Sets the debugging level.
+ *
+ * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
+ * compiler's command line options. The value must be a number.
+ */
+#ifndef XXH_DEBUGLEVEL
+#  ifdef DEBUGLEVEL /* backwards compat */
+#    define XXH_DEBUGLEVEL DEBUGLEVEL
+#  else
+#    define XXH_DEBUGLEVEL 0
+#  endif
+#endif
+
+#if (XXH_DEBUGLEVEL>=1)
+#  include <assert.h>   /* note: can still be disabled with NDEBUG */
+#  define XXH_ASSERT(c)   assert(c)
+#else
+#  define XXH_ASSERT(c)   ((void)0)
+#endif
+
+/* note: use after variable declarations */
+#ifndef XXH_STATIC_ASSERT
+#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)    /* C11 */
+#    include <assert.h>
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
+#  elif defined(__cplusplus) && (__cplusplus >= 201103L)            /* C++11 */
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
+#  else
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0)
+#  endif
+#  define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c)
+#endif
+
+/*!
+ * @internal
+ * @def XXH_COMPILER_GUARD(var)
+ * @brief Used to prevent unwanted optimizations for @p var.
+ *
+ * It uses an empty GCC inline assembly statement with a register constraint
+ * which forces @p var into a general purpose register (e.g. eax, ebx, ecx
+ * on x86) and marks it as modified.
+ *
+ * This is used in a few places to avoid unwanted autovectorization (e.g.
+ * XXH32_round()). All vectorization we want is explicit via intrinsics,
+ * and _usually_ isn't wanted elsewhere.
+ *
+ * We also use it to prevent unwanted constant folding for AArch64 in
+ * XXH3_initCustomSecret_scalar().
+ */
+#if defined(__GNUC__) || defined(__clang__)
+#  define XXH_COMPILER_GUARD(var) __asm__ __volatile__("" : "+r" (var))
+#else
+#  define XXH_COMPILER_GUARD(var) ((void)0)
+#endif
+
+/* *************************************
+*  Basic Types
+***************************************/
+#if !defined (__VMS) \
+ && (defined (__cplusplus) \
+ || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# include <stdint.h>
+  typedef uint8_t xxh_u8;
+#else
+  typedef unsigned char xxh_u8;
+#endif
+typedef XXH32_hash_t xxh_u32;
+
+#ifdef XXH_OLD_NAMES
+#  define BYTE xxh_u8
+#  define U8   xxh_u8
+#  define U32  xxh_u32
+#endif
+
+/* ***   Memory access   *** */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_read32(const void* ptr)
+ * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit native endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readLE32(const void* ptr)
+ * @brief Reads an unaligned 32-bit little endian integer from @p ptr.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit little endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readBE32(const void* ptr)
+ * @brief Reads an unaligned 32-bit big endian integer from @p ptr.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit big endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readLE32_align(const void* ptr, XXH_alignment align)
+ * @brief Like @ref XXH_readLE32(), but has an option for aligned reads.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ * Note that when @ref XXH_FORCE_ALIGN_CHECK == 0, the @p align parameter is
+ * always @ref XXH_alignment::XXH_unaligned.
+ *
+ * @param ptr The pointer to read from.
+ * @param align Whether @p ptr is aligned.
+ * @pre
+ *   If @p align == @ref XXH_alignment::XXH_aligned, @p ptr must be 4 byte
+ *   aligned.
+ * @return The 32-bit little endian integer from the bytes at @p ptr.
+ */
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE32 and XXH_readBE32.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/*
+ * Force direct memory access. Only works on CPU which support unaligned memory
+ * access in hardware.
+ */
+static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; }
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/*
+ * __pack instructions are safer but compiler specific, hence potentially
+ * problematic for some compilers.
+ *
+ * Currently only defined for GCC and ICC.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
+#endif
+static xxh_u32 XXH_read32(const void* ptr)
+{
+    typedef union { xxh_u32 u32; } __attribute__((packed)) xxh_unalign;
+    return ((const xxh_unalign*)ptr)->u32;
+}
+
+#else
+
+/*
+ * Portable and safe solution. Generally efficient.
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u32 XXH_read32(const void* memPtr)
+{
+    xxh_u32 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+
+/* ***   Endianness   *** */
+
+/*!
+ * @ingroup tuning
+ * @def XXH_CPU_LITTLE_ENDIAN
+ * @brief Whether the target is little endian.
+ *
+ * Defined to 1 if the target is little endian, or 0 if it is big endian.
+ * It can be defined externally, for example on the compiler command line.
+ *
+ * If it is not defined,
+ * a runtime check (which is usually constant folded) is used instead.
+ *
+ * @note
+ *   This is not necessarily defined to an integer constant.
+ *
+ * @see XXH_isLittleEndian() for the runtime check.
+ */
+#ifndef XXH_CPU_LITTLE_ENDIAN
+/*
+ * Try to detect endianness automatically, to avoid the nonstandard behavior
+ * in `XXH_isLittleEndian()`
+ */
+#  if defined(_WIN32) /* Windows is always little endian */ \
+     || defined(__LITTLE_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 1
+#  elif defined(__BIG_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 0
+#  else
+/*!
+ * @internal
+ * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
+ *
+ * Most compilers will constant fold this.
+ */
+static int XXH_isLittleEndian(void)
+{
+    /*
+     * Portable and well-defined behavior.
+     * Don't use static: it is detrimental to performance.
+     */
+    const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 };
+    return one.c[0];
+}
+#   define XXH_CPU_LITTLE_ENDIAN   XXH_isLittleEndian()
+#  endif
+#endif
+
+
+
+
+/* ****************************************
+*  Compiler-specific Functions and Macros
+******************************************/
+#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+
+#ifdef __has_builtin
+#  define XXH_HAS_BUILTIN(x) __has_builtin(x)
+#else
+#  define XXH_HAS_BUILTIN(x) 0
+#endif
+
+/*!
+ * @internal
+ * @def XXH_rotl32(x,r)
+ * @brief 32-bit rotate left.
+ *
+ * @param x The 32-bit integer to be rotated.
+ * @param r The number of bits to rotate.
+ * @pre
+ *   @p r > 0 && @p r < 32
+ * @note
+ *   @p x and @p r may be evaluated multiple times.
+ * @return The rotated result.
+ */
+#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \
+                               && XXH_HAS_BUILTIN(__builtin_rotateleft64)
+#  define XXH_rotl32 __builtin_rotateleft32
+#  define XXH_rotl64 __builtin_rotateleft64
+/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */
+#elif defined(_MSC_VER)
+#  define XXH_rotl32(x,r) _rotl(x,r)
+#  define XXH_rotl64(x,r) _rotl64(x,r)
+#else
+#  define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
+#  define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r))))
+#endif
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_swap32(xxh_u32 x)
+ * @brief A 32-bit byteswap.
+ *
+ * @param x The 32-bit integer to byteswap.
+ * @return @p x, byteswapped.
+ */
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap32 _byteswap_ulong
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap32 __builtin_bswap32
+#else
+static xxh_u32 XXH_swap32 (xxh_u32 x)
+{
+    return  ((x << 24) & 0xff000000 ) |
+            ((x <<  8) & 0x00ff0000 ) |
+            ((x >>  8) & 0x0000ff00 ) |
+            ((x >> 24) & 0x000000ff );
+}
+#endif
+
+
+/* ***************************
+*  Memory reads
+*****************************/
+
+/*!
+ * @internal
+ * @brief Enum to indicate whether a pointer is aligned.
+ */
+typedef enum {
+    XXH_aligned,  /*!< Aligned */
+    XXH_unaligned /*!< Possibly unaligned */
+} XXH_alignment;
+
+/*
+ * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
+ *
+ * This is ideal for older compilers which don't inline memcpy.
+ */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u32)bytePtr[1] << 8)
+         | ((xxh_u32)bytePtr[2] << 16)
+         | ((xxh_u32)bytePtr[3] << 24);
+}
+
+XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[3]
+         | ((xxh_u32)bytePtr[2] << 8)
+         | ((xxh_u32)bytePtr[1] << 16)
+         | ((xxh_u32)bytePtr[0] << 24);
+}
+
+#else
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
+}
+
+static xxh_u32 XXH_readBE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
+}
+#endif
+
+XXH_FORCE_INLINE xxh_u32
+XXH_readLE32_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned) {
+        return XXH_readLE32(ptr);
+    } else {
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr);
+    }
+}
+
+
+/* *************************************
+*  Misc
+***************************************/
+/*! @ingroup public */
+XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
+
+
+/* *******************************************************************
+*  32-bit hash functions
+*********************************************************************/
+/*!
+ * @}
+ * @defgroup xxh32_impl XXH32 implementation
+ * @ingroup impl
+ * @{
+ */
+ /* #define instead of static const, to be used as initializers */
+#define XXH_PRIME32_1  0x9E3779B1U  /*!< 0b10011110001101110111100110110001 */
+#define XXH_PRIME32_2  0x85EBCA77U  /*!< 0b10000101111010111100101001110111 */
+#define XXH_PRIME32_3  0xC2B2AE3DU  /*!< 0b11000010101100101010111000111101 */
+#define XXH_PRIME32_4  0x27D4EB2FU  /*!< 0b00100111110101001110101100101111 */
+#define XXH_PRIME32_5  0x165667B1U  /*!< 0b00010110010101100110011110110001 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME32_1 XXH_PRIME32_1
+#  define PRIME32_2 XXH_PRIME32_2
+#  define PRIME32_3 XXH_PRIME32_3
+#  define PRIME32_4 XXH_PRIME32_4
+#  define PRIME32_5 XXH_PRIME32_5
+#endif
+
+/*!
+ * @internal
+ * @brief Normal stripe processing routine.
+ *
+ * This shuffles the bits so that any bit from @p input impacts several bits in
+ * @p acc.
+ *
+ * @param acc The accumulator lane.
+ * @param input The stripe of input to mix.
+ * @return The mixed accumulator lane.
+ */
+static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
+{
+    acc += input * XXH_PRIME32_2;
+    acc  = XXH_rotl32(acc, 13);
+    acc *= XXH_PRIME32_1;
+#if (defined(__SSE4_1__) || defined(__aarch64__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
+    /*
+     * UGLY HACK:
+     * A compiler fence is the only thing that prevents GCC and Clang from
+     * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
+     * reason) without globally disabling SSE4.1.
+     *
+     * The reason we want to avoid vectorization is because despite working on
+     * 4 integers at a time, there are multiple factors slowing XXH32 down on
+     * SSE4:
+     * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
+     *   newer chips!) making it slightly slower to multiply four integers at
+     *   once compared to four integers independently. Even when pmulld was
+     *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
+     *   just to multiply unless doing a long operation.
+     *
+     * - Four instructions are required to rotate,
+     *      movqda tmp,  v // not required with VEX encoding
+     *      pslld  tmp, 13 // tmp <<= 13
+     *      psrld  v,   19 // x >>= 19
+     *      por    v,  tmp // x |= tmp
+     *   compared to one for scalar:
+     *      roll   v, 13    // reliably fast across the board
+     *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
+     *
+     * - Instruction level parallelism is actually more beneficial here because
+     *   the SIMD actually serializes this operation: While v1 is rotating, v2
+     *   can load data, while v3 can multiply. SSE forces them to operate
+     *   together.
+     *
+     * This is also enabled on AArch64, as Clang autovectorizes it incorrectly
+     * and it is pointless writing a NEON implementation that is basically the
+     * same speed as scalar for XXH32.
+     */
+    XXH_COMPILER_GUARD(acc);
+#endif
+    return acc;
+}
+
+/*!
+ * @internal
+ * @brief Mixes all bits to finalize the hash.
+ *
+ * The final mix ensures that all input bits have a chance to impact any bit in
+ * the output digest, resulting in an unbiased distribution.
+ *
+ * @param h32 The hash to avalanche.
+ * @return The avalanched hash.
+ */
+static xxh_u32 XXH32_avalanche(xxh_u32 h32)
+{
+    h32 ^= h32 >> 15;
+    h32 *= XXH_PRIME32_2;
+    h32 ^= h32 >> 13;
+    h32 *= XXH_PRIME32_3;
+    h32 ^= h32 >> 16;
+    return(h32);
+}
+
+#define XXH_get32bits(p) XXH_readLE32_align(p, align)
+
+/*!
+ * @internal
+ * @brief Processes the last 0-15 bytes of @p ptr.
+ *
+ * There may be up to 15 bytes remaining to consume from the input.
+ * This final stage will digest them to ensure that all input bytes are present
+ * in the final mix.
+ *
+ * @param h32 The hash to finalize.
+ * @param ptr The pointer to the remaining input.
+ * @param len The remaining length, modulo 16.
+ * @param align Whether @p ptr is aligned.
+ * @return The finalized hash.
+ */
+static xxh_u32
+XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+#define XXH_PROCESS1 do {                           \
+    h32 += (*ptr++) * XXH_PRIME32_5;                \
+    h32 = XXH_rotl32(h32, 11) * XXH_PRIME32_1;      \
+} while (0)
+
+#define XXH_PROCESS4 do {                           \
+    h32 += XXH_get32bits(ptr) * XXH_PRIME32_3;      \
+    ptr += 4;                                   \
+    h32  = XXH_rotl32(h32, 17) * XXH_PRIME32_4;     \
+} while (0)
+
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+
+    /* Compact rerolled version; generally faster */
+    if (!XXH32_ENDJMP) {
+        len &= 15;
+        while (len >= 4) {
+            XXH_PROCESS4;
+            len -= 4;
+        }
+        while (len > 0) {
+            XXH_PROCESS1;
+            --len;
+        }
+        return XXH32_avalanche(h32);
+    } else {
+         switch(len&15) /* or switch(bEnd - p) */ {
+           case 12:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 8:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 4:       XXH_PROCESS4;
+                         return XXH32_avalanche(h32);
+
+           case 13:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 9:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 5:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(h32);
+
+           case 14:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 10:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 6:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(h32);
+
+           case 15:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 11:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 7:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 3:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 2:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 1:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 0:       return XXH32_avalanche(h32);
+        }
+        XXH_ASSERT(0);
+        return h32;   /* reaching this point is deemed impossible */
+    }
+}
+
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1 XXH_PROCESS1
+#  define PROCESS4 XXH_PROCESS4
+#else
+#  undef XXH_PROCESS1
+#  undef XXH_PROCESS4
+#endif
+
+/*!
+ * @internal
+ * @brief The implementation for @ref XXH32().
+ *
+ * @param input , len , seed Directly passed from @ref XXH32().
+ * @param align Whether @p input is aligned.
+ * @return The calculated hash.
+ */
+XXH_FORCE_INLINE xxh_u32
+XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
+{
+    xxh_u32 h32;
+
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=16) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 15;
+        xxh_u32 v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+        xxh_u32 v2 = seed + XXH_PRIME32_2;
+        xxh_u32 v3 = seed + 0;
+        xxh_u32 v4 = seed - XXH_PRIME32_1;
+
+        do {
+            v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4;
+            v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4;
+            v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4;
+            v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4;
+        } while (input < limit);
+
+        h32 = XXH_rotl32(v1, 1)  + XXH_rotl32(v2, 7)
+            + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
+    } else {
+        h32  = seed + XXH_PRIME32_5;
+    }
+
+    h32 += (xxh_u32)len;
+
+    return XXH32_finalize(h32, input, len&15, align);
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
+{
+#if 0
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH32_state_t state;
+    XXH32_reset(&state, seed);
+    XXH32_update(&state, (const xxh_u8*)input, len);
+    return XXH32_digest(&state);
+#else
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
+            return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
+    }   }
+
+    return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
+#endif
+}
+
+
+
+/*******   Hash streaming   *******/
+/*!
+ * @ingroup xxh32_family
+ */
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
+{
+    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
+}
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
+{
+    XXH_memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
+{
+    XXH_ASSERT(statePtr != NULL);
+    memset(statePtr, 0, sizeof(*statePtr));
+    statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+    statePtr->v[1] = seed + XXH_PRIME32_2;
+    statePtr->v[2] = seed + 0;
+    statePtr->v[3] = seed - XXH_PRIME32_1;
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH32_update(XXH32_state_t* state, const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
+
+        state->total_len_32 += (XXH32_hash_t)len;
+        state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16));
+
+        if (state->memsize + len < 16)  {   /* fill in tmp buffer */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len);
+            state->memsize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
+
+        if (state->memsize) {   /* some data left from previous update */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize);
+            {   const xxh_u32* p32 = state->mem32;
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32)); p32++;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32)); p32++;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32)); p32++;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
+            }
+            p += 16-state->memsize;
+            state->memsize = 0;
+        }
+
+        if (p <= bEnd-16) {
+            const xxh_u8* const limit = bEnd - 16;
+
+            do {
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p)); p+=4;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p)); p+=4;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p)); p+=4;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p)); p+=4;
+            } while (p<=limit);
+
+        }
+
+        if (p < bEnd) {
+            XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
+    }
+
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
+{
+    xxh_u32 h32;
+
+    if (state->large_len) {
+        h32 = XXH_rotl32(state->v[0], 1)
+            + XXH_rotl32(state->v[1], 7)
+            + XXH_rotl32(state->v[2], 12)
+            + XXH_rotl32(state->v[3], 18);
+    } else {
+        h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
+    }
+
+    h32 += state->total_len_32;
+
+    return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
+}
+
+
+/*******   Canonical representation   *******/
+
+/*!
+ * @ingroup xxh32_family
+ * The default return values from XXH functions are unsigned 32 and 64 bit
+ * integers.
+ *
+ * The canonical representation uses big endian convention, the same convention
+ * as human-readable numbers (large digits first).
+ *
+ * This way, hash values can be written into a file or buffer, remaining
+ * comparable across different systems.
+ *
+ * The following functions allow transformation of hash values to and from their
+ * canonical format.
+ */
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
+{
+    /* XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t)); */
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
+}
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
+{
+    return XXH_readBE32(src);
+}
+
+
+#ifndef XXH_NO_LONG_LONG
+
+/* *******************************************************************
+*  64-bit hash functions
+*********************************************************************/
+/*!
+ * @}
+ * @ingroup impl
+ * @{
+ */
+/*******   Memory access   *******/
+
+typedef XXH64_hash_t xxh_u64;
+
+#ifdef XXH_OLD_NAMES
+#  define U64 xxh_u64
+#endif
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE64 and XXH_readBE64.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    return *(const xxh_u64*) memPtr;
+}
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/*
+ * __pack instructions are safer, but compiler specific, hence potentially
+ * problematic for some compilers.
+ *
+ * Currently only defined for GCC and ICC.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
+#endif
+static xxh_u64 XXH_read64(const void* ptr)
+{
+    typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) xxh_unalign64;
+    return ((const xxh_unalign64*)ptr)->u64;
+}
+
+#else
+
+/*
+ * Portable and safe solution. Generally efficient.
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    xxh_u64 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap64 _byteswap_uint64
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap64 __builtin_bswap64
+#else
+static xxh_u64 XXH_swap64(xxh_u64 x)
+{
+    return  ((x << 56) & 0xff00000000000000ULL) |
+            ((x << 40) & 0x00ff000000000000ULL) |
+            ((x << 24) & 0x0000ff0000000000ULL) |
+            ((x << 8)  & 0x000000ff00000000ULL) |
+            ((x >> 8)  & 0x00000000ff000000ULL) |
+            ((x >> 24) & 0x0000000000ff0000ULL) |
+            ((x >> 40) & 0x000000000000ff00ULL) |
+            ((x >> 56) & 0x00000000000000ffULL);
+}
+#endif
+
+
+/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u64)bytePtr[1] << 8)
+         | ((xxh_u64)bytePtr[2] << 16)
+         | ((xxh_u64)bytePtr[3] << 24)
+         | ((xxh_u64)bytePtr[4] << 32)
+         | ((xxh_u64)bytePtr[5] << 40)
+         | ((xxh_u64)bytePtr[6] << 48)
+         | ((xxh_u64)bytePtr[7] << 56);
+}
+
+XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[7]
+         | ((xxh_u64)bytePtr[6] << 8)
+         | ((xxh_u64)bytePtr[5] << 16)
+         | ((xxh_u64)bytePtr[4] << 24)
+         | ((xxh_u64)bytePtr[3] << 32)
+         | ((xxh_u64)bytePtr[2] << 40)
+         | ((xxh_u64)bytePtr[1] << 48)
+         | ((xxh_u64)bytePtr[0] << 56);
+}
+
+#else
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
+}
+
+static xxh_u64 XXH_readBE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
+}
+#endif
+
+XXH_FORCE_INLINE xxh_u64
+XXH_readLE64_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned)
+        return XXH_readLE64(ptr);
+    else
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr);
+}
+
+
+/*******   xxh64   *******/
+/*!
+ * @}
+ * @defgroup xxh64_impl XXH64 implementation
+ * @ingroup impl
+ * @{
+ */
+/* #define rather that static const, to be used as initializers */
+#define XXH_PRIME64_1  0x9E3779B185EBCA87ULL  /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */
+#define XXH_PRIME64_2  0xC2B2AE3D27D4EB4FULL  /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */
+#define XXH_PRIME64_3  0x165667B19E3779F9ULL  /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */
+#define XXH_PRIME64_4  0x85EBCA77C2B2AE63ULL  /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */
+#define XXH_PRIME64_5  0x27D4EB2F165667C5ULL  /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME64_1 XXH_PRIME64_1
+#  define PRIME64_2 XXH_PRIME64_2
+#  define PRIME64_3 XXH_PRIME64_3
+#  define PRIME64_4 XXH_PRIME64_4
+#  define PRIME64_5 XXH_PRIME64_5
+#endif
+
+static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
+{
+    acc += input * XXH_PRIME64_2;
+    acc  = XXH_rotl64(acc, 31);
+    acc *= XXH_PRIME64_1;
+    return acc;
+}
+
+static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
+{
+    val  = XXH64_round(0, val);
+    acc ^= val;
+    acc  = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
+    return acc;
+}
+
+static xxh_u64 XXH64_avalanche(xxh_u64 h64)
+{
+    h64 ^= h64 >> 33;
+    h64 *= XXH_PRIME64_2;
+    h64 ^= h64 >> 29;
+    h64 *= XXH_PRIME64_3;
+    h64 ^= h64 >> 32;
+    return h64;
+}
+
+
+#define XXH_get64bits(p) XXH_readLE64_align(p, align)
+
+static xxh_u64
+XXH64_finalize(xxh_u64 h64, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+    len &= 31;
+    while (len >= 8) {
+        xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
+        ptr += 8;
+        h64 ^= k1;
+        h64  = XXH_rotl64(h64,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
+        len -= 8;
+    }
+    if (len >= 4) {
+        h64 ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
+        ptr += 4;
+        h64 = XXH_rotl64(h64, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
+        len -= 4;
+    }
+    while (len > 0) {
+        h64 ^= (*ptr++) * XXH_PRIME64_5;
+        h64 = XXH_rotl64(h64, 11) * XXH_PRIME64_1;
+        --len;
+    }
+    return  XXH64_avalanche(h64);
+}
+
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1_64 XXH_PROCESS1_64
+#  define PROCESS4_64 XXH_PROCESS4_64
+#  define PROCESS8_64 XXH_PROCESS8_64
+#else
+#  undef XXH_PROCESS1_64
+#  undef XXH_PROCESS4_64
+#  undef XXH_PROCESS8_64
+#endif
+
+XXH_FORCE_INLINE xxh_u64
+XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
+{
+    xxh_u64 h64;
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=32) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 31;
+        xxh_u64 v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+        xxh_u64 v2 = seed + XXH_PRIME64_2;
+        xxh_u64 v3 = seed + 0;
+        xxh_u64 v4 = seed - XXH_PRIME64_1;
+
+        do {
+            v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8;
+            v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8;
+            v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8;
+            v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8;
+        } while (input<limit);
+
+        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
+        h64 = XXH64_mergeRound(h64, v1);
+        h64 = XXH64_mergeRound(h64, v2);
+        h64 = XXH64_mergeRound(h64, v3);
+        h64 = XXH64_mergeRound(h64, v4);
+
+    } else {
+        h64  = seed + XXH_PRIME64_5;
+    }
+
+    h64 += (xxh_u64) len;
+
+    return XXH64_finalize(h64, input, len, align);
+}
+
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t len, XXH64_hash_t seed)
+{
+#if 0
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH64_state_t state;
+    XXH64_reset(&state, seed);
+    XXH64_update(&state, (const xxh_u8*)input, len);
+    return XXH64_digest(&state);
+#else
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
+            return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
+    }   }
+
+    return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
+
+#endif
+}
+
+/*******   Hash Streaming   *******/
+
+/*! @ingroup xxh64_family*/
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
+{
+    return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
+}
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState)
+{
+    XXH_memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, XXH64_hash_t seed)
+{
+    XXH_ASSERT(statePtr != NULL);
+    memset(statePtr, 0, sizeof(*statePtr));
+    statePtr->v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+    statePtr->v[1] = seed + XXH_PRIME64_2;
+    statePtr->v[2] = seed + 0;
+    statePtr->v[3] = seed - XXH_PRIME64_1;
+    return XXH_OK;
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH64_update (XXH64_state_t* state, const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
+
+        state->total_len += len;
+
+        if (state->memsize + len < 32) {  /* fill in tmp buffer */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len);
+            state->memsize += (xxh_u32)len;
+            return XXH_OK;
+        }
+
+        if (state->memsize) {   /* tmp buffer is full */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize);
+            state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64+0));
+            state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64+1));
+            state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64+2));
+            state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64+3));
+            p += 32 - state->memsize;
+            state->memsize = 0;
+        }
+
+        if (p+32 <= bEnd) {
+            const xxh_u8* const limit = bEnd - 32;
+
+            do {
+                state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p)); p+=8;
+                state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p)); p+=8;
+                state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p)); p+=8;
+                state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p)); p+=8;
+            } while (p<=limit);
+
+        }
+
+        if (p < bEnd) {
+            XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
+    }
+
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_digest(const XXH64_state_t* state)
+{
+    xxh_u64 h64;
+
+    if (state->total_len >= 32) {
+        h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) + XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
+        h64 = XXH64_mergeRound(h64, state->v[0]);
+        h64 = XXH64_mergeRound(h64, state->v[1]);
+        h64 = XXH64_mergeRound(h64, state->v[2]);
+        h64 = XXH64_mergeRound(h64, state->v[3]);
+    } else {
+        h64  = state->v[2] /*seed*/ + XXH_PRIME64_5;
+    }
+
+    h64 += (xxh_u64) state->total_len;
+
+    return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
+}
+
+
+/******* Canonical representation   *******/
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
+{
+    /* XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t)); */
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
+{
+    return XXH_readBE64(src);
+}
+
+#ifndef XXH_NO_XXH3
+
+/* *********************************************************************
+*  XXH3
+*  New generation hash designed for speed on small keys and vectorization
+************************************************************************ */
+/*!
+ * @}
+ * @defgroup xxh3_impl XXH3 implementation
+ * @ingroup impl
+ * @{
+ */
+
+/* ===   Compiler specifics   === */
+
+#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
+#  define XXH_RESTRICT /* disable */
+#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* >= C99 */
+#  define XXH_RESTRICT   restrict
+#else
+/* Note: it might be useful to define __restrict or __restrict__ for some C++ compilers */
+#  define XXH_RESTRICT   /* disable */
+#endif
+
+#if (defined(__GNUC__) && (__GNUC__ >= 3))  \
+  || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \
+  || defined(__clang__)
+#    define XXH_likely(x) __builtin_expect(x, 1)
+#    define XXH_unlikely(x) __builtin_expect(x, 0)
+#else
+#    define XXH_likely(x) (x)
+#    define XXH_unlikely(x) (x)
+#endif
+
+#if defined(__GNUC__) || defined(__clang__)
+#  if defined(__ARM_NEON__) || defined(__ARM_NEON) \
+   || defined(__aarch64__)  || defined(_M_ARM) \
+   || defined(_M_ARM64)     || defined(_M_ARM64EC)
+#    define inline __inline__  /* circumvent a clang bug */
+#    include <arm_neon.h>
+#    undef inline
+#  elif defined(__AVX2__)
+#    include <immintrin.h>
+#  elif defined(__SSE2__)
+#    include <emmintrin.h>
+#  endif
+#endif
+
+#if defined(_MSC_VER)
+#  include <intrin.h>
+#endif
+
+/*
+ * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
+ * remaining a true 64-bit/128-bit hash function.
+ *
+ * This is done by prioritizing a subset of 64-bit operations that can be
+ * emulated without too many steps on the average 32-bit machine.
+ *
+ * For example, these two lines seem similar, and run equally fast on 64-bit:
+ *
+ *   xxh_u64 x;
+ *   x ^= (x >> 47); // good
+ *   x ^= (x >> 13); // bad
+ *
+ * However, to a 32-bit machine, there is a major difference.
+ *
+ * x ^= (x >> 47) looks like this:
+ *
+ *   x.lo ^= (x.hi >> (47 - 32));
+ *
+ * while x ^= (x >> 13) looks like this:
+ *
+ *   // note: funnel shifts are not usually cheap.
+ *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
+ *   x.hi ^= (x.hi >> 13);
+ *
+ * The first one is significantly faster than the second, simply because the
+ * shift is larger than 32. This means:
+ *  - All the bits we need are in the upper 32 bits, so we can ignore the lower
+ *    32 bits in the shift.
+ *  - The shift result will always fit in the lower 32 bits, and therefore,
+ *    we can ignore the upper 32 bits in the xor.
+ *
+ * Thanks to this optimization, XXH3 only requires these features to be efficient:
+ *
+ *  - Usable unaligned access
+ *  - A 32-bit or 64-bit ALU
+ *      - If 32-bit, a decent ADC instruction
+ *  - A 32 or 64-bit multiply with a 64-bit result
+ *  - For the 128-bit variant, a decent byteswap helps short inputs.
+ *
+ * The first two are already required by XXH32, and almost all 32-bit and 64-bit
+ * platforms which can run XXH32 can run XXH3 efficiently.
+ *
+ * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one
+ * notable exception.
+ *
+ * First of all, Thumb-1 lacks support for the UMULL instruction which
+ * performs the important long multiply. This means numerous __aeabi_lmul
+ * calls.
+ *
+ * Second of all, the 8 functional registers are just not enough.
+ * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need
+ * Lo registers, and this shuffling results in thousands more MOVs than A32.
+ *
+ * A32 and T32 don't have this limitation. They can access all 14 registers,
+ * do a 32->64 multiply with UMULL, and the flexible operand allowing free
+ * shifts is helpful, too.
+ *
+ * Therefore, we do a quick sanity check.
+ *
+ * If compiling Thumb-1 for a target which supports ARM instructions, we will
+ * emit a warning, as it is not a "sane" platform to compile for.
+ *
+ * Usually, if this happens, it is because of an accident and you probably need
+ * to specify -march, as you likely meant to compile for a newer architecture.
+ *
+ * Credit: large sections of the vectorial and asm source code paths
+ *         have been contributed by @easyaspi314
+ */
+#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM)
+#   warning "XXH3 is highly inefficient without ARM or Thumb-2."
+#endif
+
+/* ==========================================
+ * Vectorization detection
+ * ========================================== */
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @ingroup tuning
+ * @brief Overrides the vectorization implementation chosen for XXH3.
+ *
+ * Can be defined to 0 to disable SIMD or any of the values mentioned in
+ * @ref XXH_VECTOR_TYPE.
+ *
+ * If this is not defined, it uses predefined macros to determine the best
+ * implementation.
+ */
+#  define XXH_VECTOR XXH_SCALAR
+/*!
+ * @ingroup tuning
+ * @brief Possible values for @ref XXH_VECTOR.
+ *
+ * Note that these are actually implemented as macros.
+ *
+ * If this is not defined, it is detected automatically.
+ * @ref XXH_X86DISPATCH overrides this.
+ */
+enum XXH_VECTOR_TYPE /* fake enum */ {
+    XXH_SCALAR = 0,  /*!< Portable scalar version */
+    XXH_SSE2   = 1,  /*!<
+                      * SSE2 for Pentium 4, Opteron, all x86_64.
+                      *
+                      * @note SSE2 is also guaranteed on Windows 10, macOS, and
+                      * Android x86.
+                      */
+    XXH_AVX2   = 2,  /*!< AVX2 for Haswell and Bulldozer */
+    XXH_AVX512 = 3,  /*!< AVX512 for Skylake and Icelake */
+    XXH_NEON   = 4,  /*!< NEON for most ARMv7-A and all AArch64 */
+    XXH_VSX    = 5,  /*!< VSX and ZVector for POWER8/z13 (64-bit) */
+};
+/*!
+ * @ingroup tuning
+ * @brief Selects the minimum alignment for XXH3's accumulators.
+ *
+ * When using SIMD, this should match the alignment required for said vector
+ * type, so, for example, 32 for AVX2.
+ *
+ * Default: Auto detected.
+ */
+#  define XXH_ACC_ALIGN 8
+#endif
+
+/* Actual definition */
+#ifndef XXH_DOXYGEN
+#  define XXH_SCALAR 0
+#  define XXH_SSE2   1
+#  define XXH_AVX2   2
+#  define XXH_AVX512 3
+#  define XXH_NEON   4
+#  define XXH_VSX    5
+#endif
+
+#ifndef XXH_VECTOR    /* can be defined on command line */
+#  if ( \
+        defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
+     || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \
+   ) && ( \
+        defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
+    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
+   )
+#    define XXH_VECTOR XXH_NEON
+#  elif defined(__AVX512F__)
+#    define XXH_VECTOR XXH_AVX512
+#  elif defined(__AVX2__)
+#    define XXH_VECTOR XXH_AVX2
+#  elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
+#    define XXH_VECTOR XXH_SSE2
+#  elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
+     || (defined(__s390x__) && defined(__VEC__)) \
+     && defined(__GNUC__) /* TODO: IBM XL */
+#    define XXH_VECTOR XXH_VSX
+#  else
+#    define XXH_VECTOR XXH_SCALAR
+#  endif
+#endif
+
+/*
+ * Controls the alignment of the accumulator,
+ * for compatibility with aligned vector loads, which are usually faster.
+ */
+#ifndef XXH_ACC_ALIGN
+#  if defined(XXH_X86DISPATCH)
+#     define XXH_ACC_ALIGN 64  /* for compatibility with avx512 */
+#  elif XXH_VECTOR == XXH_SCALAR  /* scalar */
+#     define XXH_ACC_ALIGN 8
+#  elif XXH_VECTOR == XXH_SSE2  /* sse2 */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX2  /* avx2 */
+#     define XXH_ACC_ALIGN 32
+#  elif XXH_VECTOR == XXH_NEON  /* neon */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_VSX   /* vsx */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX512  /* avx512 */
+#     define XXH_ACC_ALIGN 64
+#  endif
+#endif
+
+#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
+    || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
+#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
+#else
+#  define XXH_SEC_ALIGN 8
+#endif
+
+/*
+ * UGLY HACK:
+ * GCC usually generates the best code with -O3 for xxHash.
+ *
+ * However, when targeting AVX2, it is overzealous in its unrolling resulting
+ * in code roughly 3/4 the speed of Clang.
+ *
+ * There are other issues, such as GCC splitting _mm256_loadu_si256 into
+ * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which
+ * only applies to Sandy and Ivy Bridge... which don't even support AVX2.
+ *
+ * That is why when compiling the AVX2 version, it is recommended to use either
+ *   -O2 -mavx2 -march=haswell
+ * or
+ *   -O2 -mavx2 -mno-avx256-split-unaligned-load
+ * for decent performance, or to use Clang instead.
+ *
+ * Fortunately, we can control the first one with a pragma that forces GCC into
+ * -O2, but the other one we can't control without "failed to inline always
+ * inline function due to target mismatch" warnings.
+ */
+#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
+  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
+#  pragma GCC push_options
+#  pragma GCC optimize("-O2")
+#endif
+
+
+#if XXH_VECTOR == XXH_NEON
+/*
+ * NEON's setup for vmlal_u32 is a little more complicated than it is on
+ * SSE2, AVX2, and VSX.
+ *
+ * While PMULUDQ and VMULEUW both perform a mask, VMLAL.U32 performs an upcast.
+ *
+ * To do the same operation, the 128-bit 'Q' register needs to be split into
+ * two 64-bit 'D' registers, performing this operation::
+ *
+ *   [                a                 |                 b                ]
+ *            |              '---------. .--------'                |
+ *            |                         x                          |
+ *            |              .---------' '--------.                |
+ *   [ a & 0xFFFFFFFF | b & 0xFFFFFFFF ],[    a >> 32     |     b >> 32    ]
+ *
+ * Due to significant changes in aarch64, the fastest method for aarch64 is
+ * completely different than the fastest method for ARMv7-A.
+ *
+ * ARMv7-A treats D registers as unions overlaying Q registers, so modifying
+ * D11 will modify the high half of Q5. This is similar to how modifying AH
+ * will only affect bits 8-15 of AX on x86.
+ *
+ * VZIP takes two registers, and puts even lanes in one register and odd lanes
+ * in the other.
+ *
+ * On ARMv7-A, this strangely modifies both parameters in place instead of
+ * taking the usual 3-operand form.
+ *
+ * Therefore, if we want to do this, we can simply use a D-form VZIP.32 on the
+ * lower and upper halves of the Q register to end up with the high and low
+ * halves where we want - all in one instruction.
+ *
+ *   vzip.32   d10, d11       @ d10 = { d10[0], d11[0] }; d11 = { d10[1], d11[1] }
+ *
+ * Unfortunately we need inline assembly for this: Instructions modifying two
+ * registers at once is not possible in GCC or Clang's IR, and they have to
+ * create a copy.
+ *
+ * aarch64 requires a different approach.
+ *
+ * In order to make it easier to write a decent compiler for aarch64, many
+ * quirks were removed, such as conditional execution.
+ *
+ * NEON was also affected by this.
+ *
+ * aarch64 cannot access the high bits of a Q-form register, and writes to a
+ * D-form register zero the high bits, similar to how writes to W-form scalar
+ * registers (or DWORD registers on x86_64) work.
+ *
+ * The formerly free vget_high intrinsics now require a vext (with a few
+ * exceptions)
+ *
+ * Additionally, VZIP was replaced by ZIP1 and ZIP2, which are the equivalent
+ * of PUNPCKL* and PUNPCKH* in SSE, respectively, in order to only modify one
+ * operand.
+ *
+ * The equivalent of the VZIP.32 on the lower and upper halves would be this
+ * mess:
+ *
+ *   ext     v2.4s, v0.4s, v0.4s, #2 // v2 = { v0[2], v0[3], v0[0], v0[1] }
+ *   zip1    v1.2s, v0.2s, v2.2s     // v1 = { v0[0], v2[0] }
+ *   zip2    v0.2s, v0.2s, v1.2s     // v0 = { v0[1], v2[1] }
+ *
+ * Instead, we use a literal downcast, vmovn_u64 (XTN), and vshrn_n_u64 (SHRN):
+ *
+ *   shrn    v1.2s, v0.2d, #32  // v1 = (uint32x2_t)(v0 >> 32);
+ *   xtn     v0.2s, v0.2d       // v0 = (uint32x2_t)(v0 & 0xFFFFFFFF);
+ *
+ * This is available on ARMv7-A, but is less efficient than a single VZIP.32.
+ */
+
+/*!
+ * Function-like macro:
+ * void XXH_SPLIT_IN_PLACE(uint64x2_t &in, uint32x2_t &outLo, uint32x2_t &outHi)
+ * {
+ *     outLo = (uint32x2_t)(in & 0xFFFFFFFF);
+ *     outHi = (uint32x2_t)(in >> 32);
+ *     in = UNDEFINED;
+ * }
+ */
+# if !defined(XXH_NO_VZIP_HACK) /* define to disable */ \
+   && (defined(__GNUC__) || defined(__clang__)) \
+   && (defined(__arm__) || defined(__thumb__) || defined(_M_ARM))
+#  define XXH_SPLIT_IN_PLACE(in, outLo, outHi)                                              \
+    do {                                                                                    \
+      /* Undocumented GCC/Clang operand modifier: %e0 = lower D half, %f0 = upper D half */ \
+      /* https://github.com/gcc-mirror/gcc/blob/38cf91e5/gcc/config/arm/arm.c#L22486 */     \
+      /* https://github.com/llvm-mirror/llvm/blob/2c4ca683/lib/Target/ARM/ARMAsmPrinter.cpp#L399 */ \
+      __asm__("vzip.32  %e0, %f0" : "+w" (in));                                             \
+      (outLo) = vget_low_u32 (vreinterpretq_u32_u64(in));                                   \
+      (outHi) = vget_high_u32(vreinterpretq_u32_u64(in));                                   \
+   } while (0)
+# else
+#  define XXH_SPLIT_IN_PLACE(in, outLo, outHi)                                            \
+    do {                                                                                  \
+      (outLo) = vmovn_u64    (in);                                                        \
+      (outHi) = vshrn_n_u64  ((in), 32);                                                  \
+    } while (0)
+# endif
+
+/*!
+ * @ingroup tuning
+ * @brief Controls the NEON to scalar ratio for XXH3
+ *
+ * On AArch64 when not optimizing for size, XXH3 will run 6 lanes using NEON and
+ * 2 lanes on scalar by default.
+ *
+ * This can be set to 2, 4, 6, or 8. ARMv7 will default to all 8 NEON lanes, as the
+ * emulated 64-bit arithmetic is too slow.
+ *
+ * Modern ARM CPUs are _very_ sensitive to how their pipelines are used.
+ *
+ * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but it can't
+ * have more than 2 NEON (F0/F1) micro-ops. If you are only using NEON instructions,
+ * you are only using 2/3 of the CPU bandwidth.
+ *
+ * This is even more noticeable on the more advanced cores like the A76 which
+ * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once.
+ *
+ * Therefore, @ref XXH3_NEON_LANES lanes will be processed using NEON, and the
+ * remaining lanes will use scalar instructions. This improves the bandwidth
+ * and also gives the integer pipelines something to do besides twiddling loop
+ * counters and pointers.
+ *
+ * This change benefits CPUs with large micro-op buffers without negatively affecting
+ * other CPUs:
+ *
+ *  | Chipset               | Dispatch type       | NEON only | 6:2 hybrid | Diff. |
+ *  |:----------------------|:--------------------|----------:|-----------:|------:|
+ *  | Snapdragon 730 (A76)  | 2 NEON/8 micro-ops  |  8.8 GB/s |  10.1 GB/s |  ~16% |
+ *  | Snapdragon 835 (A73)  | 2 NEON/3 micro-ops  |  5.1 GB/s |   5.3 GB/s |   ~5% |
+ *  | Marvell PXA1928 (A53) | In-order dual-issue |  1.9 GB/s |   1.9 GB/s |    0% |
+ *
+ * It also seems to fix some bad codegen on GCC, making it almost as fast as clang.
+ *
+ * @see XXH3_accumulate_512_neon()
+ */
+# ifndef XXH3_NEON_LANES
+#  if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \
+   && !defined(__OPTIMIZE_SIZE__)
+#   define XXH3_NEON_LANES 6
+#  else
+#   define XXH3_NEON_LANES XXH_ACC_NB
+#  endif
+# endif
+#endif  /* XXH_VECTOR == XXH_NEON */
+
+/*
+ * VSX and Z Vector helpers.
+ *
+ * This is very messy, and any pull requests to clean this up are welcome.
+ *
+ * There are a lot of problems with supporting VSX and s390x, due to
+ * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
+ */
+#if XXH_VECTOR == XXH_VSX
+#  if defined(__s390x__)
+#    include <s390intrin.h>
+#  else
+/* gcc's altivec.h can have the unwanted consequence to unconditionally
+ * #define bool, vector, and pixel keywords,
+ * with bad consequences for programs already using these keywords for other purposes.
+ * The paragraph defining these macros is skipped when __APPLE_ALTIVEC__ is defined.
+ * __APPLE_ALTIVEC__ is _generally_ defined automatically by the compiler,
+ * but it seems that, in some cases, it isn't.
+ * Force the build macro to be defined, so that keywords are not altered.
+ */
+#    if defined(__GNUC__) && !defined(__APPLE_ALTIVEC__)
+#      define __APPLE_ALTIVEC__
+#    endif
+#    include <altivec.h>
+#  endif
+
+typedef __vector unsigned long long xxh_u64x2;
+typedef __vector unsigned char xxh_u8x16;
+typedef __vector unsigned xxh_u32x4;
+
+# ifndef XXH_VSX_BE
+#  if defined(__BIG_ENDIAN__) \
+  || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_VSX_BE 1
+#  elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
+#    warning "-maltivec=be is not recommended. Please use native endianness."
+#    define XXH_VSX_BE 1
+#  else
+#    define XXH_VSX_BE 0
+#  endif
+# endif /* !defined(XXH_VSX_BE) */
+
+# if XXH_VSX_BE
+#  if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__))
+#    define XXH_vec_revb vec_revb
+#  else
+/*!
+ * A polyfill for POWER9's vec_revb().
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val)
+{
+    xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
+                                  0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
+    return vec_perm(val, val, vByteSwap);
+}
+#  endif
+# endif /* XXH_VSX_BE */
+
+/*!
+ * Performs an unaligned vector load and byte swaps it on big endian.
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
+{
+    xxh_u64x2 ret;
+    XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
+# if XXH_VSX_BE
+    ret = XXH_vec_revb(ret);
+# endif
+    return ret;
+}
+
+/*
+ * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
+ *
+ * These intrinsics weren't added until GCC 8, despite existing for a while,
+ * and they are endian dependent. Also, their meaning swap depending on version.
+ * */
+# if defined(__s390x__)
+ /* s390x is always big endian, no issue on this platform */
+#  define XXH_vec_mulo vec_mulo
+#  define XXH_vec_mule vec_mule
+# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw)
+/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
+#  define XXH_vec_mulo __builtin_altivec_vmulouw
+#  define XXH_vec_mule __builtin_altivec_vmuleuw
+# else
+/* gcc needs inline assembly */
+/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+# endif /* XXH_vec_mulo, XXH_vec_mule */
+#endif /* XXH_VECTOR == XXH_VSX */
+
+
+/* prefetch
+ * can be disabled, by declaring XXH_NO_PREFETCH build macro */
+#if defined(XXH_NO_PREFETCH)
+#  define XXH_PREFETCH(ptr)  (void)(ptr)  /* disabled */
+#else
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))  /* _mm_prefetch() not defined outside of x86/x64 */
+#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#    define XXH_PREFETCH(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
+#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
+#    define XXH_PREFETCH(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
+#  else
+#    define XXH_PREFETCH(ptr) (void)(ptr)  /* disabled */
+#  endif
+#endif  /* XXH_NO_PREFETCH */
+
+
+/* ==========================================
+ * XXH3 default settings
+ * ========================================== */
+
+#define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */
+
+#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
+#  error "default keyset is not large enough"
+#endif
+
+/*! Pseudorandom secret taken directly from FARSH. */
+XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
+    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
+    0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
+    0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
+    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c,
+    0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3,
+    0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
+    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d,
+    0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64,
+    0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
+    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e,
+    0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce,
+    0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
+};
+
+
+#ifdef XXH_OLD_NAMES
+#  define kSecret XXH3_kSecret
+#endif
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Calculates a 32-bit to 64-bit long multiply.
+ *
+ * Implemented as a macro.
+ *
+ * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't
+ * need to (but it shouldn't need to anyways, it is about 7 instructions to do
+ * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we
+ * use that instead of the normal method.
+ *
+ * If you are compiling for platforms like Thumb-1 and don't have a better option,
+ * you may also want to write your own long multiply routine here.
+ *
+ * @param x, y Numbers to be multiplied
+ * @return 64-bit product of the low 32 bits of @p x and @p y.
+ */
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64(xxh_u64 x, xxh_u64 y)
+{
+   return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
+}
+#elif defined(_MSC_VER) && defined(_M_IX86)
+#    define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
+#else
+/*
+ * Downcast + upcast is usually better than masking on older compilers like
+ * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers.
+ *
+ * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands
+ * and perform a full 64x64 multiply -- entirely redundant on 32-bit.
+ */
+#    define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
+#endif
+
+/*!
+ * @brief Calculates a 64->128-bit long multiply.
+ *
+ * Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar
+ * version.
+ *
+ * @param lhs , rhs The 64-bit integers to be multiplied
+ * @return The 128-bit result represented in an @ref XXH128_hash_t.
+ */
+static XXH128_hash_t
+XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
+{
+    /*
+     * GCC/Clang __uint128_t method.
+     *
+     * On most 64-bit targets, GCC and Clang define a __uint128_t type.
+     * This is usually the best way as it usually uses a native long 64-bit
+     * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
+     *
+     * Usually.
+     *
+     * Despite being a 32-bit platform, Clang (and emscripten) define this type
+     * despite not having the arithmetic for it. This results in a laggy
+     * compiler builtin call which calculates a full 128-bit multiply.
+     * In that case it is best to use the portable one.
+     * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
+     */
+#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \
+    && defined(__SIZEOF_INT128__) \
+    || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+
+    __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
+    XXH128_hash_t r128;
+    r128.low64  = (xxh_u64)(product);
+    r128.high64 = (xxh_u64)(product >> 64);
+    return r128;
+
+    /*
+     * MSVC for x64's _umul128 method.
+     *
+     * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct);
+     *
+     * This compiles to single operand MUL on x64.
+     */
+#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(_umul128)
+#endif
+    xxh_u64 product_high;
+    xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
+    XXH128_hash_t r128;
+    r128.low64  = product_low;
+    r128.high64 = product_high;
+    return r128;
+
+    /*
+     * MSVC for ARM64's __umulh method.
+     *
+     * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
+     */
+#elif defined(_M_ARM64) || defined(_M_ARM64EC)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(__umulh)
+#endif
+    XXH128_hash_t r128;
+    r128.low64  = lhs * rhs;
+    r128.high64 = __umulh(lhs, rhs);
+    return r128;
+
+#else
+    /*
+     * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
+     *
+     * This is a fast and simple grade school multiply, which is shown below
+     * with base 10 arithmetic instead of base 0x100000000.
+     *
+     *           9 3 // D2 lhs = 93
+     *         x 7 5 // D2 rhs = 75
+     *     ----------
+     *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
+     *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
+     *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
+     *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
+     *     ---------
+     *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
+     *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
+     *     ---------
+     *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
+     *
+     * The reasons for adding the products like this are:
+     *  1. It avoids manual carry tracking. Just like how
+     *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
+     *     This avoids a lot of complexity.
+     *
+     *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
+     *     instruction available in ARM's Digital Signal Processing extension
+     *     in 32-bit ARMv6 and later, which is shown below:
+     *
+     *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
+     *         {
+     *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
+     *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
+     *             *RdHi = (xxh_u32)(product >> 32);
+     *         }
+     *
+     *     This instruction was designed for efficient long multiplication, and
+     *     allows this to be calculated in only 4 instructions at speeds
+     *     comparable to some 64-bit ALUs.
+     *
+     *  3. It isn't terrible on other platforms. Usually this will be a couple
+     *     of 32-bit ADD/ADCs.
+     */
+
+    /* First calculate all of the cross products. */
+    xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
+    xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32,        rhs & 0xFFFFFFFF);
+    xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
+    xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32,        rhs >> 32);
+
+    /* Now add the products together. These will never overflow. */
+    xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
+    xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32)        + hi_hi;
+    xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
+
+    XXH128_hash_t r128;
+    r128.low64  = lower;
+    r128.high64 = upper;
+    return r128;
+#endif
+}
+
+/*!
+ * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it.
+ *
+ * The reason for the separate function is to prevent passing too many structs
+ * around by value. This will hopefully inline the multiply, but we don't force it.
+ *
+ * @param lhs , rhs The 64-bit integers to multiply
+ * @return The low 64 bits of the product XOR'd by the high 64 bits.
+ * @see XXH_mult64to128()
+ */
+static xxh_u64
+XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
+{
+    XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
+    return product.low64 ^ product.high64;
+}
+
+/*! Seems to produce slightly better code on GCC for some reason. */
+XXH_FORCE_INLINE xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
+{
+    XXH_ASSERT(0 <= shift && shift < 64);
+    return v64 ^ (v64 >> shift);
+}
+
+/*
+ * This is a fast avalanche stage,
+ * suitable when input bits are already partially mixed
+ */
+static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
+{
+    h64 = XXH_xorshift64(h64, 37);
+    h64 *= 0x165667919E3779F9ULL;
+    h64 = XXH_xorshift64(h64, 32);
+    return h64;
+}
+
+/*
+ * This is a stronger avalanche,
+ * inspired by Pelle Evensen's rrmxmx
+ * preferable when input has not been previously mixed
+ */
+static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
+{
+    /* this mix is inspired by Pelle Evensen's rrmxmx */
+    h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
+    h64 *= 0x9FB21C651E98DF25ULL;
+    h64 ^= (h64 >> 35) + len ;
+    h64 *= 0x9FB21C651E98DF25ULL;
+    return XXH_xorshift64(h64, 28);
+}
+
+
+/* ==========================================
+ * Short keys
+ * ==========================================
+ * One of the shortcomings of XXH32 and XXH64 was that their performance was
+ * sub-optimal on short lengths. It used an iterative algorithm which strongly
+ * favored lengths that were a multiple of 4 or 8.
+ *
+ * Instead of iterating over individual inputs, we use a set of single shot
+ * functions which piece together a range of lengths and operate in constant time.
+ *
+ * Additionally, the number of multiplies has been significantly reduced. This
+ * reduces latency, especially when emulating 64-bit multiplies on 32-bit.
+ *
+ * Depending on the platform, this may or may not be faster than XXH32, but it
+ * is almost guaranteed to be faster than XXH64.
+ */
+
+/*
+ * At very short lengths, there isn't enough input to fully hide secrets, or use
+ * the entire secret.
+ *
+ * There is also only a limited amount of mixing we can do before significantly
+ * impacting performance.
+ *
+ * Therefore, we use different sections of the secret and always mix two secret
+ * samples with an XOR. This should have no effect on performance on the
+ * seedless or withSeed variants because everything _should_ be constant folded
+ * by modern compilers.
+ *
+ * The XOR mixing hides individual parts of the secret and increases entropy.
+ *
+ * This adds an extra layer of strength for custom secrets.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combined = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combined = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combined = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8  const c1 = input[0];
+        xxh_u8  const c2 = input[len >> 1];
+        xxh_u8  const c3 = input[len - 1];
+        xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2  << 24)
+                               | ((xxh_u32)c3 <<  0) | ((xxh_u32)len << 8);
+        xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
+        return XXH64_avalanche(keyed);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input1 = XXH_readLE32(input);
+        xxh_u32 const input2 = XXH_readLE32(input + len - 4);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed;
+        xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
+        xxh_u64 const keyed = input64 ^ bitflip;
+        return XXH3_rrmxmx(keyed, len);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed;
+        xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed;
+        xxh_u64 const input_lo = XXH_readLE64(input)           ^ bitflip1;
+        xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
+        xxh_u64 const acc = len
+                          + XXH_swap64(input_lo) + input_hi
+                          + XXH3_mul128_fold64(input_lo, input_hi);
+        return XXH3_avalanche(acc);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (XXH_likely(len >  8)) return XXH3_len_9to16_64b(input, len, secret, seed);
+        if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
+        return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64)));
+    }
+}
+
+/*
+ * DISCLAIMER: There are known *seed-dependent* multicollisions here due to
+ * multiplication by zero, affecting hashes of lengths 17 to 240.
+ *
+ * However, they are very unlikely.
+ *
+ * Keep this in mind when using the unseeded XXH3_64bits() variant: As with all
+ * unseeded non-cryptographic hashes, it does not attempt to defend itself
+ * against specially crafted inputs, only random inputs.
+ *
+ * Compared to classic UMAC where a 1 in 2^31 chance of 4 consecutive bytes
+ * cancelling out the secret is taken an arbitrary number of times (addressed
+ * in XXH3_accumulate_512), this collision is very unlikely with random inputs
+ * and/or proper seeding:
+ *
+ * This only has a 1 in 2^63 chance of 8 consecutive bytes cancelling out, in a
+ * function that is only called up to 16 times per hash with up to 240 bytes of
+ * input.
+ *
+ * This is not too bad for a non-cryptographic hash function, especially with
+ * only 64 bit outputs.
+ *
+ * The 128-bit variant (which trades some speed for strength) is NOT affected
+ * by this, although it is always a good idea to use a proper seed if you care
+ * about strength.
+ */
+XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
+                                     const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64)
+{
+#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__i386__) && defined(__SSE2__)  /* x86 + SSE2 */ \
+  && !defined(XXH_ENABLE_AUTOVECTORIZE)      /* Define to disable like XXH32 hack */
+    /*
+     * UGLY HACK:
+     * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
+     * slower code.
+     *
+     * By forcing seed64 into a register, we disrupt the cost model and
+     * cause it to scalarize. See `XXH32_round()`
+     *
+     * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
+     * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
+     * GCC 9.2, despite both emitting scalar code.
+     *
+     * GCC generates much better scalar code than Clang for the rest of XXH3,
+     * which is why finding a more optimal codepath is an interest.
+     */
+    XXH_COMPILER_GUARD(seed64);
+#endif
+    {   xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64 const input_hi = XXH_readLE64(input+8);
+        return XXH3_mul128_fold64(
+            input_lo ^ (XXH_readLE64(secret)   + seed64),
+            input_hi ^ (XXH_readLE64(secret+8) - seed64)
+        );
+    }
+}
+
+/* For mid range keys, XXH3 uses a Mum-hash variant. */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                     const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                     XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc += XXH3_mix16B(input+48, secret+96, seed);
+                    acc += XXH3_mix16B(input+len-64, secret+112, seed);
+                }
+                acc += XXH3_mix16B(input+32, secret+64, seed);
+                acc += XXH3_mix16B(input+len-48, secret+80, seed);
+            }
+            acc += XXH3_mix16B(input+16, secret+32, seed);
+            acc += XXH3_mix16B(input+len-32, secret+48, seed);
+        }
+        acc += XXH3_mix16B(input+0, secret+0, seed);
+        acc += XXH3_mix16B(input+len-16, secret+16, seed);
+
+        return XXH3_avalanche(acc);
+    }
+}
+
+#define XXH3_MIDSIZE_MAX 240
+
+XXH_NO_INLINE XXH64_hash_t
+XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+    #define XXH3_MIDSIZE_STARTOFFSET 3
+    #define XXH3_MIDSIZE_LASTOFFSET  17
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+        int const nbRounds = (int)len / 16;
+        int i;
+        for (i=0; i<8; i++) {
+            acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
+        }
+        acc = XXH3_avalanche(acc);
+        XXH_ASSERT(nbRounds >= 8);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+        /*
+         * UGLY HACK:
+         * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86.
+         * In everywhere else, it uses scalar code.
+         *
+         * For 64->128-bit multiplies, even if the NEON was 100% optimal, it
+         * would still be slower than UMAAL (see XXH_mult64to128).
+         *
+         * Unfortunately, Clang doesn't handle the long multiplies properly and
+         * converts them to the nonexistent "vmulq_u64" intrinsic, which is then
+         * scalarized into an ugly mess of VMOV.32 instructions.
+         *
+         * This mess is difficult to avoid without turning autovectorization
+         * off completely, but they are usually relatively minor and/or not
+         * worth it to fix.
+         *
+         * This loop is the easiest to fix, as unlike XXH32, this pragma
+         * _actually works_ because it is a loop vectorization instead of an
+         * SLP vectorization.
+         */
+        #pragma clang loop vectorize(disable)
+#endif
+        for (i=8 ; i < nbRounds; i++) {
+            acc += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
+        }
+        /* last bytes */
+        acc += XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
+        return XXH3_avalanche(acc);
+    }
+}
+
+
+/* =======     Long Keys     ======= */
+
+#define XXH_STRIPE_LEN 64
+#define XXH_SECRET_CONSUME_RATE 8   /* nb of secret bytes consumed at each accumulation */
+#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
+
+#ifdef XXH_OLD_NAMES
+#  define STRIPE_LEN XXH_STRIPE_LEN
+#  define ACC_NB XXH_ACC_NB
+#endif
+
+XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
+{
+    if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
+    XXH_memcpy(dst, &v64, sizeof(v64));
+}
+
+/* Several intrinsic functions below are supposed to accept __int64 as argument,
+ * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
+ * However, several environments do not define __int64 type,
+ * requiring a workaround.
+ */
+#if !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+    typedef int64_t xxh_i64;
+#else
+    /* the following type must have a width of 64-bit */
+    typedef long long xxh_i64;
+#endif
+
+
+/*
+ * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
+ *
+ * It is a hardened version of UMAC, based off of FARSH's implementation.
+ *
+ * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
+ * implementations, and it is ridiculously fast.
+ *
+ * We harden it by mixing the original input to the accumulators as well as the product.
+ *
+ * This means that in the (relatively likely) case of a multiply by zero, the
+ * original input is preserved.
+ *
+ * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
+ * cross-pollination, as otherwise the upper and lower halves would be
+ * essentially independent.
+ *
+ * This doesn't matter on 64-bit hashes since they all get merged together in
+ * the end, so we skip the extra step.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
+ */
+
+#if (XXH_VECTOR == XXH_AVX512) \
+     || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
+
+#ifndef XXH_TARGET_AVX512
+# define XXH_TARGET_AVX512  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    __m512i* const xacc = (__m512i *) acc;
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+
+    {
+        /* data_vec    = input[0]; */
+        __m512i const data_vec    = _mm512_loadu_si512   (input);
+        /* key_vec     = secret[0]; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        /* data_key    = data_vec ^ key_vec; */
+        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
+        /* data_key_lo = data_key >> 32; */
+        __m512i const data_key_lo = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+        /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+        __m512i const product     = _mm512_mul_epu32     (data_key, data_key_lo);
+        /* xacc[0] += swap(data_vec); */
+        __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
+        __m512i const sum       = _mm512_add_epi64(*xacc, data_swap);
+        /* xacc[0] += product; */
+        *xacc = _mm512_add_epi64(product, sum);
+    }
+}
+
+/*
+ * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
+ *
+ * Multiplication isn't perfect, as explained by Google in HighwayHash:
+ *
+ *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
+ *  // varying degrees. In descending order of goodness, bytes
+ *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
+ *  // As expected, the upper and lower bytes are much worse.
+ *
+ * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
+ *
+ * Since our algorithm uses a pseudorandom secret to add some variance into the
+ * mix, we don't need to (or want to) mix as often or as much as HighwayHash does.
+ *
+ * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
+ * extraction.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
+ */
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+    {   __m512i* const xacc = (__m512i*) acc;
+        const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
+
+        /* xacc[0] ^= (xacc[0] >> 47) */
+        __m512i const acc_vec     = *xacc;
+        __m512i const shifted     = _mm512_srli_epi64    (acc_vec, 47);
+        __m512i const data_vec    = _mm512_xor_si512     (acc_vec, shifted);
+        /* xacc[0] ^= secret; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
+
+        /* xacc[0] *= XXH_PRIME32_1; */
+        __m512i const data_key_hi = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+        __m512i const prod_lo     = _mm512_mul_epu32     (data_key, prime32);
+        __m512i const prod_hi     = _mm512_mul_epu32     (data_key_hi, prime32);
+        *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
+    XXH_ASSERT(((size_t)customSecret & 63) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
+        __m512i const seed = _mm512_mask_set1_epi64(_mm512_set1_epi64((xxh_i64)seed64), 0xAA, (xxh_i64)(0U - seed64));
+
+        const __m512i* const src  = (const __m512i*) ((const void*) XXH3_kSecret);
+              __m512i* const dest = (      __m512i*) customSecret;
+        int i;
+        XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 63) == 0);
+        for (i=0; i < nbRounds; ++i) {
+            /* GCC has a bug, _mm512_stream_load_si512 accepts 'void*', not 'void const*',
+             * this will warn "discards 'const' qualifier". */
+            union {
+                const __m512i* cp;
+                void* p;
+            } remote_const_void;
+            remote_const_void.cp = src + i;
+            dest[i] = _mm512_add_epi64(_mm512_stream_load_si512(remote_const_void.p), seed);
+    }   }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_AVX2) \
+    || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
+
+#ifndef XXH_TARGET_AVX2
+# define XXH_TARGET_AVX2  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc    =       (__m256i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason. */
+        const         __m256i* const xinput  = (const __m256i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m256i const data_vec    = _mm256_loadu_si256    (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m256i const data_key_lo = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m256i const product     = _mm256_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
+            __m256i const sum       = _mm256_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm256_add_epi64(product, sum);
+    }   }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc = (__m256i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+        const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m256i const acc_vec     = xacc[i];
+            __m256i const shifted     = _mm256_srli_epi64    (acc_vec, 47);
+            __m256i const data_vec    = _mm256_xor_si256     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m256i const data_key_hi = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            __m256i const prod_lo     = _mm256_mul_epu32     (data_key, prime32);
+            __m256i const prod_hi     = _mm256_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
+    (void)(&XXH_writeLE64);
+    XXH_PREFETCH(customSecret);
+    {   __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64);
+
+        const __m256i* const src  = (const __m256i*) ((const void*) XXH3_kSecret);
+              __m256i*       dest = (      __m256i*) customSecret;
+
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dest);
+#       endif
+        XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 31) == 0);
+
+        /* GCC -O2 need unroll loop manually */
+        dest[0] = _mm256_add_epi64(_mm256_stream_load_si256(src+0), seed);
+        dest[1] = _mm256_add_epi64(_mm256_stream_load_si256(src+1), seed);
+        dest[2] = _mm256_add_epi64(_mm256_stream_load_si256(src+2), seed);
+        dest[3] = _mm256_add_epi64(_mm256_stream_load_si256(src+3), seed);
+        dest[4] = _mm256_add_epi64(_mm256_stream_load_si256(src+4), seed);
+        dest[5] = _mm256_add_epi64(_mm256_stream_load_si256(src+5), seed);
+    }
+}
+
+#endif
+
+/* x86dispatch always generates SSE2 */
+#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
+
+#ifndef XXH_TARGET_SSE2
+# define XXH_TARGET_SSE2  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* SSE2 is just a half-scale version of the AVX2 version. */
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc    =       (__m128i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xinput  = (const __m128i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m128i const data_vec    = _mm_loadu_si128   (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m128i const product     = _mm_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2));
+            __m128i const sum       = _mm_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm_add_epi64(product, sum);
+    }   }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc = (__m128i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+        const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m128i const acc_vec     = xacc[i];
+            __m128i const shifted     = _mm_srli_epi64    (acc_vec, 47);
+            __m128i const data_vec    = _mm_xor_si128     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            __m128i const prod_lo     = _mm_mul_epu32     (data_key, prime32);
+            __m128i const prod_hi     = _mm_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
+
+#       if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
+        /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
+        XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) };
+        __m128i const seed = _mm_load_si128((__m128i const*)seed64x2);
+#       else
+        __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
+#       endif
+        int i;
+
+        const void* const src16 = XXH3_kSecret;
+        __m128i* dst16 = (__m128i*) customSecret;
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dst16);
+#       endif
+        XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dst16 & 15) == 0);
+
+        for (i=0; i < nbRounds; ++i) {
+            dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed);
+    }   }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_NEON)
+
+/* forward declarations for the scalar routines */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input,
+                 void const* XXH_RESTRICT secret, size_t lane);
+
+XXH_FORCE_INLINE void
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+                         void const* XXH_RESTRICT secret, size_t lane);
+
+/*!
+ * @internal
+ * @brief The bulk processing loop for NEON.
+ *
+ * The NEON code path is actually partially scalar when running on AArch64. This
+ * is to optimize the pipelining and can have up to 15% speedup depending on the
+ * CPU, and it also mitigates some GCC codegen issues.
+ *
+ * @see XXH3_NEON_LANES for configuring this and details about this optimization.
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0);
+    {
+        uint64x2_t* const xacc = (uint64x2_t *) acc;
+        /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
+        uint8_t const* const xinput = (const uint8_t *) input;
+        uint8_t const* const xsecret  = (const uint8_t *) secret;
+
+        size_t i;
+        /* NEON for the first few lanes (these loops are normally interleaved) */
+        for (i=0; i < XXH3_NEON_LANES / 2; i++) {
+            /* data_vec = xinput[i]; */
+            uint8x16_t data_vec    = vld1q_u8(xinput  + (i * 16));
+            /* key_vec  = xsecret[i];  */
+            uint8x16_t key_vec     = vld1q_u8(xsecret + (i * 16));
+            uint64x2_t data_key;
+            uint32x2_t data_key_lo, data_key_hi;
+            /* xacc[i] += swap(data_vec); */
+            uint64x2_t const data64  = vreinterpretq_u64_u8(data_vec);
+            uint64x2_t const swapped = vextq_u64(data64, data64, 1);
+            xacc[i] = vaddq_u64 (xacc[i], swapped);
+            /* data_key = data_vec ^ key_vec; */
+            data_key = vreinterpretq_u64_u8(veorq_u8(data_vec, key_vec));
+            /* data_key_lo = (uint32x2_t) (data_key & 0xFFFFFFFF);
+             * data_key_hi = (uint32x2_t) (data_key >> 32);
+             * data_key = UNDEFINED; */
+            XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
+            /* xacc[i] += (uint64x2_t) data_key_lo * (uint64x2_t) data_key_hi; */
+            xacc[i] = vmlal_u32 (xacc[i], data_key_lo, data_key_hi);
+
+        }
+        /* Scalar for the remainder. This may be a zero iteration loop. */
+        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+            XXH3_scalarRound(acc, input, secret, i);
+        }
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+    {   uint64x2_t* xacc       = (uint64x2_t*) acc;
+        uint8_t const* xsecret = (uint8_t const*) secret;
+        uint32x2_t prime       = vdup_n_u32 (XXH_PRIME32_1);
+
+        size_t i;
+        /* NEON for the first few lanes (these loops are normally interleaved) */
+        for (i=0; i < XXH3_NEON_LANES / 2; i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            uint64x2_t acc_vec  = xacc[i];
+            uint64x2_t shifted  = vshrq_n_u64 (acc_vec, 47);
+            uint64x2_t data_vec = veorq_u64   (acc_vec, shifted);
+
+            /* xacc[i] ^= xsecret[i]; */
+            uint8x16_t key_vec  = vld1q_u8    (xsecret + (i * 16));
+            uint64x2_t data_key = veorq_u64   (data_vec, vreinterpretq_u64_u8(key_vec));
+
+            /* xacc[i] *= XXH_PRIME32_1 */
+            uint32x2_t data_key_lo, data_key_hi;
+            /* data_key_lo = (uint32x2_t) (xacc[i] & 0xFFFFFFFF);
+             * data_key_hi = (uint32x2_t) (xacc[i] >> 32);
+             * xacc[i] = UNDEFINED; */
+            XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
+            {   /*
+                 * prod_hi = (data_key >> 32) * XXH_PRIME32_1;
+                 *
+                 * Avoid vmul_u32 + vshll_n_u32 since Clang 6 and 7 will
+                 * incorrectly "optimize" this:
+                 *   tmp     = vmul_u32(vmovn_u64(a), vmovn_u64(b));
+                 *   shifted = vshll_n_u32(tmp, 32);
+                 * to this:
+                 *   tmp     = "vmulq_u64"(a, b); // no such thing!
+                 *   shifted = vshlq_n_u64(tmp, 32);
+                 *
+                 * However, unlike SSE, Clang lacks a 64-bit multiply routine
+                 * for NEON, and it scalarizes two 64-bit multiplies instead.
+                 *
+                 * vmull_u32 has the same timing as vmul_u32, and it avoids
+                 * this bug completely.
+                 * See https://bugs.llvm.org/show_bug.cgi?id=39967
+                 */
+                uint64x2_t prod_hi = vmull_u32 (data_key_hi, prime);
+                /* xacc[i] = prod_hi << 32; */
+                xacc[i] = vshlq_n_u64(prod_hi, 32);
+                /* xacc[i] += (prod_hi & 0xFFFFFFFF) * XXH_PRIME32_1; */
+                xacc[i] = vmlal_u32(xacc[i], data_key_lo, prime);
+            }
+        }
+        /* Scalar for the remainder. This may be a zero iteration loop. */
+        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+            XXH3_scalarScrambleRound(acc, secret, i);
+        }
+    }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_VSX)
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_vsx(  void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* presumed aligned */
+    unsigned int* const xacc = (unsigned int*) acc;
+    xxh_u64x2 const* const xinput   = (xxh_u64x2 const*) input;   /* no alignment restriction */
+    xxh_u64x2 const* const xsecret  = (xxh_u64x2 const*) secret;    /* no alignment restriction */
+    xxh_u64x2 const v32 = { 32, 32 };
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+        /* data_vec = xinput[i]; */
+        xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + i);
+        /* key_vec = xsecret[i]; */
+        xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + i);
+        xxh_u64x2 const data_key = data_vec ^ key_vec;
+        /* shuffled = (data_key << 32) | (data_key >> 32); */
+        xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
+        /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
+        xxh_u64x2 const product  = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
+        /* acc_vec = xacc[i]; */
+        xxh_u64x2 acc_vec        = (xxh_u64x2)vec_xl(0, xacc + 4 * i);
+        acc_vec += product;
+
+        /* swap high and low halves */
+#ifdef __s390x__
+        acc_vec += vec_permi(data_vec, data_vec, 2);
+#else
+        acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
+#endif
+        /* xacc[i] = acc_vec; */
+        vec_xst((xxh_u32x4)acc_vec, 0, xacc + 4 * i);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+    {         xxh_u64x2* const xacc    =       (xxh_u64x2*) acc;
+        const xxh_u64x2* const xsecret = (const xxh_u64x2*) secret;
+        /* constants */
+        xxh_u64x2 const v32  = { 32, 32 };
+        xxh_u64x2 const v47 = { 47, 47 };
+        xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 };
+        size_t i;
+        for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            xxh_u64x2 const acc_vec  = xacc[i];
+            xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
+
+            /* xacc[i] ^= xsecret[i]; */
+            xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + i);
+            xxh_u64x2 const data_key = data_vec ^ key_vec;
+
+            /* xacc[i] *= XXH_PRIME32_1 */
+            /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF);  */
+            xxh_u64x2 const prod_even  = XXH_vec_mule((xxh_u32x4)data_key, prime);
+            /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
+            xxh_u64x2 const prod_odd  = XXH_vec_mulo((xxh_u32x4)data_key, prime);
+            xacc[i] = prod_odd + (prod_even << v32);
+    }   }
+}
+
+#endif
+
+/* scalar variants - universal */
+
+/*!
+ * @internal
+ * @brief Scalar round for @ref XXH3_accumulate_512_scalar().
+ *
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
+ */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc,
+                 void const* XXH_RESTRICT input,
+                 void const* XXH_RESTRICT secret,
+                 size_t lane)
+{
+    xxh_u64* xacc = (xxh_u64*) acc;
+    xxh_u8 const* xinput  = (xxh_u8 const*) input;
+    xxh_u8 const* xsecret = (xxh_u8 const*) secret;
+    XXH_ASSERT(lane < XXH_ACC_NB);
+    XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
+    {
+        xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
+        xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
+        xacc[lane ^ 1] += data_val; /* swap adjacent lanes */
+        xacc[lane] += XXH_mult32to64(data_key & 0xFFFFFFFF, data_key >> 32);
+    }
+}
+
+/*!
+ * @internal
+ * @brief Processes a 64 byte block of data using the scalar path.
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    size_t i;
+    for (i=0; i < XXH_ACC_NB; i++) {
+        XXH3_scalarRound(acc, input, secret, i);
+    }
+}
+
+/*!
+ * @internal
+ * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
+ *
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
+ */
+XXH_FORCE_INLINE void
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+                         void const* XXH_RESTRICT secret,
+                         size_t lane)
+{
+    xxh_u64* const xacc = (xxh_u64*) acc;   /* presumed aligned */
+    const xxh_u8* const xsecret = (const xxh_u8*) secret;   /* no alignment restriction */
+    XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
+    XXH_ASSERT(lane < XXH_ACC_NB);
+    {
+        xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
+        xxh_u64 acc64 = xacc[lane];
+        acc64 = XXH_xorshift64(acc64, 47);
+        acc64 ^= key64;
+        acc64 *= XXH_PRIME32_1;
+        xacc[lane] = acc64;
+    }
+}
+
+/*!
+ * @internal
+ * @brief Scrambles the accumulators after a large chunk has been read
+ */
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    size_t i;
+    for (i=0; i < XXH_ACC_NB; i++) {
+        XXH3_scalarScrambleRound(acc, secret, i);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    /*
+     * We need a separate pointer for the hack below,
+     * which requires a non-const pointer.
+     * Any decent compiler will optimize this out otherwise.
+     */
+    const xxh_u8* kSecretPtr = XXH3_kSecret;
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+
+#if defined(__clang__) && defined(__aarch64__)
+    /*
+     * UGLY HACK:
+     * Clang generates a bunch of MOV/MOVK pairs for aarch64, and they are
+     * placed sequentially, in order, at the top of the unrolled loop.
+     *
+     * While MOVK is great for generating constants (2 cycles for a 64-bit
+     * constant compared to 4 cycles for LDR), it fights for bandwidth with
+     * the arithmetic instructions.
+     *
+     *   I   L   S
+     * MOVK
+     * MOVK
+     * MOVK
+     * MOVK
+     * ADD
+     * SUB      STR
+     *          STR
+     * By forcing loads from memory (as the asm line causes Clang to assume
+     * that XXH3_kSecretPtr has been changed), the pipelines are used more
+     * efficiently:
+     *   I   L   S
+     *      LDR
+     *  ADD LDR
+     *  SUB     STR
+     *          STR
+     *
+     * See XXH3_NEON_LANES for details on the pipsline.
+     *
+     * XXH3_64bits_withSeed, len == 256, Snapdragon 835
+     *   without hack: 2654.4 MB/s
+     *   with hack:    3202.9 MB/s
+     */
+    XXH_COMPILER_GUARD(kSecretPtr);
+#endif
+    /*
+     * Note: in debug mode, this overrides the asm optimization
+     * and Clang will emit MOVK chains again.
+     */
+    XXH_ASSERT(kSecretPtr == XXH3_kSecret);
+
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
+        int i;
+        for (i=0; i < nbRounds; i++) {
+            /*
+             * The asm hack causes Clang to assume that kSecretPtr aliases with
+             * customSecret, and on aarch64, this prevented LDP from merging two
+             * loads together for free. Putting the loads together before the stores
+             * properly generates LDP.
+             */
+            xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i)     + seed64;
+            xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64;
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i,     lo);
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi);
+    }   }
+}
+
+
+typedef void (*XXH3_f_accumulate_512)(void* XXH_RESTRICT, const void*, const void*);
+typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
+typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
+
+
+#if (XXH_VECTOR == XXH_AVX512)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx512
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
+
+#elif (XXH_VECTOR == XXH_AVX2)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
+
+#elif (XXH_VECTOR == XXH_SSE2)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_sse2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
+
+#elif (XXH_VECTOR == XXH_NEON)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_neon
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_neon
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#elif (XXH_VECTOR == XXH_VSX)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_vsx
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#else /* scalar */
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#endif
+
+
+
+#ifndef XXH_PREFETCH_DIST
+#  ifdef __clang__
+#    define XXH_PREFETCH_DIST 320
+#  else
+#    if (XXH_VECTOR == XXH_AVX512)
+#      define XXH_PREFETCH_DIST 512
+#    else
+#      define XXH_PREFETCH_DIST 384
+#    endif
+#  endif  /* __clang__ */
+#endif  /* XXH_PREFETCH_DIST */
+
+/*
+ * XXH3_accumulate()
+ * Loops over XXH3_accumulate_512().
+ * Assumption: nbStripes will not overflow the secret size
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate(     xxh_u64* XXH_RESTRICT acc,
+                const xxh_u8* XXH_RESTRICT input,
+                const xxh_u8* XXH_RESTRICT secret,
+                      size_t nbStripes,
+                      XXH3_f_accumulate_512 f_acc512)
+{
+    size_t n;
+    for (n = 0; n < nbStripes; n++ ) {
+        const xxh_u8* const in = input + n*XXH_STRIPE_LEN;
+        XXH_PREFETCH(in + XXH_PREFETCH_DIST);
+        f_acc512(acc,
+                 in,
+                 secret + n*XXH_SECRET_CONSUME_RATE);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
+                      const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate_512 f_acc512,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
+    size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
+    size_t const nb_blocks = (len - 1) / block_len;
+
+    size_t n;
+
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+
+    for (n = 0; n < nb_blocks; n++) {
+        XXH3_accumulate(acc, input + n*block_len, secret, nbStripesPerBlock, f_acc512);
+        f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
+    }
+
+    /* last partial block */
+    XXH_ASSERT(len > XXH_STRIPE_LEN);
+    {   size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
+        XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
+        XXH3_accumulate(acc, input + nb_blocks*block_len, secret, nbStripes, f_acc512);
+
+        /* last stripe */
+        {   const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
+#define XXH_SECRET_LASTACC_START 7  /* not aligned on 8, last secret is different from acc & scrambler */
+            f_acc512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
+    }   }
+}
+
+XXH_FORCE_INLINE xxh_u64
+XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret)
+{
+    return XXH3_mul128_fold64(
+               acc[0] ^ XXH_readLE64(secret),
+               acc[1] ^ XXH_readLE64(secret+8) );
+}
+
+static XXH64_hash_t
+XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start)
+{
+    xxh_u64 result64 = start;
+    size_t i = 0;
+
+    for (i = 0; i < 4; i++) {
+        result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+        /*
+         * UGLY HACK:
+         * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
+         * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
+         * XXH3_64bits, len == 256, Snapdragon 835:
+         *   without hack: 2063.7 MB/s
+         *   with hack:    2560.7 MB/s
+         */
+        XXH_COMPILER_GUARD(result64);
+#endif
+    }
+
+    return XXH3_avalanche(result64);
+}
+
+#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \
+                        XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 }
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
+                           const void* XXH_RESTRICT secret, size_t secretSize,
+                           XXH3_f_accumulate_512 f_acc512,
+                           XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc512, f_scramble);
+
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    /* do not align on 8, so that the secret is different from the accumulator */
+#define XXH_SECRET_MERGEACCS_START 11
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    return XXH3_mergeAccs(acc, (const xxh_u8*)secret + XXH_SECRET_MERGEACCS_START, (xxh_u64)len * XXH_PRIME64_1);
+}
+
+/*
+ * It's important for performance to transmit secret's size (when it's static)
+ * so that the compiler can properly optimize the vectorized loop.
+ * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                             XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * It's preferable for performance that XXH3_hashLong is not inlined,
+ * as it results in a smaller function for small data, easier to the instruction cache.
+ * Note that inside this no_inline function, we do inline the internal loop,
+ * and provide a statically defined secret size to allow optimization of vector loop.
+ */
+XXH_NO_INLINE XXH64_hash_t
+XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
+                          XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * XXH3_hashLong_64b_withSeed():
+ * Generate a custom key based on alteration of default XXH3_kSecret with the seed,
+ * and then use this key for long mode hashing.
+ *
+ * This operation is decently fast but nonetheless costs a little bit of time.
+ * Try to avoid it whenever possible (typically when seed==0).
+ *
+ * It's important for performance that XXH3_hashLong is not inlined. Not sure
+ * why (uop cache maybe?), but the difference is large and easily measurable.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
+                                    XXH64_hash_t seed,
+                                    XXH3_f_accumulate_512 f_acc512,
+                                    XXH3_f_scrambleAcc f_scramble,
+                                    XXH3_f_initCustomSecret f_initSec)
+{
+    if (seed == 0)
+        return XXH3_hashLong_64b_internal(input, len,
+                                          XXH3_kSecret, sizeof(XXH3_kSecret),
+                                          f_acc512, f_scramble);
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed);
+        return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
+                                          f_acc512, f_scramble);
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed(const void* input, size_t len,
+                           XXH64_hash_t seed, const xxh_u8* secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
+                XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
+
+
+typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t,
+                                          XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t);
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
+                     XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                     XXH3_hashLong64_f f_hashLong)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secretLen` condition is not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     * Also, note that function signature doesn't offer room to return an error.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen);
+}
+
+
+/* ===   Public entry point   === */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* input, size_t len)
+{
+    return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_internal(input, len, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
+}
+
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_64b_withSecret(input, len, seed, (const xxh_u8*)secret, secretSize);
+}
+
+
+/* ===   XXH3 streaming   === */
+
+/*
+ * Malloc's a pointer that is always aligned to align.
+ *
+ * This must be freed with `XXH_alignedFree()`.
+ *
+ * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte
+ * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2
+ * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON.
+ *
+ * This underalignment previously caused a rather obvious crash which went
+ * completely unnoticed due to XXH3_createState() not actually being tested.
+ * Credit to RedSpah for noticing this bug.
+ *
+ * The alignment is done manually: Functions like posix_memalign or _mm_malloc
+ * are avoided: To maintain portability, we would have to write a fallback
+ * like this anyways, and besides, testing for the existence of library
+ * functions without relying on external build tools is impossible.
+ *
+ * The method is simple: Overallocate, manually align, and store the offset
+ * to the original behind the returned pointer.
+ *
+ * Align must be a power of 2 and 8 <= align <= 128.
+ */
+static void* XXH_alignedMalloc(size_t s, size_t align)
+{
+    XXH_ASSERT(align <= 128 && align >= 8); /* range check */
+    XXH_ASSERT((align & (align-1)) == 0);   /* power of 2 */
+    XXH_ASSERT(s != 0 && s < (s + align));  /* empty/overflow */
+    {   /* Overallocate to make room for manual realignment and an offset byte */
+        xxh_u8* base = (xxh_u8*)XXH_malloc(s + align);
+        if (base != NULL) {
+            /*
+             * Get the offset needed to align this pointer.
+             *
+             * Even if the returned pointer is aligned, there will always be
+             * at least one byte to store the offset to the original pointer.
+             */
+            size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
+            /* Add the offset for the now-aligned pointer */
+            xxh_u8* ptr = base + offset;
+
+            XXH_ASSERT((size_t)ptr % align == 0);
+
+            /* Store the offset immediately before the returned pointer. */
+            ptr[-1] = (xxh_u8)offset;
+            return ptr;
+        }
+        return NULL;
+    }
+}
+/*
+ * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
+ * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
+ */
+static void XXH_alignedFree(void* p)
+{
+    if (p != NULL) {
+        xxh_u8* ptr = (xxh_u8*)p;
+        /* Get the offset byte we added in XXH_malloc. */
+        xxh_u8 offset = ptr[-1];
+        /* Free the original malloc'd pointer */
+        xxh_u8* base = ptr - offset;
+        XXH_free(base);
+    }
+}
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
+{
+    XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
+    if (state==NULL) return NULL;
+    XXH3_INITSTATE(state);
+    return state;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
+{
+    XXH_alignedFree(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state)
+{
+    XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
+}
+
+static void
+XXH3_reset_internal(XXH3_state_t* statePtr,
+                    XXH64_hash_t seed,
+                    const void* secret, size_t secretSize)
+{
+    size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
+    size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
+    XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
+    XXH_ASSERT(statePtr != NULL);
+    /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
+    memset((char*)statePtr + initStart, 0, initLength);
+    statePtr->acc[0] = XXH_PRIME32_3;
+    statePtr->acc[1] = XXH_PRIME64_1;
+    statePtr->acc[2] = XXH_PRIME64_2;
+    statePtr->acc[3] = XXH_PRIME64_3;
+    statePtr->acc[4] = XXH_PRIME64_4;
+    statePtr->acc[5] = XXH_PRIME32_2;
+    statePtr->acc[6] = XXH_PRIME64_5;
+    statePtr->acc[7] = XXH_PRIME32_1;
+    statePtr->seed = seed;
+    statePtr->useSeed = (seed != 0);
+    statePtr->extSecret = (const unsigned char*)secret;
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+    statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
+    statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset(XXH3_state_t* statePtr)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, secret, secretSize);
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (seed==0) return XXH3_64bits_reset(statePtr);
+    if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
+        XXH3_initCustomSecret(statePtr->customSecret, seed);
+    XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed64)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, seed64, secret, secretSize);
+    statePtr->useSeed = 1; /* always, even if seed64==0 */
+    return XXH_OK;
+}
+
+/* Note : when XXH3_consumeStripes() is invoked,
+ * there must be a guarantee that at least one more byte must be consumed from input
+ * so that the function can blindly consume all stripes using the "normal" secret segment */
+XXH_FORCE_INLINE void
+XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
+                    size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
+                    const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
+                    const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
+                    XXH3_f_accumulate_512 f_acc512,
+                    XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ASSERT(nbStripes <= nbStripesPerBlock);  /* can handle max 1 scramble per invocation */
+    XXH_ASSERT(*nbStripesSoFarPtr < nbStripesPerBlock);
+    if (nbStripesPerBlock - *nbStripesSoFarPtr <= nbStripes) {
+        /* need a scrambling operation */
+        size_t const nbStripesToEndofBlock = nbStripesPerBlock - *nbStripesSoFarPtr;
+        size_t const nbStripesAfterBlock = nbStripes - nbStripesToEndofBlock;
+        XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripesToEndofBlock, f_acc512);
+        f_scramble(acc, secret + secretLimit);
+        XXH3_accumulate(acc, input + nbStripesToEndofBlock * XXH_STRIPE_LEN, secret, nbStripesAfterBlock, f_acc512);
+        *nbStripesSoFarPtr = nbStripesAfterBlock;
+    } else {
+        XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripes, f_acc512);
+        *nbStripesSoFarPtr += nbStripes;
+    }
+}
+
+#ifndef XXH3_STREAM_USE_STACK
+# ifndef __clang__ /* clang doesn't need additional stack space */
+#   define XXH3_STREAM_USE_STACK 1
+# endif
+#endif
+/*
+ * Both XXH3_64bits_update and XXH3_128bits_update use this routine.
+ */
+XXH_FORCE_INLINE XXH_errorcode
+XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
+            const xxh_u8* XXH_RESTRICT input, size_t len,
+            XXH3_f_accumulate_512 f_acc512,
+            XXH3_f_scrambleAcc f_scramble)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    XXH_ASSERT(state != NULL);
+    {   const xxh_u8* const bEnd = input + len;
+        const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* For some reason, gcc and MSVC seem to suffer greatly
+         * when operating accumulators directly into state.
+         * Operating into stack space seems to enable proper optimization.
+         * clang, on the other hand, doesn't seem to need this trick */
+        XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8]; memcpy(acc, state->acc, sizeof(acc));
+#else
+        xxh_u64* XXH_RESTRICT const acc = state->acc;
+#endif
+        state->totalLen += len;
+        XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
+
+        /* small input : just fill in tmp buffer */
+        if (state->bufferedSize + len <= XXH3_INTERNALBUFFER_SIZE) {
+            XXH_memcpy(state->buffer + state->bufferedSize, input, len);
+            state->bufferedSize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
+
+        /* total input is now > XXH3_INTERNALBUFFER_SIZE */
+        #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
+        XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0);   /* clean multiple */
+
+        /*
+         * Internal buffer is partially filled (always, except at beginning)
+         * Complete it, then consume it.
+         */
+        if (state->bufferedSize) {
+            size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
+            XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
+            input += loadSize;
+            XXH3_consumeStripes(acc,
+                               &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                state->buffer, XXH3_INTERNALBUFFER_STRIPES,
+                                secret, state->secretLimit,
+                                f_acc512, f_scramble);
+            state->bufferedSize = 0;
+        }
+        XXH_ASSERT(input < bEnd);
+
+        /* large input to consume : ingest per full block */
+        if ((size_t)(bEnd - input) > state->nbStripesPerBlock * XXH_STRIPE_LEN) {
+            size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
+            XXH_ASSERT(state->nbStripesPerBlock >= state->nbStripesSoFar);
+            /* join to current block's end */
+            {   size_t const nbStripesToEnd = state->nbStripesPerBlock - state->nbStripesSoFar;
+                XXH_ASSERT(nbStripesToEnd <= nbStripes);
+                XXH3_accumulate(acc, input, secret + state->nbStripesSoFar * XXH_SECRET_CONSUME_RATE, nbStripesToEnd, f_acc512);
+                f_scramble(acc, secret + state->secretLimit);
+                state->nbStripesSoFar = 0;
+                input += nbStripesToEnd * XXH_STRIPE_LEN;
+                nbStripes -= nbStripesToEnd;
+            }
+            /* consume per entire blocks */
+            while(nbStripes >= state->nbStripesPerBlock) {
+                XXH3_accumulate(acc, input, secret, state->nbStripesPerBlock, f_acc512);
+                f_scramble(acc, secret + state->secretLimit);
+                input += state->nbStripesPerBlock * XXH_STRIPE_LEN;
+                nbStripes -= state->nbStripesPerBlock;
+            }
+            /* consume last partial block */
+            XXH3_accumulate(acc, input, secret, nbStripes, f_acc512);
+            input += nbStripes * XXH_STRIPE_LEN;
+            XXH_ASSERT(input < bEnd);  /* at least some bytes left */
+            state->nbStripesSoFar = nbStripes;
+            /* buffer predecessor of last partial stripe */
+            XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+            XXH_ASSERT(bEnd - input <= XXH_STRIPE_LEN);
+        } else {
+            /* content to consume <= block size */
+            /* Consume input by a multiple of internal buffer size */
+            if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
+                const xxh_u8* const limit = bEnd - XXH3_INTERNALBUFFER_SIZE;
+                do {
+                    XXH3_consumeStripes(acc,
+                                       &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                        input, XXH3_INTERNALBUFFER_STRIPES,
+                                        secret, state->secretLimit,
+                                        f_acc512, f_scramble);
+                    input += XXH3_INTERNALBUFFER_SIZE;
+                } while (input<limit);
+                /* buffer predecessor of last partial stripe */
+                XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+            }
+        }
+
+        /* Some remaining input (always) : buffer it */
+        XXH_ASSERT(input < bEnd);
+        XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
+        XXH_ASSERT(state->bufferedSize == 0);
+        XXH_memcpy(state->buffer, input, (size_t)(bEnd-input));
+        state->bufferedSize = (XXH32_hash_t)(bEnd-input);
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* save stack accumulators into state */
+        memcpy(state->acc, acc, sizeof(acc));
+#endif
+    }
+
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_update(XXH3_state_t* state, const void* input, size_t len)
+{
+    return XXH3_update(state, (const xxh_u8*)input, len,
+                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+
+XXH_FORCE_INLINE void
+XXH3_digest_long (XXH64_hash_t* acc,
+                  const XXH3_state_t* state,
+                  const unsigned char* secret)
+{
+    /*
+     * Digest on a local copy. This way, the state remains unaltered, and it can
+     * continue ingesting more input afterwards.
+     */
+    XXH_memcpy(acc, state->acc, sizeof(state->acc));
+    if (state->bufferedSize >= XXH_STRIPE_LEN) {
+        size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
+        size_t nbStripesSoFar = state->nbStripesSoFar;
+        XXH3_consumeStripes(acc,
+                           &nbStripesSoFar, state->nbStripesPerBlock,
+                            state->buffer, nbStripes,
+                            secret, state->secretLimit,
+                            XXH3_accumulate_512, XXH3_scrambleAcc);
+        /* last stripe */
+        XXH3_accumulate_512(acc,
+                            state->buffer + state->bufferedSize - XXH_STRIPE_LEN,
+                            secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+    } else {  /* bufferedSize < XXH_STRIPE_LEN */
+        xxh_u8 lastStripe[XXH_STRIPE_LEN];
+        size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
+        XXH_ASSERT(state->bufferedSize > 0);  /* there is always some input buffered */
+        XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
+        XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
+        XXH3_accumulate_512(acc,
+                            lastStripe,
+                            secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+    }
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        return XXH3_mergeAccs(acc,
+                              secret + XXH_SECRET_MERGEACCS_START,
+                              (xxh_u64)state->totalLen * XXH_PRIME64_1);
+    }
+    /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
+    if (state->useSeed)
+        return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                  secret, state->secretLimit + XXH_STRIPE_LEN);
+}
+
+
+
+/* ==========================================
+ * XXH3 128 bits (a.k.a XXH128)
+ * ==========================================
+ * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant,
+ * even without counting the significantly larger output size.
+ *
+ * For example, extra steps are taken to avoid the seed-dependent collisions
+ * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
+ *
+ * This strength naturally comes at the cost of some speed, especially on short
+ * lengths. Note that longer hashes are about as fast as the 64-bit version
+ * due to it using only a slight modification of the 64-bit loop.
+ *
+ * XXH128 is also more oriented towards 64-bit machines. It is still extremely
+ * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
+ */
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    /* A doubled version of 1to3_64b with different constants. */
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8 const c1 = input[0];
+        xxh_u8 const c2 = input[len >> 1];
+        xxh_u8 const c3 = input[len - 1];
+        xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24)
+                                | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
+        xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
+        xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed;
+        xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
+        xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
+        XXH128_hash_t h128;
+        h128.low64  = XXH64_avalanche(keyed_lo);
+        h128.high64 = XXH64_avalanche(keyed_hi);
+        return h128;
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input_lo = XXH_readLE32(input);
+        xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
+        xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed;
+        xxh_u64 const keyed = input_64 ^ bitflip;
+
+        /* Shift len to the left to ensure it is even, this avoids even multiplies. */
+        XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
+
+        m128.high64 += (m128.low64 << 1);
+        m128.low64  ^= (m128.high64 >> 3);
+
+        m128.low64   = XXH_xorshift64(m128.low64, 35);
+        m128.low64  *= 0x9FB21C651E98DF25ULL;
+        m128.low64   = XXH_xorshift64(m128.low64, 28);
+        m128.high64  = XXH3_avalanche(m128.high64);
+        return m128;
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed;
+        xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed;
+        xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64       input_hi = XXH_readLE64(input + len - 8);
+        XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
+        /*
+         * Put len in the middle of m128 to ensure that the length gets mixed to
+         * both the low and high bits in the 128x64 multiply below.
+         */
+        m128.low64 += (xxh_u64)(len - 1) << 54;
+        input_hi   ^= bitfliph;
+        /*
+         * Add the high 32 bits of input_hi to the high 32 bits of m128, then
+         * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
+         * the high 64 bits of m128.
+         *
+         * The best approach to this operation is different on 32-bit and 64-bit.
+         */
+        if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */
+            /*
+             * 32-bit optimized version, which is more readable.
+             *
+             * On 32-bit, it removes an ADC and delays a dependency between the two
+             * halves of m128.high64, but it generates an extra mask on 64-bit.
+             */
+            m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
+        } else {
+            /*
+             * 64-bit optimized (albeit more confusing) version.
+             *
+             * Uses some properties of addition and multiplication to remove the mask:
+             *
+             * Let:
+             *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
+             *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
+             *    c = XXH_PRIME32_2
+             *
+             *    a + (b * c)
+             * Inverse Property: x + y - x == y
+             *    a + (b * (1 + c - 1))
+             * Distributive Property: x * (y + z) == (x * y) + (x * z)
+             *    a + (b * 1) + (b * (c - 1))
+             * Identity Property: x * 1 == x
+             *    a + b + (b * (c - 1))
+             *
+             * Substitute a, b, and c:
+             *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             *
+             * Since input_hi.hi + input_hi.lo == input_hi, we get this:
+             *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             */
+            m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
+        }
+        /* m128 ^= XXH_swap64(m128 >> 64); */
+        m128.low64  ^= XXH_swap64(m128.high64);
+
+        {   /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
+            XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
+            h128.high64 += m128.high64 * XXH_PRIME64_2;
+
+            h128.low64   = XXH3_avalanche(h128.low64);
+            h128.high64  = XXH3_avalanche(h128.high64);
+            return h128;
+    }   }
+}
+
+/*
+ * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
+        if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
+        {   XXH128_hash_t h128;
+            xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72);
+            xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88);
+            h128.low64 = XXH64_avalanche(seed ^ bitflipl);
+            h128.high64 = XXH64_avalanche( seed ^ bitfliph);
+            return h128;
+    }   }
+}
+
+/*
+ * A bit slower than XXH3_mix16B, but handles multiply by zero better.
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
+              const xxh_u8* secret, XXH64_hash_t seed)
+{
+    acc.low64  += XXH3_mix16B (input_1, secret+0, seed);
+    acc.low64  ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
+    acc.high64 += XXH3_mix16B (input_2, secret+16, seed);
+    acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
+    return acc;
+}
+
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   XXH128_hash_t acc;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed);
+                }
+                acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed);
+            }
+            acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
+        }
+        acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
+
+XXH_NO_INLINE XXH128_hash_t
+XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                       const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                       XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+    {   XXH128_hash_t acc;
+        int const nbRounds = (int)len / 32;
+        int i;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+        for (i=0; i<4; i++) {
+            acc = XXH128_mix32B(acc,
+                                input  + (32 * i),
+                                input  + (32 * i) + 16,
+                                secret + (32 * i),
+                                seed);
+        }
+        acc.low64 = XXH3_avalanche(acc.low64);
+        acc.high64 = XXH3_avalanche(acc.high64);
+        XXH_ASSERT(nbRounds >= 4);
+        for (i=4 ; i < nbRounds; i++) {
+            acc = XXH128_mix32B(acc,
+                                input + (32 * i),
+                                input + (32 * i) + 16,
+                                secret + XXH3_MIDSIZE_STARTOFFSET + (32 * (i - 4)),
+                                seed);
+        }
+        /* last bytes */
+        acc = XXH128_mix32B(acc,
+                            input + len - 16,
+                            input + len - 32,
+                            secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
+                            0ULL - seed);
+
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
+                            const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate_512 f_acc512,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc512, f_scramble);
+
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    {   XXH128_hash_t h128;
+        h128.low64  = XXH3_mergeAccs(acc,
+                                     secret + XXH_SECRET_MERGEACCS_START,
+                                     (xxh_u64)len * XXH_PRIME64_1);
+        h128.high64 = XXH3_mergeAccs(acc,
+                                     secret + secretSize
+                                            - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                     ~((xxh_u64)len * XXH_PRIME64_2));
+        return h128;
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH128_hash_t
+XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
+                           XXH64_hash_t seed64,
+                           const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
+                                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * It's important for performance to pass @secretLen (when it's static)
+ * to the compiler, so that it can properly optimize the vectorized loop.
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                              XXH64_hash_t seed64,
+                              const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
+                                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
+                                XXH64_hash_t seed64,
+                                XXH3_f_accumulate_512 f_acc512,
+                                XXH3_f_scrambleAcc f_scramble,
+                                XXH3_f_initCustomSecret f_initSec)
+{
+    if (seed64 == 0)
+        return XXH3_hashLong_128b_internal(input, len,
+                                           XXH3_kSecret, sizeof(XXH3_kSecret),
+                                           f_acc512, f_scramble);
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed64);
+        return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
+                                           f_acc512, f_scramble);
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed(const void* input, size_t len,
+                            XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
+                XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
+
+typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
+                                            XXH64_hash_t, const void* XXH_RESTRICT, size_t);
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_128bits_internal(const void* input, size_t len,
+                      XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                      XXH3_hashLong128_f f_hl128)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secret` conditions are not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hl128(input, len, seed64, secret, secretLen);
+}
+
+
+/* ===   Public XXH128 API   === */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* input, size_t len)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_default);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 (const xxh_u8*)secret, secretSize,
+                                 XXH3_hashLong_128b_withSecret);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_internal(input, len, seed,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_withSeed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_withSeed(input, len, seed);
+}
+
+
+/* ===   XXH3 128-bit streaming   === */
+
+/*
+ * All initialization and update functions are identical to 64-bit streaming variant.
+ * The only difference is the finalization routine.
+ */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset(XXH3_state_t* statePtr)
+{
+    return XXH3_64bits_reset(statePtr);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSeed(statePtr, seed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_update(XXH3_state_t* state, const void* input, size_t len)
+{
+    return XXH3_update(state, (const xxh_u8*)input, len,
+                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+        {   XXH128_hash_t h128;
+            h128.low64  = XXH3_mergeAccs(acc,
+                                         secret + XXH_SECRET_MERGEACCS_START,
+                                         (xxh_u64)state->totalLen * XXH_PRIME64_1);
+            h128.high64 = XXH3_mergeAccs(acc,
+                                         secret + state->secretLimit + XXH_STRIPE_LEN
+                                                - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
+            return h128;
+        }
+    }
+    /* len <= XXH3_MIDSIZE_MAX : short code */
+    if (state->seed)
+        return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                   secret, state->secretLimit + XXH_STRIPE_LEN);
+}
+
+/* 128-bit utility functions */
+
+#include <string.h>   /* memcmp, memcpy */
+
+/* return : 1 is equal, 0 if different */
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
+{
+    /* note : XXH128_hash_t is compact, it has no padding byte */
+    return !(memcmp(&h1, &h2, sizeof(h1)));
+}
+
+/* This prototype is compatible with stdlib's qsort().
+ * return : >0 if *h128_1  > *h128_2
+ *          <0 if *h128_1  < *h128_2
+ *          =0 if *h128_1 == *h128_2  */
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2)
+{
+    XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
+    XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
+    int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
+    /* note : bets that, in most cases, hash values are different */
+    if (hcmp) return hcmp;
+    return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
+}
+
+
+/*======   Canonical representation   ======*/
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) {
+        hash.high64 = XXH_swap64(hash.high64);
+        hash.low64  = XXH_swap64(hash.low64);
+    }
+    XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
+    XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128_hashFromCanonical(const XXH128_canonical_t* src)
+{
+    XXH128_hash_t h;
+    h.high64 = XXH_readBE64(src);
+    h.low64  = XXH_readBE64(src->digest + 8);
+    return h;
+}
+
+
+
+/* ==========================================
+ * Secret generators
+ * ==========================================
+ */
+#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
+
+XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128)
+{
+    XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
+    XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize)
+{
+#if (XXH_DEBUGLEVEL >= 1)
+    XXH_ASSERT(secretBuffer != NULL);
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+#else
+    /* production mode, assert() are disabled */
+    if (secretBuffer == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+#endif
+
+    if (customSeedSize == 0) {
+        customSeed = XXH3_kSecret;
+        customSeedSize = XXH_SECRET_DEFAULT_SIZE;
+    }
+#if (XXH_DEBUGLEVEL >= 1)
+    XXH_ASSERT(customSeed != NULL);
+#else
+    if (customSeed == NULL) return XXH_ERROR;
+#endif
+
+    /* Fill secretBuffer with a copy of customSeed - repeat as needed */
+    {   size_t pos = 0;
+        while (pos < secretSize) {
+            size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
+            memcpy((char*)secretBuffer + pos, customSeed, toCopy);
+            pos += toCopy;
+    }   }
+
+    {   size_t const nbSeg16 = secretSize / 16;
+        size_t n;
+        XXH128_canonical_t scrambler;
+        XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
+        for (n=0; n<nbSeg16; n++) {
+            XXH128_hash_t const h128 = XXH128(&scrambler, sizeof(scrambler), n);
+            XXH3_combine16((char*)secretBuffer + n*16, h128);
+        }
+        /* last segment */
+        XXH3_combine16((char*)secretBuffer + secretSize - 16, XXH128_hashFromCanonical(&scrambler));
+    }
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed)
+{
+    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+    XXH3_initCustomSecret(secret, seed);
+    XXH_ASSERT(secretBuffer != NULL);
+    memcpy(secretBuffer, secret, XXH_SECRET_DEFAULT_SIZE);
+}
+
+
+
+/* Pop our optimization override from above */
+#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
+  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
+#  pragma GCC pop_options
+#endif
+
+#endif  /* XXH_NO_LONG_LONG */
+
+#endif  /* XXH_NO_XXH3 */
+
+/*!
+ * @}
+ */
+#endif  /* XXH_IMPLEMENTATION */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_common.c b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_common.c
new file mode 100644
index 0000000..3f04c22
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_common.c
@@ -0,0 +1,48 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#define ZSTD_DEPS_NEED_MALLOC
+#include "error_private.h"
+#include "zstd_internal.h"
+
+
+/*-****************************************
+*  Version
+******************************************/
+unsigned ZSTD_versionNumber(void) { return ZSTD_VERSION_NUMBER; }
+
+const char* ZSTD_versionString(void) { return ZSTD_VERSION_STRING; }
+
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+#undef ZSTD_isError   /* defined within zstd_internal.h */
+/*! ZSTD_isError() :
+ *  tells if a return value is an error code
+ *  symbol is required for external callers */
+unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTD_getErrorName() :
+ *  provides error code string from function result (useful for debugging) */
+const char* ZSTD_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+/*! ZSTD_getError() :
+ *  convert a `size_t` function result into a proper ZSTD_errorCode enum */
+ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
+
+/*! ZSTD_getErrorString() :
+ *  provides error code string from enum */
+const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_deps.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_deps.h
new file mode 100644
index 0000000..4d767ae
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_deps.h
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* This file provides common libc dependencies that zstd requires.
+ * The purpose is to allow replacing this file with a custom implementation
+ * to compile zstd without libc support.
+ */
+
+/* Need:
+ * NULL
+ * INT_MAX
+ * UINT_MAX
+ * ZSTD_memcpy()
+ * ZSTD_memset()
+ * ZSTD_memmove()
+ */
+#ifndef ZSTD_DEPS_COMMON
+#define ZSTD_DEPS_COMMON
+
+#include <limits.h>
+#include <stddef.h>
+#include <string.h>
+
+#if defined(__GNUC__) && __GNUC__ >= 4
+# define ZSTD_memcpy(d,s,l) __builtin_memcpy((d),(s),(l))
+# define ZSTD_memmove(d,s,l) __builtin_memmove((d),(s),(l))
+# define ZSTD_memset(p,v,l) __builtin_memset((p),(v),(l))
+#else
+# define ZSTD_memcpy(d,s,l) memcpy((d),(s),(l))
+# define ZSTD_memmove(d,s,l) memmove((d),(s),(l))
+# define ZSTD_memset(p,v,l) memset((p),(v),(l))
+#endif
+
+#endif /* ZSTD_DEPS_COMMON */
+
+/* Need:
+ * ZSTD_malloc()
+ * ZSTD_free()
+ * ZSTD_calloc()
+ */
+#ifdef ZSTD_DEPS_NEED_MALLOC
+#ifndef ZSTD_DEPS_MALLOC
+#define ZSTD_DEPS_MALLOC
+
+#include <stdlib.h>
+
+#define ZSTD_malloc(s) malloc(s)
+#define ZSTD_calloc(n,s) calloc((n), (s))
+#define ZSTD_free(p) free((p))
+
+#endif /* ZSTD_DEPS_MALLOC */
+#endif /* ZSTD_DEPS_NEED_MALLOC */
+
+/*
+ * Provides 64-bit math support.
+ * Need:
+ * U64 ZSTD_div64(U64 dividend, U32 divisor)
+ */
+#ifdef ZSTD_DEPS_NEED_MATH64
+#ifndef ZSTD_DEPS_MATH64
+#define ZSTD_DEPS_MATH64
+
+#define ZSTD_div64(dividend, divisor) ((dividend) / (divisor))
+
+#endif /* ZSTD_DEPS_MATH64 */
+#endif /* ZSTD_DEPS_NEED_MATH64 */
+
+/* Need:
+ * assert()
+ */
+#ifdef ZSTD_DEPS_NEED_ASSERT
+#ifndef ZSTD_DEPS_ASSERT
+#define ZSTD_DEPS_ASSERT
+
+#include <assert.h>
+
+#endif /* ZSTD_DEPS_ASSERT */
+#endif /* ZSTD_DEPS_NEED_ASSERT */
+
+/* Need:
+ * ZSTD_DEBUG_PRINT()
+ */
+#ifdef ZSTD_DEPS_NEED_IO
+#ifndef ZSTD_DEPS_IO
+#define ZSTD_DEPS_IO
+
+#include <stdio.h>
+#define ZSTD_DEBUG_PRINT(...) fprintf(stderr, __VA_ARGS__)
+
+#endif /* ZSTD_DEPS_IO */
+#endif /* ZSTD_DEPS_NEED_IO */
+
+/* Only requested when <stdint.h> is known to be present.
+ * Need:
+ * intptr_t
+ */
+#ifdef ZSTD_DEPS_NEED_STDINT
+#ifndef ZSTD_DEPS_STDINT
+#define ZSTD_DEPS_STDINT
+
+#include <stdint.h>
+
+#endif /* ZSTD_DEPS_STDINT */
+#endif /* ZSTD_DEPS_NEED_STDINT */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_internal.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_internal.h
new file mode 100644
index 0000000..1f942f2
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_internal.h
@@ -0,0 +1,392 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+/* this module contains definitions which must be identical
+ * across compression, decompression and dictBuilder.
+ * It also contains a few functions useful to at least 2 of them
+ * and which benefit from being inlined */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "compiler.h"
+#include "cpu.h"
+#include "mem.h"
+#include "debug.h"                 /* assert, DEBUGLOG, RAWLOG, g_debuglevel */
+#include "error_private.h"
+#define ZSTD_STATIC_LINKING_ONLY
+#include "../zstd.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#include "huf.h"
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY  /* XXH64_state_t */
+#endif
+#include "xxhash.h"                /* XXH_reset, update, digest */
+#ifndef ZSTD_NO_TRACE
+#  include "zstd_trace.h"
+#else
+#  define ZSTD_TRACE 0
+#endif
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* ---- static assert (debug) --- */
+#define ZSTD_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)
+#define ZSTD_isError ERR_isError   /* for inlining */
+#define FSE_isError  ERR_isError
+#define HUF_isError  ERR_isError
+
+
+/*-*************************************
+*  shared macros
+***************************************/
+#undef MIN
+#undef MAX
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+#define BOUNDED(min,val,max) (MAX(min,MIN(val,max)))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTD_OPT_NUM    (1<<12)
+
+#define ZSTD_REP_NUM      3                 /* number of repcodes */
+static UNUSED_ATTR const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTD_WINDOWLOG_ABSOLUTEMIN 10
+static UNUSED_ATTR const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
+static UNUSED_ATTR const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
+
+#define ZSTD_FRAMEIDSIZE 4   /* magic number size */
+
+#define ZSTD_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
+static UNUSED_ATTR const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
+typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
+
+#define ZSTD_FRAMECHECKSUMSIZE 4
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */)   /* for a non-null block */
+#define MIN_LITERALS_FOR_4_STREAMS 6
+
+typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingType_e;
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+
+#define Litbits  8
+#define LitHufLog 11
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML   52
+#define MaxLL   35
+#define DefaultMaxOff 28
+#define MaxOff  31
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+#define MaxFSELog  MAX(MAX(MLFSELog, LLFSELog), OffFSELog)
+#define MaxMLBits 16
+#define MaxLLBits 16
+
+#define ZSTD_MAX_HUF_HEADER_SIZE 128 /* header + <= 127 byte tree description */
+/* Each table cannot take more than #symbols * FSELog bits */
+#define ZSTD_MAX_FSE_HEADERS_SIZE (((MaxML + 1) * MLFSELog + (MaxLL + 1) * LLFSELog + (MaxOff + 1) * OffFSELog + 7) / 8)
+
+static UNUSED_ATTR const U8 LL_bits[MaxLL+1] = {
+     0, 0, 0, 0, 0, 0, 0, 0,
+     0, 0, 0, 0, 0, 0, 0, 0,
+     1, 1, 1, 1, 2, 2, 3, 3,
+     4, 6, 7, 8, 9,10,11,12,
+    13,14,15,16
+};
+static UNUSED_ATTR const S16 LL_defaultNorm[MaxLL+1] = {
+     4, 3, 2, 2, 2, 2, 2, 2,
+     2, 2, 2, 2, 2, 1, 1, 1,
+     2, 2, 2, 2, 2, 2, 2, 2,
+     2, 3, 2, 1, 1, 1, 1, 1,
+    -1,-1,-1,-1
+};
+#define LL_DEFAULTNORMLOG 6  /* for static allocation */
+static UNUSED_ATTR const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
+
+static UNUSED_ATTR const U8 ML_bits[MaxML+1] = {
+     0, 0, 0, 0, 0, 0, 0, 0,
+     0, 0, 0, 0, 0, 0, 0, 0,
+     0, 0, 0, 0, 0, 0, 0, 0,
+     0, 0, 0, 0, 0, 0, 0, 0,
+     1, 1, 1, 1, 2, 2, 3, 3,
+     4, 4, 5, 7, 8, 9,10,11,
+    12,13,14,15,16
+};
+static UNUSED_ATTR const S16 ML_defaultNorm[MaxML+1] = {
+     1, 4, 3, 2, 2, 2, 2, 2,
+     2, 1, 1, 1, 1, 1, 1, 1,
+     1, 1, 1, 1, 1, 1, 1, 1,
+     1, 1, 1, 1, 1, 1, 1, 1,
+     1, 1, 1, 1, 1, 1, 1, 1,
+     1, 1, 1, 1, 1, 1,-1,-1,
+    -1,-1,-1,-1,-1
+};
+#define ML_DEFAULTNORMLOG 6  /* for static allocation */
+static UNUSED_ATTR const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
+
+static UNUSED_ATTR const S16 OF_defaultNorm[DefaultMaxOff+1] = {
+     1, 1, 1, 1, 1, 1, 2, 2,
+     2, 1, 1, 1, 1, 1, 1, 1,
+     1, 1, 1, 1, 1, 1, 1, 1,
+    -1,-1,-1,-1,-1
+};
+#define OF_DEFAULTNORMLOG 5  /* for static allocation */
+static UNUSED_ATTR const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTD_copy8(void* dst, const void* src) {
+#if defined(ZSTD_ARCH_ARM_NEON)
+    vst1_u8((uint8_t*)dst, vld1_u8((const uint8_t*)src));
+#else
+    ZSTD_memcpy(dst, src, 8);
+#endif
+}
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/* Need to use memmove here since the literal buffer can now be located within
+   the dst buffer. In circumstances where the op "catches up" to where the
+   literal buffer is, there can be partial overlaps in this call on the final
+   copy if the literal is being shifted by less than 16 bytes. */
+static void ZSTD_copy16(void* dst, const void* src) {
+#if defined(ZSTD_ARCH_ARM_NEON)
+    vst1q_u8((uint8_t*)dst, vld1q_u8((const uint8_t*)src));
+#elif defined(ZSTD_ARCH_X86_SSE2)
+    _mm_storeu_si128((__m128i*)dst, _mm_loadu_si128((const __m128i*)src));
+#elif defined(__clang__)
+    ZSTD_memmove(dst, src, 16);
+#else
+    /* ZSTD_memmove is not inlined properly by gcc */
+    BYTE copy16_buf[16];
+    ZSTD_memcpy(copy16_buf, src, 16);
+    ZSTD_memcpy(dst, copy16_buf, 16);
+#endif
+}
+#define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
+
+#define WILDCOPY_OVERLENGTH 32
+#define WILDCOPY_VECLEN 16
+
+typedef enum {
+    ZSTD_no_overlap,
+    ZSTD_overlap_src_before_dst
+    /*  ZSTD_overlap_dst_before_src, */
+} ZSTD_overlap_e;
+
+/*! ZSTD_wildcopy() :
+ *  Custom version of ZSTD_memcpy(), can over read/write up to WILDCOPY_OVERLENGTH bytes (if length==0)
+ *  @param ovtype controls the overlap detection
+ *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
+ *         - ZSTD_overlap_src_before_dst: The src and dst may overlap, but they MUST be at least 8 bytes apart.
+ *           The src buffer must be before the dst buffer.
+ */
+MEM_STATIC FORCE_INLINE_ATTR
+void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e const ovtype)
+{
+    ptrdiff_t diff = (BYTE*)dst - (const BYTE*)src;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+
+    if (ovtype == ZSTD_overlap_src_before_dst && diff < WILDCOPY_VECLEN) {
+        /* Handle short offset copies. */
+        do {
+            COPY8(op, ip)
+        } while (op < oend);
+    } else {
+        assert(diff >= WILDCOPY_VECLEN || diff <= -WILDCOPY_VECLEN);
+        /* Separate out the first COPY16() call because the copy length is
+         * almost certain to be short, so the branches have different
+         * probabilities. Since it is almost certain to be short, only do
+         * one COPY16() in the first call. Then, do two calls per loop since
+         * at that point it is more likely to have a high trip count.
+         */
+        ZSTD_copy16(op, ip);
+        if (16 >= length) return;
+        op += 16;
+        ip += 16;
+        do {
+            COPY16(op, ip);
+            COPY16(op, ip);
+        }
+        while (op < oend);
+    }
+}
+
+MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const length = MIN(dstCapacity, srcSize);
+    if (length > 0) {
+        ZSTD_memcpy(dst, src, length);
+    }
+    return length;
+}
+
+/* define "workspace is too large" as this number of times larger than needed */
+#define ZSTD_WORKSPACETOOLARGE_FACTOR 3
+
+/* when workspace is continuously too large
+ * during at least this number of times,
+ * context's memory usage is considered wasteful,
+ * because it's sized to handle a worst case scenario which rarely happens.
+ * In which case, resize it down to free some memory */
+#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128
+
+/* Controls whether the input/output buffer is buffered or stable. */
+typedef enum {
+    ZSTD_bm_buffered = 0,  /* Buffer the input/output */
+    ZSTD_bm_stable = 1     /* ZSTD_inBuffer/ZSTD_outBuffer is stable */
+} ZSTD_bufferMode_e;
+
+
+/*-*******************************************
+*  Private declarations
+*********************************************/
+typedef struct seqDef_s {
+    U32 offBase;   /* offBase == Offset + ZSTD_REP_NUM, or repcode 1,2,3 */
+    U16 litLength;
+    U16 mlBase;    /* mlBase == matchLength - MINMATCH */
+} seqDef;
+
+/* Controls whether seqStore has a single "long" litLength or matchLength. See seqStore_t. */
+typedef enum {
+    ZSTD_llt_none = 0,             /* no longLengthType */
+    ZSTD_llt_literalLength = 1,    /* represents a long literal */
+    ZSTD_llt_matchLength = 2       /* represents a long match */
+} ZSTD_longLengthType_e;
+
+typedef struct {
+    seqDef* sequencesStart;
+    seqDef* sequences;      /* ptr to end of sequences */
+    BYTE*  litStart;
+    BYTE*  lit;             /* ptr to end of literals */
+    BYTE*  llCode;
+    BYTE*  mlCode;
+    BYTE*  ofCode;
+    size_t maxNbSeq;
+    size_t maxNbLit;
+
+    /* longLengthPos and longLengthType to allow us to represent either a single litLength or matchLength
+     * in the seqStore that has a value larger than U16 (if it exists). To do so, we increment
+     * the existing value of the litLength or matchLength by 0x10000.
+     */
+    ZSTD_longLengthType_e longLengthType;
+    U32                   longLengthPos;  /* Index of the sequence to apply long length modification to */
+} seqStore_t;
+
+typedef struct {
+    U32 litLength;
+    U32 matchLength;
+} ZSTD_sequenceLength;
+
+/**
+ * Returns the ZSTD_sequenceLength for the given sequences. It handles the decoding of long sequences
+ * indicated by longLengthPos and longLengthType, and adds MINMATCH back to matchLength.
+ */
+MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore, seqDef const* seq)
+{
+    ZSTD_sequenceLength seqLen;
+    seqLen.litLength = seq->litLength;
+    seqLen.matchLength = seq->mlBase + MINMATCH;
+    if (seqStore->longLengthPos == (U32)(seq - seqStore->sequencesStart)) {
+        if (seqStore->longLengthType == ZSTD_llt_literalLength) {
+            seqLen.litLength += 0x10000;
+        }
+        if (seqStore->longLengthType == ZSTD_llt_matchLength) {
+            seqLen.matchLength += 0x10000;
+        }
+    }
+    return seqLen;
+}
+
+/**
+ * Contains the compressed frame size and an upper-bound for the decompressed frame size.
+ * Note: before using `compressedSize`, check for errors using ZSTD_isError().
+ *       similarly, before using `decompressedBound`, check for errors using:
+ *          `decompressedBound != ZSTD_CONTENTSIZE_ERROR`
+ */
+typedef struct {
+    size_t nbBlocks;
+    size_t compressedSize;
+    unsigned long long decompressedBound;
+} ZSTD_frameSizeInfo;   /* decompress & legacy */
+
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);   /* compress & dictBuilder */
+int ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
+
+
+/* ZSTD_invalidateRepCodes() :
+ * ensures next compression will not use repcodes from previous block.
+ * Note : only works with regular variant;
+ *        do not use with extDict variant ! */
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx);   /* zstdmt, adaptive_compression (shouldn't get this definition from here) */
+
+
+typedef struct {
+    blockType_e blockType;
+    U32 lastBlock;
+    U32 origSize;
+} blockProperties_t;   /* declared here for decompress and fullbench */
+
+/*! ZSTD_getcBlockSize() :
+ *  Provides the size of compressed block from block header `src` */
+/* Used by: decompress, fullbench (does not get its definition from here) */
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr);
+
+/*! ZSTD_decodeSeqHeaders() :
+ *  decode sequence header from src */
+/* Used by: decompress, fullbench (does not get its definition from here) */
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
+                       const void* src, size_t srcSize);
+
+/**
+ * @returns true iff the CPU supports dynamic BMI2 dispatch.
+ */
+MEM_STATIC int ZSTD_cpuSupportsBmi2(void)
+{
+    ZSTD_cpuid_t cpuid = ZSTD_cpuid();
+    return ZSTD_cpuid_bmi1(cpuid) && ZSTD_cpuid_bmi2(cpuid);
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* ZSTD_CCOMMON_H_MODULE */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_trace.h b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_trace.h
new file mode 100644
index 0000000..da20534
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/common/zstd_trace.h
@@ -0,0 +1,163 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_TRACE_H
+#define ZSTD_TRACE_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include <stddef.h>
+
+/* weak symbol support
+ * For now, enable conservatively:
+ * - Only GNUC
+ * - Only ELF
+ * - Only x86-64, i386 and aarch64
+ * Also, explicitly disable on platforms known not to work so they aren't
+ * forgotten in the future.
+ */
+#if !defined(ZSTD_HAVE_WEAK_SYMBOLS) && \
+    defined(__GNUC__) && defined(__ELF__) && \
+    (defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86) || defined(__aarch64__)) && \
+    !defined(__APPLE__) && !defined(_WIN32) && !defined(__MINGW32__) && \
+    !defined(__CYGWIN__) && !defined(_AIX)
+#  define ZSTD_HAVE_WEAK_SYMBOLS 1
+#else
+#  define ZSTD_HAVE_WEAK_SYMBOLS 0
+#endif
+#if ZSTD_HAVE_WEAK_SYMBOLS
+#  define ZSTD_WEAK_ATTR __attribute__((__weak__))
+#else
+#  define ZSTD_WEAK_ATTR
+#endif
+
+/* Only enable tracing when weak symbols are available. */
+#ifndef ZSTD_TRACE
+#  define ZSTD_TRACE ZSTD_HAVE_WEAK_SYMBOLS
+#endif
+
+#if ZSTD_TRACE
+
+struct ZSTD_CCtx_s;
+struct ZSTD_DCtx_s;
+struct ZSTD_CCtx_params_s;
+
+typedef struct {
+    /**
+     * ZSTD_VERSION_NUMBER
+     *
+     * This is guaranteed to be the first member of ZSTD_trace.
+     * Otherwise, this struct is not stable between versions. If
+     * the version number does not match your expectation, you
+     * should not interpret the rest of the struct.
+     */
+    unsigned version;
+    /**
+     * Non-zero if streaming (de)compression is used.
+     */
+    unsigned streaming;
+    /**
+     * The dictionary ID.
+     */
+    unsigned dictionaryID;
+    /**
+     * Is the dictionary cold?
+     * Only set on decompression.
+     */
+    unsigned dictionaryIsCold;
+    /**
+     * The dictionary size or zero if no dictionary.
+     */
+    size_t dictionarySize;
+    /**
+     * The uncompressed size of the data.
+     */
+    size_t uncompressedSize;
+    /**
+     * The compressed size of the data.
+     */
+    size_t compressedSize;
+    /**
+     * The fully resolved CCtx parameters (NULL on decompression).
+     */
+    struct ZSTD_CCtx_params_s const* params;
+    /**
+     * The ZSTD_CCtx pointer (NULL on decompression).
+     */
+    struct ZSTD_CCtx_s const* cctx;
+    /**
+     * The ZSTD_DCtx pointer (NULL on compression).
+     */
+    struct ZSTD_DCtx_s const* dctx;
+} ZSTD_Trace;
+
+/**
+ * A tracing context. It must be 0 when tracing is disabled.
+ * Otherwise, any non-zero value returned by a tracing begin()
+ * function is presented to any subsequent calls to end().
+ *
+ * Any non-zero value is treated as tracing is enabled and not
+ * interpreted by the library.
+ *
+ * Two possible uses are:
+ * * A timestamp for when the begin() function was called.
+ * * A unique key identifying the (de)compression, like the
+ *   address of the [dc]ctx pointer if you need to track
+ *   more information than just a timestamp.
+ */
+typedef unsigned long long ZSTD_TraceCtx;
+
+/**
+ * Trace the beginning of a compression call.
+ * @param cctx The dctx pointer for the compression.
+ *             It can be used as a key to map begin() to end().
+ * @returns Non-zero if tracing is enabled. The return value is
+ *          passed to ZSTD_trace_compress_end().
+ */
+ZSTD_WEAK_ATTR ZSTD_TraceCtx ZSTD_trace_compress_begin(
+    struct ZSTD_CCtx_s const* cctx);
+
+/**
+ * Trace the end of a compression call.
+ * @param ctx The return value of ZSTD_trace_compress_begin().
+ * @param trace The zstd tracing info.
+ */
+ZSTD_WEAK_ATTR void ZSTD_trace_compress_end(
+    ZSTD_TraceCtx ctx,
+    ZSTD_Trace const* trace);
+
+/**
+ * Trace the beginning of a decompression call.
+ * @param dctx The dctx pointer for the decompression.
+ *             It can be used as a key to map begin() to end().
+ * @returns Non-zero if tracing is enabled. The return value is
+ *          passed to ZSTD_trace_compress_end().
+ */
+ZSTD_WEAK_ATTR ZSTD_TraceCtx ZSTD_trace_decompress_begin(
+    struct ZSTD_DCtx_s const* dctx);
+
+/**
+ * Trace the end of a decompression call.
+ * @param ctx The return value of ZSTD_trace_decompress_begin().
+ * @param trace The zstd tracing info.
+ */
+ZSTD_WEAK_ATTR void ZSTD_trace_decompress_end(
+    ZSTD_TraceCtx ctx,
+    ZSTD_Trace const* trace);
+
+#endif /* ZSTD_TRACE */
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_TRACE_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/clevels.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/clevels.h
new file mode 100644
index 0000000..c18da46
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/clevels.h
@@ -0,0 +1,134 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_CLEVELS_H
+#define ZSTD_CLEVELS_H
+
+#define ZSTD_STATIC_LINKING_ONLY  /* ZSTD_compressionParameters  */
+#include "../zstd.h"
+
+/*-=====  Pre-defined compression levels  =====-*/
+
+#define ZSTD_MAX_CLEVEL     22
+
+#ifdef __GNUC__
+__attribute__((__unused__))
+#endif
+
+static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
+{   /* "default" - for any srcSize > 256 KB */
+    /* W,  C,  H,  S,  L, TL, strat */
+    { 19, 12, 13,  1,  6,  1, ZSTD_fast    },  /* base for negative levels */
+    { 19, 13, 14,  1,  7,  0, ZSTD_fast    },  /* level  1 */
+    { 20, 15, 16,  1,  6,  0, ZSTD_fast    },  /* level  2 */
+    { 21, 16, 17,  1,  5,  0, ZSTD_dfast   },  /* level  3 */
+    { 21, 18, 18,  1,  5,  0, ZSTD_dfast   },  /* level  4 */
+    { 21, 18, 19,  3,  5,  2, ZSTD_greedy  },  /* level  5 */
+    { 21, 18, 19,  3,  5,  4, ZSTD_lazy    },  /* level  6 */
+    { 21, 19, 20,  4,  5,  8, ZSTD_lazy    },  /* level  7 */
+    { 21, 19, 20,  4,  5, 16, ZSTD_lazy2   },  /* level  8 */
+    { 22, 20, 21,  4,  5, 16, ZSTD_lazy2   },  /* level  9 */
+    { 22, 21, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 10 */
+    { 22, 21, 22,  6,  5, 16, ZSTD_lazy2   },  /* level 11 */
+    { 22, 22, 23,  6,  5, 32, ZSTD_lazy2   },  /* level 12 */
+    { 22, 22, 22,  4,  5, 32, ZSTD_btlazy2 },  /* level 13 */
+    { 22, 22, 23,  5,  5, 32, ZSTD_btlazy2 },  /* level 14 */
+    { 22, 23, 23,  6,  5, 32, ZSTD_btlazy2 },  /* level 15 */
+    { 22, 22, 22,  5,  5, 48, ZSTD_btopt   },  /* level 16 */
+    { 23, 23, 22,  5,  4, 64, ZSTD_btopt   },  /* level 17 */
+    { 23, 23, 22,  6,  3, 64, ZSTD_btultra },  /* level 18 */
+    { 23, 24, 22,  7,  3,256, ZSTD_btultra2},  /* level 19 */
+    { 25, 25, 23,  7,  3,256, ZSTD_btultra2},  /* level 20 */
+    { 26, 26, 24,  7,  3,512, ZSTD_btultra2},  /* level 21 */
+    { 27, 27, 25,  9,  3,999, ZSTD_btultra2},  /* level 22 */
+},
+{   /* for srcSize <= 256 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 18, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 18, 13, 14,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 18, 14, 14,  1,  5,  0, ZSTD_dfast   },  /* level  2 */
+    { 18, 16, 16,  1,  4,  0, ZSTD_dfast   },  /* level  3 */
+    { 18, 16, 17,  3,  5,  2, ZSTD_greedy  },  /* level  4.*/
+    { 18, 17, 18,  5,  5,  2, ZSTD_greedy  },  /* level  5.*/
+    { 18, 18, 19,  3,  5,  4, ZSTD_lazy    },  /* level  6.*/
+    { 18, 18, 19,  4,  4,  4, ZSTD_lazy    },  /* level  7 */
+    { 18, 18, 19,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 18, 18, 19,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 18, 18, 19,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 18, 18, 19,  5,  4, 12, ZSTD_btlazy2 },  /* level 11.*/
+    { 18, 19, 19,  7,  4, 12, ZSTD_btlazy2 },  /* level 12.*/
+    { 18, 18, 19,  4,  4, 16, ZSTD_btopt   },  /* level 13 */
+    { 18, 18, 19,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
+    { 18, 18, 19,  6,  3,128, ZSTD_btopt   },  /* level 15.*/
+    { 18, 19, 19,  6,  3,128, ZSTD_btultra },  /* level 16.*/
+    { 18, 19, 19,  8,  3,256, ZSTD_btultra },  /* level 17.*/
+    { 18, 19, 19,  6,  3,128, ZSTD_btultra2},  /* level 18.*/
+    { 18, 19, 19,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 18, 19, 19, 10,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 18, 19, 19, 12,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 18, 19, 19, 13,  3,999, ZSTD_btultra2},  /* level 22.*/
+},
+{   /* for srcSize <= 128 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 17, 12, 13,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 17, 13, 15,  1,  5,  0, ZSTD_fast    },  /* level  2 */
+    { 17, 15, 16,  2,  5,  0, ZSTD_dfast   },  /* level  3 */
+    { 17, 17, 17,  2,  4,  0, ZSTD_dfast   },  /* level  4 */
+    { 17, 16, 17,  3,  4,  2, ZSTD_greedy  },  /* level  5 */
+    { 17, 16, 17,  3,  4,  4, ZSTD_lazy    },  /* level  6 */
+    { 17, 16, 17,  3,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 17, 16, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 17, 16, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 17, 16, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 17, 17, 17,  5,  4,  8, ZSTD_btlazy2 },  /* level 11 */
+    { 17, 18, 17,  7,  4, 12, ZSTD_btlazy2 },  /* level 12 */
+    { 17, 18, 17,  3,  4, 12, ZSTD_btopt   },  /* level 13.*/
+    { 17, 18, 17,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
+    { 17, 18, 17,  6,  3,256, ZSTD_btopt   },  /* level 15.*/
+    { 17, 18, 17,  6,  3,128, ZSTD_btultra },  /* level 16.*/
+    { 17, 18, 17,  8,  3,256, ZSTD_btultra },  /* level 17.*/
+    { 17, 18, 17, 10,  3,512, ZSTD_btultra },  /* level 18.*/
+    { 17, 18, 17,  5,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 17, 18, 17,  7,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 17, 18, 17,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 17, 18, 17, 11,  3,999, ZSTD_btultra2},  /* level 22.*/
+},
+{   /* for srcSize <= 16 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 14, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 14, 14, 15,  1,  5,  0, ZSTD_fast    },  /* level  1 */
+    { 14, 14, 15,  1,  4,  0, ZSTD_fast    },  /* level  2 */
+    { 14, 14, 15,  2,  4,  0, ZSTD_dfast   },  /* level  3 */
+    { 14, 14, 14,  4,  4,  2, ZSTD_greedy  },  /* level  4 */
+    { 14, 14, 14,  3,  4,  4, ZSTD_lazy    },  /* level  5.*/
+    { 14, 14, 14,  4,  4,  8, ZSTD_lazy2   },  /* level  6 */
+    { 14, 14, 14,  6,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 14, 14, 14,  8,  4,  8, ZSTD_lazy2   },  /* level  8.*/
+    { 14, 15, 14,  5,  4,  8, ZSTD_btlazy2 },  /* level  9.*/
+    { 14, 15, 14,  9,  4,  8, ZSTD_btlazy2 },  /* level 10.*/
+    { 14, 15, 14,  3,  4, 12, ZSTD_btopt   },  /* level 11.*/
+    { 14, 15, 14,  4,  3, 24, ZSTD_btopt   },  /* level 12.*/
+    { 14, 15, 14,  5,  3, 32, ZSTD_btultra },  /* level 13.*/
+    { 14, 15, 15,  6,  3, 64, ZSTD_btultra },  /* level 14.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btultra },  /* level 15.*/
+    { 14, 15, 15,  5,  3, 48, ZSTD_btultra2},  /* level 16.*/
+    { 14, 15, 15,  6,  3,128, ZSTD_btultra2},  /* level 17.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btultra2},  /* level 18.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 14, 15, 15,  8,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 14, 15, 15,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 14, 15, 15, 10,  3,999, ZSTD_btultra2},  /* level 22.*/
+},
+};
+
+
+
+#endif  /* ZSTD_CLEVELS_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/fse_compress.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/fse_compress.c
new file mode 100644
index 0000000..5d37708
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/fse_compress.c
@@ -0,0 +1,624 @@
+/* ******************************************************************
+ * FSE : Finite State Entropy encoder
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include "../common/compiler.h"
+#include "../common/mem.h"        /* U32, U16, etc. */
+#include "../common/debug.h"      /* assert, DEBUGLOG */
+#include "hist.h"       /* HIST_count_wksp */
+#include "../common/bitstream.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "../common/fse.h"
+#include "../common/error_private.h"
+#define ZSTD_DEPS_NEED_MALLOC
+#define ZSTD_DEPS_NEED_MATH64
+#include "../common/zstd_deps.h"  /* ZSTD_malloc, ZSTD_free, ZSTD_memcpy, ZSTD_memset */
+#include "../common/bits.h" /* ZSTD_highbit32 */
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_isError ERR_isError
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
+ * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
+ */
+size_t FSE_buildCTable_wksp(FSE_CTable* ct,
+                      const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                            void* workSpace, size_t wkspSize)
+{
+    U32 const tableSize = 1 << tableLog;
+    U32 const tableMask = tableSize - 1;
+    void* const ptr = ct;
+    U16* const tableU16 = ( (U16*) ptr) + 2;
+    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
+    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
+    U32 const step = FSE_TABLESTEP(tableSize);
+    U32 const maxSV1 = maxSymbolValue+1;
+
+    U16* cumul = (U16*)workSpace;   /* size = maxSV1 */
+    FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSV1+1));  /* size = tableSize */
+
+    U32 highThreshold = tableSize-1;
+
+    assert(((size_t)workSpace & 1) == 0);  /* Must be 2 bytes-aligned */
+    if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
+    /* CTable header */
+    tableU16[-2] = (U16) tableLog;
+    tableU16[-1] = (U16) maxSymbolValue;
+    assert(tableLog < 16);   /* required for threshold strategy to work */
+
+    /* For explanations on how to distribute symbol values over the table :
+     * https://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+
+     #ifdef __clang_analyzer__
+     ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
+     #endif
+
+    /* symbol start positions */
+    {   U32 u;
+        cumul[0] = 0;
+        for (u=1; u <= maxSV1; u++) {
+            if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
+                cumul[u] = cumul[u-1] + 1;
+                tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
+            } else {
+                assert(normalizedCounter[u-1] >= 0);
+                cumul[u] = cumul[u-1] + (U16)normalizedCounter[u-1];
+                assert(cumul[u] >= cumul[u-1]);  /* no overflow */
+        }   }
+        cumul[maxSV1] = (U16)(tableSize+1);
+    }
+
+    /* Spread symbols */
+    if (highThreshold == tableSize - 1) {
+        /* Case for no low prob count symbols. Lay down 8 bytes at a time
+         * to reduce branch misses since we are operating on a small block
+         */
+        BYTE* const spread = tableSymbol + tableSize; /* size = tableSize + 8 (may write beyond tableSize) */
+        {   U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                assert(n>=0);
+                pos += (size_t)n;
+            }
+        }
+        /* Spread symbols across the table. Lack of lowprob symbols means that
+         * we don't need variable sized inner loop, so we can unroll the loop and
+         * reduce branch misses.
+         */
+        {   size_t position = 0;
+            size_t s;
+            size_t const unroll = 2; /* Experimentally determined optimal unroll */
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableSymbol[uPosition] = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);   /* Must have initialized all positions */
+        }
+    } else {
+        U32 position = 0;
+        U32 symbol;
+        for (symbol=0; symbol<maxSV1; symbol++) {
+            int nbOccurrences;
+            int const freq = normalizedCounter[symbol];
+            for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
+                tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
+                position = (position + step) & tableMask;
+                while (position > highThreshold)
+                    position = (position + step) & tableMask;   /* Low proba area */
+        }   }
+        assert(position==0);  /* Must have initialized all positions */
+    }
+
+    /* Build table */
+    {   U32 u; for (u=0; u<tableSize; u++) {
+        FSE_FUNCTION_TYPE s = tableSymbol[u];   /* note : static analyzer may not understand tableSymbol is properly initialized */
+        tableU16[cumul[s]++] = (U16) (tableSize+u);   /* TableU16 : sorted by symbol order; gives next state value */
+    }   }
+
+    /* Build Symbol Transformation Table */
+    {   unsigned total = 0;
+        unsigned s;
+        for (s=0; s<=maxSymbolValue; s++) {
+            switch (normalizedCounter[s])
+            {
+            case  0:
+                /* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
+                symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
+                break;
+
+            case -1:
+            case  1:
+                symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
+                assert(total <= INT_MAX);
+                symbolTT[s].deltaFindState = (int)(total - 1);
+                total ++;
+                break;
+            default :
+                assert(normalizedCounter[s] > 1);
+                {   U32 const maxBitsOut = tableLog - ZSTD_highbit32 ((U32)normalizedCounter[s]-1);
+                    U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
+                    symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
+                    symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
+                    total +=  (unsigned)normalizedCounter[s];
+    }   }   }   }
+
+#if 0  /* debug : symbol costs */
+    DEBUGLOG(5, "\n --- table statistics : ");
+    {   U32 symbol;
+        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
+            DEBUGLOG(5, "%3u: w=%3i,   maxBits=%u, fracBits=%.2f",
+                symbol, normalizedCounter[symbol],
+                FSE_getMaxNbBits(symbolTT, symbol),
+                (double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
+    }   }
+#endif
+
+    return 0;
+}
+
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/*-**************************************************************
+*  FSE NCount encoding
+****************************************************************/
+size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
+{
+    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
+                                   + 4 /* bitCount initialized at 4 */
+                                   + 2 /* first two symbols may use one additional bit each */) / 8)
+                                    + 1 /* round up to whole nb bytes */
+                                    + 2 /* additional two bytes for bitstream flush */;
+    return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
+}
+
+static size_t
+FSE_writeNCount_generic (void* header, size_t headerBufferSize,
+                   const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                         unsigned writeIsSafe)
+{
+    BYTE* const ostart = (BYTE*) header;
+    BYTE* out = ostart;
+    BYTE* const oend = ostart + headerBufferSize;
+    int nbBits;
+    const int tableSize = 1 << tableLog;
+    int remaining;
+    int threshold;
+    U32 bitStream = 0;
+    int bitCount = 0;
+    unsigned symbol = 0;
+    unsigned const alphabetSize = maxSymbolValue + 1;
+    int previousIs0 = 0;
+
+    /* Table Size */
+    bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
+    bitCount  += 4;
+
+    /* Init */
+    remaining = tableSize+1;   /* +1 for extra accuracy */
+    threshold = tableSize;
+    nbBits = tableLog+1;
+
+    while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
+        if (previousIs0) {
+            unsigned start = symbol;
+            while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
+            if (symbol == alphabetSize) break;   /* incorrect distribution */
+            while (symbol >= start+24) {
+                start+=24;
+                bitStream += 0xFFFFU << bitCount;
+                if ((!writeIsSafe) && (out > oend-2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE) bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out+=2;
+                bitStream>>=16;
+            }
+            while (symbol >= start+3) {
+                start+=3;
+                bitStream += 3 << bitCount;
+                bitCount += 2;
+            }
+            bitStream += (symbol-start) << bitCount;
+            bitCount += 2;
+            if (bitCount>16) {
+                if ((!writeIsSafe) && (out > oend - 2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE)bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out += 2;
+                bitStream >>= 16;
+                bitCount -= 16;
+        }   }
+        {   int count = normalizedCounter[symbol++];
+            int const max = (2*threshold-1) - remaining;
+            remaining -= count < 0 ? -count : count;
+            count++;   /* +1 for extra accuracy */
+            if (count>=threshold)
+                count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
+            bitStream += count << bitCount;
+            bitCount  += nbBits;
+            bitCount  -= (count<max);
+            previousIs0  = (count==1);
+            if (remaining<1) return ERROR(GENERIC);
+            while (remaining<threshold) { nbBits--; threshold>>=1; }
+        }
+        if (bitCount>16) {
+            if ((!writeIsSafe) && (out > oend - 2))
+                return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+            out[0] = (BYTE)bitStream;
+            out[1] = (BYTE)(bitStream>>8);
+            out += 2;
+            bitStream >>= 16;
+            bitCount -= 16;
+    }   }
+
+    if (remaining != 1)
+        return ERROR(GENERIC);  /* incorrect normalized distribution */
+    assert(symbol <= alphabetSize);
+
+    /* flush remaining bitStream */
+    if ((!writeIsSafe) && (out > oend - 2))
+        return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+    out[0] = (BYTE)bitStream;
+    out[1] = (BYTE)(bitStream>>8);
+    out+= (bitCount+7) /8;
+
+    return (out-ostart);
+}
+
+
+size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                  const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
+
+    if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
+        return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
+
+    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
+}
+
+
+/*-**************************************************************
+*  FSE Compression Code
+****************************************************************/
+
+/* provides the minimum logSize to safely represent a distribution */
+static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
+{
+    U32 minBitsSrc = ZSTD_highbit32((U32)(srcSize)) + 1;
+    U32 minBitsSymbols = ZSTD_highbit32(maxSymbolValue) + 2;
+    U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    return minBits;
+}
+
+unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
+{
+    U32 maxBitsSrc = ZSTD_highbit32((U32)(srcSize - 1)) - minus;
+    U32 tableLog = maxTableLog;
+    U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
+    if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
+    if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
+    if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
+    return tableLog;
+}
+
+unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+{
+    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
+}
+
+/* Secondary normalization method.
+   To be used when primary method fails. */
+
+static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue, short lowProbCount)
+{
+    short const NOT_YET_ASSIGNED = -2;
+    U32 s;
+    U32 distributed = 0;
+    U32 ToDistribute;
+
+    /* Init */
+    U32 const lowThreshold = (U32)(total >> tableLog);
+    U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
+
+    for (s=0; s<=maxSymbolValue; s++) {
+        if (count[s] == 0) {
+            norm[s]=0;
+            continue;
+        }
+        if (count[s] <= lowThreshold) {
+            norm[s] = lowProbCount;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+        if (count[s] <= lowOne) {
+            norm[s] = 1;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+
+        norm[s]=NOT_YET_ASSIGNED;
+    }
+    ToDistribute = (1 << tableLog) - distributed;
+
+    if (ToDistribute == 0)
+        return 0;
+
+    if ((total / ToDistribute) > lowOne) {
+        /* risk of rounding to zero */
+        lowOne = (U32)((total * 3) / (ToDistribute * 2));
+        for (s=0; s<=maxSymbolValue; s++) {
+            if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
+                norm[s] = 1;
+                distributed++;
+                total -= count[s];
+                continue;
+        }   }
+        ToDistribute = (1 << tableLog) - distributed;
+    }
+
+    if (distributed == maxSymbolValue+1) {
+        /* all values are pretty poor;
+           probably incompressible data (should have already been detected);
+           find max, then give all remaining points to max */
+        U32 maxV = 0, maxC = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > maxC) { maxV=s; maxC=count[s]; }
+        norm[maxV] += (short)ToDistribute;
+        return 0;
+    }
+
+    if (total == 0) {
+        /* all of the symbols were low enough for the lowOne or lowThreshold */
+        for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
+            if (norm[s] > 0) { ToDistribute--; norm[s]++; }
+        return 0;
+    }
+
+    {   U64 const vStepLog = 62 - tableLog;
+        U64 const mid = (1ULL << (vStepLog-1)) - 1;
+        U64 const rStep = ZSTD_div64((((U64)1<<vStepLog) * ToDistribute) + mid, (U32)total);   /* scale on remaining */
+        U64 tmpTotal = mid;
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (norm[s]==NOT_YET_ASSIGNED) {
+                U64 const end = tmpTotal + (count[s] * rStep);
+                U32 const sStart = (U32)(tmpTotal >> vStepLog);
+                U32 const sEnd = (U32)(end >> vStepLog);
+                U32 const weight = sEnd - sStart;
+                if (weight < 1)
+                    return ERROR(GENERIC);
+                norm[s] = (short)weight;
+                tmpTotal = end;
+    }   }   }
+
+    return 0;
+}
+
+size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
+                           const unsigned* count, size_t total,
+                           unsigned maxSymbolValue, unsigned useLowProbCount)
+{
+    /* Sanity checks */
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported size */
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
+    if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
+
+    {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
+        short const lowProbCount = useLowProbCount ? -1 : 1;
+        U64 const scale = 62 - tableLog;
+        U64 const step = ZSTD_div64((U64)1<<62, (U32)total);   /* <== here, one division ! */
+        U64 const vStep = 1ULL<<(scale-20);
+        int stillToDistribute = 1<<tableLog;
+        unsigned s;
+        unsigned largest=0;
+        short largestP=0;
+        U32 lowThreshold = (U32)(total >> tableLog);
+
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (count[s] == total) return 0;   /* rle special case */
+            if (count[s] == 0) { normalizedCounter[s]=0; continue; }
+            if (count[s] <= lowThreshold) {
+                normalizedCounter[s] = lowProbCount;
+                stillToDistribute--;
+            } else {
+                short proba = (short)((count[s]*step) >> scale);
+                if (proba<8) {
+                    U64 restToBeat = vStep * rtbTable[proba];
+                    proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
+                }
+                if (proba > largestP) { largestP=proba; largest=s; }
+                normalizedCounter[s] = proba;
+                stillToDistribute -= proba;
+        }   }
+        if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
+            /* corner case, need another normalization method */
+            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
+            if (FSE_isError(errorCode)) return errorCode;
+        }
+        else normalizedCounter[largest] += (short)stillToDistribute;
+    }
+
+#if 0
+    {   /* Print Table (debug) */
+        U32 s;
+        U32 nTotal = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
+        for (s=0; s<=maxSymbolValue; s++)
+            nTotal += abs(normalizedCounter[s]);
+        if (nTotal != (1U<<tableLog))
+            RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
+        getchar();
+    }
+#endif
+
+    return tableLog;
+}
+
+/* fake FSE_CTable, for rle input (always same symbol) */
+size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
+{
+    void* ptr = ct;
+    U16* tableU16 = ( (U16*) ptr) + 2;
+    void* FSCTptr = (U32*)ptr + 2;
+    FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
+
+    /* header */
+    tableU16[-2] = (U16) 0;
+    tableU16[-1] = (U16) symbolValue;
+
+    /* Build table */
+    tableU16[0] = 0;
+    tableU16[1] = 0;   /* just in case */
+
+    /* Build Symbol Transformation Table */
+    symbolTT[symbolValue].deltaNbBits = 0;
+    symbolTT[symbolValue].deltaFindState = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct, const unsigned fast)
+{
+    const BYTE* const istart = (const BYTE*) src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip=iend;
+
+    BIT_CStream_t bitC;
+    FSE_CState_t CState1, CState2;
+
+    /* init */
+    if (srcSize <= 2) return 0;
+    { size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
+      if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
+
+#define FSE_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+
+    if (srcSize & 1) {
+        FSE_initCState2(&CState1, ct, *--ip);
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    } else {
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_initCState2(&CState1, ct, *--ip);
+    }
+
+    /* join to mod 4 */
+    srcSize -= 2;
+    if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) {  /* test bit 2 */
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    /* 2 or 4 encoding per loop */
+    while ( ip>istart ) {
+
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 )   /* this test must be static */
+            FSE_FLUSHBITS(&bitC);
+
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) {  /* this test must be static */
+            FSE_encodeSymbol(&bitC, &CState2, *--ip);
+            FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        }
+
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    FSE_flushCState(&bitC, &CState2);
+    FSE_flushCState(&bitC, &CState1);
+    return BIT_closeCStream(&bitC);
+}
+
+size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct)
+{
+    unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
+
+    if (fast)
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
+    else
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
+}
+
+
+size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
+
+#endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/hist.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/hist.c
new file mode 100644
index 0000000..e2fb431
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/hist.c
@@ -0,0 +1,181 @@
+/* ******************************************************************
+ * hist : Histogram functions
+ * part of Finite State Entropy project
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* --- dependencies --- */
+#include "../common/mem.h"             /* U32, BYTE, etc. */
+#include "../common/debug.h"           /* assert, DEBUGLOG */
+#include "../common/error_private.h"   /* ERROR */
+#include "hist.h"
+
+
+/* --- Error management --- */
+unsigned HIST_isError(size_t code) { return ERR_isError(code); }
+
+/*-**************************************************************
+ *  Histogram functions
+ ****************************************************************/
+unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const end = ip + srcSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned largestCount=0;
+
+    ZSTD_memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
+    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
+
+    while (ip<end) {
+        assert(*ip <= maxSymbolValue);
+        count[*ip++]++;
+    }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+
+    {   U32 s;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > largestCount) largestCount = count[s];
+    }
+
+    return largestCount;
+}
+
+typedef enum { trustInput, checkMaxSymbolValue } HIST_checkInput_e;
+
+/* HIST_count_parallel_wksp() :
+ * store histogram into 4 intermediate tables, recombined at the end.
+ * this design makes better use of OoO cpus,
+ * and is noticeably faster when some values are heavily repeated.
+ * But it needs some additional workspace for intermediate tables.
+ * `workSpace` must be a U32 table of size >= HIST_WKSP_SIZE_U32.
+ * @return : largest histogram frequency,
+ *           or an error code (notably when histogram's alphabet is larger than *maxSymbolValuePtr) */
+static size_t HIST_count_parallel_wksp(
+                                unsigned* count, unsigned* maxSymbolValuePtr,
+                                const void* source, size_t sourceSize,
+                                HIST_checkInput_e check,
+                                U32* const workSpace)
+{
+    const BYTE* ip = (const BYTE*)source;
+    const BYTE* const iend = ip+sourceSize;
+    size_t const countSize = (*maxSymbolValuePtr + 1) * sizeof(*count);
+    unsigned max=0;
+    U32* const Counting1 = workSpace;
+    U32* const Counting2 = Counting1 + 256;
+    U32* const Counting3 = Counting2 + 256;
+    U32* const Counting4 = Counting3 + 256;
+
+    /* safety checks */
+    assert(*maxSymbolValuePtr <= 255);
+    if (!sourceSize) {
+        ZSTD_memset(count, 0, countSize);
+        *maxSymbolValuePtr = 0;
+        return 0;
+    }
+    ZSTD_memset(workSpace, 0, 4*256*sizeof(unsigned));
+
+    /* by stripes of 16 bytes */
+    {   U32 cached = MEM_read32(ip); ip += 4;
+        while (ip < iend-15) {
+            U32 c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+        }
+        ip-=4;
+    }
+
+    /* finish last symbols */
+    while (ip<iend) Counting1[*ip++]++;
+
+    {   U32 s;
+        for (s=0; s<256; s++) {
+            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
+            if (Counting1[s] > max) max = Counting1[s];
+    }   }
+
+    {   unsigned maxSymbolValue = 255;
+        while (!Counting1[maxSymbolValue]) maxSymbolValue--;
+        if (check && maxSymbolValue > *maxSymbolValuePtr) return ERROR(maxSymbolValue_tooSmall);
+        *maxSymbolValuePtr = maxSymbolValue;
+        ZSTD_memmove(count, Counting1, countSize);   /* in case count & Counting1 are overlapping */
+    }
+    return (size_t)max;
+}
+
+/* HIST_countFast_wksp() :
+ * Same as HIST_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` is a writable buffer which must be 4-bytes aligned,
+ * `workSpaceSize` must be >= HIST_WKSP_SIZE
+ */
+size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                          const void* source, size_t sourceSize,
+                          void* workSpace, size_t workSpaceSize)
+{
+    if (sourceSize < 1500) /* heuristic threshold */
+        return HIST_count_simple(count, maxSymbolValuePtr, source, sourceSize);
+    if ((size_t)workSpace & 3) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
+    return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, trustInput, (U32*)workSpace);
+}
+
+/* HIST_count_wksp() :
+ * Same as HIST_count(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
+size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                       const void* source, size_t sourceSize,
+                       void* workSpace, size_t workSpaceSize)
+{
+    if ((size_t)workSpace & 3) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
+    if (*maxSymbolValuePtr < 255)
+        return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, checkMaxSymbolValue, (U32*)workSpace);
+    *maxSymbolValuePtr = 255;
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace, workSpaceSize);
+}
+
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
+/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
+}
+
+size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* src, size_t srcSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters, sizeof(tmpCounters));
+}
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/hist.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/hist.h
new file mode 100644
index 0000000..887896b
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/hist.h
@@ -0,0 +1,75 @@
+/* ******************************************************************
+ * hist : Histogram functions
+ * part of Finite State Entropy project
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* --- dependencies --- */
+#include "../common/zstd_deps.h"   /* size_t */
+
+
+/* --- simple histogram functions --- */
+
+/*! HIST_count():
+ *  Provides the precise count of each byte within a table 'count'.
+ * 'count' is a table of unsigned int, of minimum size (*maxSymbolValuePtr+1).
+ *  Updates *maxSymbolValuePtr with actual largest symbol value detected.
+ * @return : count of the most frequent symbol (which isn't identified).
+ *           or an error code, which can be tested using HIST_isError().
+ *           note : if return == srcSize, there is only one symbol.
+ */
+size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                  const void* src, size_t srcSize);
+
+unsigned HIST_isError(size_t code);  /**< tells if a return value is an error code */
+
+
+/* --- advanced histogram functions --- */
+
+#define HIST_WKSP_SIZE_U32 1024
+#define HIST_WKSP_SIZE    (HIST_WKSP_SIZE_U32 * sizeof(unsigned))
+/** HIST_count_wksp() :
+ *  Same as HIST_count(), but using an externally provided scratch buffer.
+ *  Benefit is this function will use very little stack space.
+ * `workSpace` is a writable buffer which must be 4-bytes aligned,
+ * `workSpaceSize` must be >= HIST_WKSP_SIZE
+ */
+size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                       const void* src, size_t srcSize,
+                       void* workSpace, size_t workSpaceSize);
+
+/** HIST_countFast() :
+ *  same as HIST_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr.
+ *  This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr`
+ */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                      const void* src, size_t srcSize);
+
+/** HIST_countFast_wksp() :
+ *  Same as HIST_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` is a writable buffer which must be 4-bytes aligned,
+ * `workSpaceSize` must be >= HIST_WKSP_SIZE
+ */
+size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize,
+                           void* workSpace, size_t workSpaceSize);
+
+/*! HIST_count_simple() :
+ *  Same as HIST_countFast(), this function is unsafe,
+ *  and will segfault if any value within `src` is `> *maxSymbolValuePtr`.
+ *  It is also a bit slower for large inputs.
+ *  However, it does not need any additional memory (not even on stack).
+ * @return : count of the most frequent symbol.
+ *  Note this function doesn't produce any error (i.e. it must succeed).
+ */
+unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize);
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/huf_compress.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/huf_compress.c
new file mode 100644
index 0000000..2987187
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/huf_compress.c
@@ -0,0 +1,1435 @@
+/* ******************************************************************
+ * Huffman encoder, part of New Generation Entropy library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include "../common/zstd_deps.h"     /* ZSTD_memcpy, ZSTD_memset */
+#include "../common/compiler.h"
+#include "../common/bitstream.h"
+#include "hist.h"
+#define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
+#include "../common/fse.h"        /* header compression */
+#include "../common/huf.h"
+#include "../common/error_private.h"
+#include "../common/bits.h"       /* ZSTD_highbit32 */
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_isError ERR_isError
+#define HUF_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Required declarations
+****************************************************************/
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+
+/* **************************************************************
+*  Debug Traces
+****************************************************************/
+
+#if DEBUGLEVEL >= 2
+
+static size_t showU32(const U32* arr, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", arr[u]); (void)arr;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+static size_t HUF_getNbBits(HUF_CElt elt);
+
+static size_t showCTableBits(const HUF_CElt* ctable, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %zu", HUF_getNbBits(ctable[u])); (void)ctable;
+    }
+    RAWLOG(6, " \n");
+    return size;
+
+}
+
+static size_t showHNodeSymbols(const nodeElt* hnode, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", hnode[u].byte); (void)hnode;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+static size_t showHNodeBits(const nodeElt* hnode, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", hnode[u].nbBits); (void)hnode;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+#endif
+
+
+/* *******************************************************
+*  HUF : Huffman block compression
+*********************************************************/
+#define HUF_WORKSPACE_MAX_ALIGNMENT 8
+
+static void* HUF_alignUpWorkspace(void* workspace, size_t* workspaceSizePtr, size_t align)
+{
+    size_t const mask = align - 1;
+    size_t const rem = (size_t)workspace & mask;
+    size_t const add = (align - rem) & mask;
+    BYTE* const aligned = (BYTE*)workspace + add;
+    assert((align & (align - 1)) == 0); /* pow 2 */
+    assert(align <= HUF_WORKSPACE_MAX_ALIGNMENT);
+    if (*workspaceSizePtr >= add) {
+        assert(add < align);
+        assert(((size_t)aligned & mask) == 0);
+        *workspaceSizePtr -= add;
+        return aligned;
+    } else {
+        *workspaceSizePtr = 0;
+        return NULL;
+    }
+}
+
+
+/* HUF_compressWeights() :
+ * Same as FSE_compress(), but dedicated to huff0's weights compression.
+ * The use case needs much less stack memory.
+ * Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
+ */
+#define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
+
+typedef struct {
+    FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
+    U32 scratchBuffer[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(HUF_TABLELOG_MAX, MAX_FSE_TABLELOG_FOR_HUFF_HEADER)];
+    unsigned count[HUF_TABLELOG_MAX+1];
+    S16 norm[HUF_TABLELOG_MAX+1];
+} HUF_CompressWeightsWksp;
+
+static size_t
+HUF_compressWeights(void* dst, size_t dstSize,
+              const void* weightTable, size_t wtSize,
+                    void* workspace, size_t workspaceSize)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstSize;
+
+    unsigned maxSymbolValue = HUF_TABLELOG_MAX;
+    U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
+    HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
+
+    if (workspaceSize < sizeof(HUF_CompressWeightsWksp)) return ERROR(GENERIC);
+
+    /* init conditions */
+    if (wtSize <= 1) return 0;  /* Not compressible */
+
+    /* Scan input and build symbol stats */
+    {   unsigned const maxCount = HIST_count_simple(wksp->count, &maxSymbolValue, weightTable, wtSize);   /* never fails */
+        if (maxCount == wtSize) return 1;   /* only a single symbol in src : rle */
+        if (maxCount == 1) return 0;        /* each symbol present maximum once => not compressible */
+    }
+
+    tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
+    CHECK_F( FSE_normalizeCount(wksp->norm, tableLog, wksp->count, wtSize, maxSymbolValue, /* useLowProbCount */ 0) );
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, FSE_writeNCount(op, (size_t)(oend-op), wksp->norm, maxSymbolValue, tableLog) );
+        op += hSize;
+    }
+
+    /* Compress */
+    CHECK_F( FSE_buildCTable_wksp(wksp->CTable, wksp->norm, maxSymbolValue, tableLog, wksp->scratchBuffer, sizeof(wksp->scratchBuffer)) );
+    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, (size_t)(oend - op), weightTable, wtSize, wksp->CTable) );
+        if (cSize == 0) return 0;   /* not enough space for compressed data */
+        op += cSize;
+    }
+
+    return (size_t)(op-ostart);
+}
+
+static size_t HUF_getNbBits(HUF_CElt elt)
+{
+    return elt & 0xFF;
+}
+
+static size_t HUF_getNbBitsFast(HUF_CElt elt)
+{
+    return elt;
+}
+
+static size_t HUF_getValue(HUF_CElt elt)
+{
+    return elt & ~(size_t)0xFF;
+}
+
+static size_t HUF_getValueFast(HUF_CElt elt)
+{
+    return elt;
+}
+
+static void HUF_setNbBits(HUF_CElt* elt, size_t nbBits)
+{
+    assert(nbBits <= HUF_TABLELOG_ABSOLUTEMAX);
+    *elt = nbBits;
+}
+
+static void HUF_setValue(HUF_CElt* elt, size_t value)
+{
+    size_t const nbBits = HUF_getNbBits(*elt);
+    if (nbBits > 0) {
+        assert((value >> nbBits) == 0);
+        *elt |= value << (sizeof(HUF_CElt) * 8 - nbBits);
+    }
+}
+
+typedef struct {
+    HUF_CompressWeightsWksp wksp;
+    BYTE bitsToWeight[HUF_TABLELOG_MAX + 1];   /* precomputed conversion table */
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX];
+} HUF_WriteCTableWksp;
+
+size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
+                            const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog,
+                            void* workspace, size_t workspaceSize)
+{
+    HUF_CElt const* const ct = CTable + 1;
+    BYTE* op = (BYTE*)dst;
+    U32 n;
+    HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
+
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE >= sizeof(HUF_WriteCTableWksp));
+
+    /* check conditions */
+    if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+
+    /* convert to weight */
+    wksp->bitsToWeight[0] = 0;
+    for (n=1; n<huffLog+1; n++)
+        wksp->bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
+    for (n=0; n<maxSymbolValue; n++)
+        wksp->huffWeight[n] = wksp->bitsToWeight[HUF_getNbBits(ct[n])];
+
+    /* attempt weights compression by FSE */
+    if (maxDstSize < 1) return ERROR(dstSize_tooSmall);
+    {   CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, wksp->huffWeight, maxSymbolValue, &wksp->wksp, sizeof(wksp->wksp)) );
+        if ((hSize>1) & (hSize < maxSymbolValue/2)) {   /* FSE compressed */
+            op[0] = (BYTE)hSize;
+            return hSize+1;
+    }   }
+
+    /* write raw values as 4-bits (max : 15) */
+    if (maxSymbolValue > (256-128)) return ERROR(GENERIC);   /* should not happen : likely means source cannot be compressed */
+    if (((maxSymbolValue+1)/2) + 1 > maxDstSize) return ERROR(dstSize_tooSmall);   /* not enough space within dst buffer */
+    op[0] = (BYTE)(128 /*special case*/ + (maxSymbolValue-1));
+    wksp->huffWeight[maxSymbolValue] = 0;   /* to be sure it doesn't cause msan issue in final combination */
+    for (n=0; n<maxSymbolValue; n+=2)
+        op[(n/2)+1] = (BYTE)((wksp->huffWeight[n] << 4) + wksp->huffWeight[n+1]);
+    return ((maxSymbolValue+1)/2) + 1;
+}
+
+
+size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
+{
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];   /* init not required, even though some static analyzer may complain */
+    U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    HUF_CElt* const ct = CTable + 1;
+
+    /* get symbol weights */
+    CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
+    *hasZeroWeights = (rankVal[0] > 0);
+
+    /* check result */
+    if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
+
+    CTable[0] = tableLog;
+
+    /* Prepare base value per rank */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<=tableLog; n++) {
+            U32 curr = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = curr;
+    }   }
+
+    /* fill nbBits */
+    {   U32 n; for (n=0; n<nbSymbols; n++) {
+            const U32 w = huffWeight[n];
+            HUF_setNbBits(ct + n, (BYTE)(tableLog + 1 - w) & -(w != 0));
+    }   }
+
+    /* fill val */
+    {   U16 nbPerRank[HUF_TABLELOG_MAX+2]  = {0};  /* support w=0=>n=tableLog+1 */
+        U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
+        { U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[HUF_getNbBits(ct[n])]++; }
+        /* determine stating value per rank */
+        valPerRank[tableLog+1] = 0;   /* for w==0 */
+        {   U16 min = 0;
+            U32 n; for (n=tableLog; n>0; n--) {  /* start at n=tablelog <-> w=1 */
+                valPerRank[n] = min;     /* get starting value within each rank */
+                min += nbPerRank[n];
+                min >>= 1;
+        }   }
+        /* assign value within rank, symbol order */
+        { U32 n; for (n=0; n<nbSymbols; n++) HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); }
+    }
+
+    *maxSymbolValuePtr = nbSymbols - 1;
+    return readSize;
+}
+
+U32 HUF_getNbBitsFromCTable(HUF_CElt const* CTable, U32 symbolValue)
+{
+    const HUF_CElt* const ct = CTable + 1;
+    assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
+    return (U32)HUF_getNbBits(ct[symbolValue]);
+}
+
+
+/**
+ * HUF_setMaxHeight():
+ * Try to enforce @targetNbBits on the Huffman tree described in @huffNode.
+ *
+ * It attempts to convert all nodes with nbBits > @targetNbBits
+ * to employ @targetNbBits instead. Then it adjusts the tree
+ * so that it remains a valid canonical Huffman tree.
+ *
+ * @pre               The sum of the ranks of each symbol == 2^largestBits,
+ *                    where largestBits == huffNode[lastNonNull].nbBits.
+ * @post              The sum of the ranks of each symbol == 2^largestBits,
+ *                    where largestBits is the return value (expected <= targetNbBits).
+ *
+ * @param huffNode    The Huffman tree modified in place to enforce targetNbBits.
+ *                    It's presumed sorted, from most frequent to rarest symbol.
+ * @param lastNonNull The symbol with the lowest count in the Huffman tree.
+ * @param targetNbBits  The allowed number of bits, which the Huffman tree
+ *                    may not respect. After this function the Huffman tree will
+ *                    respect targetNbBits.
+ * @return            The maximum number of bits of the Huffman tree after adjustment.
+ */
+static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 targetNbBits)
+{
+    const U32 largestBits = huffNode[lastNonNull].nbBits;
+    /* early exit : no elt > targetNbBits, so the tree is already valid. */
+    if (largestBits <= targetNbBits) return largestBits;
+
+    DEBUGLOG(5, "HUF_setMaxHeight (targetNbBits = %u)", targetNbBits);
+
+    /* there are several too large elements (at least >= 2) */
+    {   int totalCost = 0;
+        const U32 baseCost = 1 << (largestBits - targetNbBits);
+        int n = (int)lastNonNull;
+
+        /* Adjust any ranks > targetNbBits to targetNbBits.
+         * Compute totalCost, which is how far the sum of the ranks is
+         * we are over 2^largestBits after adjust the offending ranks.
+         */
+        while (huffNode[n].nbBits > targetNbBits) {
+            totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
+            huffNode[n].nbBits = (BYTE)targetNbBits;
+            n--;
+        }
+        /* n stops at huffNode[n].nbBits <= targetNbBits */
+        assert(huffNode[n].nbBits <= targetNbBits);
+        /* n end at index of smallest symbol using < targetNbBits */
+        while (huffNode[n].nbBits == targetNbBits) --n;
+
+        /* renorm totalCost from 2^largestBits to 2^targetNbBits
+         * note : totalCost is necessarily a multiple of baseCost */
+        assert(((U32)totalCost & (baseCost - 1)) == 0);
+        totalCost >>= (largestBits - targetNbBits);
+        assert(totalCost > 0);
+
+        /* repay normalized cost */
+        {   U32 const noSymbol = 0xF0F0F0F0;
+            U32 rankLast[HUF_TABLELOG_MAX+2];
+
+            /* Get pos of last (smallest = lowest cum. count) symbol per rank */
+            ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
+            {   U32 currentNbBits = targetNbBits;
+                int pos;
+                for (pos=n ; pos >= 0; pos--) {
+                    if (huffNode[pos].nbBits >= currentNbBits) continue;
+                    currentNbBits = huffNode[pos].nbBits;   /* < targetNbBits */
+                    rankLast[targetNbBits-currentNbBits] = (U32)pos;
+            }   }
+
+            while (totalCost > 0) {
+                /* Try to reduce the next power of 2 above totalCost because we
+                 * gain back half the rank.
+                 */
+                U32 nBitsToDecrease = ZSTD_highbit32((U32)totalCost) + 1;
+                for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
+                    U32 const highPos = rankLast[nBitsToDecrease];
+                    U32 const lowPos = rankLast[nBitsToDecrease-1];
+                    if (highPos == noSymbol) continue;
+                    /* Decrease highPos if no symbols of lowPos or if it is
+                     * not cheaper to remove 2 lowPos than highPos.
+                     */
+                    if (lowPos == noSymbol) break;
+                    {   U32 const highTotal = huffNode[highPos].count;
+                        U32 const lowTotal = 2 * huffNode[lowPos].count;
+                        if (highTotal <= lowTotal) break;
+                }   }
+                /* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
+                assert(rankLast[nBitsToDecrease] != noSymbol || nBitsToDecrease == 1);
+                /* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
+                while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
+                    nBitsToDecrease++;
+                assert(rankLast[nBitsToDecrease] != noSymbol);
+                /* Increase the number of bits to gain back half the rank cost. */
+                totalCost -= 1 << (nBitsToDecrease-1);
+                huffNode[rankLast[nBitsToDecrease]].nbBits++;
+
+                /* Fix up the new rank.
+                 * If the new rank was empty, this symbol is now its smallest.
+                 * Otherwise, this symbol will be the largest in the new rank so no adjustment.
+                 */
+                if (rankLast[nBitsToDecrease-1] == noSymbol)
+                    rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];
+                /* Fix up the old rank.
+                 * If the symbol was at position 0, meaning it was the highest weight symbol in the tree,
+                 * it must be the only symbol in its rank, so the old rank now has no symbols.
+                 * Otherwise, since the Huffman nodes are sorted by count, the previous position is now
+                 * the smallest node in the rank. If the previous position belongs to a different rank,
+                 * then the rank is now empty.
+                 */
+                if (rankLast[nBitsToDecrease] == 0)    /* special case, reached largest symbol */
+                    rankLast[nBitsToDecrease] = noSymbol;
+                else {
+                    rankLast[nBitsToDecrease]--;
+                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != targetNbBits-nBitsToDecrease)
+                        rankLast[nBitsToDecrease] = noSymbol;   /* this rank is now empty */
+                }
+            }   /* while (totalCost > 0) */
+
+            /* If we've removed too much weight, then we have to add it back.
+             * To avoid overshooting again, we only adjust the smallest rank.
+             * We take the largest nodes from the lowest rank 0 and move them
+             * to rank 1. There's guaranteed to be enough rank 0 symbols because
+             * TODO.
+             */
+            while (totalCost < 0) {  /* Sometimes, cost correction overshoot */
+                /* special case : no rank 1 symbol (using targetNbBits-1);
+                 * let's create one from largest rank 0 (using targetNbBits).
+                 */
+                if (rankLast[1] == noSymbol) {
+                    while (huffNode[n].nbBits == targetNbBits) n--;
+                    huffNode[n+1].nbBits--;
+                    assert(n >= 0);
+                    rankLast[1] = (U32)(n+1);
+                    totalCost++;
+                    continue;
+                }
+                huffNode[ rankLast[1] + 1 ].nbBits--;
+                rankLast[1]++;
+                totalCost ++;
+            }
+        }   /* repay normalized cost */
+    }   /* there are several too large elements (at least >= 2) */
+
+    return targetNbBits;
+}
+
+typedef struct {
+    U16 base;
+    U16 curr;
+} rankPos;
+
+typedef nodeElt huffNodeTable[2 * (HUF_SYMBOLVALUE_MAX + 1)];
+
+/* Number of buckets available for HUF_sort() */
+#define RANK_POSITION_TABLE_SIZE 192
+
+typedef struct {
+  huffNodeTable huffNodeTbl;
+  rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
+} HUF_buildCTable_wksp_tables;
+
+/* RANK_POSITION_DISTINCT_COUNT_CUTOFF == Cutoff point in HUF_sort() buckets for which we use log2 bucketing.
+ * Strategy is to use as many buckets as possible for representing distinct
+ * counts while using the remainder to represent all "large" counts.
+ *
+ * To satisfy this requirement for 192 buckets, we can do the following:
+ * Let buckets 0-166 represent distinct counts of [0, 166]
+ * Let buckets 166 to 192 represent all remaining counts up to RANK_POSITION_MAX_COUNT_LOG using log2 bucketing.
+ */
+#define RANK_POSITION_MAX_COUNT_LOG 32
+#define RANK_POSITION_LOG_BUCKETS_BEGIN ((RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */)
+#define RANK_POSITION_DISTINCT_COUNT_CUTOFF (RANK_POSITION_LOG_BUCKETS_BEGIN + ZSTD_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */)
+
+/* Return the appropriate bucket index for a given count. See definition of
+ * RANK_POSITION_DISTINCT_COUNT_CUTOFF for explanation of bucketing strategy.
+ */
+static U32 HUF_getIndex(U32 const count) {
+    return (count < RANK_POSITION_DISTINCT_COUNT_CUTOFF)
+        ? count
+        : ZSTD_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
+}
+
+/* Helper swap function for HUF_quickSortPartition() */
+static void HUF_swapNodes(nodeElt* a, nodeElt* b) {
+	nodeElt tmp = *a;
+	*a = *b;
+	*b = tmp;
+}
+
+/* Returns 0 if the huffNode array is not sorted by descending count */
+MEM_STATIC int HUF_isSorted(nodeElt huffNode[], U32 const maxSymbolValue1) {
+    U32 i;
+    for (i = 1; i < maxSymbolValue1; ++i) {
+        if (huffNode[i].count > huffNode[i-1].count) {
+            return 0;
+        }
+    }
+    return 1;
+}
+
+/* Insertion sort by descending order */
+HINT_INLINE void HUF_insertionSort(nodeElt huffNode[], int const low, int const high) {
+    int i;
+    int const size = high-low+1;
+    huffNode += low;
+    for (i = 1; i < size; ++i) {
+        nodeElt const key = huffNode[i];
+        int j = i - 1;
+        while (j >= 0 && huffNode[j].count < key.count) {
+            huffNode[j + 1] = huffNode[j];
+            j--;
+        }
+        huffNode[j + 1] = key;
+    }
+}
+
+/* Pivot helper function for quicksort. */
+static int HUF_quickSortPartition(nodeElt arr[], int const low, int const high) {
+    /* Simply select rightmost element as pivot. "Better" selectors like
+     * median-of-three don't experimentally appear to have any benefit.
+     */
+    U32 const pivot = arr[high].count;
+    int i = low - 1;
+    int j = low;
+    for ( ; j < high; j++) {
+        if (arr[j].count > pivot) {
+            i++;
+            HUF_swapNodes(&arr[i], &arr[j]);
+        }
+    }
+    HUF_swapNodes(&arr[i + 1], &arr[high]);
+    return i + 1;
+}
+
+/* Classic quicksort by descending with partially iterative calls
+ * to reduce worst case callstack size.
+ */
+static void HUF_simpleQuickSort(nodeElt arr[], int low, int high) {
+    int const kInsertionSortThreshold = 8;
+    if (high - low < kInsertionSortThreshold) {
+        HUF_insertionSort(arr, low, high);
+        return;
+    }
+    while (low < high) {
+        int const idx = HUF_quickSortPartition(arr, low, high);
+        if (idx - low < high - idx) {
+            HUF_simpleQuickSort(arr, low, idx - 1);
+            low = idx + 1;
+        } else {
+            HUF_simpleQuickSort(arr, idx + 1, high);
+            high = idx - 1;
+        }
+    }
+}
+
+/**
+ * HUF_sort():
+ * Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
+ * This is a typical bucket sorting strategy that uses either quicksort or insertion sort to sort each bucket.
+ *
+ * @param[out] huffNode       Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
+ *                            Must have (maxSymbolValue + 1) entries.
+ * @param[in]  count          Histogram of the symbols.
+ * @param[in]  maxSymbolValue Maximum symbol value.
+ * @param      rankPosition   This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
+ */
+static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSymbolValue, rankPos rankPosition[]) {
+    U32 n;
+    U32 const maxSymbolValue1 = maxSymbolValue+1;
+
+    /* Compute base and set curr to base.
+     * For symbol s let lowerRank = HUF_getIndex(count[n]) and rank = lowerRank + 1.
+     * See HUF_getIndex to see bucketing strategy.
+     * We attribute each symbol to lowerRank's base value, because we want to know where
+     * each rank begins in the output, so for rank R we want to count ranks R+1 and above.
+     */
+    ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
+    for (n = 0; n < maxSymbolValue1; ++n) {
+        U32 lowerRank = HUF_getIndex(count[n]);
+        assert(lowerRank < RANK_POSITION_TABLE_SIZE - 1);
+        rankPosition[lowerRank].base++;
+    }
+
+    assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
+    /* Set up the rankPosition table */
+    for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
+        rankPosition[n-1].base += rankPosition[n].base;
+        rankPosition[n-1].curr = rankPosition[n-1].base;
+    }
+
+    /* Insert each symbol into their appropriate bucket, setting up rankPosition table. */
+    for (n = 0; n < maxSymbolValue1; ++n) {
+        U32 const c = count[n];
+        U32 const r = HUF_getIndex(c) + 1;
+        U32 const pos = rankPosition[r].curr++;
+        assert(pos < maxSymbolValue1);
+        huffNode[pos].count = c;
+        huffNode[pos].byte  = (BYTE)n;
+    }
+
+    /* Sort each bucket. */
+    for (n = RANK_POSITION_DISTINCT_COUNT_CUTOFF; n < RANK_POSITION_TABLE_SIZE - 1; ++n) {
+        int const bucketSize = rankPosition[n].curr - rankPosition[n].base;
+        U32 const bucketStartIdx = rankPosition[n].base;
+        if (bucketSize > 1) {
+            assert(bucketStartIdx < maxSymbolValue1);
+            HUF_simpleQuickSort(huffNode + bucketStartIdx, 0, bucketSize-1);
+        }
+    }
+
+    assert(HUF_isSorted(huffNode, maxSymbolValue1));
+}
+
+
+/** HUF_buildCTable_wksp() :
+ *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
+ *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as sizeof(HUF_buildCTable_wksp_tables).
+ */
+#define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
+
+/* HUF_buildTree():
+ * Takes the huffNode array sorted by HUF_sort() and builds an unlimited-depth Huffman tree.
+ *
+ * @param huffNode        The array sorted by HUF_sort(). Builds the Huffman tree in this array.
+ * @param maxSymbolValue  The maximum symbol value.
+ * @return                The smallest node in the Huffman tree (by count).
+ */
+static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
+{
+    nodeElt* const huffNode0 = huffNode - 1;
+    int nonNullRank;
+    int lowS, lowN;
+    int nodeNb = STARTNODE;
+    int n, nodeRoot;
+    DEBUGLOG(5, "HUF_buildTree (alphabet size = %u)", maxSymbolValue + 1);
+    /* init for parents */
+    nonNullRank = (int)maxSymbolValue;
+    while(huffNode[nonNullRank].count == 0) nonNullRank--;
+    lowS = nonNullRank; nodeRoot = nodeNb + lowS - 1; lowN = nodeNb;
+    huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count;
+    huffNode[lowS].parent = huffNode[lowS-1].parent = (U16)nodeNb;
+    nodeNb++; lowS-=2;
+    for (n=nodeNb; n<=nodeRoot; n++) huffNode[n].count = (U32)(1U<<30);
+    huffNode0[0].count = (U32)(1U<<31);  /* fake entry, strong barrier */
+
+    /* create parents */
+    while (nodeNb <= nodeRoot) {
+        int const n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        int const n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        huffNode[nodeNb].count = huffNode[n1].count + huffNode[n2].count;
+        huffNode[n1].parent = huffNode[n2].parent = (U16)nodeNb;
+        nodeNb++;
+    }
+
+    /* distribute weights (unlimited tree height) */
+    huffNode[nodeRoot].nbBits = 0;
+    for (n=nodeRoot-1; n>=STARTNODE; n--)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+    for (n=0; n<=nonNullRank; n++)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+
+    DEBUGLOG(6, "Initial distribution of bits completed (%zu sorted symbols)", showHNodeBits(huffNode, maxSymbolValue+1));
+
+    return nonNullRank;
+}
+
+/**
+ * HUF_buildCTableFromTree():
+ * Build the CTable given the Huffman tree in huffNode.
+ *
+ * @param[out] CTable         The output Huffman CTable.
+ * @param      huffNode       The Huffman tree.
+ * @param      nonNullRank    The last and smallest node in the Huffman tree.
+ * @param      maxSymbolValue The maximum symbol value.
+ * @param      maxNbBits      The exact maximum number of bits used in the Huffman tree.
+ */
+static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
+{
+    HUF_CElt* const ct = CTable + 1;
+    /* fill result into ctable (val, nbBits) */
+    int n;
+    U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
+    U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
+    int const alphabetSize = (int)(maxSymbolValue + 1);
+    for (n=0; n<=nonNullRank; n++)
+        nbPerRank[huffNode[n].nbBits]++;
+    /* determine starting value per rank */
+    {   U16 min = 0;
+        for (n=(int)maxNbBits; n>0; n--) {
+            valPerRank[n] = min;      /* get starting value within each rank */
+            min += nbPerRank[n];
+            min >>= 1;
+    }   }
+    for (n=0; n<alphabetSize; n++)
+        HUF_setNbBits(ct + huffNode[n].byte, huffNode[n].nbBits);   /* push nbBits per symbol, symbol order */
+    for (n=0; n<alphabetSize; n++)
+        HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++);   /* assign value within rank, symbol order */
+    CTable[0] = maxNbBits;
+}
+
+size_t
+HUF_buildCTable_wksp(HUF_CElt* CTable, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
+                     void* workSpace, size_t wkspSize)
+{
+    HUF_buildCTable_wksp_tables* const wksp_tables =
+        (HUF_buildCTable_wksp_tables*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(U32));
+    nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
+    nodeElt* const huffNode = huffNode0+1;
+    int nonNullRank;
+
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE == sizeof(HUF_buildCTable_wksp_tables));
+
+    DEBUGLOG(5, "HUF_buildCTable_wksp (alphabet size = %u)", maxSymbolValue+1);
+
+    /* safety checks */
+    if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
+        return ERROR(workSpace_tooSmall);
+    if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
+        return ERROR(maxSymbolValue_tooLarge);
+    ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
+
+    /* sort, decreasing order */
+    HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
+    DEBUGLOG(6, "sorted symbols completed (%zu symbols)", showHNodeSymbols(huffNode, maxSymbolValue+1));
+
+    /* build tree */
+    nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
+
+    /* determine and enforce maxTableLog */
+    maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
+    if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
+
+    HUF_buildCTableFromTree(CTable, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
+
+    return maxNbBits;
+}
+
+size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
+{
+    HUF_CElt const* ct = CTable + 1;
+    size_t nbBits = 0;
+    int s;
+    for (s = 0; s <= (int)maxSymbolValue; ++s) {
+        nbBits += HUF_getNbBits(ct[s]) * count[s];
+    }
+    return nbBits >> 3;
+}
+
+int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
+  HUF_CElt const* ct = CTable + 1;
+  int bad = 0;
+  int s;
+  for (s = 0; s <= (int)maxSymbolValue; ++s) {
+    bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
+  }
+  return !bad;
+}
+
+size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
+
+/** HUF_CStream_t:
+ * Huffman uses its own BIT_CStream_t implementation.
+ * There are three major differences from BIT_CStream_t:
+ *   1. HUF_addBits() takes a HUF_CElt (size_t) which is
+ *      the pair (nbBits, value) in the format:
+ *      format:
+ *        - Bits [0, 4)            = nbBits
+ *        - Bits [4, 64 - nbBits)  = 0
+ *        - Bits [64 - nbBits, 64) = value
+ *   2. The bitContainer is built from the upper bits and
+ *      right shifted. E.g. to add a new value of N bits
+ *      you right shift the bitContainer by N, then or in
+ *      the new value into the N upper bits.
+ *   3. The bitstream has two bit containers. You can add
+ *      bits to the second container and merge them into
+ *      the first container.
+ */
+
+#define HUF_BITS_IN_CONTAINER (sizeof(size_t) * 8)
+
+typedef struct {
+    size_t bitContainer[2];
+    size_t bitPos[2];
+
+    BYTE* startPtr;
+    BYTE* ptr;
+    BYTE* endPtr;
+} HUF_CStream_t;
+
+/**! HUF_initCStream():
+ * Initializes the bitstream.
+ * @returns 0 or an error code.
+ */
+static size_t HUF_initCStream(HUF_CStream_t* bitC,
+                                  void* startPtr, size_t dstCapacity)
+{
+    ZSTD_memset(bitC, 0, sizeof(*bitC));
+    bitC->startPtr = (BYTE*)startPtr;
+    bitC->ptr = bitC->startPtr;
+    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer[0]);
+    if (dstCapacity <= sizeof(bitC->bitContainer[0])) return ERROR(dstSize_tooSmall);
+    return 0;
+}
+
+/*! HUF_addBits():
+ * Adds the symbol stored in HUF_CElt elt to the bitstream.
+ *
+ * @param elt   The element we're adding. This is a (nbBits, value) pair.
+ *              See the HUF_CStream_t docs for the format.
+ * @param idx   Insert into the bitstream at this idx.
+ * @param kFast This is a template parameter. If the bitstream is guaranteed
+ *              to have at least 4 unused bits after this call it may be 1,
+ *              otherwise it must be 0. HUF_addBits() is faster when fast is set.
+ */
+FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int idx, int kFast)
+{
+    assert(idx <= 1);
+    assert(HUF_getNbBits(elt) <= HUF_TABLELOG_ABSOLUTEMAX);
+    /* This is efficient on x86-64 with BMI2 because shrx
+     * only reads the low 6 bits of the register. The compiler
+     * knows this and elides the mask. When fast is set,
+     * every operation can use the same value loaded from elt.
+     */
+    bitC->bitContainer[idx] >>= HUF_getNbBits(elt);
+    bitC->bitContainer[idx] |= kFast ? HUF_getValueFast(elt) : HUF_getValue(elt);
+    /* We only read the low 8 bits of bitC->bitPos[idx] so it
+     * doesn't matter that the high bits have noise from the value.
+     */
+    bitC->bitPos[idx] += HUF_getNbBitsFast(elt);
+    assert((bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+    /* The last 4-bits of elt are dirty if fast is set,
+     * so we must not be overwriting bits that have already been
+     * inserted into the bit container.
+     */
+#if DEBUGLEVEL >= 1
+    {
+        size_t const nbBits = HUF_getNbBits(elt);
+        size_t const dirtyBits = nbBits == 0 ? 0 : ZSTD_highbit32((U32)nbBits) + 1;
+        (void)dirtyBits;
+        /* Middle bits are 0. */
+        assert(((elt >> dirtyBits) << (dirtyBits + nbBits)) == 0);
+        /* We didn't overwrite any bits in the bit container. */
+        assert(!kFast || (bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+        (void)dirtyBits;
+    }
+#endif
+}
+
+FORCE_INLINE_TEMPLATE void HUF_zeroIndex1(HUF_CStream_t* bitC)
+{
+    bitC->bitContainer[1] = 0;
+    bitC->bitPos[1] = 0;
+}
+
+/*! HUF_mergeIndex1() :
+ * Merges the bit container @ index 1 into the bit container @ index 0
+ * and zeros the bit container @ index 1.
+ */
+FORCE_INLINE_TEMPLATE void HUF_mergeIndex1(HUF_CStream_t* bitC)
+{
+    assert((bitC->bitPos[1] & 0xFF) < HUF_BITS_IN_CONTAINER);
+    bitC->bitContainer[0] >>= (bitC->bitPos[1] & 0xFF);
+    bitC->bitContainer[0] |= bitC->bitContainer[1];
+    bitC->bitPos[0] += bitC->bitPos[1];
+    assert((bitC->bitPos[0] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+}
+
+/*! HUF_flushBits() :
+* Flushes the bits in the bit container @ index 0.
+*
+* @post bitPos will be < 8.
+* @param kFast If kFast is set then we must know a-priori that
+*              the bit container will not overflow.
+*/
+FORCE_INLINE_TEMPLATE void HUF_flushBits(HUF_CStream_t* bitC, int kFast)
+{
+    /* The upper bits of bitPos are noisy, so we must mask by 0xFF. */
+    size_t const nbBits = bitC->bitPos[0] & 0xFF;
+    size_t const nbBytes = nbBits >> 3;
+    /* The top nbBits bits of bitContainer are the ones we need. */
+    size_t const bitContainer = bitC->bitContainer[0] >> (HUF_BITS_IN_CONTAINER - nbBits);
+    /* Mask bitPos to account for the bytes we consumed. */
+    bitC->bitPos[0] &= 7;
+    assert(nbBits > 0);
+    assert(nbBits <= sizeof(bitC->bitContainer[0]) * 8);
+    assert(bitC->ptr <= bitC->endPtr);
+    MEM_writeLEST(bitC->ptr, bitContainer);
+    bitC->ptr += nbBytes;
+    assert(!kFast || bitC->ptr <= bitC->endPtr);
+    if (!kFast && bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
+    /* bitContainer doesn't need to be modified because the leftover
+     * bits are already the top bitPos bits. And we don't care about
+     * noise in the lower values.
+     */
+}
+
+/*! HUF_endMark()
+ * @returns The Huffman stream end mark: A 1-bit value = 1.
+ */
+static HUF_CElt HUF_endMark(void)
+{
+    HUF_CElt endMark;
+    HUF_setNbBits(&endMark, 1);
+    HUF_setValue(&endMark, 1);
+    return endMark;
+}
+
+/*! HUF_closeCStream() :
+ *  @return Size of CStream, in bytes,
+ *          or 0 if it could not fit into dstBuffer */
+static size_t HUF_closeCStream(HUF_CStream_t* bitC)
+{
+    HUF_addBits(bitC, HUF_endMark(), /* idx */ 0, /* kFast */ 0);
+    HUF_flushBits(bitC, /* kFast */ 0);
+    {
+        size_t const nbBits = bitC->bitPos[0] & 0xFF;
+        if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
+        return (size_t)(bitC->ptr - bitC->startPtr) + (nbBits > 0);
+    }
+}
+
+FORCE_INLINE_TEMPLATE void
+HUF_encodeSymbol(HUF_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable, int idx, int fast)
+{
+    HUF_addBits(bitCPtr, CTable[symbol], idx, fast);
+}
+
+FORCE_INLINE_TEMPLATE void
+HUF_compress1X_usingCTable_internal_body_loop(HUF_CStream_t* bitC,
+                                   const BYTE* ip, size_t srcSize,
+                                   const HUF_CElt* ct,
+                                   int kUnroll, int kFastFlush, int kLastFast)
+{
+    /* Join to kUnroll */
+    int n = (int)srcSize;
+    int rem = n % kUnroll;
+    if (rem > 0) {
+        for (; rem > 0; --rem) {
+            HUF_encodeSymbol(bitC, ip[--n], ct, 0, /* fast */ 0);
+        }
+        HUF_flushBits(bitC, kFastFlush);
+    }
+    assert(n % kUnroll == 0);
+
+    /* Join to 2 * kUnroll */
+    if (n % (2 * kUnroll)) {
+        int u;
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - u], ct, 0, 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, 0, kLastFast);
+        HUF_flushBits(bitC, kFastFlush);
+        n -= kUnroll;
+    }
+    assert(n % (2 * kUnroll) == 0);
+
+    for (; n>0; n-= 2 * kUnroll) {
+        /* Encode kUnroll symbols into the bitstream @ index 0. */
+        int u;
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - u], ct, /* idx */ 0, /* fast */ 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, /* idx */ 0, /* fast */ kLastFast);
+        HUF_flushBits(bitC, kFastFlush);
+        /* Encode kUnroll symbols into the bitstream @ index 1.
+         * This allows us to start filling the bit container
+         * without any data dependencies.
+         */
+        HUF_zeroIndex1(bitC);
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - kUnroll - u], ct, /* idx */ 1, /* fast */ 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll - kUnroll], ct, /* idx */ 1, /* fast */ kLastFast);
+        /* Merge bitstream @ index 1 into the bitstream @ index 0 */
+        HUF_mergeIndex1(bitC);
+        HUF_flushBits(bitC, kFastFlush);
+    }
+    assert(n == 0);
+
+}
+
+/**
+ * Returns a tight upper bound on the output space needed by Huffman
+ * with 8 bytes buffer to handle over-writes. If the output is at least
+ * this large we don't need to do bounds checks during Huffman encoding.
+ */
+static size_t HUF_tightCompressBound(size_t srcSize, size_t tableLog)
+{
+    return ((srcSize * tableLog) >> 3) + 8;
+}
+
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    U32 const tableLog = (U32)CTable[0];
+    HUF_CElt const* ct = CTable + 1;
+    const BYTE* ip = (const BYTE*) src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+    HUF_CStream_t bitC;
+
+    /* init */
+    if (dstSize < 8) return 0;   /* not enough space to compress */
+    { size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
+      if (HUF_isError(initErr)) return 0; }
+
+    if (dstSize < HUF_tightCompressBound(srcSize, (size_t)tableLog) || tableLog > 11)
+        HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ MEM_32bits() ? 2 : 4, /* kFast */ 0, /* kLastFast */ 0);
+    else {
+        if (MEM_32bits()) {
+            switch (tableLog) {
+            case 11:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 10: ZSTD_FALLTHROUGH;
+            case 9: ZSTD_FALLTHROUGH;
+            case 8:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            case 7: ZSTD_FALLTHROUGH;
+            default:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 3, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            }
+        } else {
+            switch (tableLog) {
+            case 11:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 10:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            case 9:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 6, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 8:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 7, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 7:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 8, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 6: ZSTD_FALLTHROUGH;
+            default:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 9, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            }
+        }
+    }
+    assert(bitC.ptr <= bitC.endPtr);
+
+    return HUF_closeCStream(&bitC);
+}
+
+#if DYNAMIC_BMI2
+
+static BMI2_TARGET_ATTRIBUTE size_t
+HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
+                                      const void* src, size_t srcSize,
+                                      const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int flags)
+{
+    if (flags & HUF_flags_bmi2) {
+        return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
+    }
+    return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
+}
+
+#else
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int flags)
+{
+    (void)flags;
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+#endif
+
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
+{
+    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
+}
+
+static size_t
+HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, int flags)
+{
+    size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
+    const BYTE* ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    if (dstSize < 6 + 1 + 1 + 1 + 8) return 0;   /* minimum space to compress successfully */
+    if (srcSize < 12) return 0;   /* no saving possible : too small input */
+    op += 6;   /* jumpTable */
+
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        MEM_writeLE16(ostart, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        MEM_writeLE16(ostart+2, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        MEM_writeLE16(ostart+4, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    assert(op <= oend);
+    assert(ip <= iend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        op += cSize;
+    }
+
+    return (size_t)(op-ostart);
+}
+
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
+{
+    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
+}
+
+typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
+
+static size_t HUF_compressCTable_internal(
+                BYTE* const ostart, BYTE* op, BYTE* const oend,
+                const void* src, size_t srcSize,
+                HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int flags)
+{
+    size_t const cSize = (nbStreams==HUF_singleStream) ?
+                         HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags) :
+                         HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags);
+    if (HUF_isError(cSize)) { return cSize; }
+    if (cSize==0) { return 0; }   /* uncompressible */
+    op += cSize;
+    /* check compressibility */
+    assert(op >= ostart);
+    if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
+    return (size_t)(op-ostart);
+}
+
+typedef struct {
+    unsigned count[HUF_SYMBOLVALUE_MAX + 1];
+    HUF_CElt CTable[HUF_CTABLE_SIZE_ST(HUF_SYMBOLVALUE_MAX)];
+    union {
+        HUF_buildCTable_wksp_tables buildCTable_wksp;
+        HUF_WriteCTableWksp writeCTable_wksp;
+        U32 hist_wksp[HIST_WKSP_SIZE_U32];
+    } wksps;
+} HUF_compress_tables_t;
+
+#define SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE 4096
+#define SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO 10  /* Must be >= 2 */
+
+unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue)
+{
+    unsigned cardinality = 0;
+    unsigned i;
+
+    for (i = 0; i < maxSymbolValue + 1; i++) {
+        if (count[i] != 0) cardinality += 1;
+    }
+
+    return cardinality;
+}
+
+unsigned HUF_minTableLog(unsigned symbolCardinality)
+{
+    U32 minBitsSymbols = ZSTD_highbit32(symbolCardinality) + 1;
+    return minBitsSymbols;
+}
+
+unsigned HUF_optimalTableLog(
+            unsigned maxTableLog,
+            size_t srcSize,
+            unsigned maxSymbolValue,
+            void* workSpace, size_t wkspSize,
+            HUF_CElt* table,
+      const unsigned* count,
+            int flags)
+{
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    assert(wkspSize >= sizeof(HUF_buildCTable_wksp_tables));
+
+    if (!(flags & HUF_flags_optimalDepth)) {
+        /* cheap evaluation, based on FSE */
+        return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    }
+
+    {   BYTE* dst = (BYTE*)workSpace + sizeof(HUF_WriteCTableWksp);
+        size_t dstSize = wkspSize - sizeof(HUF_WriteCTableWksp);
+        size_t maxBits, hSize, newSize;
+        const unsigned symbolCardinality = HUF_cardinality(count, maxSymbolValue);
+        const unsigned minTableLog = HUF_minTableLog(symbolCardinality);
+        size_t optSize = ((size_t) ~0) - 1;
+        unsigned optLog = maxTableLog, optLogGuess;
+
+        DEBUGLOG(6, "HUF_optimalTableLog: probing huf depth (srcSize=%zu)", srcSize);
+
+        /* Search until size increases */
+        for (optLogGuess = minTableLog; optLogGuess <= maxTableLog; optLogGuess++) {
+            DEBUGLOG(7, "checking for huffLog=%u", optLogGuess);
+            maxBits = HUF_buildCTable_wksp(table, count, maxSymbolValue, optLogGuess, workSpace, wkspSize);
+            if (ERR_isError(maxBits)) continue;
+
+            if (maxBits < optLogGuess && optLogGuess > minTableLog) break;
+
+            hSize = HUF_writeCTable_wksp(dst, dstSize, table, maxSymbolValue, (U32)maxBits, workSpace, wkspSize);
+
+            if (ERR_isError(hSize)) continue;
+
+            newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
+
+            if (newSize > optSize + 1) {
+                break;
+            }
+
+            if (newSize < optSize) {
+                optSize = newSize;
+                optLog = optLogGuess;
+            }
+        }
+        assert(optLog <= HUF_TABLELOG_MAX);
+        return optLog;
+    }
+}
+
+/* HUF_compress_internal() :
+ * `workSpace_align4` must be aligned on 4-bytes boundaries,
+ * and occupies the same space as a table of HUF_WORKSPACE_SIZE_U64 unsigned */
+static size_t
+HUF_compress_internal (void* dst, size_t dstSize,
+                 const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned huffLog,
+                       HUF_nbStreams_e nbStreams,
+                       void* workSpace, size_t wkspSize,
+                       HUF_CElt* oldHufTable, HUF_repeat* repeat, int flags)
+{
+    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(size_t));
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    DEBUGLOG(5, "HUF_compress_internal (srcSize=%zu)", srcSize);
+    HUF_STATIC_ASSERT(sizeof(*table) + HUF_WORKSPACE_MAX_ALIGNMENT <= HUF_WORKSPACE_SIZE);
+
+    /* checks & inits */
+    if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
+    if (!srcSize) return 0;  /* Uncompressed */
+    if (!dstSize) return 0;  /* cannot fit anything within dst budget */
+    if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);   /* current block size limit */
+    if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+    if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
+
+    /* Heuristic : If old table is valid, use it for small inputs */
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat == HUF_repeat_valid) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           nbStreams, oldHufTable, flags);
+    }
+
+    /* If uncompressible data is suspected, do a smaller sampling first */
+    DEBUG_STATIC_ASSERT(SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO >= 2);
+    if ((flags & HUF_flags_suspectUncompressible) && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
+        size_t largestTotal = 0;
+        DEBUGLOG(5, "input suspected incompressible : sampling to check");
+        {   unsigned maxSymbolValueBegin = maxSymbolValue;
+            CHECK_V_F(largestBegin, HIST_count_simple (table->count, &maxSymbolValueBegin, (const BYTE*)src, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
+            largestTotal += largestBegin;
+        }
+        {   unsigned maxSymbolValueEnd = maxSymbolValue;
+            CHECK_V_F(largestEnd, HIST_count_simple (table->count, &maxSymbolValueEnd, (const BYTE*)src + srcSize - SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
+            largestTotal += largestEnd;
+        }
+        if (largestTotal <= ((2 * SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
+    }
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->wksps.hist_wksp, sizeof(table->wksps.hist_wksp)) );
+        if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
+        if (largest <= (srcSize >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
+    }
+    DEBUGLOG(6, "histogram detail completed (%zu symbols)", showU32(table->count, maxSymbolValue+1));
+
+    /* Check validity of previous table */
+    if ( repeat
+      && *repeat == HUF_repeat_check
+      && !HUF_validateCTable(oldHufTable, table->count, maxSymbolValue)) {
+        *repeat = HUF_repeat_none;
+    }
+    /* Heuristic : use existing table for small inputs */
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat != HUF_repeat_none) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           nbStreams, oldHufTable, flags);
+    }
+
+    /* Build Huffman Tree */
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, &table->wksps, sizeof(table->wksps), table->CTable, table->count, flags);
+    {   size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
+                                            maxSymbolValue, huffLog,
+                                            &table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
+        CHECK_F(maxBits);
+        huffLog = (U32)maxBits;
+        DEBUGLOG(6, "bit distribution completed (%zu symbols)", showCTableBits(table->CTable + 1, maxSymbolValue+1));
+    }
+    /* Zero unused symbols in CTable, so we can check it for validity */
+    {
+        size_t const ctableSize = HUF_CTABLE_SIZE_ST(maxSymbolValue);
+        size_t const unusedSize = sizeof(table->CTable) - ctableSize * sizeof(HUF_CElt);
+        ZSTD_memset(table->CTable + ctableSize, 0, unusedSize);
+    }
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, HUF_writeCTable_wksp(op, dstSize, table->CTable, maxSymbolValue, huffLog,
+                                              &table->wksps.writeCTable_wksp, sizeof(table->wksps.writeCTable_wksp)) );
+        /* Check if using previous huffman table is beneficial */
+        if (repeat && *repeat != HUF_repeat_none) {
+            size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, table->count, maxSymbolValue);
+            size_t const newSize = HUF_estimateCompressedSize(table->CTable, table->count, maxSymbolValue);
+            if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
+                return HUF_compressCTable_internal(ostart, op, oend,
+                                                   src, srcSize,
+                                                   nbStreams, oldHufTable, flags);
+        }   }
+
+        /* Use the new huffman table */
+        if (hSize + 12ul >= srcSize) { return 0; }
+        op += hSize;
+        if (repeat) { *repeat = HUF_repeat_none; }
+        if (oldHufTable)
+            ZSTD_memcpy(oldHufTable, table->CTable, sizeof(table->CTable));  /* Save new table */
+    }
+    return HUF_compressCTable_internal(ostart, op, oend,
+                                       src, srcSize,
+                                       nbStreams, table->CTable, flags);
+}
+
+size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
+{
+    DEBUGLOG(5, "HUF_compress1X_repeat (srcSize = %zu)", srcSize);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_singleStream,
+                                 workSpace, wkspSize, hufTable,
+                                 repeat, flags);
+}
+
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * consider skipping quickly
+ * re-use an existing huffman compression table */
+size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
+{
+    DEBUGLOG(5, "HUF_compress4X_repeat (srcSize = %zu)", srcSize);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_fourStreams,
+                                 workSpace, wkspSize,
+                                 hufTable, repeat, flags);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress.c
new file mode 100644
index 0000000..d6133e7
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress.c
@@ -0,0 +1,7032 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
+#include "../common/zstd_deps.h"  /* INT_MAX, ZSTD_memset, ZSTD_memcpy */
+#include "../common/mem.h"
+#include "hist.h"           /* HIST_countFast_wksp */
+#define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
+#include "../common/fse.h"
+#include "../common/huf.h"
+#include "zstd_compress_internal.h"
+#include "zstd_compress_sequences.h"
+#include "zstd_compress_literals.h"
+#include "zstd_fast.h"
+#include "zstd_double_fast.h"
+#include "zstd_lazy.h"
+#include "zstd_opt.h"
+#include "zstd_ldm.h"
+#include "zstd_compress_superblock.h"
+#include  "../common/bits.h"      /* ZSTD_highbit32, ZSTD_rotateRight_U64 */
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * COMPRESS_HEAPMODE :
+ * Select how default decompression function ZSTD_compress() allocates its context,
+ * on stack (0, default), or into heap (1).
+ * Note that functions with explicit context such as ZSTD_compressCCtx() are unaffected.
+ */
+#ifndef ZSTD_COMPRESS_HEAPMODE
+#  define ZSTD_COMPRESS_HEAPMODE 0
+#endif
+
+/*!
+ * ZSTD_HASHLOG3_MAX :
+ * Maximum size of the hash table dedicated to find 3-bytes matches,
+ * in log format, aka 17 => 1 << 17 == 128Ki positions.
+ * This structure is only used in zstd_opt.
+ * Since allocation is centralized for all strategies, it has to be known here.
+ * The actual (selected) size of the hash table is then stored in ZSTD_matchState_t.hashLog3,
+ * so that zstd_opt.c doesn't need to know about this constant.
+ */
+#ifndef ZSTD_HASHLOG3_MAX
+#  define ZSTD_HASHLOG3_MAX 17
+#endif
+
+/*-*************************************
+*  Helper functions
+***************************************/
+/* ZSTD_compressBound()
+ * Note that the result from this function is only valid for
+ * the one-pass compression functions.
+ * When employing the streaming mode,
+ * if flushes are frequently altering the size of blocks,
+ * the overhead from block headers can make the compressed data larger
+ * than the return value of ZSTD_compressBound().
+ */
+size_t ZSTD_compressBound(size_t srcSize) {
+    size_t const r = ZSTD_COMPRESSBOUND(srcSize);
+    if (r==0) return ERROR(srcSize_wrong);
+    return r;
+}
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+struct ZSTD_CDict_s {
+    const void* dictContent;
+    size_t dictContentSize;
+    ZSTD_dictContentType_e dictContentType; /* The dictContentType the CDict was created with */
+    U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+    ZSTD_cwksp workspace;
+    ZSTD_matchState_t matchState;
+    ZSTD_compressedBlockState_t cBlockState;
+    ZSTD_customMem customMem;
+    U32 dictID;
+    int compressionLevel; /* 0 indicates that advanced API was used to select CDict params */
+    ZSTD_paramSwitch_e useRowMatchFinder; /* Indicates whether the CDict was created with params that would use
+                                           * row-based matchfinder. Unless the cdict is reloaded, we will use
+                                           * the same greedy/lazy matchfinder at compression time.
+                                           */
+};  /* typedef'd to ZSTD_CDict within "zstd.h" */
+
+ZSTD_CCtx* ZSTD_createCCtx(void)
+{
+    return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
+}
+
+static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
+{
+    assert(cctx != NULL);
+    ZSTD_memset(cctx, 0, sizeof(*cctx));
+    cctx->customMem = memManager;
+    cctx->bmi2 = ZSTD_cpuSupportsBmi2();
+    {   size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
+}
+
+ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
+{
+    ZSTD_STATIC_ASSERT(zcss_init==0);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_customMalloc(sizeof(ZSTD_CCtx), customMem);
+        if (!cctx) return NULL;
+        ZSTD_initCCtx(cctx, customMem);
+        return cctx;
+    }
+}
+
+ZSTD_CCtx* ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize)
+{
+    ZSTD_cwksp ws;
+    ZSTD_CCtx* cctx;
+    if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL;  /* minimum size */
+    if ((size_t)workspace & 7) return NULL;  /* must be 8-aligned */
+    ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_static_alloc);
+
+    cctx = (ZSTD_CCtx*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CCtx));
+    if (cctx == NULL) return NULL;
+
+    ZSTD_memset(cctx, 0, sizeof(ZSTD_CCtx));
+    ZSTD_cwksp_move(&cctx->workspace, &ws);
+    cctx->staticSize = workspaceSize;
+
+    /* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
+    if (!ZSTD_cwksp_check_available(&cctx->workspace, ENTROPY_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL;
+    cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
+    cctx->blockState.nextCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
+    cctx->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cctx->workspace, ENTROPY_WORKSPACE_SIZE);
+    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    return cctx;
+}
+
+/**
+ * Clears and frees all of the dictionaries in the CCtx.
+ */
+static void ZSTD_clearAllDicts(ZSTD_CCtx* cctx)
+{
+    ZSTD_customFree(cctx->localDict.dictBuffer, cctx->customMem);
+    ZSTD_freeCDict(cctx->localDict.cdict);
+    ZSTD_memset(&cctx->localDict, 0, sizeof(cctx->localDict));
+    ZSTD_memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));
+    cctx->cdict = NULL;
+}
+
+static size_t ZSTD_sizeof_localDict(ZSTD_localDict dict)
+{
+    size_t const bufferSize = dict.dictBuffer != NULL ? dict.dictSize : 0;
+    size_t const cdictSize = ZSTD_sizeof_CDict(dict.cdict);
+    return bufferSize + cdictSize;
+}
+
+static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx)
+{
+    assert(cctx != NULL);
+    assert(cctx->staticSize == 0);
+    ZSTD_clearAllDicts(cctx);
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
+#endif
+    ZSTD_cwksp_free(&cctx->workspace, cctx->customMem);
+}
+
+size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support free on NULL */
+    RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
+                    "not compatible with static CCtx");
+    {   int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
+        ZSTD_freeCCtxContent(cctx);
+        if (!cctxInWorkspace) ZSTD_customFree(cctx, cctx->customMem);
+    }
+    return 0;
+}
+
+
+static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    return ZSTDMT_sizeof_CCtx(cctx->mtctx);
+#else
+    (void)cctx;
+    return 0;
+#endif
+}
+
+
+size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support sizeof on NULL */
+    /* cctx may be in the workspace */
+    return (cctx->workspace.workspace == cctx ? 0 : sizeof(*cctx))
+           + ZSTD_cwksp_sizeof(&cctx->workspace)
+           + ZSTD_sizeof_localDict(cctx->localDict)
+           + ZSTD_sizeof_mtctx(cctx);
+}
+
+size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
+{
+    return ZSTD_sizeof_CCtx(zcs);  /* same object */
+}
+
+/* private API call, for dictBuilder only */
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
+
+/* Returns true if the strategy supports using a row based matchfinder */
+static int ZSTD_rowMatchFinderSupported(const ZSTD_strategy strategy) {
+    return (strategy >= ZSTD_greedy && strategy <= ZSTD_lazy2);
+}
+
+/* Returns true if the strategy and useRowMatchFinder mode indicate that we will use the row based matchfinder
+ * for this compression.
+ */
+static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_paramSwitch_e mode) {
+    assert(mode != ZSTD_ps_auto);
+    return ZSTD_rowMatchFinderSupported(strategy) && (mode == ZSTD_ps_enable);
+}
+
+/* Returns row matchfinder usage given an initial mode and cParams */
+static ZSTD_paramSwitch_e ZSTD_resolveRowMatchFinderMode(ZSTD_paramSwitch_e mode,
+                                                         const ZSTD_compressionParameters* const cParams) {
+#if defined(ZSTD_ARCH_X86_SSE2) || defined(ZSTD_ARCH_ARM_NEON)
+    int const kHasSIMD128 = 1;
+#else
+    int const kHasSIMD128 = 0;
+#endif
+    if (mode != ZSTD_ps_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
+    mode = ZSTD_ps_disable;
+    if (!ZSTD_rowMatchFinderSupported(cParams->strategy)) return mode;
+    if (kHasSIMD128) {
+        if (cParams->windowLog > 14) mode = ZSTD_ps_enable;
+    } else {
+        if (cParams->windowLog > 17) mode = ZSTD_ps_enable;
+    }
+    return mode;
+}
+
+/* Returns block splitter usage (generally speaking, when using slower/stronger compression modes) */
+static ZSTD_paramSwitch_e ZSTD_resolveBlockSplitterMode(ZSTD_paramSwitch_e mode,
+                                                        const ZSTD_compressionParameters* const cParams) {
+    if (mode != ZSTD_ps_auto) return mode;
+    return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17) ? ZSTD_ps_enable : ZSTD_ps_disable;
+}
+
+/* Returns 1 if the arguments indicate that we should allocate a chainTable, 0 otherwise */
+static int ZSTD_allocateChainTable(const ZSTD_strategy strategy,
+                                   const ZSTD_paramSwitch_e useRowMatchFinder,
+                                   const U32 forDDSDict) {
+    assert(useRowMatchFinder != ZSTD_ps_auto);
+    /* We always should allocate a chaintable if we are allocating a matchstate for a DDS dictionary matchstate.
+     * We do not allocate a chaintable if we are using ZSTD_fast, or are using the row-based matchfinder.
+     */
+    return forDDSDict || ((strategy != ZSTD_fast) && !ZSTD_rowMatchFinderUsed(strategy, useRowMatchFinder));
+}
+
+/* Returns ZSTD_ps_enable if compression parameters are such that we should
+ * enable long distance matching (wlog >= 27, strategy >= btopt).
+ * Returns ZSTD_ps_disable otherwise.
+ */
+static ZSTD_paramSwitch_e ZSTD_resolveEnableLdm(ZSTD_paramSwitch_e mode,
+                                 const ZSTD_compressionParameters* const cParams) {
+    if (mode != ZSTD_ps_auto) return mode;
+    return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27) ? ZSTD_ps_enable : ZSTD_ps_disable;
+}
+
+static int ZSTD_resolveExternalSequenceValidation(int mode) {
+    return mode;
+}
+
+/* Resolves maxBlockSize to the default if no value is present. */
+static size_t ZSTD_resolveMaxBlockSize(size_t maxBlockSize) {
+    if (maxBlockSize == 0) {
+        return ZSTD_BLOCKSIZE_MAX;
+    } else {
+        return maxBlockSize;
+    }
+}
+
+static ZSTD_paramSwitch_e ZSTD_resolveExternalRepcodeSearch(ZSTD_paramSwitch_e value, int cLevel) {
+    if (value != ZSTD_ps_auto) return value;
+    if (cLevel < 10) {
+        return ZSTD_ps_disable;
+    } else {
+        return ZSTD_ps_enable;
+    }
+}
+
+/* Returns 1 if compression parameters are such that CDict hashtable and chaintable indices are tagged.
+ * If so, the tags need to be removed in ZSTD_resetCCtx_byCopyingCDict. */
+static int ZSTD_CDictIndicesAreTagged(const ZSTD_compressionParameters* const cParams) {
+    return cParams->strategy == ZSTD_fast || cParams->strategy == ZSTD_dfast;
+}
+
+static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
+        ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params cctxParams;
+    /* should not matter, as all cParams are presumed properly defined */
+    ZSTD_CCtxParams_init(&cctxParams, ZSTD_CLEVEL_DEFAULT);
+    cctxParams.cParams = cParams;
+
+    /* Adjust advanced params according to cParams */
+    cctxParams.ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams.ldmParams.enableLdm, &cParams);
+    if (cctxParams.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_ldm_adjustParameters(&cctxParams.ldmParams, &cParams);
+        assert(cctxParams.ldmParams.hashLog >= cctxParams.ldmParams.bucketSizeLog);
+        assert(cctxParams.ldmParams.hashRateLog < 32);
+    }
+    cctxParams.useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams.useBlockSplitter, &cParams);
+    cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
+    cctxParams.validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams.validateSequences);
+    cctxParams.maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams.maxBlockSize);
+    cctxParams.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams.searchForExternalRepcodes,
+                                                                             cctxParams.compressionLevel);
+    assert(!ZSTD_checkCParams(cParams));
+    return cctxParams;
+}
+
+static ZSTD_CCtx_params* ZSTD_createCCtxParams_advanced(
+        ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params* params;
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+    params = (ZSTD_CCtx_params*)ZSTD_customCalloc(
+            sizeof(ZSTD_CCtx_params), customMem);
+    if (!params) { return NULL; }
+    ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
+    params->customMem = customMem;
+    return params;
+}
+
+ZSTD_CCtx_params* ZSTD_createCCtxParams(void)
+{
+    return ZSTD_createCCtxParams_advanced(ZSTD_defaultCMem);
+}
+
+size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
+{
+    if (params == NULL) { return 0; }
+    ZSTD_customFree(params, params->customMem);
+    return 0;
+}
+
+size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
+{
+    return ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
+}
+
+size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
+    RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
+    ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->compressionLevel = compressionLevel;
+    cctxParams->fParams.contentSizeFlag = 1;
+    return 0;
+}
+
+#define ZSTD_NO_CLEVEL 0
+
+/**
+ * Initializes `cctxParams` from `params` and `compressionLevel`.
+ * @param compressionLevel If params are derived from a compression level then that compression level, otherwise ZSTD_NO_CLEVEL.
+ */
+static void
+ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams,
+                        const ZSTD_parameters* params,
+                              int compressionLevel)
+{
+    assert(!ZSTD_checkCParams(params->cParams));
+    ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->cParams = params->cParams;
+    cctxParams->fParams = params->fParams;
+    /* Should not matter, as all cParams are presumed properly defined.
+     * But, set it for tracing anyway.
+     */
+    cctxParams->compressionLevel = compressionLevel;
+    cctxParams->useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams->useRowMatchFinder, &params->cParams);
+    cctxParams->useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams->useBlockSplitter, &params->cParams);
+    cctxParams->ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams->ldmParams.enableLdm, &params->cParams);
+    cctxParams->validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams->validateSequences);
+    cctxParams->maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams->maxBlockSize);
+    cctxParams->searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams->searchForExternalRepcodes, compressionLevel);
+    DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d, useBlockSplitter=%d ldm=%d",
+                cctxParams->useRowMatchFinder, cctxParams->useBlockSplitter, cctxParams->ldmParams.enableLdm);
+}
+
+size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
+{
+    RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
+    ZSTD_CCtxParams_init_internal(cctxParams, &params, ZSTD_NO_CLEVEL);
+    return 0;
+}
+
+/**
+ * Sets cctxParams' cParams and fParams from params, but otherwise leaves them alone.
+ * @param params Validated zstd parameters.
+ */
+static void ZSTD_CCtxParams_setZstdParams(
+        ZSTD_CCtx_params* cctxParams, const ZSTD_parameters* params)
+{
+    assert(!ZSTD_checkCParams(params->cParams));
+    cctxParams->cParams = params->cParams;
+    cctxParams->fParams = params->fParams;
+    /* Should not matter, as all cParams are presumed properly defined.
+     * But, set it for tracing anyway.
+     */
+    cctxParams->compressionLevel = ZSTD_NO_CLEVEL;
+}
+
+ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
+{
+    ZSTD_bounds bounds = { 0, 0, 0 };
+
+    switch(param)
+    {
+    case ZSTD_c_compressionLevel:
+        bounds.lowerBound = ZSTD_minCLevel();
+        bounds.upperBound = ZSTD_maxCLevel();
+        return bounds;
+
+    case ZSTD_c_windowLog:
+        bounds.lowerBound = ZSTD_WINDOWLOG_MIN;
+        bounds.upperBound = ZSTD_WINDOWLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_hashLog:
+        bounds.lowerBound = ZSTD_HASHLOG_MIN;
+        bounds.upperBound = ZSTD_HASHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_chainLog:
+        bounds.lowerBound = ZSTD_CHAINLOG_MIN;
+        bounds.upperBound = ZSTD_CHAINLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_searchLog:
+        bounds.lowerBound = ZSTD_SEARCHLOG_MIN;
+        bounds.upperBound = ZSTD_SEARCHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_minMatch:
+        bounds.lowerBound = ZSTD_MINMATCH_MIN;
+        bounds.upperBound = ZSTD_MINMATCH_MAX;
+        return bounds;
+
+    case ZSTD_c_targetLength:
+        bounds.lowerBound = ZSTD_TARGETLENGTH_MIN;
+        bounds.upperBound = ZSTD_TARGETLENGTH_MAX;
+        return bounds;
+
+    case ZSTD_c_strategy:
+        bounds.lowerBound = ZSTD_STRATEGY_MIN;
+        bounds.upperBound = ZSTD_STRATEGY_MAX;
+        return bounds;
+
+    case ZSTD_c_contentSizeFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_checksumFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_dictIDFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_nbWorkers:
+        bounds.lowerBound = 0;
+#ifdef ZSTD_MULTITHREAD
+        bounds.upperBound = ZSTDMT_NBWORKERS_MAX;
+#else
+        bounds.upperBound = 0;
+#endif
+        return bounds;
+
+    case ZSTD_c_jobSize:
+        bounds.lowerBound = 0;
+#ifdef ZSTD_MULTITHREAD
+        bounds.upperBound = ZSTDMT_JOBSIZE_MAX;
+#else
+        bounds.upperBound = 0;
+#endif
+        return bounds;
+
+    case ZSTD_c_overlapLog:
+#ifdef ZSTD_MULTITHREAD
+        bounds.lowerBound = ZSTD_OVERLAPLOG_MIN;
+        bounds.upperBound = ZSTD_OVERLAPLOG_MAX;
+#else
+        bounds.lowerBound = 0;
+        bounds.upperBound = 0;
+#endif
+        return bounds;
+
+    case ZSTD_c_enableDedicatedDictSearch:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_enableLongDistanceMatching:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_ldmHashLog:
+        bounds.lowerBound = ZSTD_LDM_HASHLOG_MIN;
+        bounds.upperBound = ZSTD_LDM_HASHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmMinMatch:
+        bounds.lowerBound = ZSTD_LDM_MINMATCH_MIN;
+        bounds.upperBound = ZSTD_LDM_MINMATCH_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmBucketSizeLog:
+        bounds.lowerBound = ZSTD_LDM_BUCKETSIZELOG_MIN;
+        bounds.upperBound = ZSTD_LDM_BUCKETSIZELOG_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmHashRateLog:
+        bounds.lowerBound = ZSTD_LDM_HASHRATELOG_MIN;
+        bounds.upperBound = ZSTD_LDM_HASHRATELOG_MAX;
+        return bounds;
+
+    /* experimental parameters */
+    case ZSTD_c_rsyncable:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_forceMaxWindow :
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_format:
+        ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
+        bounds.lowerBound = ZSTD_f_zstd1;
+        bounds.upperBound = ZSTD_f_zstd1_magicless;   /* note : how to ensure at compile time that this is the highest value enum ? */
+        return bounds;
+
+    case ZSTD_c_forceAttachDict:
+        ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceLoad);
+        bounds.lowerBound = ZSTD_dictDefaultAttach;
+        bounds.upperBound = ZSTD_dictForceLoad;       /* note : how to ensure at compile time that this is the highest value enum ? */
+        return bounds;
+
+    case ZSTD_c_literalCompressionMode:
+        ZSTD_STATIC_ASSERT(ZSTD_ps_auto < ZSTD_ps_enable && ZSTD_ps_enable < ZSTD_ps_disable);
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_targetCBlockSize:
+        bounds.lowerBound = ZSTD_TARGETCBLOCKSIZE_MIN;
+        bounds.upperBound = ZSTD_TARGETCBLOCKSIZE_MAX;
+        return bounds;
+
+    case ZSTD_c_srcSizeHint:
+        bounds.lowerBound = ZSTD_SRCSIZEHINT_MIN;
+        bounds.upperBound = ZSTD_SRCSIZEHINT_MAX;
+        return bounds;
+
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+        bounds.lowerBound = (int)ZSTD_bm_buffered;
+        bounds.upperBound = (int)ZSTD_bm_stable;
+        return bounds;
+
+    case ZSTD_c_blockDelimiters:
+        bounds.lowerBound = (int)ZSTD_sf_noBlockDelimiters;
+        bounds.upperBound = (int)ZSTD_sf_explicitBlockDelimiters;
+        return bounds;
+
+    case ZSTD_c_validateSequences:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_useBlockSplitter:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_useRowMatchFinder:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_deterministicRefPrefix:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_prefetchCDictTables:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_maxBlockSize:
+        bounds.lowerBound = ZSTD_BLOCKSIZE_MAX_MIN;
+        bounds.upperBound = ZSTD_BLOCKSIZE_MAX;
+        return bounds;
+
+    case ZSTD_c_searchForExternalRepcodes:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    default:
+        bounds.error = ERROR(parameter_unsupported);
+        return bounds;
+    }
+}
+
+/* ZSTD_cParam_clampBounds:
+ * Clamps the value into the bounded range.
+ */
+static size_t ZSTD_cParam_clampBounds(ZSTD_cParameter cParam, int* value)
+{
+    ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
+    if (ZSTD_isError(bounds.error)) return bounds.error;
+    if (*value < bounds.lowerBound) *value = bounds.lowerBound;
+    if (*value > bounds.upperBound) *value = bounds.upperBound;
+    return 0;
+}
+
+#define BOUNDCHECK(cParam, val) { \
+    RETURN_ERROR_IF(!ZSTD_cParam_withinBounds(cParam,val), \
+                    parameter_outOfBound, "Param out of bounds"); \
+}
+
+
+static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
+{
+    switch(param)
+    {
+    case ZSTD_c_compressionLevel:
+    case ZSTD_c_hashLog:
+    case ZSTD_c_chainLog:
+    case ZSTD_c_searchLog:
+    case ZSTD_c_minMatch:
+    case ZSTD_c_targetLength:
+    case ZSTD_c_strategy:
+        return 1;
+
+    case ZSTD_c_format:
+    case ZSTD_c_windowLog:
+    case ZSTD_c_contentSizeFlag:
+    case ZSTD_c_checksumFlag:
+    case ZSTD_c_dictIDFlag:
+    case ZSTD_c_forceMaxWindow :
+    case ZSTD_c_nbWorkers:
+    case ZSTD_c_jobSize:
+    case ZSTD_c_overlapLog:
+    case ZSTD_c_rsyncable:
+    case ZSTD_c_enableDedicatedDictSearch:
+    case ZSTD_c_enableLongDistanceMatching:
+    case ZSTD_c_ldmHashLog:
+    case ZSTD_c_ldmMinMatch:
+    case ZSTD_c_ldmBucketSizeLog:
+    case ZSTD_c_ldmHashRateLog:
+    case ZSTD_c_forceAttachDict:
+    case ZSTD_c_literalCompressionMode:
+    case ZSTD_c_targetCBlockSize:
+    case ZSTD_c_srcSizeHint:
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+    case ZSTD_c_blockDelimiters:
+    case ZSTD_c_validateSequences:
+    case ZSTD_c_useBlockSplitter:
+    case ZSTD_c_useRowMatchFinder:
+    case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_searchForExternalRepcodes:
+    default:
+        return 0;
+    }
+}
+
+size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParameter (%i, %i)", (int)param, value);
+    if (cctx->streamStage != zcss_init) {
+        if (ZSTD_isUpdateAuthorized(param)) {
+            cctx->cParamsChanged = 1;
+        } else {
+            RETURN_ERROR(stage_wrong, "can only set params in cctx init stage");
+    }   }
+
+    switch(param)
+    {
+    case ZSTD_c_nbWorkers:
+        RETURN_ERROR_IF((value!=0) && cctx->staticSize, parameter_unsupported,
+                        "MT not compatible with static alloc");
+        break;
+
+    case ZSTD_c_compressionLevel:
+    case ZSTD_c_windowLog:
+    case ZSTD_c_hashLog:
+    case ZSTD_c_chainLog:
+    case ZSTD_c_searchLog:
+    case ZSTD_c_minMatch:
+    case ZSTD_c_targetLength:
+    case ZSTD_c_strategy:
+    case ZSTD_c_ldmHashRateLog:
+    case ZSTD_c_format:
+    case ZSTD_c_contentSizeFlag:
+    case ZSTD_c_checksumFlag:
+    case ZSTD_c_dictIDFlag:
+    case ZSTD_c_forceMaxWindow:
+    case ZSTD_c_forceAttachDict:
+    case ZSTD_c_literalCompressionMode:
+    case ZSTD_c_jobSize:
+    case ZSTD_c_overlapLog:
+    case ZSTD_c_rsyncable:
+    case ZSTD_c_enableDedicatedDictSearch:
+    case ZSTD_c_enableLongDistanceMatching:
+    case ZSTD_c_ldmHashLog:
+    case ZSTD_c_ldmMinMatch:
+    case ZSTD_c_ldmBucketSizeLog:
+    case ZSTD_c_targetCBlockSize:
+    case ZSTD_c_srcSizeHint:
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+    case ZSTD_c_blockDelimiters:
+    case ZSTD_c_validateSequences:
+    case ZSTD_c_useBlockSplitter:
+    case ZSTD_c_useRowMatchFinder:
+    case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_searchForExternalRepcodes:
+        break;
+
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+    return ZSTD_CCtxParams_setParameter(&cctx->requestedParams, param, value);
+}
+
+size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
+                                    ZSTD_cParameter param, int value)
+{
+    DEBUGLOG(4, "ZSTD_CCtxParams_setParameter (%i, %i)", (int)param, value);
+    switch(param)
+    {
+    case ZSTD_c_format :
+        BOUNDCHECK(ZSTD_c_format, value);
+        CCtxParams->format = (ZSTD_format_e)value;
+        return (size_t)CCtxParams->format;
+
+    case ZSTD_c_compressionLevel : {
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        if (value == 0)
+            CCtxParams->compressionLevel = ZSTD_CLEVEL_DEFAULT; /* 0 == default */
+        else
+            CCtxParams->compressionLevel = value;
+        if (CCtxParams->compressionLevel >= 0) return (size_t)CCtxParams->compressionLevel;
+        return 0;  /* return type (size_t) cannot represent negative values */
+    }
+
+    case ZSTD_c_windowLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_windowLog, value);
+        CCtxParams->cParams.windowLog = (U32)value;
+        return CCtxParams->cParams.windowLog;
+
+    case ZSTD_c_hashLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_hashLog, value);
+        CCtxParams->cParams.hashLog = (U32)value;
+        return CCtxParams->cParams.hashLog;
+
+    case ZSTD_c_chainLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_chainLog, value);
+        CCtxParams->cParams.chainLog = (U32)value;
+        return CCtxParams->cParams.chainLog;
+
+    case ZSTD_c_searchLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_searchLog, value);
+        CCtxParams->cParams.searchLog = (U32)value;
+        return (size_t)value;
+
+    case ZSTD_c_minMatch :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_minMatch, value);
+        CCtxParams->cParams.minMatch = (U32)value;
+        return CCtxParams->cParams.minMatch;
+
+    case ZSTD_c_targetLength :
+        BOUNDCHECK(ZSTD_c_targetLength, value);
+        CCtxParams->cParams.targetLength = (U32)value;
+        return CCtxParams->cParams.targetLength;
+
+    case ZSTD_c_strategy :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_strategy, value);
+        CCtxParams->cParams.strategy = (ZSTD_strategy)value;
+        return (size_t)CCtxParams->cParams.strategy;
+
+    case ZSTD_c_contentSizeFlag :
+        /* Content size written in frame header _when known_ (default:1) */
+        DEBUGLOG(4, "set content size flag = %u", (value!=0));
+        CCtxParams->fParams.contentSizeFlag = value != 0;
+        return (size_t)CCtxParams->fParams.contentSizeFlag;
+
+    case ZSTD_c_checksumFlag :
+        /* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
+        CCtxParams->fParams.checksumFlag = value != 0;
+        return (size_t)CCtxParams->fParams.checksumFlag;
+
+    case ZSTD_c_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
+        DEBUGLOG(4, "set dictIDFlag = %u", (value!=0));
+        CCtxParams->fParams.noDictIDFlag = !value;
+        return !CCtxParams->fParams.noDictIDFlag;
+
+    case ZSTD_c_forceMaxWindow :
+        CCtxParams->forceWindow = (value != 0);
+        return (size_t)CCtxParams->forceWindow;
+
+    case ZSTD_c_forceAttachDict : {
+        const ZSTD_dictAttachPref_e pref = (ZSTD_dictAttachPref_e)value;
+        BOUNDCHECK(ZSTD_c_forceAttachDict, (int)pref);
+        CCtxParams->attachDictPref = pref;
+        return CCtxParams->attachDictPref;
+    }
+
+    case ZSTD_c_literalCompressionMode : {
+        const ZSTD_paramSwitch_e lcm = (ZSTD_paramSwitch_e)value;
+        BOUNDCHECK(ZSTD_c_literalCompressionMode, (int)lcm);
+        CCtxParams->literalCompressionMode = lcm;
+        return CCtxParams->literalCompressionMode;
+    }
+
+    case ZSTD_c_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        CCtxParams->nbWorkers = value;
+        return CCtxParams->nbWorkers;
+#endif
+
+    case ZSTD_c_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        /* Adjust to the minimum non-default value. */
+        if (value != 0 && value < ZSTDMT_JOBSIZE_MIN)
+            value = ZSTDMT_JOBSIZE_MIN;
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        assert(value >= 0);
+        CCtxParams->jobSize = value;
+        return CCtxParams->jobSize;
+#endif
+
+    case ZSTD_c_overlapLog :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
+        CCtxParams->overlapLog = value;
+        return CCtxParams->overlapLog;
+#endif
+
+    case ZSTD_c_rsyncable :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
+        CCtxParams->rsyncable = value;
+        return CCtxParams->rsyncable;
+#endif
+
+    case ZSTD_c_enableDedicatedDictSearch :
+        CCtxParams->enableDedicatedDictSearch = (value!=0);
+        return (size_t)CCtxParams->enableDedicatedDictSearch;
+
+    case ZSTD_c_enableLongDistanceMatching :
+        BOUNDCHECK(ZSTD_c_enableLongDistanceMatching, value);
+        CCtxParams->ldmParams.enableLdm = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->ldmParams.enableLdm;
+
+    case ZSTD_c_ldmHashLog :
+        if (value!=0)   /* 0 ==> auto */
+            BOUNDCHECK(ZSTD_c_ldmHashLog, value);
+        CCtxParams->ldmParams.hashLog = (U32)value;
+        return CCtxParams->ldmParams.hashLog;
+
+    case ZSTD_c_ldmMinMatch :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmMinMatch, value);
+        CCtxParams->ldmParams.minMatchLength = (U32)value;
+        return CCtxParams->ldmParams.minMatchLength;
+
+    case ZSTD_c_ldmBucketSizeLog :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmBucketSizeLog, value);
+        CCtxParams->ldmParams.bucketSizeLog = (U32)value;
+        return CCtxParams->ldmParams.bucketSizeLog;
+
+    case ZSTD_c_ldmHashRateLog :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmHashRateLog, value);
+        CCtxParams->ldmParams.hashRateLog = (U32)value;
+        return CCtxParams->ldmParams.hashRateLog;
+
+    case ZSTD_c_targetCBlockSize :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
+        CCtxParams->targetCBlockSize = (U32)value;
+        return CCtxParams->targetCBlockSize;
+
+    case ZSTD_c_srcSizeHint :
+        if (value!=0)    /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_srcSizeHint, value);
+        CCtxParams->srcSizeHint = value;
+        return (size_t)CCtxParams->srcSizeHint;
+
+    case ZSTD_c_stableInBuffer:
+        BOUNDCHECK(ZSTD_c_stableInBuffer, value);
+        CCtxParams->inBufferMode = (ZSTD_bufferMode_e)value;
+        return CCtxParams->inBufferMode;
+
+    case ZSTD_c_stableOutBuffer:
+        BOUNDCHECK(ZSTD_c_stableOutBuffer, value);
+        CCtxParams->outBufferMode = (ZSTD_bufferMode_e)value;
+        return CCtxParams->outBufferMode;
+
+    case ZSTD_c_blockDelimiters:
+        BOUNDCHECK(ZSTD_c_blockDelimiters, value);
+        CCtxParams->blockDelimiters = (ZSTD_sequenceFormat_e)value;
+        return CCtxParams->blockDelimiters;
+
+    case ZSTD_c_validateSequences:
+        BOUNDCHECK(ZSTD_c_validateSequences, value);
+        CCtxParams->validateSequences = value;
+        return CCtxParams->validateSequences;
+
+    case ZSTD_c_useBlockSplitter:
+        BOUNDCHECK(ZSTD_c_useBlockSplitter, value);
+        CCtxParams->useBlockSplitter = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->useBlockSplitter;
+
+    case ZSTD_c_useRowMatchFinder:
+        BOUNDCHECK(ZSTD_c_useRowMatchFinder, value);
+        CCtxParams->useRowMatchFinder = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->useRowMatchFinder;
+
+    case ZSTD_c_deterministicRefPrefix:
+        BOUNDCHECK(ZSTD_c_deterministicRefPrefix, value);
+        CCtxParams->deterministicRefPrefix = !!value;
+        return CCtxParams->deterministicRefPrefix;
+
+    case ZSTD_c_prefetchCDictTables:
+        BOUNDCHECK(ZSTD_c_prefetchCDictTables, value);
+        CCtxParams->prefetchCDictTables = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->prefetchCDictTables;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        BOUNDCHECK(ZSTD_c_enableSeqProducerFallback, value);
+        CCtxParams->enableMatchFinderFallback = value;
+        return CCtxParams->enableMatchFinderFallback;
+
+    case ZSTD_c_maxBlockSize:
+        if (value!=0)    /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_maxBlockSize, value);
+        CCtxParams->maxBlockSize = value;
+        return CCtxParams->maxBlockSize;
+
+    case ZSTD_c_searchForExternalRepcodes:
+        BOUNDCHECK(ZSTD_c_searchForExternalRepcodes, value);
+        CCtxParams->searchForExternalRepcodes = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->searchForExternalRepcodes;
+
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+}
+
+size_t ZSTD_CCtx_getParameter(ZSTD_CCtx const* cctx, ZSTD_cParameter param, int* value)
+{
+    return ZSTD_CCtxParams_getParameter(&cctx->requestedParams, param, value);
+}
+
+size_t ZSTD_CCtxParams_getParameter(
+        ZSTD_CCtx_params const* CCtxParams, ZSTD_cParameter param, int* value)
+{
+    switch(param)
+    {
+    case ZSTD_c_format :
+        *value = CCtxParams->format;
+        break;
+    case ZSTD_c_compressionLevel :
+        *value = CCtxParams->compressionLevel;
+        break;
+    case ZSTD_c_windowLog :
+        *value = (int)CCtxParams->cParams.windowLog;
+        break;
+    case ZSTD_c_hashLog :
+        *value = (int)CCtxParams->cParams.hashLog;
+        break;
+    case ZSTD_c_chainLog :
+        *value = (int)CCtxParams->cParams.chainLog;
+        break;
+    case ZSTD_c_searchLog :
+        *value = CCtxParams->cParams.searchLog;
+        break;
+    case ZSTD_c_minMatch :
+        *value = CCtxParams->cParams.minMatch;
+        break;
+    case ZSTD_c_targetLength :
+        *value = CCtxParams->cParams.targetLength;
+        break;
+    case ZSTD_c_strategy :
+        *value = (unsigned)CCtxParams->cParams.strategy;
+        break;
+    case ZSTD_c_contentSizeFlag :
+        *value = CCtxParams->fParams.contentSizeFlag;
+        break;
+    case ZSTD_c_checksumFlag :
+        *value = CCtxParams->fParams.checksumFlag;
+        break;
+    case ZSTD_c_dictIDFlag :
+        *value = !CCtxParams->fParams.noDictIDFlag;
+        break;
+    case ZSTD_c_forceMaxWindow :
+        *value = CCtxParams->forceWindow;
+        break;
+    case ZSTD_c_forceAttachDict :
+        *value = CCtxParams->attachDictPref;
+        break;
+    case ZSTD_c_literalCompressionMode :
+        *value = CCtxParams->literalCompressionMode;
+        break;
+    case ZSTD_c_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        assert(CCtxParams->nbWorkers == 0);
+#endif
+        *value = CCtxParams->nbWorkers;
+        break;
+    case ZSTD_c_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        assert(CCtxParams->jobSize <= INT_MAX);
+        *value = (int)CCtxParams->jobSize;
+        break;
+#endif
+    case ZSTD_c_overlapLog :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        *value = CCtxParams->overlapLog;
+        break;
+#endif
+    case ZSTD_c_rsyncable :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        *value = CCtxParams->rsyncable;
+        break;
+#endif
+    case ZSTD_c_enableDedicatedDictSearch :
+        *value = CCtxParams->enableDedicatedDictSearch;
+        break;
+    case ZSTD_c_enableLongDistanceMatching :
+        *value = CCtxParams->ldmParams.enableLdm;
+        break;
+    case ZSTD_c_ldmHashLog :
+        *value = CCtxParams->ldmParams.hashLog;
+        break;
+    case ZSTD_c_ldmMinMatch :
+        *value = CCtxParams->ldmParams.minMatchLength;
+        break;
+    case ZSTD_c_ldmBucketSizeLog :
+        *value = CCtxParams->ldmParams.bucketSizeLog;
+        break;
+    case ZSTD_c_ldmHashRateLog :
+        *value = CCtxParams->ldmParams.hashRateLog;
+        break;
+    case ZSTD_c_targetCBlockSize :
+        *value = (int)CCtxParams->targetCBlockSize;
+        break;
+    case ZSTD_c_srcSizeHint :
+        *value = (int)CCtxParams->srcSizeHint;
+        break;
+    case ZSTD_c_stableInBuffer :
+        *value = (int)CCtxParams->inBufferMode;
+        break;
+    case ZSTD_c_stableOutBuffer :
+        *value = (int)CCtxParams->outBufferMode;
+        break;
+    case ZSTD_c_blockDelimiters :
+        *value = (int)CCtxParams->blockDelimiters;
+        break;
+    case ZSTD_c_validateSequences :
+        *value = (int)CCtxParams->validateSequences;
+        break;
+    case ZSTD_c_useBlockSplitter :
+        *value = (int)CCtxParams->useBlockSplitter;
+        break;
+    case ZSTD_c_useRowMatchFinder :
+        *value = (int)CCtxParams->useRowMatchFinder;
+        break;
+    case ZSTD_c_deterministicRefPrefix:
+        *value = (int)CCtxParams->deterministicRefPrefix;
+        break;
+    case ZSTD_c_prefetchCDictTables:
+        *value = (int)CCtxParams->prefetchCDictTables;
+        break;
+    case ZSTD_c_enableSeqProducerFallback:
+        *value = CCtxParams->enableMatchFinderFallback;
+        break;
+    case ZSTD_c_maxBlockSize:
+        *value = (int)CCtxParams->maxBlockSize;
+        break;
+    case ZSTD_c_searchForExternalRepcodes:
+        *value = (int)CCtxParams->searchForExternalRepcodes;
+        break;
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+    return 0;
+}
+
+/** ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  just applies `params` into `cctx`
+ *  no action is performed, parameters are merely stored.
+ *  If ZSTDMT is enabled, parameters are pushed to cctx->mtctx.
+ *    This is possible even if a compression is ongoing.
+ *    In which case, new parameters will be applied on the fly, starting with next compression job.
+ */
+size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+        ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParametersUsingCCtxParams");
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "The context is in the wrong stage!");
+    RETURN_ERROR_IF(cctx->cdict, stage_wrong,
+                    "Can't override parameters with cdict attached (some must "
+                    "be inherited from the cdict).");
+
+    cctx->requestedParams = *params;
+    return 0;
+}
+
+size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams)
+{
+    ZSTD_STATIC_ASSERT(sizeof(cparams) == 7 * 4 /* all params are listed below */);
+    DEBUGLOG(4, "ZSTD_CCtx_setCParams");
+    /* only update if all parameters are valid */
+    FORWARD_IF_ERROR(ZSTD_checkCParams(cparams), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, cparams.windowLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_chainLog, cparams.chainLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_hashLog, cparams.hashLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_searchLog, cparams.searchLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, cparams.minMatch), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_targetLength, cparams.targetLength), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_strategy, cparams.strategy), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setFParams(ZSTD_CCtx* cctx, ZSTD_frameParameters fparams)
+{
+    ZSTD_STATIC_ASSERT(sizeof(fparams) == 3 * 4 /* all params are listed below */);
+    DEBUGLOG(4, "ZSTD_CCtx_setFParams");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_contentSizeFlag, fparams.contentSizeFlag != 0), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, fparams.checksumFlag != 0), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_dictIDFlag, fparams.noDictIDFlag == 0), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setParams(ZSTD_CCtx* cctx, ZSTD_parameters params)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParams");
+    /* First check cParams, because we want to update all or none. */
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    /* Next set fParams, because this could fail if the cctx isn't in init stage. */
+    FORWARD_IF_ERROR(ZSTD_CCtx_setFParams(cctx, params.fParams), "");
+    /* Finally set cParams, which should succeed. */
+    FORWARD_IF_ERROR(ZSTD_CCtx_setCParams(cctx, params.cParams), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %llu bytes", pledgedSrcSize);
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't set pledgedSrcSize when not in init stage.");
+    cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+    return 0;
+}
+
+static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(
+        int const compressionLevel,
+        size_t const dictSize);
+static int ZSTD_dedicatedDictSearch_isSupported(
+        const ZSTD_compressionParameters* cParams);
+static void ZSTD_dedicatedDictSearch_revertCParams(
+        ZSTD_compressionParameters* cParams);
+
+/**
+ * Initializes the local dictionary using requested parameters.
+ * NOTE: Initialization does not employ the pledged src size,
+ * because the dictionary may be used for multiple compressions.
+ */
+static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
+{
+    ZSTD_localDict* const dl = &cctx->localDict;
+    if (dl->dict == NULL) {
+        /* No local dictionary. */
+        assert(dl->dictBuffer == NULL);
+        assert(dl->cdict == NULL);
+        assert(dl->dictSize == 0);
+        return 0;
+    }
+    if (dl->cdict != NULL) {
+        /* Local dictionary already initialized. */
+        assert(cctx->cdict == dl->cdict);
+        return 0;
+    }
+    assert(dl->dictSize > 0);
+    assert(cctx->cdict == NULL);
+    assert(cctx->prefixDict.dict == NULL);
+
+    dl->cdict = ZSTD_createCDict_advanced2(
+            dl->dict,
+            dl->dictSize,
+            ZSTD_dlm_byRef,
+            dl->dictContentType,
+            &cctx->requestedParams,
+            cctx->customMem);
+    RETURN_ERROR_IF(!dl->cdict, memory_allocation, "ZSTD_createCDict_advanced failed");
+    cctx->cdict = dl->cdict;
+    return 0;
+}
+
+size_t ZSTD_CCtx_loadDictionary_advanced(
+        ZSTD_CCtx* cctx,
+        const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod,
+        ZSTD_dictContentType_e dictContentType)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't load a dictionary when cctx is not in init stage.");
+    ZSTD_clearAllDicts(cctx);  /* erase any previously set dictionary */
+    if (dict == NULL || dictSize == 0)  /* no dictionary */
+        return 0;
+    if (dictLoadMethod == ZSTD_dlm_byRef) {
+        cctx->localDict.dict = dict;
+    } else {
+        /* copy dictionary content inside CCtx to own its lifetime */
+        void* dictBuffer;
+        RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
+                        "static CCtx can't allocate for an internal copy of dictionary");
+        dictBuffer = ZSTD_customMalloc(dictSize, cctx->customMem);
+        RETURN_ERROR_IF(dictBuffer==NULL, memory_allocation,
+                        "allocation failed for dictionary content");
+        ZSTD_memcpy(dictBuffer, dict, dictSize);
+        cctx->localDict.dictBuffer = dictBuffer;  /* owned ptr to free */
+        cctx->localDict.dict = dictBuffer;        /* read-only reference */
+    }
+    cctx->localDict.dictSize = dictSize;
+    cctx->localDict.dictContentType = dictContentType;
+    return 0;
+}
+
+size_t ZSTD_CCtx_loadDictionary_byReference(
+      ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+
+size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a dict when ctx not in init stage.");
+    /* Free the existing local cdict (if any) to save memory. */
+    ZSTD_clearAllDicts(cctx);
+    cctx->cdict = cdict;
+    return 0;
+}
+
+size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a pool when ctx not in init stage.");
+    cctx->pool = pool;
+    return 0;
+}
+
+size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_CCtx_refPrefix_advanced(cctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+size_t ZSTD_CCtx_refPrefix_advanced(
+        ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a prefix when ctx not in init stage.");
+    ZSTD_clearAllDicts(cctx);
+    if (prefix != NULL && prefixSize > 0) {
+        cctx->prefixDict.dict = prefix;
+        cctx->prefixDict.dictSize = prefixSize;
+        cctx->prefixDict.dictContentType = dictContentType;
+    }
+    return 0;
+}
+
+/*! ZSTD_CCtx_reset() :
+ *  Also dumps dictionary */
+size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset)
+{
+    if ( (reset == ZSTD_reset_session_only)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        cctx->streamStage = zcss_init;
+        cctx->pledgedSrcSizePlusOne = 0;
+    }
+    if ( (reset == ZSTD_reset_parameters)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                        "Reset parameters is only possible during init stage.");
+        ZSTD_clearAllDicts(cctx);
+        ZSTD_memset(&cctx->externalMatchCtx, 0, sizeof(cctx->externalMatchCtx));
+        return ZSTD_CCtxParams_reset(&cctx->requestedParams);
+    }
+    return 0;
+}
+
+
+/** ZSTD_checkCParams() :
+    control CParam values remain within authorized range.
+    @return : 0, or an error code if one value is beyond authorized range */
+size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
+{
+    BOUNDCHECK(ZSTD_c_windowLog, (int)cParams.windowLog);
+    BOUNDCHECK(ZSTD_c_chainLog,  (int)cParams.chainLog);
+    BOUNDCHECK(ZSTD_c_hashLog,   (int)cParams.hashLog);
+    BOUNDCHECK(ZSTD_c_searchLog, (int)cParams.searchLog);
+    BOUNDCHECK(ZSTD_c_minMatch,  (int)cParams.minMatch);
+    BOUNDCHECK(ZSTD_c_targetLength,(int)cParams.targetLength);
+    BOUNDCHECK(ZSTD_c_strategy,  cParams.strategy);
+    return 0;
+}
+
+/** ZSTD_clampCParams() :
+ *  make CParam values within valid range.
+ *  @return : valid CParams */
+static ZSTD_compressionParameters
+ZSTD_clampCParams(ZSTD_compressionParameters cParams)
+{
+#   define CLAMP_TYPE(cParam, val, type) {                                \
+        ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);         \
+        if ((int)val<bounds.lowerBound) val=(type)bounds.lowerBound;      \
+        else if ((int)val>bounds.upperBound) val=(type)bounds.upperBound; \
+    }
+#   define CLAMP(cParam, val) CLAMP_TYPE(cParam, val, unsigned)
+    CLAMP(ZSTD_c_windowLog, cParams.windowLog);
+    CLAMP(ZSTD_c_chainLog,  cParams.chainLog);
+    CLAMP(ZSTD_c_hashLog,   cParams.hashLog);
+    CLAMP(ZSTD_c_searchLog, cParams.searchLog);
+    CLAMP(ZSTD_c_minMatch,  cParams.minMatch);
+    CLAMP(ZSTD_c_targetLength,cParams.targetLength);
+    CLAMP_TYPE(ZSTD_c_strategy,cParams.strategy, ZSTD_strategy);
+    return cParams;
+}
+
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
+{
+    U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
+    return hashLog - btScale;
+}
+
+/** ZSTD_dictAndWindowLog() :
+ * Returns an adjusted window log that is large enough to fit the source and the dictionary.
+ * The zstd format says that the entire dictionary is valid if one byte of the dictionary
+ * is within the window. So the hashLog and chainLog should be large enough to reference both
+ * the dictionary and the window. So we must use this adjusted dictAndWindowLog when downsizing
+ * the hashLog and windowLog.
+ * NOTE: srcSize must not be ZSTD_CONTENTSIZE_UNKNOWN.
+ */
+static U32 ZSTD_dictAndWindowLog(U32 windowLog, U64 srcSize, U64 dictSize)
+{
+    const U64 maxWindowSize = 1ULL << ZSTD_WINDOWLOG_MAX;
+    /* No dictionary ==> No change */
+    if (dictSize == 0) {
+        return windowLog;
+    }
+    assert(windowLog <= ZSTD_WINDOWLOG_MAX);
+    assert(srcSize != ZSTD_CONTENTSIZE_UNKNOWN); /* Handled in ZSTD_adjustCParams_internal() */
+    {
+        U64 const windowSize = 1ULL << windowLog;
+        U64 const dictAndWindowSize = dictSize + windowSize;
+        /* If the window size is already large enough to fit both the source and the dictionary
+         * then just use the window size. Otherwise adjust so that it fits the dictionary and
+         * the window.
+         */
+        if (windowSize >= dictSize + srcSize) {
+            return windowLog; /* Window size large enough already */
+        } else if (dictAndWindowSize >= maxWindowSize) {
+            return ZSTD_WINDOWLOG_MAX; /* Larger than max window log */
+        } else  {
+            return ZSTD_highbit32((U32)dictAndWindowSize - 1) + 1;
+        }
+    }
+}
+
+/** ZSTD_adjustCParams_internal() :
+ *  optimize `cPar` for a specified input (`srcSize` and `dictSize`).
+ *  mostly downsize to reduce memory consumption and initialization latency.
+ * `srcSize` can be ZSTD_CONTENTSIZE_UNKNOWN when not known.
+ * `mode` is the mode for parameter adjustment. See docs for `ZSTD_cParamMode_e`.
+ *  note : `srcSize==0` means 0!
+ *  condition : cPar is presumed validated (can be checked using ZSTD_checkCParams()). */
+static ZSTD_compressionParameters
+ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
+                            unsigned long long srcSize,
+                            size_t dictSize,
+                            ZSTD_cParamMode_e mode,
+                            ZSTD_paramSwitch_e useRowMatchFinder)
+{
+    const U64 minSrcSize = 513; /* (1<<9) + 1 */
+    const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
+    assert(ZSTD_checkCParams(cPar)==0);
+
+    switch (mode) {
+    case ZSTD_cpm_unknown:
+    case ZSTD_cpm_noAttachDict:
+        /* If we don't know the source size, don't make any
+         * assumptions about it. We will already have selected
+         * smaller parameters if a dictionary is in use.
+         */
+        break;
+    case ZSTD_cpm_createCDict:
+        /* Assume a small source size when creating a dictionary
+         * with an unknown source size.
+         */
+        if (dictSize && srcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+            srcSize = minSrcSize;
+        break;
+    case ZSTD_cpm_attachDict:
+        /* Dictionary has its own dedicated parameters which have
+         * already been selected. We are selecting parameters
+         * for only the source.
+         */
+        dictSize = 0;
+        break;
+    default:
+        assert(0);
+        break;
+    }
+
+    /* resize windowLog if input is small enough, to use less memory */
+    if ( (srcSize <= maxWindowResize)
+      && (dictSize <= maxWindowResize) )  {
+        U32 const tSize = (U32)(srcSize + dictSize);
+        static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
+        U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
+                            ZSTD_highbit32(tSize-1) + 1;
+        if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
+    }
+    if (srcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        U32 const dictAndWindowLog = ZSTD_dictAndWindowLog(cPar.windowLog, (U64)srcSize, (U64)dictSize);
+        U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
+        if (cPar.hashLog > dictAndWindowLog+1) cPar.hashLog = dictAndWindowLog+1;
+        if (cycleLog > dictAndWindowLog)
+            cPar.chainLog -= (cycleLog - dictAndWindowLog);
+    }
+
+    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
+        cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* minimum wlog required for valid frame header */
+
+    /* We can't use more than 32 bits of hash in total, so that means that we require:
+     * (hashLog + 8) <= 32 && (chainLog + 8) <= 32
+     */
+    if (mode == ZSTD_cpm_createCDict && ZSTD_CDictIndicesAreTagged(&cPar)) {
+        U32 const maxShortCacheHashLog = 32 - ZSTD_SHORT_CACHE_TAG_BITS;
+        if (cPar.hashLog > maxShortCacheHashLog) {
+            cPar.hashLog = maxShortCacheHashLog;
+        }
+        if (cPar.chainLog > maxShortCacheHashLog) {
+            cPar.chainLog = maxShortCacheHashLog;
+        }
+    }
+
+
+    /* At this point, we aren't 100% sure if we are using the row match finder.
+     * Unless it is explicitly disabled, conservatively assume that it is enabled.
+     * In this case it will only be disabled for small sources, so shrinking the
+     * hash log a little bit shouldn't result in any ratio loss.
+     */
+    if (useRowMatchFinder == ZSTD_ps_auto)
+        useRowMatchFinder = ZSTD_ps_enable;
+
+    /* We can't hash more than 32-bits in total. So that means that we require:
+     * (hashLog - rowLog + 8) <= 32
+     */
+    if (ZSTD_rowMatchFinderUsed(cPar.strategy, useRowMatchFinder)) {
+        /* Switch to 32-entry rows if searchLog is 5 (or more) */
+        U32 const rowLog = BOUNDED(4, cPar.searchLog, 6);
+        U32 const maxRowHashLog = 32 - ZSTD_ROW_HASH_TAG_BITS;
+        U32 const maxHashLog = maxRowHashLog + rowLog;
+        assert(cPar.hashLog >= rowLog);
+        if (cPar.hashLog > maxHashLog) {
+            cPar.hashLog = maxHashLog;
+        }
+    }
+
+    return cPar;
+}
+
+ZSTD_compressionParameters
+ZSTD_adjustCParams(ZSTD_compressionParameters cPar,
+                   unsigned long long srcSize,
+                   size_t dictSize)
+{
+    cPar = ZSTD_clampCParams(cPar);   /* resulting cPar is necessarily valid (all parameters within range) */
+    if (srcSize == 0) srcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize, ZSTD_cpm_unknown, ZSTD_ps_auto);
+}
+
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
+static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
+
+static void ZSTD_overrideCParams(
+              ZSTD_compressionParameters* cParams,
+        const ZSTD_compressionParameters* overrides)
+{
+    if (overrides->windowLog)    cParams->windowLog    = overrides->windowLog;
+    if (overrides->hashLog)      cParams->hashLog      = overrides->hashLog;
+    if (overrides->chainLog)     cParams->chainLog     = overrides->chainLog;
+    if (overrides->searchLog)    cParams->searchLog    = overrides->searchLog;
+    if (overrides->minMatch)     cParams->minMatch     = overrides->minMatch;
+    if (overrides->targetLength) cParams->targetLength = overrides->targetLength;
+    if (overrides->strategy)     cParams->strategy     = overrides->strategy;
+}
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode)
+{
+    ZSTD_compressionParameters cParams;
+    if (srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN && CCtxParams->srcSizeHint > 0) {
+      srcSizeHint = CCtxParams->srcSizeHint;
+    }
+    cParams = ZSTD_getCParams_internal(CCtxParams->compressionLevel, srcSizeHint, dictSize, mode);
+    if (CCtxParams->ldmParams.enableLdm == ZSTD_ps_enable) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
+    ZSTD_overrideCParams(&cParams, &CCtxParams->cParams);
+    assert(!ZSTD_checkCParams(cParams));
+    /* srcSizeHint == 0 means 0 */
+    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize, mode, CCtxParams->useRowMatchFinder);
+}
+
+static size_t
+ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
+                       const ZSTD_paramSwitch_e useRowMatchFinder,
+                       const U32 enableDedicatedDictSearch,
+                       const U32 forCCtx)
+{
+    /* chain table size should be 0 for fast or row-hash strategies */
+    size_t const chainSize = ZSTD_allocateChainTable(cParams->strategy, useRowMatchFinder, enableDedicatedDictSearch && !forCCtx)
+                                ? ((size_t)1 << cParams->chainLog)
+                                : 0;
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = (forCCtx && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
+    /* We don't use ZSTD_cwksp_alloc_size() here because the tables aren't
+     * surrounded by redzones in ASAN. */
+    size_t const tableSpace = chainSize * sizeof(U32)
+                            + hSize * sizeof(U32)
+                            + h3Size * sizeof(U32);
+    size_t const optPotentialSpace =
+        ZSTD_cwksp_aligned_alloc_size((MaxML+1) * sizeof(U32))
+      + ZSTD_cwksp_aligned_alloc_size((MaxLL+1) * sizeof(U32))
+      + ZSTD_cwksp_aligned_alloc_size((MaxOff+1) * sizeof(U32))
+      + ZSTD_cwksp_aligned_alloc_size((1<<Litbits) * sizeof(U32))
+      + ZSTD_cwksp_aligned_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t))
+      + ZSTD_cwksp_aligned_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
+    size_t const lazyAdditionalSpace = ZSTD_rowMatchFinderUsed(cParams->strategy, useRowMatchFinder)
+                                            ? ZSTD_cwksp_aligned_alloc_size(hSize)
+                                            : 0;
+    size_t const optSpace = (forCCtx && (cParams->strategy >= ZSTD_btopt))
+                                ? optPotentialSpace
+                                : 0;
+    size_t const slackSpace = ZSTD_cwksp_slack_space_required();
+
+    /* tables are guaranteed to be sized in multiples of 64 bytes (or 16 uint32_t) */
+    ZSTD_STATIC_ASSERT(ZSTD_HASHLOG_MIN >= 4 && ZSTD_WINDOWLOG_MIN >= 4 && ZSTD_CHAINLOG_MIN >= 4);
+    assert(useRowMatchFinder != ZSTD_ps_auto);
+
+    DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
+                (U32)chainSize, (U32)hSize, (U32)h3Size);
+    return tableSpace + optSpace + slackSpace + lazyAdditionalSpace;
+}
+
+/* Helper function for calculating memory requirements.
+ * Gives a tighter bound than ZSTD_sequenceBound() by taking minMatch into account. */
+static size_t ZSTD_maxNbSeq(size_t blockSize, unsigned minMatch, int useSequenceProducer) {
+    U32 const divider = (minMatch==3 || useSequenceProducer) ? 3 : 4;
+    return blockSize / divider;
+}
+
+static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+        const ZSTD_compressionParameters* cParams,
+        const ldmParams_t* ldmParams,
+        const int isStatic,
+        const ZSTD_paramSwitch_e useRowMatchFinder,
+        const size_t buffInSize,
+        const size_t buffOutSize,
+        const U64 pledgedSrcSize,
+        int useSequenceProducer,
+        size_t maxBlockSize)
+{
+    size_t const windowSize = (size_t) BOUNDED(1ULL, 1ULL << cParams->windowLog, pledgedSrcSize);
+    size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(maxBlockSize), windowSize);
+    size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, cParams->minMatch, useSequenceProducer);
+    size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
+                            + ZSTD_cwksp_aligned_alloc_size(maxNbSeq * sizeof(seqDef))
+                            + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
+    size_t const entropySpace = ZSTD_cwksp_alloc_size(ENTROPY_WORKSPACE_SIZE);
+    size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
+    size_t const matchStateSize = ZSTD_sizeof_matchState(cParams, useRowMatchFinder, /* enableDedicatedDictSearch */ 0, /* forCCtx */ 1);
+
+    size_t const ldmSpace = ZSTD_ldm_getTableSize(*ldmParams);
+    size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(*ldmParams, blockSize);
+    size_t const ldmSeqSpace = ldmParams->enableLdm == ZSTD_ps_enable ?
+        ZSTD_cwksp_aligned_alloc_size(maxNbLdmSeq * sizeof(rawSeq)) : 0;
+
+
+    size_t const bufferSpace = ZSTD_cwksp_alloc_size(buffInSize)
+                             + ZSTD_cwksp_alloc_size(buffOutSize);
+
+    size_t const cctxSpace = isStatic ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
+
+    size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+    size_t const externalSeqSpace = useSequenceProducer
+        ? ZSTD_cwksp_aligned_alloc_size(maxNbExternalSeq * sizeof(ZSTD_Sequence))
+        : 0;
+
+    size_t const neededSpace =
+        cctxSpace +
+        entropySpace +
+        blockStateSpace +
+        ldmSpace +
+        ldmSeqSpace +
+        matchStateSize +
+        tokenSpace +
+        bufferSpace +
+        externalSeqSpace;
+
+    DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
+    return neededSpace;
+}
+
+size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+    ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder,
+                                                                               &cParams);
+
+    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
+    /* estimateCCtxSize is for one-shot compression. So no buffers should
+     * be needed. However, we still allocate two 0-sized buffers, which can
+     * take space under ASAN. */
+    return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+        &cParams, &params->ldmParams, 1, useRowMatchFinder, 0, 0, ZSTD_CONTENTSIZE_UNKNOWN, params->useSequenceProducer, params->maxBlockSize);
+}
+
+size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params initialParams = ZSTD_makeCCtxParamsFromCParams(cParams);
+    if (ZSTD_rowMatchFinderSupported(cParams.strategy)) {
+        /* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
+        size_t noRowCCtxSize;
+        size_t rowCCtxSize;
+        initialParams.useRowMatchFinder = ZSTD_ps_disable;
+        noRowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
+        initialParams.useRowMatchFinder = ZSTD_ps_enable;
+        rowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
+        return MAX(noRowCCtxSize, rowCCtxSize);
+    } else {
+        return ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
+    }
+}
+
+static size_t ZSTD_estimateCCtxSize_internal(int compressionLevel)
+{
+    int tier = 0;
+    size_t largestSize = 0;
+    static const unsigned long long srcSizeTiers[4] = {16 KB, 128 KB, 256 KB, ZSTD_CONTENTSIZE_UNKNOWN};
+    for (; tier < 4; ++tier) {
+        /* Choose the set of cParams for a given level across all srcSizes that give the largest cctxSize */
+        ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, srcSizeTiers[tier], 0, ZSTD_cpm_noAttachDict);
+        largestSize = MAX(ZSTD_estimateCCtxSize_usingCParams(cParams), largestSize);
+    }
+    return largestSize;
+}
+
+size_t ZSTD_estimateCCtxSize(int compressionLevel)
+{
+    int level;
+    size_t memBudget = 0;
+    for (level=MIN(compressionLevel, 1); level<=compressionLevel; level++) {
+        /* Ensure monotonically increasing memory usage as compression level increases */
+        size_t const newMB = ZSTD_estimateCCtxSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
+}
+
+size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
+    {   ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+        size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(params->maxBlockSize), (size_t)1 << cParams.windowLog);
+        size_t const inBuffSize = (params->inBufferMode == ZSTD_bm_buffered)
+                ? ((size_t)1 << cParams.windowLog) + blockSize
+                : 0;
+        size_t const outBuffSize = (params->outBufferMode == ZSTD_bm_buffered)
+                ? ZSTD_compressBound(blockSize) + 1
+                : 0;
+        ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder, &params->cParams);
+
+        return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+            &cParams, &params->ldmParams, 1, useRowMatchFinder, inBuffSize, outBuffSize,
+            ZSTD_CONTENTSIZE_UNKNOWN, params->useSequenceProducer, params->maxBlockSize);
+    }
+}
+
+size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params initialParams = ZSTD_makeCCtxParamsFromCParams(cParams);
+    if (ZSTD_rowMatchFinderSupported(cParams.strategy)) {
+        /* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
+        size_t noRowCCtxSize;
+        size_t rowCCtxSize;
+        initialParams.useRowMatchFinder = ZSTD_ps_disable;
+        noRowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
+        initialParams.useRowMatchFinder = ZSTD_ps_enable;
+        rowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
+        return MAX(noRowCCtxSize, rowCCtxSize);
+    } else {
+        return ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
+    }
+}
+
+static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+    return ZSTD_estimateCStreamSize_usingCParams(cParams);
+}
+
+size_t ZSTD_estimateCStreamSize(int compressionLevel)
+{
+    int level;
+    size_t memBudget = 0;
+    for (level=MIN(compressionLevel, 1); level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCStreamSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
+}
+
+/* ZSTD_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads (non-blocking mode).
+ */
+ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_getFrameProgression(cctx->mtctx);
+    }
+#endif
+    {   ZSTD_frameProgression fp;
+        size_t const buffered = (cctx->inBuff == NULL) ? 0 :
+                                cctx->inBuffPos - cctx->inToCompress;
+        if (buffered) assert(cctx->inBuffPos >= cctx->inToCompress);
+        assert(buffered <= ZSTD_BLOCKSIZE_MAX);
+        fp.ingested = cctx->consumedSrcSize + buffered;
+        fp.consumed = cctx->consumedSrcSize;
+        fp.produced = cctx->producedCSize;
+        fp.flushed  = cctx->producedCSize;   /* simplified; some data might still be left within streaming output buffer */
+        fp.currentJobID = 0;
+        fp.nbActiveWorkers = 0;
+        return fp;
+}   }
+
+/*! ZSTD_toFlushNow()
+ *  Only useful for multithreading scenarios currently (nbWorkers >= 1).
+ */
+size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_toFlushNow(cctx->mtctx);
+    }
+#endif
+    (void)cctx;
+    return 0;   /* over-simplification; could also check if context is currently running in streaming mode, and in which case, report how many bytes are left to be flushed within output buffer */
+}
+
+static void ZSTD_assertEqualCParams(ZSTD_compressionParameters cParams1,
+                                    ZSTD_compressionParameters cParams2)
+{
+    (void)cParams1;
+    (void)cParams2;
+    assert(cParams1.windowLog    == cParams2.windowLog);
+    assert(cParams1.chainLog     == cParams2.chainLog);
+    assert(cParams1.hashLog      == cParams2.hashLog);
+    assert(cParams1.searchLog    == cParams2.searchLog);
+    assert(cParams1.minMatch     == cParams2.minMatch);
+    assert(cParams1.targetLength == cParams2.targetLength);
+    assert(cParams1.strategy     == cParams2.strategy);
+}
+
+void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
+{
+    int i;
+    for (i = 0; i < ZSTD_REP_NUM; ++i)
+        bs->rep[i] = repStartValue[i];
+    bs->entropy.huf.repeatMode = HUF_repeat_none;
+    bs->entropy.fse.offcode_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.matchlength_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.litlength_repeatMode = FSE_repeat_none;
+}
+
+/*! ZSTD_invalidateMatchState()
+ *  Invalidate all the matches in the match finder tables.
+ *  Requires nextSrc and base to be set (can be NULL).
+ */
+static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
+{
+    ZSTD_window_clear(&ms->window);
+
+    ms->nextToUpdate = ms->window.dictLimit;
+    ms->loadedDictEnd = 0;
+    ms->opt.litLengthSum = 0;  /* force reset of btopt stats */
+    ms->dictMatchState = NULL;
+}
+
+/**
+ * Controls, for this matchState reset, whether the tables need to be cleared /
+ * prepared for the coming compression (ZSTDcrp_makeClean), or whether the
+ * tables can be left unclean (ZSTDcrp_leaveDirty), because we know that a
+ * subsequent operation will overwrite the table space anyways (e.g., copying
+ * the matchState contents in from a CDict).
+ */
+typedef enum {
+    ZSTDcrp_makeClean,
+    ZSTDcrp_leaveDirty
+} ZSTD_compResetPolicy_e;
+
+/**
+ * Controls, for this matchState reset, whether indexing can continue where it
+ * left off (ZSTDirp_continue), or whether it needs to be restarted from zero
+ * (ZSTDirp_reset).
+ */
+typedef enum {
+    ZSTDirp_continue,
+    ZSTDirp_reset
+} ZSTD_indexResetPolicy_e;
+
+typedef enum {
+    ZSTD_resetTarget_CDict,
+    ZSTD_resetTarget_CCtx
+} ZSTD_resetTarget_e;
+
+/* Mixes bits in a 64 bits in a value, based on XXH3_rrmxmx */
+static U64 ZSTD_bitmix(U64 val, U64 len) {
+    val ^= ZSTD_rotateRight_U64(val, 49) ^ ZSTD_rotateRight_U64(val, 24);
+    val *= 0x9FB21C651E98DF25ULL;
+    val ^= (val >> 35) + len ;
+    val *= 0x9FB21C651E98DF25ULL;
+    return val ^ (val >> 28);
+}
+
+/* Mixes in the hashSalt and hashSaltEntropy to create a new hashSalt */
+static void ZSTD_advanceHashSalt(ZSTD_matchState_t* ms) {
+    ms->hashSalt = ZSTD_bitmix(ms->hashSalt, 8) ^ ZSTD_bitmix((U64) ms->hashSaltEntropy, 4);
+}
+
+static size_t
+ZSTD_reset_matchState(ZSTD_matchState_t* ms,
+                      ZSTD_cwksp* ws,
+                const ZSTD_compressionParameters* cParams,
+                const ZSTD_paramSwitch_e useRowMatchFinder,
+                const ZSTD_compResetPolicy_e crp,
+                const ZSTD_indexResetPolicy_e forceResetIndex,
+                const ZSTD_resetTarget_e forWho)
+{
+    /* disable chain table allocation for fast or row-based strategies */
+    size_t const chainSize = ZSTD_allocateChainTable(cParams->strategy, useRowMatchFinder,
+                                                     ms->dedicatedDictSearch && (forWho == ZSTD_resetTarget_CDict))
+                                ? ((size_t)1 << cParams->chainLog)
+                                : 0;
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = ((forWho == ZSTD_resetTarget_CCtx) && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
+
+    DEBUGLOG(4, "reset indices : %u", forceResetIndex == ZSTDirp_reset);
+    assert(useRowMatchFinder != ZSTD_ps_auto);
+    if (forceResetIndex == ZSTDirp_reset) {
+        ZSTD_window_init(&ms->window);
+        ZSTD_cwksp_mark_tables_dirty(ws);
+    }
+
+    ms->hashLog3 = hashLog3;
+    ms->lazySkipping = 0;
+
+    ZSTD_invalidateMatchState(ms);
+
+    assert(!ZSTD_cwksp_reserve_failed(ws)); /* check that allocation hasn't already failed */
+
+    ZSTD_cwksp_clear_tables(ws);
+
+    DEBUGLOG(5, "reserving table space");
+    /* table Space */
+    ms->hashTable = (U32*)ZSTD_cwksp_reserve_table(ws, hSize * sizeof(U32));
+    ms->chainTable = (U32*)ZSTD_cwksp_reserve_table(ws, chainSize * sizeof(U32));
+    ms->hashTable3 = (U32*)ZSTD_cwksp_reserve_table(ws, h3Size * sizeof(U32));
+    RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
+                    "failed a workspace allocation in ZSTD_reset_matchState");
+
+    DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_leaveDirty);
+    if (crp!=ZSTDcrp_leaveDirty) {
+        /* reset tables only */
+        ZSTD_cwksp_clean_tables(ws);
+    }
+
+    if (ZSTD_rowMatchFinderUsed(cParams->strategy, useRowMatchFinder)) {
+        /* Row match finder needs an additional table of hashes ("tags") */
+        size_t const tagTableSize = hSize;
+        /* We want to generate a new salt in case we reset a Cctx, but we always want to use
+         * 0 when we reset a Cdict */
+        if(forWho == ZSTD_resetTarget_CCtx) {
+            ms->tagTable = (BYTE*) ZSTD_cwksp_reserve_aligned_init_once(ws, tagTableSize);
+            ZSTD_advanceHashSalt(ms);
+        } else {
+            /* When we are not salting we want to always memset the memory */
+            ms->tagTable = (BYTE*) ZSTD_cwksp_reserve_aligned(ws, tagTableSize);
+            ZSTD_memset(ms->tagTable, 0, tagTableSize);
+            ms->hashSalt = 0;
+        }
+        {   /* Switch to 32-entry rows if searchLog is 5 (or more) */
+            U32 const rowLog = BOUNDED(4, cParams->searchLog, 6);
+            assert(cParams->hashLog >= rowLog);
+            ms->rowHashLog = cParams->hashLog - rowLog;
+        }
+    }
+
+    /* opt parser space */
+    if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) {
+        DEBUGLOG(4, "reserving optimal parser space");
+        ms->opt.litFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (1<<Litbits) * sizeof(unsigned));
+        ms->opt.litLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxLL+1) * sizeof(unsigned));
+        ms->opt.matchLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxML+1) * sizeof(unsigned));
+        ms->opt.offCodeFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxOff+1) * sizeof(unsigned));
+        ms->opt.matchTable = (ZSTD_match_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t));
+        ms->opt.priceTable = (ZSTD_optimal_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
+    }
+
+    ms->cParams = *cParams;
+
+    RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
+                    "failed a workspace allocation in ZSTD_reset_matchState");
+    return 0;
+}
+
+/* ZSTD_indexTooCloseToMax() :
+ * minor optimization : prefer memset() rather than reduceIndex()
+ * which is measurably slow in some circumstances (reported for Visual Studio).
+ * Works when re-using a context for a lot of smallish inputs :
+ * if all inputs are smaller than ZSTD_INDEXOVERFLOW_MARGIN,
+ * memset() will be triggered before reduceIndex().
+ */
+#define ZSTD_INDEXOVERFLOW_MARGIN (16 MB)
+static int ZSTD_indexTooCloseToMax(ZSTD_window_t w)
+{
+    return (size_t)(w.nextSrc - w.base) > (ZSTD_CURRENT_MAX - ZSTD_INDEXOVERFLOW_MARGIN);
+}
+
+/** ZSTD_dictTooBig():
+ * When dictionaries are larger than ZSTD_CHUNKSIZE_MAX they can't be loaded in
+ * one go generically. So we ensure that in that case we reset the tables to zero,
+ * so that we can load as much of the dictionary as possible.
+ */
+static int ZSTD_dictTooBig(size_t const loadedDictSize)
+{
+    return loadedDictSize > ZSTD_CHUNKSIZE_MAX;
+}
+
+/*! ZSTD_resetCCtx_internal() :
+ * @param loadedDictSize The size of the dictionary to be loaded
+ * into the context, if any. If no dictionary is used, or the
+ * dictionary is being attached / copied, then pass 0.
+ * note : `params` are assumed fully validated at this stage.
+ */
+static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
+                                      ZSTD_CCtx_params const* params,
+                                      U64 const pledgedSrcSize,
+                                      size_t const loadedDictSize,
+                                      ZSTD_compResetPolicy_e const crp,
+                                      ZSTD_buffered_policy_e const zbuff)
+{
+    ZSTD_cwksp* const ws = &zc->workspace;
+    DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u, useRowMatchFinder=%d useBlockSplitter=%d",
+                (U32)pledgedSrcSize, params->cParams.windowLog, (int)params->useRowMatchFinder, (int)params->useBlockSplitter);
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
+
+    zc->isFirstBlock = 1;
+
+    /* Set applied params early so we can modify them for LDM,
+     * and point params at the applied params.
+     */
+    zc->appliedParams = *params;
+    params = &zc->appliedParams;
+
+    assert(params->useRowMatchFinder != ZSTD_ps_auto);
+    assert(params->useBlockSplitter != ZSTD_ps_auto);
+    assert(params->ldmParams.enableLdm != ZSTD_ps_auto);
+    assert(params->maxBlockSize != 0);
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+        /* Adjust long distance matching parameters */
+        ZSTD_ldm_adjustParameters(&zc->appliedParams.ldmParams, &params->cParams);
+        assert(params->ldmParams.hashLog >= params->ldmParams.bucketSizeLog);
+        assert(params->ldmParams.hashRateLog < 32);
+    }
+
+    {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params->cParams.windowLog), pledgedSrcSize));
+        size_t const blockSize = MIN(params->maxBlockSize, windowSize);
+        size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, params->cParams.minMatch, params->useSequenceProducer);
+        size_t const buffOutSize = (zbuff == ZSTDb_buffered && params->outBufferMode == ZSTD_bm_buffered)
+                ? ZSTD_compressBound(blockSize) + 1
+                : 0;
+        size_t const buffInSize = (zbuff == ZSTDb_buffered && params->inBufferMode == ZSTD_bm_buffered)
+                ? windowSize + blockSize
+                : 0;
+        size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params->ldmParams, blockSize);
+
+        int const indexTooClose = ZSTD_indexTooCloseToMax(zc->blockState.matchState.window);
+        int const dictTooBig = ZSTD_dictTooBig(loadedDictSize);
+        ZSTD_indexResetPolicy_e needsIndexReset =
+            (indexTooClose || dictTooBig || !zc->initialized) ? ZSTDirp_reset : ZSTDirp_continue;
+
+        size_t const neededSpace =
+            ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+                &params->cParams, &params->ldmParams, zc->staticSize != 0, params->useRowMatchFinder,
+                buffInSize, buffOutSize, pledgedSrcSize, params->useSequenceProducer, params->maxBlockSize);
+        int resizeWorkspace;
+
+        FORWARD_IF_ERROR(neededSpace, "cctx size estimate failed!");
+
+        if (!zc->staticSize) ZSTD_cwksp_bump_oversized_duration(ws, 0);
+
+        {   /* Check if workspace is large enough, alloc a new one if needed */
+            int const workspaceTooSmall = ZSTD_cwksp_sizeof(ws) < neededSpace;
+            int const workspaceWasteful = ZSTD_cwksp_check_wasteful(ws, neededSpace);
+            resizeWorkspace = workspaceTooSmall || workspaceWasteful;
+            DEBUGLOG(4, "Need %zu B workspace", neededSpace);
+            DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
+
+            if (resizeWorkspace) {
+                DEBUGLOG(4, "Resize workspaceSize from %zuKB to %zuKB",
+                            ZSTD_cwksp_sizeof(ws) >> 10,
+                            neededSpace >> 10);
+
+                RETURN_ERROR_IF(zc->staticSize, memory_allocation, "static cctx : no resize");
+
+                needsIndexReset = ZSTDirp_reset;
+
+                ZSTD_cwksp_free(ws, zc->customMem);
+                FORWARD_IF_ERROR(ZSTD_cwksp_create(ws, neededSpace, zc->customMem), "");
+
+                DEBUGLOG(5, "reserving object space");
+                /* Statically sized space.
+                 * entropyWorkspace never moves,
+                 * though prev/next block swap places */
+                assert(ZSTD_cwksp_check_available(ws, 2 * sizeof(ZSTD_compressedBlockState_t)));
+                zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
+                RETURN_ERROR_IF(zc->blockState.prevCBlock == NULL, memory_allocation, "couldn't allocate prevCBlock");
+                zc->blockState.nextCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
+                RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate nextCBlock");
+                zc->entropyWorkspace = (U32*) ZSTD_cwksp_reserve_object(ws, ENTROPY_WORKSPACE_SIZE);
+                RETURN_ERROR_IF(zc->entropyWorkspace == NULL, memory_allocation, "couldn't allocate entropyWorkspace");
+        }   }
+
+        ZSTD_cwksp_clear(ws);
+
+        /* init params */
+        zc->blockState.matchState.cParams = params->cParams;
+        zc->blockState.matchState.prefetchCDictTables = params->prefetchCDictTables == ZSTD_ps_enable;
+        zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+        zc->consumedSrcSize = 0;
+        zc->producedCSize = 0;
+        if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+            zc->appliedParams.fParams.contentSizeFlag = 0;
+        DEBUGLOG(4, "pledged content size : %u ; flag : %u",
+            (unsigned)pledgedSrcSize, zc->appliedParams.fParams.contentSizeFlag);
+        zc->blockSize = blockSize;
+
+        XXH64_reset(&zc->xxhState, 0);
+        zc->stage = ZSTDcs_init;
+        zc->dictID = 0;
+        zc->dictContentSize = 0;
+
+        ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
+
+        FORWARD_IF_ERROR(ZSTD_reset_matchState(
+                &zc->blockState.matchState,
+                ws,
+                &params->cParams,
+                params->useRowMatchFinder,
+                crp,
+                needsIndexReset,
+                ZSTD_resetTarget_CCtx), "");
+
+        zc->seqStore.sequencesStart = (seqDef*)ZSTD_cwksp_reserve_aligned(ws, maxNbSeq * sizeof(seqDef));
+
+        /* ldm hash table */
+        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+            /* TODO: avoid memset? */
+            size_t const ldmHSize = ((size_t)1) << params->ldmParams.hashLog;
+            zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
+            ZSTD_memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
+            zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned(ws, maxNbLdmSeq * sizeof(rawSeq));
+            zc->maxNbLdmSequences = maxNbLdmSeq;
+
+            ZSTD_window_init(&zc->ldmState.window);
+            zc->ldmState.loadedDictEnd = 0;
+        }
+
+        /* reserve space for block-level external sequences */
+        if (params->useSequenceProducer) {
+            size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+            zc->externalMatchCtx.seqBufferCapacity = maxNbExternalSeq;
+            zc->externalMatchCtx.seqBuffer =
+                (ZSTD_Sequence*)ZSTD_cwksp_reserve_aligned(ws, maxNbExternalSeq * sizeof(ZSTD_Sequence));
+        }
+
+        /* buffers */
+
+        /* ZSTD_wildcopy() is used to copy into the literals buffer,
+         * so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes.
+         */
+        zc->seqStore.litStart = ZSTD_cwksp_reserve_buffer(ws, blockSize + WILDCOPY_OVERLENGTH);
+        zc->seqStore.maxNbLit = blockSize;
+
+        zc->bufferedPolicy = zbuff;
+        zc->inBuffSize = buffInSize;
+        zc->inBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffInSize);
+        zc->outBuffSize = buffOutSize;
+        zc->outBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffOutSize);
+
+        /* ldm bucketOffsets table */
+        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+            /* TODO: avoid memset? */
+            size_t const numBuckets =
+                  ((size_t)1) << (params->ldmParams.hashLog -
+                                  params->ldmParams.bucketSizeLog);
+            zc->ldmState.bucketOffsets = ZSTD_cwksp_reserve_buffer(ws, numBuckets);
+            ZSTD_memset(zc->ldmState.bucketOffsets, 0, numBuckets);
+        }
+
+        /* sequences storage */
+        ZSTD_referenceExternalSequences(zc, NULL, 0);
+        zc->seqStore.maxNbSeq = maxNbSeq;
+        zc->seqStore.llCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+        zc->seqStore.mlCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+        zc->seqStore.ofCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+
+        DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
+        assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace));
+
+        zc->initialized = 1;
+
+        return 0;
+    }
+}
+
+/* ZSTD_invalidateRepCodes() :
+ * ensures next compression will not use repcodes from previous block.
+ * Note : only works with regular variant;
+ *        do not use with extDict variant ! */
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
+    int i;
+    for (i=0; i<ZSTD_REP_NUM; i++) cctx->blockState.prevCBlock->rep[i] = 0;
+    assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
+}
+
+/* These are the approximate sizes for each strategy past which copying the
+ * dictionary tables into the working context is faster than using them
+ * in-place.
+ */
+static const size_t attachDictSizeCutoffs[ZSTD_STRATEGY_MAX+1] = {
+    8 KB,  /* unused */
+    8 KB,  /* ZSTD_fast */
+    16 KB, /* ZSTD_dfast */
+    32 KB, /* ZSTD_greedy */
+    32 KB, /* ZSTD_lazy */
+    32 KB, /* ZSTD_lazy2 */
+    32 KB, /* ZSTD_btlazy2 */
+    32 KB, /* ZSTD_btopt */
+    8 KB,  /* ZSTD_btultra */
+    8 KB   /* ZSTD_btultra2 */
+};
+
+static int ZSTD_shouldAttachDict(const ZSTD_CDict* cdict,
+                                 const ZSTD_CCtx_params* params,
+                                 U64 pledgedSrcSize)
+{
+    size_t cutoff = attachDictSizeCutoffs[cdict->matchState.cParams.strategy];
+    int const dedicatedDictSearch = cdict->matchState.dedicatedDictSearch;
+    return dedicatedDictSearch
+        || ( ( pledgedSrcSize <= cutoff
+            || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+            || params->attachDictPref == ZSTD_dictForceAttach )
+          && params->attachDictPref != ZSTD_dictForceCopy
+          && !params->forceWindow ); /* dictMatchState isn't correctly
+                                      * handled in _enforceMaxDist */
+}
+
+static size_t
+ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
+                        const ZSTD_CDict* cdict,
+                        ZSTD_CCtx_params params,
+                        U64 pledgedSrcSize,
+                        ZSTD_buffered_policy_e zbuff)
+{
+    DEBUGLOG(4, "ZSTD_resetCCtx_byAttachingCDict() pledgedSrcSize=%llu",
+                (unsigned long long)pledgedSrcSize);
+    {
+        ZSTD_compressionParameters adjusted_cdict_cParams = cdict->matchState.cParams;
+        unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
+        /* Resize working context table params for input only, since the dict
+         * has its own tables. */
+        /* pledgedSrcSize == 0 means 0! */
+
+        if (cdict->matchState.dedicatedDictSearch) {
+            ZSTD_dedicatedDictSearch_revertCParams(&adjusted_cdict_cParams);
+        }
+
+        params.cParams = ZSTD_adjustCParams_internal(adjusted_cdict_cParams, pledgedSrcSize,
+                                                     cdict->dictContentSize, ZSTD_cpm_attachDict,
+                                                     params.useRowMatchFinder);
+        params.cParams.windowLog = windowLog;
+        params.useRowMatchFinder = cdict->useRowMatchFinder;    /* cdict overrides */
+        FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, &params, pledgedSrcSize,
+                                                 /* loadedDictSize */ 0,
+                                                 ZSTDcrp_makeClean, zbuff), "");
+        assert(cctx->appliedParams.cParams.strategy == adjusted_cdict_cParams.strategy);
+    }
+
+    {   const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
+                                  - cdict->matchState.window.base);
+        const U32 cdictLen = cdictEnd - cdict->matchState.window.dictLimit;
+        if (cdictLen == 0) {
+            /* don't even attach dictionaries with no contents */
+            DEBUGLOG(4, "skipping attaching empty dictionary");
+        } else {
+            DEBUGLOG(4, "attaching dictionary into context");
+            cctx->blockState.matchState.dictMatchState = &cdict->matchState;
+
+            /* prep working match state so dict matches never have negative indices
+             * when they are translated to the working context's index space. */
+            if (cctx->blockState.matchState.window.dictLimit < cdictEnd) {
+                cctx->blockState.matchState.window.nextSrc =
+                    cctx->blockState.matchState.window.base + cdictEnd;
+                ZSTD_window_clear(&cctx->blockState.matchState.window);
+            }
+            /* loadedDictEnd is expressed within the referential of the active context */
+            cctx->blockState.matchState.loadedDictEnd = cctx->blockState.matchState.window.dictLimit;
+    }   }
+
+    cctx->dictID = cdict->dictID;
+    cctx->dictContentSize = cdict->dictContentSize;
+
+    /* copy block state */
+    ZSTD_memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
+}
+
+static void ZSTD_copyCDictTableIntoCCtx(U32* dst, U32 const* src, size_t tableSize,
+                                        ZSTD_compressionParameters const* cParams) {
+    if (ZSTD_CDictIndicesAreTagged(cParams)){
+        /* Remove tags from the CDict table if they are present.
+         * See docs on "short cache" in zstd_compress_internal.h for context. */
+        size_t i;
+        for (i = 0; i < tableSize; i++) {
+            U32 const taggedIndex = src[i];
+            U32 const index = taggedIndex >> ZSTD_SHORT_CACHE_TAG_BITS;
+            dst[i] = index;
+        }
+    } else {
+        ZSTD_memcpy(dst, src, tableSize * sizeof(U32));
+    }
+}
+
+static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            ZSTD_CCtx_params params,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+    const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
+
+    assert(!cdict->matchState.dedicatedDictSearch);
+    DEBUGLOG(4, "ZSTD_resetCCtx_byCopyingCDict() pledgedSrcSize=%llu",
+                (unsigned long long)pledgedSrcSize);
+
+    {   unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
+        /* Copy only compression parameters related to tables. */
+        params.cParams = *cdict_cParams;
+        params.cParams.windowLog = windowLog;
+        params.useRowMatchFinder = cdict->useRowMatchFinder;
+        FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, &params, pledgedSrcSize,
+                                                 /* loadedDictSize */ 0,
+                                                 ZSTDcrp_leaveDirty, zbuff), "");
+        assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
+        assert(cctx->appliedParams.cParams.hashLog == cdict_cParams->hashLog);
+        assert(cctx->appliedParams.cParams.chainLog == cdict_cParams->chainLog);
+    }
+
+    ZSTD_cwksp_mark_tables_dirty(&cctx->workspace);
+    assert(params.useRowMatchFinder != ZSTD_ps_auto);
+
+    /* copy tables */
+    {   size_t const chainSize = ZSTD_allocateChainTable(cdict_cParams->strategy, cdict->useRowMatchFinder, 0 /* DDS guaranteed disabled */)
+                                                            ? ((size_t)1 << cdict_cParams->chainLog)
+                                                            : 0;
+        size_t const hSize =  (size_t)1 << cdict_cParams->hashLog;
+
+        ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.hashTable,
+                                cdict->matchState.hashTable,
+                                hSize, cdict_cParams);
+
+        /* Do not copy cdict's chainTable if cctx has parameters such that it would not use chainTable */
+        if (ZSTD_allocateChainTable(cctx->appliedParams.cParams.strategy, cctx->appliedParams.useRowMatchFinder, 0 /* forDDSDict */)) {
+            ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.chainTable,
+                                    cdict->matchState.chainTable,
+                                    chainSize, cdict_cParams);
+        }
+        /* copy tag table */
+        if (ZSTD_rowMatchFinderUsed(cdict_cParams->strategy, cdict->useRowMatchFinder)) {
+            size_t const tagTableSize = hSize;
+            ZSTD_memcpy(cctx->blockState.matchState.tagTable,
+                        cdict->matchState.tagTable,
+                        tagTableSize);
+            cctx->blockState.matchState.hashSalt = cdict->matchState.hashSalt;
+        }
+    }
+
+    /* Zero the hashTable3, since the cdict never fills it */
+    {   int const h3log = cctx->blockState.matchState.hashLog3;
+        size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
+        assert(cdict->matchState.hashLog3 == 0);
+        ZSTD_memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
+    }
+
+    ZSTD_cwksp_mark_tables_clean(&cctx->workspace);
+
+    /* copy dictionary offsets */
+    {   ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
+        ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+
+    cctx->dictID = cdict->dictID;
+    cctx->dictContentSize = cdict->dictContentSize;
+
+    /* copy block state */
+    ZSTD_memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
+}
+
+/* We have a choice between copying the dictionary context into the working
+ * context, or referencing the dictionary context from the working context
+ * in-place. We decide here which strategy to use. */
+static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            const ZSTD_CCtx_params* params,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+
+    DEBUGLOG(4, "ZSTD_resetCCtx_usingCDict (pledgedSrcSize=%u)",
+                (unsigned)pledgedSrcSize);
+
+    if (ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize)) {
+        return ZSTD_resetCCtx_byAttachingCDict(
+            cctx, cdict, *params, pledgedSrcSize, zbuff);
+    } else {
+        return ZSTD_resetCCtx_byCopyingCDict(
+            cctx, cdict, *params, pledgedSrcSize, zbuff);
+    }
+}
+
+/*! ZSTD_copyCCtx_internal() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  The "context", in this case, refers to the hash and chain tables,
+ *  entropy tables, and dictionary references.
+ * `windowLog` value is enforced if != 0, otherwise value is copied from srcCCtx.
+ * @return : 0, or an error code */
+static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
+                            const ZSTD_CCtx* srcCCtx,
+                            ZSTD_frameParameters fParams,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+    RETURN_ERROR_IF(srcCCtx->stage!=ZSTDcs_init, stage_wrong,
+                    "Can't copy a ctx that's not in init stage.");
+    DEBUGLOG(5, "ZSTD_copyCCtx_internal");
+    ZSTD_memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
+    {   ZSTD_CCtx_params params = dstCCtx->requestedParams;
+        /* Copy only compression parameters related to tables. */
+        params.cParams = srcCCtx->appliedParams.cParams;
+        assert(srcCCtx->appliedParams.useRowMatchFinder != ZSTD_ps_auto);
+        assert(srcCCtx->appliedParams.useBlockSplitter != ZSTD_ps_auto);
+        assert(srcCCtx->appliedParams.ldmParams.enableLdm != ZSTD_ps_auto);
+        params.useRowMatchFinder = srcCCtx->appliedParams.useRowMatchFinder;
+        params.useBlockSplitter = srcCCtx->appliedParams.useBlockSplitter;
+        params.ldmParams = srcCCtx->appliedParams.ldmParams;
+        params.fParams = fParams;
+        params.maxBlockSize = srcCCtx->appliedParams.maxBlockSize;
+        ZSTD_resetCCtx_internal(dstCCtx, &params, pledgedSrcSize,
+                                /* loadedDictSize */ 0,
+                                ZSTDcrp_leaveDirty, zbuff);
+        assert(dstCCtx->appliedParams.cParams.windowLog == srcCCtx->appliedParams.cParams.windowLog);
+        assert(dstCCtx->appliedParams.cParams.strategy == srcCCtx->appliedParams.cParams.strategy);
+        assert(dstCCtx->appliedParams.cParams.hashLog == srcCCtx->appliedParams.cParams.hashLog);
+        assert(dstCCtx->appliedParams.cParams.chainLog == srcCCtx->appliedParams.cParams.chainLog);
+        assert(dstCCtx->blockState.matchState.hashLog3 == srcCCtx->blockState.matchState.hashLog3);
+    }
+
+    ZSTD_cwksp_mark_tables_dirty(&dstCCtx->workspace);
+
+    /* copy tables */
+    {   size_t const chainSize = ZSTD_allocateChainTable(srcCCtx->appliedParams.cParams.strategy,
+                                                         srcCCtx->appliedParams.useRowMatchFinder,
+                                                         0 /* forDDSDict */)
+                                    ? ((size_t)1 << srcCCtx->appliedParams.cParams.chainLog)
+                                    : 0;
+        size_t const hSize =  (size_t)1 << srcCCtx->appliedParams.cParams.hashLog;
+        int const h3log = srcCCtx->blockState.matchState.hashLog3;
+        size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
+
+        ZSTD_memcpy(dstCCtx->blockState.matchState.hashTable,
+               srcCCtx->blockState.matchState.hashTable,
+               hSize * sizeof(U32));
+        ZSTD_memcpy(dstCCtx->blockState.matchState.chainTable,
+               srcCCtx->blockState.matchState.chainTable,
+               chainSize * sizeof(U32));
+        ZSTD_memcpy(dstCCtx->blockState.matchState.hashTable3,
+               srcCCtx->blockState.matchState.hashTable3,
+               h3Size * sizeof(U32));
+    }
+
+    ZSTD_cwksp_mark_tables_clean(&dstCCtx->workspace);
+
+    /* copy dictionary offsets */
+    {
+        const ZSTD_matchState_t* srcMatchState = &srcCCtx->blockState.matchState;
+        ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+    dstCCtx->dictID = srcCCtx->dictID;
+    dstCCtx->dictContentSize = srcCCtx->dictContentSize;
+
+    /* copy block state */
+    ZSTD_memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
+
+    return 0;
+}
+
+/*! ZSTD_copyCCtx() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  pledgedSrcSize==0 means "unknown".
+*   @return : 0, or an error code */
+size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long long pledgedSrcSize)
+{
+    ZSTD_frameParameters fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    ZSTD_buffered_policy_e const zbuff = srcCCtx->bufferedPolicy;
+    ZSTD_STATIC_ASSERT((U32)ZSTDb_buffered==1);
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    fParams.contentSizeFlag = (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN);
+
+    return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx,
+                                fParams, pledgedSrcSize,
+                                zbuff);
+}
+
+
+#define ZSTD_ROWSIZE 16
+/*! ZSTD_reduceTable() :
+ *  reduce table indexes by `reducerValue`, or squash to zero.
+ *  PreserveMark preserves "unsorted mark" for btlazy2 strategy.
+ *  It must be set to a clear 0/1 value, to remove branch during inlining.
+ *  Presume table size is a multiple of ZSTD_ROWSIZE
+ *  to help auto-vectorization */
+FORCE_INLINE_TEMPLATE void
+ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerValue, int const preserveMark)
+{
+    int const nbRows = (int)size / ZSTD_ROWSIZE;
+    int cellNb = 0;
+    int rowNb;
+    /* Protect special index values < ZSTD_WINDOW_START_INDEX. */
+    U32 const reducerThreshold = reducerValue + ZSTD_WINDOW_START_INDEX;
+    assert((size & (ZSTD_ROWSIZE-1)) == 0);  /* multiple of ZSTD_ROWSIZE */
+    assert(size < (1U<<31));   /* can be casted to int */
+
+#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+    /* To validate that the table re-use logic is sound, and that we don't
+     * access table space that we haven't cleaned, we re-"poison" the table
+     * space every time we mark it dirty.
+     *
+     * This function however is intended to operate on those dirty tables and
+     * re-clean them. So when this function is used correctly, we can unpoison
+     * the memory it operated on. This introduces a blind spot though, since
+     * if we now try to operate on __actually__ poisoned memory, we will not
+     * detect that. */
+    __msan_unpoison(table, size * sizeof(U32));
+#endif
+
+    for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
+        int column;
+        for (column=0; column<ZSTD_ROWSIZE; column++) {
+            U32 newVal;
+            if (preserveMark && table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) {
+                /* This write is pointless, but is required(?) for the compiler
+                 * to auto-vectorize the loop. */
+                newVal = ZSTD_DUBT_UNSORTED_MARK;
+            } else if (table[cellNb] < reducerThreshold) {
+                newVal = 0;
+            } else {
+                newVal = table[cellNb] - reducerValue;
+            }
+            table[cellNb] = newVal;
+            cellNb++;
+    }   }
+}
+
+static void ZSTD_reduceTable(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 0);
+}
+
+static void ZSTD_reduceTable_btlazy2(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 1);
+}
+
+/*! ZSTD_reduceIndex() :
+*   rescale all indexes to avoid future overflow (indexes are U32) */
+static void ZSTD_reduceIndex (ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const U32 reducerValue)
+{
+    {   U32 const hSize = (U32)1 << params->cParams.hashLog;
+        ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
+    }
+
+    if (ZSTD_allocateChainTable(params->cParams.strategy, params->useRowMatchFinder, (U32)ms->dedicatedDictSearch)) {
+        U32 const chainSize = (U32)1 << params->cParams.chainLog;
+        if (params->cParams.strategy == ZSTD_btlazy2)
+            ZSTD_reduceTable_btlazy2(ms->chainTable, chainSize, reducerValue);
+        else
+            ZSTD_reduceTable(ms->chainTable, chainSize, reducerValue);
+    }
+
+    if (ms->hashLog3) {
+        U32 const h3Size = (U32)1 << ms->hashLog3;
+        ZSTD_reduceTable(ms->hashTable3, h3Size, reducerValue);
+    }
+}
+
+
+/*-*******************************************************
+*  Block entropic compression
+*********************************************************/
+
+/* See doc/zstd_compression_format.md for detailed format description */
+
+int ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
+{
+    const seqDef* const sequences = seqStorePtr->sequencesStart;
+    BYTE* const llCodeTable = seqStorePtr->llCode;
+    BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    U32 u;
+    int longOffsets = 0;
+    assert(nbSeq <= seqStorePtr->maxNbSeq);
+    for (u=0; u<nbSeq; u++) {
+        U32 const llv = sequences[u].litLength;
+        U32 const ofCode = ZSTD_highbit32(sequences[u].offBase);
+        U32 const mlv = sequences[u].mlBase;
+        llCodeTable[u] = (BYTE)ZSTD_LLcode(llv);
+        ofCodeTable[u] = (BYTE)ofCode;
+        mlCodeTable[u] = (BYTE)ZSTD_MLcode(mlv);
+        assert(!(MEM_64bits() && ofCode >= STREAM_ACCUMULATOR_MIN));
+        if (MEM_32bits() && ofCode >= STREAM_ACCUMULATOR_MIN)
+            longOffsets = 1;
+    }
+    if (seqStorePtr->longLengthType==ZSTD_llt_literalLength)
+        llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
+    if (seqStorePtr->longLengthType==ZSTD_llt_matchLength)
+        mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
+    return longOffsets;
+}
+
+/* ZSTD_useTargetCBlockSize():
+ * Returns if target compressed block size param is being used.
+ * If used, compression will do best effort to make a compressed block size to be around targetCBlockSize.
+ * Returns 1 if true, 0 otherwise. */
+static int ZSTD_useTargetCBlockSize(const ZSTD_CCtx_params* cctxParams)
+{
+    DEBUGLOG(5, "ZSTD_useTargetCBlockSize (targetCBlockSize=%zu)", cctxParams->targetCBlockSize);
+    return (cctxParams->targetCBlockSize != 0);
+}
+
+/* ZSTD_blockSplitterEnabled():
+ * Returns if block splitting param is being used
+ * If used, compression will do best effort to split a block in order to improve compression ratio.
+ * At the time this function is called, the parameter must be finalized.
+ * Returns 1 if true, 0 otherwise. */
+static int ZSTD_blockSplitterEnabled(ZSTD_CCtx_params* cctxParams)
+{
+    DEBUGLOG(5, "ZSTD_blockSplitterEnabled (useBlockSplitter=%d)", cctxParams->useBlockSplitter);
+    assert(cctxParams->useBlockSplitter != ZSTD_ps_auto);
+    return (cctxParams->useBlockSplitter == ZSTD_ps_enable);
+}
+
+/* Type returned by ZSTD_buildSequencesStatistics containing finalized symbol encoding types
+ * and size of the sequences statistics
+ */
+typedef struct {
+    U32 LLtype;
+    U32 Offtype;
+    U32 MLtype;
+    size_t size;
+    size_t lastCountSize; /* Accounts for bug in 1.3.4. More detail in ZSTD_entropyCompressSeqStore_internal() */
+    int longOffsets;
+} ZSTD_symbolEncodingTypeStats_t;
+
+/* ZSTD_buildSequencesStatistics():
+ * Returns a ZSTD_symbolEncodingTypeStats_t, or a zstd error code in the `size` field.
+ * Modifies `nextEntropy` to have the appropriate values as a side effect.
+ * nbSeq must be greater than 0.
+ *
+ * entropyWkspSize must be of size at least ENTROPY_WORKSPACE_SIZE - (MaxSeq + 1)*sizeof(U32)
+ */
+static ZSTD_symbolEncodingTypeStats_t
+ZSTD_buildSequencesStatistics(
+                const seqStore_t* seqStorePtr, size_t nbSeq,
+                const ZSTD_fseCTables_t* prevEntropy, ZSTD_fseCTables_t* nextEntropy,
+                      BYTE* dst, const BYTE* const dstEnd,
+                      ZSTD_strategy strategy, unsigned* countWorkspace,
+                      void* entropyWorkspace, size_t entropyWkspSize)
+{
+    BYTE* const ostart = dst;
+    const BYTE* const oend = dstEnd;
+    BYTE* op = ostart;
+    FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    ZSTD_symbolEncodingTypeStats_t stats;
+
+    stats.lastCountSize = 0;
+    /* convert length/distances into codes */
+    stats.longOffsets = ZSTD_seqToCodes(seqStorePtr);
+    assert(op <= oend);
+    assert(nbSeq != 0); /* ZSTD_selectEncodingType() divides by nbSeq */
+    /* build CTable for Literal Lengths */
+    {   unsigned max = MaxLL;
+        size_t const mostFrequent = HIST_countFast_wksp(countWorkspace, &max, llCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);   /* can't fail */
+        DEBUGLOG(5, "Building LL table");
+        nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
+        stats.LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode,
+                                        countWorkspace, max, mostFrequent, nbSeq,
+                                        LLFSELog, prevEntropy->litlengthCTable,
+                                        LL_defaultNorm, LL_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(set_basic < set_compressed && set_rle < set_compressed);
+        assert(!(stats.LLtype < set_compressed && nextEntropy->litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_LitLength, LLFSELog, (symbolEncodingType_e)stats.LLtype,
+                countWorkspace, max, llCodeTable, nbSeq,
+                LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                prevEntropy->litlengthCTable,
+                sizeof(prevEntropy->litlengthCTable),
+                entropyWorkspace, entropyWkspSize);
+            if (ZSTD_isError(countSize)) {
+                DEBUGLOG(3, "ZSTD_buildCTable for LitLens failed");
+                stats.size = countSize;
+                return stats;
+            }
+            if (stats.LLtype == set_compressed)
+                stats.lastCountSize = countSize;
+            op += countSize;
+            assert(op <= oend);
+    }   }
+    /* build CTable for Offsets */
+    {   unsigned max = MaxOff;
+        size_t const mostFrequent = HIST_countFast_wksp(
+            countWorkspace, &max, ofCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);  /* can't fail */
+        /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
+        ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
+        DEBUGLOG(5, "Building OF table");
+        nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
+        stats.Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode,
+                                        countWorkspace, max, mostFrequent, nbSeq,
+                                        OffFSELog, prevEntropy->offcodeCTable,
+                                        OF_defaultNorm, OF_defaultNormLog,
+                                        defaultPolicy, strategy);
+        assert(!(stats.Offtype < set_compressed && nextEntropy->offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)stats.Offtype,
+                countWorkspace, max, ofCodeTable, nbSeq,
+                OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                prevEntropy->offcodeCTable,
+                sizeof(prevEntropy->offcodeCTable),
+                entropyWorkspace, entropyWkspSize);
+            if (ZSTD_isError(countSize)) {
+                DEBUGLOG(3, "ZSTD_buildCTable for Offsets failed");
+                stats.size = countSize;
+                return stats;
+            }
+            if (stats.Offtype == set_compressed)
+                stats.lastCountSize = countSize;
+            op += countSize;
+            assert(op <= oend);
+    }   }
+    /* build CTable for MatchLengths */
+    {   unsigned max = MaxML;
+        size_t const mostFrequent = HIST_countFast_wksp(
+            countWorkspace, &max, mlCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);   /* can't fail */
+        DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
+        nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
+        stats.MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode,
+                                        countWorkspace, max, mostFrequent, nbSeq,
+                                        MLFSELog, prevEntropy->matchlengthCTable,
+                                        ML_defaultNorm, ML_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(!(stats.MLtype < set_compressed && nextEntropy->matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_MatchLength, MLFSELog, (symbolEncodingType_e)stats.MLtype,
+                countWorkspace, max, mlCodeTable, nbSeq,
+                ML_defaultNorm, ML_defaultNormLog, MaxML,
+                prevEntropy->matchlengthCTable,
+                sizeof(prevEntropy->matchlengthCTable),
+                entropyWorkspace, entropyWkspSize);
+            if (ZSTD_isError(countSize)) {
+                DEBUGLOG(3, "ZSTD_buildCTable for MatchLengths failed");
+                stats.size = countSize;
+                return stats;
+            }
+            if (stats.MLtype == set_compressed)
+                stats.lastCountSize = countSize;
+            op += countSize;
+            assert(op <= oend);
+    }   }
+    stats.size = (size_t)(op-ostart);
+    return stats;
+}
+
+/* ZSTD_entropyCompressSeqStore_internal():
+ * compresses both literals and sequences
+ * Returns compressed size of block, or a zstd error.
+ */
+#define SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO 20
+MEM_STATIC size_t
+ZSTD_entropyCompressSeqStore_internal(
+                        const seqStore_t* seqStorePtr,
+                        const ZSTD_entropyCTables_t* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                        const ZSTD_CCtx_params* cctxParams,
+                              void* dst, size_t dstCapacity,
+                              void* entropyWorkspace, size_t entropyWkspSize,
+                        const int bmi2)
+{
+    ZSTD_strategy const strategy = cctxParams->cParams.strategy;
+    unsigned* count = (unsigned*)entropyWorkspace;
+    FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
+    const seqDef* const sequences = seqStorePtr->sequencesStart;
+    const size_t nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t lastCountSize;
+    int longOffsets = 0;
+
+    entropyWorkspace = count + (MaxSeq + 1);
+    entropyWkspSize -= (MaxSeq + 1) * sizeof(*count);
+
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore_internal (nbSeq=%zu, dstCapacity=%zu)", nbSeq, dstCapacity);
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+    assert(entropyWkspSize >= HUF_WORKSPACE_SIZE);
+
+    /* Compress literals */
+    {   const BYTE* const literals = seqStorePtr->litStart;
+        size_t const numSequences = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+        size_t const numLiterals = (size_t)(seqStorePtr->lit - seqStorePtr->litStart);
+        /* Base suspicion of uncompressibility on ratio of literals to sequences */
+        unsigned const suspectUncompressible = (numSequences == 0) || (numLiterals / numSequences >= SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO);
+        size_t const litSize = (size_t)(seqStorePtr->lit - literals);
+
+        size_t const cSize = ZSTD_compressLiterals(
+                                    op, dstCapacity,
+                                    literals, litSize,
+                                    entropyWorkspace, entropyWkspSize,
+                                    &prevEntropy->huf, &nextEntropy->huf,
+                                    cctxParams->cParams.strategy,
+                                    ZSTD_literalsCompressionIsDisabled(cctxParams),
+                                    suspectUncompressible, bmi2);
+        FORWARD_IF_ERROR(cSize, "ZSTD_compressLiterals failed");
+        assert(cSize <= dstCapacity);
+        op += cSize;
+    }
+
+    /* Sequences Header */
+    RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
+                    dstSize_tooSmall, "Can't fit seq hdr in output buf!");
+    if (nbSeq < 128) {
+        *op++ = (BYTE)nbSeq;
+    } else if (nbSeq < LONGNBSEQ) {
+        op[0] = (BYTE)((nbSeq>>8) + 0x80);
+        op[1] = (BYTE)nbSeq;
+        op+=2;
+    } else {
+        op[0]=0xFF;
+        MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ));
+        op+=3;
+    }
+    assert(op <= oend);
+    if (nbSeq==0) {
+        /* Copy the old tables over as if we repeated them */
+        ZSTD_memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
+        return (size_t)(op - ostart);
+    }
+    {   BYTE* const seqHead = op++;
+        /* build stats for sequences */
+        const ZSTD_symbolEncodingTypeStats_t stats =
+                ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
+                                             &prevEntropy->fse, &nextEntropy->fse,
+                                              op, oend,
+                                              strategy, count,
+                                              entropyWorkspace, entropyWkspSize);
+        FORWARD_IF_ERROR(stats.size, "ZSTD_buildSequencesStatistics failed!");
+        *seqHead = (BYTE)((stats.LLtype<<6) + (stats.Offtype<<4) + (stats.MLtype<<2));
+        lastCountSize = stats.lastCountSize;
+        op += stats.size;
+        longOffsets = stats.longOffsets;
+    }
+
+    {   size_t const bitstreamSize = ZSTD_encodeSequences(
+                                        op, (size_t)(oend - op),
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq,
+                                        longOffsets, bmi2);
+        FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
+        op += bitstreamSize;
+        assert(op <= oend);
+        /* zstd versions <= 1.3.4 mistakenly report corruption when
+         * FSE_readNCount() receives a buffer < 4 bytes.
+         * Fixed by https://github.com/facebook/zstd/pull/1146.
+         * This can happen when the last set_compressed table present is 2
+         * bytes and the bitstream is only one byte.
+         * In this exceedingly rare case, we will simply emit an uncompressed
+         * block, since it isn't worth optimizing.
+         */
+        if (lastCountSize && (lastCountSize + bitstreamSize) < 4) {
+            /* lastCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
+            assert(lastCountSize + bitstreamSize == 3);
+            DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
+                        "emitting an uncompressed block.");
+            return 0;
+        }
+    }
+
+    DEBUGLOG(5, "compressed block size : %u", (unsigned)(op - ostart));
+    return (size_t)(op - ostart);
+}
+
+MEM_STATIC size_t
+ZSTD_entropyCompressSeqStore(
+                    const seqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          void* dst, size_t dstCapacity,
+                          size_t srcSize,
+                          void* entropyWorkspace, size_t entropyWkspSize,
+                          int bmi2)
+{
+    size_t const cSize = ZSTD_entropyCompressSeqStore_internal(
+                            seqStorePtr, prevEntropy, nextEntropy, cctxParams,
+                            dst, dstCapacity,
+                            entropyWorkspace, entropyWkspSize, bmi2);
+    if (cSize == 0) return 0;
+    /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
+     * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
+     */
+    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity)) {
+        DEBUGLOG(4, "not enough dstCapacity (%zu) for ZSTD_entropyCompressSeqStore_internal()=> do not compress block", dstCapacity);
+        return 0;  /* block not compressed */
+    }
+    FORWARD_IF_ERROR(cSize, "ZSTD_entropyCompressSeqStore_internal failed");
+
+    /* Check compressibility */
+    {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
+        if (cSize >= maxCSize) return 0;  /* block not compressed */
+    }
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore() cSize: %zu", cSize);
+    /* libzstd decoder before  > v1.5.4 is not compatible with compressed blocks of size ZSTD_BLOCKSIZE_MAX exactly.
+     * This restriction is indirectly already fulfilled by respecting ZSTD_minGain() condition above.
+     */
+    assert(cSize < ZSTD_BLOCKSIZE_MAX);
+    return cSize;
+}
+
+/* ZSTD_selectBlockCompressor() :
+ * Not static, but internal use only (used by long distance matcher)
+ * assumption : strat is a valid strategy */
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e useRowMatchFinder, ZSTD_dictMode_e dictMode)
+{
+    static const ZSTD_blockCompressor blockCompressor[4][ZSTD_STRATEGY_MAX+1] = {
+        { ZSTD_compressBlock_fast  /* default for 0 */,
+          ZSTD_compressBlock_fast,
+          ZSTD_compressBlock_doubleFast,
+          ZSTD_compressBlock_greedy,
+          ZSTD_compressBlock_lazy,
+          ZSTD_compressBlock_lazy2,
+          ZSTD_compressBlock_btlazy2,
+          ZSTD_compressBlock_btopt,
+          ZSTD_compressBlock_btultra,
+          ZSTD_compressBlock_btultra2 },
+        { ZSTD_compressBlock_fast_extDict  /* default for 0 */,
+          ZSTD_compressBlock_fast_extDict,
+          ZSTD_compressBlock_doubleFast_extDict,
+          ZSTD_compressBlock_greedy_extDict,
+          ZSTD_compressBlock_lazy_extDict,
+          ZSTD_compressBlock_lazy2_extDict,
+          ZSTD_compressBlock_btlazy2_extDict,
+          ZSTD_compressBlock_btopt_extDict,
+          ZSTD_compressBlock_btultra_extDict,
+          ZSTD_compressBlock_btultra_extDict },
+        { ZSTD_compressBlock_fast_dictMatchState  /* default for 0 */,
+          ZSTD_compressBlock_fast_dictMatchState,
+          ZSTD_compressBlock_doubleFast_dictMatchState,
+          ZSTD_compressBlock_greedy_dictMatchState,
+          ZSTD_compressBlock_lazy_dictMatchState,
+          ZSTD_compressBlock_lazy2_dictMatchState,
+          ZSTD_compressBlock_btlazy2_dictMatchState,
+          ZSTD_compressBlock_btopt_dictMatchState,
+          ZSTD_compressBlock_btultra_dictMatchState,
+          ZSTD_compressBlock_btultra_dictMatchState },
+        { NULL  /* default for 0 */,
+          NULL,
+          NULL,
+          ZSTD_compressBlock_greedy_dedicatedDictSearch,
+          ZSTD_compressBlock_lazy_dedicatedDictSearch,
+          ZSTD_compressBlock_lazy2_dedicatedDictSearch,
+          NULL,
+          NULL,
+          NULL,
+          NULL }
+    };
+    ZSTD_blockCompressor selectedCompressor;
+    ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
+
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, strat));
+    DEBUGLOG(4, "Selected block compressor: dictMode=%d strat=%d rowMatchfinder=%d", (int)dictMode, (int)strat, (int)useRowMatchFinder);
+    if (ZSTD_rowMatchFinderUsed(strat, useRowMatchFinder)) {
+        static const ZSTD_blockCompressor rowBasedBlockCompressors[4][3] = {
+            { ZSTD_compressBlock_greedy_row,
+            ZSTD_compressBlock_lazy_row,
+            ZSTD_compressBlock_lazy2_row },
+            { ZSTD_compressBlock_greedy_extDict_row,
+            ZSTD_compressBlock_lazy_extDict_row,
+            ZSTD_compressBlock_lazy2_extDict_row },
+            { ZSTD_compressBlock_greedy_dictMatchState_row,
+            ZSTD_compressBlock_lazy_dictMatchState_row,
+            ZSTD_compressBlock_lazy2_dictMatchState_row },
+            { ZSTD_compressBlock_greedy_dedicatedDictSearch_row,
+            ZSTD_compressBlock_lazy_dedicatedDictSearch_row,
+            ZSTD_compressBlock_lazy2_dedicatedDictSearch_row }
+        };
+        DEBUGLOG(4, "Selecting a row-based matchfinder");
+        assert(useRowMatchFinder != ZSTD_ps_auto);
+        selectedCompressor = rowBasedBlockCompressors[(int)dictMode][(int)strat - (int)ZSTD_greedy];
+    } else {
+        selectedCompressor = blockCompressor[(int)dictMode][(int)strat];
+    }
+    assert(selectedCompressor != NULL);
+    return selectedCompressor;
+}
+
+static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
+                                   const BYTE* anchor, size_t lastLLSize)
+{
+    ZSTD_memcpy(seqStorePtr->lit, anchor, lastLLSize);
+    seqStorePtr->lit += lastLLSize;
+}
+
+void ZSTD_resetSeqStore(seqStore_t* ssPtr)
+{
+    ssPtr->lit = ssPtr->litStart;
+    ssPtr->sequences = ssPtr->sequencesStart;
+    ssPtr->longLengthType = ZSTD_llt_none;
+}
+
+/* ZSTD_postProcessSequenceProducerResult() :
+ * Validates and post-processes sequences obtained through the external matchfinder API:
+ *   - Checks whether nbExternalSeqs represents an error condition.
+ *   - Appends a block delimiter to outSeqs if one is not already present.
+ *     See zstd.h for context regarding block delimiters.
+ * Returns the number of sequences after post-processing, or an error code. */
+static size_t ZSTD_postProcessSequenceProducerResult(
+    ZSTD_Sequence* outSeqs, size_t nbExternalSeqs, size_t outSeqsCapacity, size_t srcSize
+) {
+    RETURN_ERROR_IF(
+        nbExternalSeqs > outSeqsCapacity,
+        sequenceProducer_failed,
+        "External sequence producer returned error code %lu",
+        (unsigned long)nbExternalSeqs
+    );
+
+    RETURN_ERROR_IF(
+        nbExternalSeqs == 0 && srcSize > 0,
+        sequenceProducer_failed,
+        "Got zero sequences from external sequence producer for a non-empty src buffer!"
+    );
+
+    if (srcSize == 0) {
+        ZSTD_memset(&outSeqs[0], 0, sizeof(ZSTD_Sequence));
+        return 1;
+    }
+
+    {
+        ZSTD_Sequence const lastSeq = outSeqs[nbExternalSeqs - 1];
+
+        /* We can return early if lastSeq is already a block delimiter. */
+        if (lastSeq.offset == 0 && lastSeq.matchLength == 0) {
+            return nbExternalSeqs;
+        }
+
+        /* This error condition is only possible if the external matchfinder
+         * produced an invalid parse, by definition of ZSTD_sequenceBound(). */
+        RETURN_ERROR_IF(
+            nbExternalSeqs == outSeqsCapacity,
+            sequenceProducer_failed,
+            "nbExternalSeqs == outSeqsCapacity but lastSeq is not a block delimiter!"
+        );
+
+        /* lastSeq is not a block delimiter, so we need to append one. */
+        ZSTD_memset(&outSeqs[nbExternalSeqs], 0, sizeof(ZSTD_Sequence));
+        return nbExternalSeqs + 1;
+    }
+}
+
+/* ZSTD_fastSequenceLengthSum() :
+ * Returns sum(litLen) + sum(matchLen) + lastLits for *seqBuf*.
+ * Similar to another function in zstd_compress.c (determine_blockSize),
+ * except it doesn't check for a block delimiter to end summation.
+ * Removing the early exit allows the compiler to auto-vectorize (https://godbolt.org/z/cY1cajz9P).
+ * This function can be deleted and replaced by determine_blockSize after we resolve issue #3456. */
+static size_t ZSTD_fastSequenceLengthSum(ZSTD_Sequence const* seqBuf, size_t seqBufSize) {
+    size_t matchLenSum, litLenSum, i;
+    matchLenSum = 0;
+    litLenSum = 0;
+    for (i = 0; i < seqBufSize; i++) {
+        litLenSum += seqBuf[i].litLength;
+        matchLenSum += seqBuf[i].matchLength;
+    }
+    return litLenSum + matchLenSum;
+}
+
+typedef enum { ZSTDbss_compress, ZSTDbss_noCompress } ZSTD_buildSeqStore_e;
+
+static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
+{
+    ZSTD_matchState_t* const ms = &zc->blockState.matchState;
+    DEBUGLOG(5, "ZSTD_buildSeqStore (srcSize=%zu)", srcSize);
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+    /* Assert that we have correctly flushed the ctx params into the ms's copy */
+    ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
+    /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+     * additional 1. We need to revisit and change this logic to be more consistent */
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
+        if (zc->appliedParams.cParams.strategy >= ZSTD_btopt) {
+            ZSTD_ldm_skipRawSeqStoreBytes(&zc->externSeqStore, srcSize);
+        } else {
+            ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.minMatch);
+        }
+        return ZSTDbss_noCompress; /* don't even attempt compression below a certain srcSize */
+    }
+    ZSTD_resetSeqStore(&(zc->seqStore));
+    /* required for optimal parser to read stats from dictionary */
+    ms->opt.symbolCosts = &zc->blockState.prevCBlock->entropy;
+    /* tell the optimal parser how we expect to compress literals */
+    ms->opt.literalCompressionMode = zc->appliedParams.literalCompressionMode;
+    /* a gap between an attached dict and the current window is not safe,
+     * they must remain adjacent,
+     * and when that stops being the case, the dict must be unset */
+    assert(ms->dictMatchState == NULL || ms->loadedDictEnd == ms->window.dictLimit);
+
+    /* limited update after a very long match */
+    {   const BYTE* const base = ms->window.base;
+        const BYTE* const istart = (const BYTE*)src;
+        const U32 curr = (U32)(istart-base);
+        if (sizeof(ptrdiff_t)==8) assert(istart - base < (ptrdiff_t)(U32)(-1));   /* ensure no overflow */
+        if (curr > ms->nextToUpdate + 384)
+            ms->nextToUpdate = curr - MIN(192, (U32)(curr - ms->nextToUpdate - 384));
+    }
+
+    /* select and store sequences */
+    {   ZSTD_dictMode_e const dictMode = ZSTD_matchState_dictMode(ms);
+        size_t lastLLSize;
+        {   int i;
+            for (i = 0; i < ZSTD_REP_NUM; ++i)
+                zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
+        }
+        if (zc->externSeqStore.pos < zc->externSeqStore.size) {
+            assert(zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_disable);
+
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                zc->appliedParams.useSequenceProducer,
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&zc->externSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       zc->appliedParams.useRowMatchFinder,
+                                       src, srcSize);
+            assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
+        } else if (zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
+            rawSeqStore_t ldmSeqStore = kNullRawSeqStore;
+
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                zc->appliedParams.useSequenceProducer,
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
+            ldmSeqStore.seq = zc->ldmSequences;
+            ldmSeqStore.capacity = zc->maxNbLdmSequences;
+            /* Updates ldmSeqStore.size */
+            FORWARD_IF_ERROR(ZSTD_ldm_generateSequences(&zc->ldmState, &ldmSeqStore,
+                                               &zc->appliedParams.ldmParams,
+                                               src, srcSize), "");
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&ldmSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       zc->appliedParams.useRowMatchFinder,
+                                       src, srcSize);
+            assert(ldmSeqStore.pos == ldmSeqStore.size);
+        } else if (zc->appliedParams.useSequenceProducer) {
+            assert(
+                zc->externalMatchCtx.seqBufferCapacity >= ZSTD_sequenceBound(srcSize)
+            );
+            assert(zc->externalMatchCtx.mFinder != NULL);
+
+            {   U32 const windowSize = (U32)1 << zc->appliedParams.cParams.windowLog;
+
+                size_t const nbExternalSeqs = (zc->externalMatchCtx.mFinder)(
+                    zc->externalMatchCtx.mState,
+                    zc->externalMatchCtx.seqBuffer,
+                    zc->externalMatchCtx.seqBufferCapacity,
+                    src, srcSize,
+                    NULL, 0,  /* dict and dictSize, currently not supported */
+                    zc->appliedParams.compressionLevel,
+                    windowSize
+                );
+
+                size_t const nbPostProcessedSeqs = ZSTD_postProcessSequenceProducerResult(
+                    zc->externalMatchCtx.seqBuffer,
+                    nbExternalSeqs,
+                    zc->externalMatchCtx.seqBufferCapacity,
+                    srcSize
+                );
+
+                /* Return early if there is no error, since we don't need to worry about last literals */
+                if (!ZSTD_isError(nbPostProcessedSeqs)) {
+                    ZSTD_sequencePosition seqPos = {0,0,0};
+                    size_t const seqLenSum = ZSTD_fastSequenceLengthSum(zc->externalMatchCtx.seqBuffer, nbPostProcessedSeqs);
+                    RETURN_ERROR_IF(seqLenSum > srcSize, externalSequences_invalid, "External sequences imply too large a block!");
+                    FORWARD_IF_ERROR(
+                        ZSTD_copySequencesToSeqStoreExplicitBlockDelim(
+                            zc, &seqPos,
+                            zc->externalMatchCtx.seqBuffer, nbPostProcessedSeqs,
+                            src, srcSize,
+                            zc->appliedParams.searchForExternalRepcodes
+                        ),
+                        "Failed to copy external sequences to seqStore!"
+                    );
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(5, "Copied %lu sequences from external sequence producer to internal seqStore.", (unsigned long)nbExternalSeqs);
+                    return ZSTDbss_compress;
+                }
+
+                /* Propagate the error if fallback is disabled */
+                if (!zc->appliedParams.enableMatchFinderFallback) {
+                    return nbPostProcessedSeqs;
+                }
+
+                /* Fallback to software matchfinder */
+                {   ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy,
+                                                                                            zc->appliedParams.useRowMatchFinder,
+                                                                                            dictMode);
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(
+                        5,
+                        "External sequence producer returned error code %lu. Falling back to internal parser.",
+                        (unsigned long)nbExternalSeqs
+                    );
+                    lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
+            }   }
+        } else {   /* not long range mode and no external matchfinder */
+            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy,
+                                                                                    zc->appliedParams.useRowMatchFinder,
+                                                                                    dictMode);
+            ms->ldmSeqStore = NULL;
+            lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
+        }
+        {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
+            ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
+    }   }
+    return ZSTDbss_compress;
+}
+
+static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
+{
+    const seqStore_t* seqStore = ZSTD_getSeqStore(zc);
+    const seqDef* seqStoreSeqs = seqStore->sequencesStart;
+    size_t seqStoreSeqSize = seqStore->sequences - seqStoreSeqs;
+    size_t seqStoreLiteralsSize = (size_t)(seqStore->lit - seqStore->litStart);
+    size_t literalsRead = 0;
+    size_t lastLLSize;
+
+    ZSTD_Sequence* outSeqs = &zc->seqCollector.seqStart[zc->seqCollector.seqIndex];
+    size_t i;
+    repcodes_t updatedRepcodes;
+
+    assert(zc->seqCollector.seqIndex + 1 < zc->seqCollector.maxSequences);
+    /* Ensure we have enough space for last literals "sequence" */
+    assert(zc->seqCollector.maxSequences >= seqStoreSeqSize + 1);
+    ZSTD_memcpy(updatedRepcodes.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    for (i = 0; i < seqStoreSeqSize; ++i) {
+        U32 rawOffset = seqStoreSeqs[i].offBase - ZSTD_REP_NUM;
+        outSeqs[i].litLength = seqStoreSeqs[i].litLength;
+        outSeqs[i].matchLength = seqStoreSeqs[i].mlBase + MINMATCH;
+        outSeqs[i].rep = 0;
+
+        if (i == seqStore->longLengthPos) {
+            if (seqStore->longLengthType == ZSTD_llt_literalLength) {
+                outSeqs[i].litLength += 0x10000;
+            } else if (seqStore->longLengthType == ZSTD_llt_matchLength) {
+                outSeqs[i].matchLength += 0x10000;
+            }
+        }
+
+        if (seqStoreSeqs[i].offBase <= ZSTD_REP_NUM) {
+            /* Derive the correct offset corresponding to a repcode */
+            outSeqs[i].rep = seqStoreSeqs[i].offBase;
+            if (outSeqs[i].litLength != 0) {
+                rawOffset = updatedRepcodes.rep[outSeqs[i].rep - 1];
+            } else {
+                if (outSeqs[i].rep == 3) {
+                    rawOffset = updatedRepcodes.rep[0] - 1;
+                } else {
+                    rawOffset = updatedRepcodes.rep[outSeqs[i].rep];
+                }
+            }
+        }
+        outSeqs[i].offset = rawOffset;
+        /* seqStoreSeqs[i].offset == offCode+1, and ZSTD_updateRep() expects offCode
+           so we provide seqStoreSeqs[i].offset - 1 */
+        ZSTD_updateRep(updatedRepcodes.rep,
+                       seqStoreSeqs[i].offBase,
+                       seqStoreSeqs[i].litLength == 0);
+        literalsRead += outSeqs[i].litLength;
+    }
+    /* Insert last literals (if any exist) in the block as a sequence with ml == off == 0.
+     * If there are no last literals, then we'll emit (of: 0, ml: 0, ll: 0), which is a marker
+     * for the block boundary, according to the API.
+     */
+    assert(seqStoreLiteralsSize >= literalsRead);
+    lastLLSize = seqStoreLiteralsSize - literalsRead;
+    outSeqs[i].litLength = (U32)lastLLSize;
+    outSeqs[i].matchLength = outSeqs[i].offset = outSeqs[i].rep = 0;
+    seqStoreSeqSize++;
+    zc->seqCollector.seqIndex += seqStoreSeqSize;
+}
+
+size_t ZSTD_sequenceBound(size_t srcSize) {
+    return (srcSize / ZSTD_MINMATCH_MIN) + 1;
+}
+
+size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
+                              size_t outSeqsSize, const void* src, size_t srcSize)
+{
+    const size_t dstCapacity = ZSTD_compressBound(srcSize);
+    void* dst = ZSTD_customMalloc(dstCapacity, ZSTD_defaultCMem);
+    SeqCollector seqCollector;
+
+    RETURN_ERROR_IF(dst == NULL, memory_allocation, "NULL pointer!");
+
+    seqCollector.collectSequences = 1;
+    seqCollector.seqStart = outSeqs;
+    seqCollector.seqIndex = 0;
+    seqCollector.maxSequences = outSeqsSize;
+    zc->seqCollector = seqCollector;
+
+    ZSTD_compress2(zc, dst, dstCapacity, src, srcSize);
+    ZSTD_customFree(dst, ZSTD_defaultCMem);
+    return zc->seqCollector.seqIndex;
+}
+
+size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize) {
+    size_t in = 0;
+    size_t out = 0;
+    for (; in < seqsSize; ++in) {
+        if (sequences[in].offset == 0 && sequences[in].matchLength == 0) {
+            if (in != seqsSize - 1) {
+                sequences[in+1].litLength += sequences[in].litLength;
+            }
+        } else {
+            sequences[out] = sequences[in];
+            ++out;
+        }
+    }
+    return out;
+}
+
+/* Unrolled loop to read four size_ts of input at a time. Returns 1 if is RLE, 0 if not. */
+static int ZSTD_isRLE(const BYTE* src, size_t length) {
+    const BYTE* ip = src;
+    const BYTE value = ip[0];
+    const size_t valueST = (size_t)((U64)value * 0x0101010101010101ULL);
+    const size_t unrollSize = sizeof(size_t) * 4;
+    const size_t unrollMask = unrollSize - 1;
+    const size_t prefixLength = length & unrollMask;
+    size_t i;
+    if (length == 1) return 1;
+    /* Check if prefix is RLE first before using unrolled loop */
+    if (prefixLength && ZSTD_count(ip+1, ip, ip+prefixLength) != prefixLength-1) {
+        return 0;
+    }
+    for (i = prefixLength; i != length; i += unrollSize) {
+        size_t u;
+        for (u = 0; u < unrollSize; u += sizeof(size_t)) {
+            if (MEM_readST(ip + i + u) != valueST) {
+                return 0;
+    }   }   }
+    return 1;
+}
+
+/* Returns true if the given block may be RLE.
+ * This is just a heuristic based on the compressibility.
+ * It may return both false positives and false negatives.
+ */
+static int ZSTD_maybeRLE(seqStore_t const* seqStore)
+{
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
+    size_t const nbLits = (size_t)(seqStore->lit - seqStore->litStart);
+
+    return nbSeqs < 4 && nbLits < 10;
+}
+
+static void
+ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* const bs)
+{
+    ZSTD_compressedBlockState_t* const tmp = bs->prevCBlock;
+    bs->prevCBlock = bs->nextCBlock;
+    bs->nextCBlock = tmp;
+}
+
+/* Writes the block header */
+static void
+writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastBlock)
+{
+    U32 const cBlockHeader = cSize == 1 ?
+                        lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
+                        lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+    MEM_writeLE24(op, cBlockHeader);
+    DEBUGLOG(3, "writeBlockHeader: cSize: %zu blockSize: %zu lastBlock: %u", cSize, blockSize, lastBlock);
+}
+
+/** ZSTD_buildBlockEntropyStats_literals() :
+ *  Builds entropy for the literals.
+ *  Stores literals block type (raw, rle, compressed, repeat) and
+ *  huffman description table to hufMetadata.
+ *  Requires ENTROPY_WORKSPACE_SIZE workspace
+ * @return : size of huffman description table, or an error code
+ */
+static size_t
+ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSize,
+                               const ZSTD_hufCTables_t* prevHuf,
+                                     ZSTD_hufCTables_t* nextHuf,
+                                     ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               const int literalsCompressionIsDisabled,
+                                     void* workspace, size_t wkspSize,
+                                     int hufFlags)
+{
+    BYTE* const wkspStart = (BYTE*)workspace;
+    BYTE* const wkspEnd = wkspStart + wkspSize;
+    BYTE* const countWkspStart = wkspStart;
+    unsigned* const countWksp = (unsigned*)workspace;
+    const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
+    BYTE* const nodeWksp = countWkspStart + countWkspSize;
+    const size_t nodeWkspSize = (size_t)(wkspEnd - nodeWksp);
+    unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    unsigned huffLog = LitHufLog;
+    HUF_repeat repeat = prevHuf->repeatMode;
+    DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_literals (srcSize=%zu)", srcSize);
+
+    /* Prepare nextEntropy assuming reusing the existing table */
+    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+
+    if (literalsCompressionIsDisabled) {
+        DEBUGLOG(5, "set_basic - disabled");
+        hufMetadata->hType = set_basic;
+        return 0;
+    }
+
+    /* small ? don't even attempt compression (speed opt) */
+#ifndef COMPRESS_LITERALS_SIZE_MIN
+# define COMPRESS_LITERALS_SIZE_MIN 63  /* heuristic */
+#endif
+    {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
+        if (srcSize <= minLitSize) {
+            DEBUGLOG(5, "set_basic - too small");
+            hufMetadata->hType = set_basic;
+            return 0;
+    }   }
+
+    /* Scan input and build symbol stats */
+    {   size_t const largest =
+            HIST_count_wksp (countWksp, &maxSymbolValue,
+                            (const BYTE*)src, srcSize,
+                            workspace, wkspSize);
+        FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
+        if (largest == srcSize) {
+            /* only one literal symbol */
+            DEBUGLOG(5, "set_rle");
+            hufMetadata->hType = set_rle;
+            return 0;
+        }
+        if (largest <= (srcSize >> 7)+4) {
+            /* heuristic: likely not compressible */
+            DEBUGLOG(5, "set_basic - no gain");
+            hufMetadata->hType = set_basic;
+            return 0;
+    }   }
+
+    /* Validate the previous Huffman table */
+    if (repeat == HUF_repeat_check
+      && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
+        repeat = HUF_repeat_none;
+    }
+
+    /* Build Huffman Tree */
+    ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, nodeWksp, nodeWkspSize, nextHuf->CTable, countWksp, hufFlags);
+    assert(huffLog <= LitHufLog);
+    {   size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
+                                                    maxSymbolValue, huffLog,
+                                                    nodeWksp, nodeWkspSize);
+        FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
+        huffLog = (U32)maxBits;
+    }
+    {   /* Build and write the CTable */
+        size_t const newCSize = HUF_estimateCompressedSize(
+                (HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
+        size_t const hSize = HUF_writeCTable_wksp(
+                hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
+                (HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
+                nodeWksp, nodeWkspSize);
+        /* Check against repeating the previous CTable */
+        if (repeat != HUF_repeat_none) {
+            size_t const oldCSize = HUF_estimateCompressedSize(
+                    (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
+            if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
+                DEBUGLOG(5, "set_repeat - smaller");
+                ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                hufMetadata->hType = set_repeat;
+                return 0;
+        }   }
+        if (newCSize + hSize >= srcSize) {
+            DEBUGLOG(5, "set_basic - no gains");
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            hufMetadata->hType = set_basic;
+            return 0;
+        }
+        DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
+        hufMetadata->hType = set_compressed;
+        nextHuf->repeatMode = HUF_repeat_check;
+        return hSize;
+    }
+}
+
+
+/* ZSTD_buildDummySequencesStatistics():
+ * Returns a ZSTD_symbolEncodingTypeStats_t with all encoding types as set_basic,
+ * and updates nextEntropy to the appropriate repeatMode.
+ */
+static ZSTD_symbolEncodingTypeStats_t
+ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy)
+{
+    ZSTD_symbolEncodingTypeStats_t stats = {set_basic, set_basic, set_basic, 0, 0, 0};
+    nextEntropy->litlength_repeatMode = FSE_repeat_none;
+    nextEntropy->offcode_repeatMode = FSE_repeat_none;
+    nextEntropy->matchlength_repeatMode = FSE_repeat_none;
+    return stats;
+}
+
+/** ZSTD_buildBlockEntropyStats_sequences() :
+ *  Builds entropy for the sequences.
+ *  Stores symbol compression modes and fse table to fseMetadata.
+ *  Requires ENTROPY_WORKSPACE_SIZE wksp.
+ * @return : size of fse tables or error code */
+static size_t
+ZSTD_buildBlockEntropyStats_sequences(
+                const seqStore_t* seqStorePtr,
+                const ZSTD_fseCTables_t* prevEntropy,
+                      ZSTD_fseCTables_t* nextEntropy,
+                const ZSTD_CCtx_params* cctxParams,
+                      ZSTD_fseCTablesMetadata_t* fseMetadata,
+                      void* workspace, size_t wkspSize)
+{
+    ZSTD_strategy const strategy = cctxParams->cParams.strategy;
+    size_t const nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    BYTE* const ostart = fseMetadata->fseTablesBuffer;
+    BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
+    BYTE* op = ostart;
+    unsigned* countWorkspace = (unsigned*)workspace;
+    unsigned* entropyWorkspace = countWorkspace + (MaxSeq + 1);
+    size_t entropyWorkspaceSize = wkspSize - (MaxSeq + 1) * sizeof(*countWorkspace);
+    ZSTD_symbolEncodingTypeStats_t stats;
+
+    DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_sequences (nbSeq=%zu)", nbSeq);
+    stats = nbSeq != 0 ? ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
+                                          prevEntropy, nextEntropy, op, oend,
+                                          strategy, countWorkspace,
+                                          entropyWorkspace, entropyWorkspaceSize)
+                       : ZSTD_buildDummySequencesStatistics(nextEntropy);
+    FORWARD_IF_ERROR(stats.size, "ZSTD_buildSequencesStatistics failed!");
+    fseMetadata->llType = (symbolEncodingType_e) stats.LLtype;
+    fseMetadata->ofType = (symbolEncodingType_e) stats.Offtype;
+    fseMetadata->mlType = (symbolEncodingType_e) stats.MLtype;
+    fseMetadata->lastCountSize = stats.lastCountSize;
+    return stats.size;
+}
+
+
+/** ZSTD_buildBlockEntropyStats() :
+ *  Builds entropy for the block.
+ *  Requires workspace size ENTROPY_WORKSPACE_SIZE
+ * @return : 0 on success, or an error code
+ *  Note : also employed in superblock
+ */
+size_t ZSTD_buildBlockEntropyStats(
+            const seqStore_t* seqStorePtr,
+            const ZSTD_entropyCTables_t* prevEntropy,
+                  ZSTD_entropyCTables_t* nextEntropy,
+            const ZSTD_CCtx_params* cctxParams,
+                  ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                  void* workspace, size_t wkspSize)
+{
+    size_t const litSize = (size_t)(seqStorePtr->lit - seqStorePtr->litStart);
+    int const huf_useOptDepth = (cctxParams->cParams.strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD);
+    int const hufFlags = huf_useOptDepth ? HUF_flags_optimalDepth : 0;
+
+    entropyMetadata->hufMetadata.hufDesSize =
+        ZSTD_buildBlockEntropyStats_literals(seqStorePtr->litStart, litSize,
+                                            &prevEntropy->huf, &nextEntropy->huf,
+                                            &entropyMetadata->hufMetadata,
+                                            ZSTD_literalsCompressionIsDisabled(cctxParams),
+                                            workspace, wkspSize, hufFlags);
+
+    FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildBlockEntropyStats_literals failed");
+    entropyMetadata->fseMetadata.fseTablesSize =
+        ZSTD_buildBlockEntropyStats_sequences(seqStorePtr,
+                                              &prevEntropy->fse, &nextEntropy->fse,
+                                              cctxParams,
+                                              &entropyMetadata->fseMetadata,
+                                              workspace, wkspSize);
+    FORWARD_IF_ERROR(entropyMetadata->fseMetadata.fseTablesSize, "ZSTD_buildBlockEntropyStats_sequences failed");
+    return 0;
+}
+
+/* Returns the size estimate for the literals section (header + content) of a block */
+static size_t
+ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
+                               const ZSTD_hufCTables_t* huf,
+                               const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               void* workspace, size_t wkspSize,
+                               int writeEntropy)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    size_t literalSectionHeaderSize = 3 + (litSize >= 1 KB) + (litSize >= 16 KB);
+    U32 singleStream = litSize < 256;
+
+    if (hufMetadata->hType == set_basic) return litSize;
+    else if (hufMetadata->hType == set_rle) return 1;
+    else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
+        size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
+        if (ZSTD_isError(largest)) return litSize;
+        {   size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
+            if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
+            if (!singleStream) cLitSizeEstimate += 6; /* multi-stream huffman uses 6-byte jump table */
+            return cLitSizeEstimate + literalSectionHeaderSize;
+    }   }
+    assert(0); /* impossible */
+    return 0;
+}
+
+/* Returns the size estimate for the FSE-compressed symbols (of, ml, ll) of a block */
+static size_t
+ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
+                    const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
+                    const FSE_CTable* fseCTable,
+                    const U8* additionalBits,
+                    short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                    void* workspace, size_t wkspSize)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    const BYTE* ctp = codeTable;
+    const BYTE* const ctStart = ctp;
+    const BYTE* const ctEnd = ctStart + nbSeq;
+    size_t cSymbolTypeSizeEstimateInBits = 0;
+    unsigned max = maxCode;
+
+    HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);  /* can't fail */
+    if (type == set_basic) {
+        /* We selected this encoding type, so it must be valid. */
+        assert(max <= defaultMax);
+        (void)defaultMax;
+        cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
+    } else if (type == set_rle) {
+        cSymbolTypeSizeEstimateInBits = 0;
+    } else if (type == set_compressed || type == set_repeat) {
+        cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
+    }
+    if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) {
+        return nbSeq * 10;
+    }
+    while (ctp < ctEnd) {
+        if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
+        else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
+        ctp++;
+    }
+    return cSymbolTypeSizeEstimateInBits >> 3;
+}
+
+/* Returns the size estimate for the sequences section (header + content) of a block */
+static size_t
+ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
+                                 const BYTE* llCodeTable,
+                                 const BYTE* mlCodeTable,
+                                 size_t nbSeq,
+                                 const ZSTD_fseCTables_t* fseTables,
+                                 const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                 void* workspace, size_t wkspSize,
+                                 int writeEntropy)
+{
+    size_t sequencesSectionHeaderSize = 1 /* seqHead */ + 1 /* min seqSize size */ + (nbSeq >= 128) + (nbSeq >= LONGNBSEQ);
+    size_t cSeqSizeEstimate = 0;
+    cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, nbSeq, MaxOff,
+                                    fseTables->offcodeCTable, NULL,
+                                    OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                                    workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->llType, llCodeTable, nbSeq, MaxLL,
+                                    fseTables->litlengthCTable, LL_bits,
+                                    LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                                    workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, nbSeq, MaxML,
+                                    fseTables->matchlengthCTable, ML_bits,
+                                    ML_defaultNorm, ML_defaultNormLog, MaxML,
+                                    workspace, wkspSize);
+    if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
+    return cSeqSizeEstimate + sequencesSectionHeaderSize;
+}
+
+/* Returns the size estimate for a given stream of literals, of, ll, ml */
+static size_t
+ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
+                       const BYTE* ofCodeTable,
+                       const BYTE* llCodeTable,
+                       const BYTE* mlCodeTable,
+                       size_t nbSeq,
+                       const ZSTD_entropyCTables_t* entropy,
+                       const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                       void* workspace, size_t wkspSize,
+                       int writeLitEntropy, int writeSeqEntropy)
+{
+    size_t const literalsSize = ZSTD_estimateBlockSize_literal(literals, litSize,
+                                    &entropy->huf, &entropyMetadata->hufMetadata,
+                                    workspace, wkspSize, writeLitEntropy);
+    size_t const seqSize = ZSTD_estimateBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
+                                    nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
+                                    workspace, wkspSize, writeSeqEntropy);
+    return seqSize + literalsSize + ZSTD_blockHeaderSize;
+}
+
+/* Builds entropy statistics and uses them for blocksize estimation.
+ *
+ * @return: estimated compressed size of the seqStore, or a zstd error.
+ */
+static size_t
+ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, ZSTD_CCtx* zc)
+{
+    ZSTD_entropyCTablesMetadata_t* const entropyMetadata = &zc->blockSplitCtx.entropyMetadata;
+    DEBUGLOG(6, "ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize()");
+    FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(seqStore,
+                    &zc->blockState.prevCBlock->entropy,
+                    &zc->blockState.nextCBlock->entropy,
+                    &zc->appliedParams,
+                    entropyMetadata,
+                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE), "");
+    return ZSTD_estimateBlockSize(
+                    seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
+                    seqStore->ofCode, seqStore->llCode, seqStore->mlCode,
+                    (size_t)(seqStore->sequences - seqStore->sequencesStart),
+                    &zc->blockState.nextCBlock->entropy,
+                    entropyMetadata,
+                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
+                    (int)(entropyMetadata->hufMetadata.hType == set_compressed), 1);
+}
+
+/* Returns literals bytes represented in a seqStore */
+static size_t ZSTD_countSeqStoreLiteralsBytes(const seqStore_t* const seqStore)
+{
+    size_t literalsBytes = 0;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
+    size_t i;
+    for (i = 0; i < nbSeqs; ++i) {
+        seqDef const seq = seqStore->sequencesStart[i];
+        literalsBytes += seq.litLength;
+        if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_literalLength) {
+            literalsBytes += 0x10000;
+    }   }
+    return literalsBytes;
+}
+
+/* Returns match bytes represented in a seqStore */
+static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore)
+{
+    size_t matchBytes = 0;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
+    size_t i;
+    for (i = 0; i < nbSeqs; ++i) {
+        seqDef seq = seqStore->sequencesStart[i];
+        matchBytes += seq.mlBase + MINMATCH;
+        if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_matchLength) {
+            matchBytes += 0x10000;
+    }   }
+    return matchBytes;
+}
+
+/* Derives the seqStore that is a chunk of the originalSeqStore from [startIdx, endIdx).
+ * Stores the result in resultSeqStore.
+ */
+static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
+                               const seqStore_t* originalSeqStore,
+                                     size_t startIdx, size_t endIdx)
+{
+    *resultSeqStore = *originalSeqStore;
+    if (startIdx > 0) {
+        resultSeqStore->sequences = originalSeqStore->sequencesStart + startIdx;
+        resultSeqStore->litStart += ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+    }
+
+    /* Move longLengthPos into the correct position if necessary */
+    if (originalSeqStore->longLengthType != ZSTD_llt_none) {
+        if (originalSeqStore->longLengthPos < startIdx || originalSeqStore->longLengthPos > endIdx) {
+            resultSeqStore->longLengthType = ZSTD_llt_none;
+        } else {
+            resultSeqStore->longLengthPos -= (U32)startIdx;
+        }
+    }
+    resultSeqStore->sequencesStart = originalSeqStore->sequencesStart + startIdx;
+    resultSeqStore->sequences = originalSeqStore->sequencesStart + endIdx;
+    if (endIdx == (size_t)(originalSeqStore->sequences - originalSeqStore->sequencesStart)) {
+        /* This accounts for possible last literals if the derived chunk reaches the end of the block */
+        assert(resultSeqStore->lit == originalSeqStore->lit);
+    } else {
+        size_t const literalsBytes = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+        resultSeqStore->lit = resultSeqStore->litStart + literalsBytes;
+    }
+    resultSeqStore->llCode += startIdx;
+    resultSeqStore->mlCode += startIdx;
+    resultSeqStore->ofCode += startIdx;
+}
+
+/**
+ * Returns the raw offset represented by the combination of offBase, ll0, and repcode history.
+ * offBase must represent a repcode in the numeric representation of ZSTD_storeSeq().
+ */
+static U32
+ZSTD_resolveRepcodeToRawOffset(const U32 rep[ZSTD_REP_NUM], const U32 offBase, const U32 ll0)
+{
+    U32 const adjustedRepCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;  /* [ 0 - 3 ] */
+    assert(OFFBASE_IS_REPCODE(offBase));
+    if (adjustedRepCode == ZSTD_REP_NUM) {
+        assert(ll0);
+        /* litlength == 0 and offCode == 2 implies selection of first repcode - 1
+         * This is only valid if it results in a valid offset value, aka > 0.
+         * Note : it may happen that `rep[0]==1` in exceptional circumstances.
+         * In which case this function will return 0, which is an invalid offset.
+         * It's not an issue though, since this value will be
+         * compared and discarded within ZSTD_seqStore_resolveOffCodes().
+         */
+        return rep[0] - 1;
+    }
+    return rep[adjustedRepCode];
+}
+
+/**
+ * ZSTD_seqStore_resolveOffCodes() reconciles any possible divergences in offset history that may arise
+ * due to emission of RLE/raw blocks that disturb the offset history,
+ * and replaces any repcodes within the seqStore that may be invalid.
+ *
+ * dRepcodes are updated as would be on the decompression side.
+ * cRepcodes are updated exactly in accordance with the seqStore.
+ *
+ * Note : this function assumes seq->offBase respects the following numbering scheme :
+ *        0 : invalid
+ *        1-3 : repcode 1-3
+ *        4+ : real_offset+3
+ */
+static void
+ZSTD_seqStore_resolveOffCodes(repcodes_t* const dRepcodes, repcodes_t* const cRepcodes,
+                        const seqStore_t* const seqStore, U32 const nbSeq)
+{
+    U32 idx = 0;
+    U32 const longLitLenIdx = seqStore->longLengthType == ZSTD_llt_literalLength ? seqStore->longLengthPos : nbSeq;
+    for (; idx < nbSeq; ++idx) {
+        seqDef* const seq = seqStore->sequencesStart + idx;
+        U32 const ll0 = (seq->litLength == 0) && (idx != longLitLenIdx);
+        U32 const offBase = seq->offBase;
+        assert(offBase > 0);
+        if (OFFBASE_IS_REPCODE(offBase)) {
+            U32 const dRawOffset = ZSTD_resolveRepcodeToRawOffset(dRepcodes->rep, offBase, ll0);
+            U32 const cRawOffset = ZSTD_resolveRepcodeToRawOffset(cRepcodes->rep, offBase, ll0);
+            /* Adjust simulated decompression repcode history if we come across a mismatch. Replace
+             * the repcode with the offset it actually references, determined by the compression
+             * repcode history.
+             */
+            if (dRawOffset != cRawOffset) {
+                seq->offBase = OFFSET_TO_OFFBASE(cRawOffset);
+            }
+        }
+        /* Compression repcode history is always updated with values directly from the unmodified seqStore.
+         * Decompression repcode history may use modified seq->offset value taken from compression repcode history.
+         */
+        ZSTD_updateRep(dRepcodes->rep, seq->offBase, ll0);
+        ZSTD_updateRep(cRepcodes->rep, offBase, ll0);
+    }
+}
+
+/* ZSTD_compressSeqStore_singleBlock():
+ * Compresses a seqStore into a block with a block header, into the buffer dst.
+ *
+ * Returns the total size of that block (including header) or a ZSTD error code.
+ */
+static size_t
+ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc,
+                            const seqStore_t* const seqStore,
+                                  repcodes_t* const dRep, repcodes_t* const cRep,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                                  U32 lastBlock, U32 isPartition)
+{
+    const U32 rleMaxLength = 25;
+    BYTE* op = (BYTE*)dst;
+    const BYTE* ip = (const BYTE*)src;
+    size_t cSize;
+    size_t cSeqsSize;
+
+    /* In case of an RLE or raw block, the simulated decompression repcode history must be reset */
+    repcodes_t const dRepOriginal = *dRep;
+    DEBUGLOG(5, "ZSTD_compressSeqStore_singleBlock");
+    if (isPartition)
+        ZSTD_seqStore_resolveOffCodes(dRep, cRep, seqStore, (U32)(seqStore->sequences - seqStore->sequencesStart));
+
+    RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "Block header doesn't fit");
+    cSeqsSize = ZSTD_entropyCompressSeqStore(seqStore,
+                &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+                &zc->appliedParams,
+                op + ZSTD_blockHeaderSize, dstCapacity - ZSTD_blockHeaderSize,
+                srcSize,
+                zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
+                zc->bmi2);
+    FORWARD_IF_ERROR(cSeqsSize, "ZSTD_entropyCompressSeqStore failed!");
+
+    if (!zc->isFirstBlock &&
+        cSeqsSize < rleMaxLength &&
+        ZSTD_isRLE((BYTE const*)src, srcSize)) {
+        /* We don't want to emit our first block as a RLE even if it qualifies because
+        * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+        * This is only an issue for zstd <= v1.4.3
+        */
+        cSeqsSize = 1;
+    }
+
+    if (zc->seqCollector.collectSequences) {
+        ZSTD_copyBlockSequences(zc);
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+        return 0;
+    }
+
+    if (cSeqsSize == 0) {
+        cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
+        FORWARD_IF_ERROR(cSize, "Nocompress block failed");
+        DEBUGLOG(4, "Writing out nocompress block, size: %zu", cSize);
+        *dRep = dRepOriginal; /* reset simulated decompression repcode history */
+    } else if (cSeqsSize == 1) {
+        cSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, srcSize, lastBlock);
+        FORWARD_IF_ERROR(cSize, "RLE compress block failed");
+        DEBUGLOG(4, "Writing out RLE block, size: %zu", cSize);
+        *dRep = dRepOriginal; /* reset simulated decompression repcode history */
+    } else {
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+        writeBlockHeader(op, cSeqsSize, srcSize, lastBlock);
+        cSize = ZSTD_blockHeaderSize + cSeqsSize;
+        DEBUGLOG(4, "Writing out compressed block, size: %zu", cSize);
+    }
+
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+/* Struct to keep track of where we are in our recursive calls. */
+typedef struct {
+    U32* splitLocations;    /* Array of split indices */
+    size_t idx;             /* The current index within splitLocations being worked on */
+} seqStoreSplits;
+
+#define MIN_SEQUENCES_BLOCK_SPLITTING 300
+
+/* Helper function to perform the recursive search for block splits.
+ * Estimates the cost of seqStore prior to split, and estimates the cost of splitting the sequences in half.
+ * If advantageous to split, then we recurse down the two sub-blocks.
+ * If not, or if an error occurred in estimation, then we do not recurse.
+ *
+ * Note: The recursion depth is capped by a heuristic minimum number of sequences,
+ * defined by MIN_SEQUENCES_BLOCK_SPLITTING.
+ * In theory, this means the absolute largest recursion depth is 10 == log2(maxNbSeqInBlock/MIN_SEQUENCES_BLOCK_SPLITTING).
+ * In practice, recursion depth usually doesn't go beyond 4.
+ *
+ * Furthermore, the number of splits is capped by ZSTD_MAX_NB_BLOCK_SPLITS.
+ * At ZSTD_MAX_NB_BLOCK_SPLITS == 196 with the current existing blockSize
+ * maximum of 128 KB, this value is actually impossible to reach.
+ */
+static void
+ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t endIdx,
+                             ZSTD_CCtx* zc, const seqStore_t* origSeqStore)
+{
+    seqStore_t* const fullSeqStoreChunk = &zc->blockSplitCtx.fullSeqStoreChunk;
+    seqStore_t* const firstHalfSeqStore = &zc->blockSplitCtx.firstHalfSeqStore;
+    seqStore_t* const secondHalfSeqStore = &zc->blockSplitCtx.secondHalfSeqStore;
+    size_t estimatedOriginalSize;
+    size_t estimatedFirstHalfSize;
+    size_t estimatedSecondHalfSize;
+    size_t midIdx = (startIdx + endIdx)/2;
+
+    DEBUGLOG(5, "ZSTD_deriveBlockSplitsHelper: startIdx=%zu endIdx=%zu", startIdx, endIdx);
+    assert(endIdx >= startIdx);
+    if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= ZSTD_MAX_NB_BLOCK_SPLITS) {
+        DEBUGLOG(6, "ZSTD_deriveBlockSplitsHelper: Too few sequences (%zu)", endIdx - startIdx);
+        return;
+    }
+    ZSTD_deriveSeqStoreChunk(fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
+    ZSTD_deriveSeqStoreChunk(firstHalfSeqStore, origSeqStore, startIdx, midIdx);
+    ZSTD_deriveSeqStoreChunk(secondHalfSeqStore, origSeqStore, midIdx, endIdx);
+    estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(fullSeqStoreChunk, zc);
+    estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(firstHalfSeqStore, zc);
+    estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(secondHalfSeqStore, zc);
+    DEBUGLOG(5, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
+             estimatedOriginalSize, estimatedFirstHalfSize, estimatedSecondHalfSize);
+    if (ZSTD_isError(estimatedOriginalSize) || ZSTD_isError(estimatedFirstHalfSize) || ZSTD_isError(estimatedSecondHalfSize)) {
+        return;
+    }
+    if (estimatedFirstHalfSize + estimatedSecondHalfSize < estimatedOriginalSize) {
+        DEBUGLOG(5, "split decided at seqNb:%zu", midIdx);
+        ZSTD_deriveBlockSplitsHelper(splits, startIdx, midIdx, zc, origSeqStore);
+        splits->splitLocations[splits->idx] = (U32)midIdx;
+        splits->idx++;
+        ZSTD_deriveBlockSplitsHelper(splits, midIdx, endIdx, zc, origSeqStore);
+    }
+}
+
+/* Base recursive function.
+ * Populates a table with intra-block partition indices that can improve compression ratio.
+ *
+ * @return: number of splits made (which equals the size of the partition table - 1).
+ */
+static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq)
+{
+    seqStoreSplits splits;
+    splits.splitLocations = partitions;
+    splits.idx = 0;
+    if (nbSeq <= 4) {
+        DEBUGLOG(5, "ZSTD_deriveBlockSplits: Too few sequences to split (%u <= 4)", nbSeq);
+        /* Refuse to try and split anything with less than 4 sequences */
+        return 0;
+    }
+    ZSTD_deriveBlockSplitsHelper(&splits, 0, nbSeq, zc, &zc->seqStore);
+    splits.splitLocations[splits.idx] = nbSeq;
+    DEBUGLOG(5, "ZSTD_deriveBlockSplits: final nb partitions: %zu", splits.idx+1);
+    return splits.idx;
+}
+
+/* ZSTD_compressBlock_splitBlock():
+ * Attempts to split a given block into multiple blocks to improve compression ratio.
+ *
+ * Returns combined size of all blocks (which includes headers), or a ZSTD error code.
+ */
+static size_t
+ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc,
+                                    void* dst, size_t dstCapacity,
+                              const void* src, size_t blockSize,
+                                    U32 lastBlock, U32 nbSeq)
+{
+    size_t cSize = 0;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    size_t i = 0;
+    size_t srcBytesTotal = 0;
+    U32* const partitions = zc->blockSplitCtx.partitions; /* size == ZSTD_MAX_NB_BLOCK_SPLITS */
+    seqStore_t* const nextSeqStore = &zc->blockSplitCtx.nextSeqStore;
+    seqStore_t* const currSeqStore = &zc->blockSplitCtx.currSeqStore;
+    size_t const numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
+
+    /* If a block is split and some partitions are emitted as RLE/uncompressed, then repcode history
+     * may become invalid. In order to reconcile potentially invalid repcodes, we keep track of two
+     * separate repcode histories that simulate repcode history on compression and decompression side,
+     * and use the histories to determine whether we must replace a particular repcode with its raw offset.
+     *
+     * 1) cRep gets updated for each partition, regardless of whether the block was emitted as uncompressed
+     *    or RLE. This allows us to retrieve the offset value that an invalid repcode references within
+     *    a nocompress/RLE block.
+     * 2) dRep gets updated only for compressed partitions, and when a repcode gets replaced, will use
+     *    the replacement offset value rather than the original repcode to update the repcode history.
+     *    dRep also will be the final repcode history sent to the next block.
+     *
+     * See ZSTD_seqStore_resolveOffCodes() for more details.
+     */
+    repcodes_t dRep;
+    repcodes_t cRep;
+    ZSTD_memcpy(dRep.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    ZSTD_memcpy(cRep.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    ZSTD_memset(nextSeqStore, 0, sizeof(seqStore_t));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
+                (unsigned)zc->blockState.matchState.nextToUpdate);
+
+    if (numSplits == 0) {
+        size_t cSizeSingleBlock =
+            ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore,
+                                            &dRep, &cRep,
+                                            op, dstCapacity,
+                                            ip, blockSize,
+                                            lastBlock, 0 /* isPartition */);
+        FORWARD_IF_ERROR(cSizeSingleBlock, "Compressing single block from splitBlock_internal() failed!");
+        DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal: No splits");
+        assert(zc->blockSize <= ZSTD_BLOCKSIZE_MAX);
+        assert(cSizeSingleBlock <= zc->blockSize + ZSTD_blockHeaderSize);
+        return cSizeSingleBlock;
+    }
+
+    ZSTD_deriveSeqStoreChunk(currSeqStore, &zc->seqStore, 0, partitions[0]);
+    for (i = 0; i <= numSplits; ++i) {
+        size_t cSizeChunk;
+        U32 const lastPartition = (i == numSplits);
+        U32 lastBlockEntireSrc = 0;
+
+        size_t srcBytes = ZSTD_countSeqStoreLiteralsBytes(currSeqStore) + ZSTD_countSeqStoreMatchBytes(currSeqStore);
+        srcBytesTotal += srcBytes;
+        if (lastPartition) {
+            /* This is the final partition, need to account for possible last literals */
+            srcBytes += blockSize - srcBytesTotal;
+            lastBlockEntireSrc = lastBlock;
+        } else {
+            ZSTD_deriveSeqStoreChunk(nextSeqStore, &zc->seqStore, partitions[i], partitions[i+1]);
+        }
+
+        cSizeChunk = ZSTD_compressSeqStore_singleBlock(zc, currSeqStore,
+                                                      &dRep, &cRep,
+                                                       op, dstCapacity,
+                                                       ip, srcBytes,
+                                                       lastBlockEntireSrc, 1 /* isPartition */);
+        DEBUGLOG(5, "Estimated size: %zu vs %zu : actual size",
+                    ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(currSeqStore, zc), cSizeChunk);
+        FORWARD_IF_ERROR(cSizeChunk, "Compressing chunk failed!");
+
+        ip += srcBytes;
+        op += cSizeChunk;
+        dstCapacity -= cSizeChunk;
+        cSize += cSizeChunk;
+        *currSeqStore = *nextSeqStore;
+        assert(cSizeChunk <= zc->blockSize + ZSTD_blockHeaderSize);
+    }
+    /* cRep and dRep may have diverged during the compression.
+     * If so, we use the dRep repcodes for the next block.
+     */
+    ZSTD_memcpy(zc->blockState.prevCBlock->rep, dRep.rep, sizeof(repcodes_t));
+    return cSize;
+}
+
+static size_t
+ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
+                              void* dst, size_t dstCapacity,
+                              const void* src, size_t srcSize, U32 lastBlock)
+{
+    U32 nbSeq;
+    size_t cSize;
+    DEBUGLOG(4, "ZSTD_compressBlock_splitBlock");
+    assert(zc->appliedParams.useBlockSplitter == ZSTD_ps_enable);
+
+    {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+        FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
+        if (bss == ZSTDbss_noCompress) {
+            if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+                zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+            cSize = ZSTD_noCompressBlock(dst, dstCapacity, src, srcSize, lastBlock);
+            FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
+            DEBUGLOG(4, "ZSTD_compressBlock_splitBlock: Nocompress block");
+            return cSize;
+        }
+        nbSeq = (U32)(zc->seqStore.sequences - zc->seqStore.sequencesStart);
+    }
+
+    cSize = ZSTD_compressBlock_splitBlock_internal(zc, dst, dstCapacity, src, srcSize, lastBlock, nbSeq);
+    FORWARD_IF_ERROR(cSize, "Splitting blocks failed!");
+    return cSize;
+}
+
+static size_t
+ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
+                            void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize, U32 frame)
+{
+    /* This is an estimated upper bound for the length of an rle block.
+     * This isn't the actual upper bound.
+     * Finding the real threshold needs further investigation.
+     */
+    const U32 rleMaxLength = 25;
+    size_t cSize;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
+                (unsigned)zc->blockState.matchState.nextToUpdate);
+
+    {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+        FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
+        if (bss == ZSTDbss_noCompress) { cSize = 0; goto out; }
+    }
+
+    if (zc->seqCollector.collectSequences) {
+        ZSTD_copyBlockSequences(zc);
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+        return 0;
+    }
+
+    /* encode sequences and literals */
+    cSize = ZSTD_entropyCompressSeqStore(&zc->seqStore,
+            &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+            &zc->appliedParams,
+            dst, dstCapacity,
+            srcSize,
+            zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
+            zc->bmi2);
+
+    if (frame &&
+        /* We don't want to emit our first block as a RLE even if it qualifies because
+         * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+         * This is only an issue for zstd <= v1.4.3
+         */
+        !zc->isFirstBlock &&
+        cSize < rleMaxLength &&
+        ZSTD_isRLE(ip, srcSize))
+    {
+        cSize = 1;
+        op[0] = ip[0];
+    }
+
+out:
+    if (!ZSTD_isError(cSize) && cSize > 1) {
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+    }
+    /* We check that dictionaries have offset codes available for the first
+     * block. After the first block, the offcode table might not have large
+     * enough codes to represent the offsets in the data.
+     */
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+static size_t ZSTD_compressBlock_targetCBlockSize_body(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               const size_t bss, U32 lastBlock)
+{
+    DEBUGLOG(6, "Attempting ZSTD_compressSuperBlock()");
+    if (bss == ZSTDbss_compress) {
+        if (/* We don't want to emit our first block as a RLE even if it qualifies because
+            * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+            * This is only an issue for zstd <= v1.4.3
+            */
+            !zc->isFirstBlock &&
+            ZSTD_maybeRLE(&zc->seqStore) &&
+            ZSTD_isRLE((BYTE const*)src, srcSize))
+        {
+            return ZSTD_rleCompressBlock(dst, dstCapacity, *(BYTE const*)src, srcSize, lastBlock);
+        }
+        /* Attempt superblock compression.
+         *
+         * Note that compressed size of ZSTD_compressSuperBlock() is not bound by the
+         * standard ZSTD_compressBound(). This is a problem, because even if we have
+         * space now, taking an extra byte now could cause us to run out of space later
+         * and violate ZSTD_compressBound().
+         *
+         * Define blockBound(blockSize) = blockSize + ZSTD_blockHeaderSize.
+         *
+         * In order to respect ZSTD_compressBound() we must attempt to emit a raw
+         * uncompressed block in these cases:
+         *   * cSize == 0: Return code for an uncompressed block.
+         *   * cSize == dstSize_tooSmall: We may have expanded beyond blockBound(srcSize).
+         *     ZSTD_noCompressBlock() will return dstSize_tooSmall if we are really out of
+         *     output space.
+         *   * cSize >= blockBound(srcSize): We have expanded the block too much so
+         *     emit an uncompressed block.
+         */
+        {   size_t const cSize =
+                ZSTD_compressSuperBlock(zc, dst, dstCapacity, src, srcSize, lastBlock);
+            if (cSize != ERROR(dstSize_tooSmall)) {
+                size_t const maxCSize =
+                    srcSize - ZSTD_minGain(srcSize, zc->appliedParams.cParams.strategy);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressSuperBlock failed");
+                if (cSize != 0 && cSize < maxCSize + ZSTD_blockHeaderSize) {
+                    ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+                    return cSize;
+                }
+            }
+        }
+    } /* if (bss == ZSTDbss_compress)*/
+
+    DEBUGLOG(6, "Resorting to ZSTD_noCompressBlock()");
+    /* Superblock compression failed, attempt to emit a single no compress block.
+     * The decoder will be able to stream this block since it is uncompressed.
+     */
+    return ZSTD_noCompressBlock(dst, dstCapacity, src, srcSize, lastBlock);
+}
+
+static size_t ZSTD_compressBlock_targetCBlockSize(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               U32 lastBlock)
+{
+    size_t cSize = 0;
+    const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+    DEBUGLOG(5, "ZSTD_compressBlock_targetCBlockSize (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u, srcSize=%zu)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit, (unsigned)zc->blockState.matchState.nextToUpdate, srcSize);
+    FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
+
+    cSize = ZSTD_compressBlock_targetCBlockSize_body(zc, dst, dstCapacity, src, srcSize, bss, lastBlock);
+    FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_targetCBlockSize_body failed");
+
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms,
+                                         ZSTD_cwksp* ws,
+                                         ZSTD_CCtx_params const* params,
+                                         void const* ip,
+                                         void const* iend)
+{
+    U32 const cycleLog = ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy);
+    U32 const maxDist = (U32)1 << params->cParams.windowLog;
+    if (ZSTD_window_needOverflowCorrection(ms->window, cycleLog, maxDist, ms->loadedDictEnd, ip, iend)) {
+        U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
+        ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        ZSTD_cwksp_mark_tables_dirty(ws);
+        ZSTD_reduceIndex(ms, params, correction);
+        ZSTD_cwksp_mark_tables_clean(ws);
+        if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
+        else ms->nextToUpdate -= correction;
+        /* invalidate dictionaries on overflow correction */
+        ms->loadedDictEnd = 0;
+        ms->dictMatchState = NULL;
+    }
+}
+
+/*! ZSTD_compress_frameChunk() :
+*   Compress a chunk of data into one or multiple blocks.
+*   All blocks will be terminated, all input will be consumed.
+*   Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
+*   Frame is supposed already started (header already produced)
+*  @return : compressed size, or an error code
+*/
+static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     U32 lastFrameChunk)
+{
+    size_t blockSize = cctx->blockSize;
+    size_t remaining = srcSize;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    U32 const maxDist = (U32)1 << cctx->appliedParams.cParams.windowLog;
+
+    assert(cctx->appliedParams.cParams.windowLog <= ZSTD_WINDOWLOG_MAX);
+
+    DEBUGLOG(4, "ZSTD_compress_frameChunk (blockSize=%u)", (unsigned)blockSize);
+    if (cctx->appliedParams.fParams.checksumFlag && srcSize)
+        XXH64_update(&cctx->xxhState, src, srcSize);
+
+    while (remaining) {
+        ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
+        U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
+
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE + 1,
+                        dstSize_tooSmall,
+                        "not enough space to store compressed block");
+        if (remaining < blockSize) blockSize = remaining;
+
+        ZSTD_overflowCorrectIfNeeded(
+            ms, &cctx->workspace, &cctx->appliedParams, ip, ip + blockSize);
+        ZSTD_checkDictValidity(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
+        ZSTD_window_enforceMaxDist(&ms->window, ip, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
+
+        /* Ensure hash/chain table insertion resumes no sooner than lowlimit */
+        if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
+
+        {   size_t cSize;
+            if (ZSTD_useTargetCBlockSize(&cctx->appliedParams)) {
+                cSize = ZSTD_compressBlock_targetCBlockSize(cctx, op, dstCapacity, ip, blockSize, lastBlock);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_targetCBlockSize failed");
+                assert(cSize > 0);
+                assert(cSize <= blockSize + ZSTD_blockHeaderSize);
+            } else if (ZSTD_blockSplitterEnabled(&cctx->appliedParams)) {
+                cSize = ZSTD_compressBlock_splitBlock(cctx, op, dstCapacity, ip, blockSize, lastBlock);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_splitBlock failed");
+                assert(cSize > 0 || cctx->seqCollector.collectSequences == 1);
+            } else {
+                cSize = ZSTD_compressBlock_internal(cctx,
+                                        op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
+                                        ip, blockSize, 1 /* frame */);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_internal failed");
+
+                if (cSize == 0) {  /* block is not compressible */
+                    cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+                    FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
+                } else {
+                    U32 const cBlockHeader = cSize == 1 ?
+                        lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
+                        lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+                    MEM_writeLE24(op, cBlockHeader);
+                    cSize += ZSTD_blockHeaderSize;
+                }
+            }  /* if (ZSTD_useTargetCBlockSize(&cctx->appliedParams))*/
+
+
+            ip += blockSize;
+            assert(remaining >= blockSize);
+            remaining -= blockSize;
+            op += cSize;
+            assert(dstCapacity >= cSize);
+            dstCapacity -= cSize;
+            cctx->isFirstBlock = 0;
+            DEBUGLOG(5, "ZSTD_compress_frameChunk: adding a block of size %u",
+                        (unsigned)cSize);
+    }   }
+
+    if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
+    return (size_t)(op-ostart);
+}
+
+
+static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
+                                    const ZSTD_CCtx_params* params, U64 pledgedSrcSize, U32 dictID)
+{   BYTE* const op = (BYTE*)dst;
+    U32   const dictIDSizeCodeLength = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
+    U32   const dictIDSizeCode = params->fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength;   /* 0-3 */
+    U32   const checksumFlag = params->fParams.checksumFlag>0;
+    U32   const windowSize = (U32)1 << params->cParams.windowLog;
+    U32   const singleSegment = params->fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
+    BYTE  const windowLogByte = (BYTE)((params->cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
+    U32   const fcsCode = params->fParams.contentSizeFlag ?
+                     (pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) : 0;  /* 0-3 */
+    BYTE  const frameHeaderDescriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
+    size_t pos=0;
+
+    assert(!(params->fParams.contentSizeFlag && pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN));
+    RETURN_ERROR_IF(dstCapacity < ZSTD_FRAMEHEADERSIZE_MAX, dstSize_tooSmall,
+                    "dst buf is too small to fit worst-case frame header size.");
+    DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
+                !params->fParams.noDictIDFlag, (unsigned)dictID, (unsigned)dictIDSizeCode);
+    if (params->format == ZSTD_f_zstd1) {
+        MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
+        pos = 4;
+    }
+    op[pos++] = frameHeaderDescriptionByte;
+    if (!singleSegment) op[pos++] = windowLogByte;
+    switch(dictIDSizeCode)
+    {
+        default:
+            assert(0); /* impossible */
+            ZSTD_FALLTHROUGH;
+        case 0 : break;
+        case 1 : op[pos] = (BYTE)(dictID); pos++; break;
+        case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
+        case 3 : MEM_writeLE32(op+pos, dictID); pos+=4; break;
+    }
+    switch(fcsCode)
+    {
+        default:
+            assert(0); /* impossible */
+            ZSTD_FALLTHROUGH;
+        case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
+        case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
+        case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
+        case 3 : MEM_writeLE64(op+pos, (U64)(pledgedSrcSize)); pos+=8; break;
+    }
+    return pos;
+}
+
+/* ZSTD_writeSkippableFrame_advanced() :
+ * Writes out a skippable frame with the specified magic number variant (16 are supported),
+ * from ZSTD_MAGIC_SKIPPABLE_START to ZSTD_MAGIC_SKIPPABLE_START+15, and the desired source data.
+ *
+ * Returns the total number of bytes written, or a ZSTD error code.
+ */
+size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize, unsigned magicVariant) {
+    BYTE* op = (BYTE*)dst;
+    RETURN_ERROR_IF(dstCapacity < srcSize + ZSTD_SKIPPABLEHEADERSIZE /* Skippable frame overhead */,
+                    dstSize_tooSmall, "Not enough room for skippable frame");
+    RETURN_ERROR_IF(srcSize > (unsigned)0xFFFFFFFF, srcSize_wrong, "Src size too large for skippable frame");
+    RETURN_ERROR_IF(magicVariant > 15, parameter_outOfBound, "Skippable frame magic number variant not supported");
+
+    MEM_writeLE32(op, (U32)(ZSTD_MAGIC_SKIPPABLE_START + magicVariant));
+    MEM_writeLE32(op+4, (U32)srcSize);
+    ZSTD_memcpy(op+8, src, srcSize);
+    return srcSize + ZSTD_SKIPPABLEHEADERSIZE;
+}
+
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
+{
+    RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall,
+                    "dst buf is too small to write frame trailer empty block.");
+    {   U32 const cBlockHeader24 = 1 /*lastBlock*/ + (((U32)bt_raw)<<1);  /* 0 size */
+        MEM_writeLE24(dst, cBlockHeader24);
+        return ZSTD_blockHeaderSize;
+    }
+}
+
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
+{
+    RETURN_ERROR_IF(cctx->stage != ZSTDcs_init, stage_wrong,
+                    "wrong cctx stage");
+    RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable,
+                    parameter_unsupported,
+                    "incompatible with ldm");
+    cctx->externSeqStore.seq = seq;
+    cctx->externSeqStore.size = nbSeq;
+    cctx->externSeqStore.capacity = nbSeq;
+    cctx->externSeqStore.pos = 0;
+    cctx->externSeqStore.posInSequence = 0;
+    return 0;
+}
+
+
+static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                               U32 frame, U32 lastFrameChunk)
+{
+    ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
+    size_t fhSize = 0;
+
+    DEBUGLOG(5, "ZSTD_compressContinue_internal, stage: %u, srcSize: %u",
+                cctx->stage, (unsigned)srcSize);
+    RETURN_ERROR_IF(cctx->stage==ZSTDcs_created, stage_wrong,
+                    "missing init (ZSTD_compressBegin)");
+
+    if (frame && (cctx->stage==ZSTDcs_init)) {
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams,
+                                       cctx->pledgedSrcSizePlusOne-1, cctx->dictID);
+        FORWARD_IF_ERROR(fhSize, "ZSTD_writeFrameHeader failed");
+        assert(fhSize <= dstCapacity);
+        dstCapacity -= fhSize;
+        dst = (char*)dst + fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    if (!srcSize) return fhSize;  /* do not generate an empty block if no input */
+
+    if (!ZSTD_window_update(&ms->window, src, srcSize, ms->forceNonContiguous)) {
+        ms->forceNonContiguous = 0;
+        ms->nextToUpdate = ms->window.dictLimit;
+    }
+    if (cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_window_update(&cctx->ldmState.window, src, srcSize, /* forceNonContiguous */ 0);
+    }
+
+    if (!frame) {
+        /* overflow check and correction for block mode */
+        ZSTD_overflowCorrectIfNeeded(
+            ms, &cctx->workspace, &cctx->appliedParams,
+            src, (BYTE const*)src + srcSize);
+    }
+
+    DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (unsigned)cctx->blockSize);
+    {   size_t const cSize = frame ?
+                             ZSTD_compress_frameChunk (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
+                             ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize, 0 /* frame */);
+        FORWARD_IF_ERROR(cSize, "%s", frame ? "ZSTD_compress_frameChunk failed" : "ZSTD_compressBlock_internal failed");
+        cctx->consumedSrcSize += srcSize;
+        cctx->producedCSize += (cSize + fhSize);
+        assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+        if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+            ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
+            RETURN_ERROR_IF(
+                cctx->consumedSrcSize+1 > cctx->pledgedSrcSizePlusOne,
+                srcSize_wrong,
+                "error : pledgedSrcSize = %u, while realSrcSize >= %u",
+                (unsigned)cctx->pledgedSrcSizePlusOne-1,
+                (unsigned)cctx->consumedSrcSize);
+        }
+        return cSize + fhSize;
+    }
+}
+
+size_t ZSTD_compressContinue_public(ZSTD_CCtx* cctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (unsigned)srcSize);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
+}
+
+/* NOTE: Must just wrap ZSTD_compressContinue_public() */
+size_t ZSTD_compressContinue(ZSTD_CCtx* cctx,
+                             void* dst, size_t dstCapacity,
+                       const void* src, size_t srcSize)
+{
+    return ZSTD_compressContinue_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+static size_t ZSTD_getBlockSize_deprecated(const ZSTD_CCtx* cctx)
+{
+    ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
+    assert(!ZSTD_checkCParams(cParams));
+    return MIN(cctx->appliedParams.maxBlockSize, (size_t)1 << cParams.windowLog);
+}
+
+/* NOTE: Must just wrap ZSTD_getBlockSize_deprecated() */
+size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
+{
+    return ZSTD_getBlockSize_deprecated(cctx);
+}
+
+/* NOTE: Must just wrap ZSTD_compressBlock_deprecated() */
+size_t ZSTD_compressBlock_deprecated(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock: srcSize = %u", (unsigned)srcSize);
+    { size_t const blockSizeMax = ZSTD_getBlockSize_deprecated(cctx);
+      RETURN_ERROR_IF(srcSize > blockSizeMax, srcSize_wrong, "input is larger than a block"); }
+
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
+}
+
+/* NOTE: Must just wrap ZSTD_compressBlock_deprecated() */
+size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_deprecated(cctx, dst, dstCapacity, src, srcSize);
+}
+
+/*! ZSTD_loadDictionaryContent() :
+ *  @return : 0, or an error code
+ */
+static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
+                                         ldmState_t* ls,
+                                         ZSTD_cwksp* ws,
+                                         ZSTD_CCtx_params const* params,
+                                         const void* src, size_t srcSize,
+                                         ZSTD_dictTableLoadMethod_e dtlm,
+                                         ZSTD_tableFillPurpose_e tfp)
+{
+    const BYTE* ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+    int const loadLdmDict = params->ldmParams.enableLdm == ZSTD_ps_enable && ls != NULL;
+
+    /* Assert that the ms params match the params we're being given */
+    ZSTD_assertEqualCParams(params->cParams, ms->cParams);
+
+    {   /* Ensure large dictionaries can't cause index overflow */
+
+        /* Allow the dictionary to set indices up to exactly ZSTD_CURRENT_MAX.
+         * Dictionaries right at the edge will immediately trigger overflow
+         * correction, but I don't want to insert extra constraints here.
+         */
+        U32 maxDictSize = ZSTD_CURRENT_MAX - ZSTD_WINDOW_START_INDEX;
+
+        int const CDictTaggedIndices = ZSTD_CDictIndicesAreTagged(&params->cParams);
+        if (CDictTaggedIndices && tfp == ZSTD_tfp_forCDict) {
+            /* Some dictionary matchfinders in zstd use "short cache",
+             * which treats the lower ZSTD_SHORT_CACHE_TAG_BITS of each
+             * CDict hashtable entry as a tag rather than as part of an index.
+             * When short cache is used, we need to truncate the dictionary
+             * so that its indices don't overlap with the tag. */
+            U32 const shortCacheMaxDictSize = (1u << (32 - ZSTD_SHORT_CACHE_TAG_BITS)) - ZSTD_WINDOW_START_INDEX;
+            maxDictSize = MIN(maxDictSize, shortCacheMaxDictSize);
+            assert(!loadLdmDict);
+        }
+
+        /* If the dictionary is too large, only load the suffix of the dictionary. */
+        if (srcSize > maxDictSize) {
+            ip = iend - maxDictSize;
+            src = ip;
+            srcSize = maxDictSize;
+        }
+    }
+
+    if (srcSize > ZSTD_CHUNKSIZE_MAX) {
+        /* We must have cleared our windows when our source is this large. */
+        assert(ZSTD_window_isEmpty(ms->window));
+        if (loadLdmDict) assert(ZSTD_window_isEmpty(ls->window));
+    }
+    ZSTD_window_update(&ms->window, src, srcSize, /* forceNonContiguous */ 0);
+
+    DEBUGLOG(4, "ZSTD_loadDictionaryContent(): useRowMatchFinder=%d", (int)params->useRowMatchFinder);
+
+    if (loadLdmDict) { /* Load the entire dict into LDM matchfinders. */
+        ZSTD_window_update(&ls->window, src, srcSize, /* forceNonContiguous */ 0);
+        ls->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ls->window.base);
+        ZSTD_ldm_fillHashTable(ls, ip, iend, &params->ldmParams);
+    }
+
+    /* If the dict is larger than we can reasonably index in our tables, only load the suffix. */
+    if (params->cParams.strategy < ZSTD_btultra) {
+        U32 maxDictSize = 8U << MIN(MAX(params->cParams.hashLog, params->cParams.chainLog), 28);
+        if (srcSize > maxDictSize) {
+            ip = iend - maxDictSize;
+            src = ip;
+            srcSize = maxDictSize;
+        }
+    }
+
+    ms->nextToUpdate = (U32)(ip - ms->window.base);
+    ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
+    ms->forceNonContiguous = params->deterministicRefPrefix;
+
+    if (srcSize <= HASH_READ_SIZE) return 0;
+
+    ZSTD_overflowCorrectIfNeeded(ms, ws, params, ip, iend);
+
+    switch(params->cParams.strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable(ms, iend, dtlm, tfp);
+        break;
+    case ZSTD_dfast:
+        ZSTD_fillDoubleHashTable(ms, iend, dtlm, tfp);
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+        assert(srcSize >= HASH_READ_SIZE);
+        if (ms->dedicatedDictSearch) {
+            assert(ms->chainTable != NULL);
+            ZSTD_dedicatedDictSearch_lazy_loadDictionary(ms, iend-HASH_READ_SIZE);
+        } else {
+            assert(params->useRowMatchFinder != ZSTD_ps_auto);
+            if (params->useRowMatchFinder == ZSTD_ps_enable) {
+                size_t const tagTableSize = ((size_t)1 << params->cParams.hashLog);
+                ZSTD_memset(ms->tagTable, 0, tagTableSize);
+                ZSTD_row_update(ms, iend-HASH_READ_SIZE);
+                DEBUGLOG(4, "Using row-based hash table for lazy dict");
+            } else {
+                ZSTD_insertAndFindFirstIndex(ms, iend-HASH_READ_SIZE);
+                DEBUGLOG(4, "Using chain-based hash table for lazy dict");
+            }
+        }
+        break;
+
+    case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
+    case ZSTD_btopt:
+    case ZSTD_btultra:
+    case ZSTD_btultra2:
+        assert(srcSize >= HASH_READ_SIZE);
+        ZSTD_updateTree(ms, iend-HASH_READ_SIZE, iend);
+        break;
+
+    default:
+        assert(0);  /* not possible : not a valid strategy id */
+    }
+
+    ms->nextToUpdate = (U32)(iend - ms->window.base);
+    return 0;
+}
+
+
+/* Dictionaries that assign zero probability to symbols that show up causes problems
+ * when FSE encoding. Mark dictionaries with zero probability symbols as FSE_repeat_check
+ * and only dictionaries with 100% valid symbols can be assumed valid.
+ */
+static FSE_repeat ZSTD_dictNCountRepeat(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue)
+{
+    U32 s;
+    if (dictMaxSymbolValue < maxSymbolValue) {
+        return FSE_repeat_check;
+    }
+    for (s = 0; s <= maxSymbolValue; ++s) {
+        if (normalizedCounter[s] == 0) {
+            return FSE_repeat_check;
+        }
+    }
+    return FSE_repeat_valid;
+}
+
+size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
+                         const void* const dict, size_t dictSize)
+{
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue = MaxOff;
+    const BYTE* dictPtr = (const BYTE*)dict;    /* skip magic num and dict ID */
+    const BYTE* const dictEnd = dictPtr + dictSize;
+    dictPtr += 8;
+    bs->entropy.huf.repeatMode = HUF_repeat_check;
+
+    {   unsigned maxSymbolValue = 255;
+        unsigned hasZeroWeights = 1;
+        size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.huf.CTable, &maxSymbolValue, dictPtr,
+            dictEnd-dictPtr, &hasZeroWeights);
+
+        /* We only set the loaded table as valid if it contains all non-zero
+         * weights. Otherwise, we set it to check */
+        if (!hasZeroWeights)
+            bs->entropy.huf.repeatMode = HUF_repeat_valid;
+
+        RETURN_ERROR_IF(HUF_isError(hufHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(maxSymbolValue < 255, dictionary_corrupted, "");
+        dictPtr += hufHeaderSize;
+    }
+
+    {   unsigned offcodeLog;
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
+        /* fill all offset symbols to avoid garbage at end of table */
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.offcodeCTable,
+                offcodeNCount, MaxOff, offcodeLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
+        /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.matchlengthCTable,
+                matchlengthNCount, matchlengthMaxValue, matchlengthLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
+        bs->entropy.fse.matchlength_repeatMode = ZSTD_dictNCountRepeat(matchlengthNCount, matchlengthMaxValue, MaxML);
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.litlengthCTable,
+                litlengthNCount, litlengthMaxValue, litlengthLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
+        bs->entropy.fse.litlength_repeatMode = ZSTD_dictNCountRepeat(litlengthNCount, litlengthMaxValue, MaxLL);
+        dictPtr += litlengthHeaderSize;
+    }
+
+    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted, "");
+    bs->rep[0] = MEM_readLE32(dictPtr+0);
+    bs->rep[1] = MEM_readLE32(dictPtr+4);
+    bs->rep[2] = MEM_readLE32(dictPtr+8);
+    dictPtr += 12;
+
+    {   size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
+        U32 offcodeMax = MaxOff;
+        if (dictContentSize <= ((U32)-1) - 128 KB) {
+            U32 const maxOffset = (U32)dictContentSize + 128 KB; /* The maximum offset that must be supported */
+            offcodeMax = ZSTD_highbit32(maxOffset); /* Calculate minimum offset code required to represent maxOffset */
+        }
+        /* All offset values <= dictContentSize + 128 KB must be representable for a valid table */
+        bs->entropy.fse.offcode_repeatMode = ZSTD_dictNCountRepeat(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff));
+
+        /* All repCodes must be <= dictContentSize and != 0 */
+        {   U32 u;
+            for (u=0; u<3; u++) {
+                RETURN_ERROR_IF(bs->rep[u] == 0, dictionary_corrupted, "");
+                RETURN_ERROR_IF(bs->rep[u] > dictContentSize, dictionary_corrupted, "");
+    }   }   }
+
+    return dictPtr - (const BYTE*)dict;
+}
+
+/* Dictionary format :
+ * See :
+ * https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#dictionary-format
+ */
+/*! ZSTD_loadZstdDictionary() :
+ * @return : dictID, or an error code
+ *  assumptions : magic number supposed already checked
+ *                dictSize supposed >= 8
+ */
+static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
+                                      ZSTD_matchState_t* ms,
+                                      ZSTD_cwksp* ws,
+                                      ZSTD_CCtx_params const* params,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictTableLoadMethod_e dtlm,
+                                      ZSTD_tableFillPurpose_e tfp,
+                                      void* workspace)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+    size_t dictID;
+    size_t eSize;
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+    assert(dictSize >= 8);
+    assert(MEM_readLE32(dictPtr) == ZSTD_MAGIC_DICTIONARY);
+
+    dictID = params->fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr + 4 /* skip magic number */ );
+    eSize = ZSTD_loadCEntropy(bs, workspace, dict, dictSize);
+    FORWARD_IF_ERROR(eSize, "ZSTD_loadCEntropy failed");
+    dictPtr += eSize;
+
+    {
+        size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
+        FORWARD_IF_ERROR(ZSTD_loadDictionaryContent(
+            ms, NULL, ws, params, dictPtr, dictContentSize, dtlm, tfp), "");
+    }
+    return dictID;
+}
+
+/** ZSTD_compress_insertDictionary() :
+*   @return : dictID, or an error code */
+static size_t
+ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
+                               ZSTD_matchState_t* ms,
+                               ldmState_t* ls,
+                               ZSTD_cwksp* ws,
+                         const ZSTD_CCtx_params* params,
+                         const void* dict, size_t dictSize,
+                               ZSTD_dictContentType_e dictContentType,
+                               ZSTD_dictTableLoadMethod_e dtlm,
+                               ZSTD_tableFillPurpose_e tfp,
+                               void* workspace)
+{
+    DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
+    if ((dict==NULL) || (dictSize<8)) {
+        RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
+        return 0;
+    }
+
+    ZSTD_reset_compressedBlockState(bs);
+
+    /* dict restricted modes */
+    if (dictContentType == ZSTD_dct_rawContent)
+        return ZSTD_loadDictionaryContent(ms, ls, ws, params, dict, dictSize, dtlm, tfp);
+
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
+        if (dictContentType == ZSTD_dct_auto) {
+            DEBUGLOG(4, "raw content dictionary detected");
+            return ZSTD_loadDictionaryContent(
+                ms, ls, ws, params, dict, dictSize, dtlm, tfp);
+        }
+        RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
+        assert(0);   /* impossible */
+    }
+
+    /* dict as full zstd dictionary */
+    return ZSTD_loadZstdDictionary(
+        bs, ms, ws, params, dict, dictSize, dtlm, tfp, workspace);
+}
+
+#define ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF (128 KB)
+#define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6ULL)
+
+/*! ZSTD_compressBegin_internal() :
+ * Assumption : either @dict OR @cdict (or none) is non-NULL, never both
+ * @return : 0, or an error code */
+static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params, U64 pledgedSrcSize,
+                                    ZSTD_buffered_policy_e zbuff)
+{
+    size_t const dictContentSize = cdict ? cdict->dictContentSize : dictSize;
+#if ZSTD_TRACE
+    cctx->traceCtx = (ZSTD_trace_compress_begin != NULL) ? ZSTD_trace_compress_begin(cctx) : 0;
+#endif
+    DEBUGLOG(4, "ZSTD_compressBegin_internal: wlog=%u", params->cParams.windowLog);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+    if ( (cdict)
+      && (cdict->dictContentSize > 0)
+      && ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
+        || pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
+        || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+        || cdict->compressionLevel == 0)
+      && (params->attachDictPref != ZSTD_dictForceLoad) ) {
+        return ZSTD_resetCCtx_usingCDict(cctx, cdict, params, pledgedSrcSize, zbuff);
+    }
+
+    FORWARD_IF_ERROR( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                     dictContentSize,
+                                     ZSTDcrp_makeClean, zbuff) , "");
+    {   size_t const dictID = cdict ?
+                ZSTD_compress_insertDictionary(
+                        cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                        &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, cdict->dictContent,
+                        cdict->dictContentSize, cdict->dictContentType, dtlm,
+                        ZSTD_tfp_forCCtx, cctx->entropyWorkspace)
+              : ZSTD_compress_insertDictionary(
+                        cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                        &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, dict, dictSize,
+                        dictContentType, dtlm, ZSTD_tfp_forCCtx, cctx->entropyWorkspace);
+        FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
+        assert(dictID <= UINT_MAX);
+        cctx->dictID = (U32)dictID;
+        cctx->dictContentSize = dictContentSize;
+    }
+    return 0;
+}
+
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params,
+                                    unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_compressBegin_advanced_internal: wlog=%u", params->cParams.windowLog);
+    /* compression parameters verification and optimization */
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params->cParams) , "");
+    return ZSTD_compressBegin_internal(cctx,
+                                       dict, dictSize, dictContentType, dtlm,
+                                       cdict,
+                                       params, pledgedSrcSize,
+                                       ZSTDb_not_buffered);
+}
+
+/*! ZSTD_compressBegin_advanced() :
+*   @return : 0, or an error code */
+size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
+                             const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams;
+    ZSTD_CCtxParams_init_internal(&cctxParams, &params, ZSTD_NO_CLEVEL);
+    return ZSTD_compressBegin_advanced_internal(cctx,
+                                            dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast,
+                                            NULL /*cdict*/,
+                                            &cctxParams, pledgedSrcSize);
+}
+
+static size_t
+ZSTD_compressBegin_usingDict_deprecated(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_CCtx_params cctxParams;
+    {   ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_noAttachDict);
+        ZSTD_CCtxParams_init_internal(&cctxParams, &params, (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel);
+    }
+    DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (unsigned)dictSize);
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
+                                       &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
+}
+
+size_t
+ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+{
+    return ZSTD_compressBegin_usingDict_deprecated(cctx, dict, dictSize, compressionLevel);
+}
+
+size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
+{
+    return ZSTD_compressBegin_usingDict_deprecated(cctx, NULL, 0, compressionLevel);
+}
+
+
+/*! ZSTD_writeEpilogue() :
+*   Ends a frame.
+*   @return : nb of bytes written into dst (or an error code) */
+static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    size_t fhSize = 0;
+
+    DEBUGLOG(4, "ZSTD_writeEpilogue");
+    RETURN_ERROR_IF(cctx->stage == ZSTDcs_created, stage_wrong, "init missing");
+
+    /* special case : empty frame */
+    if (cctx->stage == ZSTDcs_init) {
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams, 0, 0);
+        FORWARD_IF_ERROR(fhSize, "ZSTD_writeFrameHeader failed");
+        dstCapacity -= fhSize;
+        op += fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    if (cctx->stage != ZSTDcs_ending) {
+        /* write one last empty block, make it the "last" block */
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1) + 0;
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for epilogue");
+        MEM_writeLE32(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+    }
+
+    if (cctx->appliedParams.fParams.checksumFlag) {
+        U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for checksum");
+        DEBUGLOG(4, "ZSTD_writeEpilogue: write checksum : %08X", (unsigned)checksum);
+        MEM_writeLE32(op, checksum);
+        op += 4;
+    }
+
+    cctx->stage = ZSTDcs_created;  /* return to "created but no init" status */
+    return op-ostart;
+}
+
+void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize)
+{
+#if ZSTD_TRACE
+    if (cctx->traceCtx && ZSTD_trace_compress_end != NULL) {
+        int const streaming = cctx->inBuffSize > 0 || cctx->outBuffSize > 0 || cctx->appliedParams.nbWorkers > 0;
+        ZSTD_Trace trace;
+        ZSTD_memset(&trace, 0, sizeof(trace));
+        trace.version = ZSTD_VERSION_NUMBER;
+        trace.streaming = streaming;
+        trace.dictionaryID = cctx->dictID;
+        trace.dictionarySize = cctx->dictContentSize;
+        trace.uncompressedSize = cctx->consumedSrcSize;
+        trace.compressedSize = cctx->producedCSize + extraCSize;
+        trace.params = &cctx->appliedParams;
+        trace.cctx = cctx;
+        ZSTD_trace_compress_end(cctx->traceCtx, &trace);
+    }
+    cctx->traceCtx = 0;
+#else
+    (void)cctx;
+    (void)extraCSize;
+#endif
+}
+
+size_t ZSTD_compressEnd_public(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize)
+{
+    size_t endResult;
+    size_t const cSize = ZSTD_compressContinue_internal(cctx,
+                                dst, dstCapacity, src, srcSize,
+                                1 /* frame mode */, 1 /* last chunk */);
+    FORWARD_IF_ERROR(cSize, "ZSTD_compressContinue_internal failed");
+    endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
+    FORWARD_IF_ERROR(endResult, "ZSTD_writeEpilogue failed");
+    assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+    if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+        ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
+        DEBUGLOG(4, "end of frame : controlling src size");
+        RETURN_ERROR_IF(
+            cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1,
+            srcSize_wrong,
+             "error : pledgedSrcSize = %u, while realSrcSize = %u",
+            (unsigned)cctx->pledgedSrcSizePlusOne-1,
+            (unsigned)cctx->consumedSrcSize);
+    }
+    ZSTD_CCtx_trace(cctx, endResult);
+    return cSize + endResult;
+}
+
+/* NOTE: Must just wrap ZSTD_compressEnd_public() */
+size_t ZSTD_compressEnd(ZSTD_CCtx* cctx,
+                        void* dst, size_t dstCapacity,
+                  const void* src, size_t srcSize)
+{
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict,size_t dictSize,
+                               ZSTD_parameters params)
+{
+    DEBUGLOG(4, "ZSTD_compress_advanced");
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    ZSTD_CCtxParams_init_internal(&cctx->simpleApiParams, &params, ZSTD_NO_CLEVEL);
+    return ZSTD_compress_advanced_internal(cctx,
+                                           dst, dstCapacity,
+                                           src, srcSize,
+                                           dict, dictSize,
+                                           &cctx->simpleApiParams);
+}
+
+/* Internal */
+size_t ZSTD_compress_advanced_internal(
+        ZSTD_CCtx* cctx,
+        void* dst, size_t dstCapacity,
+        const void* src, size_t srcSize,
+        const void* dict,size_t dictSize,
+        const ZSTD_CCtx_params* params)
+{
+    DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)", (unsigned)srcSize);
+    FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
+                         dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
+                         params, srcSize, ZSTDb_not_buffered) , "");
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict, size_t dictSize,
+                               int compressionLevel)
+{
+    {
+        ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, srcSize, dict ? dictSize : 0, ZSTD_cpm_noAttachDict);
+        assert(params.fParams.contentSizeFlag == 1);
+        ZSTD_CCtxParams_init_internal(&cctx->simpleApiParams, &params, (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT: compressionLevel);
+    }
+    DEBUGLOG(4, "ZSTD_compress_usingDict (srcSize=%u)", (unsigned)srcSize);
+    return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, dict, dictSize, &cctx->simpleApiParams);
+}
+
+size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
+                         void* dst, size_t dstCapacity,
+                   const void* src, size_t srcSize,
+                         int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (unsigned)srcSize);
+    assert(cctx != NULL);
+    return ZSTD_compress_usingDict(cctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
+}
+
+size_t ZSTD_compress(void* dst, size_t dstCapacity,
+               const void* src, size_t srcSize,
+                     int compressionLevel)
+{
+    size_t result;
+#if ZSTD_COMPRESS_HEAPMODE
+    ZSTD_CCtx* cctx = ZSTD_createCCtx();
+    RETURN_ERROR_IF(!cctx, memory_allocation, "ZSTD_createCCtx failed");
+    result = ZSTD_compressCCtx(cctx, dst, dstCapacity, src, srcSize, compressionLevel);
+    ZSTD_freeCCtx(cctx);
+#else
+    ZSTD_CCtx ctxBody;
+    ZSTD_initCCtx(&ctxBody, ZSTD_defaultCMem);
+    result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
+    ZSTD_freeCCtxContent(&ctxBody);   /* can't free ctxBody itself, as it's on stack; free only heap content */
+#endif
+    return result;
+}
+
+
+/* =====  Dictionary API  ===== */
+
+/*! ZSTD_estimateCDictSize_advanced() :
+ *  Estimate amount of memory that will be needed to create a dictionary with following arguments */
+size_t ZSTD_estimateCDictSize_advanced(
+        size_t dictSize, ZSTD_compressionParameters cParams,
+        ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (unsigned)sizeof(ZSTD_CDict));
+    return ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+         + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+         /* enableDedicatedDictSearch == 1 ensures that CDict estimation will not be too small
+          * in case we are using DDS with row-hash. */
+         + ZSTD_sizeof_matchState(&cParams, ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams),
+                                  /* enableDedicatedDictSearch */ 1, /* forCCtx */ 0)
+         + (dictLoadMethod == ZSTD_dlm_byRef ? 0
+            : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void *))));
+}
+
+size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    return ZSTD_estimateCDictSize_advanced(dictSize, cParams, ZSTD_dlm_byCopy);
+}
+
+size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support sizeof on NULL */
+    DEBUGLOG(5, "sizeof(*cdict) : %u", (unsigned)sizeof(*cdict));
+    /* cdict may be in the workspace */
+    return (cdict->workspace.workspace == cdict ? 0 : sizeof(*cdict))
+        + ZSTD_cwksp_sizeof(&cdict->workspace);
+}
+
+static size_t ZSTD_initCDict_internal(
+                    ZSTD_CDict* cdict,
+              const void* dictBuffer, size_t dictSize,
+                    ZSTD_dictLoadMethod_e dictLoadMethod,
+                    ZSTD_dictContentType_e dictContentType,
+                    ZSTD_CCtx_params params)
+{
+    DEBUGLOG(3, "ZSTD_initCDict_internal (dictContentType:%u)", (unsigned)dictContentType);
+    assert(!ZSTD_checkCParams(params.cParams));
+    cdict->matchState.cParams = params.cParams;
+    cdict->matchState.dedicatedDictSearch = params.enableDedicatedDictSearch;
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
+        cdict->dictContent = dictBuffer;
+    } else {
+         void *internalBuffer = ZSTD_cwksp_reserve_object(&cdict->workspace, ZSTD_cwksp_align(dictSize, sizeof(void*)));
+        RETURN_ERROR_IF(!internalBuffer, memory_allocation, "NULL pointer!");
+        cdict->dictContent = internalBuffer;
+        ZSTD_memcpy(internalBuffer, dictBuffer, dictSize);
+    }
+    cdict->dictContentSize = dictSize;
+    cdict->dictContentType = dictContentType;
+
+    cdict->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cdict->workspace, HUF_WORKSPACE_SIZE);
+
+
+    /* Reset the state to no dictionary */
+    ZSTD_reset_compressedBlockState(&cdict->cBlockState);
+    FORWARD_IF_ERROR(ZSTD_reset_matchState(
+        &cdict->matchState,
+        &cdict->workspace,
+        &params.cParams,
+        params.useRowMatchFinder,
+        ZSTDcrp_makeClean,
+        ZSTDirp_reset,
+        ZSTD_resetTarget_CDict), "");
+    /* (Maybe) load the dictionary
+     * Skips loading the dictionary if it is < 8 bytes.
+     */
+    {   params.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+        params.fParams.contentSizeFlag = 1;
+        {   size_t const dictID = ZSTD_compress_insertDictionary(
+                    &cdict->cBlockState, &cdict->matchState, NULL, &cdict->workspace,
+                    &params, cdict->dictContent, cdict->dictContentSize,
+                    dictContentType, ZSTD_dtlm_full, ZSTD_tfp_forCDict, cdict->entropyWorkspace);
+            FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
+            assert(dictID <= (size_t)(U32)-1);
+            cdict->dictID = (U32)dictID;
+        }
+    }
+
+    return 0;
+}
+
+static ZSTD_CDict* ZSTD_createCDict_advanced_internal(size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_compressionParameters cParams,
+                                      ZSTD_paramSwitch_e useRowMatchFinder,
+                                      U32 enableDedicatedDictSearch,
+                                      ZSTD_customMem customMem)
+{
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+
+    {   size_t const workspaceSize =
+            ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict)) +
+            ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE) +
+            ZSTD_sizeof_matchState(&cParams, useRowMatchFinder, enableDedicatedDictSearch, /* forCCtx */ 0) +
+            (dictLoadMethod == ZSTD_dlm_byRef ? 0
+             : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))));
+        void* const workspace = ZSTD_customMalloc(workspaceSize, customMem);
+        ZSTD_cwksp ws;
+        ZSTD_CDict* cdict;
+
+        if (!workspace) {
+            ZSTD_customFree(workspace, customMem);
+            return NULL;
+        }
+
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_dynamic_alloc);
+
+        cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
+        assert(cdict != NULL);
+        ZSTD_cwksp_move(&cdict->workspace, &ws);
+        cdict->customMem = customMem;
+        cdict->compressionLevel = ZSTD_NO_CLEVEL; /* signals advanced API usage */
+        cdict->useRowMatchFinder = useRowMatchFinder;
+        return cdict;
+    }
+}
+
+ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_compressionParameters cParams,
+                                      ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params cctxParams;
+    ZSTD_memset(&cctxParams, 0, sizeof(cctxParams));
+    ZSTD_CCtxParams_init(&cctxParams, 0);
+    cctxParams.cParams = cParams;
+    cctxParams.customMem = customMem;
+    return ZSTD_createCDict_advanced2(
+        dictBuffer, dictSize,
+        dictLoadMethod, dictContentType,
+        &cctxParams, customMem);
+}
+
+ZSTD_CDict* ZSTD_createCDict_advanced2(
+        const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod,
+        ZSTD_dictContentType_e dictContentType,
+        const ZSTD_CCtx_params* originalCctxParams,
+        ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params cctxParams = *originalCctxParams;
+    ZSTD_compressionParameters cParams;
+    ZSTD_CDict* cdict;
+
+    DEBUGLOG(3, "ZSTD_createCDict_advanced2, mode %u", (unsigned)dictContentType);
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    if (cctxParams.enableDedicatedDictSearch) {
+        cParams = ZSTD_dedicatedDictSearch_getCParams(
+            cctxParams.compressionLevel, dictSize);
+        ZSTD_overrideCParams(&cParams, &cctxParams.cParams);
+    } else {
+        cParams = ZSTD_getCParamsFromCCtxParams(
+            &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    }
+
+    if (!ZSTD_dedicatedDictSearch_isSupported(&cParams)) {
+        /* Fall back to non-DDSS params */
+        cctxParams.enableDedicatedDictSearch = 0;
+        cParams = ZSTD_getCParamsFromCCtxParams(
+            &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    }
+
+    DEBUGLOG(3, "ZSTD_createCDict_advanced2: DDS: %u", cctxParams.enableDedicatedDictSearch);
+    cctxParams.cParams = cParams;
+    cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
+
+    cdict = ZSTD_createCDict_advanced_internal(dictSize,
+                        dictLoadMethod, cctxParams.cParams,
+                        cctxParams.useRowMatchFinder, cctxParams.enableDedicatedDictSearch,
+                        customMem);
+
+    if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                    dict, dictSize,
+                                    dictLoadMethod, dictContentType,
+                                    cctxParams) )) {
+        ZSTD_freeCDict(cdict);
+        return NULL;
+    }
+
+    return cdict;
+}
+
+ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dict, dictSize,
+                                                  ZSTD_dlm_byCopy, ZSTD_dct_auto,
+                                                  cParams, ZSTD_defaultCMem);
+    if (cdict)
+        cdict->compressionLevel = (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
+    return cdict;
+}
+
+ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dict, dictSize,
+                                     ZSTD_dlm_byRef, ZSTD_dct_auto,
+                                     cParams, ZSTD_defaultCMem);
+    if (cdict)
+        cdict->compressionLevel = (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
+    return cdict;
+}
+
+size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = cdict->customMem;
+        int cdictInWorkspace = ZSTD_cwksp_owns_buffer(&cdict->workspace, cdict);
+        ZSTD_cwksp_free(&cdict->workspace, cMem);
+        if (!cdictInWorkspace) {
+            ZSTD_customFree(cdict, cMem);
+        }
+        return 0;
+    }
+}
+
+/*! ZSTD_initStaticCDict_advanced() :
+ *  Generate a digested dictionary in provided memory area.
+ *  workspace: The memory area to emplace the dictionary into.
+ *             Provided pointer must 8-bytes aligned.
+ *             It must outlive dictionary usage.
+ *  workspaceSize: Use ZSTD_estimateCDictSize()
+ *                 to determine how large workspace must be.
+ *  cParams : use ZSTD_getCParams() to transform a compression level
+ *            into its relevants cParams.
+ * @return : pointer to ZSTD_CDict*, or NULL if error (size too small)
+ *  Note : there is no corresponding "free" function.
+ *         Since workspace was allocated externally, it must be freed externally.
+ */
+const ZSTD_CDict* ZSTD_initStaticCDict(
+                                 void* workspace, size_t workspaceSize,
+                           const void* dict, size_t dictSize,
+                                 ZSTD_dictLoadMethod_e dictLoadMethod,
+                                 ZSTD_dictContentType_e dictContentType,
+                                 ZSTD_compressionParameters cParams)
+{
+    ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams);
+    /* enableDedicatedDictSearch == 1 ensures matchstate is not too small in case this CDict will be used for DDS + row hash */
+    size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, useRowMatchFinder, /* enableDedicatedDictSearch */ 1, /* forCCtx */ 0);
+    size_t const neededSize = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+                            + (dictLoadMethod == ZSTD_dlm_byRef ? 0
+                               : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))))
+                            + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+                            + matchStateSize;
+    ZSTD_CDict* cdict;
+    ZSTD_CCtx_params params;
+
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+
+    {
+        ZSTD_cwksp ws;
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_static_alloc);
+        cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
+        if (cdict == NULL) return NULL;
+        ZSTD_cwksp_move(&cdict->workspace, &ws);
+    }
+
+    DEBUGLOG(4, "(workspaceSize < neededSize) : (%u < %u) => %u",
+        (unsigned)workspaceSize, (unsigned)neededSize, (unsigned)(workspaceSize < neededSize));
+    if (workspaceSize < neededSize) return NULL;
+
+    ZSTD_CCtxParams_init(&params, 0);
+    params.cParams = cParams;
+    params.useRowMatchFinder = useRowMatchFinder;
+    cdict->useRowMatchFinder = useRowMatchFinder;
+    cdict->compressionLevel = ZSTD_NO_CLEVEL;
+
+    if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                              dict, dictSize,
+                                              dictLoadMethod, dictContentType,
+                                              params) ))
+        return NULL;
+
+    return cdict;
+}
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
+{
+    assert(cdict != NULL);
+    return cdict->matchState.cParams;
+}
+
+/*! ZSTD_getDictID_fromCDict() :
+ *  Provides the dictID of the dictionary loaded into `cdict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+unsigned ZSTD_getDictID_fromCDict(const ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;
+    return cdict->dictID;
+}
+
+/* ZSTD_compressBegin_usingCDict_internal() :
+ * Implementation of various ZSTD_compressBegin_usingCDict* functions.
+ */
+static size_t ZSTD_compressBegin_usingCDict_internal(
+    ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
+    ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams;
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict_internal");
+    RETURN_ERROR_IF(cdict==NULL, dictionary_wrong, "NULL pointer!");
+    /* Initialize the cctxParams from the cdict */
+    {
+        ZSTD_parameters params;
+        params.fParams = fParams;
+        params.cParams = ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
+                        || pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
+                        || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+                        || cdict->compressionLevel == 0 ) ?
+                ZSTD_getCParamsFromCDict(cdict)
+              : ZSTD_getCParams(cdict->compressionLevel,
+                                pledgedSrcSize,
+                                cdict->dictContentSize);
+        ZSTD_CCtxParams_init_internal(&cctxParams, &params, cdict->compressionLevel);
+    }
+    /* Increase window log to fit the entire dictionary and source if the
+     * source size is known. Limit the increase to 19, which is the
+     * window log for compression level 1 with the largest source size.
+     */
+    if (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        U32 const limitedSrcSize = (U32)MIN(pledgedSrcSize, 1U << 19);
+        U32 const limitedSrcLog = limitedSrcSize > 1 ? ZSTD_highbit32(limitedSrcSize - 1) + 1 : 1;
+        cctxParams.cParams.windowLog = MAX(cctxParams.cParams.windowLog, limitedSrcLog);
+    }
+    return ZSTD_compressBegin_internal(cctx,
+                                        NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast,
+                                        cdict,
+                                        &cctxParams, pledgedSrcSize,
+                                        ZSTDb_not_buffered);
+}
+
+
+/* ZSTD_compressBegin_usingCDict_advanced() :
+ * This function is DEPRECATED.
+ * cdict must be != NULL */
+size_t ZSTD_compressBegin_usingCDict_advanced(
+    ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
+    ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
+{
+    return ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, pledgedSrcSize);
+}
+
+/* ZSTD_compressBegin_usingCDict() :
+ * cdict must be != NULL */
+size_t ZSTD_compressBegin_usingCDict_deprecated(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    return ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN);
+}
+
+size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    return ZSTD_compressBegin_usingCDict_deprecated(cctx, cdict);
+}
+
+/*! ZSTD_compress_usingCDict_internal():
+ * Implementation of various ZSTD_compress_usingCDict* functions.
+ */
+static size_t ZSTD_compress_usingCDict_internal(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
+{
+    FORWARD_IF_ERROR(ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, srcSize), ""); /* will check if cdict != NULL */
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+/*! ZSTD_compress_usingCDict_advanced():
+ * This function is DEPRECATED.
+ */
+size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
+{
+    return ZSTD_compress_usingCDict_internal(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
+}
+
+/*! ZSTD_compress_usingCDict() :
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression parameters are decided at CDict creation time
+ *  while frame parameters are hardcoded */
+size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    return ZSTD_compress_usingCDict_internal(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
+}
+
+
+
+/* ******************************************************************
+*  Streaming
+********************************************************************/
+
+ZSTD_CStream* ZSTD_createCStream(void)
+{
+    DEBUGLOG(3, "ZSTD_createCStream");
+    return ZSTD_createCStream_advanced(ZSTD_defaultCMem);
+}
+
+ZSTD_CStream* ZSTD_initStaticCStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticCCtx(workspace, workspaceSize);
+}
+
+ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem)
+{   /* CStream and CCtx are now same object */
+    return ZSTD_createCCtx_advanced(customMem);
+}
+
+size_t ZSTD_freeCStream(ZSTD_CStream* zcs)
+{
+    return ZSTD_freeCCtx(zcs);   /* same object */
+}
+
+
+
+/*======   Initialization   ======*/
+
+size_t ZSTD_CStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_CStreamOutSize(void)
+{
+    return ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ;
+}
+
+static ZSTD_cParamMode_e ZSTD_getCParamMode(ZSTD_CDict const* cdict, ZSTD_CCtx_params const* params, U64 pledgedSrcSize)
+{
+    if (cdict != NULL && ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize))
+        return ZSTD_cpm_attachDict;
+    else
+        return ZSTD_cpm_noAttachDict;
+}
+
+/* ZSTD_resetCStream():
+ * pledgedSrcSize == 0 means "unknown" */
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pss)
+{
+    /* temporary : 0 interpreted as "unknown" during transition period.
+     * Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN.
+     * 0 will be interpreted as "empty" in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (unsigned)pledgedSrcSize);
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    return 0;
+}
+
+/*! ZSTD_initCStream_internal() :
+ *  Note : for lib/compress only. Used by zstdmt_compress.c.
+ *  Assumption 1 : params are valid
+ *  Assumption 2 : either dict, or cdict, is defined, not both */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                    const void* dict, size_t dictSize, const ZSTD_CDict* cdict,
+                    const ZSTD_CCtx_params* params,
+                    unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_internal");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
+    zcs->requestedParams = *params;
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+    if (dict) {
+        FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    } else {
+        /* Dictionary is cleared if !cdict */
+        FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    }
+    return 0;
+}
+
+/* ZSTD_initCStream_usingCDict_advanced() :
+ * same as ZSTD_initCStream_usingCDict(), with control over frame parameters */
+size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
+                                            const ZSTD_CDict* cdict,
+                                            ZSTD_frameParameters fParams,
+                                            unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict_advanced");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    zcs->requestedParams.fParams = fParams;
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    return 0;
+}
+
+/* note : cdict must outlive compression session */
+size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    return 0;
+}
+
+
+/* ZSTD_initCStream_advanced() :
+ * pledgedSrcSize must be exact.
+ * if srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
+ * dict is loaded with default parameters ZSTD_dct_auto and ZSTD_dlm_byCopy. */
+size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
+                                 const void* dict, size_t dictSize,
+                                 ZSTD_parameters params, unsigned long long pss)
+{
+    /* for compatibility with older programs relying on this behavior.
+     * Users should now specify ZSTD_CONTENTSIZE_UNKNOWN.
+     * This line will be removed in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0 && params.fParams.contentSizeFlag==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_initCStream_advanced");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
+    ZSTD_CCtxParams_setZstdParams(&zcs->requestedParams, &params);
+    FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    return 0;
+}
+
+size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingDict");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    return 0;
+}
+
+size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
+{
+    /* temporary : 0 interpreted as "unknown" during transition period.
+     * Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN.
+     * 0 will be interpreted as "empty" in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_initCStream_srcSize");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    return 0;
+}
+
+size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_initCStream");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    return 0;
+}
+
+/*======   Compression   ======*/
+
+static size_t ZSTD_nextInputSizeHint(const ZSTD_CCtx* cctx)
+{
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        return cctx->blockSize - cctx->stableIn_notConsumed;
+    }
+    assert(cctx->appliedParams.inBufferMode == ZSTD_bm_buffered);
+    {   size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
+        if (hintInSize==0) hintInSize = cctx->blockSize;
+        return hintInSize;
+    }
+}
+
+/** ZSTD_compressStream_generic():
+ *  internal function for all *compressStream*() variants
+ * @return : hint size for next input to complete ongoing block */
+static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                                          ZSTD_outBuffer* output,
+                                          ZSTD_inBuffer* input,
+                                          ZSTD_EndDirective const flushMode)
+{
+    const char* const istart = (assert(input != NULL), (const char*)input->src);
+    const char* const iend = (istart != NULL) ? istart + input->size : istart;
+    const char* ip = (istart != NULL) ? istart + input->pos : istart;
+    char* const ostart = (assert(output != NULL), (char*)output->dst);
+    char* const oend = (ostart != NULL) ? ostart + output->size : ostart;
+    char* op = (ostart != NULL) ? ostart + output->pos : ostart;
+    U32 someMoreWork = 1;
+
+    /* check expectations */
+    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%i, srcSize = %zu", (int)flushMode, input->size - input->pos);
+    assert(zcs != NULL);
+    if (zcs->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        assert(input->pos >= zcs->stableIn_notConsumed);
+        input->pos -= zcs->stableIn_notConsumed;
+        ip -= zcs->stableIn_notConsumed;
+        zcs->stableIn_notConsumed = 0;
+    }
+    if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+        assert(zcs->inBuff != NULL);
+        assert(zcs->inBuffSize > 0);
+    }
+    if (zcs->appliedParams.outBufferMode == ZSTD_bm_buffered) {
+        assert(zcs->outBuff !=  NULL);
+        assert(zcs->outBuffSize > 0);
+    }
+    if (input->src == NULL) assert(input->size == 0);
+    assert(input->pos <= input->size);
+    if (output->dst == NULL) assert(output->size == 0);
+    assert(output->pos <= output->size);
+    assert((U32)flushMode <= (U32)ZSTD_e_end);
+
+    while (someMoreWork) {
+        switch(zcs->streamStage)
+        {
+        case zcss_init:
+            RETURN_ERROR(init_missing, "call ZSTD_initCStream() first!");
+
+        case zcss_load:
+            if ( (flushMode == ZSTD_e_end)
+              && ( (size_t)(oend-op) >= ZSTD_compressBound(iend-ip)     /* Enough output space */
+                || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)  /* OR we are allowed to return dstSizeTooSmall */
+              && (zcs->inBuffPos == 0) ) {
+                /* shortcut to compression pass directly into output buffer */
+                size_t const cSize = ZSTD_compressEnd_public(zcs,
+                                                op, oend-op, ip, iend-ip);
+                DEBUGLOG(4, "ZSTD_compressEnd : cSize=%u", (unsigned)cSize);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressEnd failed");
+                ip = iend;
+                op += cSize;
+                zcs->frameEnded = 1;
+                ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+                someMoreWork = 0; break;
+            }
+            /* complete loading into inBuffer in buffered mode */
+            if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+                size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
+                size_t const loaded = ZSTD_limitCopy(
+                                        zcs->inBuff + zcs->inBuffPos, toLoad,
+                                        ip, iend-ip);
+                zcs->inBuffPos += loaded;
+                if (ip) ip += loaded;
+                if ( (flushMode == ZSTD_e_continue)
+                  && (zcs->inBuffPos < zcs->inBuffTarget) ) {
+                    /* not enough input to fill full block : stop here */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (zcs->inBuffPos == zcs->inToCompress) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
+            } else {
+                assert(zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+                if ( (flushMode == ZSTD_e_continue)
+                  && ( (size_t)(iend - ip) < zcs->blockSize) ) {
+                    /* can't compress a full block : stop here */
+                    zcs->stableIn_notConsumed = (size_t)(iend - ip);
+                    ip = iend;  /* pretend to have consumed input */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (ip == iend) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
+            }
+            /* compress current block (note : this stage cannot be stopped in the middle) */
+            DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
+            {   int const inputBuffered = (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered);
+                void* cDst;
+                size_t cSize;
+                size_t oSize = oend-op;
+                size_t const iSize = inputBuffered ? zcs->inBuffPos - zcs->inToCompress
+                                                   : MIN((size_t)(iend - ip), zcs->blockSize);
+                if (oSize >= ZSTD_compressBound(iSize) || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)
+                    cDst = op;   /* compress into output buffer, to skip flush stage */
+                else
+                    cDst = zcs->outBuff, oSize = zcs->outBuffSize;
+                if (inputBuffered) {
+                    unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip==iend);
+                    cSize = lastBlock ?
+                            ZSTD_compressEnd_public(zcs, cDst, oSize,
+                                        zcs->inBuff + zcs->inToCompress, iSize) :
+                            ZSTD_compressContinue_public(zcs, cDst, oSize,
+                                        zcs->inBuff + zcs->inToCompress, iSize);
+                    FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
+                    zcs->frameEnded = lastBlock;
+                    /* prepare next block */
+                    zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
+                    if (zcs->inBuffTarget > zcs->inBuffSize)
+                        zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;
+                    DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
+                            (unsigned)zcs->inBuffTarget, (unsigned)zcs->inBuffSize);
+                    if (!lastBlock)
+                        assert(zcs->inBuffTarget <= zcs->inBuffSize);
+                    zcs->inToCompress = zcs->inBuffPos;
+                } else { /* !inputBuffered, hence ZSTD_bm_stable */
+                    unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip + iSize == iend);
+                    cSize = lastBlock ?
+                            ZSTD_compressEnd_public(zcs, cDst, oSize, ip, iSize) :
+                            ZSTD_compressContinue_public(zcs, cDst, oSize, ip, iSize);
+                    /* Consume the input prior to error checking to mirror buffered mode. */
+                    if (ip) ip += iSize;
+                    FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
+                    zcs->frameEnded = lastBlock;
+                    if (lastBlock) assert(ip == iend);
+                }
+                if (cDst == op) {  /* no need to flush */
+                    op += cSize;
+                    if (zcs->frameEnded) {
+                        DEBUGLOG(5, "Frame completed directly in outBuffer");
+                        someMoreWork = 0;
+                        ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+                    }
+                    break;
+                }
+                zcs->outBuffContentSize = cSize;
+                zcs->outBuffFlushedSize = 0;
+                zcs->streamStage = zcss_flush; /* pass-through to flush stage */
+            }
+	    ZSTD_FALLTHROUGH;
+        case zcss_flush:
+            DEBUGLOG(5, "flush stage");
+            assert(zcs->appliedParams.outBufferMode == ZSTD_bm_buffered);
+            {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
+                size_t const flushed = ZSTD_limitCopy(op, (size_t)(oend-op),
+                            zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+                DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
+                            (unsigned)toFlush, (unsigned)(oend-op), (unsigned)flushed);
+                if (flushed)
+                    op += flushed;
+                zcs->outBuffFlushedSize += flushed;
+                if (toFlush!=flushed) {
+                    /* flush not fully completed, presumably because dst is too small */
+                    assert(op==oend);
+                    someMoreWork = 0;
+                    break;
+                }
+                zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
+                if (zcs->frameEnded) {
+                    DEBUGLOG(5, "Frame completed on flush");
+                    someMoreWork = 0;
+                    ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+                    break;
+                }
+                zcs->streamStage = zcss_load;
+                break;
+            }
+
+        default: /* impossible */
+            assert(0);
+        }
+    }
+
+    input->pos = ip - istart;
+    output->pos = op - ostart;
+    if (zcs->frameEnded) return 0;
+    return ZSTD_nextInputSizeHint(zcs);
+}
+
+static size_t ZSTD_nextInputSizeHint_MTorST(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers >= 1) {
+        assert(cctx->mtctx != NULL);
+        return ZSTDMT_nextInputSizeHint(cctx->mtctx);
+    }
+#endif
+    return ZSTD_nextInputSizeHint(cctx);
+
+}
+
+size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    FORWARD_IF_ERROR( ZSTD_compressStream2(zcs, output, input, ZSTD_e_continue) , "");
+    return ZSTD_nextInputSizeHint_MTorST(zcs);
+}
+
+/* After a compression call set the expected input/output buffer.
+ * This is validated at the start of the next compression call.
+ */
+static void
+ZSTD_setBufferExpectations(ZSTD_CCtx* cctx, const ZSTD_outBuffer* output, const ZSTD_inBuffer* input)
+{
+    DEBUGLOG(5, "ZSTD_setBufferExpectations (for advanced stable in/out modes)");
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        cctx->expectedInBuffer = *input;
+    }
+    if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
+        cctx->expectedOutBufferSize = output->size - output->pos;
+    }
+}
+
+/* Validate that the input/output buffers match the expectations set by
+ * ZSTD_setBufferExpectations.
+ */
+static size_t ZSTD_checkBufferStability(ZSTD_CCtx const* cctx,
+                                        ZSTD_outBuffer const* output,
+                                        ZSTD_inBuffer const* input,
+                                        ZSTD_EndDirective endOp)
+{
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        ZSTD_inBuffer const expect = cctx->expectedInBuffer;
+        if (expect.src != input->src || expect.pos != input->pos)
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableInBuffer enabled but input differs!");
+    }
+    (void)endOp;
+    if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
+        size_t const outBufferSize = output->size - output->pos;
+        if (cctx->expectedOutBufferSize != outBufferSize)
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableOutBuffer enabled but output size differs!");
+    }
+    return 0;
+}
+
+static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
+                                             ZSTD_EndDirective endOp,
+                                             size_t inSize)
+{
+    ZSTD_CCtx_params params = cctx->requestedParams;
+    ZSTD_prefixDict const prefixDict = cctx->prefixDict;
+    FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) , ""); /* Init the local dict if present. */
+    ZSTD_memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));   /* single usage */
+    assert(prefixDict.dict==NULL || cctx->cdict==NULL);    /* only one can be set */
+    if (cctx->cdict && !cctx->localDict.cdict) {
+        /* Let the cdict's compression level take priority over the requested params.
+         * But do not take the cdict's compression level if the "cdict" is actually a localDict
+         * generated from ZSTD_initLocalDict().
+         */
+        params.compressionLevel = cctx->cdict->compressionLevel;
+    }
+    DEBUGLOG(4, "ZSTD_compressStream2 : transparent init stage");
+    if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = inSize + 1;  /* auto-determine pledgedSrcSize */
+
+    {   size_t const dictSize = prefixDict.dict
+                ? prefixDict.dictSize
+                : (cctx->cdict ? cctx->cdict->dictContentSize : 0);
+        ZSTD_cParamMode_e const mode = ZSTD_getCParamMode(cctx->cdict, &params, cctx->pledgedSrcSizePlusOne - 1);
+        params.cParams = ZSTD_getCParamsFromCCtxParams(
+                &params, cctx->pledgedSrcSizePlusOne-1,
+                dictSize, mode);
+    }
+
+    params.useBlockSplitter = ZSTD_resolveBlockSplitterMode(params.useBlockSplitter, &params.cParams);
+    params.ldmParams.enableLdm = ZSTD_resolveEnableLdm(params.ldmParams.enableLdm, &params.cParams);
+    params.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params.useRowMatchFinder, &params.cParams);
+    params.validateSequences = ZSTD_resolveExternalSequenceValidation(params.validateSequences);
+    params.maxBlockSize = ZSTD_resolveMaxBlockSize(params.maxBlockSize);
+    params.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(params.searchForExternalRepcodes, params.compressionLevel);
+
+#ifdef ZSTD_MULTITHREAD
+    /* If external matchfinder is enabled, make sure to fail before checking job size (for consistency) */
+    RETURN_ERROR_IF(
+        params.useSequenceProducer == 1 && params.nbWorkers >= 1,
+        parameter_combination_unsupported,
+        "External sequence producer isn't supported with nbWorkers >= 1"
+    );
+
+    if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
+        params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
+    }
+    if (params.nbWorkers > 0) {
+#if ZSTD_TRACE
+        cctx->traceCtx = (ZSTD_trace_compress_begin != NULL) ? ZSTD_trace_compress_begin(cctx) : 0;
+#endif
+        /* mt context creation */
+        if (cctx->mtctx == NULL) {
+            DEBUGLOG(4, "ZSTD_compressStream2: creating new mtctx for nbWorkers=%u",
+                        params.nbWorkers);
+            cctx->mtctx = ZSTDMT_createCCtx_advanced((U32)params.nbWorkers, cctx->customMem, cctx->pool);
+            RETURN_ERROR_IF(cctx->mtctx == NULL, memory_allocation, "NULL pointer!");
+        }
+        /* mt compression */
+        DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
+        FORWARD_IF_ERROR( ZSTDMT_initCStream_internal(
+                    cctx->mtctx,
+                    prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
+                    cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) , "");
+        cctx->dictID = cctx->cdict ? cctx->cdict->dictID : 0;
+        cctx->dictContentSize = cctx->cdict ? cctx->cdict->dictContentSize : prefixDict.dictSize;
+        cctx->consumedSrcSize = 0;
+        cctx->producedCSize = 0;
+        cctx->streamStage = zcss_load;
+        cctx->appliedParams = params;
+    } else
+#endif  /* ZSTD_MULTITHREAD */
+    {   U64 const pledgedSrcSize = cctx->pledgedSrcSizePlusOne - 1;
+        assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+        FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
+                prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType, ZSTD_dtlm_fast,
+                cctx->cdict,
+                &params, pledgedSrcSize,
+                ZSTDb_buffered) , "");
+        assert(cctx->appliedParams.nbWorkers == 0);
+        cctx->inToCompress = 0;
+        cctx->inBuffPos = 0;
+        if (cctx->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+            /* for small input: avoid automatic flush on reaching end of block, since
+            * it would require to add a 3-bytes null block to end frame
+            */
+            cctx->inBuffTarget = cctx->blockSize + (cctx->blockSize == pledgedSrcSize);
+        } else {
+            cctx->inBuffTarget = 0;
+        }
+        cctx->outBuffContentSize = cctx->outBuffFlushedSize = 0;
+        cctx->streamStage = zcss_load;
+        cctx->frameEnded = 0;
+    }
+    return 0;
+}
+
+/* @return provides a minimum amount of data remaining to be flushed from internal buffers
+ */
+size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
+                             ZSTD_outBuffer* output,
+                             ZSTD_inBuffer* input,
+                             ZSTD_EndDirective endOp)
+{
+    DEBUGLOG(5, "ZSTD_compressStream2, endOp=%u ", (unsigned)endOp);
+    /* check conditions */
+    RETURN_ERROR_IF(output->pos > output->size, dstSize_tooSmall, "invalid output buffer");
+    RETURN_ERROR_IF(input->pos  > input->size, srcSize_wrong, "invalid input buffer");
+    RETURN_ERROR_IF((U32)endOp > (U32)ZSTD_e_end, parameter_outOfBound, "invalid endDirective");
+    assert(cctx != NULL);
+
+    /* transparent initialization stage */
+    if (cctx->streamStage == zcss_init) {
+        size_t const inputSize = input->size - input->pos;  /* no obligation to start from pos==0 */
+        size_t const totalInputSize = inputSize + cctx->stableIn_notConsumed;
+        if ( (cctx->requestedParams.inBufferMode == ZSTD_bm_stable) /* input is presumed stable, across invocations */
+          && (endOp == ZSTD_e_continue)                             /* no flush requested, more input to come */
+          && (totalInputSize < ZSTD_BLOCKSIZE_MAX) ) {              /* not even reached one block yet */
+            if (cctx->stableIn_notConsumed) {  /* not the first time */
+                /* check stable source guarantees */
+                RETURN_ERROR_IF(input->src != cctx->expectedInBuffer.src, stabilityCondition_notRespected, "stableInBuffer condition not respected: wrong src pointer");
+                RETURN_ERROR_IF(input->pos != cctx->expectedInBuffer.size, stabilityCondition_notRespected, "stableInBuffer condition not respected: externally modified pos");
+            }
+            /* pretend input was consumed, to give a sense forward progress */
+            input->pos = input->size;
+            /* save stable inBuffer, for later control, and flush/end */
+            cctx->expectedInBuffer = *input;
+            /* but actually input wasn't consumed, so keep track of position from where compression shall resume */
+            cctx->stableIn_notConsumed += inputSize;
+            /* don't initialize yet, wait for the first block of flush() order, for better parameters adaptation */
+            return ZSTD_FRAMEHEADERSIZE_MIN(cctx->requestedParams.format);  /* at least some header to produce */
+        }
+        FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, endOp, totalInputSize), "compressStream2 initialization failed");
+        ZSTD_setBufferExpectations(cctx, output, input);   /* Set initial buffer expectations now that we've initialized */
+    }
+    /* end of transparent initialization stage */
+
+    FORWARD_IF_ERROR(ZSTD_checkBufferStability(cctx, output, input, endOp), "invalid buffers");
+    /* compression stage */
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        size_t flushMin;
+        if (cctx->cParamsChanged) {
+            ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
+            cctx->cParamsChanged = 0;
+        }
+        if (cctx->stableIn_notConsumed) {
+            assert(cctx->appliedParams.inBufferMode == ZSTD_bm_stable);
+            /* some early data was skipped - make it available for consumption */
+            assert(input->pos >= cctx->stableIn_notConsumed);
+            input->pos -= cctx->stableIn_notConsumed;
+            cctx->stableIn_notConsumed = 0;
+        }
+        for (;;) {
+            size_t const ipos = input->pos;
+            size_t const opos = output->pos;
+            flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
+            cctx->consumedSrcSize += (U64)(input->pos - ipos);
+            cctx->producedCSize += (U64)(output->pos - opos);
+            if ( ZSTD_isError(flushMin)
+              || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
+                if (flushMin == 0)
+                    ZSTD_CCtx_trace(cctx, 0);
+                ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
+            }
+            FORWARD_IF_ERROR(flushMin, "ZSTDMT_compressStream_generic failed");
+
+            if (endOp == ZSTD_e_continue) {
+                /* We only require some progress with ZSTD_e_continue, not maximal progress.
+                 * We're done if we've consumed or produced any bytes, or either buffer is
+                 * full.
+                 */
+                if (input->pos != ipos || output->pos != opos || input->pos == input->size || output->pos == output->size)
+                    break;
+            } else {
+                assert(endOp == ZSTD_e_flush || endOp == ZSTD_e_end);
+                /* We require maximal progress. We're done when the flush is complete or the
+                 * output buffer is full.
+                 */
+                if (flushMin == 0 || output->pos == output->size)
+                    break;
+            }
+        }
+        DEBUGLOG(5, "completed ZSTD_compressStream2 delegating to ZSTDMT_compressStream_generic");
+        /* Either we don't require maximum forward progress, we've finished the
+         * flush, or we are out of output space.
+         */
+        assert(endOp == ZSTD_e_continue || flushMin == 0 || output->pos == output->size);
+        ZSTD_setBufferExpectations(cctx, output, input);
+        return flushMin;
+    }
+#endif /* ZSTD_MULTITHREAD */
+    FORWARD_IF_ERROR( ZSTD_compressStream_generic(cctx, output, input, endOp) , "");
+    DEBUGLOG(5, "completed ZSTD_compressStream2");
+    ZSTD_setBufferExpectations(cctx, output, input);
+    return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
+}
+
+size_t ZSTD_compressStream2_simpleArgs (
+                            ZSTD_CCtx* cctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos,
+                            ZSTD_EndDirective endOp)
+{
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
+    /* ZSTD_compressStream2() will check validity of dstPos and srcPos */
+    {   size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
+}
+
+size_t ZSTD_compress2(ZSTD_CCtx* cctx,
+                      void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTD_bufferMode_e const originalInBufferMode = cctx->requestedParams.inBufferMode;
+    ZSTD_bufferMode_e const originalOutBufferMode = cctx->requestedParams.outBufferMode;
+    DEBUGLOG(4, "ZSTD_compress2 (srcSize=%u)", (unsigned)srcSize);
+    ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
+    /* Enable stable input/output buffers. */
+    cctx->requestedParams.inBufferMode = ZSTD_bm_stable;
+    cctx->requestedParams.outBufferMode = ZSTD_bm_stable;
+    {   size_t oPos = 0;
+        size_t iPos = 0;
+        size_t const result = ZSTD_compressStream2_simpleArgs(cctx,
+                                        dst, dstCapacity, &oPos,
+                                        src, srcSize, &iPos,
+                                        ZSTD_e_end);
+        /* Reset to the original values. */
+        cctx->requestedParams.inBufferMode = originalInBufferMode;
+        cctx->requestedParams.outBufferMode = originalOutBufferMode;
+
+        FORWARD_IF_ERROR(result, "ZSTD_compressStream2_simpleArgs failed");
+        if (result != 0) {  /* compression not completed, due to lack of output space */
+            assert(oPos == dstCapacity);
+            RETURN_ERROR(dstSize_tooSmall, "");
+        }
+        assert(iPos == srcSize);   /* all input is expected consumed */
+        return oPos;
+    }
+}
+
+/* ZSTD_validateSequence() :
+ * @offCode : is presumed to follow format required by ZSTD_storeSeq()
+ * @returns a ZSTD error code if sequence is not valid
+ */
+static size_t
+ZSTD_validateSequence(U32 offCode, U32 matchLength, U32 minMatch,
+                      size_t posInSrc, U32 windowLog, size_t dictSize, int useSequenceProducer)
+{
+    U32 const windowSize = 1u << windowLog;
+    /* posInSrc represents the amount of data the decoder would decode up to this point.
+     * As long as the amount of data decoded is less than or equal to window size, offsets may be
+     * larger than the total length of output decoded in order to reference the dict, even larger than
+     * window size. After output surpasses windowSize, we're limited to windowSize offsets again.
+     */
+    size_t const offsetBound = posInSrc > windowSize ? (size_t)windowSize : posInSrc + (size_t)dictSize;
+    size_t const matchLenLowerBound = (minMatch == 3 || useSequenceProducer) ? 3 : 4;
+    RETURN_ERROR_IF(offCode > OFFSET_TO_OFFBASE(offsetBound), externalSequences_invalid, "Offset too large!");
+    /* Validate maxNbSeq is large enough for the given matchLength and minMatch */
+    RETURN_ERROR_IF(matchLength < matchLenLowerBound, externalSequences_invalid, "Matchlength too small for the minMatch");
+    return 0;
+}
+
+/* Returns an offset code, given a sequence's raw offset, the ongoing repcode array, and whether litLength == 0 */
+static U32 ZSTD_finalizeOffBase(U32 rawOffset, const U32 rep[ZSTD_REP_NUM], U32 ll0)
+{
+    U32 offBase = OFFSET_TO_OFFBASE(rawOffset);
+
+    if (!ll0 && rawOffset == rep[0]) {
+        offBase = REPCODE1_TO_OFFBASE;
+    } else if (rawOffset == rep[1]) {
+        offBase = REPCODE_TO_OFFBASE(2 - ll0);
+    } else if (rawOffset == rep[2]) {
+        offBase = REPCODE_TO_OFFBASE(3 - ll0);
+    } else if (ll0 && rawOffset == rep[0] - 1) {
+        offBase = REPCODE3_TO_OFFBASE;
+    }
+    return offBase;
+}
+
+size_t
+ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
+                                              ZSTD_sequencePosition* seqPos,
+                                        const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                        const void* src, size_t blockSize,
+                                        ZSTD_paramSwitch_e externalRepSearch)
+{
+    U32 idx = seqPos->idx;
+    U32 const startIdx = idx;
+    BYTE const* ip = (BYTE const*)(src);
+    const BYTE* const iend = ip + blockSize;
+    repcodes_t updatedRepcodes;
+    U32 dictSize;
+
+    DEBUGLOG(5, "ZSTD_copySequencesToSeqStoreExplicitBlockDelim (blockSize = %zu)", blockSize);
+
+    if (cctx->cdict) {
+        dictSize = (U32)cctx->cdict->dictContentSize;
+    } else if (cctx->prefixDict.dict) {
+        dictSize = (U32)cctx->prefixDict.dictSize;
+    } else {
+        dictSize = 0;
+    }
+    ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    for (; idx < inSeqsSize && (inSeqs[idx].matchLength != 0 || inSeqs[idx].offset != 0); ++idx) {
+        U32 const litLength = inSeqs[idx].litLength;
+        U32 const matchLength = inSeqs[idx].matchLength;
+        U32 offBase;
+
+        if (externalRepSearch == ZSTD_ps_disable) {
+            offBase = OFFSET_TO_OFFBASE(inSeqs[idx].offset);
+        } else {
+            U32 const ll0 = (litLength == 0);
+            offBase = ZSTD_finalizeOffBase(inSeqs[idx].offset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
+        }
+
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
+        if (cctx->appliedParams.validateSequences) {
+            seqPos->posInSrc += litLength + matchLength;
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch, seqPos->posInSrc,
+                                                cctx->appliedParams.cParams.windowLog, dictSize, cctx->appliedParams.useSequenceProducer),
+                                                "Sequence validation failed");
+        }
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
+                        "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
+        ip += matchLength + litLength;
+    }
+
+    /* If we skipped repcode search while parsing, we need to update repcodes now */
+    assert(externalRepSearch != ZSTD_ps_auto);
+    assert(idx >= startIdx);
+    if (externalRepSearch == ZSTD_ps_disable && idx != startIdx) {
+        U32* const rep = updatedRepcodes.rep;
+        U32 lastSeqIdx = idx - 1; /* index of last non-block-delimiter sequence */
+
+        if (lastSeqIdx >= startIdx + 2) {
+            rep[2] = inSeqs[lastSeqIdx - 2].offset;
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else if (lastSeqIdx == startIdx + 1) {
+            rep[2] = rep[0];
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else {
+            assert(lastSeqIdx == startIdx);
+            rep[2] = rep[1];
+            rep[1] = rep[0];
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        }
+    }
+
+    ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(repcodes_t));
+
+    if (inSeqs[idx].litLength) {
+        DEBUGLOG(6, "Storing last literals of size: %u", inSeqs[idx].litLength);
+        ZSTD_storeLastLiterals(&cctx->seqStore, ip, inSeqs[idx].litLength);
+        ip += inSeqs[idx].litLength;
+        seqPos->posInSrc += inSeqs[idx].litLength;
+    }
+    RETURN_ERROR_IF(ip != iend, externalSequences_invalid, "Blocksize doesn't agree with block delimiter!");
+    seqPos->idx = idx+1;
+    return 0;
+}
+
+size_t
+ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                   const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                   const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch)
+{
+    U32 idx = seqPos->idx;
+    U32 startPosInSequence = seqPos->posInSequence;
+    U32 endPosInSequence = seqPos->posInSequence + (U32)blockSize;
+    size_t dictSize;
+    BYTE const* ip = (BYTE const*)(src);
+    BYTE const* iend = ip + blockSize;  /* May be adjusted if we decide to process fewer than blockSize bytes */
+    repcodes_t updatedRepcodes;
+    U32 bytesAdjustment = 0;
+    U32 finalMatchSplit = 0;
+
+    /* TODO(embg) support fast parsing mode in noBlockDelim mode */
+    (void)externalRepSearch;
+
+    if (cctx->cdict) {
+        dictSize = cctx->cdict->dictContentSize;
+    } else if (cctx->prefixDict.dict) {
+        dictSize = cctx->prefixDict.dictSize;
+    } else {
+        dictSize = 0;
+    }
+    DEBUGLOG(5, "ZSTD_copySequencesToSeqStoreNoBlockDelim: idx: %u PIS: %u blockSize: %zu", idx, startPosInSequence, blockSize);
+    DEBUGLOG(5, "Start seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
+    ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    while (endPosInSequence && idx < inSeqsSize && !finalMatchSplit) {
+        const ZSTD_Sequence currSeq = inSeqs[idx];
+        U32 litLength = currSeq.litLength;
+        U32 matchLength = currSeq.matchLength;
+        U32 const rawOffset = currSeq.offset;
+        U32 offBase;
+
+        /* Modify the sequence depending on where endPosInSequence lies */
+        if (endPosInSequence >= currSeq.litLength + currSeq.matchLength) {
+            if (startPosInSequence >= litLength) {
+                startPosInSequence -= litLength;
+                litLength = 0;
+                matchLength -= startPosInSequence;
+            } else {
+                litLength -= startPosInSequence;
+            }
+            /* Move to the next sequence */
+            endPosInSequence -= currSeq.litLength + currSeq.matchLength;
+            startPosInSequence = 0;
+        } else {
+            /* This is the final (partial) sequence we're adding from inSeqs, and endPosInSequence
+               does not reach the end of the match. So, we have to split the sequence */
+            DEBUGLOG(6, "Require a split: diff: %u, idx: %u PIS: %u",
+                     currSeq.litLength + currSeq.matchLength - endPosInSequence, idx, endPosInSequence);
+            if (endPosInSequence > litLength) {
+                U32 firstHalfMatchLength;
+                litLength = startPosInSequence >= litLength ? 0 : litLength - startPosInSequence;
+                firstHalfMatchLength = endPosInSequence - startPosInSequence - litLength;
+                if (matchLength > blockSize && firstHalfMatchLength >= cctx->appliedParams.cParams.minMatch) {
+                    /* Only ever split the match if it is larger than the block size */
+                    U32 secondHalfMatchLength = currSeq.matchLength + currSeq.litLength - endPosInSequence;
+                    if (secondHalfMatchLength < cctx->appliedParams.cParams.minMatch) {
+                        /* Move the endPosInSequence backward so that it creates match of minMatch length */
+                        endPosInSequence -= cctx->appliedParams.cParams.minMatch - secondHalfMatchLength;
+                        bytesAdjustment = cctx->appliedParams.cParams.minMatch - secondHalfMatchLength;
+                        firstHalfMatchLength -= bytesAdjustment;
+                    }
+                    matchLength = firstHalfMatchLength;
+                    /* Flag that we split the last match - after storing the sequence, exit the loop,
+                       but keep the value of endPosInSequence */
+                    finalMatchSplit = 1;
+                } else {
+                    /* Move the position in sequence backwards so that we don't split match, and break to store
+                     * the last literals. We use the original currSeq.litLength as a marker for where endPosInSequence
+                     * should go. We prefer to do this whenever it is not necessary to split the match, or if doing so
+                     * would cause the first half of the match to be too small
+                     */
+                    bytesAdjustment = endPosInSequence - currSeq.litLength;
+                    endPosInSequence = currSeq.litLength;
+                    break;
+                }
+            } else {
+                /* This sequence ends inside the literals, break to store the last literals */
+                break;
+            }
+        }
+        /* Check if this offset can be represented with a repcode */
+        {   U32 const ll0 = (litLength == 0);
+            offBase = ZSTD_finalizeOffBase(rawOffset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
+        }
+
+        if (cctx->appliedParams.validateSequences) {
+            seqPos->posInSrc += litLength + matchLength;
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch, seqPos->posInSrc,
+                                                   cctx->appliedParams.cParams.windowLog, dictSize, cctx->appliedParams.useSequenceProducer),
+                                                   "Sequence validation failed");
+        }
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
+                        "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
+        ip += matchLength + litLength;
+        if (!finalMatchSplit)
+            idx++; /* Next Sequence */
+    }
+    DEBUGLOG(5, "Ending seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
+    assert(idx == inSeqsSize || endPosInSequence <= inSeqs[idx].litLength + inSeqs[idx].matchLength);
+    seqPos->idx = idx;
+    seqPos->posInSequence = endPosInSequence;
+    ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(repcodes_t));
+
+    iend -= bytesAdjustment;
+    if (ip != iend) {
+        /* Store any last literals */
+        U32 lastLLSize = (U32)(iend - ip);
+        assert(ip <= iend);
+        DEBUGLOG(6, "Storing last literals of size: %u", lastLLSize);
+        ZSTD_storeLastLiterals(&cctx->seqStore, ip, lastLLSize);
+        seqPos->posInSrc += lastLLSize;
+    }
+
+    return bytesAdjustment;
+}
+
+typedef size_t (*ZSTD_sequenceCopier) (ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                       const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                       const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+static ZSTD_sequenceCopier ZSTD_selectSequenceCopier(ZSTD_sequenceFormat_e mode)
+{
+    ZSTD_sequenceCopier sequenceCopier = NULL;
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_blockDelimiters, mode));
+    if (mode == ZSTD_sf_explicitBlockDelimiters) {
+        return ZSTD_copySequencesToSeqStoreExplicitBlockDelim;
+    } else if (mode == ZSTD_sf_noBlockDelimiters) {
+        return ZSTD_copySequencesToSeqStoreNoBlockDelim;
+    }
+    assert(sequenceCopier != NULL);
+    return sequenceCopier;
+}
+
+/* Discover the size of next block by searching for the delimiter.
+ * Note that a block delimiter **must** exist in this mode,
+ * otherwise it's an input error.
+ * The block size retrieved will be later compared to ensure it remains within bounds */
+static size_t
+blockSize_explicitDelimiter(const ZSTD_Sequence* inSeqs, size_t inSeqsSize, ZSTD_sequencePosition seqPos)
+{
+    int end = 0;
+    size_t blockSize = 0;
+    size_t spos = seqPos.idx;
+    DEBUGLOG(6, "blockSize_explicitDelimiter : seq %zu / %zu", spos, inSeqsSize);
+    assert(spos <= inSeqsSize);
+    while (spos < inSeqsSize) {
+        end = (inSeqs[spos].offset == 0);
+        blockSize += inSeqs[spos].litLength + inSeqs[spos].matchLength;
+        if (end) {
+            if (inSeqs[spos].matchLength != 0)
+                RETURN_ERROR(externalSequences_invalid, "delimiter format error : both matchlength and offset must be == 0");
+            break;
+        }
+        spos++;
+    }
+    if (!end)
+        RETURN_ERROR(externalSequences_invalid, "Reached end of sequences without finding a block delimiter");
+    return blockSize;
+}
+
+/* More a "target" block size */
+static size_t blockSize_noDelimiter(size_t blockSize, size_t remaining)
+{
+    int const lastBlock = (remaining <= blockSize);
+    return lastBlock ? remaining : blockSize;
+}
+
+static size_t determine_blockSize(ZSTD_sequenceFormat_e mode,
+                           size_t blockSize, size_t remaining,
+                     const ZSTD_Sequence* inSeqs, size_t inSeqsSize, ZSTD_sequencePosition seqPos)
+{
+    DEBUGLOG(6, "determine_blockSize : remainingSize = %zu", remaining);
+    if (mode == ZSTD_sf_noBlockDelimiters)
+        return blockSize_noDelimiter(blockSize, remaining);
+    {   size_t const explicitBlockSize = blockSize_explicitDelimiter(inSeqs, inSeqsSize, seqPos);
+        FORWARD_IF_ERROR(explicitBlockSize, "Error while determining block size with explicit delimiters");
+        if (explicitBlockSize > blockSize)
+            RETURN_ERROR(externalSequences_invalid, "sequences incorrectly define a too large block");
+        if (explicitBlockSize > remaining)
+            RETURN_ERROR(externalSequences_invalid, "sequences define a frame longer than source");
+        return explicitBlockSize;
+    }
+}
+
+/* Compress, block-by-block, all of the sequences given.
+ *
+ * Returns the cumulative size of all compressed blocks (including their headers),
+ * otherwise a ZSTD error.
+ */
+static size_t
+ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                          const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                          const void* src, size_t srcSize)
+{
+    size_t cSize = 0;
+    size_t remaining = srcSize;
+    ZSTD_sequencePosition seqPos = {0, 0, 0};
+
+    BYTE const* ip = (BYTE const*)src;
+    BYTE* op = (BYTE*)dst;
+    ZSTD_sequenceCopier const sequenceCopier = ZSTD_selectSequenceCopier(cctx->appliedParams.blockDelimiters);
+
+    DEBUGLOG(4, "ZSTD_compressSequences_internal srcSize: %zu, inSeqsSize: %zu", srcSize, inSeqsSize);
+    /* Special case: empty frame */
+    if (remaining == 0) {
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "No room for empty frame block header");
+        MEM_writeLE32(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+        cSize += ZSTD_blockHeaderSize;
+    }
+
+    while (remaining) {
+        size_t compressedSeqsSize;
+        size_t cBlockSize;
+        size_t additionalByteAdjustment;
+        size_t blockSize = determine_blockSize(cctx->appliedParams.blockDelimiters,
+                                        cctx->blockSize, remaining,
+                                        inSeqs, inSeqsSize, seqPos);
+        U32 const lastBlock = (blockSize == remaining);
+        FORWARD_IF_ERROR(blockSize, "Error while trying to determine block size");
+        assert(blockSize <= remaining);
+        ZSTD_resetSeqStore(&cctx->seqStore);
+        DEBUGLOG(5, "Working on new block. Blocksize: %zu (total:%zu)", blockSize, (ip - (const BYTE*)src) + blockSize);
+
+        additionalByteAdjustment = sequenceCopier(cctx, &seqPos, inSeqs, inSeqsSize, ip, blockSize, cctx->appliedParams.searchForExternalRepcodes);
+        FORWARD_IF_ERROR(additionalByteAdjustment, "Bad sequence copy");
+        blockSize -= additionalByteAdjustment;
+
+        /* If blocks are too small, emit as a nocompress block */
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        if (blockSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
+            cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
+            DEBUGLOG(5, "Block too small, writing out nocompress block: cSize: %zu", cBlockSize);
+            cSize += cBlockSize;
+            ip += blockSize;
+            op += cBlockSize;
+            remaining -= blockSize;
+            dstCapacity -= cBlockSize;
+            continue;
+        }
+
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "not enough dstCapacity to write a new compressed block");
+        compressedSeqsSize = ZSTD_entropyCompressSeqStore(&cctx->seqStore,
+                                &cctx->blockState.prevCBlock->entropy, &cctx->blockState.nextCBlock->entropy,
+                                &cctx->appliedParams,
+                                op + ZSTD_blockHeaderSize /* Leave space for block header */, dstCapacity - ZSTD_blockHeaderSize,
+                                blockSize,
+                                cctx->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
+                                cctx->bmi2);
+        FORWARD_IF_ERROR(compressedSeqsSize, "Compressing sequences of block failed");
+        DEBUGLOG(5, "Compressed sequences size: %zu", compressedSeqsSize);
+
+        if (!cctx->isFirstBlock &&
+            ZSTD_maybeRLE(&cctx->seqStore) &&
+            ZSTD_isRLE(ip, blockSize)) {
+            /* We don't want to emit our first block as a RLE even if it qualifies because
+            * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+            * This is only an issue for zstd <= v1.4.3
+            */
+            compressedSeqsSize = 1;
+        }
+
+        if (compressedSeqsSize == 0) {
+            /* ZSTD_noCompressBlock writes the block header as well */
+            cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_noCompressBlock failed");
+            DEBUGLOG(5, "Writing out nocompress block, size: %zu", cBlockSize);
+        } else if (compressedSeqsSize == 1) {
+            cBlockSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_rleCompressBlock failed");
+            DEBUGLOG(5, "Writing out RLE block, size: %zu", cBlockSize);
+        } else {
+            U32 cBlockHeader;
+            /* Error checking and repcodes update */
+            ZSTD_blockState_confirmRepcodesAndEntropyTables(&cctx->blockState);
+            if (cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+                cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+            /* Write block header into beginning of block*/
+            cBlockHeader = lastBlock + (((U32)bt_compressed)<<1) + (U32)(compressedSeqsSize << 3);
+            MEM_writeLE24(op, cBlockHeader);
+            cBlockSize = ZSTD_blockHeaderSize + compressedSeqsSize;
+            DEBUGLOG(5, "Writing out compressed block, size: %zu", cBlockSize);
+        }
+
+        cSize += cBlockSize;
+
+        if (lastBlock) {
+            break;
+        } else {
+            ip += blockSize;
+            op += cBlockSize;
+            remaining -= blockSize;
+            dstCapacity -= cBlockSize;
+            cctx->isFirstBlock = 0;
+        }
+        DEBUGLOG(5, "cSize running total: %zu (remaining dstCapacity=%zu)", cSize, dstCapacity);
+    }
+
+    DEBUGLOG(4, "cSize final total: %zu", cSize);
+    return cSize;
+}
+
+size_t ZSTD_compressSequences(ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                              const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                              const void* src, size_t srcSize)
+{
+    BYTE* op = (BYTE*)dst;
+    size_t cSize = 0;
+    size_t compressedBlocksSize = 0;
+    size_t frameHeaderSize = 0;
+
+    /* Transparent initialization stage, same as compressStream2() */
+    DEBUGLOG(4, "ZSTD_compressSequences (dstCapacity=%zu)", dstCapacity);
+    assert(cctx != NULL);
+    FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, ZSTD_e_end, srcSize), "CCtx initialization failed");
+    /* Begin writing output, starting with frame header */
+    frameHeaderSize = ZSTD_writeFrameHeader(op, dstCapacity, &cctx->appliedParams, srcSize, cctx->dictID);
+    op += frameHeaderSize;
+    dstCapacity -= frameHeaderSize;
+    cSize += frameHeaderSize;
+    if (cctx->appliedParams.fParams.checksumFlag && srcSize) {
+        XXH64_update(&cctx->xxhState, src, srcSize);
+    }
+    /* cSize includes block header size and compressed sequences size */
+    compressedBlocksSize = ZSTD_compressSequences_internal(cctx,
+                                                           op, dstCapacity,
+                                                           inSeqs, inSeqsSize,
+                                                           src, srcSize);
+    FORWARD_IF_ERROR(compressedBlocksSize, "Compressing blocks failed!");
+    cSize += compressedBlocksSize;
+    dstCapacity -= compressedBlocksSize;
+
+    if (cctx->appliedParams.fParams.checksumFlag) {
+        U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for checksum");
+        DEBUGLOG(4, "Write checksum : %08X", (unsigned)checksum);
+        MEM_writeLE32((char*)dst + cSize, checksum);
+        cSize += 4;
+    }
+
+    DEBUGLOG(4, "Final compressed size: %zu", cSize);
+    return cSize;
+}
+
+/*======   Finalize   ======*/
+
+static ZSTD_inBuffer inBuffer_forEndFlush(const ZSTD_CStream* zcs)
+{
+    const ZSTD_inBuffer nullInput = { NULL, 0, 0 };
+    const int stableInput = (zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+    return stableInput ? zcs->expectedInBuffer : nullInput;
+}
+
+/*! ZSTD_flushStream() :
+ * @return : amount of data remaining to flush */
+size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
+    input.size = input.pos; /* do not ingest more input during flush */
+    return ZSTD_compressStream2(zcs, output, &input, ZSTD_e_flush);
+}
+
+
+size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
+    size_t const remainingToFlush = ZSTD_compressStream2(zcs, output, &input, ZSTD_e_end);
+    FORWARD_IF_ERROR(remainingToFlush , "ZSTD_compressStream2(,,ZSTD_e_end) failed");
+    if (zcs->appliedParams.nbWorkers > 0) return remainingToFlush;   /* minimal estimation */
+    /* single thread mode : attempt to calculate remaining to flush more precisely */
+    {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
+        size_t const checksumSize = (size_t)(zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4);
+        size_t const toFlush = remainingToFlush + lastBlockSize + checksumSize;
+        DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (unsigned)toFlush);
+        return toFlush;
+    }
+}
+
+
+/*-=====  Pre-defined compression levels  =====-*/
+#include "clevels.h"
+
+int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
+int ZSTD_minCLevel(void) { return (int)-ZSTD_TARGETLENGTH_MAX; }
+int ZSTD_defaultCLevel(void) { return ZSTD_CLEVEL_DEFAULT; }
+
+static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(int const compressionLevel, size_t const dictSize)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, 0, dictSize, ZSTD_cpm_createCDict);
+    switch (cParams.strategy) {
+        case ZSTD_fast:
+        case ZSTD_dfast:
+            break;
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            cParams.hashLog += ZSTD_LAZY_DDSS_BUCKET_LOG;
+            break;
+        case ZSTD_btlazy2:
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            break;
+    }
+    return cParams;
+}
+
+static int ZSTD_dedicatedDictSearch_isSupported(
+        ZSTD_compressionParameters const* cParams)
+{
+    return (cParams->strategy >= ZSTD_greedy)
+        && (cParams->strategy <= ZSTD_lazy2)
+        && (cParams->hashLog > cParams->chainLog)
+        && (cParams->chainLog <= 24);
+}
+
+/**
+ * Reverses the adjustment applied to cparams when enabling dedicated dict
+ * search. This is used to recover the params set to be used in the working
+ * context. (Otherwise, those tables would also grow.)
+ */
+static void ZSTD_dedicatedDictSearch_revertCParams(
+        ZSTD_compressionParameters* cParams) {
+    switch (cParams->strategy) {
+        case ZSTD_fast:
+        case ZSTD_dfast:
+            break;
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            cParams->hashLog -= ZSTD_LAZY_DDSS_BUCKET_LOG;
+            if (cParams->hashLog < ZSTD_HASHLOG_MIN) {
+                cParams->hashLog = ZSTD_HASHLOG_MIN;
+            }
+            break;
+        case ZSTD_btlazy2:
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            break;
+    }
+}
+
+static U64 ZSTD_getCParamRowSize(U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode)
+{
+    switch (mode) {
+    case ZSTD_cpm_unknown:
+    case ZSTD_cpm_noAttachDict:
+    case ZSTD_cpm_createCDict:
+        break;
+    case ZSTD_cpm_attachDict:
+        dictSize = 0;
+        break;
+    default:
+        assert(0);
+        break;
+    }
+    {   int const unknown = srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN;
+        size_t const addedSize = unknown && dictSize > 0 ? 500 : 0;
+        return unknown && dictSize == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : srcSizeHint+dictSize+addedSize;
+    }
+}
+
+/*! ZSTD_getCParams_internal() :
+ * @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
+ *  Note: srcSizeHint 0 means 0, use ZSTD_CONTENTSIZE_UNKNOWN for unknown.
+ *        Use dictSize == 0 for unknown or unused.
+ *  Note: `mode` controls how we treat the `dictSize`. See docs for `ZSTD_cParamMode_e`. */
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode)
+{
+    U64 const rSize = ZSTD_getCParamRowSize(srcSizeHint, dictSize, mode);
+    U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);
+    int row;
+    DEBUGLOG(5, "ZSTD_getCParams_internal (cLevel=%i)", compressionLevel);
+
+    /* row */
+    if (compressionLevel == 0) row = ZSTD_CLEVEL_DEFAULT;   /* 0 == default */
+    else if (compressionLevel < 0) row = 0;   /* entry 0 is baseline for fast mode */
+    else if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
+    else row = compressionLevel;
+
+    {   ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
+        DEBUGLOG(5, "ZSTD_getCParams_internal selected tableID: %u row: %u strat: %u", tableID, row, (U32)cp.strategy);
+        /* acceleration factor */
+        if (compressionLevel < 0) {
+            int const clampedCompressionLevel = MAX(ZSTD_minCLevel(), compressionLevel);
+            cp.targetLength = (unsigned)(-clampedCompressionLevel);
+        }
+        /* refine parameters based on srcSize & dictSize */
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize, mode, ZSTD_ps_auto);
+    }
+}
+
+/*! ZSTD_getCParams() :
+ * @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
+ *  Size values are optional, provide 0 if not known or unused */
+ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
+{
+    if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
+}
+
+/*! ZSTD_getParams() :
+ *  same idea as ZSTD_getCParams()
+ * @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
+ *  Fields of `ZSTD_frameParameters` are set to default values */
+static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode) {
+    ZSTD_parameters params;
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize, mode);
+    DEBUGLOG(5, "ZSTD_getParams (cLevel=%i)", compressionLevel);
+    ZSTD_memset(&params, 0, sizeof(params));
+    params.cParams = cParams;
+    params.fParams.contentSizeFlag = 1;
+    return params;
+}
+
+/*! ZSTD_getParams() :
+ *  same idea as ZSTD_getCParams()
+ * @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
+ *  Fields of `ZSTD_frameParameters` are set to default values */
+ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
+    if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_getParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
+}
+
+void ZSTD_registerSequenceProducer(
+    ZSTD_CCtx* zc, void* mState,
+    ZSTD_sequenceProducer_F* mFinder
+) {
+    if (mFinder != NULL) {
+        ZSTD_externalMatchCtx emctx;
+        emctx.mState = mState;
+        emctx.mFinder = mFinder;
+        emctx.seqBuffer = NULL;
+        emctx.seqBufferCapacity = 0;
+        zc->externalMatchCtx = emctx;
+        zc->requestedParams.useSequenceProducer = 1;
+    } else {
+        ZSTD_memset(&zc->externalMatchCtx, 0, sizeof(zc->externalMatchCtx));
+        zc->requestedParams.useSequenceProducer = 0;
+    }
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_internal.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_internal.h
new file mode 100644
index 0000000..10f68d0
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_internal.h
@@ -0,0 +1,1532 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* This header contains definitions
+ * that shall **only** be used by modules within lib/compress.
+ */
+
+#ifndef ZSTD_COMPRESS_H
+#define ZSTD_COMPRESS_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "../common/zstd_internal.h"
+#include "zstd_cwksp.h"
+#ifdef ZSTD_MULTITHREAD
+#  include "zstdmt_compress.h"
+#endif
+#include "../common/bits.h" /* ZSTD_highbit32, ZSTD_NbCommonBytes */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-*************************************
+*  Constants
+***************************************/
+#define kSearchStrength      8
+#define HASH_READ_SIZE       8
+#define ZSTD_DUBT_UNSORTED_MARK 1   /* For btlazy2 strategy, index ZSTD_DUBT_UNSORTED_MARK==1 means "unsorted".
+                                       It could be confused for a real successor at index "1", if sorted as larger than its predecessor.
+                                       It's not a big deal though : candidate will just be sorted again.
+                                       Additionally, candidate position 1 will be lost.
+                                       But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
+                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table re-use with a different strategy.
+                                       This constant is required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
+typedef enum { zcss_init=0, zcss_load, zcss_flush } ZSTD_cStreamStage;
+
+typedef struct ZSTD_prefixDict_s {
+    const void* dict;
+    size_t dictSize;
+    ZSTD_dictContentType_e dictContentType;
+} ZSTD_prefixDict;
+
+typedef struct {
+    void* dictBuffer;
+    void const* dict;
+    size_t dictSize;
+    ZSTD_dictContentType_e dictContentType;
+    ZSTD_CDict* cdict;
+} ZSTD_localDict;
+
+typedef struct {
+    HUF_CElt CTable[HUF_CTABLE_SIZE_ST(255)];
+    HUF_repeat repeatMode;
+} ZSTD_hufCTables_t;
+
+typedef struct {
+    FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
+    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
+    FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
+    FSE_repeat offcode_repeatMode;
+    FSE_repeat matchlength_repeatMode;
+    FSE_repeat litlength_repeatMode;
+} ZSTD_fseCTables_t;
+
+typedef struct {
+    ZSTD_hufCTables_t huf;
+    ZSTD_fseCTables_t fse;
+} ZSTD_entropyCTables_t;
+
+/***********************************************
+*  Entropy buffer statistics structs and funcs *
+***********************************************/
+/** ZSTD_hufCTablesMetadata_t :
+ *  Stores Literals Block Type for a super-block in hType, and
+ *  huffman tree description in hufDesBuffer.
+ *  hufDesSize refers to the size of huffman tree description in bytes.
+ *  This metadata is populated in ZSTD_buildBlockEntropyStats_literals() */
+typedef struct {
+    symbolEncodingType_e hType;
+    BYTE hufDesBuffer[ZSTD_MAX_HUF_HEADER_SIZE];
+    size_t hufDesSize;
+} ZSTD_hufCTablesMetadata_t;
+
+/** ZSTD_fseCTablesMetadata_t :
+ *  Stores symbol compression modes for a super-block in {ll, ol, ml}Type, and
+ *  fse tables in fseTablesBuffer.
+ *  fseTablesSize refers to the size of fse tables in bytes.
+ *  This metadata is populated in ZSTD_buildBlockEntropyStats_sequences() */
+typedef struct {
+    symbolEncodingType_e llType;
+    symbolEncodingType_e ofType;
+    symbolEncodingType_e mlType;
+    BYTE fseTablesBuffer[ZSTD_MAX_FSE_HEADERS_SIZE];
+    size_t fseTablesSize;
+    size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_entropyCompressSeqStore_internal() */
+} ZSTD_fseCTablesMetadata_t;
+
+typedef struct {
+    ZSTD_hufCTablesMetadata_t hufMetadata;
+    ZSTD_fseCTablesMetadata_t fseMetadata;
+} ZSTD_entropyCTablesMetadata_t;
+
+/** ZSTD_buildBlockEntropyStats() :
+ *  Builds entropy for the block.
+ *  @return : 0 on success or error code */
+size_t ZSTD_buildBlockEntropyStats(
+                    const seqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                          void* workspace, size_t wkspSize);
+
+/*********************************
+*  Compression internals structs *
+*********************************/
+
+typedef struct {
+    U32 off;            /* Offset sumtype code for the match, using ZSTD_storeSeq() format */
+    U32 len;            /* Raw length of match */
+} ZSTD_match_t;
+
+typedef struct {
+    U32 offset;         /* Offset of sequence */
+    U32 litLength;      /* Length of literals prior to match */
+    U32 matchLength;    /* Raw length of match */
+} rawSeq;
+
+typedef struct {
+  rawSeq* seq;          /* The start of the sequences */
+  size_t pos;           /* The index in seq where reading stopped. pos <= size. */
+  size_t posInSequence; /* The position within the sequence at seq[pos] where reading
+                           stopped. posInSequence <= seq[pos].litLength + seq[pos].matchLength */
+  size_t size;          /* The number of sequences. <= capacity. */
+  size_t capacity;      /* The capacity starting from `seq` pointer */
+} rawSeqStore_t;
+
+typedef struct {
+    U32 idx;            /* Index in array of ZSTD_Sequence */
+    U32 posInSequence;  /* Position within sequence at idx */
+    size_t posInSrc;    /* Number of bytes given by sequences provided so far */
+} ZSTD_sequencePosition;
+
+UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
+
+typedef struct {
+    int price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_optimal_t;
+
+typedef enum { zop_dynamic=0, zop_predef } ZSTD_OptPrice_e;
+
+typedef struct {
+    /* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
+    unsigned* litFreq;           /* table of literals statistics, of size 256 */
+    unsigned* litLengthFreq;     /* table of litLength statistics, of size (MaxLL+1) */
+    unsigned* matchLengthFreq;   /* table of matchLength statistics, of size (MaxML+1) */
+    unsigned* offCodeFreq;       /* table of offCode statistics, of size (MaxOff+1) */
+    ZSTD_match_t* matchTable;    /* list of found matches, of size ZSTD_OPT_NUM+1 */
+    ZSTD_optimal_t* priceTable;  /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
+
+    U32  litSum;                 /* nb of literals */
+    U32  litLengthSum;           /* nb of litLength codes */
+    U32  matchLengthSum;         /* nb of matchLength codes */
+    U32  offCodeSum;             /* nb of offset codes */
+    U32  litSumBasePrice;        /* to compare to log2(litfreq) */
+    U32  litLengthSumBasePrice;  /* to compare to log2(llfreq)  */
+    U32  matchLengthSumBasePrice;/* to compare to log2(mlfreq)  */
+    U32  offCodeSumBasePrice;    /* to compare to log2(offreq)  */
+    ZSTD_OptPrice_e priceType;   /* prices can be determined dynamically, or follow a pre-defined cost structure */
+    const ZSTD_entropyCTables_t* symbolCosts;  /* pre-calculated dictionary statistics */
+    ZSTD_paramSwitch_e literalCompressionMode;
+} optState_t;
+
+typedef struct {
+  ZSTD_entropyCTables_t entropy;
+  U32 rep[ZSTD_REP_NUM];
+} ZSTD_compressedBlockState_t;
+
+typedef struct {
+    BYTE const* nextSrc;       /* next block here to continue on current prefix */
+    BYTE const* base;          /* All regular indexes relative to this position */
+    BYTE const* dictBase;      /* extDict indexes relative to this position */
+    U32 dictLimit;             /* below that point, need extDict */
+    U32 lowLimit;              /* below that point, no more valid data */
+    U32 nbOverflowCorrections; /* Number of times overflow correction has run since
+                                * ZSTD_window_init(). Useful for debugging coredumps
+                                * and for ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY.
+                                */
+} ZSTD_window_t;
+
+#define ZSTD_WINDOW_START_INDEX 2
+
+typedef struct ZSTD_matchState_t ZSTD_matchState_t;
+
+#define ZSTD_ROW_HASH_CACHE_SIZE 8       /* Size of prefetching hash cache for row-based matchfinder */
+
+struct ZSTD_matchState_t {
+    ZSTD_window_t window;   /* State for window round buffer management */
+    U32 loadedDictEnd;      /* index of end of dictionary, within context's referential.
+                             * When loadedDictEnd != 0, a dictionary is in use, and still valid.
+                             * This relies on a mechanism to set loadedDictEnd=0 when dictionary is no longer within distance.
+                             * Such mechanism is provided within ZSTD_window_enforceMaxDist() and ZSTD_checkDictValidity().
+                             * When dict referential is copied into active context (i.e. not attached),
+                             * loadedDictEnd == dictSize, since referential starts from zero.
+                             */
+    U32 nextToUpdate;       /* index from which to continue table update */
+    U32 hashLog3;           /* dispatch table for matches of len==3 : larger == faster, more memory */
+
+    U32 rowHashLog;                          /* For row-based matchfinder: Hashlog based on nb of rows in the hashTable.*/
+    BYTE* tagTable;                          /* For row-based matchFinder: A row-based table containing the hashes and head index. */
+    U32 hashCache[ZSTD_ROW_HASH_CACHE_SIZE]; /* For row-based matchFinder: a cache of hashes to improve speed */
+    U64 hashSalt;                            /* For row-based matchFinder: salts the hash for re-use of tag table */
+    U32 hashSaltEntropy;                     /* For row-based matchFinder: collects entropy for salt generation */
+
+    U32* hashTable;
+    U32* hashTable3;
+    U32* chainTable;
+
+    U32 forceNonContiguous; /* Non-zero if we should force non-contiguous load for the next window update. */
+
+    int dedicatedDictSearch;  /* Indicates whether this matchState is using the
+                               * dedicated dictionary search structure.
+                               */
+    optState_t opt;         /* optimal parser state */
+    const ZSTD_matchState_t* dictMatchState;
+    ZSTD_compressionParameters cParams;
+    const rawSeqStore_t* ldmSeqStore;
+
+    /* Controls prefetching in some dictMatchState matchfinders.
+     * This behavior is controlled from the cctx ms.
+     * This parameter has no effect in the cdict ms. */
+    int prefetchCDictTables;
+
+    /* When == 0, lazy match finders insert every position.
+     * When != 0, lazy match finders only insert positions they search.
+     * This allows them to skip much faster over incompressible data,
+     * at a small cost to compression ratio.
+     */
+    int lazySkipping;
+};
+
+typedef struct {
+    ZSTD_compressedBlockState_t* prevCBlock;
+    ZSTD_compressedBlockState_t* nextCBlock;
+    ZSTD_matchState_t matchState;
+} ZSTD_blockState_t;
+
+typedef struct {
+    U32 offset;
+    U32 checksum;
+} ldmEntry_t;
+
+typedef struct {
+    BYTE const* split;
+    U32 hash;
+    U32 checksum;
+    ldmEntry_t* bucket;
+} ldmMatchCandidate_t;
+
+#define LDM_BATCH_SIZE 64
+
+typedef struct {
+    ZSTD_window_t window;   /* State for the window round buffer management */
+    ldmEntry_t* hashTable;
+    U32 loadedDictEnd;
+    BYTE* bucketOffsets;    /* Next position in bucket to insert entry */
+    size_t splitIndices[LDM_BATCH_SIZE];
+    ldmMatchCandidate_t matchCandidates[LDM_BATCH_SIZE];
+} ldmState_t;
+
+typedef struct {
+    ZSTD_paramSwitch_e enableLdm; /* ZSTD_ps_enable to enable LDM. ZSTD_ps_auto by default */
+    U32 hashLog;            /* Log size of hashTable */
+    U32 bucketSizeLog;      /* Log bucket size for collision resolution, at most 8 */
+    U32 minMatchLength;     /* Minimum match length */
+    U32 hashRateLog;       /* Log number of entries to skip */
+    U32 windowLog;          /* Window log for the LDM */
+} ldmParams_t;
+
+typedef struct {
+    int collectSequences;
+    ZSTD_Sequence* seqStart;
+    size_t seqIndex;
+    size_t maxSequences;
+} SeqCollector;
+
+struct ZSTD_CCtx_params_s {
+    ZSTD_format_e format;
+    ZSTD_compressionParameters cParams;
+    ZSTD_frameParameters fParams;
+
+    int compressionLevel;
+    int forceWindow;           /* force back-references to respect limit of
+                                * 1<<wLog, even for dictionary */
+    size_t targetCBlockSize;   /* Tries to fit compressed block size to be around targetCBlockSize.
+                                * No target when targetCBlockSize == 0.
+                                * There is no guarantee on compressed block size */
+    int srcSizeHint;           /* User's best guess of source size.
+                                * Hint is not valid when srcSizeHint == 0.
+                                * There is no guarantee that hint is close to actual source size */
+
+    ZSTD_dictAttachPref_e attachDictPref;
+    ZSTD_paramSwitch_e literalCompressionMode;
+
+    /* Multithreading: used to pass parameters to mtctx */
+    int nbWorkers;
+    size_t jobSize;
+    int overlapLog;
+    int rsyncable;
+
+    /* Long distance matching parameters */
+    ldmParams_t ldmParams;
+
+    /* Dedicated dict search algorithm trigger */
+    int enableDedicatedDictSearch;
+
+    /* Input/output buffer modes */
+    ZSTD_bufferMode_e inBufferMode;
+    ZSTD_bufferMode_e outBufferMode;
+
+    /* Sequence compression API */
+    ZSTD_sequenceFormat_e blockDelimiters;
+    int validateSequences;
+
+    /* Block splitting */
+    ZSTD_paramSwitch_e useBlockSplitter;
+
+    /* Param for deciding whether to use row-based matchfinder */
+    ZSTD_paramSwitch_e useRowMatchFinder;
+
+    /* Always load a dictionary in ext-dict mode (not prefix mode)? */
+    int deterministicRefPrefix;
+
+    /* Internal use, for createCCtxParams() and freeCCtxParams() only */
+    ZSTD_customMem customMem;
+
+    /* Controls prefetching in some dictMatchState matchfinders */
+    ZSTD_paramSwitch_e prefetchCDictTables;
+
+    /* Controls whether zstd will fall back to an internal matchfinder
+     * if the external matchfinder returns an error code. */
+    int enableMatchFinderFallback;
+
+    /* Indicates whether an external matchfinder has been referenced.
+     * Users can't set this externally.
+     * It is set internally in ZSTD_registerSequenceProducer(). */
+    int useSequenceProducer;
+
+    /* Adjust the max block size*/
+    size_t maxBlockSize;
+
+    /* Controls repcode search in external sequence parsing */
+    ZSTD_paramSwitch_e searchForExternalRepcodes;
+};  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
+
+#define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
+#define ENTROPY_WORKSPACE_SIZE (HUF_WORKSPACE_SIZE + COMPRESS_SEQUENCES_WORKSPACE_SIZE)
+
+/**
+ * Indicates whether this compression proceeds directly from user-provided
+ * source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
+ * whether the context needs to buffer the input/output (ZSTDb_buffered).
+ */
+typedef enum {
+    ZSTDb_not_buffered,
+    ZSTDb_buffered
+} ZSTD_buffered_policy_e;
+
+/**
+ * Struct that contains all elements of block splitter that should be allocated
+ * in a wksp.
+ */
+#define ZSTD_MAX_NB_BLOCK_SPLITS 196
+typedef struct {
+    seqStore_t fullSeqStoreChunk;
+    seqStore_t firstHalfSeqStore;
+    seqStore_t secondHalfSeqStore;
+    seqStore_t currSeqStore;
+    seqStore_t nextSeqStore;
+
+    U32 partitions[ZSTD_MAX_NB_BLOCK_SPLITS];
+    ZSTD_entropyCTablesMetadata_t entropyMetadata;
+} ZSTD_blockSplitCtx;
+
+/* Context for block-level external matchfinder API */
+typedef struct {
+  void* mState;
+  ZSTD_sequenceProducer_F* mFinder;
+  ZSTD_Sequence* seqBuffer;
+  size_t seqBufferCapacity;
+} ZSTD_externalMatchCtx;
+
+struct ZSTD_CCtx_s {
+    ZSTD_compressionStage_e stage;
+    int cParamsChanged;                  /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
+    int bmi2;                            /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+    ZSTD_CCtx_params requestedParams;
+    ZSTD_CCtx_params appliedParams;
+    ZSTD_CCtx_params simpleApiParams;    /* Param storage used by the simple API - not sticky. Must only be used in top-level simple API functions for storage. */
+    U32   dictID;
+    size_t dictContentSize;
+
+    ZSTD_cwksp workspace; /* manages buffer for dynamic allocations */
+    size_t blockSize;
+    unsigned long long pledgedSrcSizePlusOne;  /* this way, 0 (default) == unknown */
+    unsigned long long consumedSrcSize;
+    unsigned long long producedCSize;
+    XXH64_state_t xxhState;
+    ZSTD_customMem customMem;
+    ZSTD_threadPool* pool;
+    size_t staticSize;
+    SeqCollector seqCollector;
+    int isFirstBlock;
+    int initialized;
+
+    seqStore_t seqStore;      /* sequences storage ptrs */
+    ldmState_t ldmState;      /* long distance matching state */
+    rawSeq* ldmSequences;     /* Storage for the ldm output sequences */
+    size_t maxNbLdmSequences;
+    rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
+    ZSTD_blockState_t blockState;
+    U32* entropyWorkspace;  /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
+
+    /* Whether we are streaming or not */
+    ZSTD_buffered_policy_e bufferedPolicy;
+
+    /* streaming */
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inToCompress;
+    size_t inBuffPos;
+    size_t inBuffTarget;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outBuffContentSize;
+    size_t outBuffFlushedSize;
+    ZSTD_cStreamStage streamStage;
+    U32    frameEnded;
+
+    /* Stable in/out buffer verification */
+    ZSTD_inBuffer expectedInBuffer;
+    size_t stableIn_notConsumed; /* nb bytes within stable input buffer that are said to be consumed but are not */
+    size_t expectedOutBufferSize;
+
+    /* Dictionary */
+    ZSTD_localDict localDict;
+    const ZSTD_CDict* cdict;
+    ZSTD_prefixDict prefixDict;   /* single-usage dictionary */
+
+    /* Multi-threading */
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_CCtx* mtctx;
+#endif
+
+    /* Tracing */
+#if ZSTD_TRACE
+    ZSTD_TraceCtx traceCtx;
+#endif
+
+    /* Workspace for block splitter */
+    ZSTD_blockSplitCtx blockSplitCtx;
+
+    /* Workspace for external matchfinder */
+    ZSTD_externalMatchCtx externalMatchCtx;
+};
+
+typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
+typedef enum { ZSTD_tfp_forCCtx, ZSTD_tfp_forCDict } ZSTD_tableFillPurpose_e;
+
+typedef enum {
+    ZSTD_noDict = 0,
+    ZSTD_extDict = 1,
+    ZSTD_dictMatchState = 2,
+    ZSTD_dedicatedDictSearch = 3
+} ZSTD_dictMode_e;
+
+typedef enum {
+    ZSTD_cpm_noAttachDict = 0,  /* Compression with ZSTD_noDict or ZSTD_extDict.
+                                 * In this mode we use both the srcSize and the dictSize
+                                 * when selecting and adjusting parameters.
+                                 */
+    ZSTD_cpm_attachDict = 1,    /* Compression with ZSTD_dictMatchState or ZSTD_dedicatedDictSearch.
+                                 * In this mode we only take the srcSize into account when selecting
+                                 * and adjusting parameters.
+                                 */
+    ZSTD_cpm_createCDict = 2,   /* Creating a CDict.
+                                 * In this mode we take both the source size and the dictionary size
+                                 * into account when selecting and adjusting the parameters.
+                                 */
+    ZSTD_cpm_unknown = 3        /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
+                                 * We don't know what these parameters are for. We default to the legacy
+                                 * behavior of taking both the source size and the dict size into account
+                                 * when selecting and adjusting parameters.
+                                 */
+} ZSTD_cParamMode_e;
+
+typedef size_t (*ZSTD_blockCompressor) (
+        ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
+
+
+MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
+{
+    static const BYTE LL_Code[64] = {  0,  1,  2,  3,  4,  5,  6,  7,
+                                       8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 16, 17, 17, 18, 18, 19, 19,
+                                      20, 20, 20, 20, 21, 21, 21, 21,
+                                      22, 22, 22, 22, 22, 22, 22, 22,
+                                      23, 23, 23, 23, 23, 23, 23, 23,
+                                      24, 24, 24, 24, 24, 24, 24, 24,
+                                      24, 24, 24, 24, 24, 24, 24, 24 };
+    static const U32 LL_deltaCode = 19;
+    return (litLength > 63) ? ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
+}
+
+/* ZSTD_MLcode() :
+ * note : mlBase = matchLength - MINMATCH;
+ *        because it's the format it's stored in seqStore->sequences */
+MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
+{
+    static const BYTE ML_Code[128] = { 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
+                                      32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
+                                      38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
+                                      40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
+                                      41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
+    static const U32 ML_deltaCode = 36;
+    return (mlBase > 127) ? ZSTD_highbit32(mlBase) + ML_deltaCode : ML_Code[mlBase];
+}
+
+/* ZSTD_cParam_withinBounds:
+ * @return 1 if value is within cParam bounds,
+ * 0 otherwise */
+MEM_STATIC int ZSTD_cParam_withinBounds(ZSTD_cParameter cParam, int value)
+{
+    ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
+    if (ZSTD_isError(bounds.error)) return 0;
+    if (value < bounds.lowerBound) return 0;
+    if (value > bounds.upperBound) return 0;
+    return 1;
+}
+
+/* ZSTD_noCompressBlock() :
+ * Writes uncompressed block to dst buffer from given src.
+ * Returns the size of the block */
+MEM_STATIC size_t
+ZSTD_noCompressBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
+{
+    U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
+    DEBUGLOG(5, "ZSTD_noCompressBlock (srcSize=%zu, dstCapacity=%zu)", srcSize, dstCapacity);
+    RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
+                    dstSize_tooSmall, "dst buf too small for uncompressed block");
+    MEM_writeLE24(dst, cBlockHeader24);
+    ZSTD_memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
+    return ZSTD_blockHeaderSize + srcSize;
+}
+
+MEM_STATIC size_t
+ZSTD_rleCompressBlock(void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
+{
+    BYTE* const op = (BYTE*)dst;
+    U32 const cBlockHeader = lastBlock + (((U32)bt_rle)<<1) + (U32)(srcSize << 3);
+    RETURN_ERROR_IF(dstCapacity < 4, dstSize_tooSmall, "");
+    MEM_writeLE24(op, cBlockHeader);
+    op[3] = src;
+    return 4;
+}
+
+
+/* ZSTD_minGain() :
+ * minimum compression required
+ * to generate a compress block or a compressed literals section.
+ * note : use same formula for both situations */
+MEM_STATIC size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
+{
+    U32 const minlog = (strat>=ZSTD_btultra) ? (U32)(strat) - 1 : 6;
+    ZSTD_STATIC_ASSERT(ZSTD_btultra == 8);
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, (int)strat));
+    return (srcSize >> minlog) + 2;
+}
+
+MEM_STATIC int ZSTD_literalsCompressionIsDisabled(const ZSTD_CCtx_params* cctxParams)
+{
+    switch (cctxParams->literalCompressionMode) {
+    case ZSTD_ps_enable:
+        return 0;
+    case ZSTD_ps_disable:
+        return 1;
+    default:
+        assert(0 /* impossible: pre-validated */);
+        ZSTD_FALLTHROUGH;
+    case ZSTD_ps_auto:
+        return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
+    }
+}
+
+/*! ZSTD_safecopyLiterals() :
+ *  memcpy() function that won't read beyond more than WILDCOPY_OVERLENGTH bytes past ilimit_w.
+ *  Only called when the sequence ends past ilimit_w, so it only needs to be optimized for single
+ *  large copies.
+ */
+static void
+ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const iend, BYTE const* ilimit_w)
+{
+    assert(iend > ilimit_w);
+    if (ip <= ilimit_w) {
+        ZSTD_wildcopy(op, ip, ilimit_w - ip, ZSTD_no_overlap);
+        op += ilimit_w - ip;
+        ip = ilimit_w;
+    }
+    while (ip < iend) *op++ = *ip++;
+}
+
+
+#define REPCODE1_TO_OFFBASE REPCODE_TO_OFFBASE(1)
+#define REPCODE2_TO_OFFBASE REPCODE_TO_OFFBASE(2)
+#define REPCODE3_TO_OFFBASE REPCODE_TO_OFFBASE(3)
+#define REPCODE_TO_OFFBASE(r) (assert((r)>=1), assert((r)<=ZSTD_REP_NUM), (r)) /* accepts IDs 1,2,3 */
+#define OFFSET_TO_OFFBASE(o)  (assert((o)>0), o + ZSTD_REP_NUM)
+#define OFFBASE_IS_OFFSET(o)  ((o) > ZSTD_REP_NUM)
+#define OFFBASE_IS_REPCODE(o) ( 1 <= (o) && (o) <= ZSTD_REP_NUM)
+#define OFFBASE_TO_OFFSET(o)  (assert(OFFBASE_IS_OFFSET(o)), (o) - ZSTD_REP_NUM)
+#define OFFBASE_TO_REPCODE(o) (assert(OFFBASE_IS_REPCODE(o)), (o))  /* returns ID 1,2,3 */
+
+/*! ZSTD_storeSeq() :
+ *  Store a sequence (litlen, litPtr, offBase and matchLength) into seqStore_t.
+ *  @offBase : Users should employ macros REPCODE_TO_OFFBASE() and OFFSET_TO_OFFBASE().
+ *  @matchLength : must be >= MINMATCH
+ *  Allowed to over-read literals up to litLimit.
+*/
+HINT_INLINE UNUSED_ATTR void
+ZSTD_storeSeq(seqStore_t* seqStorePtr,
+              size_t litLength, const BYTE* literals, const BYTE* litLimit,
+              U32 offBase,
+              size_t matchLength)
+{
+    BYTE const* const litLimit_w = litLimit - WILDCOPY_OVERLENGTH;
+    BYTE const* const litEnd = literals + litLength;
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL >= 6)
+    static const BYTE* g_start = NULL;
+    if (g_start==NULL) g_start = (const BYTE*)literals;  /* note : index only works for compression within a single segment */
+    {   U32 const pos = (U32)((const BYTE*)literals - g_start);
+        DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offBase%7u",
+               pos, (U32)litLength, (U32)matchLength, (U32)offBase);
+    }
+#endif
+    assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
+    /* copy Literals */
+    assert(seqStorePtr->maxNbLit <= 128 KB);
+    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + seqStorePtr->maxNbLit);
+    assert(literals + litLength <= litLimit);
+    if (litEnd <= litLimit_w) {
+        /* Common case we can use wildcopy.
+         * First copy 16 bytes, because literals are likely short.
+         */
+        ZSTD_STATIC_ASSERT(WILDCOPY_OVERLENGTH >= 16);
+        ZSTD_copy16(seqStorePtr->lit, literals);
+        if (litLength > 16) {
+            ZSTD_wildcopy(seqStorePtr->lit+16, literals+16, (ptrdiff_t)litLength-16, ZSTD_no_overlap);
+        }
+    } else {
+        ZSTD_safecopyLiterals(seqStorePtr->lit, literals, litEnd, litLimit_w);
+    }
+    seqStorePtr->lit += litLength;
+
+    /* literal Length */
+    if (litLength>0xFFFF) {
+        assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
+        seqStorePtr->longLengthType = ZSTD_llt_literalLength;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].litLength = (U16)litLength;
+
+    /* match offset */
+    seqStorePtr->sequences[0].offBase = offBase;
+
+    /* match Length */
+    assert(matchLength >= MINMATCH);
+    {   size_t const mlBase = matchLength - MINMATCH;
+        if (mlBase>0xFFFF) {
+            assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
+            seqStorePtr->longLengthType = ZSTD_llt_matchLength;
+            seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+        }
+        seqStorePtr->sequences[0].mlBase = (U16)mlBase;
+    }
+
+    seqStorePtr->sequences++;
+}
+
+/* ZSTD_updateRep() :
+ * updates in-place @rep (array of repeat offsets)
+ * @offBase : sum-type, using numeric representation of ZSTD_storeSeq()
+ */
+MEM_STATIC void
+ZSTD_updateRep(U32 rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
+{
+    if (OFFBASE_IS_OFFSET(offBase)) {  /* full offset */
+        rep[2] = rep[1];
+        rep[1] = rep[0];
+        rep[0] = OFFBASE_TO_OFFSET(offBase);
+    } else {   /* repcode */
+        U32 const repCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;
+        if (repCode > 0) {  /* note : if repCode==0, no change */
+            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            rep[2] = (repCode >= 2) ? rep[1] : rep[2];
+            rep[1] = rep[0];
+            rep[0] = currentOffset;
+        } else {   /* repCode == 0 */
+            /* nothing to do */
+        }
+    }
+}
+
+typedef struct repcodes_s {
+    U32 rep[3];
+} repcodes_t;
+
+MEM_STATIC repcodes_t
+ZSTD_newRep(U32 const rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
+{
+    repcodes_t newReps;
+    ZSTD_memcpy(&newReps, rep, sizeof(newReps));
+    ZSTD_updateRep(newReps.rep, offBase, ll0);
+    return newReps;
+}
+
+
+/*-*************************************
+*  Match length counter
+***************************************/
+MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
+{
+    const BYTE* const pStart = pIn;
+    const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
+
+    if (pIn < pInLoopLimit) {
+        { size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+          if (diff) return ZSTD_NbCommonBytes(diff); }
+        pIn+=sizeof(size_t); pMatch+=sizeof(size_t);
+        while (pIn < pInLoopLimit) {
+            size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+            if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
+            pIn += ZSTD_NbCommonBytes(diff);
+            return (size_t)(pIn - pStart);
+    }   }
+    if (MEM_64bits() && (pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (size_t)(pIn - pStart);
+}
+
+/** ZSTD_count_2segments() :
+ *  can count match length with `ip` & `match` in 2 different segments.
+ *  convention : on reaching mEnd, match count continue starting from iStart
+ */
+MEM_STATIC size_t
+ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
+                     const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
+{
+    const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
+    size_t const matchLength = ZSTD_count(ip, match, vEnd);
+    if (match + matchLength != mEnd) return matchLength;
+    DEBUGLOG(7, "ZSTD_count_2segments: found a 2-parts match (current length==%zu)", matchLength);
+    DEBUGLOG(7, "distance from match beginning to end dictionary = %zi", mEnd - match);
+    DEBUGLOG(7, "distance from current pos to end buffer = %zi", iEnd - ip);
+    DEBUGLOG(7, "next byte : ip==%02X, istart==%02X", ip[matchLength], *iStart);
+    DEBUGLOG(7, "final match length = %zu", matchLength + ZSTD_count(ip+matchLength, iStart, iEnd));
+    return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
+}
+
+
+/*-*************************************
+ *  Hashes
+ ***************************************/
+static const U32 prime3bytes = 506832829U;
+static U32    ZSTD_hash3(U32 u, U32 h, U32 s) { assert(h <= 32); return (((u << (32-24)) * prime3bytes) ^ s)  >> (32-h) ; }
+MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h, 0); } /* only in zstd_opt.h */
+MEM_STATIC size_t ZSTD_hash3PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash3(MEM_readLE32(ptr), h, s); }
+
+static const U32 prime4bytes = 2654435761U;
+static U32    ZSTD_hash4(U32 u, U32 h, U32 s) { assert(h <= 32); return ((u * prime4bytes) ^ s) >> (32-h) ; }
+static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_readLE32(ptr), h, 0); }
+static size_t ZSTD_hash4PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash4(MEM_readLE32(ptr), h, s); }
+
+static const U64 prime5bytes = 889523592379ULL;
+static size_t ZSTD_hash5(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u  << (64-40)) * prime5bytes) ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash5PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash5(MEM_readLE64(p), h, s); }
+
+static const U64 prime6bytes = 227718039650203ULL;
+static size_t ZSTD_hash6(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u  << (64-48)) * prime6bytes) ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash6PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash6(MEM_readLE64(p), h, s); }
+
+static const U64 prime7bytes = 58295818150454627ULL;
+static size_t ZSTD_hash7(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u  << (64-56)) * prime7bytes) ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash7PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash7(MEM_readLE64(p), h, s); }
+
+static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
+static size_t ZSTD_hash8(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u) * prime8bytes)  ^ s) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h, 0); }
+static size_t ZSTD_hash8PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash8(MEM_readLE64(p), h, s); }
+
+
+MEM_STATIC FORCE_INLINE_ATTR
+size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
+{
+    /* Although some of these hashes do support hBits up to 64, some do not.
+     * To be on the safe side, always avoid hBits > 32. */
+    assert(hBits <= 32);
+
+    switch(mls)
+    {
+    default:
+    case 4: return ZSTD_hash4Ptr(p, hBits);
+    case 5: return ZSTD_hash5Ptr(p, hBits);
+    case 6: return ZSTD_hash6Ptr(p, hBits);
+    case 7: return ZSTD_hash7Ptr(p, hBits);
+    case 8: return ZSTD_hash8Ptr(p, hBits);
+    }
+}
+
+MEM_STATIC FORCE_INLINE_ATTR
+size_t ZSTD_hashPtrSalted(const void* p, U32 hBits, U32 mls, const U64 hashSalt) {
+    /* Although some of these hashes do support hBits up to 64, some do not.
+     * To be on the safe side, always avoid hBits > 32. */
+    assert(hBits <= 32);
+
+    switch(mls)
+    {
+        default:
+        case 4: return ZSTD_hash4PtrS(p, hBits, (U32)hashSalt);
+        case 5: return ZSTD_hash5PtrS(p, hBits, hashSalt);
+        case 6: return ZSTD_hash6PtrS(p, hBits, hashSalt);
+        case 7: return ZSTD_hash7PtrS(p, hBits, hashSalt);
+        case 8: return ZSTD_hash8PtrS(p, hBits, hashSalt);
+    }
+}
+
+
+/** ZSTD_ipow() :
+ * Return base^exponent.
+ */
+static U64 ZSTD_ipow(U64 base, U64 exponent)
+{
+    U64 power = 1;
+    while (exponent) {
+      if (exponent & 1) power *= base;
+      exponent >>= 1;
+      base *= base;
+    }
+    return power;
+}
+
+#define ZSTD_ROLL_HASH_CHAR_OFFSET 10
+
+/** ZSTD_rollingHash_append() :
+ * Add the buffer to the hash value.
+ */
+static U64 ZSTD_rollingHash_append(U64 hash, void const* buf, size_t size)
+{
+    BYTE const* istart = (BYTE const*)buf;
+    size_t pos;
+    for (pos = 0; pos < size; ++pos) {
+        hash *= prime8bytes;
+        hash += istart[pos] + ZSTD_ROLL_HASH_CHAR_OFFSET;
+    }
+    return hash;
+}
+
+/** ZSTD_rollingHash_compute() :
+ * Compute the rolling hash value of the buffer.
+ */
+MEM_STATIC U64 ZSTD_rollingHash_compute(void const* buf, size_t size)
+{
+    return ZSTD_rollingHash_append(0, buf, size);
+}
+
+/** ZSTD_rollingHash_primePower() :
+ * Compute the primePower to be passed to ZSTD_rollingHash_rotate() for a hash
+ * over a window of length bytes.
+ */
+MEM_STATIC U64 ZSTD_rollingHash_primePower(U32 length)
+{
+    return ZSTD_ipow(prime8bytes, length - 1);
+}
+
+/** ZSTD_rollingHash_rotate() :
+ * Rotate the rolling hash by one byte.
+ */
+MEM_STATIC U64 ZSTD_rollingHash_rotate(U64 hash, BYTE toRemove, BYTE toAdd, U64 primePower)
+{
+    hash -= (toRemove + ZSTD_ROLL_HASH_CHAR_OFFSET) * primePower;
+    hash *= prime8bytes;
+    hash += toAdd + ZSTD_ROLL_HASH_CHAR_OFFSET;
+    return hash;
+}
+
+/*-*************************************
+*  Round buffer management
+***************************************/
+#if (ZSTD_WINDOWLOG_MAX_64 > 31)
+# error "ZSTD_WINDOWLOG_MAX is too large : would overflow ZSTD_CURRENT_MAX"
+#endif
+/* Max current allowed */
+#define ZSTD_CURRENT_MAX ((3U << 29) + (1U << ZSTD_WINDOWLOG_MAX))
+/* Maximum chunk size before overflow correction needs to be called again */
+#define ZSTD_CHUNKSIZE_MAX                                                     \
+    ( ((U32)-1)                  /* Maximum ending current index */            \
+    - ZSTD_CURRENT_MAX)          /* Maximum beginning lowLimit */
+
+/**
+ * ZSTD_window_clear():
+ * Clears the window containing the history by simply setting it to empty.
+ */
+MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
+{
+    size_t const endT = (size_t)(window->nextSrc - window->base);
+    U32 const end = (U32)endT;
+
+    window->lowLimit = end;
+    window->dictLimit = end;
+}
+
+MEM_STATIC U32 ZSTD_window_isEmpty(ZSTD_window_t const window)
+{
+    return window.dictLimit == ZSTD_WINDOW_START_INDEX &&
+           window.lowLimit == ZSTD_WINDOW_START_INDEX &&
+           (window.nextSrc - window.base) == ZSTD_WINDOW_START_INDEX;
+}
+
+/**
+ * ZSTD_window_hasExtDict():
+ * Returns non-zero if the window has a non-empty extDict.
+ */
+MEM_STATIC U32 ZSTD_window_hasExtDict(ZSTD_window_t const window)
+{
+    return window.lowLimit < window.dictLimit;
+}
+
+/**
+ * ZSTD_matchState_dictMode():
+ * Inspects the provided matchState and figures out what dictMode should be
+ * passed to the compressor.
+ */
+MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
+{
+    return ZSTD_window_hasExtDict(ms->window) ?
+        ZSTD_extDict :
+        ms->dictMatchState != NULL ?
+            (ms->dictMatchState->dedicatedDictSearch ? ZSTD_dedicatedDictSearch : ZSTD_dictMatchState) :
+            ZSTD_noDict;
+}
+
+/* Defining this macro to non-zero tells zstd to run the overflow correction
+ * code much more frequently. This is very inefficient, and should only be
+ * used for tests and fuzzers.
+ */
+#ifndef ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY
+#  ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+#    define ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY 1
+#  else
+#    define ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY 0
+#  endif
+#endif
+
+/**
+ * ZSTD_window_canOverflowCorrect():
+ * Returns non-zero if the indices are large enough for overflow correction
+ * to work correctly without impacting compression ratio.
+ */
+MEM_STATIC U32 ZSTD_window_canOverflowCorrect(ZSTD_window_t const window,
+                                              U32 cycleLog,
+                                              U32 maxDist,
+                                              U32 loadedDictEnd,
+                                              void const* src)
+{
+    U32 const cycleSize = 1u << cycleLog;
+    U32 const curr = (U32)((BYTE const*)src - window.base);
+    U32 const minIndexToOverflowCorrect = cycleSize
+                                        + MAX(maxDist, cycleSize)
+                                        + ZSTD_WINDOW_START_INDEX;
+
+    /* Adjust the min index to backoff the overflow correction frequency,
+     * so we don't waste too much CPU in overflow correction. If this
+     * computation overflows we don't really care, we just need to make
+     * sure it is at least minIndexToOverflowCorrect.
+     */
+    U32 const adjustment = window.nbOverflowCorrections + 1;
+    U32 const adjustedIndex = MAX(minIndexToOverflowCorrect * adjustment,
+                                  minIndexToOverflowCorrect);
+    U32 const indexLargeEnough = curr > adjustedIndex;
+
+    /* Only overflow correct early if the dictionary is invalidated already,
+     * so we don't hurt compression ratio.
+     */
+    U32 const dictionaryInvalidated = curr > maxDist + loadedDictEnd;
+
+    return indexLargeEnough && dictionaryInvalidated;
+}
+
+/**
+ * ZSTD_window_needOverflowCorrection():
+ * Returns non-zero if the indices are getting too large and need overflow
+ * protection.
+ */
+MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
+                                                  U32 cycleLog,
+                                                  U32 maxDist,
+                                                  U32 loadedDictEnd,
+                                                  void const* src,
+                                                  void const* srcEnd)
+{
+    U32 const curr = (U32)((BYTE const*)srcEnd - window.base);
+    if (ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY) {
+        if (ZSTD_window_canOverflowCorrect(window, cycleLog, maxDist, loadedDictEnd, src)) {
+            return 1;
+        }
+    }
+    return curr > ZSTD_CURRENT_MAX;
+}
+
+/**
+ * ZSTD_window_correctOverflow():
+ * Reduces the indices to protect from index overflow.
+ * Returns the correction made to the indices, which must be applied to every
+ * stored index.
+ *
+ * The least significant cycleLog bits of the indices must remain the same,
+ * which may be 0. Every index up to maxDist in the past must be valid.
+ */
+MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
+                                           U32 maxDist, void const* src)
+{
+    /* preemptive overflow correction:
+     * 1. correction is large enough:
+     *    lowLimit > (3<<29) ==> current > 3<<29 + 1<<windowLog
+     *    1<<windowLog <= newCurrent < 1<<chainLog + 1<<windowLog
+     *
+     *    current - newCurrent
+     *    > (3<<29 + 1<<windowLog) - (1<<windowLog + 1<<chainLog)
+     *    > (3<<29) - (1<<chainLog)
+     *    > (3<<29) - (1<<30)             (NOTE: chainLog <= 30)
+     *    > 1<<29
+     *
+     * 2. (ip+ZSTD_CHUNKSIZE_MAX - cctx->base) doesn't overflow:
+     *    After correction, current is less than (1<<chainLog + 1<<windowLog).
+     *    In 64-bit mode we are safe, because we have 64-bit ptrdiff_t.
+     *    In 32-bit mode we are safe, because (chainLog <= 29), so
+     *    ip+ZSTD_CHUNKSIZE_MAX - cctx->base < 1<<32.
+     * 3. (cctx->lowLimit + 1<<windowLog) < 1<<32:
+     *    windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
+     */
+    U32 const cycleSize = 1u << cycleLog;
+    U32 const cycleMask = cycleSize - 1;
+    U32 const curr = (U32)((BYTE const*)src - window->base);
+    U32 const currentCycle = curr & cycleMask;
+    /* Ensure newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX. */
+    U32 const currentCycleCorrection = currentCycle < ZSTD_WINDOW_START_INDEX
+                                     ? MAX(cycleSize, ZSTD_WINDOW_START_INDEX)
+                                     : 0;
+    U32 const newCurrent = currentCycle
+                         + currentCycleCorrection
+                         + MAX(maxDist, cycleSize);
+    U32 const correction = curr - newCurrent;
+    /* maxDist must be a power of two so that:
+     *   (newCurrent & cycleMask) == (curr & cycleMask)
+     * This is required to not corrupt the chains / binary tree.
+     */
+    assert((maxDist & (maxDist - 1)) == 0);
+    assert((curr & cycleMask) == (newCurrent & cycleMask));
+    assert(curr > newCurrent);
+    if (!ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY) {
+        /* Loose bound, should be around 1<<29 (see above) */
+        assert(correction > 1<<28);
+    }
+
+    window->base += correction;
+    window->dictBase += correction;
+    if (window->lowLimit < correction + ZSTD_WINDOW_START_INDEX) {
+        window->lowLimit = ZSTD_WINDOW_START_INDEX;
+    } else {
+        window->lowLimit -= correction;
+    }
+    if (window->dictLimit < correction + ZSTD_WINDOW_START_INDEX) {
+        window->dictLimit = ZSTD_WINDOW_START_INDEX;
+    } else {
+        window->dictLimit -= correction;
+    }
+
+    /* Ensure we can still reference the full window. */
+    assert(newCurrent >= maxDist);
+    assert(newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX);
+    /* Ensure that lowLimit and dictLimit didn't underflow. */
+    assert(window->lowLimit <= newCurrent);
+    assert(window->dictLimit <= newCurrent);
+
+    ++window->nbOverflowCorrections;
+
+    DEBUGLOG(4, "Correction of 0x%x bytes to lowLimit=0x%x", correction,
+             window->lowLimit);
+    return correction;
+}
+
+/**
+ * ZSTD_window_enforceMaxDist():
+ * Updates lowLimit so that:
+ *    (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
+ *
+ * It ensures index is valid as long as index >= lowLimit.
+ * This must be called before a block compression call.
+ *
+ * loadedDictEnd is only defined if a dictionary is in use for current compression.
+ * As the name implies, loadedDictEnd represents the index at end of dictionary.
+ * The value lies within context's referential, it can be directly compared to blockEndIdx.
+ *
+ * If loadedDictEndPtr is NULL, no dictionary is in use, and we use loadedDictEnd == 0.
+ * If loadedDictEndPtr is not NULL, we set it to zero after updating lowLimit.
+ * This is because dictionaries are allowed to be referenced fully
+ * as long as the last byte of the dictionary is in the window.
+ * Once input has progressed beyond window size, dictionary cannot be referenced anymore.
+ *
+ * In normal dict mode, the dictionary lies between lowLimit and dictLimit.
+ * In dictMatchState mode, lowLimit and dictLimit are the same,
+ * and the dictionary is below them.
+ * forceWindow and dictMatchState are therefore incompatible.
+ */
+MEM_STATIC void
+ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
+                     const void* blockEnd,
+                           U32   maxDist,
+                           U32*  loadedDictEndPtr,
+                     const ZSTD_matchState_t** dictMatchStatePtr)
+{
+    U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
+    U32 const loadedDictEnd = (loadedDictEndPtr != NULL) ? *loadedDictEndPtr : 0;
+    DEBUGLOG(5, "ZSTD_window_enforceMaxDist: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
+                (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
+
+    /* - When there is no dictionary : loadedDictEnd == 0.
+         In which case, the test (blockEndIdx > maxDist) is merely to avoid
+         overflowing next operation `newLowLimit = blockEndIdx - maxDist`.
+       - When there is a standard dictionary :
+         Index referential is copied from the dictionary,
+         which means it starts from 0.
+         In which case, loadedDictEnd == dictSize,
+         and it makes sense to compare `blockEndIdx > maxDist + dictSize`
+         since `blockEndIdx` also starts from zero.
+       - When there is an attached dictionary :
+         loadedDictEnd is expressed within the referential of the context,
+         so it can be directly compared against blockEndIdx.
+    */
+    if (blockEndIdx > maxDist + loadedDictEnd) {
+        U32 const newLowLimit = blockEndIdx - maxDist;
+        if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
+        if (window->dictLimit < window->lowLimit) {
+            DEBUGLOG(5, "Update dictLimit to match lowLimit, from %u to %u",
+                        (unsigned)window->dictLimit, (unsigned)window->lowLimit);
+            window->dictLimit = window->lowLimit;
+        }
+        /* On reaching window size, dictionaries are invalidated */
+        if (loadedDictEndPtr) *loadedDictEndPtr = 0;
+        if (dictMatchStatePtr) *dictMatchStatePtr = NULL;
+    }
+}
+
+/* Similar to ZSTD_window_enforceMaxDist(),
+ * but only invalidates dictionary
+ * when input progresses beyond window size.
+ * assumption : loadedDictEndPtr and dictMatchStatePtr are valid (non NULL)
+ *              loadedDictEnd uses same referential as window->base
+ *              maxDist is the window size */
+MEM_STATIC void
+ZSTD_checkDictValidity(const ZSTD_window_t* window,
+                       const void* blockEnd,
+                             U32   maxDist,
+                             U32*  loadedDictEndPtr,
+                       const ZSTD_matchState_t** dictMatchStatePtr)
+{
+    assert(loadedDictEndPtr != NULL);
+    assert(dictMatchStatePtr != NULL);
+    {   U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
+        U32 const loadedDictEnd = *loadedDictEndPtr;
+        DEBUGLOG(5, "ZSTD_checkDictValidity: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
+                    (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
+        assert(blockEndIdx >= loadedDictEnd);
+
+        if (blockEndIdx > loadedDictEnd + maxDist || loadedDictEnd != window->dictLimit) {
+            /* On reaching window size, dictionaries are invalidated.
+             * For simplification, if window size is reached anywhere within next block,
+             * the dictionary is invalidated for the full block.
+             *
+             * We also have to invalidate the dictionary if ZSTD_window_update() has detected
+             * non-contiguous segments, which means that loadedDictEnd != window->dictLimit.
+             * loadedDictEnd may be 0, if forceWindow is true, but in that case we never use
+             * dictMatchState, so setting it to NULL is not a problem.
+             */
+            DEBUGLOG(6, "invalidating dictionary for current block (distance > windowSize)");
+            *loadedDictEndPtr = 0;
+            *dictMatchStatePtr = NULL;
+        } else {
+            if (*loadedDictEndPtr != 0) {
+                DEBUGLOG(6, "dictionary considered valid for current block");
+    }   }   }
+}
+
+MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
+    ZSTD_memset(window, 0, sizeof(*window));
+    window->base = (BYTE const*)" ";
+    window->dictBase = (BYTE const*)" ";
+    ZSTD_STATIC_ASSERT(ZSTD_DUBT_UNSORTED_MARK < ZSTD_WINDOW_START_INDEX); /* Start above ZSTD_DUBT_UNSORTED_MARK */
+    window->dictLimit = ZSTD_WINDOW_START_INDEX;    /* start from >0, so that 1st position is valid */
+    window->lowLimit = ZSTD_WINDOW_START_INDEX;     /* it ensures first and later CCtx usages compress the same */
+    window->nextSrc = window->base + ZSTD_WINDOW_START_INDEX;   /* see issue #1241 */
+    window->nbOverflowCorrections = 0;
+}
+
+/**
+ * ZSTD_window_update():
+ * Updates the window by appending [src, src + srcSize) to the window.
+ * If it is not contiguous, the current prefix becomes the extDict, and we
+ * forget about the extDict. Handles overlap of the prefix and extDict.
+ * Returns non-zero if the segment is contiguous.
+ */
+MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
+                                  void const* src, size_t srcSize,
+                                  int forceNonContiguous)
+{
+    BYTE const* const ip = (BYTE const*)src;
+    U32 contiguous = 1;
+    DEBUGLOG(5, "ZSTD_window_update");
+    if (srcSize == 0)
+        return contiguous;
+    assert(window->base != NULL);
+    assert(window->dictBase != NULL);
+    /* Check if blocks follow each other */
+    if (src != window->nextSrc || forceNonContiguous) {
+        /* not contiguous */
+        size_t const distanceFromBase = (size_t)(window->nextSrc - window->base);
+        DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u", window->dictLimit);
+        window->lowLimit = window->dictLimit;
+        assert(distanceFromBase == (size_t)(U32)distanceFromBase);  /* should never overflow */
+        window->dictLimit = (U32)distanceFromBase;
+        window->dictBase = window->base;
+        window->base = ip - distanceFromBase;
+        /* ms->nextToUpdate = window->dictLimit; */
+        if (window->dictLimit - window->lowLimit < HASH_READ_SIZE) window->lowLimit = window->dictLimit;   /* too small extDict */
+        contiguous = 0;
+    }
+    window->nextSrc = ip + srcSize;
+    /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
+    if ( (ip+srcSize > window->dictBase + window->lowLimit)
+       & (ip < window->dictBase + window->dictLimit)) {
+        ptrdiff_t const highInputIdx = (ip + srcSize) - window->dictBase;
+        U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)window->dictLimit) ? window->dictLimit : (U32)highInputIdx;
+        window->lowLimit = lowLimitMax;
+        DEBUGLOG(5, "Overlapping extDict and input : new lowLimit = %u", window->lowLimit);
+    }
+    return contiguous;
+}
+
+/**
+ * Returns the lowest allowed match index. It may either be in the ext-dict or the prefix.
+ */
+MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
+{
+    U32 const maxDistance = 1U << windowLog;
+    U32 const lowestValid = ms->window.lowLimit;
+    U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
+    U32 const isDictionary = (ms->loadedDictEnd != 0);
+    /* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
+     * is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
+     * valid for the entire block. So this check is sufficient to find the lowest valid match index.
+     */
+    U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
+    return matchLowest;
+}
+
+/**
+ * Returns the lowest allowed match index in the prefix.
+ */
+MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
+{
+    U32    const maxDistance = 1U << windowLog;
+    U32    const lowestValid = ms->window.dictLimit;
+    U32    const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
+    U32    const isDictionary = (ms->loadedDictEnd != 0);
+    /* When computing the lowest prefix index we need to take the dictionary into account to handle
+     * the edge case where the dictionary and the source are contiguous in memory.
+     */
+    U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
+    return matchLowest;
+}
+
+
+
+/* debug functions */
+#if (DEBUGLEVEL>=2)
+
+MEM_STATIC double ZSTD_fWeight(U32 rawStat)
+{
+    U32 const fp_accuracy = 8;
+    U32 const fp_multiplier = (1 << fp_accuracy);
+    U32 const newStat = rawStat + 1;
+    U32 const hb = ZSTD_highbit32(newStat);
+    U32 const BWeight = hb * fp_multiplier;
+    U32 const FWeight = (newStat << fp_accuracy) >> hb;
+    U32 const weight = BWeight + FWeight;
+    assert(hb + fp_accuracy < 31);
+    return (double)weight / fp_multiplier;
+}
+
+/* display a table content,
+ * listing each element, its frequency, and its predicted bit cost */
+MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
+{
+    unsigned u, sum;
+    for (u=0, sum=0; u<=max; u++) sum += table[u];
+    DEBUGLOG(2, "total nb elts: %u", sum);
+    for (u=0; u<=max; u++) {
+        DEBUGLOG(2, "%2u: %5u  (%.2f)",
+                u, table[u], ZSTD_fWeight(sum) - ZSTD_fWeight(table[u]) );
+    }
+}
+
+#endif
+
+/* Short Cache */
+
+/* Normally, zstd matchfinders follow this flow:
+ *     1. Compute hash at ip
+ *     2. Load index from hashTable[hash]
+ *     3. Check if *ip == *(base + index)
+ * In dictionary compression, loading *(base + index) is often an L2 or even L3 miss.
+ *
+ * Short cache is an optimization which allows us to avoid step 3 most of the time
+ * when the data doesn't actually match. With short cache, the flow becomes:
+ *     1. Compute (hash, currentTag) at ip. currentTag is an 8-bit independent hash at ip.
+ *     2. Load (index, matchTag) from hashTable[hash]. See ZSTD_writeTaggedIndex to understand how this works.
+ *     3. Only if currentTag == matchTag, check *ip == *(base + index). Otherwise, continue.
+ *
+ * Currently, short cache is only implemented in CDict hashtables. Thus, its use is limited to
+ * dictMatchState matchfinders.
+ */
+#define ZSTD_SHORT_CACHE_TAG_BITS 8
+#define ZSTD_SHORT_CACHE_TAG_MASK ((1u << ZSTD_SHORT_CACHE_TAG_BITS) - 1)
+
+/* Helper function for ZSTD_fillHashTable and ZSTD_fillDoubleHashTable.
+ * Unpacks hashAndTag into (hash, tag), then packs (index, tag) into hashTable[hash]. */
+MEM_STATIC void ZSTD_writeTaggedIndex(U32* const hashTable, size_t hashAndTag, U32 index) {
+    size_t const hash = hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
+    U32 const tag = (U32)(hashAndTag & ZSTD_SHORT_CACHE_TAG_MASK);
+    assert(index >> (32 - ZSTD_SHORT_CACHE_TAG_BITS) == 0);
+    hashTable[hash] = (index << ZSTD_SHORT_CACHE_TAG_BITS) | tag;
+}
+
+/* Helper function for short cache matchfinders.
+ * Unpacks tag1 and tag2 from lower bits of packedTag1 and packedTag2, then checks if the tags match. */
+MEM_STATIC int ZSTD_comparePackedTags(size_t packedTag1, size_t packedTag2) {
+    U32 const tag1 = packedTag1 & ZSTD_SHORT_CACHE_TAG_MASK;
+    U32 const tag2 = packedTag2 & ZSTD_SHORT_CACHE_TAG_MASK;
+    return tag1 == tag2;
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+/* ===============================================================
+ * Shared internal declarations
+ * These prototypes may be called from sources not in lib/compress
+ * =============================================================== */
+
+/* ZSTD_loadCEntropy() :
+ * dict : must point at beginning of a valid zstd dictionary.
+ * return : size of dictionary header (size of magic number + dict ID + entropy tables)
+ * assumptions : magic number supposed already checked
+ *               and dictSize >= 8 */
+size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
+                         const void* const dict, size_t dictSize);
+
+void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
+
+/* ==============================================================
+ * Private declarations
+ * These prototypes shall only be called from within lib/compress
+ * ============================================================== */
+
+/* ZSTD_getCParamsFromCCtxParams() :
+ * cParams are built depending on compressionLevel, src size hints,
+ * LDM and manually set compression parameters.
+ * Note: srcSizeHint == 0 means 0!
+ */
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
+
+/*! ZSTD_initCStream_internal() :
+ *  Private use only. Init streaming operation.
+ *  expects params to be valid.
+ *  must receive dict, or cdict, or none, but not both.
+ *  @return : 0, or an error code */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                     const void* dict, size_t dictSize,
+                     const ZSTD_CDict* cdict,
+                     const ZSTD_CCtx_params* params, unsigned long long pledgedSrcSize);
+
+void ZSTD_resetSeqStore(seqStore_t* ssPtr);
+
+/*! ZSTD_getCParamsFromCDict() :
+ *  as the name implies */
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
+
+/* ZSTD_compressBegin_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params,
+                                    unsigned long long pledgedSrcSize);
+
+/* ZSTD_compress_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compress_advanced_internal(ZSTD_CCtx* cctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict,size_t dictSize,
+                                 const ZSTD_CCtx_params* params);
+
+
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity);
+
+
+/* ZSTD_referenceExternalSequences() :
+ * Must be called before starting a compression operation.
+ * seqs must parse a prefix of the source.
+ * This cannot be used when long range matching is enabled.
+ * Zstd will use these sequences, and pass the literals to a secondary block
+ * compressor.
+ * @return : An error code on failure.
+ * NOTE: seqs are not verified! Invalid sequences can cause out-of-bounds memory
+ * access and data corruption.
+ */
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
+
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat);
+
+/** ZSTD_CCtx_trace() :
+ *  Trace the end of a compression call.
+ */
+void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize);
+
+/* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
+ * ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
+ * Note that the block delimiter must include the last literals of the block.
+ */
+size_t
+ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
+                                              ZSTD_sequencePosition* seqPos,
+                                        const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                        const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+
+/* Returns the number of bytes to move the current read position back by.
+ * Only non-zero if we ended up splitting a sequence.
+ * Otherwise, it may return a ZSTD error if something went wrong.
+ *
+ * This function will attempt to scan through blockSize bytes
+ * represented by the sequences in @inSeqs,
+ * storing any (partial) sequences.
+ *
+ * Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
+ * avoid splitting a match, or to avoid splitting a match such that it would produce a match
+ * smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
+ */
+size_t
+ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                   const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                   const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+
+
+/* ===============================================================
+ * Deprecated definitions that are still used internally to avoid
+ * deprecation warnings. These functions are exactly equivalent to
+ * their public variants, but avoid the deprecation warnings.
+ * =============================================================== */
+
+size_t ZSTD_compressBegin_usingCDict_deprecated(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
+
+size_t ZSTD_compressContinue_public(ZSTD_CCtx* cctx,
+                                    void* dst, size_t dstCapacity,
+                              const void* src, size_t srcSize);
+
+size_t ZSTD_compressEnd_public(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_deprecated(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+#endif /* ZSTD_COMPRESS_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_literals.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_literals.c
new file mode 100644
index 0000000..bfd4f11
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_literals.c
@@ -0,0 +1,235 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*-*************************************
+ *  Dependencies
+ ***************************************/
+#include "zstd_compress_literals.h"
+
+
+/* **************************************************************
+*  Debug Traces
+****************************************************************/
+#if DEBUGLEVEL >= 2
+
+static size_t showHexa(const void* src, size_t srcSize)
+{
+    const BYTE* const ip = (const BYTE*)src;
+    size_t u;
+    for (u=0; u<srcSize; u++) {
+        RAWLOG(5, " %02X", ip[u]); (void)ip;
+    }
+    RAWLOG(5, " \n");
+    return srcSize;
+}
+
+#endif
+
+
+/* **************************************************************
+*  Literals compression - special cases
+****************************************************************/
+size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    DEBUGLOG(5, "ZSTD_noCompressLiterals: srcSize=%zu, dstCapacity=%zu", srcSize, dstCapacity);
+
+    RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
+            break;
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
+            break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
+    }
+
+    ZSTD_memcpy(ostart + flSize, src, srcSize);
+    DEBUGLOG(5, "Raw (uncompressed) literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
+    return srcSize + flSize;
+}
+
+static int allBytesIdentical(const void* src, size_t srcSize)
+{
+    assert(srcSize >= 1);
+    assert(src != NULL);
+    {   const BYTE b = ((const BYTE*)src)[0];
+        size_t p;
+        for (p=1; p<srcSize; p++) {
+            if (((const BYTE*)src)[p] != b) return 0;
+        }
+        return 1;
+    }
+}
+
+size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    assert(dstCapacity >= 4); (void)dstCapacity;
+    assert(allBytesIdentical(src, srcSize));
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
+            break;
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
+            break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
+    }
+
+    ostart[flSize] = *(const BYTE*)src;
+    DEBUGLOG(5, "RLE : Repeated Literal (%02X: %u times) -> %u bytes encoded", ((const BYTE*)src)[0], (U32)srcSize, (U32)flSize + 1);
+    return flSize+1;
+}
+
+/* ZSTD_minLiteralsToCompress() :
+ * returns minimal amount of literals
+ * for literal compression to even be attempted.
+ * Minimum is made tighter as compression strategy increases.
+ */
+static size_t
+ZSTD_minLiteralsToCompress(ZSTD_strategy strategy, HUF_repeat huf_repeat)
+{
+    assert((int)strategy >= 0);
+    assert((int)strategy <= 9);
+    /* btultra2 : min 8 bytes;
+     * then 2x larger for each successive compression strategy
+     * max threshold 64 bytes */
+    {   int const shift = MIN(9-(int)strategy, 3);
+        size_t const mintc = (huf_repeat == HUF_repeat_valid) ? 6 : (size_t)8 << shift;
+        DEBUGLOG(7, "minLiteralsToCompress = %zu", mintc);
+        return mintc;
+    }
+}
+
+size_t ZSTD_compressLiterals (
+                  void* dst, size_t dstCapacity,
+            const void* src, size_t srcSize,
+                  void* entropyWorkspace, size_t entropyWorkspaceSize,
+            const ZSTD_hufCTables_t* prevHuf,
+                  ZSTD_hufCTables_t* nextHuf,
+                  ZSTD_strategy strategy,
+                  int disableLiteralCompression,
+                  int suspectUncompressible,
+                  int bmi2)
+{
+    size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
+    BYTE*  const ostart = (BYTE*)dst;
+    U32 singleStream = srcSize < 256;
+    symbolEncodingType_e hType = set_compressed;
+    size_t cLitSize;
+
+    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i, srcSize=%u, dstCapacity=%zu)",
+                disableLiteralCompression, (U32)srcSize, dstCapacity);
+
+    DEBUGLOG(6, "Completed literals listing (%zu bytes)", showHexa(src, srcSize));
+
+    /* Prepare nextEntropy assuming reusing the existing table */
+    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+
+    if (disableLiteralCompression)
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+
+    /* if too small, don't even attempt compression (speed opt) */
+    if (srcSize < ZSTD_minLiteralsToCompress(strategy, prevHuf->repeatMode))
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+
+    RETURN_ERROR_IF(dstCapacity < lhSize+1, dstSize_tooSmall, "not enough space for compression");
+    {   HUF_repeat repeat = prevHuf->repeatMode;
+        int const flags = 0
+            | (bmi2 ? HUF_flags_bmi2 : 0)
+            | (strategy < ZSTD_lazy && srcSize <= 1024 ? HUF_flags_preferRepeat : 0)
+            | (strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD ? HUF_flags_optimalDepth : 0)
+            | (suspectUncompressible ? HUF_flags_suspectUncompressible : 0);
+
+        typedef size_t (*huf_compress_f)(void*, size_t, const void*, size_t, unsigned, unsigned, void*, size_t, HUF_CElt*, HUF_repeat*, int);
+        huf_compress_f huf_compress;
+        if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
+        huf_compress = singleStream ? HUF_compress1X_repeat : HUF_compress4X_repeat;
+        cLitSize = huf_compress(ostart+lhSize, dstCapacity-lhSize,
+                                src, srcSize,
+                                HUF_SYMBOLVALUE_MAX, LitHufLog,
+                                entropyWorkspace, entropyWorkspaceSize,
+                                (HUF_CElt*)nextHuf->CTable,
+                                &repeat, flags);
+        DEBUGLOG(5, "%zu literals compressed into %zu bytes (before header)", srcSize, cLitSize);
+        if (repeat != HUF_repeat_none) {
+            /* reused the existing table */
+            DEBUGLOG(5, "reusing statistics from previous huffman block");
+            hType = set_repeat;
+        }
+    }
+
+    {   size_t const minGain = ZSTD_minGain(srcSize, strategy);
+        if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }   }
+    if (cLitSize==1) {
+        /* A return value of 1 signals that the alphabet consists of a single symbol.
+         * However, in some rare circumstances, it could be the compressed size (a single byte).
+         * For that outcome to have a chance to happen, it's necessary that `srcSize < 8`.
+         * (it's also necessary to not generate statistics).
+         * Therefore, in such a case, actively check that all bytes are identical. */
+        if ((srcSize >= 8) || allBytesIdentical(src, srcSize)) {
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }   }
+
+    if (hType == set_compressed) {
+        /* using a newly constructed table */
+        nextHuf->repeatMode = HUF_repeat_check;
+    }
+
+    /* Build header */
+    switch(lhSize)
+    {
+    case 3: /* 2 - 2 - 10 - 10 */
+        if (!singleStream) assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
+        {   U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
+            MEM_writeLE24(ostart, lhc);
+            break;
+        }
+    case 4: /* 2 - 2 - 14 - 14 */
+        assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
+        {   U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
+            MEM_writeLE32(ostart, lhc);
+            break;
+        }
+    case 5: /* 2 - 2 - 18 - 18 */
+        assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
+        {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
+            MEM_writeLE32(ostart, lhc);
+            ostart[4] = (BYTE)(cLitSize >> 10);
+            break;
+        }
+    default:  /* not possible : lhSize is {3,4,5} */
+        assert(0);
+    }
+    DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)srcSize, (U32)(lhSize+cLitSize));
+    return lhSize+cLitSize;
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_literals.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_literals.h
new file mode 100644
index 0000000..b060c8a
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_literals.h
@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPRESS_LITERALS_H
+#define ZSTD_COMPRESS_LITERALS_H
+
+#include "zstd_compress_internal.h" /* ZSTD_hufCTables_t, ZSTD_minGain() */
+
+
+size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/* ZSTD_compressRleLiteralsBlock() :
+ * Conditions :
+ * - All bytes in @src are identical
+ * - dstCapacity >= 4 */
+size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/* ZSTD_compressLiterals():
+ * @entropyWorkspace: must be aligned on 4-bytes boundaries
+ * @entropyWorkspaceSize : must be >= HUF_WORKSPACE_SIZE
+ * @suspectUncompressible: sampling checks, to potentially skip huffman coding
+ */
+size_t ZSTD_compressLiterals (void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                              void* entropyWorkspace, size_t entropyWorkspaceSize,
+                        const ZSTD_hufCTables_t* prevHuf,
+                              ZSTD_hufCTables_t* nextHuf,
+                              ZSTD_strategy strategy, int disableLiteralCompression,
+                              int suspectUncompressible,
+                              int bmi2);
+
+#endif /* ZSTD_COMPRESS_LITERALS_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_sequences.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_sequences.c
new file mode 100644
index 0000000..8872d4d
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_sequences.c
@@ -0,0 +1,442 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*-*************************************
+ *  Dependencies
+ ***************************************/
+#include "zstd_compress_sequences.h"
+
+/**
+ * -log2(x / 256) lookup table for x in [0, 256).
+ * If x == 0: Return 0
+ * Else: Return floor(-log2(x / 256) * 256)
+ */
+static unsigned const kInverseProbabilityLog256[256] = {
+    0,    2048, 1792, 1642, 1536, 1453, 1386, 1329, 1280, 1236, 1197, 1162,
+    1130, 1100, 1073, 1047, 1024, 1001, 980,  960,  941,  923,  906,  889,
+    874,  859,  844,  830,  817,  804,  791,  779,  768,  756,  745,  734,
+    724,  714,  704,  694,  685,  676,  667,  658,  650,  642,  633,  626,
+    618,  610,  603,  595,  588,  581,  574,  567,  561,  554,  548,  542,
+    535,  529,  523,  517,  512,  506,  500,  495,  489,  484,  478,  473,
+    468,  463,  458,  453,  448,  443,  438,  434,  429,  424,  420,  415,
+    411,  407,  402,  398,  394,  390,  386,  382,  377,  373,  370,  366,
+    362,  358,  354,  350,  347,  343,  339,  336,  332,  329,  325,  322,
+    318,  315,  311,  308,  305,  302,  298,  295,  292,  289,  286,  282,
+    279,  276,  273,  270,  267,  264,  261,  258,  256,  253,  250,  247,
+    244,  241,  239,  236,  233,  230,  228,  225,  222,  220,  217,  215,
+    212,  209,  207,  204,  202,  199,  197,  194,  192,  190,  187,  185,
+    182,  180,  178,  175,  173,  171,  168,  166,  164,  162,  159,  157,
+    155,  153,  151,  149,  146,  144,  142,  140,  138,  136,  134,  132,
+    130,  128,  126,  123,  121,  119,  117,  115,  114,  112,  110,  108,
+    106,  104,  102,  100,  98,   96,   94,   93,   91,   89,   87,   85,
+    83,   82,   80,   78,   76,   74,   73,   71,   69,   67,   66,   64,
+    62,   61,   59,   57,   55,   54,   52,   50,   49,   47,   46,   44,
+    42,   41,   39,   37,   36,   34,   33,   31,   30,   28,   26,   25,
+    23,   22,   20,   19,   17,   16,   14,   13,   11,   10,   8,    7,
+    5,    4,    2,    1,
+};
+
+static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
+  void const* ptr = ctable;
+  U16 const* u16ptr = (U16 const*)ptr;
+  U32 const maxSymbolValue = MEM_read16(u16ptr + 1);
+  return maxSymbolValue;
+}
+
+/**
+ * Returns true if we should use ncount=-1 else we should
+ * use ncount=1 for low probability symbols instead.
+ */
+static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
+{
+    /* Heuristic: This should cover most blocks <= 16K and
+     * start to fade out after 16K to about 32K depending on
+     * compressibility.
+     */
+    return nbSeq >= 2048;
+}
+
+/**
+ * Returns the cost in bytes of encoding the normalized count header.
+ * Returns an error if any of the helper functions return an error.
+ */
+static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
+                              size_t const nbSeq, unsigned const FSELog)
+{
+    BYTE wksp[FSE_NCOUNTBOUND];
+    S16 norm[MaxSeq + 1];
+    const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+    FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
+    return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution described by count
+ * using the entropy bound.
+ */
+static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t const total)
+{
+    unsigned cost = 0;
+    unsigned s;
+
+    assert(total > 0);
+    for (s = 0; s <= max; ++s) {
+        unsigned norm = (unsigned)((256 * count[s]) / total);
+        if (count[s] != 0 && norm == 0)
+            norm = 1;
+        assert(count[s] < total);
+        cost += count[s] * kInverseProbabilityLog256[norm];
+    }
+    return cost >> 8;
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using ctable.
+ * Returns an error if ctable cannot represent all the symbols in count.
+ */
+size_t ZSTD_fseBitCost(
+    FSE_CTable const* ctable,
+    unsigned const* count,
+    unsigned const max)
+{
+    unsigned const kAccuracyLog = 8;
+    size_t cost = 0;
+    unsigned s;
+    FSE_CState_t cstate;
+    FSE_initCState(&cstate, ctable);
+    if (ZSTD_getFSEMaxSymbolValue(ctable) < max) {
+        DEBUGLOG(5, "Repeat FSE_CTable has maxSymbolValue %u < %u",
+                    ZSTD_getFSEMaxSymbolValue(ctable), max);
+        return ERROR(GENERIC);
+    }
+    for (s = 0; s <= max; ++s) {
+        unsigned const tableLog = cstate.stateLog;
+        unsigned const badCost = (tableLog + 1) << kAccuracyLog;
+        unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
+        if (count[s] == 0)
+            continue;
+        if (bitCost >= badCost) {
+            DEBUGLOG(5, "Repeat FSE_CTable has Prob[%u] == 0", s);
+            return ERROR(GENERIC);
+        }
+        cost += (size_t)count[s] * bitCost;
+    }
+    return cost >> kAccuracyLog;
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using the
+ * table described by norm. The max symbol support by norm is assumed >= max.
+ * norm must be valid for every symbol with non-zero probability in count.
+ */
+size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
+                             unsigned const* count, unsigned const max)
+{
+    unsigned const shift = 8 - accuracyLog;
+    size_t cost = 0;
+    unsigned s;
+    assert(accuracyLog <= 8);
+    for (s = 0; s <= max; ++s) {
+        unsigned const normAcc = (norm[s] != -1) ? (unsigned)norm[s] : 1;
+        unsigned const norm256 = normAcc << shift;
+        assert(norm256 > 0);
+        assert(norm256 < 256);
+        cost += count[s] * kInverseProbabilityLog256[norm256];
+    }
+    return cost >> 8;
+}
+
+symbolEncodingType_e
+ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
+        size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
+        FSE_CTable const* prevCTable,
+        short const* defaultNorm, U32 defaultNormLog,
+        ZSTD_defaultPolicy_e const isDefaultAllowed,
+        ZSTD_strategy const strategy)
+{
+    ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
+    if (mostFrequent == nbSeq) {
+        *repeatMode = FSE_repeat_none;
+        if (isDefaultAllowed && nbSeq <= 2) {
+            /* Prefer set_basic over set_rle when there are 2 or fewer symbols,
+             * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
+             * If basic encoding isn't possible, always choose RLE.
+             */
+            DEBUGLOG(5, "Selected set_basic");
+            return set_basic;
+        }
+        DEBUGLOG(5, "Selected set_rle");
+        return set_rle;
+    }
+    if (strategy < ZSTD_lazy) {
+        if (isDefaultAllowed) {
+            size_t const staticFse_nbSeq_max = 1000;
+            size_t const mult = 10 - strategy;
+            size_t const baseLog = 3;
+            size_t const dynamicFse_nbSeq_min = (((size_t)1 << defaultNormLog) * mult) >> baseLog;  /* 28-36 for offset, 56-72 for lengths */
+            assert(defaultNormLog >= 5 && defaultNormLog <= 6);  /* xx_DEFAULTNORMLOG */
+            assert(mult <= 9 && mult >= 7);
+            if ( (*repeatMode == FSE_repeat_valid)
+              && (nbSeq < staticFse_nbSeq_max) ) {
+                DEBUGLOG(5, "Selected set_repeat");
+                return set_repeat;
+            }
+            if ( (nbSeq < dynamicFse_nbSeq_min)
+              || (mostFrequent < (nbSeq >> (defaultNormLog-1))) ) {
+                DEBUGLOG(5, "Selected set_basic");
+                /* The format allows default tables to be repeated, but it isn't useful.
+                 * When using simple heuristics to select encoding type, we don't want
+                 * to confuse these tables with dictionaries. When running more careful
+                 * analysis, we don't need to waste time checking both repeating tables
+                 * and default tables.
+                 */
+                *repeatMode = FSE_repeat_none;
+                return set_basic;
+            }
+        }
+    } else {
+        size_t const basicCost = isDefaultAllowed ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, count, max) : ERROR(GENERIC);
+        size_t const repeatCost = *repeatMode != FSE_repeat_none ? ZSTD_fseBitCost(prevCTable, count, max) : ERROR(GENERIC);
+        size_t const NCountCost = ZSTD_NCountCost(count, max, nbSeq, FSELog);
+        size_t const compressedCost = (NCountCost << 3) + ZSTD_entropyCost(count, max, nbSeq);
+
+        if (isDefaultAllowed) {
+            assert(!ZSTD_isError(basicCost));
+            assert(!(*repeatMode == FSE_repeat_valid && ZSTD_isError(repeatCost)));
+        }
+        assert(!ZSTD_isError(NCountCost));
+        assert(compressedCost < ERROR(maxCode));
+        DEBUGLOG(5, "Estimated bit costs: basic=%u\trepeat=%u\tcompressed=%u",
+                    (unsigned)basicCost, (unsigned)repeatCost, (unsigned)compressedCost);
+        if (basicCost <= repeatCost && basicCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_basic");
+            assert(isDefaultAllowed);
+            *repeatMode = FSE_repeat_none;
+            return set_basic;
+        }
+        if (repeatCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_repeat");
+            assert(!ZSTD_isError(repeatCost));
+            return set_repeat;
+        }
+        assert(compressedCost < basicCost && compressedCost < repeatCost);
+    }
+    DEBUGLOG(5, "Selected set_compressed");
+    *repeatMode = FSE_repeat_check;
+    return set_compressed;
+}
+
+typedef struct {
+    S16 norm[MaxSeq + 1];
+    U32 wksp[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(MaxSeq, MaxFSELog)];
+} ZSTD_BuildCTableWksp;
+
+size_t
+ZSTD_buildCTable(void* dst, size_t dstCapacity,
+                FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
+                unsigned* count, U32 max,
+                const BYTE* codeTable, size_t nbSeq,
+                const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                const FSE_CTable* prevCTable, size_t prevCTableSize,
+                void* entropyWorkspace, size_t entropyWorkspaceSize)
+{
+    BYTE* op = (BYTE*)dst;
+    const BYTE* const oend = op + dstCapacity;
+    DEBUGLOG(6, "ZSTD_buildCTable (dstCapacity=%u)", (unsigned)dstCapacity);
+
+    switch (type) {
+    case set_rle:
+        FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max), "");
+        RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall, "not enough space");
+        *op = codeTable[0];
+        return 1;
+    case set_repeat:
+        ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
+        return 0;
+    case set_basic:
+        FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), "");  /* note : could be pre-calculated */
+        return 0;
+    case set_compressed: {
+        ZSTD_BuildCTableWksp* wksp = (ZSTD_BuildCTableWksp*)entropyWorkspace;
+        size_t nbSeq_1 = nbSeq;
+        const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+        if (count[codeTable[nbSeq-1]] > 1) {
+            count[codeTable[nbSeq-1]]--;
+            nbSeq_1--;
+        }
+        assert(nbSeq_1 > 1);
+        assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
+        (void)entropyWorkspaceSize;
+        FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "FSE_normalizeCount failed");
+        assert(oend >= op);
+        {   size_t const NCountSize = FSE_writeNCount(op, (size_t)(oend - op), wksp->norm, max, tableLog);   /* overflow protected */
+            FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
+            FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "FSE_buildCTable_wksp failed");
+            return NCountSize;
+        }
+    }
+    default: assert(0); RETURN_ERROR(GENERIC, "impossible to reach");
+    }
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_encodeSequences_body(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    BIT_CStream_t blockStream;
+    FSE_CState_t  stateMatchLength;
+    FSE_CState_t  stateOffsetBits;
+    FSE_CState_t  stateLitLength;
+
+    RETURN_ERROR_IF(
+        ERR_isError(BIT_initCStream(&blockStream, dst, dstCapacity)),
+        dstSize_tooSmall, "not enough space remaining");
+    DEBUGLOG(6, "available space for bitstream : %i  (dstCapacity=%u)",
+                (int)(blockStream.endPtr - blockStream.startPtr),
+                (unsigned)dstCapacity);
+
+    /* first symbols */
+    FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].mlBase, ML_bits[mlCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    if (longOffsets) {
+        U32 const ofBits = ofCodeTable[nbSeq-1];
+        unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+        if (extraBits) {
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, extraBits);
+            BIT_flushBits(&blockStream);
+        }
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offBase >> extraBits,
+                    ofBits - extraBits);
+    } else {
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, ofCodeTable[nbSeq-1]);
+    }
+    BIT_flushBits(&blockStream);
+
+    {   size_t n;
+        for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
+            BYTE const llCode = llCodeTable[n];
+            BYTE const ofCode = ofCodeTable[n];
+            BYTE const mlCode = mlCodeTable[n];
+            U32  const llBits = LL_bits[llCode];
+            U32  const ofBits = ofCode;
+            U32  const mlBits = ML_bits[mlCode];
+            DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
+                        (unsigned)sequences[n].litLength,
+                        (unsigned)sequences[n].mlBase + MINMATCH,
+                        (unsigned)sequences[n].offBase);
+                                                                            /* 32b*/  /* 64b*/
+                                                                            /* (7)*/  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
+            FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
+            if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
+            if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
+                BIT_flushBits(&blockStream);                                /* (7)*/
+            BIT_addBits(&blockStream, sequences[n].litLength, llBits);
+            if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
+            BIT_addBits(&blockStream, sequences[n].mlBase, mlBits);
+            if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
+            if (longOffsets) {
+                unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+                if (extraBits) {
+                    BIT_addBits(&blockStream, sequences[n].offBase, extraBits);
+                    BIT_flushBits(&blockStream);                            /* (7)*/
+                }
+                BIT_addBits(&blockStream, sequences[n].offBase >> extraBits,
+                            ofBits - extraBits);                            /* 31 */
+            } else {
+                BIT_addBits(&blockStream, sequences[n].offBase, ofBits);     /* 31 */
+            }
+            BIT_flushBits(&blockStream);                                    /* (7)*/
+            DEBUGLOG(7, "remaining space : %i", (int)(blockStream.endPtr - blockStream.ptr));
+    }   }
+
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
+    FSE_flushCState(&blockStream, &stateMatchLength);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
+    FSE_flushCState(&blockStream, &stateOffsetBits);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
+    FSE_flushCState(&blockStream, &stateLitLength);
+
+    {   size_t const streamSize = BIT_closeCStream(&blockStream);
+        RETURN_ERROR_IF(streamSize==0, dstSize_tooSmall, "not enough space");
+        return streamSize;
+    }
+}
+
+static size_t
+ZSTD_encodeSequences_default(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+
+#if DYNAMIC_BMI2
+
+static BMI2_TARGET_ATTRIBUTE size_t
+ZSTD_encodeSequences_bmi2(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+#endif
+
+size_t ZSTD_encodeSequences(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
+{
+    DEBUGLOG(5, "ZSTD_encodeSequences: dstCapacity = %u", (unsigned)dstCapacity);
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
+                                         CTable_MatchLength, mlCodeTable,
+                                         CTable_OffsetBits, ofCodeTable,
+                                         CTable_LitLength, llCodeTable,
+                                         sequences, nbSeq, longOffsets);
+    }
+#endif
+    (void)bmi2;
+    return ZSTD_encodeSequences_default(dst, dstCapacity,
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq, longOffsets);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_sequences.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_sequences.h
new file mode 100644
index 0000000..4a3a05d
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_sequences.h
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPRESS_SEQUENCES_H
+#define ZSTD_COMPRESS_SEQUENCES_H
+
+#include "../common/fse.h" /* FSE_repeat, FSE_CTable */
+#include "../common/zstd_internal.h" /* symbolEncodingType_e, ZSTD_strategy */
+
+typedef enum {
+    ZSTD_defaultDisallowed = 0,
+    ZSTD_defaultAllowed = 1
+} ZSTD_defaultPolicy_e;
+
+symbolEncodingType_e
+ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
+        size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
+        FSE_CTable const* prevCTable,
+        short const* defaultNorm, U32 defaultNormLog,
+        ZSTD_defaultPolicy_e const isDefaultAllowed,
+        ZSTD_strategy const strategy);
+
+size_t
+ZSTD_buildCTable(void* dst, size_t dstCapacity,
+                FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
+                unsigned* count, U32 max,
+                const BYTE* codeTable, size_t nbSeq,
+                const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                const FSE_CTable* prevCTable, size_t prevCTableSize,
+                void* entropyWorkspace, size_t entropyWorkspaceSize);
+
+size_t ZSTD_encodeSequences(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2);
+
+size_t ZSTD_fseBitCost(
+    FSE_CTable const* ctable,
+    unsigned const* count,
+    unsigned const max);
+
+size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
+                             unsigned const* count, unsigned const max);
+#endif /* ZSTD_COMPRESS_SEQUENCES_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_superblock.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_superblock.c
new file mode 100644
index 0000000..638c4ac
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_superblock.c
@@ -0,0 +1,577 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*-*************************************
+ *  Dependencies
+ ***************************************/
+#include "zstd_compress_superblock.h"
+
+#include "../common/zstd_internal.h"  /* ZSTD_getSequenceLength */
+#include "hist.h"                     /* HIST_countFast_wksp */
+#include "zstd_compress_internal.h"   /* ZSTD_[huf|fse|entropy]CTablesMetadata_t */
+#include "zstd_compress_sequences.h"
+#include "zstd_compress_literals.h"
+
+/** ZSTD_compressSubBlock_literal() :
+ *  Compresses literals section for a sub-block.
+ *  When we have to write the Huffman table we will sometimes choose a header
+ *  size larger than necessary. This is because we have to pick the header size
+ *  before we know the table size + compressed size, so we have a bound on the
+ *  table size. If we guessed incorrectly, we fall back to uncompressed literals.
+ *
+ *  We write the header when writeEntropy=1 and set entropyWritten=1 when we succeeded
+ *  in writing the header, otherwise it is set to 0.
+ *
+ *  hufMetadata->hType has literals block type info.
+ *      If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
+ *      If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
+ *      If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
+ *      If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
+ *      and the following sub-blocks' literals sections will be Treeless_Literals_Block.
+ *  @return : compressed size of literals section of a sub-block
+ *            Or 0 if unable to compress.
+ *            Or error code */
+static size_t
+ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
+                              const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                              const BYTE* literals, size_t litSize,
+                              void* dst, size_t dstSize,
+                              const int bmi2, int writeEntropy, int* entropyWritten)
+{
+    size_t const header = writeEntropy ? 200 : 0;
+    size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart + lhSize;
+    U32 const singleStream = lhSize == 3;
+    symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
+    size_t cLitSize = 0;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
+
+    *entropyWritten = 0;
+    if (litSize == 0 || hufMetadata->hType == set_basic) {
+      DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal");
+      return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
+    } else if (hufMetadata->hType == set_rle) {
+      DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
+      return ZSTD_compressRleLiteralsBlock(dst, dstSize, literals, litSize);
+    }
+
+    assert(litSize > 0);
+    assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
+
+    if (writeEntropy && hufMetadata->hType == set_compressed) {
+        ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
+        op += hufMetadata->hufDesSize;
+        cLitSize += hufMetadata->hufDesSize;
+        DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
+    }
+
+    {   int const flags = bmi2 ? HUF_flags_bmi2 : 0;
+        const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable, flags)
+                                          : HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable, flags);
+        op += cSize;
+        cLitSize += cSize;
+        if (cSize == 0 || ERR_isError(cSize)) {
+            DEBUGLOG(5, "Failed to write entropy tables %s", ZSTD_getErrorName(cSize));
+            return 0;
+        }
+        /* If we expand and we aren't writing a header then emit uncompressed */
+        if (!writeEntropy && cLitSize >= litSize) {
+            DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal because uncompressible");
+            return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
+        }
+        /* If we are writing headers then allow expansion that doesn't change our header size. */
+        if (lhSize < (size_t)(3 + (cLitSize >= 1 KB) + (cLitSize >= 16 KB))) {
+            assert(cLitSize > litSize);
+            DEBUGLOG(5, "Literals expanded beyond allowed header size");
+            return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
+        }
+        DEBUGLOG(5, "ZSTD_compressSubBlock_literal (cSize=%zu)", cSize);
+    }
+
+    /* Build header */
+    switch(lhSize)
+    {
+    case 3: /* 2 - 2 - 10 - 10 */
+        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
+            MEM_writeLE24(ostart, lhc);
+            break;
+        }
+    case 4: /* 2 - 2 - 14 - 14 */
+        {   U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
+            MEM_writeLE32(ostart, lhc);
+            break;
+        }
+    case 5: /* 2 - 2 - 18 - 18 */
+        {   U32 const lhc = hType + (3 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<22);
+            MEM_writeLE32(ostart, lhc);
+            ostart[4] = (BYTE)(cLitSize >> 10);
+            break;
+        }
+    default:  /* not possible : lhSize is {3,4,5} */
+        assert(0);
+    }
+    *entropyWritten = 1;
+    DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
+    return op-ostart;
+}
+
+static size_t
+ZSTD_seqDecompressedSize(seqStore_t const* seqStore,
+                   const seqDef* sequences, size_t nbSeq,
+                         size_t litSize, int lastSequence)
+{
+    const seqDef* const sstart = sequences;
+    const seqDef* const send = sequences + nbSeq;
+    const seqDef* sp = sstart;
+    size_t matchLengthSum = 0;
+    size_t litLengthSum = 0;
+    (void)(litLengthSum); /* suppress unused variable warning on some environments */
+    while (send-sp > 0) {
+        ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
+        litLengthSum += seqLen.litLength;
+        matchLengthSum += seqLen.matchLength;
+        sp++;
+    }
+    assert(litLengthSum <= litSize);
+    if (!lastSequence) {
+        assert(litLengthSum == litSize);
+    }
+    return matchLengthSum + litSize;
+}
+
+/** ZSTD_compressSubBlock_sequences() :
+ *  Compresses sequences section for a sub-block.
+ *  fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
+ *  symbol compression modes for the super-block.
+ *  The first successfully compressed block will have these in its header.
+ *  We set entropyWritten=1 when we succeed in compressing the sequences.
+ *  The following sub-blocks will always have repeat mode.
+ *  @return : compressed size of sequences section of a sub-block
+ *            Or 0 if it is unable to compress
+ *            Or error code. */
+static size_t
+ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
+                                const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                const seqDef* sequences, size_t nbSeq,
+                                const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
+                                const ZSTD_CCtx_params* cctxParams,
+                                void* dst, size_t dstCapacity,
+                                const int bmi2, int writeEntropy, int* entropyWritten)
+{
+    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    BYTE* seqHead;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
+
+    *entropyWritten = 0;
+    /* Sequences Header */
+    RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
+                    dstSize_tooSmall, "");
+    if (nbSeq < 0x7F)
+        *op++ = (BYTE)nbSeq;
+    else if (nbSeq < LONGNBSEQ)
+        op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
+    else
+        op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
+    if (nbSeq==0) {
+        return op - ostart;
+    }
+
+    /* seqHead : flags for FSE encoding type */
+    seqHead = op++;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
+
+    if (writeEntropy) {
+        const U32 LLtype = fseMetadata->llType;
+        const U32 Offtype = fseMetadata->ofType;
+        const U32 MLtype = fseMetadata->mlType;
+        DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
+        *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
+        ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
+        op += fseMetadata->fseTablesSize;
+    } else {
+        const U32 repeat = set_repeat;
+        *seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
+    }
+
+    {   size_t const bitstreamSize = ZSTD_encodeSequences(
+                                        op, oend - op,
+                                        fseTables->matchlengthCTable, mlCode,
+                                        fseTables->offcodeCTable, ofCode,
+                                        fseTables->litlengthCTable, llCode,
+                                        sequences, nbSeq,
+                                        longOffsets, bmi2);
+        FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
+        op += bitstreamSize;
+        /* zstd versions <= 1.3.4 mistakenly report corruption when
+         * FSE_readNCount() receives a buffer < 4 bytes.
+         * Fixed by https://github.com/facebook/zstd/pull/1146.
+         * This can happen when the last set_compressed table present is 2
+         * bytes and the bitstream is only one byte.
+         * In this exceedingly rare case, we will simply emit an uncompressed
+         * block, since it isn't worth optimizing.
+         */
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+        if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
+            /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
+            assert(fseMetadata->lastCountSize + bitstreamSize == 3);
+            DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
+                        "emitting an uncompressed block.");
+            return 0;
+        }
+#endif
+        DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
+    }
+
+    /* zstd versions <= 1.4.0 mistakenly report error when
+     * sequences section body size is less than 3 bytes.
+     * Fixed by https://github.com/facebook/zstd/pull/1664.
+     * This can happen when the previous sequences section block is compressed
+     * with rle mode and the current block's sequences section is compressed
+     * with repeat mode where sequences section body size can be 1 byte.
+     */
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    if (op-seqHead < 4) {
+        DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
+                    "an uncompressed block when sequences are < 4 bytes");
+        return 0;
+    }
+#endif
+
+    *entropyWritten = 1;
+    return op - ostart;
+}
+
+/** ZSTD_compressSubBlock() :
+ *  Compresses a single sub-block.
+ *  @return : compressed size of the sub-block
+ *            Or 0 if it failed to compress. */
+static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
+                                    const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                                    const seqDef* sequences, size_t nbSeq,
+                                    const BYTE* literals, size_t litSize,
+                                    const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
+                                    const ZSTD_CCtx_params* cctxParams,
+                                    void* dst, size_t dstCapacity,
+                                    const int bmi2,
+                                    int writeLitEntropy, int writeSeqEntropy,
+                                    int* litEntropyWritten, int* seqEntropyWritten,
+                                    U32 lastBlock)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart + ZSTD_blockHeaderSize;
+    DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
+                litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
+    {   size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
+                                                        &entropyMetadata->hufMetadata, literals, litSize,
+                                                        op, oend-op, bmi2, writeLitEntropy, litEntropyWritten);
+        FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
+        if (cLitSize == 0) return 0;
+        op += cLitSize;
+    }
+    {   size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
+                                                  &entropyMetadata->fseMetadata,
+                                                  sequences, nbSeq,
+                                                  llCode, mlCode, ofCode,
+                                                  cctxParams,
+                                                  op, oend-op,
+                                                  bmi2, writeSeqEntropy, seqEntropyWritten);
+        FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
+        if (cSeqSize == 0) return 0;
+        op += cSeqSize;
+    }
+    /* Write block header */
+    {   size_t cSize = (op-ostart)-ZSTD_blockHeaderSize;
+        U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+        MEM_writeLE24(ostart, cBlockHeader24);
+    }
+    return op-ostart;
+}
+
+static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
+                                                const ZSTD_hufCTables_t* huf,
+                                                const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                                                void* workspace, size_t wkspSize,
+                                                int writeEntropy)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    unsigned maxSymbolValue = 255;
+    size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
+
+    if (hufMetadata->hType == set_basic) return litSize;
+    else if (hufMetadata->hType == set_rle) return 1;
+    else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
+        size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
+        if (ZSTD_isError(largest)) return litSize;
+        {   size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
+            if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
+            return cLitSizeEstimate + literalSectionHeaderSize;
+    }   }
+    assert(0); /* impossible */
+    return 0;
+}
+
+static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
+                        const BYTE* codeTable, unsigned maxCode,
+                        size_t nbSeq, const FSE_CTable* fseCTable,
+                        const U8* additionalBits,
+                        short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                        void* workspace, size_t wkspSize)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    const BYTE* ctp = codeTable;
+    const BYTE* const ctStart = ctp;
+    const BYTE* const ctEnd = ctStart + nbSeq;
+    size_t cSymbolTypeSizeEstimateInBits = 0;
+    unsigned max = maxCode;
+
+    HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);  /* can't fail */
+    if (type == set_basic) {
+        /* We selected this encoding type, so it must be valid. */
+        assert(max <= defaultMax);
+        cSymbolTypeSizeEstimateInBits = max <= defaultMax
+                ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
+                : ERROR(GENERIC);
+    } else if (type == set_rle) {
+        cSymbolTypeSizeEstimateInBits = 0;
+    } else if (type == set_compressed || type == set_repeat) {
+        cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
+    }
+    if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
+    while (ctp < ctEnd) {
+        if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
+        else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
+        ctp++;
+    }
+    return cSymbolTypeSizeEstimateInBits / 8;
+}
+
+static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
+                                                  const BYTE* llCodeTable,
+                                                  const BYTE* mlCodeTable,
+                                                  size_t nbSeq,
+                                                  const ZSTD_fseCTables_t* fseTables,
+                                                  const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                                  void* workspace, size_t wkspSize,
+                                                  int writeEntropy)
+{
+    size_t const sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
+    size_t cSeqSizeEstimate = 0;
+    if (nbSeq == 0) return sequencesSectionHeaderSize;
+    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
+                                         nbSeq, fseTables->offcodeCTable, NULL,
+                                         OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                                         workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
+                                         nbSeq, fseTables->litlengthCTable, LL_bits,
+                                         LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                                         workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
+                                         nbSeq, fseTables->matchlengthCTable, ML_bits,
+                                         ML_defaultNorm, ML_defaultNormLog, MaxML,
+                                         workspace, wkspSize);
+    if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
+    return cSeqSizeEstimate + sequencesSectionHeaderSize;
+}
+
+static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
+                                        const BYTE* ofCodeTable,
+                                        const BYTE* llCodeTable,
+                                        const BYTE* mlCodeTable,
+                                        size_t nbSeq,
+                                        const ZSTD_entropyCTables_t* entropy,
+                                        const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                                        void* workspace, size_t wkspSize,
+                                        int writeLitEntropy, int writeSeqEntropy) {
+    size_t cSizeEstimate = 0;
+    cSizeEstimate += ZSTD_estimateSubBlockSize_literal(literals, litSize,
+                                                         &entropy->huf, &entropyMetadata->hufMetadata,
+                                                         workspace, wkspSize, writeLitEntropy);
+    cSizeEstimate += ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
+                                                         nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
+                                                         workspace, wkspSize, writeSeqEntropy);
+    return cSizeEstimate + ZSTD_blockHeaderSize;
+}
+
+static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
+{
+    if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
+        return 1;
+    if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
+        return 1;
+    if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
+        return 1;
+    return 0;
+}
+
+/** ZSTD_compressSubBlock_multi() :
+ *  Breaks super-block into multiple sub-blocks and compresses them.
+ *  Entropy will be written to the first block.
+ *  The following blocks will use repeat mode to compress.
+ *  All sub-blocks are compressed blocks (no raw or rle blocks).
+ *  @return : compressed size of the super block (which is multiple ZSTD blocks)
+ *            Or 0 if it failed to compress. */
+static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
+                            const ZSTD_compressedBlockState_t* prevCBlock,
+                            ZSTD_compressedBlockState_t* nextCBlock,
+                            const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                            const ZSTD_CCtx_params* cctxParams,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const int bmi2, U32 lastBlock,
+                            void* workspace, size_t wkspSize)
+{
+    const seqDef* const sstart = seqStorePtr->sequencesStart;
+    const seqDef* const send = seqStorePtr->sequences;
+    const seqDef* sp = sstart;
+    const BYTE* const lstart = seqStorePtr->litStart;
+    const BYTE* const lend = seqStorePtr->lit;
+    const BYTE* lp = lstart;
+    BYTE const* ip = (BYTE const*)src;
+    BYTE const* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    const BYTE* llCodePtr = seqStorePtr->llCode;
+    const BYTE* mlCodePtr = seqStorePtr->mlCode;
+    const BYTE* ofCodePtr = seqStorePtr->ofCode;
+    size_t targetCBlockSize = cctxParams->targetCBlockSize;
+    size_t litSize, seqCount;
+    int writeLitEntropy = entropyMetadata->hufMetadata.hType == set_compressed;
+    int writeSeqEntropy = 1;
+    int lastSequence = 0;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_multi (litSize=%u, nbSeq=%u)",
+                (unsigned)(lend-lp), (unsigned)(send-sstart));
+
+    litSize = 0;
+    seqCount = 0;
+    do {
+        size_t cBlockSizeEstimate = 0;
+        if (sstart == send) {
+            lastSequence = 1;
+        } else {
+            const seqDef* const sequence = sp + seqCount;
+            lastSequence = sequence == send - 1;
+            litSize += ZSTD_getSequenceLength(seqStorePtr, sequence).litLength;
+            seqCount++;
+        }
+        if (lastSequence) {
+            assert(lp <= lend);
+            assert(litSize <= (size_t)(lend - lp));
+            litSize = (size_t)(lend - lp);
+        }
+        /* I think there is an optimization opportunity here.
+         * Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
+         * since it recalculates estimate from scratch.
+         * For example, it would recount literal distribution and symbol codes every time.
+         */
+        cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
+                                                       &nextCBlock->entropy, entropyMetadata,
+                                                       workspace, wkspSize, writeLitEntropy, writeSeqEntropy);
+        if (cBlockSizeEstimate > targetCBlockSize || lastSequence) {
+            int litEntropyWritten = 0;
+            int seqEntropyWritten = 0;
+            const size_t decompressedSize = ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, lastSequence);
+            const size_t cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
+                                                       sp, seqCount,
+                                                       lp, litSize,
+                                                       llCodePtr, mlCodePtr, ofCodePtr,
+                                                       cctxParams,
+                                                       op, oend-op,
+                                                       bmi2, writeLitEntropy, writeSeqEntropy,
+                                                       &litEntropyWritten, &seqEntropyWritten,
+                                                       lastBlock && lastSequence);
+            FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
+            if (cSize > 0 && cSize < decompressedSize) {
+                DEBUGLOG(5, "Committed the sub-block");
+                assert(ip + decompressedSize <= iend);
+                ip += decompressedSize;
+                sp += seqCount;
+                lp += litSize;
+                op += cSize;
+                llCodePtr += seqCount;
+                mlCodePtr += seqCount;
+                ofCodePtr += seqCount;
+                litSize = 0;
+                seqCount = 0;
+                /* Entropy only needs to be written once */
+                if (litEntropyWritten) {
+                    writeLitEntropy = 0;
+                }
+                if (seqEntropyWritten) {
+                    writeSeqEntropy = 0;
+                }
+            }
+        }
+    } while (!lastSequence);
+    if (writeLitEntropy) {
+        DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
+        ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
+    }
+    if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
+        /* If we haven't written our entropy tables, then we've violated our contract and
+         * must emit an uncompressed block.
+         */
+        DEBUGLOG(5, "ZSTD_compressSubBlock_multi has sequence entropy tables unwritten");
+        return 0;
+    }
+    if (ip < iend) {
+        size_t const cSize = ZSTD_noCompressBlock(op, oend - op, ip, iend - ip, lastBlock);
+        DEBUGLOG(5, "ZSTD_compressSubBlock_multi last sub-block uncompressed, %zu bytes", (size_t)(iend - ip));
+        FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
+        assert(cSize != 0);
+        op += cSize;
+        /* We have to regenerate the repcodes because we've skipped some sequences */
+        if (sp < send) {
+            seqDef const* seq;
+            repcodes_t rep;
+            ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
+            for (seq = sstart; seq < sp; ++seq) {
+                ZSTD_updateRep(rep.rep, seq->offBase, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
+            }
+            ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
+        }
+    }
+    DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
+    return op-ostart;
+}
+
+size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               void const* src, size_t srcSize,
+                               unsigned lastBlock) {
+    ZSTD_entropyCTablesMetadata_t entropyMetadata;
+
+    FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
+          &zc->blockState.prevCBlock->entropy,
+          &zc->blockState.nextCBlock->entropy,
+          &zc->appliedParams,
+          &entropyMetadata,
+          zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
+
+    return ZSTD_compressSubBlock_multi(&zc->seqStore,
+            zc->blockState.prevCBlock,
+            zc->blockState.nextCBlock,
+            &entropyMetadata,
+            &zc->appliedParams,
+            dst, dstCapacity,
+            src, srcSize,
+            zc->bmi2, lastBlock,
+            zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_superblock.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_superblock.h
new file mode 100644
index 0000000..8e494f0
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_compress_superblock.h
@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPRESS_ADVANCED_H
+#define ZSTD_COMPRESS_ADVANCED_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+
+#include "../zstd.h" /* ZSTD_CCtx */
+
+/*-*************************************
+*  Target Compressed Block Size
+***************************************/
+
+/* ZSTD_compressSuperBlock() :
+ * Used to compress a super block when targetCBlockSize is being used.
+ * The given block will be compressed into multiple sub blocks that are around targetCBlockSize. */
+size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               void const* src, size_t srcSize,
+                               unsigned lastBlock);
+
+#endif /* ZSTD_COMPRESS_ADVANCED_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_cwksp.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_cwksp.h
new file mode 100644
index 0000000..cc7fb1c
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_cwksp.h
@@ -0,0 +1,742 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_CWKSP_H
+#define ZSTD_CWKSP_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customFree */
+#include "../common/zstd_internal.h"
+#include "../common/portability_macros.h"
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-*************************************
+*  Constants
+***************************************/
+
+/* Since the workspace is effectively its own little malloc implementation /
+ * arena, when we run under ASAN, we should similarly insert redzones between
+ * each internal element of the workspace, so ASAN will catch overruns that
+ * reach outside an object but that stay inside the workspace.
+ *
+ * This defines the size of that redzone.
+ */
+#ifndef ZSTD_CWKSP_ASAN_REDZONE_SIZE
+#define ZSTD_CWKSP_ASAN_REDZONE_SIZE 128
+#endif
+
+
+/* Set our tables and aligneds to align by 64 bytes */
+#define ZSTD_CWKSP_ALIGNMENT_BYTES 64
+
+/*-*************************************
+*  Structures
+***************************************/
+typedef enum {
+    ZSTD_cwksp_alloc_objects,
+    ZSTD_cwksp_alloc_aligned_init_once,
+    ZSTD_cwksp_alloc_aligned,
+    ZSTD_cwksp_alloc_buffers
+} ZSTD_cwksp_alloc_phase_e;
+
+/**
+ * Used to describe whether the workspace is statically allocated (and will not
+ * necessarily ever be freed), or if it's dynamically allocated and we can
+ * expect a well-formed caller to free this.
+ */
+typedef enum {
+    ZSTD_cwksp_dynamic_alloc,
+    ZSTD_cwksp_static_alloc
+} ZSTD_cwksp_static_alloc_e;
+
+/**
+ * Zstd fits all its internal datastructures into a single continuous buffer,
+ * so that it only needs to perform a single OS allocation (or so that a buffer
+ * can be provided to it and it can perform no allocations at all). This buffer
+ * is called the workspace.
+ *
+ * Several optimizations complicate that process of allocating memory ranges
+ * from this workspace for each internal datastructure:
+ *
+ * - These different internal datastructures have different setup requirements:
+ *
+ *   - The static objects need to be cleared once and can then be trivially
+ *     reused for each compression.
+ *
+ *   - Various buffers don't need to be initialized at all--they are always
+ *     written into before they're read.
+ *
+ *   - The matchstate tables have a unique requirement that they don't need
+ *     their memory to be totally cleared, but they do need the memory to have
+ *     some bound, i.e., a guarantee that all values in the memory they've been
+ *     allocated is less than some maximum value (which is the starting value
+ *     for the indices that they will then use for compression). When this
+ *     guarantee is provided to them, they can use the memory without any setup
+ *     work. When it can't, they have to clear the area.
+ *
+ * - These buffers also have different alignment requirements.
+ *
+ * - We would like to reuse the objects in the workspace for multiple
+ *   compressions without having to perform any expensive reallocation or
+ *   reinitialization work.
+ *
+ * - We would like to be able to efficiently reuse the workspace across
+ *   multiple compressions **even when the compression parameters change** and
+ *   we need to resize some of the objects (where possible).
+ *
+ * To attempt to manage this buffer, given these constraints, the ZSTD_cwksp
+ * abstraction was created. It works as follows:
+ *
+ * Workspace Layout:
+ *
+ * [                        ... workspace ...                           ]
+ * [objects][tables ->] free space [<- buffers][<- aligned][<- init once]
+ *
+ * The various objects that live in the workspace are divided into the
+ * following categories, and are allocated separately:
+ *
+ * - Static objects: this is optionally the enclosing ZSTD_CCtx or ZSTD_CDict,
+ *   so that literally everything fits in a single buffer. Note: if present,
+ *   this must be the first object in the workspace, since ZSTD_customFree{CCtx,
+ *   CDict}() rely on a pointer comparison to see whether one or two frees are
+ *   required.
+ *
+ * - Fixed size objects: these are fixed-size, fixed-count objects that are
+ *   nonetheless "dynamically" allocated in the workspace so that we can
+ *   control how they're initialized separately from the broader ZSTD_CCtx.
+ *   Examples:
+ *   - Entropy Workspace
+ *   - 2 x ZSTD_compressedBlockState_t
+ *   - CDict dictionary contents
+ *
+ * - Tables: these are any of several different datastructures (hash tables,
+ *   chain tables, binary trees) that all respect a common format: they are
+ *   uint32_t arrays, all of whose values are between 0 and (nextSrc - base).
+ *   Their sizes depend on the cparams. These tables are 64-byte aligned.
+ *
+ * - Init once: these buffers require to be initialized at least once before
+ *   use. They should be used when we want to skip memory initialization
+ *   while not triggering memory checkers (like Valgrind) when reading from
+ *   from this memory without writing to it first.
+ *   These buffers should be used carefully as they might contain data
+ *   from previous compressions.
+ *   Buffers are aligned to 64 bytes.
+ *
+ * - Aligned: these buffers don't require any initialization before they're
+ *   used. The user of the buffer should make sure they write into a buffer
+ *   location before reading from it.
+ *   Buffers are aligned to 64 bytes.
+ *
+ * - Buffers: these buffers are used for various purposes that don't require
+ *   any alignment or initialization before they're used. This means they can
+ *   be moved around at no cost for a new compression.
+ *
+ * Allocating Memory:
+ *
+ * The various types of objects must be allocated in order, so they can be
+ * correctly packed into the workspace buffer. That order is:
+ *
+ * 1. Objects
+ * 2. Init once / Tables
+ * 3. Aligned / Tables
+ * 4. Buffers / Tables
+ *
+ * Attempts to reserve objects of different types out of order will fail.
+ */
+typedef struct {
+    void* workspace;
+    void* workspaceEnd;
+
+    void* objectEnd;
+    void* tableEnd;
+    void* tableValidEnd;
+    void* allocStart;
+    void* initOnceStart;
+
+    BYTE allocFailed;
+    int workspaceOversizedDuration;
+    ZSTD_cwksp_alloc_phase_e phase;
+    ZSTD_cwksp_static_alloc_e isStatic;
+} ZSTD_cwksp;
+
+/*-*************************************
+*  Functions
+***************************************/
+
+MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws);
+MEM_STATIC void*  ZSTD_cwksp_initialAllocStart(ZSTD_cwksp* ws);
+
+MEM_STATIC void ZSTD_cwksp_assert_internal_consistency(ZSTD_cwksp* ws) {
+    (void)ws;
+    assert(ws->workspace <= ws->objectEnd);
+    assert(ws->objectEnd <= ws->tableEnd);
+    assert(ws->objectEnd <= ws->tableValidEnd);
+    assert(ws->tableEnd <= ws->allocStart);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    assert(ws->allocStart <= ws->workspaceEnd);
+    assert(ws->initOnceStart <= ZSTD_cwksp_initialAllocStart(ws));
+    assert(ws->workspace <= ws->initOnceStart);
+#if ZSTD_MEMORY_SANITIZER
+    {
+        intptr_t const offset = __msan_test_shadow(ws->initOnceStart,
+            (U8*)ZSTD_cwksp_initialAllocStart(ws) - (U8*)ws->initOnceStart);
+#if defined(ZSTD_MSAN_PRINT)
+        if(offset!=-1) {
+            __msan_print_shadow((U8*)ws->initOnceStart + offset - 8, 32);
+        }
+#endif
+        assert(offset==-1);
+    };
+#endif
+}
+
+/**
+ * Align must be a power of 2.
+ */
+MEM_STATIC size_t ZSTD_cwksp_align(size_t size, size_t const align) {
+    size_t const mask = align - 1;
+    assert((align & mask) == 0);
+    return (size + mask) & ~mask;
+}
+
+/**
+ * Use this to determine how much space in the workspace we will consume to
+ * allocate this object. (Normally it should be exactly the size of the object,
+ * but under special conditions, like ASAN, where we pad each object, it might
+ * be larger.)
+ *
+ * Since tables aren't currently redzoned, you don't need to call through this
+ * to figure out how much space you need for the matchState tables. Everything
+ * else is though.
+ *
+ * Do not use for sizing aligned buffers. Instead, use ZSTD_cwksp_aligned_alloc_size().
+ */
+MEM_STATIC size_t ZSTD_cwksp_alloc_size(size_t size) {
+    if (size == 0)
+        return 0;
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    return size + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+#else
+    return size;
+#endif
+}
+
+/**
+ * Returns an adjusted alloc size that is the nearest larger multiple of 64 bytes.
+ * Used to determine the number of bytes required for a given "aligned".
+ */
+MEM_STATIC size_t ZSTD_cwksp_aligned_alloc_size(size_t size) {
+    return ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(size, ZSTD_CWKSP_ALIGNMENT_BYTES));
+}
+
+/**
+ * Returns the amount of additional space the cwksp must allocate
+ * for internal purposes (currently only alignment).
+ */
+MEM_STATIC size_t ZSTD_cwksp_slack_space_required(void) {
+    /* For alignment, the wksp will always allocate an additional 2*ZSTD_CWKSP_ALIGNMENT_BYTES
+     * bytes to align the beginning of tables section and end of buffers;
+     */
+    size_t const slackSpace = ZSTD_CWKSP_ALIGNMENT_BYTES * 2;
+    return slackSpace;
+}
+
+
+/**
+ * Return the number of additional bytes required to align a pointer to the given number of bytes.
+ * alignBytes must be a power of two.
+ */
+MEM_STATIC size_t ZSTD_cwksp_bytes_to_align_ptr(void* ptr, const size_t alignBytes) {
+    size_t const alignBytesMask = alignBytes - 1;
+    size_t const bytes = (alignBytes - ((size_t)ptr & (alignBytesMask))) & alignBytesMask;
+    assert((alignBytes & alignBytesMask) == 0);
+    assert(bytes < alignBytes);
+    return bytes;
+}
+
+/**
+ * Returns the initial value for allocStart which is used to determine the position from
+ * which we can allocate from the end of the workspace.
+ */
+MEM_STATIC void*  ZSTD_cwksp_initialAllocStart(ZSTD_cwksp* ws) {
+    return (void*)((size_t)ws->workspaceEnd & ~(ZSTD_CWKSP_ALIGNMENT_BYTES-1));
+}
+
+/**
+ * Internal function. Do not use directly.
+ * Reserves the given number of bytes within the aligned/buffer segment of the wksp,
+ * which counts from the end of the wksp (as opposed to the object/table segment).
+ *
+ * Returns a pointer to the beginning of that space.
+ */
+MEM_STATIC void*
+ZSTD_cwksp_reserve_internal_buffer_space(ZSTD_cwksp* ws, size_t const bytes)
+{
+    void* const alloc = (BYTE*)ws->allocStart - bytes;
+    void* const bottom = ws->tableEnd;
+    DEBUGLOG(5, "cwksp: reserving %p %zd bytes, %zd bytes remaining",
+        alloc, bytes, ZSTD_cwksp_available_space(ws) - bytes);
+    ZSTD_cwksp_assert_internal_consistency(ws);
+    assert(alloc >= bottom);
+    if (alloc < bottom) {
+        DEBUGLOG(4, "cwksp: alloc failed!");
+        ws->allocFailed = 1;
+        return NULL;
+    }
+    /* the area is reserved from the end of wksp.
+     * If it overlaps with tableValidEnd, it voids guarantees on values' range */
+    if (alloc < ws->tableValidEnd) {
+        ws->tableValidEnd = alloc;
+    }
+    ws->allocStart = alloc;
+    return alloc;
+}
+
+/**
+ * Moves the cwksp to the next phase, and does any necessary allocations.
+ * cwksp initialization must necessarily go through each phase in order.
+ * Returns a 0 on success, or zstd error
+ */
+MEM_STATIC size_t
+ZSTD_cwksp_internal_advance_phase(ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase)
+{
+    assert(phase >= ws->phase);
+    if (phase > ws->phase) {
+        /* Going from allocating objects to allocating initOnce / tables */
+        if (ws->phase < ZSTD_cwksp_alloc_aligned_init_once &&
+            phase >= ZSTD_cwksp_alloc_aligned_init_once) {
+            ws->tableValidEnd = ws->objectEnd;
+            ws->initOnceStart = ZSTD_cwksp_initialAllocStart(ws);
+
+            {   /* Align the start of the tables to 64 bytes. Use [0, 63] bytes */
+                void *const alloc = ws->objectEnd;
+                size_t const bytesToAlign = ZSTD_cwksp_bytes_to_align_ptr(alloc, ZSTD_CWKSP_ALIGNMENT_BYTES);
+                void *const objectEnd = (BYTE *) alloc + bytesToAlign;
+                DEBUGLOG(5, "reserving table alignment addtl space: %zu", bytesToAlign);
+                RETURN_ERROR_IF(objectEnd > ws->workspaceEnd, memory_allocation,
+                                "table phase - alignment initial allocation failed!");
+                ws->objectEnd = objectEnd;
+                ws->tableEnd = objectEnd;  /* table area starts being empty */
+                if (ws->tableValidEnd < ws->tableEnd) {
+                    ws->tableValidEnd = ws->tableEnd;
+                }
+            }
+        }
+        ws->phase = phase;
+        ZSTD_cwksp_assert_internal_consistency(ws);
+    }
+    return 0;
+}
+
+/**
+ * Returns whether this object/buffer/etc was allocated in this workspace.
+ */
+MEM_STATIC int ZSTD_cwksp_owns_buffer(const ZSTD_cwksp* ws, const void* ptr)
+{
+    return (ptr != NULL) && (ws->workspace <= ptr) && (ptr < ws->workspaceEnd);
+}
+
+/**
+ * Internal function. Do not use directly.
+ */
+MEM_STATIC void*
+ZSTD_cwksp_reserve_internal(ZSTD_cwksp* ws, size_t bytes, ZSTD_cwksp_alloc_phase_e phase)
+{
+    void* alloc;
+    if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase)) || bytes == 0) {
+        return NULL;
+    }
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* over-reserve space */
+    bytes += 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+#endif
+
+    alloc = ZSTD_cwksp_reserve_internal_buffer_space(ws, bytes);
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
+     * either size. */
+    if (alloc) {
+        alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+        if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+            /* We need to keep the redzone poisoned while unpoisoning the bytes that
+             * are actually allocated. */
+            __asan_unpoison_memory_region(alloc, bytes - 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE);
+        }
+    }
+#endif
+
+    return alloc;
+}
+
+/**
+ * Reserves and returns unaligned memory.
+ */
+MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes)
+{
+    return (BYTE*)ZSTD_cwksp_reserve_internal(ws, bytes, ZSTD_cwksp_alloc_buffers);
+}
+
+/**
+ * Reserves and returns memory sized on and aligned on ZSTD_CWKSP_ALIGNMENT_BYTES (64 bytes).
+ * This memory has been initialized at least once in the past.
+ * This doesn't mean it has been initialized this time, and it might contain data from previous
+ * operations.
+ * The main usage is for algorithms that might need read access into uninitialized memory.
+ * The algorithm must maintain safety under these conditions and must make sure it doesn't
+ * leak any of the past data (directly or in side channels).
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_aligned_init_once(ZSTD_cwksp* ws, size_t bytes)
+{
+    size_t const alignedBytes = ZSTD_cwksp_align(bytes, ZSTD_CWKSP_ALIGNMENT_BYTES);
+    void* ptr = ZSTD_cwksp_reserve_internal(ws, alignedBytes, ZSTD_cwksp_alloc_aligned_init_once);
+    assert(((size_t)ptr & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
+    if(ptr && ptr < ws->initOnceStart) {
+        /* We assume the memory following the current allocation is either:
+         * 1. Not usable as initOnce memory (end of workspace)
+         * 2. Another initOnce buffer that has been allocated before (and so was previously memset)
+         * 3. An ASAN redzone, in which case we don't want to write on it
+         * For these reasons it should be fine to not explicitly zero every byte up to ws->initOnceStart.
+         * Note that we assume here that MSAN and ASAN cannot run in the same time. */
+        ZSTD_memset(ptr, 0, MIN((size_t)((U8*)ws->initOnceStart - (U8*)ptr), alignedBytes));
+        ws->initOnceStart = ptr;
+    }
+#if ZSTD_MEMORY_SANITIZER
+    assert(__msan_test_shadow(ptr, bytes) == -1);
+#endif
+    return ptr;
+}
+
+/**
+ * Reserves and returns memory sized on and aligned on ZSTD_CWKSP_ALIGNMENT_BYTES (64 bytes).
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes)
+{
+    void* ptr = ZSTD_cwksp_reserve_internal(ws, ZSTD_cwksp_align(bytes, ZSTD_CWKSP_ALIGNMENT_BYTES),
+                                            ZSTD_cwksp_alloc_aligned);
+    assert(((size_t)ptr & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
+    return ptr;
+}
+
+/**
+ * Aligned on 64 bytes. These buffers have the special property that
+ * their values remain constrained, allowing us to re-use them without
+ * memset()-ing them.
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes)
+{
+    const ZSTD_cwksp_alloc_phase_e phase = ZSTD_cwksp_alloc_aligned_init_once;
+    void* alloc;
+    void* end;
+    void* top;
+
+    /* We can only start allocating tables after we are done reserving space for objects at the
+     * start of the workspace */
+    if(ws->phase < phase) {
+        if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase))) {
+            return NULL;
+        }
+    }
+    alloc = ws->tableEnd;
+    end = (BYTE *)alloc + bytes;
+    top = ws->allocStart;
+
+    DEBUGLOG(5, "cwksp: reserving %p table %zd bytes, %zd bytes remaining",
+        alloc, bytes, ZSTD_cwksp_available_space(ws) - bytes);
+    assert((bytes & (sizeof(U32)-1)) == 0);
+    ZSTD_cwksp_assert_internal_consistency(ws);
+    assert(end <= top);
+    if (end > top) {
+        DEBUGLOG(4, "cwksp: table alloc failed!");
+        ws->allocFailed = 1;
+        return NULL;
+    }
+    ws->tableEnd = end;
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+        __asan_unpoison_memory_region(alloc, bytes);
+    }
+#endif
+
+    assert((bytes & (ZSTD_CWKSP_ALIGNMENT_BYTES-1)) == 0);
+    assert(((size_t)alloc & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
+    return alloc;
+}
+
+/**
+ * Aligned on sizeof(void*).
+ * Note : should happen only once, at workspace first initialization
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes)
+{
+    size_t const roundedBytes = ZSTD_cwksp_align(bytes, sizeof(void*));
+    void* alloc = ws->objectEnd;
+    void* end = (BYTE*)alloc + roundedBytes;
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* over-reserve space */
+    end = (BYTE *)end + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+#endif
+
+    DEBUGLOG(4,
+        "cwksp: reserving %p object %zd bytes (rounded to %zd), %zd bytes remaining",
+        alloc, bytes, roundedBytes, ZSTD_cwksp_available_space(ws) - roundedBytes);
+    assert((size_t)alloc % ZSTD_ALIGNOF(void*) == 0);
+    assert(bytes % ZSTD_ALIGNOF(void*) == 0);
+    ZSTD_cwksp_assert_internal_consistency(ws);
+    /* we must be in the first phase, no advance is possible */
+    if (ws->phase != ZSTD_cwksp_alloc_objects || end > ws->workspaceEnd) {
+        DEBUGLOG(3, "cwksp: object alloc failed!");
+        ws->allocFailed = 1;
+        return NULL;
+    }
+    ws->objectEnd = end;
+    ws->tableEnd = end;
+    ws->tableValidEnd = end;
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
+     * either size. */
+    alloc = (BYTE*)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+        __asan_unpoison_memory_region(alloc, bytes);
+    }
+#endif
+
+    return alloc;
+}
+
+MEM_STATIC void ZSTD_cwksp_mark_tables_dirty(ZSTD_cwksp* ws)
+{
+    DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_dirty");
+
+#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+    /* To validate that the table re-use logic is sound, and that we don't
+     * access table space that we haven't cleaned, we re-"poison" the table
+     * space every time we mark it dirty.
+     * Since tableValidEnd space and initOnce space may overlap we don't poison
+     * the initOnce portion as it break its promise. This means that this poisoning
+     * check isn't always applied fully. */
+    {
+        size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
+        assert(__msan_test_shadow(ws->objectEnd, size) == -1);
+        if((BYTE*)ws->tableValidEnd < (BYTE*)ws->initOnceStart) {
+            __msan_poison(ws->objectEnd, size);
+        } else {
+            assert(ws->initOnceStart >= ws->objectEnd);
+            __msan_poison(ws->objectEnd, (BYTE*)ws->initOnceStart - (BYTE*)ws->objectEnd);
+        }
+    }
+#endif
+
+    assert(ws->tableValidEnd >= ws->objectEnd);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    ws->tableValidEnd = ws->objectEnd;
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+MEM_STATIC void ZSTD_cwksp_mark_tables_clean(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_clean");
+    assert(ws->tableValidEnd >= ws->objectEnd);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    if (ws->tableValidEnd < ws->tableEnd) {
+        ws->tableValidEnd = ws->tableEnd;
+    }
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+/**
+ * Zero the part of the allocated tables not already marked clean.
+ */
+MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: ZSTD_cwksp_clean_tables");
+    assert(ws->tableValidEnd >= ws->objectEnd);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    if (ws->tableValidEnd < ws->tableEnd) {
+        ZSTD_memset(ws->tableValidEnd, 0, (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd));
+    }
+    ZSTD_cwksp_mark_tables_clean(ws);
+}
+
+/**
+ * Invalidates table allocations.
+ * All other allocations remain valid.
+ */
+MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: clearing tables!");
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* We don't do this when the workspace is statically allocated, because
+     * when that is the case, we have no capability to hook into the end of the
+     * workspace's lifecycle to unpoison the memory.
+     */
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+        size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
+        __asan_poison_memory_region(ws->objectEnd, size);
+    }
+#endif
+
+    ws->tableEnd = ws->objectEnd;
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+/**
+ * Invalidates all buffer, aligned, and table allocations.
+ * Object allocations remain valid.
+ */
+MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: clearing!");
+
+#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+    /* To validate that the context re-use logic is sound, and that we don't
+     * access stuff that this compression hasn't initialized, we re-"poison"
+     * the workspace except for the areas in which we expect memory re-use
+     * without initialization (objects, valid tables area and init once
+     * memory). */
+    {
+        if((BYTE*)ws->tableValidEnd < (BYTE*)ws->initOnceStart) {
+            size_t size = (BYTE*)ws->initOnceStart - (BYTE*)ws->tableValidEnd;
+            __msan_poison(ws->tableValidEnd, size);
+        }
+    }
+#endif
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* We don't do this when the workspace is statically allocated, because
+     * when that is the case, we have no capability to hook into the end of the
+     * workspace's lifecycle to unpoison the memory.
+     */
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+        size_t size = (BYTE*)ws->workspaceEnd - (BYTE*)ws->objectEnd;
+        __asan_poison_memory_region(ws->objectEnd, size);
+    }
+#endif
+
+    ws->tableEnd = ws->objectEnd;
+    ws->allocStart = ZSTD_cwksp_initialAllocStart(ws);
+    ws->allocFailed = 0;
+    if (ws->phase > ZSTD_cwksp_alloc_aligned_init_once) {
+        ws->phase = ZSTD_cwksp_alloc_aligned_init_once;
+    }
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+/**
+ * The provided workspace takes ownership of the buffer [start, start+size).
+ * Any existing values in the workspace are ignored (the previously managed
+ * buffer, if present, must be separately freed).
+ */
+MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size, ZSTD_cwksp_static_alloc_e isStatic) {
+    DEBUGLOG(4, "cwksp: init'ing workspace with %zd bytes", size);
+    assert(((size_t)start & (sizeof(void*)-1)) == 0); /* ensure correct alignment */
+    ws->workspace = start;
+    ws->workspaceEnd = (BYTE*)start + size;
+    ws->objectEnd = ws->workspace;
+    ws->tableValidEnd = ws->objectEnd;
+    ws->initOnceStart = ZSTD_cwksp_initialAllocStart(ws);
+    ws->phase = ZSTD_cwksp_alloc_objects;
+    ws->isStatic = isStatic;
+    ZSTD_cwksp_clear(ws);
+    ws->workspaceOversizedDuration = 0;
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+MEM_STATIC size_t ZSTD_cwksp_create(ZSTD_cwksp* ws, size_t size, ZSTD_customMem customMem) {
+    void* workspace = ZSTD_customMalloc(size, customMem);
+    DEBUGLOG(4, "cwksp: creating new workspace with %zd bytes", size);
+    RETURN_ERROR_IF(workspace == NULL, memory_allocation, "NULL pointer!");
+    ZSTD_cwksp_init(ws, workspace, size, ZSTD_cwksp_dynamic_alloc);
+    return 0;
+}
+
+MEM_STATIC void ZSTD_cwksp_free(ZSTD_cwksp* ws, ZSTD_customMem customMem) {
+    void *ptr = ws->workspace;
+    DEBUGLOG(4, "cwksp: freeing workspace");
+    ZSTD_memset(ws, 0, sizeof(ZSTD_cwksp));
+    ZSTD_customFree(ptr, customMem);
+}
+
+/**
+ * Moves the management of a workspace from one cwksp to another. The src cwksp
+ * is left in an invalid state (src must be re-init()'ed before it's used again).
+ */
+MEM_STATIC void ZSTD_cwksp_move(ZSTD_cwksp* dst, ZSTD_cwksp* src) {
+    *dst = *src;
+    ZSTD_memset(src, 0, sizeof(ZSTD_cwksp));
+}
+
+MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace);
+}
+
+MEM_STATIC size_t ZSTD_cwksp_used(const ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->workspace)
+         + (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->allocStart);
+}
+
+MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
+    return ws->allocFailed;
+}
+
+/*-*************************************
+*  Functions Checking Free Space
+***************************************/
+
+/* ZSTD_alignmentSpaceWithinBounds() :
+ * Returns if the estimated space needed for a wksp is within an acceptable limit of the
+ * actual amount of space used.
+ */
+MEM_STATIC int ZSTD_cwksp_estimated_space_within_bounds(const ZSTD_cwksp *const ws, size_t const estimatedSpace) {
+    /* We have an alignment space between objects and tables between tables and buffers, so we can have up to twice
+     * the alignment bytes difference between estimation and actual usage */
+    return (estimatedSpace - ZSTD_cwksp_slack_space_required()) <= ZSTD_cwksp_used(ws) &&
+           ZSTD_cwksp_used(ws) <= estimatedSpace;
+}
+
+
+MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->allocStart - (BYTE*)ws->tableEnd);
+}
+
+MEM_STATIC int ZSTD_cwksp_check_available(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    return ZSTD_cwksp_available_space(ws) >= additionalNeededSpace;
+}
+
+MEM_STATIC int ZSTD_cwksp_check_too_large(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    return ZSTD_cwksp_check_available(
+        ws, additionalNeededSpace * ZSTD_WORKSPACETOOLARGE_FACTOR);
+}
+
+MEM_STATIC int ZSTD_cwksp_check_wasteful(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    return ZSTD_cwksp_check_too_large(ws, additionalNeededSpace)
+        && ws->workspaceOversizedDuration > ZSTD_WORKSPACETOOLARGE_MAXDURATION;
+}
+
+MEM_STATIC void ZSTD_cwksp_bump_oversized_duration(
+        ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    if (ZSTD_cwksp_check_too_large(ws, additionalNeededSpace)) {
+        ws->workspaceOversizedDuration++;
+    } else {
+        ws->workspaceOversizedDuration = 0;
+    }
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_CWKSP_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_double_fast.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_double_fast.c
new file mode 100644
index 0000000..0ad88ff
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_double_fast.c
@@ -0,0 +1,758 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_double_fast.h"
+
+static void ZSTD_fillDoubleHashTableForCDict(ZSTD_matchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    U32  const mls = cParams->minMatch;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHashAndTag = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHashAndTag = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0) {
+                ZSTD_writeTaggedIndex(hashSmall, smHashAndTag, curr + i);
+            }
+            if (i == 0 || hashLarge[lgHashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {
+                ZSTD_writeTaggedIndex(hashLarge, lgHashAndTag, curr + i);
+            }
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
+    }   }
+}
+
+static void ZSTD_fillDoubleHashTableForCCtx(ZSTD_matchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32  const mls = cParams->minMatch;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0)
+                hashSmall[smHash] = curr + i;
+            if (i == 0 || hashLarge[lgHash] == 0)
+                hashLarge[lgHash] = curr + i;
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
+        }   }
+}
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    if (tfp == ZSTD_tfp_forCDict) {
+        ZSTD_fillDoubleHashTableForCDict(ms, end, dtlm);
+    } else {
+        ZSTD_fillDoubleHashTableForCCtx(ms, end, dtlm);
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_doubleFast_noDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize, U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* anchor = istart;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    /* presumes that, if there is a dictionary, it must be using Attach mode */
+    const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    size_t mLength;
+    U32 offset;
+    U32 curr;
+
+    /* how many positions to search before increasing step size */
+    const size_t kStepIncr = 1 << kSearchStrength;
+    /* the position at which to increment the step size if no match is found */
+    const BYTE* nextStep;
+    size_t step; /* the current step size */
+
+    size_t hl0; /* the long hash at ip */
+    size_t hl1; /* the long hash at ip1 */
+
+    U32 idxl0; /* the long match index for ip */
+    U32 idxl1; /* the long match index for ip1 */
+
+    const BYTE* matchl0; /* the long match for ip */
+    const BYTE* matchs0; /* the short match for ip */
+    const BYTE* matchl1; /* the long match for ip1 */
+
+    const BYTE* ip = istart; /* the current position */
+    const BYTE* ip1; /* the next position */
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
+
+    /* init */
+    ip += ((ip - prefixLowest) == 0);
+    {
+        U32 const current = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
+        U32 const maxRep = current - windowLow;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+
+    /* Outer Loop: one iteration per match found and stored */
+    while (1) {
+        step = 1;
+        nextStep = ip + kStepIncr;
+        ip1 = ip + step;
+
+        if (ip1 > ilimit) {
+            goto _cleanup;
+        }
+
+        hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
+        idxl0 = hashLong[hl0];
+        matchl0 = base + idxl0;
+
+        /* Inner Loop: one iteration per search / position */
+        do {
+            const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
+            const U32 idxs0 = hashSmall[hs0];
+            curr = (U32)(ip-base);
+            matchs0 = base + idxs0;
+
+            hashLong[hl0] = hashSmall[hs0] = curr;   /* update hash tables */
+
+            /* check noDict repcode */
+            if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+                mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+                ip++;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+                goto _match_stored;
+            }
+
+            hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
+
+            if (idxl0 > prefixLowestIndex) {
+                /* check prefix long match */
+                if (MEM_read64(matchl0) == MEM_read64(ip)) {
+                    mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
+                    offset = (U32)(ip-matchl0);
+                    while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
+                    goto _match_found;
+                }
+            }
+
+            idxl1 = hashLong[hl1];
+            matchl1 = base + idxl1;
+
+            if (idxs0 > prefixLowestIndex) {
+                /* check prefix short match */
+                if (MEM_read32(matchs0) == MEM_read32(ip)) {
+                    goto _search_next_long;
+                }
+            }
+
+            if (ip1 >= nextStep) {
+                PREFETCH_L1(ip1 + 64);
+                PREFETCH_L1(ip1 + 128);
+                step++;
+                nextStep += kStepIncr;
+            }
+            ip = ip1;
+            ip1 += step;
+
+            hl0 = hl1;
+            idxl0 = idxl1;
+            matchl0 = matchl1;
+    #if defined(__aarch64__)
+            PREFETCH_L1(ip+256);
+    #endif
+        } while (ip1 <= ilimit);
+
+_cleanup:
+        /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+         * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+        offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+        /* save reps for next block */
+        rep[0] = offset_1 ? offset_1 : offsetSaved1;
+        rep[1] = offset_2 ? offset_2 : offsetSaved2;
+
+        /* Return the last literals size */
+        return (size_t)(iend - anchor);
+
+_search_next_long:
+
+        /* check prefix long +1 match */
+        if (idxl1 > prefixLowestIndex) {
+            if (MEM_read64(matchl1) == MEM_read64(ip1)) {
+                ip = ip1;
+                mLength = ZSTD_count(ip+8, matchl1+8, iend) + 8;
+                offset = (U32)(ip-matchl1);
+                while (((ip>anchor) & (matchl1>prefixLowest)) && (ip[-1] == matchl1[-1])) { ip--; matchl1--; mLength++; } /* catch up */
+                goto _match_found;
+            }
+        }
+
+        /* if no long +1 match, explore the short match we found */
+        mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
+        offset = (U32)(ip - matchs0);
+        while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* catch up */
+
+        /* fall-through */
+
+_match_found: /* requires ip, offset, mLength */
+        offset_2 = offset_1;
+        offset_1 = offset;
+
+        if (step < 4) {
+            /* It is unsafe to write this value back to the hashtable when ip1 is
+             * greater than or equal to the new ip we will have after we're done
+             * processing this match. Rather than perform that test directly
+             * (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
+             * more predictable test. The minmatch even if we take a short match is
+             * 4 bytes, so as long as step, the distance between ip and ip1
+             * (initially) is less than 4, we know ip1 < new ip. */
+            hashLong[hl1] = (U32)(ip1 - base);
+        }
+
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+_match_stored:
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = curr+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                    & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
+                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+        }
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    /* presumes that, if there is a dictionary, it must be using Attach mode */
+    const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
+    const U32* const dictHashLong  = dms->hashTable;
+    const U32* const dictHashSmall = dms->chainTable;
+    const U32 dictStartIndex       = dms->window.dictLimit;
+    const BYTE* const dictBase     = dms->window.base;
+    const BYTE* const dictStart    = dictBase + dictStartIndex;
+    const BYTE* const dictEnd      = dms->window.nextSrc;
+    const U32 dictIndexDelta       = prefixLowestIndex - (U32)(dictEnd - dictBase);
+    const U32 dictHBitsL           = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const U32 dictHBitsS           = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
+
+    /* if a dictionary is attached, it must be within window range */
+    assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
+
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
+        size_t const chainTableBytes = (((size_t)1) << dictCParams->chainLog) * sizeof(U32);
+        PREFETCH_AREA(dictHashLong, hashTableBytes)
+        PREFETCH_AREA(dictHashSmall, chainTableBytes)
+    }
+
+    /* init */
+    ip += (dictAndPrefixLength == 0);
+
+    /* dictMatchState repCode checks don't currently handle repCode == 0
+     * disabling. */
+    assert(offset_1 <= dictAndPrefixLength);
+    assert(offset_2 <= dictAndPrefixLength);
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        U32 offset;
+        size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
+        size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
+        size_t const dictHashAndTagL = ZSTD_hashPtr(ip, dictHBitsL, 8);
+        size_t const dictHashAndTagS = ZSTD_hashPtr(ip, dictHBitsS, mls);
+        U32 const dictMatchIndexAndTagL = dictHashLong[dictHashAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS];
+        U32 const dictMatchIndexAndTagS = dictHashSmall[dictHashAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS];
+        int const dictTagsMatchL = ZSTD_comparePackedTags(dictMatchIndexAndTagL, dictHashAndTagL);
+        int const dictTagsMatchS = ZSTD_comparePackedTags(dictMatchIndexAndTagS, dictHashAndTagS);
+        U32 const curr = (U32)(ip-base);
+        U32 const matchIndexL = hashLong[h2];
+        U32 matchIndexS = hashSmall[h];
+        const BYTE* matchLong = base + matchIndexL;
+        const BYTE* match = base + matchIndexS;
+        const U32 repIndex = curr + 1 - offset_1;
+        const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
+        hashLong[h2] = hashSmall[h] = curr;   /* update hash tables */
+
+        /* check repcode */
+        if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+            goto _match_stored;
+        }
+
+        if (matchIndexL > prefixLowestIndex) {
+            /* check prefix long match */
+            if (MEM_read64(matchLong) == MEM_read64(ip)) {
+                mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
+                offset = (U32)(ip-matchLong);
+                while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
+                goto _match_found;
+            }
+        } else if (dictTagsMatchL) {
+            /* check dictMatchState long match */
+            U32 const dictMatchIndexL = dictMatchIndexAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS;
+            const BYTE* dictMatchL = dictBase + dictMatchIndexL;
+            assert(dictMatchL < dictEnd);
+
+            if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
+                mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
+                offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
+                while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
+                goto _match_found;
+        }   }
+
+        if (matchIndexS > prefixLowestIndex) {
+            /* check prefix short match */
+            if (MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+            }
+        } else if (dictTagsMatchS) {
+            /* check dictMatchState short match */
+            U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
+            match = dictBase + dictMatchIndexS;
+            matchIndexS = dictMatchIndexS + dictIndexDelta;
+
+            if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+        }   }
+
+        ip += ((ip-anchor) >> kSearchStrength) + 1;
+#if defined(__aarch64__)
+        PREFETCH_L1(ip+256);
+#endif
+        continue;
+
+_search_next_long:
+        {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+            size_t const dictHashAndTagL3 = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
+            U32 const matchIndexL3 = hashLong[hl3];
+            U32 const dictMatchIndexAndTagL3 = dictHashLong[dictHashAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS];
+            int const dictTagsMatchL3 = ZSTD_comparePackedTags(dictMatchIndexAndTagL3, dictHashAndTagL3);
+            const BYTE* matchL3 = base + matchIndexL3;
+            hashLong[hl3] = curr + 1;
+
+            /* check prefix long +1 match */
+            if (matchIndexL3 > prefixLowestIndex) {
+                if (MEM_read64(matchL3) == MEM_read64(ip+1)) {
+                    mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
+                    ip++;
+                    offset = (U32)(ip-matchL3);
+                    while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
+                    goto _match_found;
+                }
+            } else if (dictTagsMatchL3) {
+                /* check dict long +1 match */
+                U32 const dictMatchIndexL3 = dictMatchIndexAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS;
+                const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
+                assert(dictMatchL3 < dictEnd);
+                if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
+                    mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
+                    ip++;
+                    offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
+                    while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
+                    goto _match_found;
+        }   }   }
+
+        /* if no long +1 match, explore the short match we found */
+        if (matchIndexS < prefixLowestIndex) {
+            mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
+            offset = (U32)(curr - matchIndexS);
+            while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+        } else {
+            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+            offset = (U32)(ip - match);
+            while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+        }
+
+_match_found:
+        offset_2 = offset_1;
+        offset_1 = offset;
+
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+_match_stored:
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = curr+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
+                        dictBase + repIndex2 - dictIndexDelta :
+                        base + repIndex2;
+                if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+            }
+        }
+    }   /* while (ip < ilimit) */
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+#define ZSTD_GEN_DFAST_FN(dictMode, mls)                                                                 \
+    static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls(                                      \
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                          \
+            void const* src, size_t srcSize)                                                             \
+    {                                                                                                    \
+        return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
+    }
+
+ZSTD_GEN_DFAST_FN(noDict, 4)
+ZSTD_GEN_DFAST_FN(noDict, 5)
+ZSTD_GEN_DFAST_FN(noDict, 6)
+ZSTD_GEN_DFAST_FN(noDict, 7)
+
+ZSTD_GEN_DFAST_FN(dictMatchState, 4)
+ZSTD_GEN_DFAST_FN(dictMatchState, 5)
+ZSTD_GEN_DFAST_FN(dictMatchState, 6)
+ZSTD_GEN_DFAST_FN(dictMatchState, 7)
+
+
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    const U32 mls = ms->cParams.minMatch;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
+    }
+}
+
+
+size_t ZSTD_compressBlock_doubleFast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    const U32 mls = ms->cParams.minMatch;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
+    }
+}
+
+
+static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
+    const U32   dictStartIndex = lowLimit;
+    const U32   dictLimit = ms->window.dictLimit;
+    const U32   prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictStart = dictBase + dictStartIndex;
+    const BYTE* const dictEnd = dictBase + prefixStartIndex;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
+
+    /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
+    if (prefixStartIndex == dictStartIndex)
+        return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
+        const U32 matchIndex = hashSmall[hSmall];
+        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+
+        const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
+        const U32 matchLongIndex = hashLong[hLong];
+        const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* matchLong = matchLongBase + matchLongIndex;
+
+        const U32 curr = (U32)(ip-base);
+        const U32 repIndex = curr + 1 - offset_1;   /* offset_1 expected <= curr +1 */
+        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* const repMatch = repBase + repIndex;
+        size_t mLength;
+        hashSmall[hSmall] = hashLong[hLong] = curr;   /* update hash table */
+
+        if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
+            & (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
+          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+        } else {
+            if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+                const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
+                const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
+                offset = curr - matchLongIndex;
+                while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+            } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
+                size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndex3 = hashLong[h3];
+                const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
+                const BYTE* match3 = match3Base + matchIndex3;
+                U32 offset;
+                hashLong[h3] = curr + 1;
+                if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
+                    const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
+                    ip++;
+                    offset = curr+1 - matchIndex3;
+                    while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
+                } else {
+                    const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
+                    offset = curr - matchIndex;
+                    while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                }
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+            } else {
+                ip += ((ip-anchor) >> kSearchStrength) + 1;
+                continue;
+        }   }
+
+        /* move to next sequence start */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = curr+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3)   /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
+                    & (offset_2 <= current2 - dictStartIndex))
+                  && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+ZSTD_GEN_DFAST_FN(extDict, 4)
+ZSTD_GEN_DFAST_FN(extDict, 5)
+ZSTD_GEN_DFAST_FN(extDict, 6)
+ZSTD_GEN_DFAST_FN(extDict, 7)
+
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    U32 const mls = ms->cParams.minMatch;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
+    }
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_double_fast.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_double_fast.h
new file mode 100644
index 0000000..6f0047c
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_double_fast.h
@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_DOUBLE_FAST_H
+#define ZSTD_DOUBLE_FAST_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "../common/mem.h"      /* U32 */
+#include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm,
+                              ZSTD_tableFillPurpose_e tfp);
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_doubleFast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_DOUBLE_FAST_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_fast.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_fast.c
new file mode 100644
index 0000000..5f2c6a2
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_fast.c
@@ -0,0 +1,960 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
+#include "zstd_fast.h"
+
+static void ZSTD_fillHashTableForCDict(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hBits = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    U32  const mls = cParams->minMatch;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Currently, we always use ZSTD_dtlm_full for filling CDict tables.
+     * Feel free to remove this assert if there's a good reason! */
+    assert(dtlm == ZSTD_dtlm_full);
+
+    /* Always insert every fastHashFillStep position into the hash table.
+     * Insert the other positions if their hash entry is empty.
+     */
+    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        {   size_t const hashAndTag = ZSTD_hashPtr(ip, hBits, mls);
+            ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr);   }
+
+        if (dtlm == ZSTD_dtlm_fast) continue;
+        /* Only load extra positions for ZSTD_dtlm_full */
+        {   U32 p;
+            for (p = 1; p < fastHashFillStep; ++p) {
+                size_t const hashAndTag = ZSTD_hashPtr(ip + p, hBits, mls);
+                if (hashTable[hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {  /* not yet filled */
+                    ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr + p);
+                }   }   }   }
+}
+
+static void ZSTD_fillHashTableForCCtx(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hBits = cParams->hashLog;
+    U32  const mls = cParams->minMatch;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Currently, we always use ZSTD_dtlm_fast for filling CCtx tables.
+     * Feel free to remove this assert if there's a good reason! */
+    assert(dtlm == ZSTD_dtlm_fast);
+
+    /* Always insert every fastHashFillStep position into the hash table.
+     * Insert the other positions if their hash entry is empty.
+     */
+    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
+        hashTable[hash0] = curr;
+        if (dtlm == ZSTD_dtlm_fast) continue;
+        /* Only load extra positions for ZSTD_dtlm_full */
+        {   U32 p;
+            for (p = 1; p < fastHashFillStep; ++p) {
+                size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
+                if (hashTable[hash] == 0) {  /* not yet filled */
+                    hashTable[hash] = curr + p;
+    }   }   }   }
+}
+
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    if (tfp == ZSTD_tfp_forCDict) {
+        ZSTD_fillHashTableForCDict(ms, end, dtlm);
+    } else {
+        ZSTD_fillHashTableForCCtx(ms, end, dtlm);
+    }
+}
+
+
+/**
+ * If you squint hard enough (and ignore repcodes), the search operation at any
+ * given position is broken into 4 stages:
+ *
+ * 1. Hash   (map position to hash value via input read)
+ * 2. Lookup (map hash val to index via hashtable read)
+ * 3. Load   (map index to value at that position via input read)
+ * 4. Compare
+ *
+ * Each of these steps involves a memory read at an address which is computed
+ * from the previous step. This means these steps must be sequenced and their
+ * latencies are cumulative.
+ *
+ * Rather than do 1->2->3->4 sequentially for a single position before moving
+ * onto the next, this implementation interleaves these operations across the
+ * next few positions:
+ *
+ * R = Repcode Read & Compare
+ * H = Hash
+ * T = Table Lookup
+ * M = Match Read & Compare
+ *
+ * Pos | Time -->
+ * ----+-------------------
+ * N   | ... M
+ * N+1 | ...   TM
+ * N+2 |    R H   T M
+ * N+3 |         H    TM
+ * N+4 |           R H   T M
+ * N+5 |                H   ...
+ * N+6 |                  R ...
+ *
+ * This is very much analogous to the pipelining of execution in a CPU. And just
+ * like a CPU, we have to dump the pipeline when we find a match (i.e., take a
+ * branch).
+ *
+ * When this happens, we throw away our current state, and do the following prep
+ * to re-enter the loop:
+ *
+ * Pos | Time -->
+ * ----+-------------------
+ * N   | H T
+ * N+1 |  H
+ *
+ * This is also the work we do at the beginning to enter the loop initially.
+ */
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_compressBlock_fast_noDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls, U32 const hasStep)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    size_t const stepSize = hasStep ? (cParams->targetLength + !(cParams->targetLength) + 1) : 2;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+
+    const BYTE* anchor = istart;
+    const BYTE* ip0 = istart;
+    const BYTE* ip1;
+    const BYTE* ip2;
+    const BYTE* ip3;
+    U32 current0;
+
+    U32 rep_offset1 = rep[0];
+    U32 rep_offset2 = rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    size_t hash0; /* hash for ip0 */
+    size_t hash1; /* hash for ip1 */
+    U32 idx; /* match idx for ip0 */
+    U32 mval; /* src value at match idx */
+
+    U32 offcode;
+    const BYTE* match0;
+    size_t mLength;
+
+    /* ip0 and ip1 are always adjacent. The targetLength skipping and
+     * uncompressibility acceleration is applied to every other position,
+     * matching the behavior of #1562. step therefore represents the gap
+     * between pairs of positions, from ip0 to ip2 or ip1 to ip3. */
+    size_t step;
+    const BYTE* nextStep;
+    const size_t kStepIncr = (1 << (kSearchStrength - 1));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
+    ip0 += (ip0 == prefixStart);
+    {   U32 const curr = (U32)(ip0 - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
+        U32 const maxRep = curr - windowLow;
+        if (rep_offset2 > maxRep) offsetSaved2 = rep_offset2, rep_offset2 = 0;
+        if (rep_offset1 > maxRep) offsetSaved1 = rep_offset1, rep_offset1 = 0;
+    }
+
+    /* start each op */
+_start: /* Requires: ip0 */
+
+    step = stepSize;
+    nextStep = ip0 + kStepIncr;
+
+    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
+    ip1 = ip0 + 1;
+    ip2 = ip0 + step;
+    ip3 = ip2 + 1;
+
+    if (ip3 >= ilimit) {
+        goto _cleanup;
+    }
+
+    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+    hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+
+    idx = hashTable[hash0];
+
+    do {
+        /* load repcode match for ip[2]*/
+        const U32 rval = MEM_read32(ip2 - rep_offset1);
+
+        /* write back hash table entry */
+        current0 = (U32)(ip0 - base);
+        hashTable[hash0] = current0;
+
+        /* check repcode at ip[2] */
+        if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {
+            ip0 = ip2;
+            match0 = ip0 - rep_offset1;
+            mLength = ip0[-1] == match0[-1];
+            ip0 -= mLength;
+            match0 -= mLength;
+            offcode = REPCODE1_TO_OFFBASE;
+            mLength += 4;
+
+            /* First write next hash table entry; we've already calculated it.
+             * This write is known to be safe because the ip1 is before the
+             * repcode (ip2). */
+            hashTable[hash1] = (U32)(ip1 - base);
+
+            goto _match;
+        }
+
+        /* load match for ip[0] */
+        if (idx >= prefixStartIndex) {
+            mval = MEM_read32(base + idx);
+        } else {
+            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
+        }
+
+        /* check match at ip[0] */
+        if (MEM_read32(ip0) == mval) {
+            /* found a match! */
+
+            /* First write next hash table entry; we've already calculated it.
+             * This write is known to be safe because the ip1 == ip0 + 1, so
+             * we know we will resume searching after ip1 */
+            hashTable[hash1] = (U32)(ip1 - base);
+
+            goto _offset;
+        }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip3;
+
+        /* write back hash table entry */
+        current0 = (U32)(ip0 - base);
+        hashTable[hash0] = current0;
+
+        /* load match for ip[0] */
+        if (idx >= prefixStartIndex) {
+            mval = MEM_read32(base + idx);
+        } else {
+            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
+        }
+
+        /* check match at ip[0] */
+        if (MEM_read32(ip0) == mval) {
+            /* found a match! */
+
+            /* first write next hash table entry; we've already calculated it */
+            if (step <= 4) {
+                /* We need to avoid writing an index into the hash table >= the
+                 * position at which we will pick up our searching after we've
+                 * taken this match.
+                 *
+                 * The minimum possible match has length 4, so the earliest ip0
+                 * can be after we take this match will be the current ip0 + 4.
+                 * ip1 is ip0 + step - 1. If ip1 is >= ip0 + 4, we can't safely
+                 * write this position.
+                 */
+                hashTable[hash1] = (U32)(ip1 - base);
+            }
+
+            goto _offset;
+        }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip0 + step;
+        ip3 = ip1 + step;
+
+        /* calculate step */
+        if (ip2 >= nextStep) {
+            step++;
+            PREFETCH_L1(ip1 + 64);
+            PREFETCH_L1(ip1 + 128);
+            nextStep += kStepIncr;
+        }
+    } while (ip3 < ilimit);
+
+_cleanup:
+    /* Note that there are probably still a couple positions we could search.
+     * However, it seems to be a meaningful performance hit to try to search
+     * them. So let's not. */
+
+    /* When the repcodes are outside of the prefix, we set them to zero before the loop.
+     * When the offsets are still zero, we need to restore them after the block to have a correct
+     * repcode history. If only one offset was invalid, it is easy. The tricky case is when both
+     * offsets were invalid. We need to figure out which offset to refill with.
+     *     - If both offsets are zero they are in the same order.
+     *     - If both offsets are non-zero, we won't restore the offsets from `offsetSaved[12]`.
+     *     - If only one is zero, we need to decide which offset to restore.
+     *         - If rep_offset1 is non-zero, then rep_offset2 must be offsetSaved1.
+     *         - It is impossible for rep_offset2 to be non-zero.
+     *
+     * So if rep_offset1 started invalid (offsetSaved1 != 0) and became valid (rep_offset1 != 0), then
+     * set rep[0] = rep_offset1 and rep[1] = offsetSaved1.
+     */
+    offsetSaved2 = ((offsetSaved1 != 0) && (rep_offset1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+    /* save reps for next block */
+    rep[0] = rep_offset1 ? rep_offset1 : offsetSaved1;
+    rep[1] = rep_offset2 ? rep_offset2 : offsetSaved2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+
+_offset: /* Requires: ip0, idx */
+
+    /* Compute the offset code. */
+    match0 = base + idx;
+    rep_offset2 = rep_offset1;
+    rep_offset1 = (U32)(ip0-match0);
+    offcode = OFFSET_TO_OFFBASE(rep_offset1);
+    mLength = 4;
+
+    /* Count the backwards match length. */
+    while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {
+        ip0--;
+        match0--;
+        mLength++;
+    }
+
+_match: /* Requires: ip0, match0, offcode */
+
+    /* Count the forward length. */
+    mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);
+
+    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
+
+    ip0 += mLength;
+    anchor = ip0;
+
+    /* Fill table and check for immediate repcode. */
+    if (ip0 <= ilimit) {
+        /* Fill Table */
+        assert(base+current0+2 > istart);  /* check base overflow */
+        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
+        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+
+        if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */
+            while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;
+                { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
+                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
+                ip0 += rLength;
+                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
+                anchor = ip0;
+                continue;   /* faster when present (confirmed on gcc-8) ... (?) */
+    }   }   }
+
+    goto _start;
+}
+
+#define ZSTD_GEN_FAST_FN(dictMode, mls, step)                                                            \
+    static size_t ZSTD_compressBlock_fast_##dictMode##_##mls##_##step(                                      \
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                    \
+            void const* src, size_t srcSize)                                                       \
+    {                                                                                              \
+        return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls, step); \
+    }
+
+ZSTD_GEN_FAST_FN(noDict, 4, 1)
+ZSTD_GEN_FAST_FN(noDict, 5, 1)
+ZSTD_GEN_FAST_FN(noDict, 6, 1)
+ZSTD_GEN_FAST_FN(noDict, 7, 1)
+
+ZSTD_GEN_FAST_FN(noDict, 4, 0)
+ZSTD_GEN_FAST_FN(noDict, 5, 0)
+ZSTD_GEN_FAST_FN(noDict, 6, 0)
+ZSTD_GEN_FAST_FN(noDict, 7, 0)
+
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    U32 const mls = ms->cParams.minMatch;
+    assert(ms->dictMatchState == NULL);
+    if (ms->cParams.targetLength > 1) {
+        switch(mls)
+        {
+        default: /* includes case 3 */
+        case 4 :
+            return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize);
+        case 5 :
+            return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize);
+        case 6 :
+            return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize);
+        case 7 :
+            return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize);
+        }
+    } else {
+        switch(mls)
+        {
+        default: /* includes case 3 */
+        case 4 :
+            return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize);
+        case 5 :
+            return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize);
+        case 6 :
+            return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize);
+        case 7 :
+            return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize);
+        }
+
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_fast_dictMatchState_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip0 = istart;
+    const BYTE* ip1 = ip0 + stepSize; /* we assert below that stepSize >= 1 */
+    const BYTE* anchor = istart;
+    const U32   prefixStartIndex = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
+    const U32* const dictHashTable = dms->hashTable;
+    const U32 dictStartIndex       = dms->window.dictLimit;
+    const BYTE* const dictBase     = dms->window.base;
+    const BYTE* const dictStart    = dictBase + dictStartIndex;
+    const BYTE* const dictEnd      = dms->window.nextSrc;
+    const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);
+    const U32 dictAndPrefixLength  = (U32)(istart - prefixStart + dictEnd - dictStart);
+    const U32 dictHBits            = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+
+    /* if a dictionary is still attached, it necessarily means that
+     * it is within window size. So we just check it. */
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    assert(endIndex - prefixStartIndex <= maxDistance);
+    (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */
+
+    (void)hasStep; /* not currently specialized on whether it's accelerated */
+
+    /* ensure there will be no underflow
+     * when translating a dict index into a local index */
+    assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
+
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
+        PREFETCH_AREA(dictHashTable, hashTableBytes)
+    }
+
+    /* init */
+    DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
+    ip0 += (dictAndPrefixLength == 0);
+    /* dictMatchState repCode checks don't currently handle repCode == 0
+     * disabling. */
+    assert(offset_1 <= dictAndPrefixLength);
+    assert(offset_2 <= dictAndPrefixLength);
+
+    /* Outer search loop */
+    assert(stepSize >= 1);
+    while (ip1 <= ilimit) {   /* repcode check at (ip0 + 1) is safe because ip0 < ip1 */
+        size_t mLength;
+        size_t hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+
+        size_t const dictHashAndTag0 = ZSTD_hashPtr(ip0, dictHBits, mls);
+        U32 dictMatchIndexAndTag = dictHashTable[dictHashAndTag0 >> ZSTD_SHORT_CACHE_TAG_BITS];
+        int dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag0);
+
+        U32 matchIndex = hashTable[hash0];
+        U32 curr = (U32)(ip0 - base);
+        size_t step = stepSize;
+        const size_t kStepIncr = 1 << kSearchStrength;
+        const BYTE* nextStep = ip0 + kStepIncr;
+
+        /* Inner search loop */
+        while (1) {
+            const BYTE* match = base + matchIndex;
+            const U32 repIndex = curr + 1 - offset_1;
+            const BYTE* repMatch = (repIndex < prefixStartIndex) ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+            const size_t hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+            size_t const dictHashAndTag1 = ZSTD_hashPtr(ip1, dictHBits, mls);
+            hashTable[hash0] = curr;   /* update hash table */
+
+            if (((U32) ((prefixStartIndex - 1) - repIndex) >=
+                 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
+                && (MEM_read32(repMatch) == MEM_read32(ip0 + 1))) {
+                const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+                mLength = ZSTD_count_2segments(ip0 + 1 + 4, repMatch + 4, iend, repMatchEnd, prefixStart) + 4;
+                ip0++;
+                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+                break;
+            }
+
+            if (dictTagsMatch) {
+                /* Found a possible dict match */
+                const U32 dictMatchIndex = dictMatchIndexAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
+                const BYTE* dictMatch = dictBase + dictMatchIndex;
+                if (dictMatchIndex > dictStartIndex &&
+                    MEM_read32(dictMatch) == MEM_read32(ip0)) {
+                    /* To replicate extDict parse behavior, we only use dict matches when the normal matchIndex is invalid */
+                    if (matchIndex <= prefixStartIndex) {
+                        U32 const offset = (U32) (curr - dictMatchIndex - dictIndexDelta);
+                        mLength = ZSTD_count_2segments(ip0 + 4, dictMatch + 4, iend, dictEnd, prefixStart) + 4;
+                        while (((ip0 > anchor) & (dictMatch > dictStart))
+                            && (ip0[-1] == dictMatch[-1])) {
+                            ip0--;
+                            dictMatch--;
+                            mLength++;
+                        } /* catch up */
+                        offset_2 = offset_1;
+                        offset_1 = offset;
+                        ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+                        break;
+                    }
+                }
+            }
+
+            if (matchIndex > prefixStartIndex && MEM_read32(match) == MEM_read32(ip0)) {
+                /* found a regular match */
+                U32 const offset = (U32) (ip0 - match);
+                mLength = ZSTD_count(ip0 + 4, match + 4, iend) + 4;
+                while (((ip0 > anchor) & (match > prefixStart))
+                       && (ip0[-1] == match[-1])) {
+                    ip0--;
+                    match--;
+                    mLength++;
+                } /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+                break;
+            }
+
+            /* Prepare for next iteration */
+            dictMatchIndexAndTag = dictHashTable[dictHashAndTag1 >> ZSTD_SHORT_CACHE_TAG_BITS];
+            dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag1);
+            matchIndex = hashTable[hash1];
+
+            if (ip1 >= nextStep) {
+                step++;
+                nextStep += kStepIncr;
+            }
+            ip0 = ip1;
+            ip1 = ip1 + step;
+            if (ip1 > ilimit) goto _cleanup;
+
+            curr = (U32)(ip0 - base);
+            hash0 = hash1;
+        }   /* end inner search loop */
+
+        /* match found */
+        assert(mLength);
+        ip0 += mLength;
+        anchor = ip0;
+
+        if (ip0 <= ilimit) {
+            /* Fill Table */
+            assert(base+curr+2 > istart);  /* check base overflow */
+            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */
+            hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+
+            /* check immediate repcode */
+            while (ip0 <= ilimit) {
+                U32 const current2 = (U32)(ip0-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
+                        dictBase - dictIndexDelta + repIndex2 :
+                        base + repIndex2;
+                if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
+                   && (MEM_read32(repMatch2) == MEM_read32(ip0))) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = current2;
+                    ip0 += repLength2;
+                    anchor = ip0;
+                    continue;
+                }
+                break;
+            }
+        }
+
+        /* Prepare for next iteration */
+        assert(ip0 == anchor);
+        ip1 = ip0 + stepSize;
+    }
+
+_cleanup:
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+
+ZSTD_GEN_FAST_FN(dictMatchState, 4, 0)
+ZSTD_GEN_FAST_FN(dictMatchState, 5, 0)
+ZSTD_GEN_FAST_FN(dictMatchState, 6, 0)
+ZSTD_GEN_FAST_FN(dictMatchState, 7, 0)
+
+size_t ZSTD_compressBlock_fast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    U32 const mls = ms->cParams.minMatch;
+    assert(ms->dictMatchState != NULL);
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize);
+    }
+}
+
+
+static size_t ZSTD_compressBlock_fast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* anchor = istart;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
+    const U32   dictStartIndex = lowLimit;
+    const BYTE* const dictStart = dictBase + dictStartIndex;
+    const U32   dictLimit = ms->window.dictLimit;
+    const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const dictEnd = dictBase + prefixStartIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    const BYTE* ip0 = istart;
+    const BYTE* ip1;
+    const BYTE* ip2;
+    const BYTE* ip3;
+    U32 current0;
+
+
+    size_t hash0; /* hash for ip0 */
+    size_t hash1; /* hash for ip1 */
+    U32 idx; /* match idx for ip0 */
+    const BYTE* idxBase; /* base pointer for idx */
+
+    U32 offcode;
+    const BYTE* match0;
+    size_t mLength;
+    const BYTE* matchEnd = 0; /* initialize to avoid warning, assert != 0 later */
+
+    size_t step;
+    const BYTE* nextStep;
+    const size_t kStepIncr = (1 << (kSearchStrength - 1));
+
+    (void)hasStep; /* not currently specialized on whether it's accelerated */
+
+    DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);
+
+    /* switch to "regular" variant if extDict is invalidated due to maxDistance */
+    if (prefixStartIndex == dictStartIndex)
+        return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);
+
+    {   U32 const curr = (U32)(ip0 - base);
+        U32 const maxRep = curr - dictStartIndex;
+        if (offset_2 >= maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 >= maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+
+    /* start each op */
+_start: /* Requires: ip0 */
+
+    step = stepSize;
+    nextStep = ip0 + kStepIncr;
+
+    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
+    ip1 = ip0 + 1;
+    ip2 = ip0 + step;
+    ip3 = ip2 + 1;
+
+    if (ip3 >= ilimit) {
+        goto _cleanup;
+    }
+
+    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+    hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+
+    idx = hashTable[hash0];
+    idxBase = idx < prefixStartIndex ? dictBase : base;
+
+    do {
+        {   /* load repcode match for ip[2] */
+            U32 const current2 = (U32)(ip2 - base);
+            U32 const repIndex = current2 - offset_1;
+            const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+            U32 rval;
+            if ( ((U32)(prefixStartIndex - repIndex) >= 4) /* intentional underflow */
+                 & (offset_1 > 0) ) {
+                rval = MEM_read32(repBase + repIndex);
+            } else {
+                rval = MEM_read32(ip2) ^ 1; /* guaranteed to not match. */
+            }
+
+            /* write back hash table entry */
+            current0 = (U32)(ip0 - base);
+            hashTable[hash0] = current0;
+
+            /* check repcode at ip[2] */
+            if (MEM_read32(ip2) == rval) {
+                ip0 = ip2;
+                match0 = repBase + repIndex;
+                matchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+                assert((match0 != prefixStart) & (match0 != dictStart));
+                mLength = ip0[-1] == match0[-1];
+                ip0 -= mLength;
+                match0 -= mLength;
+                offcode = REPCODE1_TO_OFFBASE;
+                mLength += 4;
+                goto _match;
+        }   }
+
+        {   /* load match for ip[0] */
+            U32 const mval = idx >= dictStartIndex ?
+                    MEM_read32(idxBase + idx) :
+                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */
+
+            /* check match at ip[0] */
+            if (MEM_read32(ip0) == mval) {
+                /* found a match! */
+                goto _offset;
+        }   }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+        idxBase = idx < prefixStartIndex ? dictBase : base;
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip3;
+
+        /* write back hash table entry */
+        current0 = (U32)(ip0 - base);
+        hashTable[hash0] = current0;
+
+        {   /* load match for ip[0] */
+            U32 const mval = idx >= dictStartIndex ?
+                    MEM_read32(idxBase + idx) :
+                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */
+
+            /* check match at ip[0] */
+            if (MEM_read32(ip0) == mval) {
+                /* found a match! */
+                goto _offset;
+        }   }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+        idxBase = idx < prefixStartIndex ? dictBase : base;
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip0 + step;
+        ip3 = ip1 + step;
+
+        /* calculate step */
+        if (ip2 >= nextStep) {
+            step++;
+            PREFETCH_L1(ip1 + 64);
+            PREFETCH_L1(ip1 + 128);
+            nextStep += kStepIncr;
+        }
+    } while (ip3 < ilimit);
+
+_cleanup:
+    /* Note that there are probably still a couple positions we could search.
+     * However, it seems to be a meaningful performance hit to try to search
+     * them. So let's not. */
+
+    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved1;
+    rep[1] = offset_2 ? offset_2 : offsetSaved2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+
+_offset: /* Requires: ip0, idx, idxBase */
+
+    /* Compute the offset code. */
+    {   U32 const offset = current0 - idx;
+        const BYTE* const lowMatchPtr = idx < prefixStartIndex ? dictStart : prefixStart;
+        matchEnd = idx < prefixStartIndex ? dictEnd : iend;
+        match0 = idxBase + idx;
+        offset_2 = offset_1;
+        offset_1 = offset;
+        offcode = OFFSET_TO_OFFBASE(offset);
+        mLength = 4;
+
+        /* Count the backwards match length. */
+        while (((ip0>anchor) & (match0>lowMatchPtr)) && (ip0[-1] == match0[-1])) {
+            ip0--;
+            match0--;
+            mLength++;
+    }   }
+
+_match: /* Requires: ip0, match0, offcode, matchEnd */
+
+    /* Count the forward length. */
+    assert(matchEnd != 0);
+    mLength += ZSTD_count_2segments(ip0 + mLength, match0 + mLength, iend, matchEnd, prefixStart);
+
+    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
+
+    ip0 += mLength;
+    anchor = ip0;
+
+    /* write next hash table entry */
+    if (ip1 < ip0) {
+        hashTable[hash1] = (U32)(ip1 - base);
+    }
+
+    /* Fill table and check for immediate repcode. */
+    if (ip0 <= ilimit) {
+        /* Fill Table */
+        assert(base+current0+2 > istart);  /* check base overflow */
+        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
+        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+
+        while (ip0 <= ilimit) {
+            U32 const repIndex2 = (U32)(ip0-base) - offset_2;
+            const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+            if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 > 0))  /* intentional underflow */
+                 && (MEM_read32(repMatch2) == MEM_read32(ip0)) ) {
+                const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
+                ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
+                ip0 += repLength2;
+                anchor = ip0;
+                continue;
+            }
+            break;
+    }   }
+
+    goto _start;
+}
+
+ZSTD_GEN_FAST_FN(extDict, 4, 0)
+ZSTD_GEN_FAST_FN(extDict, 5, 0)
+ZSTD_GEN_FAST_FN(extDict, 6, 0)
+ZSTD_GEN_FAST_FN(extDict, 7, 0)
+
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    U32 const mls = ms->cParams.minMatch;
+    assert(ms->dictMatchState == NULL);
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize);
+    }
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_fast.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_fast.h
new file mode 100644
index 0000000..9e4236b
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_fast.h
@@ -0,0 +1,38 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_FAST_H
+#define ZSTD_FAST_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "../common/mem.h"      /* U32 */
+#include "zstd_compress_internal.h"
+
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        void const* end, ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp);
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_fast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_FAST_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_lazy.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_lazy.c
new file mode 100644
index 0000000..5ba88e8
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_lazy.c
@@ -0,0 +1,2157 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_lazy.h"
+#include "../common/bits.h" /* ZSTD_countTrailingZeros64 */
+
+#define kLazySkippingStep 8
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+
+static void
+ZSTD_updateDUBT(ZSTD_matchState_t* ms,
+                const BYTE* ip, const BYTE* iend,
+                U32 mls)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hashLog = cParams->hashLog;
+
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
+
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    if (idx != target)
+        DEBUGLOG(7, "ZSTD_updateDUBT, from %u to %u (dictLimit:%u)",
+                    idx, target, ms->window.dictLimit);
+    assert(ip + 8 <= iend);   /* condition for ZSTD_hashPtr */
+    (void)iend;
+
+    assert(idx >= ms->window.dictLimit);   /* condition for valid base+idx */
+    for ( ; idx < target ; idx++) {
+        size_t const h  = ZSTD_hashPtr(base + idx, hashLog, mls);   /* assumption : ip + 8 <= iend */
+        U32    const matchIndex = hashTable[h];
+
+        U32*   const nextCandidatePtr = bt + 2*(idx&btMask);
+        U32*   const sortMarkPtr  = nextCandidatePtr + 1;
+
+        DEBUGLOG(8, "ZSTD_updateDUBT: insert %u", idx);
+        hashTable[h] = idx;   /* Update Hash Table */
+        *nextCandidatePtr = matchIndex;   /* update BT like a chain */
+        *sortMarkPtr = ZSTD_DUBT_UNSORTED_MARK;
+    }
+    ms->nextToUpdate = target;
+}
+
+
+/** ZSTD_insertDUBT1() :
+ *  sort one already inserted but unsorted position
+ *  assumption : curr >= btlow == (curr - btmask)
+ *  doesn't fail */
+static void
+ZSTD_insertDUBT1(const ZSTD_matchState_t* ms,
+                 U32 curr, const BYTE* inputEnd,
+                 U32 nbCompares, U32 btLow,
+                 const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const ip = (curr>=dictLimit) ? base + curr : dictBase + curr;
+    const BYTE* const iend = (curr>=dictLimit) ? inputEnd : dictBase + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    U32* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 matchIndex = *smallerPtr;   /* this candidate is unsorted : next sorted candidate is reached through *smallerPtr, while *largerPtr contains previous unsorted candidate (which is already saved and can be overwritten) */
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowValid = ms->window.lowLimit;
+    U32 const maxDistance = 1U << cParams->windowLog;
+    U32 const windowLow = (curr - windowValid > maxDistance) ? curr - maxDistance : windowValid;
+
+
+    DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
+                curr, dictLimit, windowLow);
+    assert(curr >= btLow);
+    assert(ip < iend);   /* condition for ZSTD_count */
+
+    for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < curr);
+        /* note : all candidates are now supposed sorted,
+         * but it's still possible to have nextPtr[1] == ZSTD_DUBT_UNSORTED_MARK
+         * when a real index has the same value as ZSTD_DUBT_UNSORTED_MARK */
+
+        if ( (dictMode != ZSTD_extDict)
+          || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
+          || (curr < dictLimit) /* both in extDict */) {
+            const BYTE* const mBase = ( (dictMode != ZSTD_extDict)
+                                     || (matchIndex+matchLength >= dictLimit)) ?
+                                        base : dictBase;
+            assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
+                 || (curr < dictLimit) );
+            match = mBase + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* preparation for next read of match[matchLength] */
+        }
+
+        DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ",
+                    curr, matchIndex, (U32)matchLength);
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is smaller : next => %u",
+                        matchIndex, btLow, nextPtr[1]);
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is larger => %u",
+                        matchIndex, btLow, nextPtr[0]);
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+}
+
+
+static size_t
+ZSTD_DUBT_findBetterDictMatch (
+        const ZSTD_matchState_t* ms,
+        const BYTE* const ip, const BYTE* const iend,
+        size_t* offsetPtr,
+        size_t bestLength,
+        U32 nbCompares,
+        U32 const mls,
+        const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_matchState_t * const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dmsCParams = &dms->cParams;
+    const U32 * const dictHashTable = dms->hashTable;
+    U32         const hashLog = dmsCParams->hashLog;
+    size_t      const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32               dictMatchIndex = dictHashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+    U32         const curr = (U32)(ip-base);
+    const BYTE* const dictBase = dms->window.base;
+    const BYTE* const dictEnd = dms->window.nextSrc;
+    U32         const dictHighLimit = (U32)(dms->window.nextSrc - dms->window.base);
+    U32         const dictLowLimit = dms->window.lowLimit;
+    U32         const dictIndexDelta = ms->window.lowLimit - dictHighLimit;
+
+    U32*        const dictBt = dms->chainTable;
+    U32         const btLog  = dmsCParams->chainLog - 1;
+    U32         const btMask = (1 << btLog) - 1;
+    U32         const btLow = (btMask >= dictHighLimit - dictLowLimit) ? dictLowLimit : dictHighLimit - btMask;
+
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+
+    (void)dictMode;
+    assert(dictMode == ZSTD_dictMatchState);
+
+    for (; nbCompares && (dictMatchIndex > dictLowLimit); --nbCompares) {
+        U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        const BYTE* match = dictBase + dictMatchIndex;
+        matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+        if (dictMatchIndex+matchLength >= dictHighLimit)
+            match = base + dictMatchIndex + dictIndexDelta;   /* to prepare for next usage of match[matchLength] */
+
+        if (matchLength > bestLength) {
+            U32 matchIndex = dictMatchIndex + dictIndexDelta;
+            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
+                DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
+                    curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, OFFSET_TO_OFFBASE(curr - matchIndex), dictMatchIndex, matchIndex);
+                bestLength = matchLength, *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+            }
+            if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
+                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+            }
+        }
+
+        if (match[matchLength] < ip[matchLength]) {
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+        } else {
+            /* match is larger than current */
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
+            commonLengthLarger = matchLength;
+            dictMatchIndex = nextPtr[0];
+        }
+    }
+
+    if (bestLength >= MINMATCH) {
+        U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offsetPtr); (void)mIndex;
+        DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                    curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
+    }
+    return bestLength;
+
+}
+
+
+static size_t
+ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iend,
+                        size_t* offBasePtr,
+                        U32 const mls,
+                        const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32          matchIndex  = hashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    U32    const curr = (U32)(ip-base);
+    U32    const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
+
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32    const btLow = (btMask >= curr) ? 0 : curr - btMask;
+    U32    const unsortLimit = MAX(btLow, windowLow);
+
+    U32*         nextCandidate = bt + 2*(matchIndex&btMask);
+    U32*         unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+    U32          nbCompares = 1U << cParams->searchLog;
+    U32          nbCandidates = nbCompares;
+    U32          previousCandidate = 0;
+
+    DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", curr);
+    assert(ip <= iend-8);   /* required for h calculation */
+    assert(dictMode != ZSTD_dedicatedDictSearch);
+
+    /* reach end of unsorted candidates list */
+    while ( (matchIndex > unsortLimit)
+         && (*unsortedMark == ZSTD_DUBT_UNSORTED_MARK)
+         && (nbCandidates > 1) ) {
+        DEBUGLOG(8, "ZSTD_DUBT_findBestMatch: candidate %u is unsorted",
+                    matchIndex);
+        *unsortedMark = previousCandidate;  /* the unsortedMark becomes a reversed chain, to move up back to original position */
+        previousCandidate = matchIndex;
+        matchIndex = *nextCandidate;
+        nextCandidate = bt + 2*(matchIndex&btMask);
+        unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+        nbCandidates --;
+    }
+
+    /* nullify last candidate if it's still unsorted
+     * simplification, detrimental to compression ratio, beneficial for speed */
+    if ( (matchIndex > unsortLimit)
+      && (*unsortedMark==ZSTD_DUBT_UNSORTED_MARK) ) {
+        DEBUGLOG(7, "ZSTD_DUBT_findBestMatch: nullify last unsorted candidate %u",
+                    matchIndex);
+        *nextCandidate = *unsortedMark = 0;
+    }
+
+    /* batch sort stacked candidates */
+    matchIndex = previousCandidate;
+    while (matchIndex) {  /* will end on matchIndex == 0 */
+        U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1;
+        U32 const nextCandidateIdx = *nextCandidateIdxPtr;
+        ZSTD_insertDUBT1(ms, matchIndex, iend,
+                         nbCandidates, unsortLimit, dictMode);
+        matchIndex = nextCandidateIdx;
+        nbCandidates++;
+    }
+
+    /* find longest match */
+    {   size_t commonLengthSmaller = 0, commonLengthLarger = 0;
+        const BYTE* const dictBase = ms->window.dictBase;
+        const U32 dictLimit = ms->window.dictLimit;
+        const BYTE* const dictEnd = dictBase + dictLimit;
+        const BYTE* const prefixStart = base + dictLimit;
+        U32* smallerPtr = bt + 2*(curr&btMask);
+        U32* largerPtr  = bt + 2*(curr&btMask) + 1;
+        U32 matchEndIdx = curr + 8 + 1;
+        U32 dummy32;   /* to be nullified at the end */
+        size_t bestLength = 0;
+
+        matchIndex  = hashTable[h];
+        hashTable[h] = curr;   /* Update Hash Table */
+
+        for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
+            U32* const nextPtr = bt + 2*(matchIndex & btMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match;
+
+            if ((dictMode != ZSTD_extDict) || (matchIndex+matchLength >= dictLimit)) {
+                match = base + matchIndex;
+                matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+            } else {
+                match = dictBase + matchIndex;
+                matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+                if (matchIndex+matchLength >= dictLimit)
+                    match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+            }
+
+            if (matchLength > bestLength) {
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr - matchIndex + 1) - ZSTD_highbit32((U32)*offBasePtr)) )
+                    bestLength = matchLength, *offBasePtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+                if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+                    if (dictMode == ZSTD_dictMatchState) {
+                        nbCompares = 0; /* in addition to avoiding checking any
+                                         * further in this loop, make sure we
+                                         * skip checking in the dictionary. */
+                    }
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+                }
+            }
+
+            if (match[matchLength] < ip[matchLength]) {
+                /* match is smaller than current */
+                *smallerPtr = matchIndex;             /* update smaller idx */
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+                matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                *largerPtr = matchIndex;
+                commonLengthLarger = matchLength;
+                if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                largerPtr = nextPtr;
+                matchIndex = nextPtr[0];
+        }   }
+
+        *smallerPtr = *largerPtr = 0;
+
+        assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+        if (dictMode == ZSTD_dictMatchState && nbCompares) {
+            bestLength = ZSTD_DUBT_findBetterDictMatch(
+                    ms, ip, iend,
+                    offBasePtr, bestLength, nbCompares,
+                    mls, dictMode);
+        }
+
+        assert(matchEndIdx > curr+8); /* ensure nextToUpdate is increased */
+        ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
+        if (bestLength >= MINMATCH) {
+            U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offBasePtr); (void)mIndex;
+            DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                        curr, (U32)bestLength, (U32)*offBasePtr, mIndex);
+        }
+        return bestLength;
+    }
+}
+
+
+/** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_BtFindBestMatch( ZSTD_matchState_t* ms,
+                const BYTE* const ip, const BYTE* const iLimit,
+                      size_t* offBasePtr,
+                const U32 mls /* template */,
+                const ZSTD_dictMode_e dictMode)
+{
+    DEBUGLOG(7, "ZSTD_BtFindBestMatch");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateDUBT(ms, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offBasePtr, mls, dictMode);
+}
+
+/***********************************
+* Dedicated dict search
+***********************************/
+
+void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const BYTE* const ip)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32* const hashTable = ms->hashTable;
+    U32* const chainTable = ms->chainTable;
+    U32 const chainSize = 1 << ms->cParams.chainLog;
+    U32 idx = ms->nextToUpdate;
+    U32 const minChain = chainSize < target - idx ? target - chainSize : idx;
+    U32 const bucketSize = 1 << ZSTD_LAZY_DDSS_BUCKET_LOG;
+    U32 const cacheSize = bucketSize - 1;
+    U32 const chainAttempts = (1 << ms->cParams.searchLog) - cacheSize;
+    U32 const chainLimit = chainAttempts > 255 ? 255 : chainAttempts;
+
+    /* We know the hashtable is oversized by a factor of `bucketSize`.
+     * We are going to temporarily pretend `bucketSize == 1`, keeping only a
+     * single entry. We will use the rest of the space to construct a temporary
+     * chaintable.
+     */
+    U32 const hashLog = ms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
+    U32* const tmpHashTable = hashTable;
+    U32* const tmpChainTable = hashTable + ((size_t)1 << hashLog);
+    U32 const tmpChainSize = (U32)((1 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1) << hashLog;
+    U32 const tmpMinChain = tmpChainSize < target ? target - tmpChainSize : idx;
+    U32 hashIdx;
+
+    assert(ms->cParams.chainLog <= 24);
+    assert(ms->cParams.hashLog > ms->cParams.chainLog);
+    assert(idx != 0);
+    assert(tmpMinChain <= minChain);
+
+    /* fill conventional hash table and conventional chain table */
+    for ( ; idx < target; idx++) {
+        U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch);
+        if (idx >= tmpMinChain) {
+            tmpChainTable[idx - tmpMinChain] = hashTable[h];
+        }
+        tmpHashTable[h] = idx;
+    }
+
+    /* sort chains into ddss chain table */
+    {
+        U32 chainPos = 0;
+        for (hashIdx = 0; hashIdx < (1U << hashLog); hashIdx++) {
+            U32 count;
+            U32 countBeyondMinChain = 0;
+            U32 i = tmpHashTable[hashIdx];
+            for (count = 0; i >= tmpMinChain && count < cacheSize; count++) {
+                /* skip through the chain to the first position that won't be
+                 * in the hash cache bucket */
+                if (i < minChain) {
+                    countBeyondMinChain++;
+                }
+                i = tmpChainTable[i - tmpMinChain];
+            }
+            if (count == cacheSize) {
+                for (count = 0; count < chainLimit;) {
+                    if (i < minChain) {
+                        if (!i || ++countBeyondMinChain > cacheSize) {
+                            /* only allow pulling `cacheSize` number of entries
+                             * into the cache or chainTable beyond `minChain`,
+                             * to replace the entries pulled out of the
+                             * chainTable into the cache. This lets us reach
+                             * back further without increasing the total number
+                             * of entries in the chainTable, guaranteeing the
+                             * DDSS chain table will fit into the space
+                             * allocated for the regular one. */
+                            break;
+                        }
+                    }
+                    chainTable[chainPos++] = i;
+                    count++;
+                    if (i < tmpMinChain) {
+                        break;
+                    }
+                    i = tmpChainTable[i - tmpMinChain];
+                }
+            } else {
+                count = 0;
+            }
+            if (count) {
+                tmpHashTable[hashIdx] = ((chainPos - count) << 8) + count;
+            } else {
+                tmpHashTable[hashIdx] = 0;
+            }
+        }
+        assert(chainPos <= chainSize); /* I believe this is guaranteed... */
+    }
+
+    /* move chain pointers into the last entry of each hash bucket */
+    for (hashIdx = (1 << hashLog); hashIdx; ) {
+        U32 const bucketIdx = --hashIdx << ZSTD_LAZY_DDSS_BUCKET_LOG;
+        U32 const chainPackedPointer = tmpHashTable[hashIdx];
+        U32 i;
+        for (i = 0; i < cacheSize; i++) {
+            hashTable[bucketIdx + i] = 0;
+        }
+        hashTable[bucketIdx + bucketSize - 1] = chainPackedPointer;
+    }
+
+    /* fill the buckets of the hash table */
+    for (idx = ms->nextToUpdate; idx < target; idx++) {
+        U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch)
+                   << ZSTD_LAZY_DDSS_BUCKET_LOG;
+        U32 i;
+        /* Shift hash cache down 1. */
+        for (i = cacheSize - 1; i; i--)
+            hashTable[h + i] = hashTable[h + i - 1];
+        hashTable[h] = idx;
+    }
+
+    ms->nextToUpdate = target;
+}
+
+/* Returns the longest match length found in the dedicated dict search structure.
+ * If none are longer than the argument ml, then ml will be returned.
+ */
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nbAttempts,
+                                            const ZSTD_matchState_t* const dms,
+                                            const BYTE* const ip, const BYTE* const iLimit,
+                                            const BYTE* const prefixStart, const U32 curr,
+                                            const U32 dictLimit, const size_t ddsIdx) {
+    const U32 ddsLowestIndex  = dms->window.dictLimit;
+    const BYTE* const ddsBase = dms->window.base;
+    const BYTE* const ddsEnd  = dms->window.nextSrc;
+    const U32 ddsSize         = (U32)(ddsEnd - ddsBase);
+    const U32 ddsIndexDelta   = dictLimit - ddsSize;
+    const U32 bucketSize      = (1 << ZSTD_LAZY_DDSS_BUCKET_LOG);
+    const U32 bucketLimit     = nbAttempts < bucketSize - 1 ? nbAttempts : bucketSize - 1;
+    U32 ddsAttempt;
+    U32 matchIndex;
+
+    for (ddsAttempt = 0; ddsAttempt < bucketSize - 1; ddsAttempt++) {
+        PREFETCH_L1(ddsBase + dms->hashTable[ddsIdx + ddsAttempt]);
+    }
+
+    {
+        U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
+        U32 const chainIndex = chainPackedPointer >> 8;
+
+        PREFETCH_L1(&dms->chainTable[chainIndex]);
+    }
+
+    for (ddsAttempt = 0; ddsAttempt < bucketLimit; ddsAttempt++) {
+        size_t currentMl=0;
+        const BYTE* match;
+        matchIndex = dms->hashTable[ddsIdx + ddsAttempt];
+        match = ddsBase + matchIndex;
+
+        if (!matchIndex) {
+            return ml;
+        }
+
+        /* guaranteed by table construction */
+        (void)ddsLowestIndex;
+        assert(matchIndex >= ddsLowestIndex);
+        assert(match+4 <= ddsEnd);
+        if (MEM_read32(match) == MEM_read32(ip)) {
+            /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+            currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
+        }
+
+        /* save best solution */
+        if (currentMl > ml) {
+            ml = currentMl;
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
+            if (ip+currentMl == iLimit) {
+                /* best possible, avoids read overflow on next attempt */
+                return ml;
+            }
+        }
+    }
+
+    {
+        U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
+        U32 chainIndex = chainPackedPointer >> 8;
+        U32 const chainLength = chainPackedPointer & 0xFF;
+        U32 const chainAttempts = nbAttempts - ddsAttempt;
+        U32 const chainLimit = chainAttempts > chainLength ? chainLength : chainAttempts;
+        U32 chainAttempt;
+
+        for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++) {
+            PREFETCH_L1(ddsBase + dms->chainTable[chainIndex + chainAttempt]);
+        }
+
+        for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++, chainIndex++) {
+            size_t currentMl=0;
+            const BYTE* match;
+            matchIndex = dms->chainTable[chainIndex];
+            match = ddsBase + matchIndex;
+
+            /* guaranteed by table construction */
+            assert(matchIndex >= ddsLowestIndex);
+            assert(match+4 <= ddsEnd);
+            if (MEM_read32(match) == MEM_read32(ip)) {
+                /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
+            }
+
+            /* save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+        }
+    }
+    return ml;
+}
+
+
+/* *********************************
+*  Hash Chain
+***********************************/
+#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & (mask)]
+
+/* Update chains up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+FORCE_INLINE_TEMPLATE U32 ZSTD_insertAndFindFirstIndex_internal(
+                        ZSTD_matchState_t* ms,
+                        const ZSTD_compressionParameters* const cParams,
+                        const BYTE* ip, U32 const mls, U32 const lazySkipping)
+{
+    U32* const hashTable  = ms->hashTable;
+    const U32 hashLog = cParams->hashLog;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainMask = (1 << cParams->chainLog) - 1;
+    const BYTE* const base = ms->window.base;
+    const U32 target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    while(idx < target) { /* catch up */
+        size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
+        NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
+        hashTable[h] = idx;
+        idx++;
+        /* Stop inserting every position when in the lazy skipping mode. */
+        if (lazySkipping)
+            break;
+    }
+
+    ms->nextToUpdate = target;
+    return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
+}
+
+U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip) {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch, /* lazySkipping*/ 0);
+}
+
+/* inlining is important to hardwire a hot branch (template emulation) */
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_HcFindBestMatch(
+                        ZSTD_matchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls, const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainSize = (1 << cParams->chainLog);
+    const U32 chainMask = chainSize-1;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32 curr = (U32)(ip-base);
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 lowestValid = ms->window.lowLimit;
+    const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
+    const U32 isDictionary = (ms->loadedDictEnd != 0);
+    const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
+    const U32 minChain = curr > chainSize ? curr - chainSize : 0;
+    U32 nbAttempts = 1U << cParams->searchLog;
+    size_t ml=4-1;
+
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const U32 ddsHashLog = dictMode == ZSTD_dedicatedDictSearch
+                         ? dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
+    const size_t ddsIdx = dictMode == ZSTD_dedicatedDictSearch
+                        ? ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
+
+    U32 matchIndex;
+
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        const U32* entry = &dms->hashTable[ddsIdx];
+        PREFETCH_L1(entry);
+    }
+
+    /* HC4 match finder */
+    matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls, ms->lazySkipping);
+
+    for ( ; (matchIndex>=lowLimit) & (nbAttempts>0) ; nbAttempts--) {
+        size_t currentMl=0;
+        if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
+            const BYTE* const match = base + matchIndex;
+            assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
+            /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+            if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
+                currentMl = ZSTD_count(ip, match, iLimit);
+        } else {
+            const BYTE* const match = dictBase + matchIndex;
+            assert(match+4 <= dictEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
+        }
+
+        /* save best solution */
+        if (currentMl > ml) {
+            ml = currentMl;
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+            if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+        }
+
+        if (matchIndex <= minChain) break;
+        matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
+    }
+
+    assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts, dms,
+                                                  ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
+    } else if (dictMode == ZSTD_dictMatchState) {
+        const U32* const dmsChainTable = dms->chainTable;
+        const U32 dmsChainSize         = (1 << dms->cParams.chainLog);
+        const U32 dmsChainMask         = dmsChainSize - 1;
+        const U32 dmsLowestIndex       = dms->window.dictLimit;
+        const BYTE* const dmsBase      = dms->window.base;
+        const BYTE* const dmsEnd       = dms->window.nextSrc;
+        const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
+        const U32 dmsIndexDelta        = dictLimit - dmsSize;
+        const U32 dmsMinChain = dmsSize > dmsChainSize ? dmsSize - dmsChainSize : 0;
+
+        matchIndex = dms->hashTable[ZSTD_hashPtr(ip, dms->cParams.hashLog, mls)];
+
+        for ( ; (matchIndex>=dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) {
+            size_t currentMl=0;
+            const BYTE* const match = dmsBase + matchIndex;
+            assert(match+4 <= dmsEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
+
+            /* save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                assert(curr > matchIndex + dmsIndexDelta);
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+
+            if (matchIndex <= dmsMinChain) break;
+
+            matchIndex = dmsChainTable[matchIndex & dmsChainMask];
+        }
+    }
+
+    return ml;
+}
+
+/* *********************************
+* (SIMD) Row-based matchfinder
+***********************************/
+/* Constants for row-based hash */
+#define ZSTD_ROW_HASH_TAG_MASK ((1u << ZSTD_ROW_HASH_TAG_BITS) - 1)
+#define ZSTD_ROW_HASH_MAX_ENTRIES 64    /* absolute maximum number of entries per row, for all configurations */
+
+#define ZSTD_ROW_HASH_CACHE_MASK (ZSTD_ROW_HASH_CACHE_SIZE - 1)
+
+typedef U64 ZSTD_VecMask;   /* Clarifies when we are interacting with a U64 representing a mask of matches */
+
+/* ZSTD_VecMask_next():
+ * Starting from the LSB, returns the idx of the next non-zero bit.
+ * Basically counting the nb of trailing zeroes.
+ */
+MEM_STATIC U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
+    return ZSTD_countTrailingZeros64(val);
+}
+
+/* ZSTD_row_nextIndex():
+ * Returns the next index to insert at within a tagTable row, and updates the "head"
+ * value to reflect the update. Essentially cycles backwards from [1, {entries per row})
+ */
+FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextIndex(BYTE* const tagRow, U32 const rowMask) {
+    U32 next = (*tagRow-1) & rowMask;
+    next += (next == 0) ? rowMask : 0; /* skip first position */
+    *tagRow = (BYTE)next;
+    return next;
+}
+
+/* ZSTD_isAligned():
+ * Checks that a pointer is aligned to "align" bytes which must be a power of 2.
+ */
+MEM_STATIC int ZSTD_isAligned(void const* ptr, size_t align) {
+    assert((align & (align - 1)) == 0);
+    return (((size_t)ptr) & (align - 1)) == 0;
+}
+
+/* ZSTD_row_prefetch():
+ * Performs prefetching for the hashTable and tagTable at a given row.
+ */
+FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, BYTE const* tagTable, U32 const relRow, U32 const rowLog) {
+    PREFETCH_L1(hashTable + relRow);
+    if (rowLog >= 5) {
+        PREFETCH_L1(hashTable + relRow + 16);
+        /* Note: prefetching more of the hash table does not appear to be beneficial for 128-entry rows */
+    }
+    PREFETCH_L1(tagTable + relRow);
+    if (rowLog == 6) {
+        PREFETCH_L1(tagTable + relRow + 32);
+    }
+    assert(rowLog == 4 || rowLog == 5 || rowLog == 6);
+    assert(ZSTD_isAligned(hashTable + relRow, 64));                 /* prefetched hash row always 64-byte aligned */
+    assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on correct multiple of bytes (32,64,128) */
+}
+
+/* ZSTD_row_fillHashCache():
+ * Fill up the hash cache starting at idx, prefetching up to ZSTD_ROW_HASH_CACHE_SIZE entries,
+ * but not beyond iLimit.
+ */
+FORCE_INLINE_TEMPLATE void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const BYTE* base,
+                                   U32 const rowLog, U32 const mls,
+                                   U32 idx, const BYTE* const iLimit)
+{
+    U32 const* const hashTable = ms->hashTable;
+    BYTE const* const tagTable = ms->tagTable;
+    U32 const hashLog = ms->rowHashLog;
+    U32 const maxElemsToPrefetch = (base + idx) > iLimit ? 0 : (U32)(iLimit - (base + idx) + 1);
+    U32 const lim = idx + MIN(ZSTD_ROW_HASH_CACHE_SIZE, maxElemsToPrefetch);
+
+    for (; idx < lim; ++idx) {
+        U32 const hash = (U32)ZSTD_hashPtrSalted(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
+        U32 const row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
+        ms->hashCache[idx & ZSTD_ROW_HASH_CACHE_MASK] = hash;
+    }
+
+    DEBUGLOG(6, "ZSTD_row_fillHashCache(): [%u %u %u %u %u %u %u %u]", ms->hashCache[0], ms->hashCache[1],
+                                                     ms->hashCache[2], ms->hashCache[3], ms->hashCache[4],
+                                                     ms->hashCache[5], ms->hashCache[6], ms->hashCache[7]);
+}
+
+/* ZSTD_row_nextCachedHash():
+ * Returns the hash of base + idx, and replaces the hash in the hash cache with the byte at
+ * base + idx + ZSTD_ROW_HASH_CACHE_SIZE. Also prefetches the appropriate rows from hashTable and tagTable.
+ */
+FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTable,
+                                                  BYTE const* tagTable, BYTE const* base,
+                                                  U32 idx, U32 const hashLog,
+                                                  U32 const rowLog, U32 const mls,
+                                                  U64 const hashSalt)
+{
+    U32 const newHash = (U32)ZSTD_hashPtrSalted(base+idx+ZSTD_ROW_HASH_CACHE_SIZE, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
+    U32 const row = (newHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+    ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
+    {   U32 const hash = cache[idx & ZSTD_ROW_HASH_CACHE_MASK];
+        cache[idx & ZSTD_ROW_HASH_CACHE_MASK] = newHash;
+        return hash;
+    }
+}
+
+/* ZSTD_row_update_internalImpl():
+ * Updates the hash table with positions starting from updateStartIdx until updateEndIdx.
+ */
+FORCE_INLINE_TEMPLATE void ZSTD_row_update_internalImpl(ZSTD_matchState_t* ms,
+                                                        U32 updateStartIdx, U32 const updateEndIdx,
+                                                        U32 const mls, U32 const rowLog,
+                                                        U32 const rowMask, U32 const useCache)
+{
+    U32* const hashTable = ms->hashTable;
+    BYTE* const tagTable = ms->tagTable;
+    U32 const hashLog = ms->rowHashLog;
+    const BYTE* const base = ms->window.base;
+
+    DEBUGLOG(6, "ZSTD_row_update_internalImpl(): updateStartIdx=%u, updateEndIdx=%u", updateStartIdx, updateEndIdx);
+    for (; updateStartIdx < updateEndIdx; ++updateStartIdx) {
+        U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls, ms->hashSalt)
+                                  : (U32)ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
+        U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        U32* const row = hashTable + relRow;
+        BYTE* tagRow = tagTable + relRow;
+        U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
+
+        assert(hash == ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt));
+        tagRow[pos] = hash & ZSTD_ROW_HASH_TAG_MASK;
+        row[pos] = updateStartIdx;
+    }
+}
+
+/* ZSTD_row_update_internal():
+ * Inserts the byte at ip into the appropriate position in the hash table, and updates ms->nextToUpdate.
+ * Skips sections of long matches as is necessary.
+ */
+FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const BYTE* ip,
+                                                    U32 const mls, U32 const rowLog,
+                                                    U32 const rowMask, U32 const useCache)
+{
+    U32 idx = ms->nextToUpdate;
+    const BYTE* const base = ms->window.base;
+    const U32 target = (U32)(ip - base);
+    const U32 kSkipThreshold = 384;
+    const U32 kMaxMatchStartPositionsToUpdate = 96;
+    const U32 kMaxMatchEndPositionsToUpdate = 32;
+
+    if (useCache) {
+        /* Only skip positions when using hash cache, i.e.
+         * if we are loading a dict, don't skip anything.
+         * If we decide to skip, then we only update a set number
+         * of positions at the beginning and end of the match.
+         */
+        if (UNLIKELY(target - idx > kSkipThreshold)) {
+            U32 const bound = idx + kMaxMatchStartPositionsToUpdate;
+            ZSTD_row_update_internalImpl(ms, idx, bound, mls, rowLog, rowMask, useCache);
+            idx = target - kMaxMatchEndPositionsToUpdate;
+            ZSTD_row_fillHashCache(ms, base, rowLog, mls, idx, ip+1);
+        }
+    }
+    assert(target >= idx);
+    ZSTD_row_update_internalImpl(ms, idx, target, mls, rowLog, rowMask, useCache);
+    ms->nextToUpdate = target;
+}
+
+/* ZSTD_row_update():
+ * External wrapper for ZSTD_row_update_internal(). Used for filling the hashtable during dictionary
+ * processing.
+ */
+void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip) {
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
+    const U32 rowMask = (1u << rowLog) - 1;
+    const U32 mls = MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */);
+
+    DEBUGLOG(5, "ZSTD_row_update(), rowLog=%u", rowLog);
+    ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* don't use cache */);
+}
+
+/* Returns the mask width of bits group of which will be set to 1. Given not all
+ * architectures have easy movemask instruction, this helps to iterate over
+ * groups of bits easier and faster.
+ */
+FORCE_INLINE_TEMPLATE U32
+ZSTD_row_matchMaskGroupWidth(const U32 rowEntries)
+{
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
+    (void)rowEntries;
+#if defined(ZSTD_ARCH_ARM_NEON)
+    /* NEON path only works for little endian */
+    if (!MEM_isLittleEndian()) {
+        return 1;
+    }
+    if (rowEntries == 16) {
+        return 4;
+    }
+    if (rowEntries == 32) {
+        return 2;
+    }
+    if (rowEntries == 64) {
+        return 1;
+    }
+#endif
+    return 1;
+}
+
+#if defined(ZSTD_ARCH_X86_SSE2)
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getSSEMask(int nbChunks, const BYTE* const src, const BYTE tag, const U32 head)
+{
+    const __m128i comparisonMask = _mm_set1_epi8((char)tag);
+    int matches[4] = {0};
+    int i;
+    assert(nbChunks == 1 || nbChunks == 2 || nbChunks == 4);
+    for (i=0; i<nbChunks; i++) {
+        const __m128i chunk = _mm_loadu_si128((const __m128i*)(const void*)(src + 16*i));
+        const __m128i equalMask = _mm_cmpeq_epi8(chunk, comparisonMask);
+        matches[i] = _mm_movemask_epi8(equalMask);
+    }
+    if (nbChunks == 1) return ZSTD_rotateRight_U16((U16)matches[0], head);
+    if (nbChunks == 2) return ZSTD_rotateRight_U32((U32)matches[1] << 16 | (U32)matches[0], head);
+    assert(nbChunks == 4);
+    return ZSTD_rotateRight_U64((U64)matches[3] << 48 | (U64)matches[2] << 32 | (U64)matches[1] << 16 | (U64)matches[0], head);
+}
+#endif
+
+#if defined(ZSTD_ARCH_ARM_NEON)
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getNEONMask(const U32 rowEntries, const BYTE* const src, const BYTE tag, const U32 headGrouped)
+{
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    if (rowEntries == 16) {
+        /* vshrn_n_u16 shifts by 4 every u16 and narrows to 8 lower bits.
+         * After that groups of 4 bits represent the equalMask. We lower
+         * all bits except the highest in these groups by doing AND with
+         * 0x88 = 0b10001000.
+         */
+        const uint8x16_t chunk = vld1q_u8(src);
+        const uint16x8_t equalMask = vreinterpretq_u16_u8(vceqq_u8(chunk, vdupq_n_u8(tag)));
+        const uint8x8_t res = vshrn_n_u16(equalMask, 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0);
+        return ZSTD_rotateRight_U64(matches, headGrouped) & 0x8888888888888888ull;
+    } else if (rowEntries == 32) {
+        /* Same idea as with rowEntries == 16 but doing AND with
+         * 0x55 = 0b01010101.
+         */
+        const uint16x8x2_t chunk = vld2q_u16((const uint16_t*)(const void*)src);
+        const uint8x16_t chunk0 = vreinterpretq_u8_u16(chunk.val[0]);
+        const uint8x16_t chunk1 = vreinterpretq_u8_u16(chunk.val[1]);
+        const uint8x16_t dup = vdupq_n_u8(tag);
+        const uint8x8_t t0 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk0, dup)), 6);
+        const uint8x8_t t1 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk1, dup)), 6);
+        const uint8x8_t res = vsli_n_u8(t0, t1, 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0) ;
+        return ZSTD_rotateRight_U64(matches, headGrouped) & 0x5555555555555555ull;
+    } else { /* rowEntries == 64 */
+        const uint8x16x4_t chunk = vld4q_u8(src);
+        const uint8x16_t dup = vdupq_n_u8(tag);
+        const uint8x16_t cmp0 = vceqq_u8(chunk.val[0], dup);
+        const uint8x16_t cmp1 = vceqq_u8(chunk.val[1], dup);
+        const uint8x16_t cmp2 = vceqq_u8(chunk.val[2], dup);
+        const uint8x16_t cmp3 = vceqq_u8(chunk.val[3], dup);
+
+        const uint8x16_t t0 = vsriq_n_u8(cmp1, cmp0, 1);
+        const uint8x16_t t1 = vsriq_n_u8(cmp3, cmp2, 1);
+        const uint8x16_t t2 = vsriq_n_u8(t1, t0, 2);
+        const uint8x16_t t3 = vsriq_n_u8(t2, t2, 4);
+        const uint8x8_t t4 = vshrn_n_u16(vreinterpretq_u16_u8(t3), 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(t4), 0);
+        return ZSTD_rotateRight_U64(matches, headGrouped);
+    }
+}
+#endif
+
+/* Returns a ZSTD_VecMask (U64) that has the nth group (determined by
+ * ZSTD_row_matchMaskGroupWidth) of bits set to 1 if the newly-computed "tag"
+ * matches the hash at the nth position in a row of the tagTable.
+ * Each row is a circular buffer beginning at the value of "headGrouped". So we
+ * must rotate the "matches" bitfield to match up with the actual layout of the
+ * entries within the hashTable */
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 headGrouped, const U32 rowEntries)
+{
+    const BYTE* const src = tagRow;
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
+    assert(ZSTD_row_matchMaskGroupWidth(rowEntries) * rowEntries <= sizeof(ZSTD_VecMask) * 8);
+
+#if defined(ZSTD_ARCH_X86_SSE2)
+
+    return ZSTD_row_getSSEMask(rowEntries / 16, src, tag, headGrouped);
+
+#else /* SW or NEON-LE */
+
+# if defined(ZSTD_ARCH_ARM_NEON)
+  /* This NEON path only works for little endian - otherwise use SWAR below */
+    if (MEM_isLittleEndian()) {
+        return ZSTD_row_getNEONMask(rowEntries, src, tag, headGrouped);
+    }
+# endif /* ZSTD_ARCH_ARM_NEON */
+    /* SWAR */
+    {   const int chunkSize = sizeof(size_t);
+        const size_t shiftAmount = ((chunkSize * 8) - chunkSize);
+        const size_t xFF = ~((size_t)0);
+        const size_t x01 = xFF / 0xFF;
+        const size_t x80 = x01 << 7;
+        const size_t splatChar = tag * x01;
+        ZSTD_VecMask matches = 0;
+        int i = rowEntries - chunkSize;
+        assert((sizeof(size_t) == 4) || (sizeof(size_t) == 8));
+        if (MEM_isLittleEndian()) { /* runtime check so have two loops */
+            const size_t extractMagic = (xFF / 0x7F) >> chunkSize;
+            do {
+                size_t chunk = MEM_readST(&src[i]);
+                chunk ^= splatChar;
+                chunk = (((chunk | x80) - x01) | chunk) & x80;
+                matches <<= chunkSize;
+                matches |= (chunk * extractMagic) >> shiftAmount;
+                i -= chunkSize;
+            } while (i >= 0);
+        } else { /* big endian: reverse bits during extraction */
+            const size_t msb = xFF ^ (xFF >> 1);
+            const size_t extractMagic = (msb / 0x1FF) | msb;
+            do {
+                size_t chunk = MEM_readST(&src[i]);
+                chunk ^= splatChar;
+                chunk = (((chunk | x80) - x01) | chunk) & x80;
+                matches <<= chunkSize;
+                matches |= ((chunk >> 7) * extractMagic) >> shiftAmount;
+                i -= chunkSize;
+            } while (i >= 0);
+        }
+        matches = ~matches;
+        if (rowEntries == 16) {
+            return ZSTD_rotateRight_U16((U16)matches, headGrouped);
+        } else if (rowEntries == 32) {
+            return ZSTD_rotateRight_U32((U32)matches, headGrouped);
+        } else {
+            return ZSTD_rotateRight_U64((U64)matches, headGrouped);
+        }
+    }
+#endif
+}
+
+/* The high-level approach of the SIMD row based match finder is as follows:
+ * - Figure out where to insert the new entry:
+ *      - Generate a hash from a byte along with an additional 1-byte "short hash". The additional byte is our "tag"
+ *      - The hashTable is effectively split into groups or "rows" of 16 or 32 entries of U32, and the hash determines
+ *        which row to insert into.
+ *      - Determine the correct position within the row to insert the entry into. Each row of 16 or 32 can
+ *        be considered as a circular buffer with a "head" index that resides in the tagTable.
+ *      - Also insert the "tag" into the equivalent row and position in the tagTable.
+ *          - Note: The tagTable has 17 or 33 1-byte entries per row, due to 16 or 32 tags, and 1 "head" entry.
+ *                  The 17 or 33 entry rows are spaced out to occur every 32 or 64 bytes, respectively,
+ *                  for alignment/performance reasons, leaving some bytes unused.
+ * - Use SIMD to efficiently compare the tags in the tagTable to the 1-byte "short hash" and
+ *   generate a bitfield that we can cycle through to check the collisions in the hash table.
+ * - Pick the longest match.
+ */
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_RowFindBestMatch(
+                        ZSTD_matchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls, const ZSTD_dictMode_e dictMode,
+                        const U32 rowLog)
+{
+    U32* const hashTable = ms->hashTable;
+    BYTE* const tagTable = ms->tagTable;
+    U32* const hashCache = ms->hashCache;
+    const U32 hashLog = ms->rowHashLog;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32 curr = (U32)(ip-base);
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 lowestValid = ms->window.lowLimit;
+    const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
+    const U32 isDictionary = (ms->loadedDictEnd != 0);
+    const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
+    const U32 rowEntries = (1U << rowLog);
+    const U32 rowMask = rowEntries - 1;
+    const U32 cappedSearchLog = MIN(cParams->searchLog, rowLog); /* nb of searches is capped at nb entries per row */
+    const U32 groupWidth = ZSTD_row_matchMaskGroupWidth(rowEntries);
+    const U64 hashSalt = ms->hashSalt;
+    U32 nbAttempts = 1U << cappedSearchLog;
+    size_t ml=4-1;
+    U32 hash;
+
+    /* DMS/DDS variables that may be referenced laster */
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+
+    /* Initialize the following variables to satisfy static analyzer */
+    size_t ddsIdx = 0;
+    U32 ddsExtraAttempts = 0; /* cctx hash tables are limited in searches, but allow extra searches into DDS */
+    U32 dmsTag = 0;
+    U32* dmsRow = NULL;
+    BYTE* dmsTagRow = NULL;
+
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        const U32 ddsHashLog = dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
+        {   /* Prefetch DDS hashtable entry */
+            ddsIdx = ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG;
+            PREFETCH_L1(&dms->hashTable[ddsIdx]);
+        }
+        ddsExtraAttempts = cParams->searchLog > rowLog ? 1U << (cParams->searchLog - rowLog) : 0;
+    }
+
+    if (dictMode == ZSTD_dictMatchState) {
+        /* Prefetch DMS rows */
+        U32* const dmsHashTable = dms->hashTable;
+        BYTE* const dmsTagTable = dms->tagTable;
+        U32 const dmsHash = (U32)ZSTD_hashPtr(ip, dms->rowHashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
+        U32 const dmsRelRow = (dmsHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        dmsTag = dmsHash & ZSTD_ROW_HASH_TAG_MASK;
+        dmsTagRow = (BYTE*)(dmsTagTable + dmsRelRow);
+        dmsRow = dmsHashTable + dmsRelRow;
+        ZSTD_row_prefetch(dmsHashTable, dmsTagTable, dmsRelRow, rowLog);
+    }
+
+    /* Update the hashTable and tagTable up to (but not including) ip */
+    if (!ms->lazySkipping) {
+        ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 1 /* useCache */);
+        hash = ZSTD_row_nextCachedHash(hashCache, hashTable, tagTable, base, curr, hashLog, rowLog, mls, hashSalt);
+    } else {
+        /* Stop inserting every position when in the lazy skipping mode.
+         * The hash cache is also not kept up to date in this mode.
+         */
+        hash = (U32)ZSTD_hashPtrSalted(ip, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
+        ms->nextToUpdate = curr;
+    }
+    ms->hashSaltEntropy += hash; /* collect salt entropy */
+
+    {   /* Get the hash for ip, compute the appropriate row */
+        U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        U32 const tag = hash & ZSTD_ROW_HASH_TAG_MASK;
+        U32* const row = hashTable + relRow;
+        BYTE* tagRow = (BYTE*)(tagTable + relRow);
+        U32 const headGrouped = (*tagRow & rowMask) * groupWidth;
+        U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
+        size_t numMatches = 0;
+        size_t currMatch = 0;
+        ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, headGrouped, rowEntries);
+
+        /* Cycle through the matches and prefetch */
+        for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
+            U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
+            U32 const matchIndex = row[matchPos];
+            if(matchPos == 0) continue;
+            assert(numMatches < rowEntries);
+            if (matchIndex < lowLimit)
+                break;
+            if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
+                PREFETCH_L1(base + matchIndex);
+            } else {
+                PREFETCH_L1(dictBase + matchIndex);
+            }
+            matchBuffer[numMatches++] = matchIndex;
+            --nbAttempts;
+        }
+
+        /* Speed opt: insert current byte into hashtable too. This allows us to avoid one iteration of the loop
+           in ZSTD_row_update_internal() at the next search. */
+        {
+            U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
+            tagRow[pos] = (BYTE)tag;
+            row[pos] = ms->nextToUpdate++;
+        }
+
+        /* Return the longest match */
+        for (; currMatch < numMatches; ++currMatch) {
+            U32 const matchIndex = matchBuffer[currMatch];
+            size_t currentMl=0;
+            assert(matchIndex < curr);
+            assert(matchIndex >= lowLimit);
+
+            if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
+                const BYTE* const match = base + matchIndex;
+                assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
+                /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+                if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
+                    currentMl = ZSTD_count(ip, match, iLimit);
+            } else {
+                const BYTE* const match = dictBase + matchIndex;
+                assert(match+4 <= dictEnd);
+                if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                    currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
+            }
+
+            /* Save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+        }
+    }
+
+    assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts + ddsExtraAttempts, dms,
+                                                  ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
+    } else if (dictMode == ZSTD_dictMatchState) {
+        /* TODO: Measure and potentially add prefetching to DMS */
+        const U32 dmsLowestIndex       = dms->window.dictLimit;
+        const BYTE* const dmsBase      = dms->window.base;
+        const BYTE* const dmsEnd       = dms->window.nextSrc;
+        const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
+        const U32 dmsIndexDelta        = dictLimit - dmsSize;
+
+        {   U32 const headGrouped = (*dmsTagRow & rowMask) * groupWidth;
+            U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
+            size_t numMatches = 0;
+            size_t currMatch = 0;
+            ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, headGrouped, rowEntries);
+
+            for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
+                U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
+                U32 const matchIndex = dmsRow[matchPos];
+                if(matchPos == 0) continue;
+                if (matchIndex < dmsLowestIndex)
+                    break;
+                PREFETCH_L1(dmsBase + matchIndex);
+                matchBuffer[numMatches++] = matchIndex;
+                --nbAttempts;
+            }
+
+            /* Return the longest match */
+            for (; currMatch < numMatches; ++currMatch) {
+                U32 const matchIndex = matchBuffer[currMatch];
+                size_t currentMl=0;
+                assert(matchIndex >= dmsLowestIndex);
+                assert(matchIndex < curr);
+
+                {   const BYTE* const match = dmsBase + matchIndex;
+                    assert(match+4 <= dmsEnd);
+                    if (MEM_read32(match) == MEM_read32(ip))
+                        currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
+                }
+
+                if (currentMl > ml) {
+                    ml = currentMl;
+                    assert(curr > matchIndex + dmsIndexDelta);
+                    *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
+                    if (ip+currentMl == iLimit) break;
+                }
+            }
+        }
+    }
+    return ml;
+}
+
+
+/**
+ * Generate search functions templated on (dictMode, mls, rowLog).
+ * These functions are outlined for code size & compilation time.
+ * ZSTD_searchMax() dispatches to the correct implementation function.
+ *
+ * TODO: The start of the search function involves loading and calculating a
+ * bunch of constants from the ZSTD_matchState_t. These computations could be
+ * done in an initialization function, and saved somewhere in the match state.
+ * Then we could pass a pointer to the saved state instead of the match state,
+ * and avoid duplicate computations.
+ *
+ * TODO: Move the match re-winding into searchMax. This improves compression
+ * ratio, and unlocks further simplifications with the next TODO.
+ *
+ * TODO: Try moving the repcode search into searchMax. After the re-winding
+ * and repcode search are in searchMax, there is no more logic in the match
+ * finder loop that requires knowledge about the dictMode. So we should be
+ * able to avoid force inlining it, and we can join the extDict loop with
+ * the single segment loop. It should go in searchMax instead of its own
+ * function to avoid having multiple virtual function calls per search.
+ */
+
+#define ZSTD_BT_SEARCH_FN(dictMode, mls) ZSTD_BtFindBestMatch_##dictMode##_##mls
+#define ZSTD_HC_SEARCH_FN(dictMode, mls) ZSTD_HcFindBestMatch_##dictMode##_##mls
+#define ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog) ZSTD_RowFindBestMatch_##dictMode##_##mls##_##rowLog
+
+#define ZSTD_SEARCH_FN_ATTRS FORCE_NOINLINE
+
+#define GEN_ZSTD_BT_SEARCH_FN(dictMode, mls)                                           \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_BT_SEARCH_FN(dictMode, mls)(                      \
+            ZSTD_matchState_t* ms,                                                     \
+            const BYTE* ip, const BYTE* const iLimit,                                  \
+            size_t* offBasePtr)                                                        \
+    {                                                                                  \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                           \
+        return ZSTD_BtFindBestMatch(ms, ip, iLimit, offBasePtr, mls, ZSTD_##dictMode); \
+    }                                                                                  \
+
+#define GEN_ZSTD_HC_SEARCH_FN(dictMode, mls)                                          \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_HC_SEARCH_FN(dictMode, mls)(                     \
+            ZSTD_matchState_t* ms,                                                    \
+            const BYTE* ip, const BYTE* const iLimit,                                 \
+            size_t* offsetPtr)                                                        \
+    {                                                                                 \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                          \
+        return ZSTD_HcFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode); \
+    }                                                                                 \
+
+#define GEN_ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)                                          \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(                     \
+            ZSTD_matchState_t* ms,                                                             \
+            const BYTE* ip, const BYTE* const iLimit,                                          \
+            size_t* offsetPtr)                                                                 \
+    {                                                                                          \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                                   \
+        assert(MAX(4, MIN(6, ms->cParams.searchLog)) == rowLog);                               \
+        return ZSTD_RowFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode, rowLog); \
+    }                                                                                          \
+
+#define ZSTD_FOR_EACH_ROWLOG(X, dictMode, mls) \
+    X(dictMode, mls, 4)                        \
+    X(dictMode, mls, 5)                        \
+    X(dictMode, mls, 6)
+
+#define ZSTD_FOR_EACH_MLS_ROWLOG(X, dictMode) \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 4)      \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 5)      \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 6)
+
+#define ZSTD_FOR_EACH_MLS(X, dictMode) \
+    X(dictMode, 4)                     \
+    X(dictMode, 5)                     \
+    X(dictMode, 6)
+
+#define ZSTD_FOR_EACH_DICT_MODE(X, ...) \
+    X(__VA_ARGS__, noDict)              \
+    X(__VA_ARGS__, extDict)             \
+    X(__VA_ARGS__, dictMatchState)      \
+    X(__VA_ARGS__, dedicatedDictSearch)
+
+/* Generate row search fns for each combination of (dictMode, mls, rowLog) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS_ROWLOG, GEN_ZSTD_ROW_SEARCH_FN)
+/* Generate binary Tree search fns for each combination of (dictMode, mls) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_BT_SEARCH_FN)
+/* Generate hash chain search fns for each combination of (dictMode, mls) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_HC_SEARCH_FN)
+
+typedef enum { search_hashChain=0, search_binaryTree=1, search_rowHash=2 } searchMethod_e;
+
+#define GEN_ZSTD_CALL_BT_SEARCH_FN(dictMode, mls)                         \
+    case mls:                                                             \
+        return ZSTD_BT_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
+#define GEN_ZSTD_CALL_HC_SEARCH_FN(dictMode, mls)                         \
+    case mls:                                                             \
+        return ZSTD_HC_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
+#define GEN_ZSTD_CALL_ROW_SEARCH_FN(dictMode, mls, rowLog)                         \
+    case rowLog:                                                                   \
+        return ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(ms, ip, iend, offsetPtr);
+
+#define ZSTD_SWITCH_MLS(X, dictMode)   \
+    switch (mls) {                     \
+        ZSTD_FOR_EACH_MLS(X, dictMode) \
+    }
+
+#define ZSTD_SWITCH_ROWLOG(dictMode, mls)                                    \
+    case mls:                                                                \
+        switch (rowLog) {                                                    \
+            ZSTD_FOR_EACH_ROWLOG(GEN_ZSTD_CALL_ROW_SEARCH_FN, dictMode, mls) \
+        }                                                                    \
+        ZSTD_UNREACHABLE;                                                    \
+        break;
+
+#define ZSTD_SWITCH_SEARCH_METHOD(dictMode)                       \
+    switch (searchMethod) {                                       \
+        case search_hashChain:                                    \
+            ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_HC_SEARCH_FN, dictMode) \
+            break;                                                \
+        case search_binaryTree:                                   \
+            ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_BT_SEARCH_FN, dictMode) \
+            break;                                                \
+        case search_rowHash:                                      \
+            ZSTD_SWITCH_MLS(ZSTD_SWITCH_ROWLOG, dictMode)         \
+            break;                                                \
+    }                                                             \
+    ZSTD_UNREACHABLE;
+
+/**
+ * Searches for the longest match at @p ip.
+ * Dispatches to the correct implementation function based on the
+ * (searchMethod, dictMode, mls, rowLog). We use switch statements
+ * here instead of using an indirect function call through a function
+ * pointer because after Spectre and Meltdown mitigations, indirect
+ * function calls can be very costly, especially in the kernel.
+ *
+ * NOTE: dictMode and searchMethod should be templated, so those switch
+ * statements should be optimized out. Only the mls & rowLog switches
+ * should be left.
+ *
+ * @param ms The match state.
+ * @param ip The position to search at.
+ * @param iend The end of the input data.
+ * @param[out] offsetPtr Stores the match offset into this pointer.
+ * @param mls The minimum search length, in the range [4, 6].
+ * @param rowLog The row log (if applicable), in the range [4, 6].
+ * @param searchMethod The search method to use (templated).
+ * @param dictMode The dictMode (templated).
+ *
+ * @returns The length of the longest match found, or < mls if no match is found.
+ * If a match is found its offset is stored in @p offsetPtr.
+ */
+FORCE_INLINE_TEMPLATE size_t ZSTD_searchMax(
+    ZSTD_matchState_t* ms,
+    const BYTE* ip,
+    const BYTE* iend,
+    size_t* offsetPtr,
+    U32 const mls,
+    U32 const rowLog,
+    searchMethod_e const searchMethod,
+    ZSTD_dictMode_e const dictMode)
+{
+    if (dictMode == ZSTD_noDict) {
+        ZSTD_SWITCH_SEARCH_METHOD(noDict)
+    } else if (dictMode == ZSTD_extDict) {
+        ZSTD_SWITCH_SEARCH_METHOD(extDict)
+    } else if (dictMode == ZSTD_dictMatchState) {
+        ZSTD_SWITCH_SEARCH_METHOD(dictMatchState)
+    } else if (dictMode == ZSTD_dedicatedDictSearch) {
+        ZSTD_SWITCH_SEARCH_METHOD(dedicatedDictSearch)
+    }
+    ZSTD_UNREACHABLE;
+    return 0;
+}
+
+/* *******************************
+*  Common parser - lazy strategy
+*********************************/
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_compressBlock_lazy_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        const void* src, size_t srcSize,
+                        const searchMethod_e searchMethod, const U32 depth,
+                        ZSTD_dictMode_e const dictMode)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = (searchMethod == search_rowHash) ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32 prefixLowestIndex = ms->window.dictLimit;
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+    const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
+
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    const int isDMS = dictMode == ZSTD_dictMatchState;
+    const int isDDS = dictMode == ZSTD_dedicatedDictSearch;
+    const int isDxS = isDMS || isDDS;
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const U32 dictLowestIndex      = isDxS ? dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = isDxS ? dms->window.base : NULL;
+    const BYTE* const dictLowest   = isDxS ? dictBase + dictLowestIndex : NULL;
+    const BYTE* const dictEnd      = isDxS ? dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = isDxS ?
+                                     prefixLowestIndex - (U32)(dictEnd - dictBase) :
+                                     0;
+    const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u) (searchFunc=%u)", (U32)dictMode, (U32)searchMethod);
+    ip += (dictAndPrefixLength == 0);
+    if (dictMode == ZSTD_noDict) {
+        U32 const curr = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, ms->cParams.windowLog);
+        U32 const maxRep = curr - windowLow;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+    if (isDxS) {
+        /* dictMatchState repCode checks don't currently handle repCode == 0
+         * disabling. */
+        assert(offset_1 <= dictAndPrefixLength);
+        assert(offset_2 <= dictAndPrefixLength);
+    }
+
+    /* Reset the lazy skipping state */
+    ms->lazySkipping = 0;
+
+    if (searchMethod == search_rowHash) {
+        ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+    }
+
+    /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offBase = REPCODE1_TO_OFFBASE;
+        const BYTE* start=ip+1;
+        DEBUGLOG(7, "search baseline (depth 0)");
+
+        /* check repCode */
+        if (isDxS) {
+            const U32 repIndex = (U32)(ip - base) + 1 - offset_1;
+            const BYTE* repMatch = ((dictMode == ZSTD_dictMatchState || dictMode == ZSTD_dedicatedDictSearch)
+                                && repIndex < prefixLowestIndex) ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+            if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+                const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                if (depth==0) goto _storeSequence;
+            }
+        }
+        if ( dictMode == ZSTD_noDict
+          && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
+            matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            if (depth==0) goto _storeSequence;
+        }
+
+        /* first search (depth 0) */
+        {   size_t offbaseFound = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offbaseFound, mls, rowLog, searchMethod, dictMode);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offBase = offbaseFound;
+        }
+
+        if (matchLength < 4) {
+            size_t const step = ((size_t)(ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */;
+            ip += step;
+            /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
+             * In this mode we stop inserting every position into our tables, and only insert
+             * positions that we search, which is one in step positions.
+             * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
+             * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
+             * triggered once we've gone 2KB without finding any matches.
+             */
+            ms->lazySkipping = step > kLazySkippingStep;
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            DEBUGLOG(7, "search depth 1");
+            ip ++;
+            if ( (dictMode == ZSTD_noDict)
+              && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                int const gain2 = (int)(mlRep * 3);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
+                if ((mlRep >= 4) && (gain2 > gain1))
+                    matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+            }
+            if (isDxS) {
+                const U32 repIndex = (U32)(ip - base) - offset_1;
+                const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
+                if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                    && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                    int const gain2 = (int)(mlRep * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                }
+            }
+            {   size_t ofbCandidate=999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                DEBUGLOG(7, "search depth 2");
+                ip ++;
+                if ( (dictMode == ZSTD_noDict)
+                  && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                    size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                    int const gain2 = (int)(mlRep * 4);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                }
+                if (isDxS) {
+                    const U32 repIndex = (U32)(ip - base) - offset_1;
+                    const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+                    if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                        && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                        const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                        size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                        int const gain2 = (int)(mlRep * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
+                        if ((mlRep >= 4) && (gain2 > gain1))
+                            matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                    }
+                }
+                {   size_t ofbCandidate=999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* NOTE:
+         * Pay attention that `start[-value]` can lead to strange undefined behavior
+         * notably if `value` is unsigned, resulting in a large positive `-value`.
+         */
+        /* catch up */
+        if (OFFBASE_IS_OFFSET(offBase)) {
+            if (dictMode == ZSTD_noDict) {
+                while ( ((start > anchor) & (start - OFFBASE_TO_OFFSET(offBase) > prefixLowest))
+                     && (start[-1] == (start-OFFBASE_TO_OFFSET(offBase))[-1]) )  /* only search for offset within prefix */
+                    { start--; matchLength++; }
+            }
+            if (isDxS) {
+                U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
+                const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
+                const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
+                while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            }
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
+        }
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = (size_t)(start - anchor);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
+            anchor = ip = start + matchLength;
+        }
+        if (ms->lazySkipping) {
+            /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
+            if (searchMethod == search_rowHash) {
+                ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+            }
+            ms->lazySkipping = 0;
+        }
+
+        /* check immediate repcode */
+        if (isDxS) {
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex = current2 - offset_2;
+                const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                        dictBase - dictIndexDelta + repIndex :
+                        base + repIndex;
+                if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex) >= 3 /* intentional overflow */)
+                   && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4;
+                    offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
+                    ip += matchLength;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+            }
+        }
+
+        if (dictMode == ZSTD_noDict) {
+            while ( ((ip <= ilimit) & (offset_2>0))
+                 && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
+                /* store sequence */
+                matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap repcodes */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved1;
+    rep[1] = offset_2 ? offset_2 : offsetSaved2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_greedy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dictMatchState);
+}
+
+
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dedicatedDictSearch);
+}
+
+/* Row-based matchfinder */
+size_t ZSTD_compressBlock_lazy2_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_greedy_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy_dictMatchState_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_greedy_dictMatchState_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dictMatchState);
+}
+
+
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dedicatedDictSearch);
+}
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_lazy_extDict_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        const void* src, size_t srcSize,
+                        const searchMethod_e searchMethod, const U32 depth)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = searchMethod == search_rowHash ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictEnd  = dictBase + dictLimit;
+    const BYTE* const dictStart  = dictBase + ms->window.lowLimit;
+    const U32 windowLog = ms->cParams.windowLog;
+    const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
+
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic (searchFunc=%u)", (U32)searchMethod);
+
+    /* Reset the lazy skipping state */
+    ms->lazySkipping = 0;
+
+    /* init */
+    ip += (ip == prefixStart);
+    if (searchMethod == search_rowHash) {
+        ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+    }
+
+    /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offBase = REPCODE1_TO_OFFBASE;
+        const BYTE* start=ip+1;
+        U32 curr = (U32)(ip-base);
+
+        /* check repCode */
+        {   const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr+1, windowLog);
+            const U32 repIndex = (U32)(curr+1 - offset_1);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow */
+               & (offset_1 <= curr+1 - windowLow) ) /* note: we are searching at curr+1 */
+            if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                if (depth==0) goto _storeSequence;
+        }   }
+
+        /* first search (depth 0) */
+        {   size_t ofbCandidate = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offBase = ofbCandidate;
+        }
+
+        if (matchLength < 4) {
+            size_t const step = ((size_t)(ip-anchor) >> kSearchStrength);
+            ip += step + 1;   /* jump faster over incompressible sections */
+            /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
+             * In this mode we stop inserting every position into our tables, and only insert
+             * positions that we search, which is one in step positions.
+             * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
+             * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
+             * triggered once we've gone 2KB without finding any matches.
+             */
+            ms->lazySkipping = step > kLazySkippingStep;
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            curr++;
+            /* check repCode */
+            if (offBase) {
+                const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
+                const U32 repIndex = (U32)(curr - offset_1);
+                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                const BYTE* const repMatch = repBase + repIndex;
+                if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments  */
+                   & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+                if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                    /* repcode detected */
+                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                    size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                    int const gain2 = (int)(repLength * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
+                    if ((repLength >= 4) && (gain2 > gain1))
+                        matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
+            }   }
+
+            /* search match, depth 1 */
+            {   size_t ofbCandidate = 999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                curr++;
+                /* check repCode */
+                if (offBase) {
+                    const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
+                    const U32 repIndex = (U32)(curr - offset_1);
+                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                    const BYTE* const repMatch = repBase + repIndex;
+                    if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments  */
+                       & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+                    if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                        /* repcode detected */
+                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                        size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                        int const gain2 = (int)(repLength * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
+                        if ((repLength >= 4) && (gain2 > gain1))
+                            matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                }   }
+
+                /* search match, depth 2 */
+                {   size_t ofbCandidate = 999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* catch up */
+        if (OFFBASE_IS_OFFSET(offBase)) {
+            U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
+            const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
+            const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
+            while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
+        }
+
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = (size_t)(start - anchor);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
+            anchor = ip = start + matchLength;
+        }
+        if (ms->lazySkipping) {
+            /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
+            if (searchMethod == search_rowHash) {
+                ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+            }
+            ms->lazySkipping = 0;
+        }
+
+        /* check immediate repcode */
+        while (ip <= ilimit) {
+            const U32 repCurrent = (U32)(ip-base);
+            const U32 windowLow = ZSTD_getLowestMatchIndex(ms, repCurrent, windowLog);
+            const U32 repIndex = repCurrent - offset_2;
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments  */
+               & (offset_2 <= repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+            if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset history */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+            break;
+    }   }
+
+    /* Save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0);
+}
+
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1);
+}
+
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2);
+}
+
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2);
+}
+
+size_t ZSTD_compressBlock_greedy_extDict_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0);
+}
+
+size_t ZSTD_compressBlock_lazy_extDict_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1);
+}
+
+size_t ZSTD_compressBlock_lazy2_extDict_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_lazy.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_lazy.h
new file mode 100644
index 0000000..3bde673
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_lazy.h
@@ -0,0 +1,127 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LAZY_H
+#define ZSTD_LAZY_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"
+
+/**
+ * Dedicated Dictionary Search Structure bucket log. In the
+ * ZSTD_dedicatedDictSearch mode, the hashTable has
+ * 2 ** ZSTD_LAZY_DDSS_BUCKET_LOG entries in each bucket, rather than just
+ * one.
+ */
+#define ZSTD_LAZY_DDSS_BUCKET_LOG 2
+
+#define ZSTD_ROW_HASH_TAG_BITS 8        /* nb bits to use for the tag */
+
+U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
+void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip);
+
+void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const BYTE* const ip);
+
+void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue);  /*! used in ZSTD_reduceIndex(). preemptively increase value of ZSTD_DUBT_UNSORTED_MARK */
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_dictMatchState_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_dictMatchState_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_extDict_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_extDict_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_extDict_row(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_LAZY_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm.c
new file mode 100644
index 0000000..3d74ff1
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm.c
@@ -0,0 +1,724 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_ldm.h"
+
+#include "../common/debug.h"
+#include "../common/xxhash.h"
+#include "zstd_fast.h"          /* ZSTD_fillHashTable() */
+#include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
+#include "zstd_ldm_geartab.h"
+
+#define LDM_BUCKET_SIZE_LOG 3
+#define LDM_MIN_MATCH_LENGTH 64
+#define LDM_HASH_RLOG 7
+
+typedef struct {
+    U64 rolling;
+    U64 stopMask;
+} ldmRollingHashState_t;
+
+/** ZSTD_ldm_gear_init():
+ *
+ * Initializes the rolling hash state such that it will honor the
+ * settings in params. */
+static void ZSTD_ldm_gear_init(ldmRollingHashState_t* state, ldmParams_t const* params)
+{
+    unsigned maxBitsInMask = MIN(params->minMatchLength, 64);
+    unsigned hashRateLog = params->hashRateLog;
+
+    state->rolling = ~(U32)0;
+
+    /* The choice of the splitting criterion is subject to two conditions:
+     *   1. it has to trigger on average every 2^(hashRateLog) bytes;
+     *   2. ideally, it has to depend on a window of minMatchLength bytes.
+     *
+     * In the gear hash algorithm, bit n depends on the last n bytes;
+     * so in order to obtain a good quality splitting criterion it is
+     * preferable to use bits with high weight.
+     *
+     * To match condition 1 we use a mask with hashRateLog bits set
+     * and, because of the previous remark, we make sure these bits
+     * have the highest possible weight while still respecting
+     * condition 2.
+     */
+    if (hashRateLog > 0 && hashRateLog <= maxBitsInMask) {
+        state->stopMask = (((U64)1 << hashRateLog) - 1) << (maxBitsInMask - hashRateLog);
+    } else {
+        /* In this degenerate case we simply honor the hash rate. */
+        state->stopMask = ((U64)1 << hashRateLog) - 1;
+    }
+}
+
+/** ZSTD_ldm_gear_reset()
+ * Feeds [data, data + minMatchLength) into the hash without registering any
+ * splits. This effectively resets the hash state. This is used when skipping
+ * over data, either at the beginning of a block, or skipping sections.
+ */
+static void ZSTD_ldm_gear_reset(ldmRollingHashState_t* state,
+                                BYTE const* data, size_t minMatchLength)
+{
+    U64 hash = state->rolling;
+    size_t n = 0;
+
+#define GEAR_ITER_ONCE() do {                                  \
+        hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
+        n += 1;                                                \
+    } while (0)
+    while (n + 3 < minMatchLength) {
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+    }
+    while (n < minMatchLength) {
+        GEAR_ITER_ONCE();
+    }
+#undef GEAR_ITER_ONCE
+}
+
+/** ZSTD_ldm_gear_feed():
+ *
+ * Registers in the splits array all the split points found in the first
+ * size bytes following the data pointer. This function terminates when
+ * either all the data has been processed or LDM_BATCH_SIZE splits are
+ * present in the splits array.
+ *
+ * Precondition: The splits array must not be full.
+ * Returns: The number of bytes processed. */
+static size_t ZSTD_ldm_gear_feed(ldmRollingHashState_t* state,
+                                 BYTE const* data, size_t size,
+                                 size_t* splits, unsigned* numSplits)
+{
+    size_t n;
+    U64 hash, mask;
+
+    hash = state->rolling;
+    mask = state->stopMask;
+    n = 0;
+
+#define GEAR_ITER_ONCE() do { \
+        hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
+        n += 1; \
+        if (UNLIKELY((hash & mask) == 0)) { \
+            splits[*numSplits] = n; \
+            *numSplits += 1; \
+            if (*numSplits == LDM_BATCH_SIZE) \
+                goto done; \
+        } \
+    } while (0)
+
+    while (n + 3 < size) {
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+    }
+    while (n < size) {
+        GEAR_ITER_ONCE();
+    }
+
+#undef GEAR_ITER_ONCE
+
+done:
+    state->rolling = hash;
+    return n;
+}
+
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams)
+{
+    params->windowLog = cParams->windowLog;
+    ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
+    DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
+    if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
+    if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
+    if (params->hashLog == 0) {
+        params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
+        assert(params->hashLog <= ZSTD_HASHLOG_MAX);
+    }
+    if (params->hashRateLog == 0) {
+        params->hashRateLog = params->windowLog < params->hashLog
+                                   ? 0
+                                   : params->windowLog - params->hashLog;
+    }
+    params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
+}
+
+size_t ZSTD_ldm_getTableSize(ldmParams_t params)
+{
+    size_t const ldmHSize = ((size_t)1) << params.hashLog;
+    size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
+    size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
+    size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
+                           + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
+    return params.enableLdm == ZSTD_ps_enable ? totalSize : 0;
+}
+
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
+{
+    return params.enableLdm == ZSTD_ps_enable ? (maxChunkSize / params.minMatchLength) : 0;
+}
+
+/** ZSTD_ldm_getBucket() :
+ *  Returns a pointer to the start of the bucket associated with hash. */
+static ldmEntry_t* ZSTD_ldm_getBucket(
+        ldmState_t* ldmState, size_t hash, ldmParams_t const ldmParams)
+{
+    return ldmState->hashTable + (hash << ldmParams.bucketSizeLog);
+}
+
+/** ZSTD_ldm_insertEntry() :
+ *  Insert the entry with corresponding hash into the hash table */
+static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
+                                 size_t const hash, const ldmEntry_t entry,
+                                 ldmParams_t const ldmParams)
+{
+    BYTE* const pOffset = ldmState->bucketOffsets + hash;
+    unsigned const offset = *pOffset;
+
+    *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + offset) = entry;
+    *pOffset = (BYTE)((offset + 1) & ((1u << ldmParams.bucketSizeLog) - 1));
+
+}
+
+/** ZSTD_ldm_countBackwardsMatch() :
+ *  Returns the number of bytes that match backwards before pIn and pMatch.
+ *
+ *  We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
+static size_t ZSTD_ldm_countBackwardsMatch(
+            const BYTE* pIn, const BYTE* pAnchor,
+            const BYTE* pMatch, const BYTE* pMatchBase)
+{
+    size_t matchLength = 0;
+    while (pIn > pAnchor && pMatch > pMatchBase && pIn[-1] == pMatch[-1]) {
+        pIn--;
+        pMatch--;
+        matchLength++;
+    }
+    return matchLength;
+}
+
+/** ZSTD_ldm_countBackwardsMatch_2segments() :
+ *  Returns the number of bytes that match backwards from pMatch,
+ *  even with the backwards match spanning 2 different segments.
+ *
+ *  On reaching `pMatchBase`, start counting from mEnd */
+static size_t ZSTD_ldm_countBackwardsMatch_2segments(
+                    const BYTE* pIn, const BYTE* pAnchor,
+                    const BYTE* pMatch, const BYTE* pMatchBase,
+                    const BYTE* pExtDictStart, const BYTE* pExtDictEnd)
+{
+    size_t matchLength = ZSTD_ldm_countBackwardsMatch(pIn, pAnchor, pMatch, pMatchBase);
+    if (pMatch - matchLength != pMatchBase || pMatchBase == pExtDictStart) {
+        /* If backwards match is entirely in the extDict or prefix, immediately return */
+        return matchLength;
+    }
+    DEBUGLOG(7, "ZSTD_ldm_countBackwardsMatch_2segments: found 2-parts backwards match (length in prefix==%zu)", matchLength);
+    matchLength += ZSTD_ldm_countBackwardsMatch(pIn - matchLength, pAnchor, pExtDictEnd, pExtDictStart);
+    DEBUGLOG(7, "final backwards match length = %zu", matchLength);
+    return matchLength;
+}
+
+/** ZSTD_ldm_fillFastTables() :
+ *
+ *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
+ *  This is similar to ZSTD_loadDictionaryContent.
+ *
+ *  The tables for the other strategies are filled within their
+ *  block compressors. */
+static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
+                                      void const* end)
+{
+    const BYTE* const iend = (const BYTE*)end;
+
+    switch(ms->cParams.strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
+        break;
+
+    case ZSTD_dfast:
+        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+    case ZSTD_btlazy2:
+    case ZSTD_btopt:
+    case ZSTD_btultra:
+    case ZSTD_btultra2:
+        break;
+    default:
+        assert(0);  /* not possible : not a valid strategy id */
+    }
+
+    return 0;
+}
+
+void ZSTD_ldm_fillHashTable(
+            ldmState_t* ldmState, const BYTE* ip,
+            const BYTE* iend, ldmParams_t const* params)
+{
+    U32 const minMatchLength = params->minMatchLength;
+    U32 const hBits = params->hashLog - params->bucketSizeLog;
+    BYTE const* const base = ldmState->window.base;
+    BYTE const* const istart = ip;
+    ldmRollingHashState_t hashState;
+    size_t* const splits = ldmState->splitIndices;
+    unsigned numSplits;
+
+    DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
+
+    ZSTD_ldm_gear_init(&hashState, params);
+    while (ip < iend) {
+        size_t hashed;
+        unsigned n;
+
+        numSplits = 0;
+        hashed = ZSTD_ldm_gear_feed(&hashState, ip, iend - ip, splits, &numSplits);
+
+        for (n = 0; n < numSplits; n++) {
+            if (ip + splits[n] >= istart + minMatchLength) {
+                BYTE const* const split = ip + splits[n] - minMatchLength;
+                U64 const xxhash = XXH64(split, minMatchLength, 0);
+                U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
+                ldmEntry_t entry;
+
+                entry.offset = (U32)(split - base);
+                entry.checksum = (U32)(xxhash >> 32);
+                ZSTD_ldm_insertEntry(ldmState, hash, entry, *params);
+            }
+        }
+
+        ip += hashed;
+    }
+}
+
+
+/** ZSTD_ldm_limitTableUpdate() :
+ *
+ *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
+ *  if it is far way
+ *  (after a long match, only update tables a limited amount). */
+static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
+{
+    U32 const curr = (U32)(anchor - ms->window.base);
+    if (curr > ms->nextToUpdate + 1024) {
+        ms->nextToUpdate =
+            curr - MIN(512, curr - ms->nextToUpdate - 1024);
+    }
+}
+
+static size_t ZSTD_ldm_generateSequences_internal(
+        ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    /* LDM parameters */
+    int const extDict = ZSTD_window_hasExtDict(ldmState->window);
+    U32 const minMatchLength = params->minMatchLength;
+    U32 const entsPerBucket = 1U << params->bucketSizeLog;
+    U32 const hBits = params->hashLog - params->bucketSizeLog;
+    /* Prefix and extDict parameters */
+    U32 const dictLimit = ldmState->window.dictLimit;
+    U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
+    BYTE const* const base = ldmState->window.base;
+    BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
+    BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
+    BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
+    BYTE const* const lowPrefixPtr = base + dictLimit;
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    BYTE const* const ilimit = iend - HASH_READ_SIZE;
+    /* Input positions */
+    BYTE const* anchor = istart;
+    BYTE const* ip = istart;
+    /* Rolling hash state */
+    ldmRollingHashState_t hashState;
+    /* Arrays for staged-processing */
+    size_t* const splits = ldmState->splitIndices;
+    ldmMatchCandidate_t* const candidates = ldmState->matchCandidates;
+    unsigned numSplits;
+
+    if (srcSize < minMatchLength)
+        return iend - anchor;
+
+    /* Initialize the rolling hash state with the first minMatchLength bytes */
+    ZSTD_ldm_gear_init(&hashState, params);
+    ZSTD_ldm_gear_reset(&hashState, ip, minMatchLength);
+    ip += minMatchLength;
+
+    while (ip < ilimit) {
+        size_t hashed;
+        unsigned n;
+
+        numSplits = 0;
+        hashed = ZSTD_ldm_gear_feed(&hashState, ip, ilimit - ip,
+                                    splits, &numSplits);
+
+        for (n = 0; n < numSplits; n++) {
+            BYTE const* const split = ip + splits[n] - minMatchLength;
+            U64 const xxhash = XXH64(split, minMatchLength, 0);
+            U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
+
+            candidates[n].split = split;
+            candidates[n].hash = hash;
+            candidates[n].checksum = (U32)(xxhash >> 32);
+            candidates[n].bucket = ZSTD_ldm_getBucket(ldmState, hash, *params);
+            PREFETCH_L1(candidates[n].bucket);
+        }
+
+        for (n = 0; n < numSplits; n++) {
+            size_t forwardMatchLength = 0, backwardMatchLength = 0,
+                   bestMatchLength = 0, mLength;
+            U32 offset;
+            BYTE const* const split = candidates[n].split;
+            U32 const checksum = candidates[n].checksum;
+            U32 const hash = candidates[n].hash;
+            ldmEntry_t* const bucket = candidates[n].bucket;
+            ldmEntry_t const* cur;
+            ldmEntry_t const* bestEntry = NULL;
+            ldmEntry_t newEntry;
+
+            newEntry.offset = (U32)(split - base);
+            newEntry.checksum = checksum;
+
+            /* If a split point would generate a sequence overlapping with
+             * the previous one, we merely register it in the hash table and
+             * move on */
+            if (split < anchor) {
+                ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
+                continue;
+            }
+
+            for (cur = bucket; cur < bucket + entsPerBucket; cur++) {
+                size_t curForwardMatchLength, curBackwardMatchLength,
+                       curTotalMatchLength;
+                if (cur->checksum != checksum || cur->offset <= lowestIndex) {
+                    continue;
+                }
+                if (extDict) {
+                    BYTE const* const curMatchBase =
+                        cur->offset < dictLimit ? dictBase : base;
+                    BYTE const* const pMatch = curMatchBase + cur->offset;
+                    BYTE const* const matchEnd =
+                        cur->offset < dictLimit ? dictEnd : iend;
+                    BYTE const* const lowMatchPtr =
+                        cur->offset < dictLimit ? dictStart : lowPrefixPtr;
+                    curForwardMatchLength =
+                        ZSTD_count_2segments(split, pMatch, iend, matchEnd, lowPrefixPtr);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength = ZSTD_ldm_countBackwardsMatch_2segments(
+                            split, anchor, pMatch, lowMatchPtr, dictStart, dictEnd);
+                } else { /* !extDict */
+                    BYTE const* const pMatch = base + cur->offset;
+                    curForwardMatchLength = ZSTD_count(split, pMatch, iend);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(split, anchor, pMatch, lowPrefixPtr);
+                }
+                curTotalMatchLength = curForwardMatchLength + curBackwardMatchLength;
+
+                if (curTotalMatchLength > bestMatchLength) {
+                    bestMatchLength = curTotalMatchLength;
+                    forwardMatchLength = curForwardMatchLength;
+                    backwardMatchLength = curBackwardMatchLength;
+                    bestEntry = cur;
+                }
+            }
+
+            /* No match found -- insert an entry into the hash table
+             * and process the next candidate match */
+            if (bestEntry == NULL) {
+                ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
+                continue;
+            }
+
+            /* Match found */
+            offset = (U32)(split - base) - bestEntry->offset;
+            mLength = forwardMatchLength + backwardMatchLength;
+            {
+                rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
+
+                /* Out of sequence storage */
+                if (rawSeqStore->size == rawSeqStore->capacity)
+                    return ERROR(dstSize_tooSmall);
+                seq->litLength = (U32)(split - backwardMatchLength - anchor);
+                seq->matchLength = (U32)mLength;
+                seq->offset = offset;
+                rawSeqStore->size++;
+            }
+
+            /* Insert the current entry into the hash table --- it must be
+             * done after the previous block to avoid clobbering bestEntry */
+            ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
+
+            anchor = split + forwardMatchLength;
+
+            /* If we find a match that ends after the data that we've hashed
+             * then we have a repeating, overlapping, pattern. E.g. all zeros.
+             * If one repetition of the pattern matches our `stopMask` then all
+             * repetitions will. We don't need to insert them all into out table,
+             * only the first one. So skip over overlapping matches.
+             * This is a major speed boost (20x) for compressing a single byte
+             * repeated, when that byte ends up in the table.
+             */
+            if (anchor > ip + hashed) {
+                ZSTD_ldm_gear_reset(&hashState, anchor - minMatchLength, minMatchLength);
+                /* Continue the outer loop at anchor (ip + hashed == anchor). */
+                ip = anchor - hashed;
+                break;
+            }
+        }
+
+        ip += hashed;
+    }
+
+    return iend - anchor;
+}
+
+/*! ZSTD_ldm_reduceTable() :
+ *  reduce table indexes by `reducerValue` */
+static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
+                                 U32 const reducerValue)
+{
+    U32 u;
+    for (u = 0; u < size; u++) {
+        if (table[u].offset < reducerValue) table[u].offset = 0;
+        else table[u].offset -= reducerValue;
+    }
+}
+
+size_t ZSTD_ldm_generateSequences(
+        ldmState_t* ldmState, rawSeqStore_t* sequences,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    U32 const maxDist = 1U << params->windowLog;
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    size_t const kMaxChunkSize = 1 << 20;
+    size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
+    size_t chunk;
+    size_t leftoverSize = 0;
+
+    assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
+    /* Check that ZSTD_window_update() has been called for this chunk prior
+     * to passing it to this function.
+     */
+    assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
+    /* The input could be very large (in zstdmt), so it must be broken up into
+     * chunks to enforce the maximum distance and handle overflow correction.
+     */
+    assert(sequences->pos <= sequences->size);
+    assert(sequences->size <= sequences->capacity);
+    for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
+        BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
+        size_t const remaining = (size_t)(iend - chunkStart);
+        BYTE const *const chunkEnd =
+            (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
+        size_t const chunkSize = chunkEnd - chunkStart;
+        size_t newLeftoverSize;
+        size_t const prevSize = sequences->size;
+
+        assert(chunkStart < iend);
+        /* 1. Perform overflow correction if necessary. */
+        if (ZSTD_window_needOverflowCorrection(ldmState->window, 0, maxDist, ldmState->loadedDictEnd, chunkStart, chunkEnd)) {
+            U32 const ldmHSize = 1U << params->hashLog;
+            U32 const correction = ZSTD_window_correctOverflow(
+                &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
+            ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
+            /* invalidate dictionaries on overflow correction */
+            ldmState->loadedDictEnd = 0;
+        }
+        /* 2. We enforce the maximum offset allowed.
+         *
+         * kMaxChunkSize should be small enough that we don't lose too much of
+         * the window through early invalidation.
+         * TODO: * Test the chunk size.
+         *       * Try invalidation after the sequence generation and test the
+         *         offset against maxDist directly.
+         *
+         * NOTE: Because of dictionaries + sequence splitting we MUST make sure
+         * that any offset used is valid at the END of the sequence, since it may
+         * be split into two sequences. This condition holds when using
+         * ZSTD_window_enforceMaxDist(), but if we move to checking offsets
+         * against maxDist directly, we'll have to carefully handle that case.
+         */
+        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
+        /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
+        newLeftoverSize = ZSTD_ldm_generateSequences_internal(
+            ldmState, sequences, params, chunkStart, chunkSize);
+        if (ZSTD_isError(newLeftoverSize))
+            return newLeftoverSize;
+        /* 4. We add the leftover literals from previous iterations to the first
+         *    newly generated sequence, or add the `newLeftoverSize` if none are
+         *    generated.
+         */
+        /* Prepend the leftover literals from the last call */
+        if (prevSize < sequences->size) {
+            sequences->seq[prevSize].litLength += (U32)leftoverSize;
+            leftoverSize = newLeftoverSize;
+        } else {
+            assert(newLeftoverSize == chunkSize);
+            leftoverSize += chunkSize;
+        }
+    }
+    return 0;
+}
+
+void
+ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch)
+{
+    while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
+        if (srcSize <= seq->litLength) {
+            /* Skip past srcSize literals */
+            seq->litLength -= (U32)srcSize;
+            return;
+        }
+        srcSize -= seq->litLength;
+        seq->litLength = 0;
+        if (srcSize < seq->matchLength) {
+            /* Skip past the first srcSize of the match */
+            seq->matchLength -= (U32)srcSize;
+            if (seq->matchLength < minMatch) {
+                /* The match is too short, omit it */
+                if (rawSeqStore->pos + 1 < rawSeqStore->size) {
+                    seq[1].litLength += seq[0].matchLength;
+                }
+                rawSeqStore->pos++;
+            }
+            return;
+        }
+        srcSize -= seq->matchLength;
+        seq->matchLength = 0;
+        rawSeqStore->pos++;
+    }
+}
+
+/**
+ * If the sequence length is longer than remaining then the sequence is split
+ * between this block and the next.
+ *
+ * Returns the current sequence to handle, or if the rest of the block should
+ * be literals, it returns a sequence with offset == 0.
+ */
+static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
+                                 U32 const remaining, U32 const minMatch)
+{
+    rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
+    assert(sequence.offset > 0);
+    /* Likely: No partial sequence */
+    if (remaining >= sequence.litLength + sequence.matchLength) {
+        rawSeqStore->pos++;
+        return sequence;
+    }
+    /* Cut the sequence short (offset == 0 ==> rest is literals). */
+    if (remaining <= sequence.litLength) {
+        sequence.offset = 0;
+    } else if (remaining < sequence.litLength + sequence.matchLength) {
+        sequence.matchLength = remaining - sequence.litLength;
+        if (sequence.matchLength < minMatch) {
+            sequence.offset = 0;
+        }
+    }
+    /* Skip past `remaining` bytes for the future sequences. */
+    ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
+    return sequence;
+}
+
+void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
+    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
+    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
+        if (currPos >= currSeq.litLength + currSeq.matchLength) {
+            currPos -= currSeq.litLength + currSeq.matchLength;
+            rawSeqStore->pos++;
+        } else {
+            rawSeqStore->posInSequence = currPos;
+            break;
+        }
+    }
+    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
+        rawSeqStore->posInSequence = 0;
+    }
+}
+
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+    ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+    ZSTD_paramSwitch_e useRowMatchFinder,
+    void const* src, size_t srcSize)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    unsigned const minMatch = cParams->minMatch;
+    ZSTD_blockCompressor const blockCompressor =
+        ZSTD_selectBlockCompressor(cParams->strategy, useRowMatchFinder, ZSTD_matchState_dictMode(ms));
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    /* Input positions */
+    BYTE const* ip = istart;
+
+    DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
+    /* If using opt parser, use LDMs only as candidates rather than always accepting them */
+    if (cParams->strategy >= ZSTD_btopt) {
+        size_t lastLLSize;
+        ms->ldmSeqStore = rawSeqStore;
+        lastLLSize = blockCompressor(ms, seqStore, rep, src, srcSize);
+        ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore, srcSize);
+        return lastLLSize;
+    }
+
+    assert(rawSeqStore->pos <= rawSeqStore->size);
+    assert(rawSeqStore->size <= rawSeqStore->capacity);
+    /* Loop through each sequence and apply the block compressor to the literals */
+    while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
+        /* maybeSplitSequence updates rawSeqStore->pos */
+        rawSeq const sequence = maybeSplitSequence(rawSeqStore,
+                                                   (U32)(iend - ip), minMatch);
+        int i;
+        /* End signal */
+        if (sequence.offset == 0)
+            break;
+
+        assert(ip + sequence.litLength + sequence.matchLength <= iend);
+
+        /* Fill tables for block compressor */
+        ZSTD_ldm_limitTableUpdate(ms, ip);
+        ZSTD_ldm_fillFastTables(ms, ip);
+        /* Run the block compressor */
+        DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
+        {
+            size_t const newLitLength =
+                blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
+            ip += sequence.litLength;
+            /* Update the repcodes */
+            for (i = ZSTD_REP_NUM - 1; i > 0; i--)
+                rep[i] = rep[i-1];
+            rep[0] = sequence.offset;
+            /* Store the sequence */
+            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
+                          OFFSET_TO_OFFBASE(sequence.offset),
+                          sequence.matchLength);
+            ip += sequence.matchLength;
+        }
+    }
+    /* Fill the tables for the block compressor */
+    ZSTD_ldm_limitTableUpdate(ms, ip);
+    ZSTD_ldm_fillFastTables(ms, ip);
+    /* Compress the last literals */
+    return blockCompressor(ms, seqStore, rep, ip, iend - ip);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm.h
new file mode 100644
index 0000000..f147021
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm.h
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LDM_H
+#define ZSTD_LDM_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"   /* ldmParams_t, U32 */
+#include "../zstd.h"   /* ZSTD_CCtx, size_t */
+
+/*-*************************************
+*  Long distance matching
+***************************************/
+
+#define ZSTD_LDM_DEFAULT_WINDOW_LOG ZSTD_WINDOWLOG_LIMIT_DEFAULT
+
+void ZSTD_ldm_fillHashTable(
+            ldmState_t* state, const BYTE* ip,
+            const BYTE* iend, ldmParams_t const* params);
+
+/**
+ * ZSTD_ldm_generateSequences():
+ *
+ * Generates the sequences using the long distance match finder.
+ * Generates long range matching sequences in `sequences`, which parse a prefix
+ * of the source. `sequences` must be large enough to store every sequence,
+ * which can be checked with `ZSTD_ldm_getMaxNbSeq()`.
+ * @returns 0 or an error code.
+ *
+ * NOTE: The user must have called ZSTD_window_update() for all of the input
+ * they have, even if they pass it to ZSTD_ldm_generateSequences() in chunks.
+ * NOTE: This function returns an error if it runs out of space to store
+ *       sequences.
+ */
+size_t ZSTD_ldm_generateSequences(
+            ldmState_t* ldms, rawSeqStore_t* sequences,
+            ldmParams_t const* params, void const* src, size_t srcSize);
+
+/**
+ * ZSTD_ldm_blockCompress():
+ *
+ * Compresses a block using the predefined sequences, along with a secondary
+ * block compressor. The literals section of every sequence is passed to the
+ * secondary block compressor, and those sequences are interspersed with the
+ * predefined sequences. Returns the length of the last literals.
+ * Updates `rawSeqStore.pos` to indicate how many sequences have been consumed.
+ * `rawSeqStore.seq` may also be updated to split the last sequence between two
+ * blocks.
+ * @return The length of the last literals.
+ *
+ * NOTE: The source must be at most the maximum block size, but the predefined
+ * sequences can be any size, and may be longer than the block. In the case that
+ * they are longer than the block, the last sequences may need to be split into
+ * two. We handle that case correctly, and update `rawSeqStore` appropriately.
+ * NOTE: This function does not return any errors.
+ */
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+            ZSTD_paramSwitch_e useRowMatchFinder,
+            void const* src, size_t srcSize);
+
+/**
+ * ZSTD_ldm_skipSequences():
+ *
+ * Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
+ * Avoids emitting matches less than `minMatch` bytes.
+ * Must be called for data that is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
+    U32 const minMatch);
+
+/* ZSTD_ldm_skipRawSeqStoreBytes():
+ * Moves forward in rawSeqStore by nbBytes, updating fields 'pos' and 'posInSequence'.
+ * Not to be used in conjunction with ZSTD_ldm_skipSequences().
+ * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes);
+
+/** ZSTD_ldm_getTableSize() :
+ *  Estimate the space needed for long distance matching tables or 0 if LDM is
+ *  disabled.
+ */
+size_t ZSTD_ldm_getTableSize(ldmParams_t params);
+
+/** ZSTD_ldm_getSeqSpace() :
+ *  Return an upper bound on the number of sequences that can be produced by
+ *  the long distance matcher, or 0 if LDM is disabled.
+ */
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize);
+
+/** ZSTD_ldm_adjustParameters() :
+ *  If the params->hashRateLog is not set, set it to its default value based on
+ *  windowLog and params->hashLog.
+ *
+ *  Ensures that params->bucketSizeLog is <= params->hashLog (setting it to
+ *  params->hashLog if it is not).
+ *
+ *  Ensures that the minMatchLength >= targetLength during optimal parsing.
+ */
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_FAST_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm_geartab.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm_geartab.h
new file mode 100644
index 0000000..ef34bc5
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_ldm_geartab.h
@@ -0,0 +1,106 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LDM_GEARTAB_H
+#define ZSTD_LDM_GEARTAB_H
+
+#include "../common/compiler.h" /* UNUSED_ATTR */
+#include "../common/mem.h"      /* U64 */
+
+static UNUSED_ATTR const U64 ZSTD_ldm_gearTab[256] = {
+    0xf5b8f72c5f77775c, 0x84935f266b7ac412, 0xb647ada9ca730ccc,
+    0xb065bb4b114fb1de, 0x34584e7e8c3a9fd0, 0x4e97e17c6ae26b05,
+    0x3a03d743bc99a604, 0xcecd042422c4044f, 0x76de76c58524259e,
+    0x9c8528f65badeaca, 0x86563706e2097529, 0x2902475fa375d889,
+    0xafb32a9739a5ebe6, 0xce2714da3883e639, 0x21eaf821722e69e,
+    0x37b628620b628,    0x49a8d455d88caf5,  0x8556d711e6958140,
+    0x4f7ae74fc605c1f,  0x829f0c3468bd3a20, 0x4ffdc885c625179e,
+    0x8473de048a3daf1b, 0x51008822b05646b2, 0x69d75d12b2d1cc5f,
+    0x8c9d4a19159154bc, 0xc3cc10f4abbd4003, 0xd06ddc1cecb97391,
+    0xbe48e6e7ed80302e, 0x3481db31cee03547, 0xacc3f67cdaa1d210,
+    0x65cb771d8c7f96cc, 0x8eb27177055723dd, 0xc789950d44cd94be,
+    0x934feadc3700b12b, 0x5e485f11edbdf182, 0x1e2e2a46fd64767a,
+    0x2969ca71d82efa7c, 0x9d46e9935ebbba2e, 0xe056b67e05e6822b,
+    0x94d73f55739d03a0, 0xcd7010bdb69b5a03, 0x455ef9fcd79b82f4,
+    0x869cb54a8749c161, 0x38d1a4fa6185d225, 0xb475166f94bbe9bb,
+    0xa4143548720959f1, 0x7aed4780ba6b26ba, 0xd0ce264439e02312,
+    0x84366d746078d508, 0xa8ce973c72ed17be, 0x21c323a29a430b01,
+    0x9962d617e3af80ee, 0xab0ce91d9c8cf75b, 0x530e8ee6d19a4dbc,
+    0x2ef68c0cf53f5d72, 0xc03a681640a85506, 0x496e4e9f9c310967,
+    0x78580472b59b14a0, 0x273824c23b388577, 0x66bf923ad45cb553,
+    0x47ae1a5a2492ba86, 0x35e304569e229659, 0x4765182a46870b6f,
+    0x6cbab625e9099412, 0xddac9a2e598522c1, 0x7172086e666624f2,
+    0xdf5003ca503b7837, 0x88c0c1db78563d09, 0x58d51865acfc289d,
+    0x177671aec65224f1, 0xfb79d8a241e967d7, 0x2be1e101cad9a49a,
+    0x6625682f6e29186b, 0x399553457ac06e50, 0x35dffb4c23abb74,
+    0x429db2591f54aade, 0xc52802a8037d1009, 0x6acb27381f0b25f3,
+    0xf45e2551ee4f823b, 0x8b0ea2d99580c2f7, 0x3bed519cbcb4e1e1,
+    0xff452823dbb010a,  0x9d42ed614f3dd267, 0x5b9313c06257c57b,
+    0xa114b8008b5e1442, 0xc1fe311c11c13d4b, 0x66e8763ea34c5568,
+    0x8b982af1c262f05d, 0xee8876faaa75fbb7, 0x8a62a4d0d172bb2a,
+    0xc13d94a3b7449a97, 0x6dbbba9dc15d037c, 0xc786101f1d92e0f1,
+    0xd78681a907a0b79b, 0xf61aaf2962c9abb9, 0x2cfd16fcd3cb7ad9,
+    0x868c5b6744624d21, 0x25e650899c74ddd7, 0xba042af4a7c37463,
+    0x4eb1a539465a3eca, 0xbe09dbf03b05d5ca, 0x774e5a362b5472ba,
+    0x47a1221229d183cd, 0x504b0ca18ef5a2df, 0xdffbdfbde2456eb9,
+    0x46cd2b2fbee34634, 0xf2aef8fe819d98c3, 0x357f5276d4599d61,
+    0x24a5483879c453e3, 0x88026889192b4b9,  0x28da96671782dbec,
+    0x4ef37c40588e9aaa, 0x8837b90651bc9fb3, 0xc164f741d3f0e5d6,
+    0xbc135a0a704b70ba, 0x69cd868f7622ada,  0xbc37ba89e0b9c0ab,
+    0x47c14a01323552f6, 0x4f00794bacee98bb, 0x7107de7d637a69d5,
+    0x88af793bb6f2255e, 0xf3c6466b8799b598, 0xc288c616aa7f3b59,
+    0x81ca63cf42fca3fd, 0x88d85ace36a2674b, 0xd056bd3792389e7,
+    0xe55c396c4e9dd32d, 0xbefb504571e6c0a6, 0x96ab32115e91e8cc,
+    0xbf8acb18de8f38d1, 0x66dae58801672606, 0x833b6017872317fb,
+    0xb87c16f2d1c92864, 0xdb766a74e58b669c, 0x89659f85c61417be,
+    0xc8daad856011ea0c, 0x76a4b565b6fe7eae, 0xa469d085f6237312,
+    0xaaf0365683a3e96c, 0x4dbb746f8424f7b8, 0x638755af4e4acc1,
+    0x3d7807f5bde64486, 0x17be6d8f5bbb7639, 0x903f0cd44dc35dc,
+    0x67b672eafdf1196c, 0xa676ff93ed4c82f1, 0x521d1004c5053d9d,
+    0x37ba9ad09ccc9202, 0x84e54d297aacfb51, 0xa0b4b776a143445,
+    0x820d471e20b348e,  0x1874383cb83d46dc, 0x97edeec7a1efe11c,
+    0xb330e50b1bdc42aa, 0x1dd91955ce70e032, 0xa514cdb88f2939d5,
+    0x2791233fd90db9d3, 0x7b670a4cc50f7a9b, 0x77c07d2a05c6dfa5,
+    0xe3778b6646d0a6fa, 0xb39c8eda47b56749, 0x933ed448addbef28,
+    0xaf846af6ab7d0bf4, 0xe5af208eb666e49,  0x5e6622f73534cd6a,
+    0x297daeca42ef5b6e, 0x862daef3d35539a6, 0xe68722498f8e1ea9,
+    0x981c53093dc0d572, 0xfa09b0bfbf86fbf5, 0x30b1e96166219f15,
+    0x70e7d466bdc4fb83, 0x5a66736e35f2a8e9, 0xcddb59d2b7c1baef,
+    0xd6c7d247d26d8996, 0xea4e39eac8de1ba3, 0x539c8bb19fa3aff2,
+    0x9f90e4c5fd508d8,  0xa34e5956fbaf3385, 0x2e2f8e151d3ef375,
+    0x173691e9b83faec1, 0xb85a8d56bf016379, 0x8382381267408ae3,
+    0xb90f901bbdc0096d, 0x7c6ad32933bcec65, 0x76bb5e2f2c8ad595,
+    0x390f851a6cf46d28, 0xc3e6064da1c2da72, 0xc52a0c101cfa5389,
+    0xd78eaf84a3fbc530, 0x3781b9e2288b997e, 0x73c2f6dea83d05c4,
+    0x4228e364c5b5ed7,  0x9d7a3edf0da43911, 0x8edcfeda24686756,
+    0x5e7667a7b7a9b3a1, 0x4c4f389fa143791d, 0xb08bc1023da7cddc,
+    0x7ab4be3ae529b1cc, 0x754e6132dbe74ff9, 0x71635442a839df45,
+    0x2f6fb1643fbe52de, 0x961e0a42cf7a8177, 0xf3b45d83d89ef2ea,
+    0xee3de4cf4a6e3e9b, 0xcd6848542c3295e7, 0xe4cee1664c78662f,
+    0x9947548b474c68c4, 0x25d73777a5ed8b0b, 0xc915b1d636b7fc,
+    0x21c2ba75d9b0d2da, 0x5f6b5dcf608a64a1, 0xdcf333255ff9570c,
+    0x633b922418ced4ee, 0xc136dde0b004b34a, 0x58cc83b05d4b2f5a,
+    0x5eb424dda28e42d2, 0x62df47369739cd98, 0xb4e0b42485e4ce17,
+    0x16e1f0c1f9a8d1e7, 0x8ec3916707560ebf, 0x62ba6e2df2cc9db3,
+    0xcbf9f4ff77d83a16, 0x78d9d7d07d2bbcc4, 0xef554ce1e02c41f4,
+    0x8d7581127eccf94d, 0xa9b53336cb3c8a05, 0x38c42c0bf45c4f91,
+    0x640893cdf4488863, 0x80ec34bc575ea568, 0x39f324f5b48eaa40,
+    0xe9d9ed1f8eff527f, 0x9224fc058cc5a214, 0xbaba00b04cfe7741,
+    0x309a9f120fcf52af, 0xa558f3ec65626212, 0x424bec8b7adabe2f,
+    0x41622513a6aea433, 0xb88da2d5324ca798, 0xd287733b245528a4,
+    0x9a44697e6d68aec3, 0x7b1093be2f49bb28, 0x50bbec632e3d8aad,
+    0x6cd90723e1ea8283, 0x897b9e7431b02bf3, 0x219efdcb338a7047,
+    0x3b0311f0a27c0656, 0xdb17bf91c0db96e7, 0x8cd4fd6b4e85a5b2,
+    0xfab071054ba6409d, 0x40d6fe831fa9dfd9, 0xaf358debad7d791e,
+    0xeb8d0e25a65e3e58, 0xbbcbd3df14e08580, 0xcf751f27ecdab2b,
+    0x2b4da14f2613d8f4
+};
+
+#endif /* ZSTD_LDM_GEARTAB_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_opt.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_opt.c
new file mode 100644
index 0000000..f02a760
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_opt.c
@@ -0,0 +1,1472 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "hist.h"
+#include "zstd_opt.h"
+
+
+#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
+#define ZSTD_MAX_PRICE     (1<<30)
+
+#define ZSTD_PREDEF_THRESHOLD 8   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
+
+
+/*-*************************************
+*  Price functions for optimal parser
+***************************************/
+
+#if 0    /* approximation at bit level (for tests) */
+#  define BITCOST_ACCURACY 0
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat, opt) ((void)(opt), ZSTD_bitWeight(stat))
+#elif 0  /* fractional bit accuracy (for tests) */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) ((void)(opt), ZSTD_fracWeight(stat))
+#else    /* opt==approx, ultra==accurate */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) ((opt) ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
+#endif
+
+/* ZSTD_bitWeight() :
+ * provide estimated "cost" of a stat in full bits only */
+MEM_STATIC U32 ZSTD_bitWeight(U32 stat)
+{
+    return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);
+}
+
+/* ZSTD_fracWeight() :
+ * provide fractional-bit "cost" of a stat,
+ * using linear interpolation approximation */
+MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
+{
+    U32 const stat = rawStat + 1;
+    U32 const hb = ZSTD_highbit32(stat);
+    U32 const BWeight = hb * BITCOST_MULTIPLIER;
+    /* Fweight was meant for "Fractional weight"
+     * but it's effectively a value between 1 and 2
+     * using fixed point arithmetic */
+    U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;
+    U32 const weight = BWeight + FWeight;
+    assert(hb + BITCOST_ACCURACY < 31);
+    return weight;
+}
+
+#if (DEBUGLEVEL>=2)
+/* debugging function,
+ * @return price in bytes as fractional value
+ * for debug messages only */
+MEM_STATIC double ZSTD_fCost(int price)
+{
+    return (double)price / (BITCOST_MULTIPLIER*8);
+}
+#endif
+
+static int ZSTD_compressedLiterals(optState_t const* const optPtr)
+{
+    return optPtr->literalCompressionMode != ZSTD_ps_disable;
+}
+
+static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
+{
+    if (ZSTD_compressedLiterals(optPtr))
+        optPtr->litSumBasePrice = WEIGHT(optPtr->litSum, optLevel);
+    optPtr->litLengthSumBasePrice = WEIGHT(optPtr->litLengthSum, optLevel);
+    optPtr->matchLengthSumBasePrice = WEIGHT(optPtr->matchLengthSum, optLevel);
+    optPtr->offCodeSumBasePrice = WEIGHT(optPtr->offCodeSum, optLevel);
+}
+
+
+static U32 sum_u32(const unsigned table[], size_t nbElts)
+{
+    size_t n;
+    U32 total = 0;
+    for (n=0; n<nbElts; n++) {
+        total += table[n];
+    }
+    return total;
+}
+
+typedef enum { base_0possible=0, base_1guaranteed=1 } base_directive_e;
+
+static U32
+ZSTD_downscaleStats(unsigned* table, U32 lastEltIndex, U32 shift, base_directive_e base1)
+{
+    U32 s, sum=0;
+    DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)",
+            (unsigned)lastEltIndex+1, (unsigned)shift );
+    assert(shift < 30);
+    for (s=0; s<lastEltIndex+1; s++) {
+        unsigned const base = base1 ? 1 : (table[s]>0);
+        unsigned const newStat = base + (table[s] >> shift);
+        sum += newStat;
+        table[s] = newStat;
+    }
+    return sum;
+}
+
+/* ZSTD_scaleStats() :
+ * reduce all elt frequencies in table if sum too large
+ * return the resulting sum of elements */
+static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
+{
+    U32 const prevsum = sum_u32(table, lastEltIndex+1);
+    U32 const factor = prevsum >> logTarget;
+    DEBUGLOG(5, "ZSTD_scaleStats (nbElts=%u, target=%u)", (unsigned)lastEltIndex+1, (unsigned)logTarget);
+    assert(logTarget < 30);
+    if (factor <= 1) return prevsum;
+    return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor), base_1guaranteed);
+}
+
+/* ZSTD_rescaleFreqs() :
+ * if first block (detected by optPtr->litLengthSum == 0) : init statistics
+ *    take hints from dictionary if there is one
+ *    and init from zero if there is none,
+ *    using src for literals stats, and baseline stats for sequence symbols
+ * otherwise downscale existing stats, to be used as seed for next block.
+ */
+static void
+ZSTD_rescaleFreqs(optState_t* const optPtr,
+            const BYTE* const src, size_t const srcSize,
+                  int const optLevel)
+{
+    int const compressedLiterals = ZSTD_compressedLiterals(optPtr);
+    DEBUGLOG(5, "ZSTD_rescaleFreqs (srcSize=%u)", (unsigned)srcSize);
+    optPtr->priceType = zop_dynamic;
+
+    if (optPtr->litLengthSum == 0) {  /* no literals stats collected -> first block assumed -> init */
+
+        /* heuristic: use pre-defined stats for too small inputs */
+        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {
+            DEBUGLOG(5, "srcSize <= %i : use predefined stats", ZSTD_PREDEF_THRESHOLD);
+            optPtr->priceType = zop_predef;
+        }
+
+        assert(optPtr->symbolCosts != NULL);
+        if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) {
+
+            /* huffman stats covering the full value set : table presumed generated by dictionary */
+            optPtr->priceType = zop_dynamic;
+
+            if (compressedLiterals) {
+                /* generate literals statistics from huffman table */
+                unsigned lit;
+                assert(optPtr->litFreq != NULL);
+                optPtr->litSum = 0;
+                for (lit=0; lit<=MaxLit; lit++) {
+                    U32 const scaleLog = 11;   /* scale to 2K */
+                    U32 const bitCost = HUF_getNbBitsFromCTable(optPtr->symbolCosts->huf.CTable, lit);
+                    assert(bitCost <= scaleLog);
+                    optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litSum += optPtr->litFreq[lit];
+            }   }
+
+            {   unsigned ll;
+                FSE_CState_t llstate;
+                FSE_initCState(&llstate, optPtr->symbolCosts->fse.litlengthCTable);
+                optPtr->litLengthSum = 0;
+                for (ll=0; ll<=MaxLL; ll++) {
+                    U32 const scaleLog = 10;   /* scale to 1K */
+                    U32 const bitCost = FSE_getMaxNbBits(llstate.symbolTT, ll);
+                    assert(bitCost < scaleLog);
+                    optPtr->litLengthFreq[ll] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litLengthSum += optPtr->litLengthFreq[ll];
+            }   }
+
+            {   unsigned ml;
+                FSE_CState_t mlstate;
+                FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);
+                optPtr->matchLengthSum = 0;
+                for (ml=0; ml<=MaxML; ml++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(mlstate.symbolTT, ml);
+                    assert(bitCost < scaleLog);
+                    optPtr->matchLengthFreq[ml] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->matchLengthSum += optPtr->matchLengthFreq[ml];
+            }   }
+
+            {   unsigned of;
+                FSE_CState_t ofstate;
+                FSE_initCState(&ofstate, optPtr->symbolCosts->fse.offcodeCTable);
+                optPtr->offCodeSum = 0;
+                for (of=0; of<=MaxOff; of++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(ofstate.symbolTT, of);
+                    assert(bitCost < scaleLog);
+                    optPtr->offCodeFreq[of] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->offCodeSum += optPtr->offCodeFreq[of];
+            }   }
+
+        } else {  /* first block, no dictionary */
+
+            assert(optPtr->litFreq != NULL);
+            if (compressedLiterals) {
+                /* base initial cost of literals on direct frequency within src */
+                unsigned lit = MaxLit;
+                HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
+                optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8, base_0possible);
+            }
+
+            {   unsigned const baseLLfreqs[MaxLL+1] = {
+                    4, 2, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1
+                };
+                ZSTD_memcpy(optPtr->litLengthFreq, baseLLfreqs, sizeof(baseLLfreqs));
+                optPtr->litLengthSum = sum_u32(baseLLfreqs, MaxLL+1);
+            }
+
+            {   unsigned ml;
+                for (ml=0; ml<=MaxML; ml++)
+                    optPtr->matchLengthFreq[ml] = 1;
+            }
+            optPtr->matchLengthSum = MaxML+1;
+
+            {   unsigned const baseOFCfreqs[MaxOff+1] = {
+                    6, 2, 1, 1, 2, 3, 4, 4,
+                    4, 3, 2, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1
+                };
+                ZSTD_memcpy(optPtr->offCodeFreq, baseOFCfreqs, sizeof(baseOFCfreqs));
+                optPtr->offCodeSum = sum_u32(baseOFCfreqs, MaxOff+1);
+            }
+
+        }
+
+    } else {   /* new block : scale down accumulated statistics */
+
+        if (compressedLiterals)
+            optPtr->litSum = ZSTD_scaleStats(optPtr->litFreq, MaxLit, 12);
+        optPtr->litLengthSum = ZSTD_scaleStats(optPtr->litLengthFreq, MaxLL, 11);
+        optPtr->matchLengthSum = ZSTD_scaleStats(optPtr->matchLengthFreq, MaxML, 11);
+        optPtr->offCodeSum = ZSTD_scaleStats(optPtr->offCodeFreq, MaxOff, 11);
+    }
+
+    ZSTD_setBasePrices(optPtr, optLevel);
+}
+
+/* ZSTD_rawLiteralsCost() :
+ * price of literals (only) in specified segment (which length can be 0).
+ * does not include price of literalLength symbol */
+static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
+                                const optState_t* const optPtr,
+                                int optLevel)
+{
+    if (litLength == 0) return 0;
+
+    if (!ZSTD_compressedLiterals(optPtr))
+        return (litLength << 3) * BITCOST_MULTIPLIER;  /* Uncompressed - 8 bytes per literal. */
+
+    if (optPtr->priceType == zop_predef)
+        return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */
+
+    /* dynamic statistics */
+    {   U32 price = optPtr->litSumBasePrice * litLength;
+        U32 const litPriceMax = optPtr->litSumBasePrice - BITCOST_MULTIPLIER;
+        U32 u;
+        assert(optPtr->litSumBasePrice >= BITCOST_MULTIPLIER);
+        for (u=0; u < litLength; u++) {
+            U32 litPrice = WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+            if (UNLIKELY(litPrice > litPriceMax)) litPrice = litPriceMax;
+            price -= litPrice;
+        }
+        return price;
+    }
+}
+
+/* ZSTD_litLengthPrice() :
+ * cost of literalLength symbol */
+static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr, int optLevel)
+{
+    assert(litLength <= ZSTD_BLOCKSIZE_MAX);
+    if (optPtr->priceType == zop_predef)
+        return WEIGHT(litLength, optLevel);
+
+    /* ZSTD_LLcode() can't compute litLength price for sizes >= ZSTD_BLOCKSIZE_MAX
+     * because it isn't representable in the zstd format.
+     * So instead just pretend it would cost 1 bit more than ZSTD_BLOCKSIZE_MAX - 1.
+     * In such a case, the block would be all literals.
+     */
+    if (litLength == ZSTD_BLOCKSIZE_MAX)
+        return BITCOST_MULTIPLIER + ZSTD_litLengthPrice(ZSTD_BLOCKSIZE_MAX - 1, optPtr, optLevel);
+
+    /* dynamic statistics */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        return (LL_bits[llCode] * BITCOST_MULTIPLIER)
+             + optPtr->litLengthSumBasePrice
+             - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
+    }
+}
+
+/* ZSTD_getMatchPrice() :
+ * Provides the cost of the match part (offset + matchLength) of a sequence.
+ * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
+ * @offBase : sumtype, representing an offset or a repcode, and using numeric representation of ZSTD_storeSeq()
+ * @optLevel: when <2, favors small offset for decompression speed (improved cache efficiency)
+ */
+FORCE_INLINE_TEMPLATE U32
+ZSTD_getMatchPrice(U32 const offBase,
+                   U32 const matchLength,
+             const optState_t* const optPtr,
+                   int const optLevel)
+{
+    U32 price;
+    U32 const offCode = ZSTD_highbit32(offBase);
+    U32 const mlBase = matchLength - MINMATCH;
+    assert(matchLength >= MINMATCH);
+
+    if (optPtr->priceType == zop_predef)  /* fixed scheme, does not use statistics */
+        return WEIGHT(mlBase, optLevel)
+             + ((16 + offCode) * BITCOST_MULTIPLIER); /* emulated offset cost */
+
+    /* dynamic statistics */
+    price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));
+    if ((optLevel<2) /*static*/ && offCode >= 20)
+        price += (offCode-19)*2 * BITCOST_MULTIPLIER; /* handicap for long distance offsets, favor decompression speed */
+
+    /* match Length */
+    {   U32 const mlCode = ZSTD_MLcode(mlBase);
+        price += (ML_bits[mlCode] * BITCOST_MULTIPLIER) + (optPtr->matchLengthSumBasePrice - WEIGHT(optPtr->matchLengthFreq[mlCode], optLevel));
+    }
+
+    price += BITCOST_MULTIPLIER / 5;   /* heuristic : make matches a bit more costly to favor less sequences -> faster decompression speed */
+
+    DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);
+    return price;
+}
+
+/* ZSTD_updateStats() :
+ * assumption : literals + litLength <= iend */
+static void ZSTD_updateStats(optState_t* const optPtr,
+                             U32 litLength, const BYTE* literals,
+                             U32 offBase, U32 matchLength)
+{
+    /* literals */
+    if (ZSTD_compressedLiterals(optPtr)) {
+        U32 u;
+        for (u=0; u < litLength; u++)
+            optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+        optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
+    }
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        optPtr->litLengthFreq[llCode]++;
+        optPtr->litLengthSum++;
+    }
+
+    /* offset code : follows storeSeq() numeric representation */
+    {   U32 const offCode = ZSTD_highbit32(offBase);
+        assert(offCode <= MaxOff);
+        optPtr->offCodeFreq[offCode]++;
+        optPtr->offCodeSum++;
+    }
+
+    /* match Length */
+    {   U32 const mlBase = matchLength - MINMATCH;
+        U32 const mlCode = ZSTD_MLcode(mlBase);
+        optPtr->matchLengthFreq[mlCode]++;
+        optPtr->matchLengthSum++;
+    }
+}
+
+
+/* ZSTD_readMINMATCH() :
+ * function safe only for comparisons
+ * assumption : memPtr must be at least 4 bytes before end of buffer */
+MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
+{
+    switch (length)
+    {
+    default :
+    case 4 : return MEM_read32(memPtr);
+    case 3 : if (MEM_isLittleEndian())
+                return MEM_read32(memPtr)<<8;
+             else
+                return MEM_read32(memPtr)>>8;
+    }
+}
+
+
+/* Update hashTable3 up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+static U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_matchState_t* ms,
+                                              U32* nextToUpdate3,
+                                              const BYTE* const ip)
+{
+    U32* const hashTable3 = ms->hashTable3;
+    U32 const hashLog3 = ms->hashLog3;
+    const BYTE* const base = ms->window.base;
+    U32 idx = *nextToUpdate3;
+    U32 const target = (U32)(ip - base);
+    size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
+    assert(hashLog3 > 0);
+
+    while(idx < target) {
+        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
+        idx++;
+    }
+
+    *nextToUpdate3 = target;
+    return hashTable3[hash3];
+}
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+/** ZSTD_insertBt1() : add one or multiple positions to tree.
+ * @param ip assumed <= iend-8 .
+ * @param target The target of ZSTD_updateTree_internal() - we are filling to this position
+ * @return : nb of positions added */
+static U32 ZSTD_insertBt1(
+                const ZSTD_matchState_t* ms,
+                const BYTE* const ip, const BYTE* const iend,
+                U32 const target,
+                U32 const mls, const int extDict)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32 matchIndex = hashTable[h];
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    const U32 curr = (U32)(ip-base);
+    const U32 btLow = btMask >= curr ? 0 : curr - btMask;
+    U32* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 dummy32;   /* to be nullified at the end */
+    /* windowLow is based on target because
+     * we only need positions that will be in the window at the end of the tree update.
+     */
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, target, cParams->windowLog);
+    U32 matchEndIdx = curr+8+1;
+    size_t bestLength = 8;
+    U32 nbCompares = 1U << cParams->searchLog;
+#ifdef ZSTD_C_PREDICT
+    U32 predictedSmall = *(bt + 2*((curr-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((curr-1)&btMask) + 1);
+    predictedSmall += (predictedSmall>0);
+    predictedLarge += (predictedLarge>0);
+#endif /* ZSTD_C_PREDICT */
+
+    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
+
+    assert(curr <= target);
+    assert(ip <= iend-8);   /* required for h calculation */
+    hashTable[h] = curr;   /* Update Hash Table */
+
+    assert(windowLow > 0);
+    for (; nbCompares && (matchIndex >= windowLow); --nbCompares) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < curr);
+
+#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
+        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
+        if (matchIndex == predictedSmall) {
+            /* no need to check length, result known */
+            *smallerPtr = matchIndex;
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
+            continue;
+        }
+        if (matchIndex == predictedLarge) {
+            *largerPtr = matchIndex;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
+            continue;
+        }
+#endif
+
+        if (!extDict || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if actually extDict */
+            match = base + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            bestLength = matchLength;
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+        }
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+    {   U32 positions = 0;
+        if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+        assert(matchEndIdx > curr + 8);
+        return MAX(positions, matchEndIdx - (curr + 8));
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+void ZSTD_updateTree_internal(
+                ZSTD_matchState_t* ms,
+                const BYTE* const ip, const BYTE* const iend,
+                const U32 mls, const ZSTD_dictMode_e dictMode)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+    DEBUGLOG(6, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
+                idx, target, dictMode);
+
+    while(idx < target) {
+        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, target, mls, dictMode == ZSTD_extDict);
+        assert(idx < (U32)(idx + forward));
+        idx += forward;
+    }
+    assert((size_t)(ip - base) <= (size_t)(U32)(-1));
+    assert((size_t)(iend - base) <= (size_t)(U32)(-1));
+    ms->nextToUpdate = target;
+}
+
+void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend) {
+    ZSTD_updateTree_internal(ms, ip, iend, ms->cParams.minMatch, ZSTD_noDict);
+}
+
+FORCE_INLINE_TEMPLATE U32
+ZSTD_insertBtAndGetAllMatches (
+                ZSTD_match_t* matches,  /* store result (found matches) in this table (presumed large enough) */
+                ZSTD_matchState_t* ms,
+                U32* nextToUpdate3,
+                const BYTE* const ip, const BYTE* const iLimit,
+                const ZSTD_dictMode_e dictMode,
+                const U32 rep[ZSTD_REP_NUM],
+                const U32 ll0,  /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
+                const U32 lengthToBeat,
+                const U32 mls /* template */)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    const BYTE* const base = ms->window.base;
+    U32 const curr = (U32)(ip-base);
+    U32 const hashLog = cParams->hashLog;
+    U32 const minMatch = (mls==3) ? 3 : 4;
+    U32* const hashTable = ms->hashTable;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32 matchIndex  = hashTable[h];
+    U32* const bt   = ms->chainTable;
+    U32 const btLog = cParams->chainLog - 1;
+    U32 const btMask= (1U << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const dictBase = ms->window.dictBase;
+    U32 const dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    U32 const btLow = (btMask >= curr) ? 0 : curr - btMask;
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
+    U32 const matchLow = windowLow ? windowLow : 1;
+    U32* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = bt + 2*(curr&btMask) + 1;
+    U32 matchEndIdx = curr+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
+    U32 dummy32;   /* to be nullified at the end */
+    U32 mnum = 0;
+    U32 nbCompares = 1U << cParams->searchLog;
+
+    const ZSTD_matchState_t* dms    = dictMode == ZSTD_dictMatchState ? ms->dictMatchState : NULL;
+    const ZSTD_compressionParameters* const dmsCParams =
+                                      dictMode == ZSTD_dictMatchState ? &dms->cParams : NULL;
+    const BYTE* const dmsBase       = dictMode == ZSTD_dictMatchState ? dms->window.base : NULL;
+    const BYTE* const dmsEnd        = dictMode == ZSTD_dictMatchState ? dms->window.nextSrc : NULL;
+    U32         const dmsHighLimit  = dictMode == ZSTD_dictMatchState ? (U32)(dmsEnd - dmsBase) : 0;
+    U32         const dmsLowLimit   = dictMode == ZSTD_dictMatchState ? dms->window.lowLimit : 0;
+    U32         const dmsIndexDelta = dictMode == ZSTD_dictMatchState ? windowLow - dmsHighLimit : 0;
+    U32         const dmsHashLog    = dictMode == ZSTD_dictMatchState ? dmsCParams->hashLog : hashLog;
+    U32         const dmsBtLog      = dictMode == ZSTD_dictMatchState ? dmsCParams->chainLog - 1 : btLog;
+    U32         const dmsBtMask     = dictMode == ZSTD_dictMatchState ? (1U << dmsBtLog) - 1 : 0;
+    U32         const dmsBtLow      = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
+
+    size_t bestLength = lengthToBeat-1;
+    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", curr);
+
+    /* check repCode */
+    assert(ll0 <= 1);   /* necessarily 1 or 0 */
+    {   U32 const lastR = ZSTD_REP_NUM + ll0;
+        U32 repCode;
+        for (repCode = ll0; repCode < lastR; repCode++) {
+            U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            U32 const repIndex = curr - repOffset;
+            U32 repLen = 0;
+            assert(curr >= dictLimit);
+            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < curr-dictLimit) {  /* equivalent to `curr > repIndex >= dictLimit` */
+                /* We must validate the repcode offset because when we're using a dictionary the
+                 * valid offset range shrinks when the dictionary goes out of bounds.
+                 */
+                if ((repIndex >= windowLow) & (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch))) {
+                    repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
+                }
+            } else {  /* repIndex < dictLimit || repIndex >= curr */
+                const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
+                                             dmsBase + repIndex - dmsIndexDelta :
+                                             dictBase + repIndex;
+                assert(curr >= windowLow);
+                if ( dictMode == ZSTD_extDict
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - windowLow)  /* equivalent to `curr > repIndex >= windowLow` */
+                     & (((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */)
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
+                }
+                if (dictMode == ZSTD_dictMatchState
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `curr > repIndex >= dmsLowLimit` */
+                     & ((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dmsEnd, prefixStart) + minMatch;
+            }   }
+            /* save longer solution */
+            if (repLen > bestLength) {
+                DEBUGLOG(8, "found repCode %u (ll0:%u, offset:%u) of length %u",
+                            repCode, ll0, repOffset, repLen);
+                bestLength = repLen;
+                matches[mnum].off = REPCODE_TO_OFFBASE(repCode - ll0 + 1);  /* expect value between 1 and 3 */
+                matches[mnum].len = (U32)repLen;
+                mnum++;
+                if ( (repLen > sufficient_len)
+                   | (ip+repLen == iLimit) ) {  /* best possible */
+                    return mnum;
+    }   }   }   }
+
+    /* HC3 match finder */
+    if ((mls == 3) /*static*/ && (bestLength < mls)) {
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
+        if ((matchIndex3 >= matchLow)
+          & (curr - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
+            size_t mlen;
+            if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
+                const BYTE* const match = base + matchIndex3;
+                mlen = ZSTD_count(ip, match, iLimit);
+            } else {
+                const BYTE* const match = dictBase + matchIndex3;
+                mlen = ZSTD_count_2segments(ip, match, iLimit, dictEnd, prefixStart);
+            }
+
+            /* save best solution */
+            if (mlen >= mls /* == 3 > bestLength */) {
+                DEBUGLOG(8, "found small match with hlog3, of length %u",
+                            (U32)mlen);
+                bestLength = mlen;
+                assert(curr > matchIndex3);
+                assert(mnum==0);  /* no prior solution */
+                matches[0].off = OFFSET_TO_OFFBASE(curr - matchIndex3);
+                matches[0].len = (U32)mlen;
+                mnum = 1;
+                if ( (mlen > sufficient_len) |
+                     (ip+mlen == iLimit) ) {  /* best possible length */
+                    ms->nextToUpdate = curr+1;  /* skip insertion */
+                    return 1;
+        }   }   }
+        /* no dictMatchState lookup: dicts don't have a populated HC3 table */
+    }  /* if (mls == 3) */
+
+    hashTable[h] = curr;   /* Update Hash Table */
+
+    for (; nbCompares && (matchIndex >= matchLow); --nbCompares) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        const BYTE* match;
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(curr > matchIndex);
+
+        if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
+            match = base + matchIndex;
+            if (matchIndex >= dictLimit) assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
+        } else {
+            match = dictBase + matchIndex;
+            assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* prepare for match[matchLength] read */
+        }
+
+        if (matchLength > bestLength) {
+            DEBUGLOG(8, "found match of length %u at distance %u (offBase=%u)",
+                    (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
+            assert(matchEndIdx > matchIndex);
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+            bestLength = matchLength;
+            matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
+            matches[mnum].len = (U32)matchLength;
+            mnum++;
+            if ( (matchLength > ZSTD_OPT_NUM)
+               | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                if (dictMode == ZSTD_dictMatchState) nbCompares = 0; /* break should also skip searching dms */
+                break; /* drop, to preserve bt consistency (miss a little bit of compression) */
+        }   }
+
+        if (match[matchLength] < ip[matchLength]) {
+            /* match smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new candidate => larger than match, which was smaller than current */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous, closer to current */
+        } else {
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+
+    assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+    if (dictMode == ZSTD_dictMatchState && nbCompares) {
+        size_t const dmsH = ZSTD_hashPtr(ip, dmsHashLog, mls);
+        U32 dictMatchIndex = dms->hashTable[dmsH];
+        const U32* const dmsBt = dms->chainTable;
+        commonLengthSmaller = commonLengthLarger = 0;
+        for (; nbCompares && (dictMatchIndex > dmsLowLimit); --nbCompares) {
+            const U32* const nextPtr = dmsBt + 2*(dictMatchIndex & dmsBtMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match = dmsBase + dictMatchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dmsEnd, prefixStart);
+            if (dictMatchIndex+matchLength >= dmsHighLimit)
+                match = base + dictMatchIndex + dmsIndexDelta;   /* to prepare for next usage of match[matchLength] */
+
+            if (matchLength > bestLength) {
+                matchIndex = dictMatchIndex + dmsIndexDelta;
+                DEBUGLOG(8, "found dms match of length %u at distance %u (offBase=%u)",
+                        (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                bestLength = matchLength;
+                matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
+                matches[mnum].len = (U32)matchLength;
+                mnum++;
+                if ( (matchLength > ZSTD_OPT_NUM)
+                   | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+            }   }
+
+            if (dictMatchIndex <= dmsBtLow) { break; }   /* beyond tree size, stop the search */
+            if (match[matchLength] < ip[matchLength]) {
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                commonLengthLarger = matchLength;
+                dictMatchIndex = nextPtr[0];
+    }   }   }  /* if (dictMode == ZSTD_dictMatchState) */
+
+    assert(matchEndIdx > curr+8);
+    ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
+    return mnum;
+}
+
+typedef U32 (*ZSTD_getAllMatchesFn)(
+    ZSTD_match_t*,
+    ZSTD_matchState_t*,
+    U32*,
+    const BYTE*,
+    const BYTE*,
+    const U32 rep[ZSTD_REP_NUM],
+    U32 const ll0,
+    U32 const lengthToBeat);
+
+FORCE_INLINE_TEMPLATE U32 ZSTD_btGetAllMatches_internal(
+        ZSTD_match_t* matches,
+        ZSTD_matchState_t* ms,
+        U32* nextToUpdate3,
+        const BYTE* ip,
+        const BYTE* const iHighLimit,
+        const U32 rep[ZSTD_REP_NUM],
+        U32 const ll0,
+        U32 const lengthToBeat,
+        const ZSTD_dictMode_e dictMode,
+        const U32 mls)
+{
+    assert(BOUNDED(3, ms->cParams.minMatch, 6) == mls);
+    DEBUGLOG(8, "ZSTD_BtGetAllMatches(dictMode=%d, mls=%u)", (int)dictMode, mls);
+    if (ip < ms->window.base + ms->nextToUpdate)
+        return 0;   /* skipped area */
+    ZSTD_updateTree_internal(ms, ip, iHighLimit, mls, dictMode);
+    return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, mls);
+}
+
+#define ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls) ZSTD_btGetAllMatches_##dictMode##_##mls
+
+#define GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, mls)            \
+    static U32 ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls)(      \
+            ZSTD_match_t* matches,                             \
+            ZSTD_matchState_t* ms,                             \
+            U32* nextToUpdate3,                                \
+            const BYTE* ip,                                    \
+            const BYTE* const iHighLimit,                      \
+            const U32 rep[ZSTD_REP_NUM],                       \
+            U32 const ll0,                                     \
+            U32 const lengthToBeat)                            \
+    {                                                          \
+        return ZSTD_btGetAllMatches_internal(                  \
+                matches, ms, nextToUpdate3, ip, iHighLimit,    \
+                rep, ll0, lengthToBeat, ZSTD_##dictMode, mls); \
+    }
+
+#define GEN_ZSTD_BT_GET_ALL_MATCHES(dictMode)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 3)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 4)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 5)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 6)
+
+GEN_ZSTD_BT_GET_ALL_MATCHES(noDict)
+GEN_ZSTD_BT_GET_ALL_MATCHES(extDict)
+GEN_ZSTD_BT_GET_ALL_MATCHES(dictMatchState)
+
+#define ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMode)  \
+    {                                            \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 3), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 4), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 5), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 6)  \
+    }
+
+static ZSTD_getAllMatchesFn
+ZSTD_selectBtGetAllMatches(ZSTD_matchState_t const* ms, ZSTD_dictMode_e const dictMode)
+{
+    ZSTD_getAllMatchesFn const getAllMatchesFns[3][4] = {
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(noDict),
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(extDict),
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMatchState)
+    };
+    U32 const mls = BOUNDED(3, ms->cParams.minMatch, 6);
+    assert((U32)dictMode < 3);
+    assert(mls - 3 < 4);
+    return getAllMatchesFns[(int)dictMode][mls - 3];
+}
+
+/*************************
+*  LDM helper functions  *
+*************************/
+
+/* Struct containing info needed to make decision about ldm inclusion */
+typedef struct {
+    rawSeqStore_t seqStore;   /* External match candidates store for this block */
+    U32 startPosInBlock;      /* Start position of the current match candidate */
+    U32 endPosInBlock;        /* End position of the current match candidate */
+    U32 offset;               /* Offset of the match candidate */
+} ZSTD_optLdm_t;
+
+/* ZSTD_optLdm_skipRawSeqStoreBytes():
+ * Moves forward in @rawSeqStore by @nbBytes,
+ * which will update the fields 'pos' and 'posInSequence'.
+ */
+static void ZSTD_optLdm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes)
+{
+    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
+    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
+        if (currPos >= currSeq.litLength + currSeq.matchLength) {
+            currPos -= currSeq.litLength + currSeq.matchLength;
+            rawSeqStore->pos++;
+        } else {
+            rawSeqStore->posInSequence = currPos;
+            break;
+        }
+    }
+    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
+        rawSeqStore->posInSequence = 0;
+    }
+}
+
+/* ZSTD_opt_getNextMatchAndUpdateSeqStore():
+ * Calculates the beginning and end of the next match in the current block.
+ * Updates 'pos' and 'posInSequence' of the ldmSeqStore.
+ */
+static void
+ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
+                                       U32 blockBytesRemaining)
+{
+    rawSeq currSeq;
+    U32 currBlockEndPos;
+    U32 literalsBytesRemaining;
+    U32 matchBytesRemaining;
+
+    /* Setting match end position to MAX to ensure we never use an LDM during this block */
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        return;
+    }
+    /* Calculate appropriate bytes left in matchLength and litLength
+     * after adjusting based on ldmSeqStore->posInSequence */
+    currSeq = optLdm->seqStore.seq[optLdm->seqStore.pos];
+    assert(optLdm->seqStore.posInSequence <= currSeq.litLength + currSeq.matchLength);
+    currBlockEndPos = currPosInBlock + blockBytesRemaining;
+    literalsBytesRemaining = (optLdm->seqStore.posInSequence < currSeq.litLength) ?
+            currSeq.litLength - (U32)optLdm->seqStore.posInSequence :
+            0;
+    matchBytesRemaining = (literalsBytesRemaining == 0) ?
+            currSeq.matchLength - ((U32)optLdm->seqStore.posInSequence - currSeq.litLength) :
+            currSeq.matchLength;
+
+    /* If there are more literal bytes than bytes remaining in block, no ldm is possible */
+    if (literalsBytesRemaining >= blockBytesRemaining) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, blockBytesRemaining);
+        return;
+    }
+
+    /* Matches may be < MINMATCH by this process. In that case, we will reject them
+       when we are deciding whether or not to add the ldm */
+    optLdm->startPosInBlock = currPosInBlock + literalsBytesRemaining;
+    optLdm->endPosInBlock = optLdm->startPosInBlock + matchBytesRemaining;
+    optLdm->offset = currSeq.offset;
+
+    if (optLdm->endPosInBlock > currBlockEndPos) {
+        /* Match ends after the block ends, we can't use the whole match */
+        optLdm->endPosInBlock = currBlockEndPos;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, currBlockEndPos - currPosInBlock);
+    } else {
+        /* Consume nb of bytes equal to size of sequence left */
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, literalsBytesRemaining + matchBytesRemaining);
+    }
+}
+
+/* ZSTD_optLdm_maybeAddMatch():
+ * Adds a match if it's long enough,
+ * based on it's 'matchStartPosInBlock' and 'matchEndPosInBlock',
+ * into 'matches'. Maintains the correct ordering of 'matches'.
+ */
+static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
+                                      const ZSTD_optLdm_t* optLdm, U32 currPosInBlock)
+{
+    U32 const posDiff = currPosInBlock - optLdm->startPosInBlock;
+    /* Note: ZSTD_match_t actually contains offBase and matchLength (before subtracting MINMATCH) */
+    U32 const candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
+
+    /* Ensure that current block position is not outside of the match */
+    if (currPosInBlock < optLdm->startPosInBlock
+      || currPosInBlock >= optLdm->endPosInBlock
+      || candidateMatchLength < MINMATCH) {
+        return;
+    }
+
+    if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
+        U32 const candidateOffBase = OFFSET_TO_OFFBASE(optLdm->offset);
+        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offBase: %u matchLength %u) at block position=%u",
+                 candidateOffBase, candidateMatchLength, currPosInBlock);
+        matches[*nbMatches].len = candidateMatchLength;
+        matches[*nbMatches].off = candidateOffBase;
+        (*nbMatches)++;
+    }
+}
+
+/* ZSTD_optLdm_processMatchCandidate():
+ * Wrapper function to update ldm seq store and call ldm functions as necessary.
+ */
+static void
+ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm,
+                                  ZSTD_match_t* matches, U32* nbMatches,
+                                  U32 currPosInBlock, U32 remainingBytes)
+{
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        return;
+    }
+
+    if (currPosInBlock >= optLdm->endPosInBlock) {
+        if (currPosInBlock > optLdm->endPosInBlock) {
+            /* The position at which ZSTD_optLdm_processMatchCandidate() is called is not necessarily
+             * at the end of a match from the ldm seq store, and will often be some bytes
+             * over beyond matchEndPosInBlock. As such, we need to correct for these "overshoots"
+             */
+            U32 const posOvershoot = currPosInBlock - optLdm->endPosInBlock;
+            ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
+        }
+        ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
+    }
+    ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock);
+}
+
+
+/*-*******************************
+*  Optimal parser
+*********************************/
+
+static U32 ZSTD_totalLen(ZSTD_optimal_t sol)
+{
+    return sol.litlen + sol.mlen;
+}
+
+#if 0 /* debug */
+
+static void
+listStats(const U32* table, int lastEltID)
+{
+    int const nbElts = lastEltID + 1;
+    int enb;
+    for (enb=0; enb < nbElts; enb++) {
+        (void)table;
+        /* RAWLOG(2, "%3i:%3i,  ", enb, table[enb]); */
+        RAWLOG(2, "%4i,", table[enb]);
+    }
+    RAWLOG(2, " \n");
+}
+
+#endif
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
+                               seqStore_t* seqStore,
+                               U32 rep[ZSTD_REP_NUM],
+                         const void* src, size_t srcSize,
+                         const int optLevel,
+                         const ZSTD_dictMode_e dictMode)
+{
+    optState_t* const optStatePtr = &ms->opt;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+
+    ZSTD_getAllMatchesFn getAllMatches = ZSTD_selectBtGetAllMatches(ms, dictMode);
+
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    U32 const minMatch = (cParams->minMatch == 3) ? 3 : 4;
+    U32 nextToUpdate3 = ms->nextToUpdate;
+
+    ZSTD_optimal_t* const opt = optStatePtr->priceTable;
+    ZSTD_match_t* const matches = optStatePtr->matchTable;
+    ZSTD_optimal_t lastSequence;
+    ZSTD_optLdm_t optLdm;
+
+    ZSTD_memset(&lastSequence, 0, sizeof(ZSTD_optimal_t));
+
+    optLdm.seqStore = ms->ldmSeqStore ? *ms->ldmSeqStore : kNullRawSeqStore;
+    optLdm.endPosInBlock = optLdm.startPosInBlock = optLdm.offset = 0;
+    ZSTD_opt_getNextMatchAndUpdateSeqStore(&optLdm, (U32)(ip-istart), (U32)(iend-ip));
+
+    /* init */
+    DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
+                (U32)(ip - base), ms->window.dictLimit, ms->nextToUpdate);
+    assert(optLevel <= 2);
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize, optLevel);
+    ip += (ip==prefixStart);
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        U32 cur, last_pos = 0;
+
+        /* find first match */
+        {   U32 const litlen = (U32)(ip - anchor);
+            U32 const ll0 = !litlen;
+            U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, ip, iend, rep, ll0, minMatch);
+            ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                              (U32)(ip-istart), (U32)(iend - ip));
+            if (!nbMatches) { ip++; continue; }
+
+            /* initialize opt[0] */
+            { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
+            opt[0].mlen = 0;  /* means is_a_literal */
+            opt[0].litlen = litlen;
+            /* We don't need to include the actual price of the literals because
+             * it is static for the duration of the forward pass, and is included
+             * in every price. We include the literal length to avoid negative
+             * prices when we subtract the previous literal length.
+             */
+            opt[0].price = (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
+
+            /* large match -> immediate encoding */
+            {   U32 const maxML = matches[nbMatches-1].len;
+                U32 const maxOffBase = matches[nbMatches-1].off;
+                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffBase=%u at cPos=%u => start new series",
+                            nbMatches, maxML, maxOffBase, (U32)(ip-prefixStart));
+
+                if (maxML > sufficient_len) {
+                    lastSequence.litlen = litlen;
+                    lastSequence.mlen = maxML;
+                    lastSequence.off = maxOffBase;
+                    DEBUGLOG(6, "large match (%u>%u), immediate encoding",
+                                maxML, sufficient_len);
+                    cur = 0;
+                    last_pos = ZSTD_totalLen(lastSequence);
+                    goto _shortestPath;
+            }   }
+
+            /* set prices for first matches starting position == 0 */
+            assert(opt[0].price >= 0);
+            {   U32 const literalsPrice = (U32)opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
+                U32 pos;
+                U32 matchNb;
+                for (pos = 1; pos < minMatch; pos++) {
+                    opt[pos].price = ZSTD_MAX_PRICE;   /* mlen, litlen and price will be fixed during forward scanning */
+                }
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offBase = matches[matchNb].off;
+                    U32 const end = matches[matchNb].len;
+                    for ( ; pos <= end ; pos++ ) {
+                        U32 const matchPrice = ZSTD_getMatchPrice(offBase, pos, optStatePtr, optLevel);
+                        U32 const sequencePrice = literalsPrice + matchPrice;
+                        DEBUGLOG(7, "rPos:%u => set initial price : %.2f",
+                                    pos, ZSTD_fCost((int)sequencePrice));
+                        opt[pos].mlen = pos;
+                        opt[pos].off = offBase;
+                        opt[pos].litlen = litlen;
+                        opt[pos].price = (int)sequencePrice;
+                }   }
+                last_pos = pos-1;
+            }
+        }
+
+        /* check further positions */
+        for (cur = 1; cur <= last_pos; cur++) {
+            const BYTE* const inr = ip + cur;
+            assert(cur < ZSTD_OPT_NUM);
+            DEBUGLOG(7, "cPos:%zi==rPos:%u", inr-istart, cur)
+
+            /* Fix current position with one literal if cheaper */
+            {   U32 const litlen = (opt[cur-1].mlen == 0) ? opt[cur-1].litlen + 1 : 1;
+                int const price = opt[cur-1].price
+                                + (int)ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
+                                + (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
+                                - (int)ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
+                assert(price < 1000000000); /* overflow check */
+                if (price <= opt[cur].price) {
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
+                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price), litlen,
+                                opt[cur-1].rep[0], opt[cur-1].rep[1], opt[cur-1].rep[2]);
+                    opt[cur].mlen = 0;
+                    opt[cur].off = 0;
+                    opt[cur].litlen = litlen;
+                    opt[cur].price = price;
+                } else {
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u : literal would cost more (%.2f>%.2f) (hist:%u,%u,%u)",
+                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price),
+                                opt[cur].rep[0], opt[cur].rep[1], opt[cur].rep[2]);
+                }
+            }
+
+            /* Set the repcodes of the current position. We must do it here
+             * because we rely on the repcodes of the 2nd to last sequence being
+             * correct to set the next chunks repcodes during the backward
+             * traversal.
+             */
+            ZSTD_STATIC_ASSERT(sizeof(opt[cur].rep) == sizeof(repcodes_t));
+            assert(cur >= opt[cur].mlen);
+            if (opt[cur].mlen != 0) {
+                U32 const prev = cur - opt[cur].mlen;
+                repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
+                ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
+            } else {
+                ZSTD_memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
+            }
+
+            /* last match must start at a minimum distance of 8 from oend */
+            if (inr > ilimit) continue;
+
+            if (cur == last_pos) break;
+
+            if ( (optLevel==0) /*static_test*/
+              && (opt[cur+1].price <= opt[cur].price + (BITCOST_MULTIPLIER/2)) ) {
+                DEBUGLOG(7, "move to next rPos:%u : price is <=", cur+1);
+                continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
+            }
+
+            assert(opt[cur].price >= 0);
+            {   U32 const ll0 = (opt[cur].mlen != 0);
+                U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
+                U32 const previousPrice = (U32)opt[cur].price;
+                U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
+                U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, inr, iend, opt[cur].rep, ll0, minMatch);
+                U32 matchNb;
+
+                ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                                  (U32)(inr-istart), (U32)(iend-inr));
+
+                if (!nbMatches) {
+                    DEBUGLOG(7, "rPos:%u : no match found", cur);
+                    continue;
+                }
+
+                {   U32 const maxML = matches[nbMatches-1].len;
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u, found %u matches, of maxLength=%u",
+                                inr-istart, cur, nbMatches, maxML);
+
+                    if ( (maxML > sufficient_len)
+                      || (cur + maxML >= ZSTD_OPT_NUM) ) {
+                        lastSequence.mlen = maxML;
+                        lastSequence.off = matches[nbMatches-1].off;
+                        lastSequence.litlen = litlen;
+                        cur -= (opt[cur].mlen==0) ? opt[cur].litlen : 0;  /* last sequence is actually only literals, fix cur to last match - note : may underflow, in which case, it's first sequence, and it's okay */
+                        last_pos = cur + ZSTD_totalLen(lastSequence);
+                        if (cur > ZSTD_OPT_NUM) cur = 0;   /* underflow => first match */
+                        goto _shortestPath;
+                }   }
+
+                /* set prices using matches found at position == cur */
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    U32 const lastML = matches[matchNb].len;
+                    U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
+                    U32 mlen;
+
+                    DEBUGLOG(7, "testing match %u => offBase=%4u, mlen=%2u, llen=%2u",
+                                matchNb, matches[matchNb].off, lastML, litlen);
+
+                    for (mlen = lastML; mlen >= startML; mlen--) {  /* scan downward */
+                        U32 const pos = cur + mlen;
+                        int const price = (int)basePrice + (int)ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
+
+                        if ((pos > last_pos) || (price < opt[pos].price)) {
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            while (last_pos < pos) { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; }   /* fill empty positions */
+                            opt[pos].mlen = mlen;
+                            opt[pos].off = offset;
+                            opt[pos].litlen = litlen;
+                            opt[pos].price = price;
+                        } else {
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new price is worse (%.2f>=%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            if (optLevel==0) break;  /* early update abort; gets ~+10% speed for about -0.01 ratio loss */
+                        }
+            }   }   }
+        }  /* for (cur = 1; cur <= last_pos; cur++) */
+
+        lastSequence = opt[last_pos];
+        cur = last_pos > ZSTD_totalLen(lastSequence) ? last_pos - ZSTD_totalLen(lastSequence) : 0;  /* single sequence, and it starts before `ip` */
+        assert(cur < ZSTD_OPT_NUM);  /* control overflow*/
+
+_shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
+        assert(opt[0].mlen == 0);
+
+        /* Set the next chunk's repcodes based on the repcodes of the beginning
+         * of the last match, and the last sequence. This avoids us having to
+         * update them while traversing the sequences.
+         */
+        if (lastSequence.mlen != 0) {
+            repcodes_t const reps = ZSTD_newRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
+            ZSTD_memcpy(rep, &reps, sizeof(reps));
+        } else {
+            ZSTD_memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
+        }
+
+        {   U32 const storeEnd = cur + 1;
+            U32 storeStart = storeEnd;
+            U32 seqPos = cur;
+
+            DEBUGLOG(6, "start reverse traversal (last_pos:%u, cur:%u)",
+                        last_pos, cur); (void)last_pos;
+            assert(storeEnd < ZSTD_OPT_NUM);
+            DEBUGLOG(6, "last sequence copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                        storeEnd, lastSequence.litlen, lastSequence.mlen, lastSequence.off);
+            opt[storeEnd] = lastSequence;
+            while (seqPos > 0) {
+                U32 const backDist = ZSTD_totalLen(opt[seqPos]);
+                storeStart--;
+                DEBUGLOG(6, "sequence from rPos=%u copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                            seqPos, storeStart, opt[seqPos].litlen, opt[seqPos].mlen, opt[seqPos].off);
+                opt[storeStart] = opt[seqPos];
+                seqPos = (seqPos > backDist) ? seqPos - backDist : 0;
+            }
+
+            /* save sequences */
+            DEBUGLOG(6, "sending selected sequences into seqStore")
+            {   U32 storePos;
+                for (storePos=storeStart; storePos <= storeEnd; storePos++) {
+                    U32 const llen = opt[storePos].litlen;
+                    U32 const mlen = opt[storePos].mlen;
+                    U32 const offBase = opt[storePos].off;
+                    U32 const advance = llen + mlen;
+                    DEBUGLOG(6, "considering seq starting at %zi, llen=%u, mlen=%u",
+                                anchor - istart, (unsigned)llen, (unsigned)mlen);
+
+                    if (mlen==0) {  /* only literals => must be last "sequence", actually starting a new stream of sequences */
+                        assert(storePos == storeEnd);   /* must be last sequence */
+                        ip = anchor + llen;     /* last "sequence" is a bunch of literals => don't progress anchor */
+                        continue;   /* will finish */
+                    }
+
+                    assert(anchor + llen <= iend);
+                    ZSTD_updateStats(optStatePtr, llen, anchor, offBase, mlen);
+                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offBase, mlen);
+                    anchor += advance;
+                    ip = anchor;
+            }   }
+            ZSTD_setBasePrices(optStatePtr, optLevel);
+        }
+    }   /* while (ip < ilimit) */
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+static size_t ZSTD_compressBlock_opt0(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /* optLevel */, dictMode);
+}
+
+static size_t ZSTD_compressBlock_opt2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /* optLevel */, dictMode);
+}
+
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btopt");
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
+}
+
+
+
+
+/* ZSTD_initStats_ultra():
+ * make a first compression pass, just to seed stats with more accurate starting values.
+ * only works on first block, with no dictionary and no ldm.
+ * this function cannot error out, its narrow contract must be respected.
+ */
+static void
+ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
+                     seqStore_t* seqStore,
+                     U32 rep[ZSTD_REP_NUM],
+               const void* src, size_t srcSize)
+{
+    U32 tmpRep[ZSTD_REP_NUM];  /* updated rep codes will sink here */
+    ZSTD_memcpy(tmpRep, rep, sizeof(tmpRep));
+
+    DEBUGLOG(4, "ZSTD_initStats_ultra (srcSize=%zu)", srcSize);
+    assert(ms->opt.litLengthSum == 0);    /* first block */
+    assert(seqStore->sequences == seqStore->sequencesStart);   /* no ldm */
+    assert(ms->window.dictLimit == ms->window.lowLimit);   /* no dictionary */
+    assert(ms->window.dictLimit - ms->nextToUpdate <= 1);  /* no prefix (note: intentional overflow, defined as 2-complement) */
+
+    ZSTD_compressBlock_opt2(ms, seqStore, tmpRep, src, srcSize, ZSTD_noDict);   /* generate stats into ms->opt*/
+
+    /* invalidate first scan from history, only keep entropy stats */
+    ZSTD_resetSeqStore(seqStore);
+    ms->window.base -= srcSize;
+    ms->window.dictLimit += (U32)srcSize;
+    ms->window.lowLimit = ms->window.dictLimit;
+    ms->nextToUpdate = ms->window.dictLimit;
+
+}
+
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btultra (srcSize=%zu)", srcSize);
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_btultra2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    U32 const curr = (U32)((const BYTE*)src - ms->window.base);
+    DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
+
+    /* 2-passes strategy:
+     * this strategy makes a first pass over first block to collect statistics
+     * in order to seed next round's statistics with it.
+     * After 1st pass, function forgets history, and starts a new block.
+     * Consequently, this can only work if no data has been previously loaded in tables,
+     * aka, no dictionary, no prefix, no ldm preprocessing.
+     * The compression ratio gain is generally small (~0.5% on first block),
+    ** the cost is 2x cpu time on first block. */
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+    if ( (ms->opt.litLengthSum==0)   /* first block */
+      && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
+      && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
+      && (curr == ms->window.dictLimit)    /* start of frame, nothing already loaded nor skipped */
+      && (srcSize > ZSTD_PREDEF_THRESHOLD) /* input large enough to not employ default stats */
+      ) {
+        ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
+    }
+
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_btopt_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_btultra_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
+}
+
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
+}
+
+/* note : no btultra2 variant for extDict nor dictMatchState,
+ * because btultra2 is not meant to work with dictionaries
+ * and is only specific for the first block (no prefix) */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_opt.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_opt.h
new file mode 100644
index 0000000..342e5a3
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstd_opt.h
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_OPT_H
+#define ZSTD_OPT_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"
+
+/* used in ZSTD_loadDictionaryContent() */
+void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend);
+
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+
+size_t ZSTD_compressBlock_btopt_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+        /* note : no btultra2 variant for extDict nor dictMatchState,
+         * because btultra2 is not meant to work with dictionaries
+         * and is only specific for the first block (no prefix) */
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_OPT_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstdmt_compress.c b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstdmt_compress.c
new file mode 100644
index 0000000..6786075
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstdmt_compress.c
@@ -0,0 +1,1867 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ======   Compiler specifics   ====== */
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4204)   /* disable: C4204: non-constant aggregate initializer */
+#endif
+
+
+/* ======   Constants   ====== */
+#define ZSTDMT_OVERLAPLOG_DEFAULT 0
+
+
+/* ======   Dependencies   ====== */
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memset, INT_MAX, UINT_MAX */
+#include "../common/mem.h"         /* MEM_STATIC */
+#include "../common/pool.h"        /* threadpool */
+#include "../common/threading.h"   /* mutex */
+#include "zstd_compress_internal.h"  /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
+#include "zstd_ldm.h"
+#include "zstdmt_compress.h"
+
+/* Guards code to support resizing the SeqPool.
+ * We will want to resize the SeqPool to save memory in the future.
+ * Until then, comment the code out since it is unused.
+ */
+#define ZSTD_RESIZE_SEQPOOL 0
+
+/* ======   Debug   ====== */
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL>=2) \
+    && !defined(_MSC_VER) \
+    && !defined(__MINGW32__)
+
+#  include <stdio.h>
+#  include <unistd.h>
+#  include <sys/times.h>
+
+#  define DEBUG_PRINTHEX(l,p,n) {            \
+    unsigned debug_u;                        \
+    for (debug_u=0; debug_u<(n); debug_u++)  \
+        RAWLOG(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
+    RAWLOG(l, " \n");                        \
+}
+
+static unsigned long long GetCurrentClockTimeMicroseconds(void)
+{
+   static clock_t _ticksPerSecond = 0;
+   if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);
+
+   {   struct tms junk; clock_t newTicks = (clock_t) times(&junk);
+       return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond);
+}  }
+
+#define MUTEX_WAIT_TIME_DLEVEL 6
+#define ZSTD_PTHREAD_MUTEX_LOCK(mutex) {          \
+    if (DEBUGLEVEL >= MUTEX_WAIT_TIME_DLEVEL) {   \
+        unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds(); \
+        ZSTD_pthread_mutex_lock(mutex);           \
+        {   unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
+            unsigned long long const elapsedTime = (afterTime-beforeTime); \
+            if (elapsedTime > 1000) {  /* or whatever threshold you like; I'm using 1 millisecond here */ \
+                DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \
+                   elapsedTime, #mutex);          \
+        }   }                                     \
+    } else {                                      \
+        ZSTD_pthread_mutex_lock(mutex);           \
+    }                                             \
+}
+
+#else
+
+#  define ZSTD_PTHREAD_MUTEX_LOCK(m) ZSTD_pthread_mutex_lock(m)
+#  define DEBUG_PRINTHEX(l,p,n) {}
+
+#endif
+
+
+/* =====   Buffer Pool   ===== */
+/* a single Buffer Pool can be invoked from multiple threads in parallel */
+
+typedef struct buffer_s {
+    void* start;
+    size_t capacity;
+} buffer_t;
+
+static const buffer_t g_nullBuffer = { NULL, 0 };
+
+typedef struct ZSTDMT_bufferPool_s {
+    ZSTD_pthread_mutex_t poolMutex;
+    size_t bufferSize;
+    unsigned totalBuffers;
+    unsigned nbBuffers;
+    ZSTD_customMem cMem;
+    buffer_t bTable[1];   /* variable size */
+} ZSTDMT_bufferPool;
+
+static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned maxNbBuffers, ZSTD_customMem cMem)
+{
+    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_customCalloc(
+        sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
+    if (bufPool==NULL) return NULL;
+    if (ZSTD_pthread_mutex_init(&bufPool->poolMutex, NULL)) {
+        ZSTD_customFree(bufPool, cMem);
+        return NULL;
+    }
+    bufPool->bufferSize = 64 KB;
+    bufPool->totalBuffers = maxNbBuffers;
+    bufPool->nbBuffers = 0;
+    bufPool->cMem = cMem;
+    return bufPool;
+}
+
+static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    unsigned u;
+    DEBUGLOG(3, "ZSTDMT_freeBufferPool (address:%08X)", (U32)(size_t)bufPool);
+    if (!bufPool) return;   /* compatibility with free on NULL */
+    for (u=0; u<bufPool->totalBuffers; u++) {
+        DEBUGLOG(4, "free buffer %2u (address:%08X)", u, (U32)(size_t)bufPool->bTable[u].start);
+        ZSTD_customFree(bufPool->bTable[u].start, bufPool->cMem);
+    }
+    ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
+    ZSTD_customFree(bufPool, bufPool->cMem);
+}
+
+/* only works at initialization, not during compression */
+static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const poolSize = sizeof(*bufPool)
+                          + (bufPool->totalBuffers - 1) * sizeof(buffer_t);
+    unsigned u;
+    size_t totalBufferSize = 0;
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    for (u=0; u<bufPool->totalBuffers; u++)
+        totalBufferSize += bufPool->bTable[u].capacity;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+
+    return poolSize + totalBufferSize;
+}
+
+/* ZSTDMT_setBufferSize() :
+ * all future buffers provided by this buffer pool will have _at least_ this size
+ * note : it's better for all buffers to have same size,
+ * as they become freely interchangeable, reducing malloc/free usages and memory fragmentation */
+static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* const bufPool, size_t const bSize)
+{
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    DEBUGLOG(4, "ZSTDMT_setBufferSize: bSize = %u", (U32)bSize);
+    bufPool->bufferSize = bSize;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+}
+
+
+static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool, unsigned maxNbBuffers)
+{
+    if (srcBufPool==NULL) return NULL;
+    if (srcBufPool->totalBuffers >= maxNbBuffers) /* good enough */
+        return srcBufPool;
+    /* need a larger buffer pool */
+    {   ZSTD_customMem const cMem = srcBufPool->cMem;
+        size_t const bSize = srcBufPool->bufferSize;   /* forward parameters */
+        ZSTDMT_bufferPool* newBufPool;
+        ZSTDMT_freeBufferPool(srcBufPool);
+        newBufPool = ZSTDMT_createBufferPool(maxNbBuffers, cMem);
+        if (newBufPool==NULL) return newBufPool;
+        ZSTDMT_setBufferSize(newBufPool, bSize);
+        return newBufPool;
+    }
+}
+
+/** ZSTDMT_getBuffer() :
+ *  assumption : bufPool must be valid
+ * @return : a buffer, with start pointer and size
+ *  note: allocation may fail, in this case, start==NULL and size==0 */
+static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const bSize = bufPool->bufferSize;
+    DEBUGLOG(5, "ZSTDMT_getBuffer: bSize = %u", (U32)bufPool->bufferSize);
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers) {   /* try to use an existing buffer */
+        buffer_t const buf = bufPool->bTable[--(bufPool->nbBuffers)];
+        size_t const availBufferSize = buf.capacity;
+        bufPool->bTable[bufPool->nbBuffers] = g_nullBuffer;
+        if ((availBufferSize >= bSize) & ((availBufferSize>>3) <= bSize)) {
+            /* large enough, but not too much */
+            DEBUGLOG(5, "ZSTDMT_getBuffer: provide buffer %u of size %u",
+                        bufPool->nbBuffers, (U32)buf.capacity);
+            ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+            return buf;
+        }
+        /* size conditions not respected : scratch this buffer, create new one */
+        DEBUGLOG(5, "ZSTDMT_getBuffer: existing buffer does not meet size conditions => freeing");
+        ZSTD_customFree(buf.start, bufPool->cMem);
+    }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+    /* create new buffer */
+    DEBUGLOG(5, "ZSTDMT_getBuffer: create a new buffer");
+    {   buffer_t buffer;
+        void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
+        buffer.start = start;   /* note : start can be NULL if malloc fails ! */
+        buffer.capacity = (start==NULL) ? 0 : bSize;
+        if (start==NULL) {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: buffer allocation failure !!");
+        } else {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: created buffer of size %u", (U32)bSize);
+        }
+        return buffer;
+    }
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+/** ZSTDMT_resizeBuffer() :
+ * assumption : bufPool must be valid
+ * @return : a buffer that is at least the buffer pool buffer size.
+ *           If a reallocation happens, the data in the input buffer is copied.
+ */
+static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
+{
+    size_t const bSize = bufPool->bufferSize;
+    if (buffer.capacity < bSize) {
+        void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
+        buffer_t newBuffer;
+        newBuffer.start = start;
+        newBuffer.capacity = start == NULL ? 0 : bSize;
+        if (start != NULL) {
+            assert(newBuffer.capacity >= buffer.capacity);
+            ZSTD_memcpy(newBuffer.start, buffer.start, buffer.capacity);
+            DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
+            return newBuffer;
+        }
+        DEBUGLOG(5, "ZSTDMT_resizeBuffer: buffer allocation failure !!");
+    }
+    return buffer;
+}
+#endif
+
+/* store buffer for later re-use, up to pool capacity */
+static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
+{
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer");
+    if (buf.start == NULL) return;   /* compatible with release on NULL */
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers < bufPool->totalBuffers) {
+        bufPool->bTable[bufPool->nbBuffers++] = buf;  /* stored for later use */
+        DEBUGLOG(5, "ZSTDMT_releaseBuffer: stored buffer of size %u in slot %u",
+                    (U32)buf.capacity, (U32)(bufPool->nbBuffers-1));
+        ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+        return;
+    }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+    /* Reached bufferPool capacity (should not happen) */
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer: pool capacity reached => freeing ");
+    ZSTD_customFree(buf.start, bufPool->cMem);
+}
+
+/* We need 2 output buffers per worker since each dstBuff must be flushed after it is released.
+ * The 3 additional buffers are as follows:
+ *   1 buffer for input loading
+ *   1 buffer for "next input" when submitting current one
+ *   1 buffer stuck in queue */
+#define BUF_POOL_MAX_NB_BUFFERS(nbWorkers) (2*(nbWorkers) + 3)
+
+/* After a worker releases its rawSeqStore, it is immediately ready for reuse.
+ * So we only need one seq buffer per worker. */
+#define SEQ_POOL_MAX_NB_BUFFERS(nbWorkers) (nbWorkers)
+
+/* =====   Seq Pool Wrapper   ====== */
+
+typedef ZSTDMT_bufferPool ZSTDMT_seqPool;
+
+static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
+{
+    return ZSTDMT_sizeof_bufferPool(seqPool);
+}
+
+static rawSeqStore_t bufferToSeq(buffer_t buffer)
+{
+    rawSeqStore_t seq = kNullRawSeqStore;
+    seq.seq = (rawSeq*)buffer.start;
+    seq.capacity = buffer.capacity / sizeof(rawSeq);
+    return seq;
+}
+
+static buffer_t seqToBuffer(rawSeqStore_t seq)
+{
+    buffer_t buffer;
+    buffer.start = seq.seq;
+    buffer.capacity = seq.capacity * sizeof(rawSeq);
+    return buffer;
+}
+
+static rawSeqStore_t ZSTDMT_getSeq(ZSTDMT_seqPool* seqPool)
+{
+    if (seqPool->bufferSize == 0) {
+        return kNullRawSeqStore;
+    }
+    return bufferToSeq(ZSTDMT_getBuffer(seqPool));
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+static rawSeqStore_t ZSTDMT_resizeSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  return bufferToSeq(ZSTDMT_resizeBuffer(seqPool, seqToBuffer(seq)));
+}
+#endif
+
+static void ZSTDMT_releaseSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  ZSTDMT_releaseBuffer(seqPool, seqToBuffer(seq));
+}
+
+static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
+{
+  ZSTDMT_setBufferSize(seqPool, nbSeq * sizeof(rawSeq));
+}
+
+static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(SEQ_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
+    if (seqPool == NULL) return NULL;
+    ZSTDMT_setNbSeq(seqPool, 0);
+    return seqPool;
+}
+
+static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
+{
+    ZSTDMT_freeBufferPool(seqPool);
+}
+
+static ZSTDMT_seqPool* ZSTDMT_expandSeqPool(ZSTDMT_seqPool* pool, U32 nbWorkers)
+{
+    return ZSTDMT_expandBufferPool(pool, SEQ_POOL_MAX_NB_BUFFERS(nbWorkers));
+}
+
+
+/* =====   CCtx Pool   ===== */
+/* a single CCtx Pool can be invoked from multiple threads in parallel */
+
+typedef struct {
+    ZSTD_pthread_mutex_t poolMutex;
+    int totalCCtx;
+    int availCCtx;
+    ZSTD_customMem cMem;
+    ZSTD_CCtx* cctx[1];   /* variable size */
+} ZSTDMT_CCtxPool;
+
+/* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */
+static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
+{
+    int cid;
+    for (cid=0; cid<pool->totalCCtx; cid++)
+        ZSTD_freeCCtx(pool->cctx[cid]);  /* note : compatible with free on NULL */
+    ZSTD_pthread_mutex_destroy(&pool->poolMutex);
+    ZSTD_customFree(pool, pool->cMem);
+}
+
+/* ZSTDMT_createCCtxPool() :
+ * implies nbWorkers >= 1 , checked by caller ZSTDMT_createCCtx() */
+static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(int nbWorkers,
+                                              ZSTD_customMem cMem)
+{
+    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_customCalloc(
+        sizeof(ZSTDMT_CCtxPool) + (nbWorkers-1)*sizeof(ZSTD_CCtx*), cMem);
+    assert(nbWorkers > 0);
+    if (!cctxPool) return NULL;
+    if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
+        ZSTD_customFree(cctxPool, cMem);
+        return NULL;
+    }
+    cctxPool->cMem = cMem;
+    cctxPool->totalCCtx = nbWorkers;
+    cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */
+    cctxPool->cctx[0] = ZSTD_createCCtx_advanced(cMem);
+    if (!cctxPool->cctx[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }
+    DEBUGLOG(3, "cctxPool created, with %u workers", nbWorkers);
+    return cctxPool;
+}
+
+static ZSTDMT_CCtxPool* ZSTDMT_expandCCtxPool(ZSTDMT_CCtxPool* srcPool,
+                                              int nbWorkers)
+{
+    if (srcPool==NULL) return NULL;
+    if (nbWorkers <= srcPool->totalCCtx) return srcPool;   /* good enough */
+    /* need a larger cctx pool */
+    {   ZSTD_customMem const cMem = srcPool->cMem;
+        ZSTDMT_freeCCtxPool(srcPool);
+        return ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    }
+}
+
+/* only works during initialization phase, not during compression */
+static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
+{
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    {   unsigned const nbWorkers = cctxPool->totalCCtx;
+        size_t const poolSize = sizeof(*cctxPool)
+                                + (nbWorkers-1) * sizeof(ZSTD_CCtx*);
+        unsigned u;
+        size_t totalCCtxSize = 0;
+        for (u=0; u<nbWorkers; u++) {
+            totalCCtxSize += ZSTD_sizeof_CCtx(cctxPool->cctx[u]);
+        }
+        ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+        assert(nbWorkers > 0);
+        return poolSize + totalCCtxSize;
+    }
+}
+
+static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool)
+{
+    DEBUGLOG(5, "ZSTDMT_getCCtx");
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    if (cctxPool->availCCtx) {
+        cctxPool->availCCtx--;
+        {   ZSTD_CCtx* const cctx = cctxPool->cctx[cctxPool->availCCtx];
+            ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+            return cctx;
+    }   }
+    ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+    DEBUGLOG(5, "create one more CCtx");
+    return ZSTD_createCCtx_advanced(cctxPool->cMem);   /* note : can be NULL, when creation fails ! */
+}
+
+static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return;   /* compatibility with release on NULL */
+    ZSTD_pthread_mutex_lock(&pool->poolMutex);
+    if (pool->availCCtx < pool->totalCCtx)
+        pool->cctx[pool->availCCtx++] = cctx;
+    else {
+        /* pool overflow : should not happen, since totalCCtx==nbWorkers */
+        DEBUGLOG(4, "CCtx pool overflow : free cctx");
+        ZSTD_freeCCtx(cctx);
+    }
+    ZSTD_pthread_mutex_unlock(&pool->poolMutex);
+}
+
+/* ====   Serial State   ==== */
+
+typedef struct {
+    void const* start;
+    size_t size;
+} range_t;
+
+typedef struct {
+    /* All variables in the struct are protected by mutex. */
+    ZSTD_pthread_mutex_t mutex;
+    ZSTD_pthread_cond_t cond;
+    ZSTD_CCtx_params params;
+    ldmState_t ldmState;
+    XXH64_state_t xxhState;
+    unsigned nextJobID;
+    /* Protects ldmWindow.
+     * Must be acquired after the main mutex when acquiring both.
+     */
+    ZSTD_pthread_mutex_t ldmWindowMutex;
+    ZSTD_pthread_cond_t ldmWindowCond;  /* Signaled when ldmWindow is updated */
+    ZSTD_window_t ldmWindow;  /* A thread-safe copy of ldmState.window */
+} serialState_t;
+
+static int
+ZSTDMT_serialState_reset(serialState_t* serialState,
+                         ZSTDMT_seqPool* seqPool,
+                         ZSTD_CCtx_params params,
+                         size_t jobSize,
+                         const void* dict, size_t const dictSize,
+                         ZSTD_dictContentType_e dictContentType)
+{
+    /* Adjust parameters */
+    if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
+        DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
+        assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
+        assert(params.ldmParams.hashRateLog < 32);
+    } else {
+        ZSTD_memset(&params.ldmParams, 0, sizeof(params.ldmParams));
+    }
+    serialState->nextJobID = 0;
+    if (params.fParams.checksumFlag)
+        XXH64_reset(&serialState->xxhState, 0);
+    if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_customMem cMem = params.customMem;
+        unsigned const hashLog = params.ldmParams.hashLog;
+        size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
+        unsigned const bucketLog =
+            params.ldmParams.hashLog - params.ldmParams.bucketSizeLog;
+        unsigned const prevBucketLog =
+            serialState->params.ldmParams.hashLog -
+            serialState->params.ldmParams.bucketSizeLog;
+        size_t const numBuckets = (size_t)1 << bucketLog;
+        /* Size the seq pool tables */
+        ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, jobSize));
+        /* Reset the window */
+        ZSTD_window_init(&serialState->ldmState.window);
+        /* Resize tables and output space if necessary. */
+        if (serialState->ldmState.hashTable == NULL || serialState->params.ldmParams.hashLog < hashLog) {
+            ZSTD_customFree(serialState->ldmState.hashTable, cMem);
+            serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_customMalloc(hashSize, cMem);
+        }
+        if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
+            ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
+            serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_customMalloc(numBuckets, cMem);
+        }
+        if (!serialState->ldmState.hashTable || !serialState->ldmState.bucketOffsets)
+            return 1;
+        /* Zero the tables */
+        ZSTD_memset(serialState->ldmState.hashTable, 0, hashSize);
+        ZSTD_memset(serialState->ldmState.bucketOffsets, 0, numBuckets);
+
+        /* Update window state and fill hash table with dict */
+        serialState->ldmState.loadedDictEnd = 0;
+        if (dictSize > 0) {
+            if (dictContentType == ZSTD_dct_rawContent) {
+                BYTE const* const dictEnd = (const BYTE*)dict + dictSize;
+                ZSTD_window_update(&serialState->ldmState.window, dict, dictSize, /* forceNonContiguous */ 0);
+                ZSTD_ldm_fillHashTable(&serialState->ldmState, (const BYTE*)dict, dictEnd, &params.ldmParams);
+                serialState->ldmState.loadedDictEnd = params.forceWindow ? 0 : (U32)(dictEnd - serialState->ldmState.window.base);
+            } else {
+                /* don't even load anything */
+            }
+        }
+
+        /* Initialize serialState's copy of ldmWindow. */
+        serialState->ldmWindow = serialState->ldmState.window;
+    }
+
+    serialState->params = params;
+    serialState->params.jobSize = (U32)jobSize;
+    return 0;
+}
+
+static int ZSTDMT_serialState_init(serialState_t* serialState)
+{
+    int initError = 0;
+    ZSTD_memset(serialState, 0, sizeof(*serialState));
+    initError |= ZSTD_pthread_mutex_init(&serialState->mutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->cond, NULL);
+    initError |= ZSTD_pthread_mutex_init(&serialState->ldmWindowMutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->ldmWindowCond, NULL);
+    return initError;
+}
+
+static void ZSTDMT_serialState_free(serialState_t* serialState)
+{
+    ZSTD_customMem cMem = serialState->params.customMem;
+    ZSTD_pthread_mutex_destroy(&serialState->mutex);
+    ZSTD_pthread_cond_destroy(&serialState->cond);
+    ZSTD_pthread_mutex_destroy(&serialState->ldmWindowMutex);
+    ZSTD_pthread_cond_destroy(&serialState->ldmWindowCond);
+    ZSTD_customFree(serialState->ldmState.hashTable, cMem);
+    ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
+}
+
+static void ZSTDMT_serialState_update(serialState_t* serialState,
+                                      ZSTD_CCtx* jobCCtx, rawSeqStore_t seqStore,
+                                      range_t src, unsigned jobID)
+{
+    /* Wait for our turn */
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    while (serialState->nextJobID < jobID) {
+        DEBUGLOG(5, "wait for serialState->cond");
+        ZSTD_pthread_cond_wait(&serialState->cond, &serialState->mutex);
+    }
+    /* A future job may error and skip our job */
+    if (serialState->nextJobID == jobID) {
+        /* It is now our turn, do any processing necessary */
+        if (serialState->params.ldmParams.enableLdm == ZSTD_ps_enable) {
+            size_t error;
+            assert(seqStore.seq != NULL && seqStore.pos == 0 &&
+                   seqStore.size == 0 && seqStore.capacity > 0);
+            assert(src.size <= serialState->params.jobSize);
+            ZSTD_window_update(&serialState->ldmState.window, src.start, src.size, /* forceNonContiguous */ 0);
+            error = ZSTD_ldm_generateSequences(
+                &serialState->ldmState, &seqStore,
+                &serialState->params.ldmParams, src.start, src.size);
+            /* We provide a large enough buffer to never fail. */
+            assert(!ZSTD_isError(error)); (void)error;
+            /* Update ldmWindow to match the ldmState.window and signal the main
+             * thread if it is waiting for a buffer.
+             */
+            ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+            serialState->ldmWindow = serialState->ldmState.window;
+            ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+            ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+        }
+        if (serialState->params.fParams.checksumFlag && src.size > 0)
+            XXH64_update(&serialState->xxhState, src.start, src.size);
+    }
+    /* Now it is the next jobs turn */
+    serialState->nextJobID++;
+    ZSTD_pthread_cond_broadcast(&serialState->cond);
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
+    if (seqStore.size > 0) {
+        size_t const err = ZSTD_referenceExternalSequences(
+            jobCCtx, seqStore.seq, seqStore.size);
+        assert(serialState->params.ldmParams.enableLdm == ZSTD_ps_enable);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
+}
+
+static void ZSTDMT_serialState_ensureFinished(serialState_t* serialState,
+                                              unsigned jobID, size_t cSize)
+{
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    if (serialState->nextJobID <= jobID) {
+        assert(ZSTD_isError(cSize)); (void)cSize;
+        DEBUGLOG(5, "Skipping past job %u because of error", jobID);
+        serialState->nextJobID = jobID + 1;
+        ZSTD_pthread_cond_broadcast(&serialState->cond);
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+        ZSTD_window_clear(&serialState->ldmWindow);
+        ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+        ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+    }
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
+}
+
+
+/* ------------------------------------------ */
+/* =====          Worker thread         ===== */
+/* ------------------------------------------ */
+
+static const range_t kNullRange = { NULL, 0 };
+
+typedef struct {
+    size_t   consumed;                   /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx */
+    size_t   cSize;                      /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx, then set0 by mtctx */
+    ZSTD_pthread_mutex_t job_mutex;      /* Thread-safe - used by mtctx and worker */
+    ZSTD_pthread_cond_t job_cond;        /* Thread-safe - used by mtctx and worker */
+    ZSTDMT_CCtxPool* cctxPool;           /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_bufferPool* bufPool;          /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_seqPool* seqPool;             /* Thread-safe - used by mtctx and (all) workers */
+    serialState_t* serial;               /* Thread-safe - used by mtctx and (all) workers */
+    buffer_t dstBuff;                    /* set by worker (or mtctx), then read by worker & mtctx, then modified by mtctx => no barrier */
+    range_t prefix;                      /* set by mtctx, then read by worker & mtctx => no barrier */
+    range_t src;                         /* set by mtctx, then read by worker & mtctx => no barrier */
+    unsigned jobID;                      /* set by mtctx, then read by worker => no barrier */
+    unsigned firstJob;                   /* set by mtctx, then read by worker => no barrier */
+    unsigned lastJob;                    /* set by mtctx, then read by worker => no barrier */
+    ZSTD_CCtx_params params;             /* set by mtctx, then read by worker => no barrier */
+    const ZSTD_CDict* cdict;             /* set by mtctx, then read by worker => no barrier */
+    unsigned long long fullFrameSize;    /* set by mtctx, then read by worker => no barrier */
+    size_t   dstFlushed;                 /* used only by mtctx */
+    unsigned frameChecksumNeeded;        /* used only by mtctx */
+} ZSTDMT_jobDescription;
+
+#define JOB_ERROR(e) {                          \
+    ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);   \
+    job->cSize = e;                             \
+    ZSTD_pthread_mutex_unlock(&job->job_mutex); \
+    goto _endJob;                               \
+}
+
+/* ZSTDMT_compressionJob() is a POOL_function type */
+static void ZSTDMT_compressionJob(void* jobDescription)
+{
+    ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
+    ZSTD_CCtx_params jobParams = job->params;   /* do not modify job->params ! copy it, modify the copy */
+    ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(job->cctxPool);
+    rawSeqStore_t rawSeqStore = ZSTDMT_getSeq(job->seqPool);
+    buffer_t dstBuff = job->dstBuff;
+    size_t lastCBlockSize = 0;
+
+    /* resources */
+    if (cctx==NULL) JOB_ERROR(ERROR(memory_allocation));
+    if (dstBuff.start == NULL) {   /* streaming job : doesn't provide a dstBuffer */
+        dstBuff = ZSTDMT_getBuffer(job->bufPool);
+        if (dstBuff.start==NULL) JOB_ERROR(ERROR(memory_allocation));
+        job->dstBuff = dstBuff;   /* this value can be read in ZSTDMT_flush, when it copies the whole job */
+    }
+    if (jobParams.ldmParams.enableLdm == ZSTD_ps_enable && rawSeqStore.seq == NULL)
+        JOB_ERROR(ERROR(memory_allocation));
+
+    /* Don't compute the checksum for chunks, since we compute it externally,
+     * but write it in the header.
+     */
+    if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
+    /* Don't run LDM for the chunks, since we handle it externally */
+    jobParams.ldmParams.enableLdm = ZSTD_ps_disable;
+    /* Correct nbWorkers to 0. */
+    jobParams.nbWorkers = 0;
+
+
+    /* init */
+    if (job->cdict) {
+        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast, job->cdict, &jobParams, job->fullFrameSize);
+        assert(job->firstJob);  /* only allowed for first job */
+        if (ZSTD_isError(initError)) JOB_ERROR(initError);
+    } else {  /* srcStart points at reloaded section */
+        U64 const pledgedSrcSize = job->firstJob ? job->fullFrameSize : job->src.size;
+        {   size_t const forceWindowError = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_forceMaxWindow, !job->firstJob);
+            if (ZSTD_isError(forceWindowError)) JOB_ERROR(forceWindowError);
+        }
+        if (!job->firstJob) {
+            size_t const err = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_deterministicRefPrefix, 0);
+            if (ZSTD_isError(err)) JOB_ERROR(err);
+        }
+        {   size_t const initError = ZSTD_compressBegin_advanced_internal(cctx,
+                                        job->prefix.start, job->prefix.size, ZSTD_dct_rawContent, /* load dictionary in "content-only" mode (no header analysis) */
+                                        ZSTD_dtlm_fast,
+                                        NULL, /*cdict*/
+                                        &jobParams, pledgedSrcSize);
+            if (ZSTD_isError(initError)) JOB_ERROR(initError);
+    }   }
+
+    /* Perform serial step as early as possible, but after CCtx initialization */
+    ZSTDMT_serialState_update(job->serial, cctx, rawSeqStore, job->src, job->jobID);
+
+    if (!job->firstJob) {  /* flush and overwrite frame header when it's not first job */
+        size_t const hSize = ZSTD_compressContinue_public(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
+        if (ZSTD_isError(hSize)) JOB_ERROR(hSize);
+        DEBUGLOG(5, "ZSTDMT_compressionJob: flush and overwrite %u bytes of frame header (not first job)", (U32)hSize);
+        ZSTD_invalidateRepCodes(cctx);
+    }
+
+    /* compress */
+    {   size_t const chunkSize = 4*ZSTD_BLOCKSIZE_MAX;
+        int const nbChunks = (int)((job->src.size + (chunkSize-1)) / chunkSize);
+        const BYTE* ip = (const BYTE*) job->src.start;
+        BYTE* const ostart = (BYTE*)dstBuff.start;
+        BYTE* op = ostart;
+        BYTE* oend = op + dstBuff.capacity;
+        int chunkNb;
+        if (sizeof(size_t) > sizeof(int)) assert(job->src.size < ((size_t)INT_MAX) * chunkSize);   /* check overflow */
+        DEBUGLOG(5, "ZSTDMT_compressionJob: compress %u bytes in %i blocks", (U32)job->src.size, nbChunks);
+        assert(job->cSize == 0);
+        for (chunkNb = 1; chunkNb < nbChunks; chunkNb++) {
+            size_t const cSize = ZSTD_compressContinue_public(cctx, op, oend-op, ip, chunkSize);
+            if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
+            ip += chunkSize;
+            op += cSize; assert(op < oend);
+            /* stats */
+            ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+            job->cSize += cSize;
+            job->consumed = chunkSize * chunkNb;
+            DEBUGLOG(5, "ZSTDMT_compressionJob: compress new block : cSize==%u bytes (total: %u)",
+                        (U32)cSize, (U32)job->cSize);
+            ZSTD_pthread_cond_signal(&job->job_cond);   /* warns some more data is ready to be flushed */
+            ZSTD_pthread_mutex_unlock(&job->job_mutex);
+        }
+        /* last block */
+        assert(chunkSize > 0);
+        assert((chunkSize & (chunkSize - 1)) == 0);  /* chunkSize must be power of 2 for mask==(chunkSize-1) to work */
+        if ((nbChunks > 0) | job->lastJob /*must output a "last block" flag*/ ) {
+            size_t const lastBlockSize1 = job->src.size & (chunkSize-1);
+            size_t const lastBlockSize = ((lastBlockSize1==0) & (job->src.size>=chunkSize)) ? chunkSize : lastBlockSize1;
+            size_t const cSize = (job->lastJob) ?
+                 ZSTD_compressEnd_public(cctx, op, oend-op, ip, lastBlockSize) :
+                 ZSTD_compressContinue_public(cctx, op, oend-op, ip, lastBlockSize);
+            if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
+            lastCBlockSize = cSize;
+    }   }
+    if (!job->firstJob) {
+        /* Double check that we don't have an ext-dict, because then our
+         * repcode invalidation doesn't work.
+         */
+        assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
+    }
+    ZSTD_CCtx_trace(cctx, 0);
+
+_endJob:
+    ZSTDMT_serialState_ensureFinished(job->serial, job->jobID, job->cSize);
+    if (job->prefix.size > 0)
+        DEBUGLOG(5, "Finished with prefix: %zx", (size_t)job->prefix.start);
+    DEBUGLOG(5, "Finished with source: %zx", (size_t)job->src.start);
+    /* release resources */
+    ZSTDMT_releaseSeq(job->seqPool, rawSeqStore);
+    ZSTDMT_releaseCCtx(job->cctxPool, cctx);
+    /* report */
+    ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+    if (ZSTD_isError(job->cSize)) assert(lastCBlockSize == 0);
+    job->cSize += lastCBlockSize;
+    job->consumed = job->src.size;  /* when job->consumed == job->src.size , compression job is presumed completed */
+    ZSTD_pthread_cond_signal(&job->job_cond);
+    ZSTD_pthread_mutex_unlock(&job->job_mutex);
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+typedef struct {
+    range_t prefix;         /* read-only non-owned prefix buffer */
+    buffer_t buffer;
+    size_t filled;
+} inBuff_t;
+
+typedef struct {
+  BYTE* buffer;     /* The round input buffer. All jobs get references
+                     * to pieces of the buffer. ZSTDMT_tryGetInputRange()
+                     * handles handing out job input buffers, and makes
+                     * sure it doesn't overlap with any pieces still in use.
+                     */
+  size_t capacity;  /* The capacity of buffer. */
+  size_t pos;       /* The position of the current inBuff in the round
+                     * buffer. Updated past the end if the inBuff once
+                     * the inBuff is sent to the worker thread.
+                     * pos <= capacity.
+                     */
+} roundBuff_t;
+
+static const roundBuff_t kNullRoundBuff = {NULL, 0, 0};
+
+#define RSYNC_LENGTH 32
+/* Don't create chunks smaller than the zstd block size.
+ * This stops us from regressing compression ratio too much,
+ * and ensures our output fits in ZSTD_compressBound().
+ *
+ * If this is shrunk < ZSTD_BLOCKSIZELOG_MIN then
+ * ZSTD_COMPRESSBOUND() will need to be updated.
+ */
+#define RSYNC_MIN_BLOCK_LOG ZSTD_BLOCKSIZELOG_MAX
+#define RSYNC_MIN_BLOCK_SIZE (1<<RSYNC_MIN_BLOCK_LOG)
+
+typedef struct {
+  U64 hash;
+  U64 hitMask;
+  U64 primePower;
+} rsyncState_t;
+
+struct ZSTDMT_CCtx_s {
+    POOL_ctx* factory;
+    ZSTDMT_jobDescription* jobs;
+    ZSTDMT_bufferPool* bufPool;
+    ZSTDMT_CCtxPool* cctxPool;
+    ZSTDMT_seqPool* seqPool;
+    ZSTD_CCtx_params params;
+    size_t targetSectionSize;
+    size_t targetPrefixSize;
+    int jobReady;        /* 1 => one job is already prepared, but pool has shortage of workers. Don't create a new job. */
+    inBuff_t inBuff;
+    roundBuff_t roundBuff;
+    serialState_t serial;
+    rsyncState_t rsync;
+    unsigned jobIDMask;
+    unsigned doneJobID;
+    unsigned nextJobID;
+    unsigned frameEnded;
+    unsigned allJobsCompleted;
+    unsigned long long frameContentSize;
+    unsigned long long consumed;
+    unsigned long long produced;
+    ZSTD_customMem cMem;
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
+    unsigned providedFactory: 1;
+};
+
+static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZSTD_customMem cMem)
+{
+    U32 jobNb;
+    if (jobTable == NULL) return;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        ZSTD_pthread_mutex_destroy(&jobTable[jobNb].job_mutex);
+        ZSTD_pthread_cond_destroy(&jobTable[jobNb].job_cond);
+    }
+    ZSTD_customFree(jobTable, cMem);
+}
+
+/* ZSTDMT_allocJobsTable()
+ * allocate and init a job table.
+ * update *nbJobsPtr to next power of 2 value, as size of table */
+static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_customMem cMem)
+{
+    U32 const nbJobsLog2 = ZSTD_highbit32(*nbJobsPtr) + 1;
+    U32 const nbJobs = 1 << nbJobsLog2;
+    U32 jobNb;
+    ZSTDMT_jobDescription* const jobTable = (ZSTDMT_jobDescription*)
+                ZSTD_customCalloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
+    int initError = 0;
+    if (jobTable==NULL) return NULL;
+    *nbJobsPtr = nbJobs;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        initError |= ZSTD_pthread_mutex_init(&jobTable[jobNb].job_mutex, NULL);
+        initError |= ZSTD_pthread_cond_init(&jobTable[jobNb].job_cond, NULL);
+    }
+    if (initError != 0) {
+        ZSTDMT_freeJobsTable(jobTable, nbJobs, cMem);
+        return NULL;
+    }
+    return jobTable;
+}
+
+static size_t ZSTDMT_expandJobsTable (ZSTDMT_CCtx* mtctx, U32 nbWorkers) {
+    U32 nbJobs = nbWorkers + 2;
+    if (nbJobs > mtctx->jobIDMask+1) {  /* need more job capacity */
+        ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+        mtctx->jobIDMask = 0;
+        mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, mtctx->cMem);
+        if (mtctx->jobs==NULL) return ERROR(memory_allocation);
+        assert((nbJobs != 0) && ((nbJobs & (nbJobs - 1)) == 0));  /* ensure nbJobs is a power of 2 */
+        mtctx->jobIDMask = nbJobs - 1;
+    }
+    return 0;
+}
+
+
+/* ZSTDMT_CCtxParam_setNbWorkers():
+ * Internal use only */
+static size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
+{
+    return ZSTD_CCtxParams_setParameter(params, ZSTD_c_nbWorkers, (int)nbWorkers);
+}
+
+MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
+{
+    ZSTDMT_CCtx* mtctx;
+    U32 nbJobs = nbWorkers + 2;
+    int initError;
+    DEBUGLOG(3, "ZSTDMT_createCCtx_advanced (nbWorkers = %u)", nbWorkers);
+
+    if (nbWorkers < 1) return NULL;
+    nbWorkers = MIN(nbWorkers , ZSTDMT_NBWORKERS_MAX);
+    if ((cMem.customAlloc!=NULL) ^ (cMem.customFree!=NULL))
+        /* invalid custom allocator */
+        return NULL;
+
+    mtctx = (ZSTDMT_CCtx*) ZSTD_customCalloc(sizeof(ZSTDMT_CCtx), cMem);
+    if (!mtctx) return NULL;
+    ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
+    mtctx->cMem = cMem;
+    mtctx->allJobsCompleted = 1;
+    if (pool != NULL) {
+      mtctx->factory = pool;
+      mtctx->providedFactory = 1;
+    }
+    else {
+      mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
+      mtctx->providedFactory = 0;
+    }
+    mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, cMem);
+    assert(nbJobs > 0); assert((nbJobs & (nbJobs - 1)) == 0);  /* ensure nbJobs is a power of 2 */
+    mtctx->jobIDMask = nbJobs - 1;
+    mtctx->bufPool = ZSTDMT_createBufferPool(BUF_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
+    mtctx->cctxPool = ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    mtctx->seqPool = ZSTDMT_createSeqPool(nbWorkers, cMem);
+    initError = ZSTDMT_serialState_init(&mtctx->serial);
+    mtctx->roundBuff = kNullRoundBuff;
+    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool | !mtctx->seqPool | initError) {
+        ZSTDMT_freeCCtx(mtctx);
+        return NULL;
+    }
+    DEBUGLOG(3, "mt_cctx created, for %u threads", nbWorkers);
+    return mtctx;
+}
+
+ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
+{
+#ifdef ZSTD_MULTITHREAD
+    return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem, pool);
+#else
+    (void)nbWorkers;
+    (void)cMem;
+    (void)pool;
+    return NULL;
+#endif
+}
+
+
+/* ZSTDMT_releaseAllJobResources() :
+ * note : ensure all workers are killed first ! */
+static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
+{
+    unsigned jobID;
+    DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
+    for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
+        /* Copy the mutex/cond out */
+        ZSTD_pthread_mutex_t const mutex = mtctx->jobs[jobID].job_mutex;
+        ZSTD_pthread_cond_t const cond = mtctx->jobs[jobID].job_cond;
+
+        DEBUGLOG(4, "job%02u: release dst address %08X", jobID, (U32)(size_t)mtctx->jobs[jobID].dstBuff.start);
+        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
+
+        /* Clear the job description, but keep the mutex/cond */
+        ZSTD_memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
+        mtctx->jobs[jobID].job_mutex = mutex;
+        mtctx->jobs[jobID].job_cond = cond;
+    }
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
+    mtctx->allJobsCompleted = 1;
+}
+
+static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
+{
+    DEBUGLOG(4, "ZSTDMT_waitForAllJobsCompleted");
+    while (mtctx->doneJobID < mtctx->nextJobID) {
+        unsigned const jobID = mtctx->doneJobID & mtctx->jobIDMask;
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
+        while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
+            DEBUGLOG(4, "waiting for jobCompleted signal from job %u", mtctx->doneJobID);   /* we want to block when waiting for data to flush */
+            ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
+        }
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
+        mtctx->doneJobID++;
+    }
+}
+
+size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
+{
+    if (mtctx==NULL) return 0;   /* compatible with free on NULL */
+    if (!mtctx->providedFactory)
+        POOL_free(mtctx->factory);   /* stop and free worker threads */
+    ZSTDMT_releaseAllJobResources(mtctx);  /* release job resources into pools first */
+    ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+    ZSTDMT_freeBufferPool(mtctx->bufPool);
+    ZSTDMT_freeCCtxPool(mtctx->cctxPool);
+    ZSTDMT_freeSeqPool(mtctx->seqPool);
+    ZSTDMT_serialState_free(&mtctx->serial);
+    ZSTD_freeCDict(mtctx->cdictLocal);
+    if (mtctx->roundBuff.buffer)
+        ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
+    ZSTD_customFree(mtctx, mtctx->cMem);
+    return 0;
+}
+
+size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
+{
+    if (mtctx == NULL) return 0;   /* supports sizeof NULL */
+    return sizeof(*mtctx)
+            + POOL_sizeof(mtctx->factory)
+            + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
+            + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
+            + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
+            + ZSTDMT_sizeof_seqPool(mtctx->seqPool)
+            + ZSTD_sizeof_CDict(mtctx->cdictLocal)
+            + mtctx->roundBuff.capacity;
+}
+
+
+/* ZSTDMT_resize() :
+ * @return : error code if fails, 0 on success */
+static size_t ZSTDMT_resize(ZSTDMT_CCtx* mtctx, unsigned nbWorkers)
+{
+    if (POOL_resize(mtctx->factory, nbWorkers)) return ERROR(memory_allocation);
+    FORWARD_IF_ERROR( ZSTDMT_expandJobsTable(mtctx, nbWorkers) , "");
+    mtctx->bufPool = ZSTDMT_expandBufferPool(mtctx->bufPool, BUF_POOL_MAX_NB_BUFFERS(nbWorkers));
+    if (mtctx->bufPool == NULL) return ERROR(memory_allocation);
+    mtctx->cctxPool = ZSTDMT_expandCCtxPool(mtctx->cctxPool, nbWorkers);
+    if (mtctx->cctxPool == NULL) return ERROR(memory_allocation);
+    mtctx->seqPool = ZSTDMT_expandSeqPool(mtctx->seqPool, nbWorkers);
+    if (mtctx->seqPool == NULL) return ERROR(memory_allocation);
+    ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
+    return 0;
+}
+
+
+/*! ZSTDMT_updateCParams_whileCompressing() :
+ *  Updates a selected set of compression parameters, remaining compatible with currently active frame.
+ *  New parameters will be applied to next compression job. */
+void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams)
+{
+    U32 const saved_wlog = mtctx->params.cParams.windowLog;   /* Do not modify windowLog while compressing */
+    int const compressionLevel = cctxParams->compressionLevel;
+    DEBUGLOG(5, "ZSTDMT_updateCParams_whileCompressing (level:%i)",
+                compressionLevel);
+    mtctx->params.compressionLevel = compressionLevel;
+    {   ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+        cParams.windowLog = saved_wlog;
+        mtctx->params.cParams = cParams;
+    }
+}
+
+/* ZSTDMT_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads.
+ * Note : mutex will be acquired during statistics collection inside workers. */
+ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx)
+{
+    ZSTD_frameProgression fps;
+    DEBUGLOG(5, "ZSTDMT_getFrameProgression");
+    fps.ingested = mtctx->consumed + mtctx->inBuff.filled;
+    fps.consumed = mtctx->consumed;
+    fps.produced = fps.flushed = mtctx->produced;
+    fps.currentJobID = mtctx->nextJobID;
+    fps.nbActiveWorkers = 0;
+    {   unsigned jobNb;
+        unsigned lastJobNb = mtctx->nextJobID + mtctx->jobReady; assert(mtctx->jobReady <= 1);
+        DEBUGLOG(6, "ZSTDMT_getFrameProgression: jobs: from %u to <%u (jobReady:%u)",
+                    mtctx->doneJobID, lastJobNb, mtctx->jobReady)
+        for (jobNb = mtctx->doneJobID ; jobNb < lastJobNb ; jobNb++) {
+            unsigned const wJobID = jobNb & mtctx->jobIDMask;
+            ZSTDMT_jobDescription* jobPtr = &mtctx->jobs[wJobID];
+            ZSTD_pthread_mutex_lock(&jobPtr->job_mutex);
+            {   size_t const cResult = jobPtr->cSize;
+                size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
+                size_t const flushed = ZSTD_isError(cResult) ? 0 : jobPtr->dstFlushed;
+                assert(flushed <= produced);
+                fps.ingested += jobPtr->src.size;
+                fps.consumed += jobPtr->consumed;
+                fps.produced += produced;
+                fps.flushed  += flushed;
+                fps.nbActiveWorkers += (jobPtr->consumed < jobPtr->src.size);
+            }
+            ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        }
+    }
+    return fps;
+}
+
+
+size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
+{
+    size_t toFlush;
+    unsigned const jobID = mtctx->doneJobID;
+    assert(jobID <= mtctx->nextJobID);
+    if (jobID == mtctx->nextJobID) return 0;   /* no active job => nothing to flush */
+
+    /* look into oldest non-fully-flushed job */
+    {   unsigned const wJobID = jobID & mtctx->jobIDMask;
+        ZSTDMT_jobDescription* const jobPtr = &mtctx->jobs[wJobID];
+        ZSTD_pthread_mutex_lock(&jobPtr->job_mutex);
+        {   size_t const cResult = jobPtr->cSize;
+            size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
+            size_t const flushed = ZSTD_isError(cResult) ? 0 : jobPtr->dstFlushed;
+            assert(flushed <= produced);
+            assert(jobPtr->consumed <= jobPtr->src.size);
+            toFlush = produced - flushed;
+            /* if toFlush==0, nothing is available to flush.
+             * However, jobID is expected to still be active:
+             * if jobID was already completed and fully flushed,
+             * ZSTDMT_flushProduced() should have already moved onto next job.
+             * Therefore, some input has not yet been consumed. */
+            if (toFlush==0) {
+                assert(jobPtr->consumed < jobPtr->src.size);
+            }
+        }
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+    }
+
+    return toFlush;
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
+{
+    unsigned jobLog;
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+        /* In Long Range Mode, the windowLog is typically oversized.
+         * In which case, it's preferable to determine the jobSize
+         * based on cycleLog instead. */
+        jobLog = MAX(21, ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy) + 3);
+    } else {
+        jobLog = MAX(20, params->cParams.windowLog + 2);
+    }
+    return MIN(jobLog, (unsigned)ZSTDMT_JOBLOG_MAX);
+}
+
+static int ZSTDMT_overlapLog_default(ZSTD_strategy strat)
+{
+    switch(strat)
+    {
+        case ZSTD_btultra2:
+            return 9;
+        case ZSTD_btultra:
+        case ZSTD_btopt:
+            return 8;
+        case ZSTD_btlazy2:
+        case ZSTD_lazy2:
+            return 7;
+        case ZSTD_lazy:
+        case ZSTD_greedy:
+        case ZSTD_dfast:
+        case ZSTD_fast:
+        default:;
+    }
+    return 6;
+}
+
+static int ZSTDMT_overlapLog(int ovlog, ZSTD_strategy strat)
+{
+    assert(0 <= ovlog && ovlog <= 9);
+    if (ovlog == 0) return ZSTDMT_overlapLog_default(strat);
+    return ovlog;
+}
+
+static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
+{
+    int const overlapRLog = 9 - ZSTDMT_overlapLog(params->overlapLog, params->cParams.strategy);
+    int ovLog = (overlapRLog >= 8) ? 0 : (params->cParams.windowLog - overlapRLog);
+    assert(0 <= overlapRLog && overlapRLog <= 8);
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+        /* In Long Range Mode, the windowLog is typically oversized.
+         * In which case, it's preferable to determine the jobSize
+         * based on chainLog instead.
+         * Then, ovLog becomes a fraction of the jobSize, rather than windowSize */
+        ovLog = MIN(params->cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2)
+                - overlapRLog;
+    }
+    assert(0 <= ovLog && ovLog <= ZSTD_WINDOWLOG_MAX);
+    DEBUGLOG(4, "overlapLog : %i", params->overlapLog);
+    DEBUGLOG(4, "overlap size : %i", 1 << ovLog);
+    return (ovLog==0) ? 0 : (size_t)1 << ovLog;
+}
+
+/* ====================================== */
+/* =======      Streaming API     ======= */
+/* ====================================== */
+
+size_t ZSTDMT_initCStream_internal(
+        ZSTDMT_CCtx* mtctx,
+        const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
+        const ZSTD_CDict* cdict, ZSTD_CCtx_params params,
+        unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal (pledgedSrcSize=%u, nbWorkers=%u, cctxPool=%u)",
+                (U32)pledgedSrcSize, params.nbWorkers, mtctx->cctxPool->totalCCtx);
+
+    /* params supposed partially fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+
+    /* init */
+    if (params.nbWorkers != mtctx->params.nbWorkers)
+        FORWARD_IF_ERROR( ZSTDMT_resize(mtctx, params.nbWorkers) , "");
+
+    if (params.jobSize != 0 && params.jobSize < ZSTDMT_JOBSIZE_MIN) params.jobSize = ZSTDMT_JOBSIZE_MIN;
+    if (params.jobSize > (size_t)ZSTDMT_JOBSIZE_MAX) params.jobSize = (size_t)ZSTDMT_JOBSIZE_MAX;
+
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal: %u workers", params.nbWorkers);
+
+    if (mtctx->allJobsCompleted == 0) {   /* previous compression not correctly finished */
+        ZSTDMT_waitForAllJobsCompleted(mtctx);
+        ZSTDMT_releaseAllJobResources(mtctx);
+        mtctx->allJobsCompleted = 1;
+    }
+
+    mtctx->params = params;
+    mtctx->frameContentSize = pledgedSrcSize;
+    if (dict) {
+        ZSTD_freeCDict(mtctx->cdictLocal);
+        mtctx->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                                    ZSTD_dlm_byCopy, dictContentType, /* note : a loadPrefix becomes an internal CDict */
+                                                    params.cParams, mtctx->cMem);
+        mtctx->cdict = mtctx->cdictLocal;
+        if (mtctx->cdictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        ZSTD_freeCDict(mtctx->cdictLocal);
+        mtctx->cdictLocal = NULL;
+        mtctx->cdict = cdict;
+    }
+
+    mtctx->targetPrefixSize = ZSTDMT_computeOverlapSize(&params);
+    DEBUGLOG(4, "overlapLog=%i => %u KB", params.overlapLog, (U32)(mtctx->targetPrefixSize>>10));
+    mtctx->targetSectionSize = params.jobSize;
+    if (mtctx->targetSectionSize == 0) {
+        mtctx->targetSectionSize = 1ULL << ZSTDMT_computeTargetJobLog(&params);
+    }
+    assert(mtctx->targetSectionSize <= (size_t)ZSTDMT_JOBSIZE_MAX);
+
+    if (params.rsyncable) {
+        /* Aim for the targetsectionSize as the average job size. */
+        U32 const jobSizeKB = (U32)(mtctx->targetSectionSize >> 10);
+        U32 const rsyncBits = (assert(jobSizeKB >= 1), ZSTD_highbit32(jobSizeKB) + 10);
+        /* We refuse to create jobs < RSYNC_MIN_BLOCK_SIZE bytes, so make sure our
+         * expected job size is at least 4x larger. */
+        assert(rsyncBits >= RSYNC_MIN_BLOCK_LOG + 2);
+        DEBUGLOG(4, "rsyncLog = %u", rsyncBits);
+        mtctx->rsync.hash = 0;
+        mtctx->rsync.hitMask = (1ULL << rsyncBits) - 1;
+        mtctx->rsync.primePower = ZSTD_rollingHash_primePower(RSYNC_LENGTH);
+    }
+    if (mtctx->targetSectionSize < mtctx->targetPrefixSize) mtctx->targetSectionSize = mtctx->targetPrefixSize;  /* job size must be >= overlap size */
+    DEBUGLOG(4, "Job Size : %u KB (note : set to %u)", (U32)(mtctx->targetSectionSize>>10), (U32)params.jobSize);
+    DEBUGLOG(4, "inBuff Size : %u KB", (U32)(mtctx->targetSectionSize>>10));
+    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
+    {
+        /* If ldm is enabled we need windowSize space. */
+        size_t const windowSize = mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable ? (1U << mtctx->params.cParams.windowLog) : 0;
+        /* Two buffers of slack, plus extra space for the overlap
+         * This is the minimum slack that LDM works with. One extra because
+         * flush might waste up to targetSectionSize-1 bytes. Another extra
+         * for the overlap (if > 0), then one to fill which doesn't overlap
+         * with the LDM window.
+         */
+        size_t const nbSlackBuffers = 2 + (mtctx->targetPrefixSize > 0);
+        size_t const slackSize = mtctx->targetSectionSize * nbSlackBuffers;
+        /* Compute the total size, and always have enough slack */
+        size_t const nbWorkers = MAX(mtctx->params.nbWorkers, 1);
+        size_t const sectionsSize = mtctx->targetSectionSize * nbWorkers;
+        size_t const capacity = MAX(windowSize, sectionsSize) + slackSize;
+        if (mtctx->roundBuff.capacity < capacity) {
+            if (mtctx->roundBuff.buffer)
+                ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
+            mtctx->roundBuff.buffer = (BYTE*)ZSTD_customMalloc(capacity, mtctx->cMem);
+            if (mtctx->roundBuff.buffer == NULL) {
+                mtctx->roundBuff.capacity = 0;
+                return ERROR(memory_allocation);
+            }
+            mtctx->roundBuff.capacity = capacity;
+        }
+    }
+    DEBUGLOG(4, "roundBuff capacity : %u KB", (U32)(mtctx->roundBuff.capacity>>10));
+    mtctx->roundBuff.pos = 0;
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
+    mtctx->inBuff.prefix = kNullRange;
+    mtctx->doneJobID = 0;
+    mtctx->nextJobID = 0;
+    mtctx->frameEnded = 0;
+    mtctx->allJobsCompleted = 0;
+    mtctx->consumed = 0;
+    mtctx->produced = 0;
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, mtctx->targetSectionSize,
+                                 dict, dictSize, dictContentType))
+        return ERROR(memory_allocation);
+    return 0;
+}
+
+
+/* ZSTDMT_writeLastEmptyBlock()
+ * Write a single empty block with an end-of-frame to finish a frame.
+ * Job must be created from streaming variant.
+ * This function is always successful if expected conditions are fulfilled.
+ */
+static void ZSTDMT_writeLastEmptyBlock(ZSTDMT_jobDescription* job)
+{
+    assert(job->lastJob == 1);
+    assert(job->src.size == 0);   /* last job is empty -> will be simplified into a last empty block */
+    assert(job->firstJob == 0);   /* cannot be first job, as it also needs to create frame header */
+    assert(job->dstBuff.start == NULL);   /* invoked from streaming variant only (otherwise, dstBuff might be user's output) */
+    job->dstBuff = ZSTDMT_getBuffer(job->bufPool);
+    if (job->dstBuff.start == NULL) {
+      job->cSize = ERROR(memory_allocation);
+      return;
+    }
+    assert(job->dstBuff.capacity >= ZSTD_blockHeaderSize);   /* no buffer should ever be that small */
+    job->src = kNullRange;
+    job->cSize = ZSTD_writeLastEmptyBlock(job->dstBuff.start, job->dstBuff.capacity);
+    assert(!ZSTD_isError(job->cSize));
+    assert(job->consumed == 0);
+}
+
+static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* mtctx, size_t srcSize, ZSTD_EndDirective endOp)
+{
+    unsigned const jobID = mtctx->nextJobID & mtctx->jobIDMask;
+    int const endFrame = (endOp == ZSTD_e_end);
+
+    if (mtctx->nextJobID > mtctx->doneJobID + mtctx->jobIDMask) {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: will not create new job : table is full");
+        assert((mtctx->nextJobID & mtctx->jobIDMask) == (mtctx->doneJobID & mtctx->jobIDMask));
+        return 0;
+    }
+
+    if (!mtctx->jobReady) {
+        BYTE const* src = (BYTE const*)mtctx->inBuff.buffer.start;
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: preparing job %u to compress %u bytes with %u preload ",
+                    mtctx->nextJobID, (U32)srcSize, (U32)mtctx->inBuff.prefix.size);
+        mtctx->jobs[jobID].src.start = src;
+        mtctx->jobs[jobID].src.size = srcSize;
+        assert(mtctx->inBuff.filled >= srcSize);
+        mtctx->jobs[jobID].prefix = mtctx->inBuff.prefix;
+        mtctx->jobs[jobID].consumed = 0;
+        mtctx->jobs[jobID].cSize = 0;
+        mtctx->jobs[jobID].params = mtctx->params;
+        mtctx->jobs[jobID].cdict = mtctx->nextJobID==0 ? mtctx->cdict : NULL;
+        mtctx->jobs[jobID].fullFrameSize = mtctx->frameContentSize;
+        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+        mtctx->jobs[jobID].cctxPool = mtctx->cctxPool;
+        mtctx->jobs[jobID].bufPool = mtctx->bufPool;
+        mtctx->jobs[jobID].seqPool = mtctx->seqPool;
+        mtctx->jobs[jobID].serial = &mtctx->serial;
+        mtctx->jobs[jobID].jobID = mtctx->nextJobID;
+        mtctx->jobs[jobID].firstJob = (mtctx->nextJobID==0);
+        mtctx->jobs[jobID].lastJob = endFrame;
+        mtctx->jobs[jobID].frameChecksumNeeded = mtctx->params.fParams.checksumFlag && endFrame && (mtctx->nextJobID>0);
+        mtctx->jobs[jobID].dstFlushed = 0;
+
+        /* Update the round buffer pos and clear the input buffer to be reset */
+        mtctx->roundBuff.pos += srcSize;
+        mtctx->inBuff.buffer = g_nullBuffer;
+        mtctx->inBuff.filled = 0;
+        /* Set the prefix */
+        if (!endFrame) {
+            size_t const newPrefixSize = MIN(srcSize, mtctx->targetPrefixSize);
+            mtctx->inBuff.prefix.start = src + srcSize - newPrefixSize;
+            mtctx->inBuff.prefix.size = newPrefixSize;
+        } else {   /* endFrame==1 => no need for another input buffer */
+            mtctx->inBuff.prefix = kNullRange;
+            mtctx->frameEnded = endFrame;
+            if (mtctx->nextJobID == 0) {
+                /* single job exception : checksum is already calculated directly within worker thread */
+                mtctx->params.fParams.checksumFlag = 0;
+        }   }
+
+        if ( (srcSize == 0)
+          && (mtctx->nextJobID>0)/*single job must also write frame header*/ ) {
+            DEBUGLOG(5, "ZSTDMT_createCompressionJob: creating a last empty block to end frame");
+            assert(endOp == ZSTD_e_end);  /* only possible case : need to end the frame with an empty last block */
+            ZSTDMT_writeLastEmptyBlock(mtctx->jobs + jobID);
+            mtctx->nextJobID++;
+            return 0;
+        }
+    }
+
+    DEBUGLOG(5, "ZSTDMT_createCompressionJob: posting job %u : %u bytes  (end:%u, jobNb == %u (mod:%u))",
+                mtctx->nextJobID,
+                (U32)mtctx->jobs[jobID].src.size,
+                mtctx->jobs[jobID].lastJob,
+                mtctx->nextJobID,
+                jobID);
+    if (POOL_tryAdd(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[jobID])) {
+        mtctx->nextJobID++;
+        mtctx->jobReady = 0;
+    } else {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: no worker available for job %u", mtctx->nextJobID);
+        mtctx->jobReady = 1;
+    }
+    return 0;
+}
+
+
+/*! ZSTDMT_flushProduced() :
+ *  flush whatever data has been produced but not yet flushed in current job.
+ *  move to next job if current one is fully flushed.
+ * `output` : `pos` will be updated with amount of data flushed .
+ * `blockToFlush` : if >0, the function will block and wait if there is no data available to flush .
+ * @return : amount of data remaining within internal buffer, 0 if no more, 1 if unknown but > 0, or an error code */
+static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, unsigned blockToFlush, ZSTD_EndDirective end)
+{
+    unsigned const wJobID = mtctx->doneJobID & mtctx->jobIDMask;
+    DEBUGLOG(5, "ZSTDMT_flushProduced (blocking:%u , job %u <= %u)",
+                blockToFlush, mtctx->doneJobID, mtctx->nextJobID);
+    assert(output->size >= output->pos);
+
+    ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+    if (  blockToFlush
+      && (mtctx->doneJobID < mtctx->nextJobID) ) {
+        assert(mtctx->jobs[wJobID].dstFlushed <= mtctx->jobs[wJobID].cSize);
+        while (mtctx->jobs[wJobID].dstFlushed == mtctx->jobs[wJobID].cSize) {  /* nothing to flush */
+            if (mtctx->jobs[wJobID].consumed == mtctx->jobs[wJobID].src.size) {
+                DEBUGLOG(5, "job %u is completely consumed (%u == %u) => don't wait for cond, there will be none",
+                            mtctx->doneJobID, (U32)mtctx->jobs[wJobID].consumed, (U32)mtctx->jobs[wJobID].src.size);
+                break;
+            }
+            DEBUGLOG(5, "waiting for something to flush from job %u (currently flushed: %u bytes)",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+            ZSTD_pthread_cond_wait(&mtctx->jobs[wJobID].job_cond, &mtctx->jobs[wJobID].job_mutex);  /* block when nothing to flush but some to come */
+    }   }
+
+    /* try to flush something */
+    {   size_t cSize = mtctx->jobs[wJobID].cSize;                  /* shared */
+        size_t const srcConsumed = mtctx->jobs[wJobID].consumed;   /* shared */
+        size_t const srcSize = mtctx->jobs[wJobID].src.size;       /* read-only, could be done after mutex lock, but no-declaration-after-statement */
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        if (ZSTD_isError(cSize)) {
+            DEBUGLOG(5, "ZSTDMT_flushProduced: job %u : compression error detected : %s",
+                        mtctx->doneJobID, ZSTD_getErrorName(cSize));
+            ZSTDMT_waitForAllJobsCompleted(mtctx);
+            ZSTDMT_releaseAllJobResources(mtctx);
+            return cSize;
+        }
+        /* add frame checksum if necessary (can only happen once) */
+        assert(srcConsumed <= srcSize);
+        if ( (srcConsumed == srcSize)   /* job completed -> worker no longer active */
+          && mtctx->jobs[wJobID].frameChecksumNeeded ) {
+            U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
+            DEBUGLOG(4, "ZSTDMT_flushProduced: writing checksum : %08X \n", checksum);
+            MEM_writeLE32((char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].cSize, checksum);
+            cSize += 4;
+            mtctx->jobs[wJobID].cSize += 4;  /* can write this shared value, as worker is no longer active */
+            mtctx->jobs[wJobID].frameChecksumNeeded = 0;
+        }
+
+        if (cSize > 0) {   /* compression is ongoing or completed */
+            size_t const toFlush = MIN(cSize - mtctx->jobs[wJobID].dstFlushed, output->size - output->pos);
+            DEBUGLOG(5, "ZSTDMT_flushProduced: Flushing %u bytes from job %u (completion:%u/%u, generated:%u)",
+                        (U32)toFlush, mtctx->doneJobID, (U32)srcConsumed, (U32)srcSize, (U32)cSize);
+            assert(mtctx->doneJobID < mtctx->nextJobID);
+            assert(cSize >= mtctx->jobs[wJobID].dstFlushed);
+            assert(mtctx->jobs[wJobID].dstBuff.start != NULL);
+            if (toFlush > 0) {
+                ZSTD_memcpy((char*)output->dst + output->pos,
+                    (const char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].dstFlushed,
+                    toFlush);
+            }
+            output->pos += toFlush;
+            mtctx->jobs[wJobID].dstFlushed += toFlush;  /* can write : this value is only used by mtctx */
+
+            if ( (srcConsumed == srcSize)    /* job is completed */
+              && (mtctx->jobs[wJobID].dstFlushed == cSize) ) {   /* output buffer fully flushed => free this job position */
+                DEBUGLOG(5, "Job %u completed (%u bytes), moving to next one",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+                ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[wJobID].dstBuff);
+                DEBUGLOG(5, "dstBuffer released");
+                mtctx->jobs[wJobID].dstBuff = g_nullBuffer;
+                mtctx->jobs[wJobID].cSize = 0;   /* ensure this job slot is considered "not started" in future check */
+                mtctx->consumed += srcSize;
+                mtctx->produced += cSize;
+                mtctx->doneJobID++;
+        }   }
+
+        /* return value : how many bytes left in buffer ; fake it to 1 when unknown but >0 */
+        if (cSize > mtctx->jobs[wJobID].dstFlushed) return (cSize - mtctx->jobs[wJobID].dstFlushed);
+        if (srcSize > srcConsumed) return 1;   /* current job not completely compressed */
+    }
+    if (mtctx->doneJobID < mtctx->nextJobID) return 1;   /* some more jobs ongoing */
+    if (mtctx->jobReady) return 1;      /* one job is ready to push, just not yet in the list */
+    if (mtctx->inBuff.filled > 0) return 1;   /* input is not empty, and still needs to be converted into a job */
+    mtctx->allJobsCompleted = mtctx->frameEnded;   /* all jobs are entirely flushed => if this one is last one, frame is completed */
+    if (end == ZSTD_e_end) return !mtctx->frameEnded;  /* for ZSTD_e_end, question becomes : is frame completed ? instead of : are internal buffers fully flushed ? */
+    return 0;   /* internal buffers fully flushed */
+}
+
+/**
+ * Returns the range of data used by the earliest job that is not yet complete.
+ * If the data of the first job is broken up into two segments, we cover both
+ * sections.
+ */
+static range_t ZSTDMT_getInputDataInUse(ZSTDMT_CCtx* mtctx)
+{
+    unsigned const firstJobID = mtctx->doneJobID;
+    unsigned const lastJobID = mtctx->nextJobID;
+    unsigned jobID;
+
+    for (jobID = firstJobID; jobID < lastJobID; ++jobID) {
+        unsigned const wJobID = jobID & mtctx->jobIDMask;
+        size_t consumed;
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+        consumed = mtctx->jobs[wJobID].consumed;
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+
+        if (consumed < mtctx->jobs[wJobID].src.size) {
+            range_t range = mtctx->jobs[wJobID].prefix;
+            if (range.size == 0) {
+                /* Empty prefix */
+                range = mtctx->jobs[wJobID].src;
+            }
+            /* Job source in multiple segments not supported yet */
+            assert(range.start <= mtctx->jobs[wJobID].src.start);
+            return range;
+        }
+    }
+    return kNullRange;
+}
+
+/**
+ * Returns non-zero iff buffer and range overlap.
+ */
+static int ZSTDMT_isOverlapped(buffer_t buffer, range_t range)
+{
+    BYTE const* const bufferStart = (BYTE const*)buffer.start;
+    BYTE const* const rangeStart = (BYTE const*)range.start;
+
+    if (rangeStart == NULL || bufferStart == NULL)
+        return 0;
+
+    {
+        BYTE const* const bufferEnd = bufferStart + buffer.capacity;
+        BYTE const* const rangeEnd = rangeStart + range.size;
+
+        /* Empty ranges cannot overlap */
+        if (bufferStart == bufferEnd || rangeStart == rangeEnd)
+            return 0;
+
+        return bufferStart < rangeEnd && rangeStart < bufferEnd;
+    }
+}
+
+static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
+{
+    range_t extDict;
+    range_t prefix;
+
+    DEBUGLOG(5, "ZSTDMT_doesOverlapWindow");
+    extDict.start = window.dictBase + window.lowLimit;
+    extDict.size = window.dictLimit - window.lowLimit;
+
+    prefix.start = window.base + window.dictLimit;
+    prefix.size = window.nextSrc - (window.base + window.dictLimit);
+    DEBUGLOG(5, "extDict [0x%zx, 0x%zx)",
+                (size_t)extDict.start,
+                (size_t)extDict.start + extDict.size);
+    DEBUGLOG(5, "prefix  [0x%zx, 0x%zx)",
+                (size_t)prefix.start,
+                (size_t)prefix.start + prefix.size);
+
+    return ZSTDMT_isOverlapped(buffer, extDict)
+        || ZSTDMT_isOverlapped(buffer, prefix);
+}
+
+static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
+{
+    if (mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
+        DEBUGLOG(5, "ZSTDMT_waitForLdmComplete");
+        DEBUGLOG(5, "source  [0x%zx, 0x%zx)",
+                    (size_t)buffer.start,
+                    (size_t)buffer.start + buffer.capacity);
+        ZSTD_PTHREAD_MUTEX_LOCK(mutex);
+        while (ZSTDMT_doesOverlapWindow(buffer, mtctx->serial.ldmWindow)) {
+            DEBUGLOG(5, "Waiting for LDM to finish...");
+            ZSTD_pthread_cond_wait(&mtctx->serial.ldmWindowCond, mutex);
+        }
+        DEBUGLOG(6, "Done waiting for LDM to finish");
+        ZSTD_pthread_mutex_unlock(mutex);
+    }
+}
+
+/**
+ * Attempts to set the inBuff to the next section to fill.
+ * If any part of the new section is still in use we give up.
+ * Returns non-zero if the buffer is filled.
+ */
+static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
+{
+    range_t const inUse = ZSTDMT_getInputDataInUse(mtctx);
+    size_t const spaceLeft = mtctx->roundBuff.capacity - mtctx->roundBuff.pos;
+    size_t const target = mtctx->targetSectionSize;
+    buffer_t buffer;
+
+    DEBUGLOG(5, "ZSTDMT_tryGetInputRange");
+    assert(mtctx->inBuff.buffer.start == NULL);
+    assert(mtctx->roundBuff.capacity >= target);
+
+    if (spaceLeft < target) {
+        /* ZSTD_invalidateRepCodes() doesn't work for extDict variants.
+         * Simply copy the prefix to the beginning in that case.
+         */
+        BYTE* const start = (BYTE*)mtctx->roundBuff.buffer;
+        size_t const prefixSize = mtctx->inBuff.prefix.size;
+
+        buffer.start = start;
+        buffer.capacity = prefixSize;
+        if (ZSTDMT_isOverlapped(buffer, inUse)) {
+            DEBUGLOG(5, "Waiting for buffer...");
+            return 0;
+        }
+        ZSTDMT_waitForLdmComplete(mtctx, buffer);
+        ZSTD_memmove(start, mtctx->inBuff.prefix.start, prefixSize);
+        mtctx->inBuff.prefix.start = start;
+        mtctx->roundBuff.pos = prefixSize;
+    }
+    buffer.start = mtctx->roundBuff.buffer + mtctx->roundBuff.pos;
+    buffer.capacity = target;
+
+    if (ZSTDMT_isOverlapped(buffer, inUse)) {
+        DEBUGLOG(5, "Waiting for buffer...");
+        return 0;
+    }
+    assert(!ZSTDMT_isOverlapped(buffer, mtctx->inBuff.prefix));
+
+    ZSTDMT_waitForLdmComplete(mtctx, buffer);
+
+    DEBUGLOG(5, "Using prefix range [%zx, %zx)",
+                (size_t)mtctx->inBuff.prefix.start,
+                (size_t)mtctx->inBuff.prefix.start + mtctx->inBuff.prefix.size);
+    DEBUGLOG(5, "Using source range [%zx, %zx)",
+                (size_t)buffer.start,
+                (size_t)buffer.start + buffer.capacity);
+
+
+    mtctx->inBuff.buffer = buffer;
+    mtctx->inBuff.filled = 0;
+    assert(mtctx->roundBuff.pos + buffer.capacity <= mtctx->roundBuff.capacity);
+    return 1;
+}
+
+typedef struct {
+  size_t toLoad;  /* The number of bytes to load from the input. */
+  int flush;      /* Boolean declaring if we must flush because we found a synchronization point. */
+} syncPoint_t;
+
+/**
+ * Searches through the input for a synchronization point. If one is found, we
+ * will instruct the caller to flush, and return the number of bytes to load.
+ * Otherwise, we will load as many bytes as possible and instruct the caller
+ * to continue as normal.
+ */
+static syncPoint_t
+findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
+{
+    BYTE const* const istart = (BYTE const*)input.src + input.pos;
+    U64 const primePower = mtctx->rsync.primePower;
+    U64 const hitMask = mtctx->rsync.hitMask;
+
+    syncPoint_t syncPoint;
+    U64 hash;
+    BYTE const* prev;
+    size_t pos;
+
+    syncPoint.toLoad = MIN(input.size - input.pos, mtctx->targetSectionSize - mtctx->inBuff.filled);
+    syncPoint.flush = 0;
+    if (!mtctx->params.rsyncable)
+        /* Rsync is disabled. */
+        return syncPoint;
+    if (mtctx->inBuff.filled + input.size - input.pos < RSYNC_MIN_BLOCK_SIZE)
+        /* We don't emit synchronization points if it would produce too small blocks.
+         * We don't have enough input to find a synchronization point, so don't look.
+         */
+        return syncPoint;
+    if (mtctx->inBuff.filled + syncPoint.toLoad < RSYNC_LENGTH)
+        /* Not enough to compute the hash.
+         * We will miss any synchronization points in this RSYNC_LENGTH byte
+         * window. However, since it depends only in the internal buffers, if the
+         * state is already synchronized, we will remain synchronized.
+         * Additionally, the probability that we miss a synchronization point is
+         * low: RSYNC_LENGTH / targetSectionSize.
+         */
+        return syncPoint;
+    /* Initialize the loop variables. */
+    if (mtctx->inBuff.filled < RSYNC_MIN_BLOCK_SIZE) {
+        /* We don't need to scan the first RSYNC_MIN_BLOCK_SIZE positions
+         * because they can't possibly be a sync point. So we can start
+         * part way through the input buffer.
+         */
+        pos = RSYNC_MIN_BLOCK_SIZE - mtctx->inBuff.filled;
+        if (pos >= RSYNC_LENGTH) {
+            prev = istart + pos - RSYNC_LENGTH;
+            hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
+        } else {
+            assert(mtctx->inBuff.filled >= RSYNC_LENGTH);
+            prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
+            hash = ZSTD_rollingHash_compute(prev + pos, (RSYNC_LENGTH - pos));
+            hash = ZSTD_rollingHash_append(hash, istart, pos);
+        }
+    } else {
+        /* We have enough bytes buffered to initialize the hash,
+         * and have processed enough bytes to find a sync point.
+         * Start scanning at the beginning of the input.
+         */
+        assert(mtctx->inBuff.filled >= RSYNC_MIN_BLOCK_SIZE);
+        assert(RSYNC_MIN_BLOCK_SIZE >= RSYNC_LENGTH);
+        pos = 0;
+        prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
+        hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
+        if ((hash & hitMask) == hitMask) {
+            /* We're already at a sync point so don't load any more until
+             * we're able to flush this sync point.
+             * This likely happened because the job table was full so we
+             * couldn't add our job.
+             */
+            syncPoint.toLoad = 0;
+            syncPoint.flush = 1;
+            return syncPoint;
+        }
+    }
+    /* Starting with the hash of the previous RSYNC_LENGTH bytes, roll
+     * through the input. If we hit a synchronization point, then cut the
+     * job off, and tell the compressor to flush the job. Otherwise, load
+     * all the bytes and continue as normal.
+     * If we go too long without a synchronization point (targetSectionSize)
+     * then a block will be emitted anyways, but this is okay, since if we
+     * are already synchronized we will remain synchronized.
+     */
+    assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
+    for (; pos < syncPoint.toLoad; ++pos) {
+        BYTE const toRemove = pos < RSYNC_LENGTH ? prev[pos] : istart[pos - RSYNC_LENGTH];
+        /* This assert is very expensive, and Debian compiles with asserts enabled.
+         * So disable it for now. We can get similar coverage by checking it at the
+         * beginning & end of the loop.
+         * assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
+         */
+        hash = ZSTD_rollingHash_rotate(hash, toRemove, istart[pos], primePower);
+        assert(mtctx->inBuff.filled + pos >= RSYNC_MIN_BLOCK_SIZE);
+        if ((hash & hitMask) == hitMask) {
+            syncPoint.toLoad = pos + 1;
+            syncPoint.flush = 1;
+            ++pos; /* for assert */
+            break;
+        }
+    }
+    assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
+    return syncPoint;
+}
+
+size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx)
+{
+    size_t hintInSize = mtctx->targetSectionSize - mtctx->inBuff.filled;
+    if (hintInSize==0) hintInSize = mtctx->targetSectionSize;
+    return hintInSize;
+}
+
+/** ZSTDMT_compressStream_generic() :
+ *  internal use only - exposed to be invoked from zstd_compress.c
+ *  assumption : output and input are valid (pos <= size)
+ * @return : minimum amount of data remaining to flush, 0 if none */
+size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                     ZSTD_outBuffer* output,
+                                     ZSTD_inBuffer* input,
+                                     ZSTD_EndDirective endOp)
+{
+    unsigned forwardInputProgress = 0;
+    DEBUGLOG(5, "ZSTDMT_compressStream_generic (endOp=%u, srcSize=%u)",
+                (U32)endOp, (U32)(input->size - input->pos));
+    assert(output->pos <= output->size);
+    assert(input->pos  <= input->size);
+
+    if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
+        /* current frame being ended. Only flush/end are allowed */
+        return ERROR(stage_wrong);
+    }
+
+    /* fill input buffer */
+    if ( (!mtctx->jobReady)
+      && (input->size > input->pos) ) {   /* support NULL input */
+        if (mtctx->inBuff.buffer.start == NULL) {
+            assert(mtctx->inBuff.filled == 0); /* Can't fill an empty buffer */
+            if (!ZSTDMT_tryGetInputRange(mtctx)) {
+                /* It is only possible for this operation to fail if there are
+                 * still compression jobs ongoing.
+                 */
+                DEBUGLOG(5, "ZSTDMT_tryGetInputRange failed");
+                assert(mtctx->doneJobID != mtctx->nextJobID);
+            } else
+                DEBUGLOG(5, "ZSTDMT_tryGetInputRange completed successfully : mtctx->inBuff.buffer.start = %p", mtctx->inBuff.buffer.start);
+        }
+        if (mtctx->inBuff.buffer.start != NULL) {
+            syncPoint_t const syncPoint = findSynchronizationPoint(mtctx, *input);
+            if (syncPoint.flush && endOp == ZSTD_e_continue) {
+                endOp = ZSTD_e_flush;
+            }
+            assert(mtctx->inBuff.buffer.capacity >= mtctx->targetSectionSize);
+            DEBUGLOG(5, "ZSTDMT_compressStream_generic: adding %u bytes on top of %u to buffer of size %u",
+                        (U32)syncPoint.toLoad, (U32)mtctx->inBuff.filled, (U32)mtctx->targetSectionSize);
+            ZSTD_memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
+            input->pos += syncPoint.toLoad;
+            mtctx->inBuff.filled += syncPoint.toLoad;
+            forwardInputProgress = syncPoint.toLoad>0;
+        }
+    }
+    if ((input->pos < input->size) && (endOp == ZSTD_e_end)) {
+        /* Can't end yet because the input is not fully consumed.
+            * We are in one of these cases:
+            * - mtctx->inBuff is NULL & empty: we couldn't get an input buffer so don't create a new job.
+            * - We filled the input buffer: flush this job but don't end the frame.
+            * - We hit a synchronization point: flush this job but don't end the frame.
+            */
+        assert(mtctx->inBuff.filled == 0 || mtctx->inBuff.filled == mtctx->targetSectionSize || mtctx->params.rsyncable);
+        endOp = ZSTD_e_flush;
+    }
+
+    if ( (mtctx->jobReady)
+      || (mtctx->inBuff.filled >= mtctx->targetSectionSize)  /* filled enough : let's compress */
+      || ((endOp != ZSTD_e_continue) && (mtctx->inBuff.filled > 0))  /* something to flush : let's go */
+      || ((endOp == ZSTD_e_end) && (!mtctx->frameEnded)) ) {   /* must finish the frame with a zero-size block */
+        size_t const jobSize = mtctx->inBuff.filled;
+        assert(mtctx->inBuff.filled <= mtctx->targetSectionSize);
+        FORWARD_IF_ERROR( ZSTDMT_createCompressionJob(mtctx, jobSize, endOp) , "");
+    }
+
+    /* check for potential compressed data ready to be flushed */
+    {   size_t const remainingToFlush = ZSTDMT_flushProduced(mtctx, output, !forwardInputProgress, endOp); /* block if there was no forward input progress */
+        if (input->pos < input->size) return MAX(remainingToFlush, 1);  /* input not consumed : do not end flush yet */
+        DEBUGLOG(5, "end of ZSTDMT_compressStream_generic: remainingToFlush = %u", (U32)remainingToFlush);
+        return remainingToFlush;
+    }
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstdmt_compress.h b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstdmt_compress.h
new file mode 100644
index 0000000..ed4dc0e
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/compress/zstdmt_compress.h
@@ -0,0 +1,113 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ #ifndef ZSTDMT_COMPRESS_H
+ #define ZSTDMT_COMPRESS_H
+
+ #if defined (__cplusplus)
+ extern "C" {
+ #endif
+
+
+/* Note : This is an internal API.
+ *        These APIs used to be exposed with ZSTDLIB_API,
+ *        because it used to be the only way to invoke MT compression.
+ *        Now, you must use ZSTD_compress2 and ZSTD_compressStream2() instead.
+ *
+ *        This API requires ZSTD_MULTITHREAD to be defined during compilation,
+ *        otherwise ZSTDMT_createCCtx*() will fail.
+ */
+
+/* ===   Dependencies   === */
+#include "../common/zstd_deps.h"   /* size_t */
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters */
+#include "../zstd.h"            /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
+
+
+/* ===   Constants   === */
+#ifndef ZSTDMT_NBWORKERS_MAX /* a different value can be selected at compile time */
+#  define ZSTDMT_NBWORKERS_MAX ((sizeof(void*)==4) /*32-bit*/ ? 64 : 256)
+#endif
+#ifndef ZSTDMT_JOBSIZE_MIN   /* a different value can be selected at compile time */
+#  define ZSTDMT_JOBSIZE_MIN (512 KB)
+#endif
+#define ZSTDMT_JOBLOG_MAX   (MEM_32bits() ? 29 : 30)
+#define ZSTDMT_JOBSIZE_MAX  (MEM_32bits() ? (512 MB) : (1024 MB))
+
+
+/* ========================================================
+ * ===  Private interface, for use by ZSTD_compress.c   ===
+ * ===  Not exposed in libzstd. Never invoke directly   ===
+ * ======================================================== */
+
+/* ===   Memory management   === */
+typedef struct ZSTDMT_CCtx_s ZSTDMT_CCtx;
+/* Requires ZSTD_MULTITHREAD to be defined during compilation, otherwise it will return NULL. */
+ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
+                                        ZSTD_customMem cMem,
+					ZSTD_threadPool *pool);
+size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
+
+size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
+
+/* ===   Streaming functions   === */
+
+size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx);
+
+/*! ZSTDMT_initCStream_internal() :
+ *  Private use only. Init streaming operation.
+ *  expects params to be valid.
+ *  must receive dict, or cdict, or none, but not both.
+ *  mtctx can be freshly constructed or reused from a prior compression.
+ *  If mtctx is reused, memory allocations from the prior compression may not be freed,
+ *  even if they are not needed for the current compression.
+ *  @return : 0, or an error code */
+size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* mtctx,
+                    const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
+                    const ZSTD_CDict* cdict,
+                    ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
+
+/*! ZSTDMT_compressStream_generic() :
+ *  Combines ZSTDMT_compressStream() with optional ZSTDMT_flushStream() or ZSTDMT_endStream()
+ *  depending on flush directive.
+ * @return : minimum amount of data still to be flushed
+ *           0 if fully flushed
+ *           or an error code
+ *  note : needs to be init using any ZSTD_initCStream*() variant */
+size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                     ZSTD_outBuffer* output,
+                                     ZSTD_inBuffer* input,
+                                     ZSTD_EndDirective endOp);
+
+ /*! ZSTDMT_toFlushNow()
+  *  Tell how many bytes are ready to be flushed immediately.
+  *  Probe the oldest active job (not yet entirely flushed) and check its output buffer.
+  *  If return 0, it means there is no active job,
+  *  or, it means oldest job is still active, but everything produced has been flushed so far,
+  *  therefore flushing is limited by speed of oldest job. */
+size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx);
+
+/*! ZSTDMT_updateCParams_whileCompressing() :
+ *  Updates only a selected set of compression parameters, to remain compatible with current frame.
+ *  New parameters will be applied to next compression job. */
+void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams);
+
+/*! ZSTDMT_getFrameProgression():
+ *  tells how much data has been consumed (input) and produced (output) for current frame.
+ *  able to count progression inside worker threads.
+ */
+ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* ZSTDMT_COMPRESS_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/huf_decompress.c b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/huf_decompress.c
new file mode 100644
index 0000000..5b217ac
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/huf_decompress.c
@@ -0,0 +1,1882 @@
+/* ******************************************************************
+ * huff0 huffman decoder,
+ * part of Finite State Entropy library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* **************************************************************
+*  Dependencies
+****************************************************************/
+#include "../common/zstd_deps.h"  /* ZSTD_memcpy, ZSTD_memset */
+#include "../common/compiler.h"
+#include "../common/bitstream.h"  /* BIT_* */
+#include "../common/fse.h"        /* to compress headers */
+#include "../common/huf.h"
+#include "../common/error_private.h"
+#include "../common/zstd_internal.h"
+#include "../common/bits.h"       /* ZSTD_highbit32, ZSTD_countTrailingZeros64 */
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+
+#define HUF_DECODER_FAST_TABLELOG 11
+
+/* **************************************************************
+*  Macros
+****************************************************************/
+
+/* These two optional macros force the use one way or another of the two
+ * Huffman decompression implementations. You can't force in both directions
+ * at the same time.
+ */
+#if defined(HUF_FORCE_DECOMPRESS_X1) && \
+    defined(HUF_FORCE_DECOMPRESS_X2)
+#error "Cannot force the use of the X1 and X2 decoders at the same time!"
+#endif
+
+/* When DYNAMIC_BMI2 is enabled, fast decoders are only called when bmi2 is
+ * supported at runtime, so we can add the BMI2 target attribute.
+ * When it is disabled, we will still get BMI2 if it is enabled statically.
+ */
+#if DYNAMIC_BMI2
+# define HUF_FAST_BMI2_ATTRS BMI2_TARGET_ATTRIBUTE
+#else
+# define HUF_FAST_BMI2_ATTRS
+#endif
+
+#ifdef __cplusplus
+# define HUF_EXTERN_C extern "C"
+#else
+# define HUF_EXTERN_C
+#endif
+#define HUF_ASM_DECL HUF_EXTERN_C
+
+#if DYNAMIC_BMI2
+# define HUF_NEED_BMI2_FUNCTION 1
+#else
+# define HUF_NEED_BMI2_FUNCTION 0
+#endif
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_isError ERR_isError
+
+
+/* **************************************************************
+*  Byte alignment for workSpace management
+****************************************************************/
+#define HUF_ALIGN(x, a)         HUF_ALIGN_MASK((x), (a) - 1)
+#define HUF_ALIGN_MASK(x, mask) (((x) + (mask)) & ~(mask))
+
+
+/* **************************************************************
+*  BMI2 Variant Wrappers
+****************************************************************/
+typedef size_t (*HUF_DecompressUsingDTableFn)(void *dst, size_t dstSize,
+                                              const void *cSrc,
+                                              size_t cSrcSize,
+                                              const HUF_DTable *DTable);
+
+#if DYNAMIC_BMI2
+
+#define HUF_DGEN(fn)                                                        \
+                                                                            \
+    static size_t fn##_default(                                             \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static BMI2_TARGET_ATTRIBUTE size_t fn##_bmi2(                          \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
+    {                                                                       \
+        if (flags & HUF_flags_bmi2) {                                       \
+            return fn##_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);         \
+        }                                                                   \
+        return fn##_default(dst, dstSize, cSrc, cSrcSize, DTable);          \
+    }
+
+#else
+
+#define HUF_DGEN(fn)                                                        \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
+    {                                                                       \
+        (void)flags;                                                        \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }
+
+#endif
+
+
+/*-***************************/
+/*  generic DTableDesc       */
+/*-***************************/
+typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
+
+static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
+{
+    DTableDesc dtd;
+    ZSTD_memcpy(&dtd, table, sizeof(dtd));
+    return dtd;
+}
+
+static size_t HUF_initFastDStream(BYTE const* ip) {
+    BYTE const lastByte = ip[7];
+    size_t const bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
+    size_t const value = MEM_readLEST(ip) | 1;
+    assert(bitsConsumed <= 8);
+    assert(sizeof(size_t) == 8);
+    return value << bitsConsumed;
+}
+
+
+/**
+ * The input/output arguments to the Huffman fast decoding loop:
+ *
+ * ip [in/out] - The input pointers, must be updated to reflect what is consumed.
+ * op [in/out] - The output pointers, must be updated to reflect what is written.
+ * bits [in/out] - The bitstream containers, must be updated to reflect the current state.
+ * dt [in] - The decoding table.
+ * ilimit [in] - The input limit, stop when any input pointer is below ilimit.
+ * oend [in] - The end of the output stream. op[3] must not cross oend.
+ * iend [in] - The end of each input stream. ip[i] may cross iend[i],
+ *             as long as it is above ilimit, but that indicates corruption.
+ */
+typedef struct {
+    BYTE const* ip[4];
+    BYTE* op[4];
+    U64 bits[4];
+    void const* dt;
+    BYTE const* ilimit;
+    BYTE* oend;
+    BYTE const* iend[4];
+} HUF_DecompressFastArgs;
+
+typedef void (*HUF_DecompressFastLoopFn)(HUF_DecompressFastArgs*);
+
+/**
+ * Initializes args for the fast decoding loop.
+ * @returns 1 on success
+ *          0 if the fallback implementation should be used.
+ *          Or an error code on failure.
+ */
+static size_t HUF_DecompressFastArgs_init(HUF_DecompressFastArgs* args, void* dst, size_t dstSize, void const* src, size_t srcSize, const HUF_DTable* DTable)
+{
+    void const* dt = DTable + 1;
+    U32 const dtLog = HUF_getDTableDesc(DTable).tableLog;
+
+    const BYTE* const ilimit = (const BYTE*)src + 6 + 8;
+
+    BYTE* const oend = (BYTE*)dst + dstSize;
+
+    /* The fast decoding loop assumes 64-bit little-endian.
+     * This condition is false on x32.
+     */
+    if (!MEM_isLittleEndian() || MEM_32bits())
+        return 0;
+
+    /* strict minimum : jump table + 1 byte per stream */
+    if (srcSize < 10)
+        return ERROR(corruption_detected);
+
+    /* Must have at least 8 bytes per stream because we don't handle initializing smaller bit containers.
+     * If table log is not correct at this point, fallback to the old decoder.
+     * On small inputs we don't have enough data to trigger the fast loop, so use the old decoder.
+     */
+    if (dtLog != HUF_DECODER_FAST_TABLELOG)
+        return 0;
+
+    /* Read the jump table. */
+    {
+        const BYTE* const istart = (const BYTE*)src;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = srcSize - (length1 + length2 + length3 + 6);
+        args->iend[0] = istart + 6;  /* jumpTable */
+        args->iend[1] = args->iend[0] + length1;
+        args->iend[2] = args->iend[1] + length2;
+        args->iend[3] = args->iend[2] + length3;
+
+        /* HUF_initFastDStream() requires this, and this small of an input
+         * won't benefit from the ASM loop anyways.
+         * length1 must be >= 16 so that ip[0] >= ilimit before the loop
+         * starts.
+         */
+        if (length1 < 16 || length2 < 8 || length3 < 8 || length4 < 8)
+            return 0;
+        if (length4 > srcSize) return ERROR(corruption_detected);   /* overflow */
+    }
+    /* ip[] contains the position that is currently loaded into bits[]. */
+    args->ip[0] = args->iend[1] - sizeof(U64);
+    args->ip[1] = args->iend[2] - sizeof(U64);
+    args->ip[2] = args->iend[3] - sizeof(U64);
+    args->ip[3] = (BYTE const*)src + srcSize - sizeof(U64);
+
+    /* op[] contains the output pointers. */
+    args->op[0] = (BYTE*)dst;
+    args->op[1] = args->op[0] + (dstSize+3)/4;
+    args->op[2] = args->op[1] + (dstSize+3)/4;
+    args->op[3] = args->op[2] + (dstSize+3)/4;
+
+    /* No point to call the ASM loop for tiny outputs. */
+    if (args->op[3] >= oend)
+        return 0;
+
+    /* bits[] is the bit container.
+        * It is read from the MSB down to the LSB.
+        * It is shifted left as it is read, and zeros are
+        * shifted in. After the lowest valid bit a 1 is
+        * set, so that CountTrailingZeros(bits[]) can be used
+        * to count how many bits we've consumed.
+        */
+    args->bits[0] = HUF_initFastDStream(args->ip[0]);
+    args->bits[1] = HUF_initFastDStream(args->ip[1]);
+    args->bits[2] = HUF_initFastDStream(args->ip[2]);
+    args->bits[3] = HUF_initFastDStream(args->ip[3]);
+
+    /* If ip[] >= ilimit, it is guaranteed to be safe to
+        * reload bits[]. It may be beyond its section, but is
+        * guaranteed to be valid (>= istart).
+        */
+    args->ilimit = ilimit;
+
+    args->oend = oend;
+    args->dt = dt;
+
+    return 1;
+}
+
+static size_t HUF_initRemainingDStream(BIT_DStream_t* bit, HUF_DecompressFastArgs const* args, int stream, BYTE* segmentEnd)
+{
+    /* Validate that we haven't overwritten. */
+    if (args->op[stream] > segmentEnd)
+        return ERROR(corruption_detected);
+    /* Validate that we haven't read beyond iend[].
+        * Note that ip[] may be < iend[] because the MSB is
+        * the next bit to read, and we may have consumed 100%
+        * of the stream, so down to iend[i] - 8 is valid.
+        */
+    if (args->ip[stream] < args->iend[stream] - 8)
+        return ERROR(corruption_detected);
+
+    /* Construct the BIT_DStream_t. */
+    assert(sizeof(size_t) == 8);
+    bit->bitContainer = MEM_readLEST(args->ip[stream]);
+    bit->bitsConsumed = ZSTD_countTrailingZeros64(args->bits[stream]);
+    bit->start = (const char*)args->iend[0];
+    bit->limitPtr = bit->start + sizeof(size_t);
+    bit->ptr = (const char*)args->ip[stream];
+
+    return 0;
+}
+
+
+#ifndef HUF_FORCE_DECOMPRESS_X2
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+typedef struct { BYTE nbBits; BYTE byte; } HUF_DEltX1;   /* single-symbol decoding */
+
+/**
+ * Packs 4 HUF_DEltX1 structs into a U64. This is used to lay down 4 entries at
+ * a time.
+ */
+static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
+    U64 D4;
+    if (MEM_isLittleEndian()) {
+        D4 = (U64)((symbol << 8) + nbBits);
+    } else {
+        D4 = (U64)(symbol + (nbBits << 8));
+    }
+    assert(D4 < (1U << 16));
+    D4 *= 0x0001000100010001ULL;
+    return D4;
+}
+
+/**
+ * Increase the tableLog to targetTableLog and rescales the stats.
+ * If tableLog > targetTableLog this is a no-op.
+ * @returns New tableLog
+ */
+static U32 HUF_rescaleStats(BYTE* huffWeight, U32* rankVal, U32 nbSymbols, U32 tableLog, U32 targetTableLog)
+{
+    if (tableLog > targetTableLog)
+        return tableLog;
+    if (tableLog < targetTableLog) {
+        U32 const scale = targetTableLog - tableLog;
+        U32 s;
+        /* Increase the weight for all non-zero probability symbols by scale. */
+        for (s = 0; s < nbSymbols; ++s) {
+            huffWeight[s] += (BYTE)((huffWeight[s] == 0) ? 0 : scale);
+        }
+        /* Update rankVal to reflect the new weights.
+         * All weights except 0 get moved to weight + scale.
+         * Weights [1, scale] are empty.
+         */
+        for (s = targetTableLog; s > scale; --s) {
+            rankVal[s] = rankVal[s - scale];
+        }
+        for (s = scale; s > 0; --s) {
+            rankVal[s] = 0;
+        }
+    }
+    return targetTableLog;
+}
+
+typedef struct {
+        U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 rankStart[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 statsWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+        BYTE symbols[HUF_SYMBOLVALUE_MAX + 1];
+        BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
+} HUF_ReadDTableX1_Workspace;
+
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags)
+{
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    size_t iSize;
+    void* const dtPtr = DTable + 1;
+    HUF_DEltX1* const dt = (HUF_DEltX1*)dtPtr;
+    HUF_ReadDTableX1_Workspace* wksp = (HUF_ReadDTableX1_Workspace*)workSpace;
+
+    DEBUG_STATIC_ASSERT(HUF_DECOMPRESS_WORKSPACE_SIZE >= sizeof(*wksp));
+    if (sizeof(*wksp) > wkspSize) return ERROR(tableLog_tooLarge);
+
+    DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
+    /* ZSTD_memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), flags);
+    if (HUF_isError(iSize)) return iSize;
+
+
+    /* Table header */
+    {   DTableDesc dtd = HUF_getDTableDesc(DTable);
+        U32 const maxTableLog = dtd.maxTableLog + 1;
+        U32 const targetTableLog = MIN(maxTableLog, HUF_DECODER_FAST_TABLELOG);
+        tableLog = HUF_rescaleStats(wksp->huffWeight, wksp->rankVal, nbSymbols, tableLog, targetTableLog);
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, Huffman tree cannot fit in */
+        dtd.tableType = 0;
+        dtd.tableLog = (BYTE)tableLog;
+        ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
+    }
+
+    /* Compute symbols and rankStart given rankVal:
+     *
+     * rankVal already contains the number of values of each weight.
+     *
+     * symbols contains the symbols ordered by weight. First are the rankVal[0]
+     * weight 0 symbols, followed by the rankVal[1] weight 1 symbols, and so on.
+     * symbols[0] is filled (but unused) to avoid a branch.
+     *
+     * rankStart contains the offset where each rank belongs in the DTable.
+     * rankStart[0] is not filled because there are no entries in the table for
+     * weight 0.
+     */
+    {   int n;
+        U32 nextRankStart = 0;
+        int const unroll = 4;
+        int const nLimit = (int)nbSymbols - unroll + 1;
+        for (n=0; n<(int)tableLog+1; n++) {
+            U32 const curr = nextRankStart;
+            nextRankStart += wksp->rankVal[n];
+            wksp->rankStart[n] = curr;
+        }
+        for (n=0; n < nLimit; n += unroll) {
+            int u;
+            for (u=0; u < unroll; ++u) {
+                size_t const w = wksp->huffWeight[n+u];
+                wksp->symbols[wksp->rankStart[w]++] = (BYTE)(n+u);
+            }
+        }
+        for (; n < (int)nbSymbols; ++n) {
+            size_t const w = wksp->huffWeight[n];
+            wksp->symbols[wksp->rankStart[w]++] = (BYTE)n;
+        }
+    }
+
+    /* fill DTable
+     * We fill all entries of each weight in order.
+     * That way length is a constant for each iteration of the outer loop.
+     * We can switch based on the length to a different inner loop which is
+     * optimized for that particular case.
+     */
+    {   U32 w;
+        int symbol = wksp->rankVal[0];
+        int rankStart = 0;
+        for (w=1; w<tableLog+1; ++w) {
+            int const symbolCount = wksp->rankVal[w];
+            int const length = (1 << w) >> 1;
+            int uStart = rankStart;
+            BYTE const nbBits = (BYTE)(tableLog + 1 - w);
+            int s;
+            int u;
+            switch (length) {
+            case 1:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart] = D;
+                    uStart += 1;
+                }
+                break;
+            case 2:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart+0] = D;
+                    dt[uStart+1] = D;
+                    uStart += 2;
+                }
+                break;
+            case 4:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    uStart += 4;
+                }
+                break;
+            case 8:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    MEM_write64(dt + uStart + 4, D4);
+                    uStart += 8;
+                }
+                break;
+            default:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    for (u=0; u < length; u += 16) {
+                        MEM_write64(dt + uStart + u + 0, D4);
+                        MEM_write64(dt + uStart + u + 4, D4);
+                        MEM_write64(dt + uStart + u + 8, D4);
+                        MEM_write64(dt + uStart + u + 12, D4);
+                    }
+                    assert(u == length);
+                    uStart += length;
+                }
+                break;
+            }
+            symbol += symbolCount;
+            rankStart += symbolCount * length;
+        }
+    }
+    return iSize;
+}
+
+FORCE_INLINE_TEMPLATE BYTE
+HUF_decodeSymbolX1(BIT_DStream_t* Dstream, const HUF_DEltX1* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    BYTE const c = dt[val].byte;
+    BIT_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX1(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX1_1(ptr, DStreamPtr)  \
+    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX1_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
+
+HINT_INLINE size_t
+HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX1* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    if ((pEnd - p) > 3) {
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
+            HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+        }
+    } else {
+        BIT_reloadDStream(bitDPtr);
+    }
+
+    /* [0-3] symbols remaining */
+    if (MEM_32bits())
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd))
+            HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+
+    return (size_t)(pEnd-pStart);
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X1_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const void* dtPtr = DTable + 1;
+    const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
+    BIT_DStream_t bitD;
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    U32 const dtLog = dtd.tableLog;
+
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    HUF_decodeStreamX1(op, &bitD, oend, dt, dtLog);
+
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+/* HUF_decompress4X1_usingDTable_internal_body():
+ * Conditions :
+ * @dstSize >= 6
+ */
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X1_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        BYTE* const olimit = oend - 3;
+        const void* const dtPtr = DTable + 1;
+        const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+        U32 endSignal = 1;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        if (opStart4 > oend) return ERROR(corruption_detected);      /* overflow */
+        if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
+        if ((size_t)(oend - op4) >= sizeof(size_t)) {
+            for ( ; (endSignal) & (op4 < olimit) ; ) {
+                HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
+                endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+            }
+        }
+
+        /* check corruption */
+        /* note : should not be necessary : op# advance in lock step, and we control op4.
+         *        but curiously, binary generated by gcc 7.2 & 7.3 with -mbmi2 runs faster when >=1 test is present */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX1(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX1(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX1(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX1(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+#if HUF_NEED_BMI2_FUNCTION
+static BMI2_TARGET_ATTRIBUTE
+size_t HUF_decompress4X1_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+#endif
+
+static
+size_t HUF_decompress4X1_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
+
+HUF_ASM_DECL void HUF_decompress4X1_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
+
+#endif
+
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X1_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
+{
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    U16 const* const dtable = (U16 const*)args->dt;
+    BYTE* const oend = args->oend;
+    BYTE const* const ilimit = args->ilimit;
+
+    /* Copy the arguments to local variables */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&ip), &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+        int symbol;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= (stream == 3 ? oend : op[stream + 1]));
+            assert(ip[stream] >= ilimit);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each iteration produces 5 output symbols per stream */
+            size_t const oiters = (size_t)(oend - op[3]) / 5;
+            /* Each iteration consumes up to 11 bits * 5 = 55 bits < 7 bytes
+             * per stream.
+             */
+            size_t const iiters = (size_t)(ip[0] - ilimit) / 7;
+            /* We can safely run iters iterations before running bounds checks */
+            size_t const iters = MIN(oiters, iiters);
+            size_t const symbols = iters * 5;
+
+            /* We can simply check that op[3] < olimit, instead of checking all
+             * of our bounds, since we can't hit the other bounds until we've run
+             * iters iterations, which only happens when op[3] == olimit.
+             */
+            olimit = op[3] + symbols;
+
+            /* Exit fast decoding loop once we get close to the end. */
+            if (op[3] + 20 > olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
+
+        do {
+            /* Decode 5 symbols in each of the 4 streams */
+            for (symbol = 0; symbol < 5; ++symbol) {
+                for (stream = 0; stream < 4; ++stream) {
+                    int const index = (int)(bits[stream] >> 53);
+                    int const entry = (int)dtable[index];
+                    bits[stream] <<= (entry & 63);
+                    op[stream][symbol] = (BYTE)((entry >> 8) & 0xFF);
+                }
+            }
+            /* Reload the bitstreams */
+            for (stream = 0; stream < 4; ++stream) {
+                int const ctz = ZSTD_countTrailingZeros64(bits[stream]);
+                int const nbBits = ctz & 7;
+                int const nbBytes = ctz >> 3;
+                op[stream] += 5;
+                ip[stream] -= nbBytes;
+                bits[stream] = MEM_read64(ip[stream]) | 1;
+                bits[stream] <<= nbBits;
+            }
+        } while (op[3] < olimit);
+    }
+
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&args->ip), &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
+}
+
+/**
+ * @returns @p dstSize on success (>= 6)
+ *          0 if the fallback implementation should be used
+ *          An error if an error occurred
+ */
+static HUF_FAST_BMI2_ATTRS
+size_t
+HUF_decompress4X1_usingDTable_internal_fast(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn)
+{
+    void const* dt = DTable + 1;
+    const BYTE* const iend = (const BYTE*)cSrc + 6;
+    BYTE* const oend = (BYTE*)dst + dstSize;
+    HUF_DecompressFastArgs args;
+    {   size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        FORWARD_IF_ERROR(ret, "Failed to init fast loop args");
+        if (ret == 0)
+            return 0;
+    }
+
+    assert(args.ip[0] >= args.ilimit);
+    loopFn(&args);
+
+    /* Our loop guarantees that ip[] >= ilimit and that we haven't
+    * overwritten any op[].
+    */
+    assert(args.ip[0] >= iend);
+    assert(args.ip[1] >= iend);
+    assert(args.ip[2] >= iend);
+    assert(args.ip[3] >= iend);
+    assert(args.op[3] <= oend);
+    (void)iend;
+
+    /* finish bit streams one by one. */
+    {   size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* segmentEnd = (BYTE*)dst;
+        int i;
+        for (i = 0; i < 4; ++i) {
+            BIT_DStream_t bit;
+            if (segmentSize <= (size_t)(oend - segmentEnd))
+                segmentEnd += segmentSize;
+            else
+                segmentEnd = oend;
+            FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
+            /* Decompress and validate that we've produced exactly the expected length. */
+            args.op[i] += HUF_decodeStreamX1(args.op[i], &bit, segmentEnd, (HUF_DEltX1 const*)dt, HUF_DECODER_FAST_TABLELOG);
+            if (args.op[i] != segmentEnd) return ERROR(corruption_detected);
+        }
+    }
+
+    /* decoded size */
+    assert(dstSize != 0);
+    return dstSize;
+}
+
+HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
+
+static size_t HUF_decompress4X1_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
+{
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X1_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X1_usingDTable_internal_fast_c_loop;
+
+#if DYNAMIC_BMI2
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X1_usingDTable_internal_bmi2;
+# if ZSTD_ENABLE_ASM_X86_64_BMI2
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+        }
+# endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+    }
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+    }
+#endif
+
+    if (!(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X1_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+static size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
+}
+
+#endif /* HUF_FORCE_DECOMPRESS_X2 */
+
+
+#ifndef HUF_FORCE_DECOMPRESS_X1
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX2;  /* double-symbols decoding */
+typedef struct { BYTE symbol; } sortedSymbol_t;
+typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
+typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
+
+/**
+ * Constructs a HUF_DEltX2 in a U32.
+ */
+static U32 HUF_buildDEltX2U32(U32 symbol, U32 nbBits, U32 baseSeq, int level)
+{
+    U32 seq;
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, sequence) == 0);
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, nbBits) == 2);
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, length) == 3);
+    DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U32));
+    if (MEM_isLittleEndian()) {
+        seq = level == 1 ? symbol : (baseSeq + (symbol << 8));
+        return seq + (nbBits << 16) + ((U32)level << 24);
+    } else {
+        seq = level == 1 ? (symbol << 8) : ((baseSeq << 8) + symbol);
+        return (seq << 16) + (nbBits << 8) + (U32)level;
+    }
+}
+
+/**
+ * Constructs a HUF_DEltX2.
+ */
+static HUF_DEltX2 HUF_buildDEltX2(U32 symbol, U32 nbBits, U32 baseSeq, int level)
+{
+    HUF_DEltX2 DElt;
+    U32 const val = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
+    DEBUG_STATIC_ASSERT(sizeof(DElt) == sizeof(val));
+    ZSTD_memcpy(&DElt, &val, sizeof(val));
+    return DElt;
+}
+
+/**
+ * Constructs 2 HUF_DEltX2s and packs them into a U64.
+ */
+static U64 HUF_buildDEltX2U64(U32 symbol, U32 nbBits, U16 baseSeq, int level)
+{
+    U32 DElt = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
+    return (U64)DElt + ((U64)DElt << 32);
+}
+
+/**
+ * Fills the DTable rank with all the symbols from [begin, end) that are each
+ * nbBits long.
+ *
+ * @param DTableRank The start of the rank in the DTable.
+ * @param begin The first symbol to fill (inclusive).
+ * @param end The last symbol to fill (exclusive).
+ * @param nbBits Each symbol is nbBits long.
+ * @param tableLog The table log.
+ * @param baseSeq If level == 1 { 0 } else { the first level symbol }
+ * @param level The level in the table. Must be 1 or 2.
+ */
+static void HUF_fillDTableX2ForWeight(
+    HUF_DEltX2* DTableRank,
+    sortedSymbol_t const* begin, sortedSymbol_t const* end,
+    U32 nbBits, U32 tableLog,
+    U16 baseSeq, int const level)
+{
+    U32 const length = 1U << ((tableLog - nbBits) & 0x1F /* quiet static-analyzer */);
+    const sortedSymbol_t* ptr;
+    assert(level >= 1 && level <= 2);
+    switch (length) {
+    case 1:
+        for (ptr = begin; ptr != end; ++ptr) {
+            HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
+            *DTableRank++ = DElt;
+        }
+        break;
+    case 2:
+        for (ptr = begin; ptr != end; ++ptr) {
+            HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
+            DTableRank[0] = DElt;
+            DTableRank[1] = DElt;
+            DTableRank += 2;
+        }
+        break;
+    case 4:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+            DTableRank += 4;
+        }
+        break;
+    case 8:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
+            DTableRank += 8;
+        }
+        break;
+    default:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            HUF_DEltX2* const DTableRankEnd = DTableRank + length;
+            for (; DTableRank != DTableRankEnd; DTableRank += 8) {
+                ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
+            }
+        }
+        break;
+    }
+}
+
+/* HUF_fillDTableX2Level2() :
+ * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
+static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 targetLog, const U32 consumedBits,
+                           const U32* rankVal, const int minWeight, const int maxWeight1,
+                           const sortedSymbol_t* sortedSymbols, U32 const* rankStart,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    /* Fill skipped values (all positions up to rankVal[minWeight]).
+     * These are positions only get a single symbol because the combined weight
+     * is too large.
+     */
+    if (minWeight>1) {
+        U32 const length = 1U << ((targetLog - consumedBits) & 0x1F /* quiet static-analyzer */);
+        U64 const DEltX2 = HUF_buildDEltX2U64(baseSeq, consumedBits, /* baseSeq */ 0, /* level */ 1);
+        int const skipSize = rankVal[minWeight];
+        assert(length > 1);
+        assert((U32)skipSize < length);
+        switch (length) {
+        case 2:
+            assert(skipSize == 1);
+            ZSTD_memcpy(DTable, &DEltX2, sizeof(DEltX2));
+            break;
+        case 4:
+            assert(skipSize <= 4);
+            ZSTD_memcpy(DTable + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTable + 2, &DEltX2, sizeof(DEltX2));
+            break;
+        default:
+            {
+                int i;
+                for (i = 0; i < skipSize; i += 8) {
+                    ZSTD_memcpy(DTable + i + 0, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 2, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 4, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 6, &DEltX2, sizeof(DEltX2));
+                }
+            }
+        }
+    }
+
+    /* Fill each of the second level symbols by weight. */
+    {
+        int w;
+        for (w = minWeight; w < maxWeight1; ++w) {
+            int const begin = rankStart[w];
+            int const end = rankStart[w+1];
+            U32 const nbBits = nbBitsBaseline - w;
+            U32 const totalBits = nbBits + consumedBits;
+            HUF_fillDTableX2ForWeight(
+                DTable + rankVal[w],
+                sortedSymbols + begin, sortedSymbols + end,
+                totalBits, targetLog,
+                baseSeq, /* level */ 2);
+        }
+    }
+}
+
+static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList,
+                           const U32* rankStart, rankValCol_t* rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32* const rankVal = rankValOrigin[0];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    int w;
+    int const wEnd = (int)maxWeight + 1;
+
+    /* Fill DTable in order of weight. */
+    for (w = 1; w < wEnd; ++w) {
+        int const begin = (int)rankStart[w];
+        int const end = (int)rankStart[w+1];
+        U32 const nbBits = nbBitsBaseline - w;
+
+        if (targetLog-nbBits >= minBits) {
+            /* Enough room for a second symbol. */
+            int start = rankVal[w];
+            U32 const length = 1U << ((targetLog - nbBits) & 0x1F /* quiet static-analyzer */);
+            int minWeight = nbBits + scaleLog;
+            int s;
+            if (minWeight < 1) minWeight = 1;
+            /* Fill the DTable for every symbol of weight w.
+             * These symbols get at least 1 second symbol.
+             */
+            for (s = begin; s != end; ++s) {
+                HUF_fillDTableX2Level2(
+                    DTable + start, targetLog, nbBits,
+                    rankValOrigin[nbBits], minWeight, wEnd,
+                    sortedList, rankStart,
+                    nbBitsBaseline, sortedList[s].symbol);
+                start += length;
+            }
+        } else {
+            /* Only a single symbol. */
+            HUF_fillDTableX2ForWeight(
+                DTable + rankVal[w],
+                sortedList + begin, sortedList + end,
+                nbBits, targetLog,
+                /* baseSeq */ 0, /* level */ 1);
+        }
+    }
+}
+
+typedef struct {
+    rankValCol_t rankVal[HUF_TABLELOG_MAX];
+    U32 rankStats[HUF_TABLELOG_MAX + 1];
+    U32 rankStart0[HUF_TABLELOG_MAX + 3];
+    sortedSymbol_t sortedSymbol[HUF_SYMBOLVALUE_MAX + 1];
+    BYTE weightList[HUF_SYMBOLVALUE_MAX + 1];
+    U32 calleeWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+} HUF_ReadDTableX2_Workspace;
+
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
+                       const void* src, size_t srcSize,
+                             void* workSpace, size_t wkspSize, int flags)
+{
+    U32 tableLog, maxW, nbSymbols;
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    U32 maxTableLog = dtd.maxTableLog;
+    size_t iSize;
+    void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+    U32 *rankStart;
+
+    HUF_ReadDTableX2_Workspace* const wksp = (HUF_ReadDTableX2_Workspace*)workSpace;
+
+    if (sizeof(*wksp) > wkspSize) return ERROR(GENERIC);
+
+    rankStart = wksp->rankStart0 + 1;
+    ZSTD_memset(wksp->rankStats, 0, sizeof(wksp->rankStats));
+    ZSTD_memset(wksp->rankStart0, 0, sizeof(wksp->rankStart0));
+
+    DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
+    if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    /* ZSTD_memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), flags);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+    if (tableLog <= HUF_DECODER_FAST_TABLELOG && maxTableLog > HUF_DECODER_FAST_TABLELOG) maxTableLog = HUF_DECODER_FAST_TABLELOG;
+
+    /* find maxWeight */
+    for (maxW = tableLog; wksp->rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 curr = nextRankStart;
+            nextRankStart += wksp->rankStats[w];
+            rankStart[w] = curr;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        rankStart[maxW+1] = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = wksp->weightList[s];
+            U32 const r = rankStart[w]++;
+            wksp->sortedSymbol[r].symbol = (BYTE)s;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = wksp->rankVal[0];
+        {   int const rescale = (maxTableLog-tableLog) - 1;   /* tableLog <= maxTableLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 curr = nextRankVal;
+                nextRankVal += wksp->rankStats[w] << (w+rescale);
+                rankVal0[w] = curr;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < maxTableLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = wksp->rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUF_fillDTableX2(dt, maxTableLog,
+                   wksp->sortedSymbol,
+                   wksp->rankStart0, wksp->rankVal, maxW,
+                   tableLog+1);
+
+    dtd.tableLog = (BYTE)maxTableLog;
+    dtd.tableType = 1;
+    ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
+    return iSize;
+}
+
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    ZSTD_memcpy(op, &dt[val].sequence, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    ZSTD_memcpy(op, &dt[val].sequence, 1);
+    if (dt[val].length==1) {
+        BIT_skipBits(DStream, dt[val].nbBits);
+    } else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
+        }
+    }
+    return 1;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+
+HINT_INLINE size_t
+HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
+                const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    if ((size_t)(pEnd - p) >= sizeof(bitDPtr->bitContainer)) {
+        if (dtLog <= 11 && MEM_64bits()) {
+            /* up to 10 symbols at a time */
+            while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-9)) {
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+            }
+        } else {
+            /* up to 8 symbols at a time */
+            while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
+                HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+            }
+        }
+    } else {
+        BIT_reloadDStream(bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    if ((size_t)(pEnd - p) >= 2) {
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+        while (p <= pEnd-2)
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+    }
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX2(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BIT_DStream_t bitD;
+
+    /* Init */
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        HUF_decodeStreamX2(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+/* HUF_decompress4X2_usingDTable_internal_body():
+ * Conditions:
+ * @dstSize >= 6
+ */
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        BYTE* const olimit = oend - (sizeof(size_t)-1);
+        const void* const dtPtr = DTable+1;
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal = 1;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);  /* overflow */
+        if (opStart4 > oend) return ERROR(corruption_detected);     /* overflow */
+        if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        if ((size_t)(oend - op4) >= sizeof(size_t)) {
+            for ( ; (endSignal) & (op4 < olimit); ) {
+#if defined(__clang__) && (defined(__x86_64__) || defined(__i386__))
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+                endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+                endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+#else
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+                endSignal = (U32)LIKELY((U32)
+                            (BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished));
+#endif
+            }
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+#if HUF_NEED_BMI2_FUNCTION
+static BMI2_TARGET_ATTRIBUTE
+size_t HUF_decompress4X2_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+#endif
+
+static
+size_t HUF_decompress4X2_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
+
+HUF_ASM_DECL void HUF_decompress4X2_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
+
+#endif
+
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X2_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
+{
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    BYTE* oend[4];
+    HUF_DEltX2 const* const dtable = (HUF_DEltX2 const*)args->dt;
+    BYTE const* const ilimit = args->ilimit;
+
+    /* Copy the arguments to local registers. */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&ip), &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    oend[0] = op[1];
+    oend[1] = op[2];
+    oend[2] = op[3];
+    oend[3] = args->oend;
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+        int symbol;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= oend[stream]);
+            assert(ip[stream] >= ilimit);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each loop does 5 table lookups for each of the 4 streams.
+             * Each table lookup consumes up to 11 bits of input, and produces
+             * up to 2 bytes of output.
+             */
+            /* We can consume up to 7 bytes of input per iteration per stream.
+             * We also know that each input pointer is >= ip[0]. So we can run
+             * iters loops before running out of input.
+             */
+            size_t iters = (size_t)(ip[0] - ilimit) / 7;
+            /* Each iteration can produce up to 10 bytes of output per stream.
+             * Each output stream my advance at different rates. So take the
+             * minimum number of safe iterations among all the output streams.
+             */
+            for (stream = 0; stream < 4; ++stream) {
+                size_t const oiters = (size_t)(oend[stream] - op[stream]) / 10;
+                iters = MIN(iters, oiters);
+            }
+
+            /* Each iteration produces at least 5 output symbols. So until
+             * op[3] crosses olimit, we know we haven't executed iters
+             * iterations yet. This saves us maintaining an iters counter,
+             * at the expense of computing the remaining # of iterations
+             * more frequently.
+             */
+            olimit = op[3] + (iters * 5);
+
+            /* Exit the fast decoding loop if we are too close to the end. */
+            if (op[3] + 10 > olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
+
+        do {
+            /* Do 5 table lookups for each of the first 3 streams */
+            for (symbol = 0; symbol < 5; ++symbol) {
+                for (stream = 0; stream < 3; ++stream) {
+                    int const index = (int)(bits[stream] >> 53);
+                    HUF_DEltX2 const entry = dtable[index];
+                    MEM_write16(op[stream], entry.sequence);
+                    bits[stream] <<= (entry.nbBits);
+                    op[stream] += (entry.length);
+                }
+            }
+            /* Do 1 table lookup from the final stream */
+            {
+                int const index = (int)(bits[3] >> 53);
+                HUF_DEltX2 const entry = dtable[index];
+                MEM_write16(op[3], entry.sequence);
+                bits[3] <<= (entry.nbBits);
+                op[3] += (entry.length);
+            }
+            /* Do 4 table lookups from the final stream & reload bitstreams */
+            for (stream = 0; stream < 4; ++stream) {
+                /* Do a table lookup from the final stream.
+                 * This is interleaved with the reloading to reduce register
+                 * pressure. This shouldn't be necessary, but compilers can
+                 * struggle with codegen with high register pressure.
+                 */
+                {
+                    int const index = (int)(bits[3] >> 53);
+                    HUF_DEltX2 const entry = dtable[index];
+                    MEM_write16(op[3], entry.sequence);
+                    bits[3] <<= (entry.nbBits);
+                    op[3] += (entry.length);
+                }
+                /* Reload the bistreams. The final bitstream must be reloaded
+                 * after the 5th symbol was decoded.
+                 */
+                {
+                    int const ctz = ZSTD_countTrailingZeros64(bits[stream]);
+                    int const nbBits = ctz & 7;
+                    int const nbBytes = ctz >> 3;
+                    ip[stream] -= nbBytes;
+                    bits[stream] = MEM_read64(ip[stream]) | 1;
+                    bits[stream] <<= nbBits;
+                }
+            }
+        } while (op[3] < olimit);
+    }
+
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&args->ip), &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
+}
+
+
+static HUF_FAST_BMI2_ATTRS size_t
+HUF_decompress4X2_usingDTable_internal_fast(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn) {
+    void const* dt = DTable + 1;
+    const BYTE* const iend = (const BYTE*)cSrc + 6;
+    BYTE* const oend = (BYTE*)dst + dstSize;
+    HUF_DecompressFastArgs args;
+    {
+        size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        FORWARD_IF_ERROR(ret, "Failed to init asm args");
+        if (ret == 0)
+            return 0;
+    }
+
+    assert(args.ip[0] >= args.ilimit);
+    loopFn(&args);
+
+    /* note : op4 already verified within main loop */
+    assert(args.ip[0] >= iend);
+    assert(args.ip[1] >= iend);
+    assert(args.ip[2] >= iend);
+    assert(args.ip[3] >= iend);
+    assert(args.op[3] <= oend);
+    (void)iend;
+
+    /* finish bitStreams one by one */
+    {
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* segmentEnd = (BYTE*)dst;
+        int i;
+        for (i = 0; i < 4; ++i) {
+            BIT_DStream_t bit;
+            if (segmentSize <= (size_t)(oend - segmentEnd))
+                segmentEnd += segmentSize;
+            else
+                segmentEnd = oend;
+            FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
+            args.op[i] += HUF_decodeStreamX2(args.op[i], &bit, segmentEnd, (HUF_DEltX2 const*)dt, HUF_DECODER_FAST_TABLELOG);
+            if (args.op[i] != segmentEnd)
+                return ERROR(corruption_detected);
+        }
+    }
+
+    /* decoded size */
+    return dstSize;
+}
+
+static size_t HUF_decompress4X2_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
+{
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X2_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X2_usingDTable_internal_fast_c_loop;
+
+#if DYNAMIC_BMI2
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X2_usingDTable_internal_bmi2;
+# if ZSTD_ENABLE_ASM_X86_64_BMI2
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+        }
+# endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+    }
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+    }
+#endif
+
+    if (!(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X2_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
+
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize,
+                                               workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, flags);
+}
+
+static size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize,
+                                         workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
+}
+
+#endif /* HUF_FORCE_DECOMPRESS_X1 */
+
+
+/* ***********************************/
+/* Universal decompression selectors */
+/* ***********************************/
+
+
+#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][2 /* single, double */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}},  /* Q==1 : impossible */
+    {{ 150,216}, { 381,119}},   /* Q == 2 : 12-18% */
+    {{ 170,205}, { 514,112}},   /* Q == 3 : 18-25% */
+    {{ 177,199}, { 539,110}},   /* Q == 4 : 25-32% */
+    {{ 197,194}, { 644,107}},   /* Q == 5 : 32-38% */
+    {{ 221,192}, { 735,107}},   /* Q == 6 : 38-44% */
+    {{ 256,189}, { 881,106}},   /* Q == 7 : 44-50% */
+    {{ 359,188}, {1167,109}},   /* Q == 8 : 50-56% */
+    {{ 582,187}, {1570,114}},   /* Q == 9 : 56-62% */
+    {{ 688,187}, {1712,122}},   /* Q ==10 : 62-69% */
+    {{ 825,186}, {1965,136}},   /* Q ==11 : 69-75% */
+    {{ 976,185}, {2131,150}},   /* Q ==12 : 75-81% */
+    {{1180,186}, {2070,175}},   /* Q ==13 : 81-87% */
+    {{1377,185}, {1731,202}},   /* Q ==14 : 87-93% */
+    {{1412,185}, {1695,202}},   /* Q ==15 : 93-99% */
+};
+#endif
+
+/** HUF_selectDecoder() :
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
+ *  Assumption : 0 < dstSize <= 128 KB */
+U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
+{
+    assert(dstSize > 0);
+    assert(dstSize <= 128*1024);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dstSize;
+    (void)cSrcSize;
+    return 0;
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dstSize;
+    (void)cSrcSize;
+    return 1;
+#else
+    /* decoder timing evaluation */
+    {   U32 const Q = (cSrcSize >= dstSize) ? 15 : (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+        U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+        DTime1 += DTime1 >> 5;  /* small advantage to algorithm using less memory, to reduce cache eviction */
+        return DTime1 < DTime0;
+    }
+#endif
+}
+
+size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                  const void* cSrc, size_t cSrcSize,
+                                  void* workSpace, size_t wkspSize, int flags)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags);
+#else
+        return algoNb ? HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags):
+                        HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags);
+#endif
+    }
+}
+
+
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#else
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#endif
+}
+
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
+}
+#endif
+
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#else
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#endif
+}
+
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize == 0) return ERROR(corruption_detected);
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
+#else
+        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags) :
+                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
+#endif
+    }
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/huf_decompress_amd64.S b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/huf_decompress_amd64.S
new file mode 100644
index 0000000..671624f
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/huf_decompress_amd64.S
@@ -0,0 +1,576 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "../common/portability_macros.h"
+
+/* Stack marking
+ * ref: https://wiki.gentoo.org/wiki/Hardened/GNU_stack_quickstart
+ */
+#if defined(__ELF__) && defined(__GNUC__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
+
+/* Calling convention:
+ *
+ * %rdi contains the first argument: HUF_DecompressAsmArgs*.
+ * %rbp isn't maintained (no frame pointer).
+ * %rsp contains the stack pointer that grows down.
+ *      No red-zone is assumed, only addresses >= %rsp are used.
+ * All register contents are preserved.
+ *
+ * TODO: Support Windows calling convention.
+ */
+
+ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
+ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
+ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
+ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
+.global HUF_decompress4X1_usingDTable_internal_fast_asm_loop
+.global HUF_decompress4X2_usingDTable_internal_fast_asm_loop
+.global _HUF_decompress4X1_usingDTable_internal_fast_asm_loop
+.global _HUF_decompress4X2_usingDTable_internal_fast_asm_loop
+.text
+
+/* Sets up register mappings for clarity.
+ * op[], bits[], dtable & ip[0] each get their own register.
+ * ip[1,2,3] & olimit alias var[].
+ * %rax is a scratch register.
+ */
+
+#define op0    rsi
+#define op1    rbx
+#define op2    rcx
+#define op3    rdi
+
+#define ip0    r8
+#define ip1    r9
+#define ip2    r10
+#define ip3    r11
+
+#define bits0  rbp
+#define bits1  rdx
+#define bits2  r12
+#define bits3  r13
+#define dtable r14
+#define olimit r15
+
+/* var[] aliases ip[1,2,3] & olimit
+ * ip[1,2,3] are saved every iteration.
+ * olimit is only used in compute_olimit.
+ */
+#define var0   r15
+#define var1   r9
+#define var2   r10
+#define var3   r11
+
+/* 32-bit var registers */
+#define vard0  r15d
+#define vard1  r9d
+#define vard2  r10d
+#define vard3  r11d
+
+/* Calls X(N) for each stream 0, 1, 2, 3. */
+#define FOR_EACH_STREAM(X) \
+    X(0);                  \
+    X(1);                  \
+    X(2);                  \
+    X(3)
+
+/* Calls X(N, idx) for each stream 0, 1, 2, 3. */
+#define FOR_EACH_STREAM_WITH_INDEX(X, idx) \
+    X(0, idx);                             \
+    X(1, idx);                             \
+    X(2, idx);                             \
+    X(3, idx)
+
+/* Define both _HUF_* & HUF_* symbols because MacOS
+ * C symbols are prefixed with '_' & Linux symbols aren't.
+ */
+_HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
+HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
+    ZSTD_CET_ENDBRANCH
+    /* Save all registers - even if they are callee saved for simplicity. */
+    push %rax
+    push %rbx
+    push %rcx
+    push %rdx
+    push %rbp
+    push %rsi
+    push %rdi
+    push %r8
+    push %r9
+    push %r10
+    push %r11
+    push %r12
+    push %r13
+    push %r14
+    push %r15
+
+    /* Read HUF_DecompressAsmArgs* args from %rax */
+    movq %rdi, %rax
+    movq  0(%rax), %ip0
+    movq  8(%rax), %ip1
+    movq 16(%rax), %ip2
+    movq 24(%rax), %ip3
+    movq 32(%rax), %op0
+    movq 40(%rax), %op1
+    movq 48(%rax), %op2
+    movq 56(%rax), %op3
+    movq 64(%rax), %bits0
+    movq 72(%rax), %bits1
+    movq 80(%rax), %bits2
+    movq 88(%rax), %bits3
+    movq 96(%rax), %dtable
+    push %rax      /* argument */
+    push 104(%rax) /* ilimit */
+    push 112(%rax) /* oend */
+    push %olimit   /* olimit space */
+
+    subq $24, %rsp
+
+.L_4X1_compute_olimit:
+    /* Computes how many iterations we can do safely
+     * %r15, %rax may be clobbered
+     * rbx, rdx must be saved
+     * op3 & ip0 mustn't be clobbered
+     */
+    movq %rbx, 0(%rsp)
+    movq %rdx, 8(%rsp)
+
+    movq 32(%rsp), %rax /* rax = oend */
+    subq %op3,    %rax  /* rax = oend - op3 */
+
+    /* r15 = (oend - op3) / 5 */
+    movabsq $-3689348814741910323, %rdx
+    mulq %rdx
+    movq %rdx, %r15
+    shrq $2, %r15
+
+    movq %ip0,     %rax /* rax = ip0 */
+    movq 40(%rsp), %rdx /* rdx = ilimit */
+    subq %rdx,     %rax /* rax = ip0 - ilimit */
+    movq %rax,     %rbx /* rbx = ip0 - ilimit */
+
+    /* rdx = (ip0 - ilimit) / 7 */
+    movabsq $2635249153387078803, %rdx
+    mulq %rdx
+    subq %rdx, %rbx
+    shrq %rbx
+    addq %rbx, %rdx
+    shrq $2, %rdx
+
+    /* r15 = min(%rdx, %r15) */
+    cmpq %rdx, %r15
+    cmova %rdx, %r15
+
+    /* r15 = r15 * 5 */
+    leaq (%r15, %r15, 4), %r15
+
+    /* olimit = op3 + r15 */
+    addq %op3, %olimit
+
+    movq 8(%rsp), %rdx
+    movq 0(%rsp), %rbx
+
+    /* If (op3 + 20 > olimit) */
+    movq %op3, %rax    /* rax = op3 */
+    addq $20,  %rax    /* rax = op3 + 20 */
+    cmpq %rax, %olimit /* op3 + 20 > olimit */
+    jb .L_4X1_exit
+
+    /* If (ip1 < ip0) go to exit */
+    cmpq %ip0, %ip1
+    jb .L_4X1_exit
+
+    /* If (ip2 < ip1) go to exit */
+    cmpq %ip1, %ip2
+    jb .L_4X1_exit
+
+    /* If (ip3 < ip2) go to exit */
+    cmpq %ip2, %ip3
+    jb .L_4X1_exit
+
+/* Reads top 11 bits from bits[n]
+ * Loads dt[bits[n]] into var[n]
+ */
+#define GET_NEXT_DELT(n)                \
+    movq $53, %var##n;                  \
+    shrxq %var##n, %bits##n, %var##n;   \
+    movzwl (%dtable,%var##n,2),%vard##n
+
+/* var[n] must contain the DTable entry computed with GET_NEXT_DELT
+ * Moves var[n] to %rax
+ * bits[n] <<= var[n] & 63
+ * op[n][idx] = %rax >> 8
+ * %ah is a way to access bits [8, 16) of %rax
+ */
+#define DECODE_FROM_DELT(n, idx)       \
+    movq %var##n, %rax;                \
+    shlxq %var##n, %bits##n, %bits##n; \
+    movb %ah, idx(%op##n)
+
+/* Assumes GET_NEXT_DELT has been called.
+ * Calls DECODE_FROM_DELT then GET_NEXT_DELT
+ */
+#define DECODE_AND_GET_NEXT(n, idx) \
+    DECODE_FROM_DELT(n, idx);       \
+    GET_NEXT_DELT(n)                \
+
+/* // ctz & nbBytes is stored in bits[n]
+ * // nbBits is stored in %rax
+ * ctz  = CTZ[bits[n]]
+ * nbBits  = ctz & 7
+ * nbBytes = ctz >> 3
+ * op[n]  += 5
+ * ip[n]  -= nbBytes
+ * // Note: x86-64 is little-endian ==> no bswap
+ * bits[n] = MEM_readST(ip[n]) | 1
+ * bits[n] <<= nbBits
+ */
+#define RELOAD_BITS(n)             \
+    bsfq %bits##n, %bits##n;       \
+    movq %bits##n, %rax;           \
+    andq $7, %rax;                 \
+    shrq $3, %bits##n;             \
+    leaq 5(%op##n), %op##n;        \
+    subq %bits##n, %ip##n;         \
+    movq (%ip##n), %bits##n;       \
+    orq $1, %bits##n;              \
+    shlx %rax, %bits##n, %bits##n
+
+    /* Store clobbered variables on the stack */
+    movq %olimit, 24(%rsp)
+    movq %ip1, 0(%rsp)
+    movq %ip2, 8(%rsp)
+    movq %ip3, 16(%rsp)
+
+    /* Call GET_NEXT_DELT for each stream */
+    FOR_EACH_STREAM(GET_NEXT_DELT)
+
+    .p2align 6
+
+.L_4X1_loop_body:
+    /* Decode 5 symbols in each of the 4 streams (20 total)
+     * Must have called GET_NEXT_DELT for each stream
+     */
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 0)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 1)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 2)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 3)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_FROM_DELT, 4)
+
+    /* Load ip[1,2,3] from stack (var[] aliases them)
+     * ip[] is needed for RELOAD_BITS
+     * Each will be stored back to the stack after RELOAD
+     */
+    movq 0(%rsp), %ip1
+    movq 8(%rsp), %ip2
+    movq 16(%rsp), %ip3
+
+    /* Reload each stream & fetch the next table entry
+     * to prepare for the next iteration
+     */
+    RELOAD_BITS(0)
+    GET_NEXT_DELT(0)
+
+    RELOAD_BITS(1)
+    movq %ip1, 0(%rsp)
+    GET_NEXT_DELT(1)
+
+    RELOAD_BITS(2)
+    movq %ip2, 8(%rsp)
+    GET_NEXT_DELT(2)
+
+    RELOAD_BITS(3)
+    movq %ip3, 16(%rsp)
+    GET_NEXT_DELT(3)
+
+    /* If op3 < olimit: continue the loop */
+    cmp %op3, 24(%rsp)
+    ja .L_4X1_loop_body
+
+    /* Reload ip[1,2,3] from stack */
+    movq 0(%rsp), %ip1
+    movq 8(%rsp), %ip2
+    movq 16(%rsp), %ip3
+
+    /* Re-compute olimit */
+    jmp .L_4X1_compute_olimit
+
+#undef GET_NEXT_DELT
+#undef DECODE_FROM_DELT
+#undef DECODE
+#undef RELOAD_BITS
+.L_4X1_exit:
+    addq $24, %rsp
+
+    /* Restore stack (oend & olimit) */
+    pop %rax /* olimit */
+    pop %rax /* oend */
+    pop %rax /* ilimit */
+    pop %rax /* arg */
+
+    /* Save ip / op / bits */
+    movq %ip0,  0(%rax)
+    movq %ip1,  8(%rax)
+    movq %ip2, 16(%rax)
+    movq %ip3, 24(%rax)
+    movq %op0, 32(%rax)
+    movq %op1, 40(%rax)
+    movq %op2, 48(%rax)
+    movq %op3, 56(%rax)
+    movq %bits0, 64(%rax)
+    movq %bits1, 72(%rax)
+    movq %bits2, 80(%rax)
+    movq %bits3, 88(%rax)
+
+    /* Restore registers */
+    pop %r15
+    pop %r14
+    pop %r13
+    pop %r12
+    pop %r11
+    pop %r10
+    pop %r9
+    pop %r8
+    pop %rdi
+    pop %rsi
+    pop %rbp
+    pop %rdx
+    pop %rcx
+    pop %rbx
+    pop %rax
+    ret
+
+_HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
+HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
+    ZSTD_CET_ENDBRANCH
+    /* Save all registers - even if they are callee saved for simplicity. */
+    push %rax
+    push %rbx
+    push %rcx
+    push %rdx
+    push %rbp
+    push %rsi
+    push %rdi
+    push %r8
+    push %r9
+    push %r10
+    push %r11
+    push %r12
+    push %r13
+    push %r14
+    push %r15
+
+    movq %rdi, %rax
+    movq  0(%rax), %ip0
+    movq  8(%rax), %ip1
+    movq 16(%rax), %ip2
+    movq 24(%rax), %ip3
+    movq 32(%rax), %op0
+    movq 40(%rax), %op1
+    movq 48(%rax), %op2
+    movq 56(%rax), %op3
+    movq 64(%rax), %bits0
+    movq 72(%rax), %bits1
+    movq 80(%rax), %bits2
+    movq 88(%rax), %bits3
+    movq 96(%rax), %dtable
+    push %rax      /* argument */
+    push %rax      /* olimit */
+    push 104(%rax) /* ilimit */
+
+    movq 112(%rax), %rax
+    push %rax /* oend3 */
+
+    movq %op3, %rax
+    push %rax /* oend2 */
+
+    movq %op2, %rax
+    push %rax /* oend1 */
+
+    movq %op1, %rax
+    push %rax /* oend0 */
+
+    /* Scratch space */
+    subq $8, %rsp
+
+.L_4X2_compute_olimit:
+    /* Computes how many iterations we can do safely
+     * %r15, %rax may be clobbered
+     * rdx must be saved
+     * op[1,2,3,4] & ip0 mustn't be clobbered
+     */
+    movq %rdx, 0(%rsp)
+
+    /* We can consume up to 7 input bytes each iteration. */
+    movq %ip0,     %rax  /* rax = ip0 */
+    movq 40(%rsp), %rdx  /* rdx = ilimit */
+    subq %rdx,     %rax  /* rax = ip0 - ilimit */
+    movq %rax,    %r15   /* r15 = ip0 - ilimit */
+
+    /* rdx = rax / 7 */
+    movabsq $2635249153387078803, %rdx
+    mulq %rdx
+    subq %rdx, %r15
+    shrq %r15
+    addq %r15, %rdx
+    shrq $2, %rdx
+
+    /* r15 = (ip0 - ilimit) / 7 */
+    movq %rdx, %r15
+
+    /* r15 = min(r15, min(oend0 - op0, oend1 - op1, oend2 - op2, oend3 - op3) / 10) */
+    movq 8(%rsp),  %rax /* rax = oend0 */
+    subq %op0,     %rax /* rax = oend0 - op0 */
+    movq 16(%rsp), %rdx /* rdx = oend1 */
+    subq %op1,     %rdx /* rdx = oend1 - op1 */
+
+    cmpq  %rax,    %rdx
+    cmova %rax,    %rdx /* rdx = min(%rdx, %rax) */
+
+    movq 24(%rsp), %rax /* rax = oend2 */
+    subq %op2,     %rax /* rax = oend2 - op2 */
+
+    cmpq  %rax,    %rdx
+    cmova %rax,    %rdx /* rdx = min(%rdx, %rax) */
+
+    movq 32(%rsp), %rax /* rax = oend3 */
+    subq %op3,     %rax /* rax = oend3 - op3 */
+
+    cmpq  %rax,    %rdx
+    cmova %rax,    %rdx /* rdx = min(%rdx, %rax) */
+
+    movabsq $-3689348814741910323, %rax
+    mulq %rdx
+    shrq $3,       %rdx /* rdx = rdx / 10 */
+
+    /* r15 = min(%rdx, %r15) */
+    cmpq  %rdx, %r15
+    cmova %rdx, %r15
+
+    /* olimit = op3 + 5 * r15 */
+    movq %r15, %rax
+    leaq (%op3, %rax, 4), %olimit
+    addq %rax, %olimit
+
+    movq 0(%rsp), %rdx
+
+    /* If (op3 + 10 > olimit) */
+    movq %op3, %rax    /* rax = op3 */
+    addq $10,  %rax    /* rax = op3 + 10 */
+    cmpq %rax, %olimit /* op3 + 10 > olimit */
+    jb .L_4X2_exit
+
+    /* If (ip1 < ip0) go to exit */
+    cmpq %ip0, %ip1
+    jb .L_4X2_exit
+
+    /* If (ip2 < ip1) go to exit */
+    cmpq %ip1, %ip2
+    jb .L_4X2_exit
+
+    /* If (ip3 < ip2) go to exit */
+    cmpq %ip2, %ip3
+    jb .L_4X2_exit
+
+#define DECODE(n, idx)              \
+    movq %bits##n, %rax;            \
+    shrq $53, %rax;                 \
+    movzwl 0(%dtable,%rax,4),%r8d;  \
+    movzbl 2(%dtable,%rax,4),%r15d; \
+    movzbl 3(%dtable,%rax,4),%eax;  \
+    movw %r8w, (%op##n);            \
+    shlxq %r15, %bits##n, %bits##n; \
+    addq %rax, %op##n
+
+#define RELOAD_BITS(n)              \
+    bsfq %bits##n, %bits##n;        \
+    movq %bits##n, %rax;            \
+    shrq $3, %bits##n;              \
+    andq $7, %rax;                  \
+    subq %bits##n, %ip##n;          \
+    movq (%ip##n), %bits##n;        \
+    orq $1, %bits##n;               \
+    shlxq %rax, %bits##n, %bits##n
+
+
+    movq %olimit, 48(%rsp)
+
+    .p2align 6
+
+.L_4X2_loop_body:
+    /* We clobber r8, so store it on the stack */
+    movq %r8, 0(%rsp)
+
+    /* Decode 5 symbols from each of the 4 streams (20 symbols total). */
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 0)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 1)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 2)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 3)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 4)
+
+    /* Reload r8 */
+    movq 0(%rsp), %r8
+
+    FOR_EACH_STREAM(RELOAD_BITS)
+
+    cmp %op3, 48(%rsp)
+    ja .L_4X2_loop_body
+    jmp .L_4X2_compute_olimit
+
+#undef DECODE
+#undef RELOAD_BITS
+.L_4X2_exit:
+    addq $8, %rsp
+    /* Restore stack (oend & olimit) */
+    pop %rax /* oend0 */
+    pop %rax /* oend1 */
+    pop %rax /* oend2 */
+    pop %rax /* oend3 */
+    pop %rax /* ilimit */
+    pop %rax /* olimit */
+    pop %rax /* arg */
+
+    /* Save ip / op / bits */
+    movq %ip0,  0(%rax)
+    movq %ip1,  8(%rax)
+    movq %ip2, 16(%rax)
+    movq %ip3, 24(%rax)
+    movq %op0, 32(%rax)
+    movq %op1, 40(%rax)
+    movq %op2, 48(%rax)
+    movq %op3, 56(%rax)
+    movq %bits0, 64(%rax)
+    movq %bits1, 72(%rax)
+    movq %bits2, 80(%rax)
+    movq %bits3, 88(%rax)
+
+    /* Restore registers */
+    pop %r15
+    pop %r14
+    pop %r13
+    pop %r12
+    pop %r11
+    pop %r10
+    pop %r9
+    pop %r8
+    pop %rdi
+    pop %rsi
+    pop %rbp
+    pop %rdx
+    pop %rcx
+    pop %rbx
+    pop %rax
+    ret
+
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_ddict.c b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_ddict.c
new file mode 100644
index 0000000..309ec0d
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_ddict.c
@@ -0,0 +1,244 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* zstd_ddict.c :
+ * concentrates all logic that needs to know the internals of ZSTD_DDict object */
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customFree */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/cpu.h"         /* bmi2 */
+#include "../common/mem.h"         /* low level memory routines */
+#define FSE_STATIC_LINKING_ONLY
+#include "../common/fse.h"
+#include "../common/huf.h"
+#include "zstd_decompress_internal.h"
+#include "zstd_ddict.h"
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+#  include "../legacy/zstd_legacy.h"
+#endif
+
+
+
+/*-*******************************************************
+*  Types
+*********************************************************/
+struct ZSTD_DDict_s {
+    void* dictBuffer;
+    const void* dictContent;
+    size_t dictSize;
+    ZSTD_entropyDTables_t entropy;
+    U32 dictID;
+    U32 entropyPresent;
+    ZSTD_customMem cMem;
+};  /* typedef'd to ZSTD_DDict within "zstd.h" */
+
+const void* ZSTD_DDict_dictContent(const ZSTD_DDict* ddict)
+{
+    assert(ddict != NULL);
+    return ddict->dictContent;
+}
+
+size_t ZSTD_DDict_dictSize(const ZSTD_DDict* ddict)
+{
+    assert(ddict != NULL);
+    return ddict->dictSize;
+}
+
+void ZSTD_copyDDictParameters(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    DEBUGLOG(4, "ZSTD_copyDDictParameters");
+    assert(dctx != NULL);
+    assert(ddict != NULL);
+    dctx->dictID = ddict->dictID;
+    dctx->prefixStart = ddict->dictContent;
+    dctx->virtualStart = ddict->dictContent;
+    dctx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
+    dctx->previousDstEnd = dctx->dictEnd;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentBeginForFuzzing = dctx->prefixStart;
+    dctx->dictContentEndForFuzzing = dctx->previousDstEnd;
+#endif
+    if (ddict->entropyPresent) {
+        dctx->litEntropy = 1;
+        dctx->fseEntropy = 1;
+        dctx->LLTptr = ddict->entropy.LLTable;
+        dctx->MLTptr = ddict->entropy.MLTable;
+        dctx->OFTptr = ddict->entropy.OFTable;
+        dctx->HUFptr = ddict->entropy.hufTable;
+        dctx->entropy.rep[0] = ddict->entropy.rep[0];
+        dctx->entropy.rep[1] = ddict->entropy.rep[1];
+        dctx->entropy.rep[2] = ddict->entropy.rep[2];
+    } else {
+        dctx->litEntropy = 0;
+        dctx->fseEntropy = 0;
+    }
+}
+
+
+static size_t
+ZSTD_loadEntropy_intoDDict(ZSTD_DDict* ddict,
+                           ZSTD_dictContentType_e dictContentType)
+{
+    ddict->dictID = 0;
+    ddict->entropyPresent = 0;
+    if (dictContentType == ZSTD_dct_rawContent) return 0;
+
+    if (ddict->dictSize < 8) {
+        if (dictContentType == ZSTD_dct_fullDict)
+            return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+        return 0;   /* pure content mode */
+    }
+    {   U32 const magic = MEM_readLE32(ddict->dictContent);
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
+            if (dictContentType == ZSTD_dct_fullDict)
+                return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+            return 0;   /* pure content mode */
+        }
+    }
+    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_FRAMEIDSIZE);
+
+    /* load entropy tables */
+    RETURN_ERROR_IF(ZSTD_isError(ZSTD_loadDEntropy(
+            &ddict->entropy, ddict->dictContent, ddict->dictSize)),
+        dictionary_corrupted, "");
+    ddict->entropyPresent = 1;
+    return 0;
+}
+
+
+static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType)
+{
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dict) || (!dictSize)) {
+        ddict->dictBuffer = NULL;
+        ddict->dictContent = dict;
+        if (!dict) dictSize = 0;
+    } else {
+        void* const internalBuffer = ZSTD_customMalloc(dictSize, ddict->cMem);
+        ddict->dictBuffer = internalBuffer;
+        ddict->dictContent = internalBuffer;
+        if (!internalBuffer) return ERROR(memory_allocation);
+        ZSTD_memcpy(internalBuffer, dict, dictSize);
+    }
+    ddict->dictSize = dictSize;
+    ddict->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);  /* cover both little and big endian */
+
+    /* parse dictionary content */
+    FORWARD_IF_ERROR( ZSTD_loadEntropy_intoDDict(ddict, dictContentType) , "");
+
+    return 0;
+}
+
+ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_customMem customMem)
+{
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+
+    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_customMalloc(sizeof(ZSTD_DDict), customMem);
+        if (ddict == NULL) return NULL;
+        ddict->cMem = customMem;
+        {   size_t const initResult = ZSTD_initDDict_internal(ddict,
+                                            dict, dictSize,
+                                            dictLoadMethod, dictContentType);
+            if (ZSTD_isError(initResult)) {
+                ZSTD_freeDDict(ddict);
+                return NULL;
+        }   }
+        return ddict;
+    }
+}
+
+/*! ZSTD_createDDict() :
+*   Create a digested dictionary, to start decompression without startup delay.
+*   `dict` content is copied inside DDict.
+*   Consequently, `dict` can be released after `ZSTD_DDict` creation */
+ZSTD_DDict* ZSTD_createDDict(const void* dict, size_t dictSize)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTD_createDDict_advanced(dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto, allocator);
+}
+
+/*! ZSTD_createDDict_byReference() :
+ *  Create a digested dictionary, to start decompression without startup delay.
+ *  Dictionary content is simply referenced, it will be accessed during decompression.
+ *  Warning : dictBuffer must outlive DDict (DDict must be freed before dictBuffer) */
+ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTD_createDDict_advanced(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto, allocator);
+}
+
+
+const ZSTD_DDict* ZSTD_initStaticDDict(
+                                void* sBuffer, size_t sBufferSize,
+                                const void* dict, size_t dictSize,
+                                ZSTD_dictLoadMethod_e dictLoadMethod,
+                                ZSTD_dictContentType_e dictContentType)
+{
+    size_t const neededSpace = sizeof(ZSTD_DDict)
+                             + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+    ZSTD_DDict* const ddict = (ZSTD_DDict*)sBuffer;
+    assert(sBuffer != NULL);
+    assert(dict != NULL);
+    if ((size_t)sBuffer & 7) return NULL;   /* 8-aligned */
+    if (sBufferSize < neededSpace) return NULL;
+    if (dictLoadMethod == ZSTD_dlm_byCopy) {
+        ZSTD_memcpy(ddict+1, dict, dictSize);  /* local copy */
+        dict = ddict+1;
+    }
+    if (ZSTD_isError( ZSTD_initDDict_internal(ddict,
+                                              dict, dictSize,
+                                              ZSTD_dlm_byRef, dictContentType) ))
+        return NULL;
+    return ddict;
+}
+
+
+size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = ddict->cMem;
+        ZSTD_customFree(ddict->dictBuffer, cMem);
+        ZSTD_customFree(ddict, cMem);
+        return 0;
+    }
+}
+
+/*! ZSTD_estimateDDictSize() :
+ *  Estimate amount of memory that will be needed to create a dictionary for decompression.
+ *  Note : dictionary created by reference using ZSTD_dlm_byRef are smaller */
+size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    return sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+}
+
+size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*ddict) + (ddict->dictBuffer ? ddict->dictSize : 0) ;
+}
+
+/*! ZSTD_getDictID_fromDDict() :
+ *  Provides the dictID of the dictionary loaded into `ddict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;
+    return ddict->dictID;
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_ddict.h b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_ddict.h
new file mode 100644
index 0000000..c4ca887
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_ddict.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#ifndef ZSTD_DDICT_H
+#define ZSTD_DDICT_H
+
+/*-*******************************************************
+ *  Dependencies
+ *********************************************************/
+#include "../common/zstd_deps.h"   /* size_t */
+#include "../zstd.h"     /* ZSTD_DDict, and several public functions */
+
+
+/*-*******************************************************
+ *  Interface
+ *********************************************************/
+
+/* note: several prototypes are already published in `zstd.h` :
+ * ZSTD_createDDict()
+ * ZSTD_createDDict_byReference()
+ * ZSTD_createDDict_advanced()
+ * ZSTD_freeDDict()
+ * ZSTD_initStaticDDict()
+ * ZSTD_sizeof_DDict()
+ * ZSTD_estimateDDictSize()
+ * ZSTD_getDictID_fromDict()
+ */
+
+const void* ZSTD_DDict_dictContent(const ZSTD_DDict* ddict);
+size_t ZSTD_DDict_dictSize(const ZSTD_DDict* ddict);
+
+void ZSTD_copyDDictParameters(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
+
+
+#endif /* ZSTD_DDICT_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress.c b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress.c
new file mode 100644
index 0000000..7bc2713
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress.c
@@ -0,0 +1,2355 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTD_decompress() allocates its context,
+ * on stack (0), or into heap (1, default; requires malloc()).
+ * Note that functions with explicit context such as ZSTD_decompressDCtx() are unaffected.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif
+
+/*!
+*  LEGACY_SUPPORT :
+*  if set to 1+, ZSTD_decompress() can decode older formats (v0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 0
+#endif
+
+/*!
+ *  MAXWINDOWSIZE_DEFAULT :
+ *  maximum window size accepted by DStream __by default__.
+ *  Frames requiring more memory will be rejected.
+ *  It's possible to set a different limit using ZSTD_DCtx_setMaxWindowSize().
+ */
+#ifndef ZSTD_MAXWINDOWSIZE_DEFAULT
+#  define ZSTD_MAXWINDOWSIZE_DEFAULT (((U32)1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT) + 1)
+#endif
+
+/*!
+ *  NO_FORWARD_PROGRESS_MAX :
+ *  maximum allowed nb of calls to ZSTD_decompressStream()
+ *  without any forward progress
+ *  (defined as: no byte read from input, and no byte flushed to output)
+ *  before triggering an error.
+ */
+#ifndef ZSTD_NO_FORWARD_PROGRESS_MAX
+#  define ZSTD_NO_FORWARD_PROGRESS_MAX 16
+#endif
+
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/mem.h"         /* low level memory routines */
+#define FSE_STATIC_LINKING_ONLY
+#include "../common/fse.h"
+#include "../common/huf.h"
+#include "../common/xxhash.h" /* XXH64_reset, XXH64_update, XXH64_digest, XXH64 */
+#include "../common/zstd_internal.h"  /* blockProperties_t */
+#include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
+#include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
+#include "zstd_decompress_block.h"   /* ZSTD_decompressBlock_internal */
+#include "../common/bits.h"  /* ZSTD_highbit32 */
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+#  include "../legacy/zstd_legacy.h"
+#endif
+
+
+
+/*************************************
+ * Multiple DDicts Hashset internals *
+ *************************************/
+
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT 4
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3  /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
+                                                    * Currently, that means a 0.75 load factor.
+                                                    * So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
+                                                    * the load factor of the ddict hash set.
+                                                    */
+
+#define DDICT_HASHSET_TABLE_BASE_SIZE 64
+#define DDICT_HASHSET_RESIZE_FACTOR 2
+
+/* Hash function to determine starting position of dict insertion within the table
+ * Returns an index between [0, hashSet->ddictPtrTableSize]
+ */
+static size_t ZSTD_DDictHashSet_getIndex(const ZSTD_DDictHashSet* hashSet, U32 dictID) {
+    const U64 hash = XXH64(&dictID, sizeof(U32), 0);
+    /* DDict ptr table size is a multiple of 2, use size - 1 as mask to get index within [0, hashSet->ddictPtrTableSize) */
+    return hash & (hashSet->ddictPtrTableSize - 1);
+}
+
+/* Adds DDict to a hashset without resizing it.
+ * If inserting a DDict with a dictID that already exists in the set, replaces the one in the set.
+ * Returns 0 if successful, or a zstd error code if something went wrong.
+ */
+static size_t ZSTD_DDictHashSet_emplaceDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict) {
+    const U32 dictID = ZSTD_getDictID_fromDDict(ddict);
+    size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
+    const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
+    RETURN_ERROR_IF(hashSet->ddictPtrCount == hashSet->ddictPtrTableSize, GENERIC, "Hash set is full!");
+    DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
+    while (hashSet->ddictPtrTable[idx] != NULL) {
+        /* Replace existing ddict if inserting ddict with same dictID */
+        if (ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]) == dictID) {
+            DEBUGLOG(4, "DictID already exists, replacing rather than adding");
+            hashSet->ddictPtrTable[idx] = ddict;
+            return 0;
+        }
+        idx &= idxRangeMask;
+        idx++;
+    }
+    DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
+    hashSet->ddictPtrTable[idx] = ddict;
+    hashSet->ddictPtrCount++;
+    return 0;
+}
+
+/* Expands hash table by factor of DDICT_HASHSET_RESIZE_FACTOR and
+ * rehashes all values, allocates new table, frees old table.
+ * Returns 0 on success, otherwise a zstd error code.
+ */
+static size_t ZSTD_DDictHashSet_expand(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
+    size_t newTableSize = hashSet->ddictPtrTableSize * DDICT_HASHSET_RESIZE_FACTOR;
+    const ZSTD_DDict** newTable = (const ZSTD_DDict**)ZSTD_customCalloc(sizeof(ZSTD_DDict*) * newTableSize, customMem);
+    const ZSTD_DDict** oldTable = hashSet->ddictPtrTable;
+    size_t oldTableSize = hashSet->ddictPtrTableSize;
+    size_t i;
+
+    DEBUGLOG(4, "Expanding DDict hash table! Old size: %zu new size: %zu", oldTableSize, newTableSize);
+    RETURN_ERROR_IF(!newTable, memory_allocation, "Expanded hashset allocation failed!");
+    hashSet->ddictPtrTable = newTable;
+    hashSet->ddictPtrTableSize = newTableSize;
+    hashSet->ddictPtrCount = 0;
+    for (i = 0; i < oldTableSize; ++i) {
+        if (oldTable[i] != NULL) {
+            FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, oldTable[i]), "");
+        }
+    }
+    ZSTD_customFree((void*)oldTable, customMem);
+    DEBUGLOG(4, "Finished re-hash");
+    return 0;
+}
+
+/* Fetches a DDict with the given dictID
+ * Returns the ZSTD_DDict* with the requested dictID. If it doesn't exist, then returns NULL.
+ */
+static const ZSTD_DDict* ZSTD_DDictHashSet_getDDict(ZSTD_DDictHashSet* hashSet, U32 dictID) {
+    size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
+    const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
+    DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
+    for (;;) {
+        size_t currDictID = ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]);
+        if (currDictID == dictID || currDictID == 0) {
+            /* currDictID == 0 implies a NULL ddict entry */
+            break;
+        } else {
+            idx &= idxRangeMask;    /* Goes to start of table when we reach the end */
+            idx++;
+        }
+    }
+    DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
+    return hashSet->ddictPtrTable[idx];
+}
+
+/* Allocates space for and returns a ddict hash set
+ * The hash set's ZSTD_DDict* table has all values automatically set to NULL to begin with.
+ * Returns NULL if allocation failed.
+ */
+static ZSTD_DDictHashSet* ZSTD_createDDictHashSet(ZSTD_customMem customMem) {
+    ZSTD_DDictHashSet* ret = (ZSTD_DDictHashSet*)ZSTD_customMalloc(sizeof(ZSTD_DDictHashSet), customMem);
+    DEBUGLOG(4, "Allocating new hash set");
+    if (!ret)
+        return NULL;
+    ret->ddictPtrTable = (const ZSTD_DDict**)ZSTD_customCalloc(DDICT_HASHSET_TABLE_BASE_SIZE * sizeof(ZSTD_DDict*), customMem);
+    if (!ret->ddictPtrTable) {
+        ZSTD_customFree(ret, customMem);
+        return NULL;
+    }
+    ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
+    ret->ddictPtrCount = 0;
+    return ret;
+}
+
+/* Frees the table of ZSTD_DDict* within a hashset, then frees the hashset itself.
+ * Note: The ZSTD_DDict* within the table are NOT freed.
+ */
+static void ZSTD_freeDDictHashSet(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
+    DEBUGLOG(4, "Freeing ddict hash set");
+    if (hashSet && hashSet->ddictPtrTable) {
+        ZSTD_customFree((void*)hashSet->ddictPtrTable, customMem);
+    }
+    if (hashSet) {
+        ZSTD_customFree(hashSet, customMem);
+    }
+}
+
+/* Public function: Adds a DDict into the ZSTD_DDictHashSet, possibly triggering a resize of the hash set.
+ * Returns 0 on success, or a ZSTD error.
+ */
+static size_t ZSTD_DDictHashSet_addDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict, ZSTD_customMem customMem) {
+    DEBUGLOG(4, "Adding dict ID: %u to hashset with - Count: %zu Tablesize: %zu", ZSTD_getDictID_fromDDict(ddict), hashSet->ddictPtrCount, hashSet->ddictPtrTableSize);
+    if (hashSet->ddictPtrCount * DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT / hashSet->ddictPtrTableSize * DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT != 0) {
+        FORWARD_IF_ERROR(ZSTD_DDictHashSet_expand(hashSet, customMem), "");
+    }
+    FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, ddict), "");
+    return 0;
+}
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support sizeof NULL */
+    return sizeof(*dctx)
+           + ZSTD_sizeof_DDict(dctx->ddictLocal)
+           + dctx->inBuffSize + dctx->outBuffSize;
+}
+
+size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
+
+
+static size_t ZSTD_startingInputLength(ZSTD_format_e format)
+{
+    size_t const startingInputLength = ZSTD_FRAMEHEADERSIZE_PREFIX(format);
+    /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
+    assert( (format == ZSTD_f_zstd1) || (format == ZSTD_f_zstd1_magicless) );
+    return startingInputLength;
+}
+
+static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
+{
+    assert(dctx->streamStage == zdss_init);
+    dctx->format = ZSTD_f_zstd1;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+    dctx->outBufferMode = ZSTD_bm_buffered;
+    dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
+    dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
+    dctx->disableHufAsm = 0;
+}
+
+static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
+{
+    dctx->staticSize  = 0;
+    dctx->ddict       = NULL;
+    dctx->ddictLocal  = NULL;
+    dctx->dictEnd     = NULL;
+    dctx->ddictIsCold = 0;
+    dctx->dictUses = ZSTD_dont_use;
+    dctx->inBuff      = NULL;
+    dctx->inBuffSize  = 0;
+    dctx->outBuffSize = 0;
+    dctx->streamStage = zdss_init;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+    dctx->legacyContext = NULL;
+    dctx->previousLegacyVersion = 0;
+#endif
+    dctx->noForwardProgress = 0;
+    dctx->oversizedDuration = 0;
+#if DYNAMIC_BMI2
+    dctx->bmi2 = ZSTD_cpuSupportsBmi2();
+#endif
+    dctx->ddictSet = NULL;
+    ZSTD_DCtx_resetParameters(dctx);
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentEndForFuzzing = NULL;
+#endif
+}
+
+ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
+{
+    ZSTD_DCtx* const dctx = (ZSTD_DCtx*) workspace;
+
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    if (workspaceSize < sizeof(ZSTD_DCtx)) return NULL;  /* minimum size */
+
+    ZSTD_initDCtx_internal(dctx);
+    dctx->staticSize = workspaceSize;
+    dctx->inBuff = (char*)(dctx+1);
+    return dctx;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx_internal(ZSTD_customMem customMem) {
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+
+    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_customMalloc(sizeof(*dctx), customMem);
+        if (!dctx) return NULL;
+        dctx->customMem = customMem;
+        ZSTD_initDCtx_internal(dctx);
+        return dctx;
+    }
+}
+
+ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDCtx_internal(customMem);
+}
+
+ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    DEBUGLOG(3, "ZSTD_createDCtx");
+    return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
+}
+
+static void ZSTD_clearDict(ZSTD_DCtx* dctx)
+{
+    ZSTD_freeDDict(dctx->ddictLocal);
+    dctx->ddictLocal = NULL;
+    dctx->ddict = NULL;
+    dctx->dictUses = ZSTD_dont_use;
+}
+
+size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support free on NULL */
+    RETURN_ERROR_IF(dctx->staticSize, memory_allocation, "not compatible with static DCtx");
+    {   ZSTD_customMem const cMem = dctx->customMem;
+        ZSTD_clearDict(dctx);
+        ZSTD_customFree(dctx->inBuff, cMem);
+        dctx->inBuff = NULL;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (dctx->legacyContext)
+            ZSTD_freeLegacyStreamContext(dctx->legacyContext, dctx->previousLegacyVersion);
+#endif
+        if (dctx->ddictSet) {
+            ZSTD_freeDDictHashSet(dctx->ddictSet, cMem);
+            dctx->ddictSet = NULL;
+        }
+        ZSTD_customFree(dctx, cMem);
+        return 0;
+    }
+}
+
+/* no longer useful */
+void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
+{
+    size_t const toCopy = (size_t)((char*)(&dstDCtx->inBuff) - (char*)dstDCtx);
+    ZSTD_memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
+}
+
+/* Given a dctx with a digested frame params, re-selects the correct ZSTD_DDict based on
+ * the requested dict ID from the frame. If there exists a reference to the correct ZSTD_DDict, then
+ * accordingly sets the ddict to be used to decompress the frame.
+ *
+ * If no DDict is found, then no action is taken, and the ZSTD_DCtx::ddict remains as-is.
+ *
+ * ZSTD_d_refMultipleDDicts must be enabled for this function to be called.
+ */
+static void ZSTD_DCtx_selectFrameDDict(ZSTD_DCtx* dctx) {
+    assert(dctx->refMultipleDDicts && dctx->ddictSet);
+    DEBUGLOG(4, "Adjusting DDict based on requested dict ID from frame");
+    if (dctx->ddict) {
+        const ZSTD_DDict* frameDDict = ZSTD_DDictHashSet_getDDict(dctx->ddictSet, dctx->fParams.dictID);
+        if (frameDDict) {
+            DEBUGLOG(4, "DDict found!");
+            ZSTD_clearDict(dctx);
+            dctx->dictID = dctx->fParams.dictID;
+            dctx->ddict = frameDDict;
+            dctx->dictUses = ZSTD_use_indefinitely;
+        }
+    }
+}
+
+
+/*-*************************************************************
+ *   Frame header decoding
+ ***************************************************************/
+
+/*! ZSTD_isFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
+ *  Note 3 : Skippable Frame Identifiers are considered valid. */
+unsigned ZSTD_isFrame(const void* buffer, size_t size)
+{
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
+    {   U32 const magic = MEM_readLE32(buffer);
+        if (magic == ZSTD_MAGICNUMBER) return 1;
+        if ((magic & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+    }
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(buffer, size)) return 1;
+#endif
+    return 0;
+}
+
+/*! ZSTD_isSkippableFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ */
+unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size)
+{
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
+    {   U32 const magic = MEM_readLE32(buffer);
+        if ((magic & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+    }
+    return 0;
+}
+
+/** ZSTD_frameHeaderSize_internal() :
+ *  srcSize must be large enough to reach header size fields.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless.
+ * @return : size of the Frame Header
+ *           or an error code, which can be tested with ZSTD_isError() */
+static size_t ZSTD_frameHeaderSize_internal(const void* src, size_t srcSize, ZSTD_format_e format)
+{
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+    RETURN_ERROR_IF(srcSize < minInputSize, srcSize_wrong, "");
+
+    {   BYTE const fhd = ((const BYTE*)src)[minInputSize-1];
+        U32 const dictID= fhd & 3;
+        U32 const singleSegment = (fhd >> 5) & 1;
+        U32 const fcsId = fhd >> 6;
+        return minInputSize + !singleSegment
+             + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
+             + (singleSegment && !fcsId);
+    }
+}
+
+/** ZSTD_frameHeaderSize() :
+ *  srcSize must be >= ZSTD_frameHeaderSize_prefix.
+ * @return : size of the Frame Header,
+ *           or an error code (if srcSize is too small) */
+size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
+{
+    return ZSTD_frameHeaderSize_internal(src, srcSize, ZSTD_f_zstd1);
+}
+
+
+/** ZSTD_getFrameHeader_advanced() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+**           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+
+    DEBUGLOG(5, "ZSTD_getFrameHeader_advanced: minInputSize = %zu, srcSize = %zu", minInputSize, srcSize);
+
+    if (srcSize > 0) {
+        /* note : technically could be considered an assert(), since it's an invalid entry */
+        RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter : src==NULL, but srcSize>0");
+    }
+    if (srcSize < minInputSize) {
+        if (srcSize > 0 && format != ZSTD_f_zstd1_magicless) {
+            /* when receiving less than @minInputSize bytes,
+             * control these bytes at least correspond to a supported magic number
+             * in order to error out early if they don't.
+            **/
+            size_t const toCopy = MIN(4, srcSize);
+            unsigned char hbuf[4]; MEM_writeLE32(hbuf, ZSTD_MAGICNUMBER);
+            assert(src != NULL);
+            ZSTD_memcpy(hbuf, src, toCopy);
+            if ( MEM_readLE32(hbuf) != ZSTD_MAGICNUMBER ) {
+                /* not a zstd frame : let's check if it's a skippable frame */
+                MEM_writeLE32(hbuf, ZSTD_MAGIC_SKIPPABLE_START);
+                ZSTD_memcpy(hbuf, src, toCopy);
+                if ((MEM_readLE32(hbuf) & ZSTD_MAGIC_SKIPPABLE_MASK) != ZSTD_MAGIC_SKIPPABLE_START) {
+                    RETURN_ERROR(prefix_unknown,
+                                "first bytes don't correspond to any supported magic number");
+        }   }   }
+        return minInputSize;
+    }
+
+    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzers may not understand that zfhPtr will be read only if return value is zero, since they are 2 different signals */
+    if ( (format != ZSTD_f_zstd1_magicless)
+      && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
+        if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+            /* skippable frame */
+            if (srcSize < ZSTD_SKIPPABLEHEADERSIZE)
+                return ZSTD_SKIPPABLEHEADERSIZE; /* magic number + frame length */
+            ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
+            zfhPtr->frameType = ZSTD_skippableFrame;
+            return 0;
+        }
+        RETURN_ERROR(prefix_unknown, "");
+    }
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    {   size_t const fhsize = ZSTD_frameHeaderSize_internal(src, srcSize, format);
+        if (srcSize < fhsize) return fhsize;
+        zfhPtr->headerSize = (U32)fhsize;
+    }
+
+    {   BYTE const fhdByte = ip[minInputSize-1];
+        size_t pos = minInputSize;
+        U32 const dictIDSizeCode = fhdByte&3;
+        U32 const checksumFlag = (fhdByte>>2)&1;
+        U32 const singleSegment = (fhdByte>>5)&1;
+        U32 const fcsID = fhdByte>>6;
+        U64 windowSize = 0;
+        U32 dictID = 0;
+        U64 frameContentSize = ZSTD_CONTENTSIZE_UNKNOWN;
+        RETURN_ERROR_IF((fhdByte & 0x08) != 0, frameParameter_unsupported,
+                        "reserved bits, must be zero");
+
+        if (!singleSegment) {
+            BYTE const wlByte = ip[pos++];
+            U32 const windowLog = (wlByte >> 3) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
+            RETURN_ERROR_IF(windowLog > ZSTD_WINDOWLOG_MAX, frameParameter_windowTooLarge, "");
+            windowSize = (1ULL << windowLog);
+            windowSize += (windowSize >> 3) * (wlByte&7);
+        }
+        switch(dictIDSizeCode)
+        {
+            default:
+                assert(0);  /* impossible */
+                ZSTD_FALLTHROUGH;
+            case 0 : break;
+            case 1 : dictID = ip[pos]; pos++; break;
+            case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
+            case 3 : dictID = MEM_readLE32(ip+pos); pos+=4; break;
+        }
+        switch(fcsID)
+        {
+            default:
+                assert(0);  /* impossible */
+                ZSTD_FALLTHROUGH;
+            case 0 : if (singleSegment) frameContentSize = ip[pos]; break;
+            case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
+            case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
+            case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
+        }
+        if (singleSegment) windowSize = frameContentSize;
+
+        zfhPtr->frameType = ZSTD_frame;
+        zfhPtr->frameContentSize = frameContentSize;
+        zfhPtr->windowSize = windowSize;
+        zfhPtr->blockSizeMax = (unsigned) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+        zfhPtr->dictID = dictID;
+        zfhPtr->checksumFlag = checksumFlag;
+    }
+    return 0;
+}
+
+/** ZSTD_getFrameHeader() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : this function does not consume input, it only reads it.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize)
+{
+    return ZSTD_getFrameHeader_advanced(zfhPtr, src, srcSize, ZSTD_f_zstd1);
+}
+
+/** ZSTD_getFrameContentSize() :
+ *  compatible with legacy mode
+ * @return : decompressed size of the single frame pointed to be `src` if known, otherwise
+ *         - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *         - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small) */
+unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize)) {
+        unsigned long long const ret = ZSTD_getDecompressedSize_legacy(src, srcSize);
+        return ret == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : ret;
+    }
+#endif
+    {   ZSTD_frameHeader zfh;
+        if (ZSTD_getFrameHeader(&zfh, src, srcSize) != 0)
+            return ZSTD_CONTENTSIZE_ERROR;
+        if (zfh.frameType == ZSTD_skippableFrame) {
+            return 0;
+        } else {
+            return zfh.frameContentSize;
+    }   }
+}
+
+static size_t readSkippableFrameSize(void const* src, size_t srcSize)
+{
+    size_t const skippableHeaderSize = ZSTD_SKIPPABLEHEADERSIZE;
+    U32 sizeU32;
+
+    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong, "");
+
+    sizeU32 = MEM_readLE32((BYTE const*)src + ZSTD_FRAMEIDSIZE);
+    RETURN_ERROR_IF((U32)(sizeU32 + ZSTD_SKIPPABLEHEADERSIZE) < sizeU32,
+                    frameParameter_unsupported, "");
+    {   size_t const skippableSize = skippableHeaderSize + sizeU32;
+        RETURN_ERROR_IF(skippableSize > srcSize, srcSize_wrong, "");
+        return skippableSize;
+    }
+}
+
+/*! ZSTD_readSkippableFrame() :
+ * Retrieves content of a skippable frame, and writes it to dst buffer.
+ *
+ * The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
+ * i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START.  This can be NULL if the caller is not interested
+ * in the magicVariant.
+ *
+ * Returns an error if destination buffer is not large enough, or if this is not a valid skippable frame.
+ *
+ * @return : number of bytes written or a ZSTD error.
+ */
+size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity,
+                               unsigned* magicVariant,  /* optional, can be NULL */
+                         const void* src, size_t srcSize)
+{
+    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong, "");
+
+    {   U32 const magicNumber = MEM_readLE32(src);
+        size_t skippableFrameSize = readSkippableFrameSize(src, srcSize);
+        size_t skippableContentSize = skippableFrameSize - ZSTD_SKIPPABLEHEADERSIZE;
+
+        /* check input validity */
+        RETURN_ERROR_IF(!ZSTD_isSkippableFrame(src, srcSize), frameParameter_unsupported, "");
+        RETURN_ERROR_IF(skippableFrameSize < ZSTD_SKIPPABLEHEADERSIZE || skippableFrameSize > srcSize, srcSize_wrong, "");
+        RETURN_ERROR_IF(skippableContentSize > dstCapacity, dstSize_tooSmall, "");
+
+        /* deliver payload */
+        if (skippableContentSize > 0  && dst != NULL)
+            ZSTD_memcpy(dst, (const BYTE *)src + ZSTD_SKIPPABLEHEADERSIZE, skippableContentSize);
+        if (magicVariant != NULL)
+            *magicVariant = magicNumber - ZSTD_MAGIC_SKIPPABLE_START;
+        return skippableContentSize;
+    }
+}
+
+/** ZSTD_findDecompressedSize() :
+ *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
+ *      skippable frames
+ *  note: compatible with legacy mode
+ * @return : decompressed size of the frames contained */
+unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
+{
+    unsigned long long totalDstSize = 0;
+
+    while (srcSize >= ZSTD_startingInputLength(ZSTD_f_zstd1)) {
+        U32 const magicNumber = MEM_readLE32(src);
+
+        if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+            size_t const skippableSize = readSkippableFrameSize(src, srcSize);
+            if (ZSTD_isError(skippableSize)) return ZSTD_CONTENTSIZE_ERROR;
+            assert(skippableSize <= srcSize);
+
+            src = (const BYTE *)src + skippableSize;
+            srcSize -= skippableSize;
+            continue;
+        }
+
+        {   unsigned long long const fcs = ZSTD_getFrameContentSize(src, srcSize);
+            if (fcs >= ZSTD_CONTENTSIZE_ERROR) return fcs;
+
+            if (totalDstSize + fcs < totalDstSize)
+                return ZSTD_CONTENTSIZE_ERROR; /* check for overflow */
+            totalDstSize += fcs;
+        }
+        /* skip to next frame */
+        {   size_t const frameSrcSize = ZSTD_findFrameCompressedSize(src, srcSize);
+            if (ZSTD_isError(frameSrcSize)) return ZSTD_CONTENTSIZE_ERROR;
+            assert(frameSrcSize <= srcSize);
+
+            src = (const BYTE *)src + frameSrcSize;
+            srcSize -= frameSrcSize;
+        }
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    if (srcSize) return ZSTD_CONTENTSIZE_ERROR;
+
+    return totalDstSize;
+}
+
+/** ZSTD_getDecompressedSize() :
+ *  compatible with legacy mode
+ * @return : decompressed size if known, 0 otherwise
+             note : 0 can mean any of the following :
+                   - frame content is empty
+                   - decompressed size field is not present in frame header
+                   - frame header unknown / not supported
+                   - frame header not complete (`srcSize` too small) */
+unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
+{
+    unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_ERROR < ZSTD_CONTENTSIZE_UNKNOWN);
+    return (ret >= ZSTD_CONTENTSIZE_ERROR) ? 0 : ret;
+}
+
+
+/** ZSTD_decodeFrameHeader() :
+ * `headerSize` must be the size provided by ZSTD_frameHeaderSize().
+ * If multiple DDict references are enabled, also will choose the correct DDict to use.
+ * @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
+{
+    size_t const result = ZSTD_getFrameHeader_advanced(&(dctx->fParams), src, headerSize, dctx->format);
+    if (ZSTD_isError(result)) return result;    /* invalid header */
+    RETURN_ERROR_IF(result>0, srcSize_wrong, "headerSize too small");
+
+    /* Reference DDict requested by frame if dctx references multiple ddicts */
+    if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts && dctx->ddictSet) {
+        ZSTD_DCtx_selectFrameDDict(dctx);
+    }
+
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    /* Skip the dictID check in fuzzing mode, because it makes the search
+     * harder.
+     */
+    RETURN_ERROR_IF(dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID),
+                    dictionary_wrong, "");
+#endif
+    dctx->validateChecksum = (dctx->fParams.checksumFlag && !dctx->forceIgnoreChecksum) ? 1 : 0;
+    if (dctx->validateChecksum) XXH64_reset(&dctx->xxhState, 0);
+    dctx->processedCSize += headerSize;
+    return 0;
+}
+
+static ZSTD_frameSizeInfo ZSTD_errorFrameSizeInfo(size_t ret)
+{
+    ZSTD_frameSizeInfo frameSizeInfo;
+    frameSizeInfo.compressedSize = ret;
+    frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
+    return frameSizeInfo;
+}
+
+static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize)
+{
+    ZSTD_frameSizeInfo frameSizeInfo;
+    ZSTD_memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize))
+        return ZSTD_findFrameSizeInfoLegacy(src, srcSize);
+#endif
+
+    if ((srcSize >= ZSTD_SKIPPABLEHEADERSIZE)
+        && (MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+        frameSizeInfo.compressedSize = readSkippableFrameSize(src, srcSize);
+        assert(ZSTD_isError(frameSizeInfo.compressedSize) ||
+               frameSizeInfo.compressedSize <= srcSize);
+        return frameSizeInfo;
+    } else {
+        const BYTE* ip = (const BYTE*)src;
+        const BYTE* const ipstart = ip;
+        size_t remainingSize = srcSize;
+        size_t nbBlocks = 0;
+        ZSTD_frameHeader zfh;
+
+        /* Extract Frame Header */
+        {   size_t const ret = ZSTD_getFrameHeader(&zfh, src, srcSize);
+            if (ZSTD_isError(ret))
+                return ZSTD_errorFrameSizeInfo(ret);
+            if (ret > 0)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
+        }
+
+        ip += zfh.headerSize;
+        remainingSize -= zfh.headerSize;
+
+        /* Iterate over each block */
+        while (1) {
+            blockProperties_t blockProperties;
+            size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+            if (ZSTD_isError(cBlockSize))
+                return ZSTD_errorFrameSizeInfo(cBlockSize);
+
+            if (ZSTD_blockHeaderSize + cBlockSize > remainingSize)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
+
+            ip += ZSTD_blockHeaderSize + cBlockSize;
+            remainingSize -= ZSTD_blockHeaderSize + cBlockSize;
+            nbBlocks++;
+
+            if (blockProperties.lastBlock) break;
+        }
+
+        /* Final frame content checksum */
+        if (zfh.checksumFlag) {
+            if (remainingSize < 4)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
+            ip += 4;
+        }
+
+        frameSizeInfo.nbBlocks = nbBlocks;
+        frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
+        frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
+                                        ? zfh.frameContentSize
+                                        : (unsigned long long)nbBlocks * zfh.blockSizeMax;
+        return frameSizeInfo;
+    }
+}
+
+/** ZSTD_findFrameCompressedSize() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame, ZSTD legacy frame, or skippable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the compressed size of the frame starting at `src` */
+size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
+{
+    ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
+    return frameSizeInfo.compressedSize;
+}
+
+/** ZSTD_decompressBound() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame or a skippeable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the maximum decompressed size of the compressed source
+ */
+unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
+{
+    unsigned long long bound = 0;
+    /* Iterate over each frame */
+    while (srcSize > 0) {
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
+        size_t const compressedSize = frameSizeInfo.compressedSize;
+        unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
+        if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
+            return ZSTD_CONTENTSIZE_ERROR;
+        assert(srcSize >= compressedSize);
+        src = (const BYTE*)src + compressedSize;
+        srcSize -= compressedSize;
+        bound += decompressedBound;
+    }
+    return bound;
+}
+
+size_t ZSTD_decompressionMargin(void const* src, size_t srcSize)
+{
+    size_t margin = 0;
+    unsigned maxBlockSize = 0;
+
+    /* Iterate over each frame */
+    while (srcSize > 0) {
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
+        size_t const compressedSize = frameSizeInfo.compressedSize;
+        unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
+        ZSTD_frameHeader zfh;
+
+        FORWARD_IF_ERROR(ZSTD_getFrameHeader(&zfh, src, srcSize), "");
+        if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
+            return ERROR(corruption_detected);
+
+        if (zfh.frameType == ZSTD_frame) {
+            /* Add the frame header to our margin */
+            margin += zfh.headerSize;
+            /* Add the checksum to our margin */
+            margin += zfh.checksumFlag ? 4 : 0;
+            /* Add 3 bytes per block */
+            margin += 3 * frameSizeInfo.nbBlocks;
+
+            /* Compute the max block size */
+            maxBlockSize = MAX(maxBlockSize, zfh.blockSizeMax);
+        } else {
+            assert(zfh.frameType == ZSTD_skippableFrame);
+            /* Add the entire skippable frame size to our margin. */
+            margin += compressedSize;
+        }
+
+        assert(srcSize >= compressedSize);
+        src = (const BYTE*)src + compressedSize;
+        srcSize -= compressedSize;
+    }
+
+    /* Add the max block size back to the margin. */
+    margin += maxBlockSize;
+
+    return margin;
+}
+
+/*-*************************************************************
+ *   Frame decoding
+ ***************************************************************/
+
+/** ZSTD_insertBlock() :
+ *  insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
+{
+    DEBUGLOG(5, "ZSTD_insertBlock: %u bytes", (unsigned)blockSize);
+    ZSTD_checkContinuity(dctx, blockStart, blockSize);
+    dctx->previousDstEnd = (const char*)blockStart + blockSize;
+    return blockSize;
+}
+
+
+static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
+                          const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_copyRawBlock");
+    RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
+    if (dst == NULL) {
+        if (srcSize == 0) return 0;
+        RETURN_ERROR(dstBuffer_null, "");
+    }
+    ZSTD_memmove(dst, src, srcSize);
+    return srcSize;
+}
+
+static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
+                               BYTE b,
+                               size_t regenSize)
+{
+    RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
+    if (dst == NULL) {
+        if (regenSize == 0) return 0;
+        RETURN_ERROR(dstBuffer_null, "");
+    }
+    ZSTD_memset(dst, b, regenSize);
+    return regenSize;
+}
+
+static void ZSTD_DCtx_trace_end(ZSTD_DCtx const* dctx, U64 uncompressedSize, U64 compressedSize, unsigned streaming)
+{
+#if ZSTD_TRACE
+    if (dctx->traceCtx && ZSTD_trace_decompress_end != NULL) {
+        ZSTD_Trace trace;
+        ZSTD_memset(&trace, 0, sizeof(trace));
+        trace.version = ZSTD_VERSION_NUMBER;
+        trace.streaming = streaming;
+        if (dctx->ddict) {
+            trace.dictionaryID = ZSTD_getDictID_fromDDict(dctx->ddict);
+            trace.dictionarySize = ZSTD_DDict_dictSize(dctx->ddict);
+            trace.dictionaryIsCold = dctx->ddictIsCold;
+        }
+        trace.uncompressedSize = (size_t)uncompressedSize;
+        trace.compressedSize = (size_t)compressedSize;
+        trace.dctx = dctx;
+        ZSTD_trace_decompress_end(dctx->traceCtx, &trace);
+    }
+#else
+    (void)dctx;
+    (void)uncompressedSize;
+    (void)compressedSize;
+    (void)streaming;
+#endif
+}
+
+
+/*! ZSTD_decompressFrame() :
+ * @dctx must be properly initialized
+ *  will update *srcPtr and *srcSizePtr,
+ *  to make *srcPtr progress by one frame. */
+static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
+                                   void* dst, size_t dstCapacity,
+                             const void** srcPtr, size_t *srcSizePtr)
+{
+    const BYTE* const istart = (const BYTE*)(*srcPtr);
+    const BYTE* ip = istart;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = dstCapacity != 0 ? ostart + dstCapacity : ostart;
+    BYTE* op = ostart;
+    size_t remainingSrcSize = *srcSizePtr;
+
+    DEBUGLOG(4, "ZSTD_decompressFrame (srcSize:%i)", (int)*srcSizePtr);
+
+    /* check */
+    RETURN_ERROR_IF(
+        remainingSrcSize < ZSTD_FRAMEHEADERSIZE_MIN(dctx->format)+ZSTD_blockHeaderSize,
+        srcSize_wrong, "");
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize_internal(
+                ip, ZSTD_FRAMEHEADERSIZE_PREFIX(dctx->format), dctx->format);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+        RETURN_ERROR_IF(remainingSrcSize < frameHeaderSize+ZSTD_blockHeaderSize,
+                        srcSize_wrong, "");
+        FORWARD_IF_ERROR( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) , "");
+        ip += frameHeaderSize; remainingSrcSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        BYTE* oBlockEnd = oend;
+        size_t decodedSize;
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSrcSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSrcSize -= ZSTD_blockHeaderSize;
+        RETURN_ERROR_IF(cBlockSize > remainingSrcSize, srcSize_wrong, "");
+
+        if (ip >= op && ip < oBlockEnd) {
+            /* We are decompressing in-place. Limit the output pointer so that we
+             * don't overwrite the block that we are currently reading. This will
+             * fail decompression if the input & output pointers aren't spaced
+             * far enough apart.
+             *
+             * This is important to set, even when the pointers are far enough
+             * apart, because ZSTD_decompressBlock_internal() can decide to store
+             * literals in the output buffer, after the block it is decompressing.
+             * Since we don't want anything to overwrite our input, we have to tell
+             * ZSTD_decompressBlock_internal to never write past ip.
+             *
+             * See ZSTD_allocateLiteralsBuffer() for reference.
+             */
+            oBlockEnd = op + (ip - op);
+        }
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oBlockEnd-op), ip, cBlockSize, /* frame */ 1, not_streaming);
+            break;
+        case bt_raw :
+            /* Use oend instead of oBlockEnd because this function is safe to overlap. It uses memmove. */
+            decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
+            break;
+        case bt_rle :
+            decodedSize = ZSTD_setRleBlock(op, (size_t)(oBlockEnd-op), *ip, blockProperties.origSize);
+            break;
+        case bt_reserved :
+        default:
+            RETURN_ERROR(corruption_detected, "invalid block type");
+        }
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        if (dctx->validateChecksum)
+            XXH64_update(&dctx->xxhState, op, decodedSize);
+        if (decodedSize != 0)
+            op += decodedSize;
+        assert(ip != NULL);
+        ip += cBlockSize;
+        remainingSrcSize -= cBlockSize;
+        if (blockProperties.lastBlock) break;
+    }
+
+    if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        RETURN_ERROR_IF((U64)(op-ostart) != dctx->fParams.frameContentSize,
+                        corruption_detected, "");
+    }
+    if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
+        RETURN_ERROR_IF(remainingSrcSize<4, checksum_wrong, "");
+        if (!dctx->forceIgnoreChecksum) {
+            U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
+            U32 checkRead;
+            checkRead = MEM_readLE32(ip);
+            RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
+        }
+        ip += 4;
+        remainingSrcSize -= 4;
+    }
+    ZSTD_DCtx_trace_end(dctx, (U64)(op-ostart), (U64)(ip-istart), /* streaming */ 0);
+    /* Allow caller to get size read */
+    DEBUGLOG(4, "ZSTD_decompressFrame: decompressed frame of size %zi, consuming %zi bytes of input", op-ostart, ip - (const BYTE*)*srcPtr);
+    *srcPtr = ip;
+    *srcSizePtr = remainingSrcSize;
+    return (size_t)(op-ostart);
+}
+
+static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize,
+                                  const void* dict, size_t dictSize,
+                                  const ZSTD_DDict* ddict)
+{
+    void* const dststart = dst;
+    int moreThan1Frame = 0;
+
+    DEBUGLOG(5, "ZSTD_decompressMultiFrame");
+    assert(dict==NULL || ddict==NULL);  /* either dict or ddict set, not both */
+
+    if (ddict) {
+        dict = ZSTD_DDict_dictContent(ddict);
+        dictSize = ZSTD_DDict_dictSize(ddict);
+    }
+
+    while (srcSize >= ZSTD_startingInputLength(dctx->format)) {
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (ZSTD_isLegacy(src, srcSize)) {
+            size_t decodedSize;
+            size_t const frameSize = ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+            if (ZSTD_isError(frameSize)) return frameSize;
+            RETURN_ERROR_IF(dctx->staticSize, memory_allocation,
+                "legacy support is not compatible with static dctx");
+
+            decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
+            if (ZSTD_isError(decodedSize)) return decodedSize;
+
+            assert(decodedSize <= dstCapacity);
+            dst = (BYTE*)dst + decodedSize;
+            dstCapacity -= decodedSize;
+
+            src = (const BYTE*)src + frameSize;
+            srcSize -= frameSize;
+
+            continue;
+        }
+#endif
+
+        if (srcSize >= 4) {
+            U32 const magicNumber = MEM_readLE32(src);
+            DEBUGLOG(5, "reading magic number %08X", (unsigned)magicNumber);
+            if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+                /* skippable frame detected : skip it */
+                size_t const skippableSize = readSkippableFrameSize(src, srcSize);
+                FORWARD_IF_ERROR(skippableSize, "invalid skippable frame");
+                assert(skippableSize <= srcSize);
+
+                src = (const BYTE *)src + skippableSize;
+                srcSize -= skippableSize;
+                continue; /* check next frame */
+        }   }
+
+        if (ddict) {
+            /* we were called from ZSTD_decompress_usingDDict */
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(dctx, ddict), "");
+        } else {
+            /* this will initialize correctly with no dict if dict == NULL, so
+             * use this in all cases but ddict */
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize), "");
+        }
+        ZSTD_checkContinuity(dctx, dst, dstCapacity);
+
+        {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
+                                                    &src, &srcSize);
+            RETURN_ERROR_IF(
+                (ZSTD_getErrorCode(res) == ZSTD_error_prefix_unknown)
+             && (moreThan1Frame==1),
+                srcSize_wrong,
+                "At least one frame successfully completed, "
+                "but following bytes are garbage: "
+                "it's more likely to be a srcSize error, "
+                "specifying more input bytes than size of frame(s). "
+                "Note: one could be unlucky, it might be a corruption error instead, "
+                "happening right at the place where we expect zstd magic bytes. "
+                "But this is _much_ less likely than a srcSize field error.");
+            if (ZSTD_isError(res)) return res;
+            assert(res <= dstCapacity);
+            if (res != 0)
+                dst = (BYTE*)dst + res;
+            dstCapacity -= res;
+        }
+        moreThan1Frame = 1;
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    RETURN_ERROR_IF(srcSize, srcSize_wrong, "input not entirely consumed");
+
+    return (size_t)((BYTE*)dst - (BYTE*)dststart);
+}
+
+size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                           const void* src, size_t srcSize,
+                           const void* dict, size_t dictSize)
+{
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize, dict, dictSize, NULL);
+}
+
+
+static ZSTD_DDict const* ZSTD_getDDict(ZSTD_DCtx* dctx)
+{
+    switch (dctx->dictUses) {
+    default:
+        assert(0 /* Impossible */);
+        ZSTD_FALLTHROUGH;
+    case ZSTD_dont_use:
+        ZSTD_clearDict(dctx);
+        return NULL;
+    case ZSTD_use_indefinitely:
+        return dctx->ddict;
+    case ZSTD_use_once:
+        dctx->dictUses = ZSTD_dont_use;
+        return dctx->ddict;
+    }
+}
+
+size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTD_decompress_usingDDict(dctx, dst, dstCapacity, src, srcSize, ZSTD_getDDict(dctx));
+}
+
+
+size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
+    size_t regenSize;
+    ZSTD_DCtx* const dctx =  ZSTD_createDCtx_internal(ZSTD_defaultCMem);
+    RETURN_ERROR_IF(dctx==NULL, memory_allocation, "NULL pointer!");
+    regenSize = ZSTD_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTD_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTD_DCtx dctx;
+    ZSTD_initDCtx_internal(&dctx);
+    return ZSTD_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+
+/*-**************************************
+*   Advanced Streaming Decompression API
+*   Bufferless and synchronous
+****************************************/
+size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
+
+/**
+ * Similar to ZSTD_nextSrcSizeToDecompress(), but when a block input can be streamed, we
+ * allow taking a partial block as the input. Currently only raw uncompressed blocks can
+ * be streamed.
+ *
+ * For blocks that can be streamed, this allows us to reduce the latency until we produce
+ * output, and avoid copying the input.
+ *
+ * @param inputSize - The total amount of input that the caller currently has.
+ */
+static size_t ZSTD_nextSrcSizeToDecompressWithInputSize(ZSTD_DCtx* dctx, size_t inputSize) {
+    if (!(dctx->stage == ZSTDds_decompressBlock || dctx->stage == ZSTDds_decompressLastBlock))
+        return dctx->expected;
+    if (dctx->bType != bt_raw)
+        return dctx->expected;
+    return BOUNDED(1, inputSize, dctx->expected);
+}
+
+ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
+    switch(dctx->stage)
+    {
+    default:   /* should not happen */
+        assert(0);
+        ZSTD_FALLTHROUGH;
+    case ZSTDds_getFrameHeaderSize:
+        ZSTD_FALLTHROUGH;
+    case ZSTDds_decodeFrameHeader:
+        return ZSTDnit_frameHeader;
+    case ZSTDds_decodeBlockHeader:
+        return ZSTDnit_blockHeader;
+    case ZSTDds_decompressBlock:
+        return ZSTDnit_block;
+    case ZSTDds_decompressLastBlock:
+        return ZSTDnit_lastBlock;
+    case ZSTDds_checkChecksum:
+        return ZSTDnit_checksum;
+    case ZSTDds_decodeSkippableHeader:
+        ZSTD_FALLTHROUGH;
+    case ZSTDds_skipFrame:
+        return ZSTDnit_skippableFrame;
+    }
+}
+
+static int ZSTD_isSkipFrame(ZSTD_DCtx* dctx) { return dctx->stage == ZSTDds_skipFrame; }
+
+/** ZSTD_decompressContinue() :
+ *  srcSize : must be the exact nb of bytes expected (see ZSTD_nextSrcSizeToDecompress())
+ *  @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
+ *            or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_decompressContinue (srcSize:%u)", (unsigned)srcSize);
+    /* Sanity check */
+    RETURN_ERROR_IF(srcSize != ZSTD_nextSrcSizeToDecompressWithInputSize(dctx, srcSize), srcSize_wrong, "not allowed");
+    ZSTD_checkContinuity(dctx, dst, dstCapacity);
+
+    dctx->processedCSize += srcSize;
+
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        assert(src != NULL);
+        if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
+            assert(srcSize >= ZSTD_FRAMEIDSIZE);  /* to read skippable magic number */
+            if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
+                ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
+                dctx->expected = ZSTD_SKIPPABLEHEADERSIZE - srcSize;  /* remaining to load to get full skippable frame header */
+                dctx->stage = ZSTDds_decodeSkippableHeader;
+                return 0;
+        }   }
+        dctx->headerSize = ZSTD_frameHeaderSize_internal(src, srcSize, dctx->format);
+        if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
+        ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
+        dctx->expected = dctx->headerSize - srcSize;
+        dctx->stage = ZSTDds_decodeFrameHeader;
+        return 0;
+
+    case ZSTDds_decodeFrameHeader:
+        assert(src != NULL);
+        ZSTD_memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
+        FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize), "");
+        dctx->expected = ZSTD_blockHeaderSize;
+        dctx->stage = ZSTDds_decodeBlockHeader;
+        return 0;
+
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+            if (ZSTD_isError(cBlockSize)) return cBlockSize;
+            RETURN_ERROR_IF(cBlockSize > dctx->fParams.blockSizeMax, corruption_detected, "Block Size Exceeds Maximum");
+            dctx->expected = cBlockSize;
+            dctx->bType = bp.blockType;
+            dctx->rleSize = bp.origSize;
+            if (cBlockSize) {
+                dctx->stage = bp.lastBlock ? ZSTDds_decompressLastBlock : ZSTDds_decompressBlock;
+                return 0;
+            }
+            /* empty block */
+            if (bp.lastBlock) {
+                if (dctx->fParams.checksumFlag) {
+                    dctx->expected = 4;
+                    dctx->stage = ZSTDds_checkChecksum;
+                } else {
+                    dctx->expected = 0; /* end of frame */
+                    dctx->stage = ZSTDds_getFrameHeaderSize;
+                }
+            } else {
+                dctx->expected = ZSTD_blockHeaderSize;  /* jump to next header */
+                dctx->stage = ZSTDds_decodeBlockHeader;
+            }
+            return 0;
+        }
+
+    case ZSTDds_decompressLastBlock:
+    case ZSTDds_decompressBlock:
+        DEBUGLOG(5, "ZSTD_decompressContinue: case ZSTDds_decompressBlock");
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
+                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1, is_streaming);
+                dctx->expected = 0;  /* Streaming not supported */
+                break;
+            case bt_raw :
+                assert(srcSize <= dctx->expected);
+                rSize = ZSTD_copyRawBlock(dst, dstCapacity, src, srcSize);
+                FORWARD_IF_ERROR(rSize, "ZSTD_copyRawBlock failed");
+                assert(rSize == srcSize);
+                dctx->expected -= rSize;
+                break;
+            case bt_rle :
+                rSize = ZSTD_setRleBlock(dst, dstCapacity, *(const BYTE*)src, dctx->rleSize);
+                dctx->expected = 0;  /* Streaming not supported */
+                break;
+            case bt_reserved :   /* should never happen */
+            default:
+                RETURN_ERROR(corruption_detected, "invalid block type");
+            }
+            FORWARD_IF_ERROR(rSize, "");
+            RETURN_ERROR_IF(rSize > dctx->fParams.blockSizeMax, corruption_detected, "Decompressed Block Size Exceeds Maximum");
+            DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (unsigned)rSize);
+            dctx->decodedSize += rSize;
+            if (dctx->validateChecksum) XXH64_update(&dctx->xxhState, dst, rSize);
+            dctx->previousDstEnd = (char*)dst + rSize;
+
+            /* Stay on the same stage until we are finished streaming the block. */
+            if (dctx->expected > 0) {
+                return rSize;
+            }
+
+            if (dctx->stage == ZSTDds_decompressLastBlock) {   /* end of frame */
+                DEBUGLOG(4, "ZSTD_decompressContinue: decoded size from frame : %u", (unsigned)dctx->decodedSize);
+                RETURN_ERROR_IF(
+                    dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                 && dctx->decodedSize != dctx->fParams.frameContentSize,
+                    corruption_detected, "");
+                if (dctx->fParams.checksumFlag) {  /* another round for frame checksum */
+                    dctx->expected = 4;
+                    dctx->stage = ZSTDds_checkChecksum;
+                } else {
+                    ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
+                    dctx->expected = 0;   /* ends here */
+                    dctx->stage = ZSTDds_getFrameHeaderSize;
+                }
+            } else {
+                dctx->stage = ZSTDds_decodeBlockHeader;
+                dctx->expected = ZSTD_blockHeaderSize;
+            }
+            return rSize;
+        }
+
+    case ZSTDds_checkChecksum:
+        assert(srcSize == 4);  /* guaranteed by dctx->expected */
+        {
+            if (dctx->validateChecksum) {
+                U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
+                U32 const check32 = MEM_readLE32(src);
+                DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
+                RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
+            }
+            ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
+            dctx->expected = 0;
+            dctx->stage = ZSTDds_getFrameHeaderSize;
+            return 0;
+        }
+
+    case ZSTDds_decodeSkippableHeader:
+        assert(src != NULL);
+        assert(srcSize <= ZSTD_SKIPPABLEHEADERSIZE);
+        ZSTD_memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize);   /* complete skippable header */
+        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
+        dctx->stage = ZSTDds_skipFrame;
+        return 0;
+
+    case ZSTDds_skipFrame:
+        dctx->expected = 0;
+        dctx->stage = ZSTDds_getFrameHeaderSize;
+        return 0;
+
+    default:
+        assert(0);   /* impossible */
+        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
+    }
+}
+
+
+static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->virtualStart = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+    dctx->prefixStart = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentBeginForFuzzing = dctx->prefixStart;
+    dctx->dictContentEndForFuzzing = dctx->previousDstEnd;
+#endif
+    return 0;
+}
+
+/*! ZSTD_loadDEntropy() :
+ *  dict : must point at beginning of a valid zstd dictionary.
+ * @return : size of entropy tables read */
+size_t
+ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
+                  const void* const dict, size_t const dictSize)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+
+    RETURN_ERROR_IF(dictSize <= 8, dictionary_corrupted, "dict is too small");
+    assert(MEM_readLE32(dict) == ZSTD_MAGIC_DICTIONARY);   /* dict must be valid */
+    dictPtr += 8;   /* skip header = magic + dictID */
+
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, OFTable) == offsetof(ZSTD_entropyDTables_t, LLTable) + sizeof(entropy->LLTable));
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, MLTable) == offsetof(ZSTD_entropyDTables_t, OFTable) + sizeof(entropy->OFTable));
+    ZSTD_STATIC_ASSERT(sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable) >= HUF_DECOMPRESS_WORKSPACE_SIZE);
+    {   void* const workspace = &entropy->LLTable;   /* use fse tables as temporary workspace; implies fse tables are grouped together */
+        size_t const workspaceSize = sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable);
+#ifdef HUF_FORCE_DECOMPRESS_X1
+        /* in minimal huffman, we always use X1 variants */
+        size_t const hSize = HUF_readDTableX1_wksp(entropy->hufTable,
+                                                dictPtr, dictEnd - dictPtr,
+                                                workspace, workspaceSize, /* flags */ 0);
+#else
+        size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
+                                                dictPtr, (size_t)(dictEnd - dictPtr),
+                                                workspace, workspaceSize, /* flags */ 0);
+#endif
+        RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
+        dictPtr += hSize;
+    }
+
+    {   short offcodeNCount[MaxOff+1];
+        unsigned offcodeMaxValue = MaxOff, offcodeLog;
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeMaxValue > MaxOff, dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->OFTable,
+                            offcodeNCount, offcodeMaxValue,
+                            OF_base, OF_bits,
+                            offcodeLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */0);
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthMaxValue > MaxML, dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->MLTable,
+                            matchlengthNCount, matchlengthMaxValue,
+                            ML_base, ML_bits,
+                            matchlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthMaxValue > MaxLL, dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->LLTable,
+                            litlengthNCount, litlengthMaxValue,
+                            LL_base, LL_bits,
+                            litlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
+        dictPtr += litlengthHeaderSize;
+    }
+
+    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted, "");
+    {   int i;
+        size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
+        for (i=0; i<3; i++) {
+            U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
+            RETURN_ERROR_IF(rep==0 || rep > dictContentSize,
+                            dictionary_corrupted, "");
+            entropy->rep[i] = rep;
+    }   }
+
+    return (size_t)(dictPtr - (const BYTE*)dict);
+}
+
+static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return ZSTD_refDictContent(dctx, dict, dictSize);
+    {   U32 const magic = MEM_readLE32(dict);
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
+            return ZSTD_refDictContent(dctx, dict, dictSize);   /* pure content mode */
+    }   }
+    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
+
+    /* load entropy tables */
+    {   size_t const eSize = ZSTD_loadDEntropy(&dctx->entropy, dict, dictSize);
+        RETURN_ERROR_IF(ZSTD_isError(eSize), dictionary_corrupted, "");
+        dict = (const char*)dict + eSize;
+        dictSize -= eSize;
+    }
+    dctx->litEntropy = dctx->fseEntropy = 1;
+
+    /* reference dictionary content */
+    return ZSTD_refDictContent(dctx, dict, dictSize);
+}
+
+size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
+{
+    assert(dctx != NULL);
+#if ZSTD_TRACE
+    dctx->traceCtx = (ZSTD_trace_decompress_begin != NULL) ? ZSTD_trace_decompress_begin(dctx) : 0;
+#endif
+    dctx->expected = ZSTD_startingInputLength(dctx->format);  /* dctx->format must be properly set */
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->processedCSize = 0;
+    dctx->decodedSize = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->prefixStart = NULL;
+    dctx->virtualStart = NULL;
+    dctx->dictEnd = NULL;
+    dctx->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);  /* cover both little and big endian */
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    dctx->bType = bt_reserved;
+    ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
+    ZSTD_memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    dctx->LLTptr = dctx->entropy.LLTable;
+    dctx->MLTptr = dctx->entropy.MLTable;
+    dctx->OFTptr = dctx->entropy.OFTable;
+    dctx->HUFptr = dctx->entropy.hufTable;
+    return 0;
+}
+
+size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
+    if (dict && dictSize)
+        RETURN_ERROR_IF(
+            ZSTD_isError(ZSTD_decompress_insertDictionary(dctx, dict, dictSize)),
+            dictionary_corrupted, "");
+    return 0;
+}
+
+
+/* ======   ZSTD_DDict   ====== */
+
+size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    DEBUGLOG(4, "ZSTD_decompressBegin_usingDDict");
+    assert(dctx != NULL);
+    if (ddict) {
+        const char* const dictStart = (const char*)ZSTD_DDict_dictContent(ddict);
+        size_t const dictSize = ZSTD_DDict_dictSize(ddict);
+        const void* const dictEnd = dictStart + dictSize;
+        dctx->ddictIsCold = (dctx->dictEnd != dictEnd);
+        DEBUGLOG(4, "DDict is %s",
+                    dctx->ddictIsCold ? "~cold~" : "hot!");
+    }
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
+    if (ddict) {   /* NULL ddict is equivalent to no dictionary */
+        ZSTD_copyDDictParameters(dctx, ddict);
+    }
+    return 0;
+}
+
+/*! ZSTD_getDictID_fromDict() :
+ *  Provides the dictID stored within dictionary.
+ *  if @return == 0, the dictionary is not conformant with Zstandard specification.
+ *  It can still be loaded, but as a content-only dictionary. */
+unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return 0;
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) return 0;
+    return MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
+}
+
+/*! ZSTD_getDictID_fromFrame() :
+ *  Provides the dictID required to decompress frame stored within `src`.
+ *  If @return == 0, the dictID could not be decoded.
+ *  This could for one of the following reasons :
+ *  - The frame does not require a dictionary (most common case).
+ *  - The frame was built with dictID intentionally removed.
+ *    Needed dictionary is a hidden piece of information.
+ *    Note : this use case also happens when using a non-conformant dictionary.
+ *  - `srcSize` is too small, and as a result, frame header could not be decoded.
+ *    Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
+ *  - This is not a Zstandard frame.
+ *  When identifying the exact failure cause, it's possible to use
+ *  ZSTD_getFrameHeader(), which will provide a more precise error code. */
+unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
+{
+    ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0, 0, 0 };
+    size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
+    if (ZSTD_isError(hError)) return 0;
+    return zfp.dictID;
+}
+
+
+/*! ZSTD_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Use dictionary without significant overhead. */
+size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const ZSTD_DDict* ddict)
+{
+    /* pass content and size in case legacy frames are encountered */
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize,
+                                     NULL, 0,
+                                     ddict);
+}
+
+
+/*=====================================
+*   Streaming decompression
+*====================================*/
+
+ZSTD_DStream* ZSTD_createDStream(void)
+{
+    DEBUGLOG(3, "ZSTD_createDStream");
+    return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
+}
+
+ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticDCtx(workspace, workspaceSize);
+}
+
+ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDCtx_internal(customMem);
+}
+
+size_t ZSTD_freeDStream(ZSTD_DStream* zds)
+{
+    return ZSTD_freeDCtx(zds);
+}
+
+
+/* ***  Initialization  *** */
+
+size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize; }
+size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx,
+                                   const void* dict, size_t dictSize,
+                                         ZSTD_dictLoadMethod_e dictLoadMethod,
+                                         ZSTD_dictContentType_e dictContentType)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    ZSTD_clearDict(dctx);
+    if (dict && dictSize != 0) {
+        dctx->ddictLocal = ZSTD_createDDict_advanced(dict, dictSize, dictLoadMethod, dictContentType, dctx->customMem);
+        RETURN_ERROR_IF(dctx->ddictLocal == NULL, memory_allocation, "NULL pointer!");
+        dctx->ddict = dctx->ddictLocal;
+        dctx->dictUses = ZSTD_use_indefinitely;
+    }
+    return 0;
+}
+
+size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    FORWARD_IF_ERROR(ZSTD_DCtx_loadDictionary_advanced(dctx, prefix, prefixSize, ZSTD_dlm_byRef, dictContentType), "");
+    dctx->dictUses = ZSTD_use_once;
+    return 0;
+}
+
+size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_DCtx_refPrefix_advanced(dctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+
+/* ZSTD_initDStream_usingDict() :
+ * return : expected size, aka ZSTD_startingInputLength().
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize)
+{
+    DEBUGLOG(4, "ZSTD_initDStream_usingDict");
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(zds, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) , "");
+    return ZSTD_startingInputLength(zds->format);
+}
+
+/* note : this variant can't fail */
+size_t ZSTD_initDStream(ZSTD_DStream* zds)
+{
+    DEBUGLOG(4, "ZSTD_initDStream");
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(zds, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_DCtx_refDDict(zds, NULL), "");
+    return ZSTD_startingInputLength(zds->format);
+}
+
+/* ZSTD_initDStream_usingDDict() :
+ * ddict will just be referenced, and must outlive decompression session
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
+{
+    DEBUGLOG(4, "ZSTD_initDStream_usingDDict");
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) , "");
+    return ZSTD_startingInputLength(dctx->format);
+}
+
+/* ZSTD_resetDStream() :
+ * return : expected size, aka ZSTD_startingInputLength().
+ * this function cannot fail */
+size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
+{
+    DEBUGLOG(4, "ZSTD_resetDStream");
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only), "");
+    return ZSTD_startingInputLength(dctx->format);
+}
+
+
+size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    ZSTD_clearDict(dctx);
+    if (ddict) {
+        dctx->ddict = ddict;
+        dctx->dictUses = ZSTD_use_indefinitely;
+        if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts) {
+            if (dctx->ddictSet == NULL) {
+                dctx->ddictSet = ZSTD_createDDictHashSet(dctx->customMem);
+                if (!dctx->ddictSet) {
+                    RETURN_ERROR(memory_allocation, "Failed to allocate memory for hash set!");
+                }
+            }
+            assert(!dctx->staticSize);  /* Impossible: ddictSet cannot have been allocated if static dctx */
+            FORWARD_IF_ERROR(ZSTD_DDictHashSet_addDDict(dctx->ddictSet, ddict, dctx->customMem), "");
+        }
+    }
+    return 0;
+}
+
+/* ZSTD_DCtx_setMaxWindowSize() :
+ * note : no direct equivalence in ZSTD_DCtx_setParameter,
+ * since this version sets windowSize, and the other sets windowLog */
+size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
+{
+    ZSTD_bounds const bounds = ZSTD_dParam_getBounds(ZSTD_d_windowLogMax);
+    size_t const min = (size_t)1 << bounds.lowerBound;
+    size_t const max = (size_t)1 << bounds.upperBound;
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    RETURN_ERROR_IF(maxWindowSize < min, parameter_outOfBound, "");
+    RETURN_ERROR_IF(maxWindowSize > max, parameter_outOfBound, "");
+    dctx->maxWindowSize = maxWindowSize;
+    return 0;
+}
+
+size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
+{
+    return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, (int)format);
+}
+
+ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
+{
+    ZSTD_bounds bounds = { 0, 0, 0 };
+    switch(dParam) {
+        case ZSTD_d_windowLogMax:
+            bounds.lowerBound = ZSTD_WINDOWLOG_ABSOLUTEMIN;
+            bounds.upperBound = ZSTD_WINDOWLOG_MAX;
+            return bounds;
+        case ZSTD_d_format:
+            bounds.lowerBound = (int)ZSTD_f_zstd1;
+            bounds.upperBound = (int)ZSTD_f_zstd1_magicless;
+            ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
+            return bounds;
+        case ZSTD_d_stableOutBuffer:
+            bounds.lowerBound = (int)ZSTD_bm_buffered;
+            bounds.upperBound = (int)ZSTD_bm_stable;
+            return bounds;
+        case ZSTD_d_forceIgnoreChecksum:
+            bounds.lowerBound = (int)ZSTD_d_validateChecksum;
+            bounds.upperBound = (int)ZSTD_d_ignoreChecksum;
+            return bounds;
+        case ZSTD_d_refMultipleDDicts:
+            bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
+            bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
+            return bounds;
+        case ZSTD_d_disableHuffmanAssembly:
+            bounds.lowerBound = 0;
+            bounds.upperBound = 1;
+            return bounds;
+
+        default:;
+    }
+    bounds.error = ERROR(parameter_unsupported);
+    return bounds;
+}
+
+/* ZSTD_dParam_withinBounds:
+ * @return 1 if value is within dParam bounds,
+ * 0 otherwise */
+static int ZSTD_dParam_withinBounds(ZSTD_dParameter dParam, int value)
+{
+    ZSTD_bounds const bounds = ZSTD_dParam_getBounds(dParam);
+    if (ZSTD_isError(bounds.error)) return 0;
+    if (value < bounds.lowerBound) return 0;
+    if (value > bounds.upperBound) return 0;
+    return 1;
+}
+
+#define CHECK_DBOUNDS(p,v) {                \
+    RETURN_ERROR_IF(!ZSTD_dParam_withinBounds(p, v), parameter_outOfBound, ""); \
+}
+
+size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value)
+{
+    switch (param) {
+        case ZSTD_d_windowLogMax:
+            *value = (int)ZSTD_highbit32((U32)dctx->maxWindowSize);
+            return 0;
+        case ZSTD_d_format:
+            *value = (int)dctx->format;
+            return 0;
+        case ZSTD_d_stableOutBuffer:
+            *value = (int)dctx->outBufferMode;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            *value = (int)dctx->forceIgnoreChecksum;
+            return 0;
+        case ZSTD_d_refMultipleDDicts:
+            *value = (int)dctx->refMultipleDDicts;
+            return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            *value = (int)dctx->disableHufAsm;
+            return 0;
+        default:;
+    }
+    RETURN_ERROR(parameter_unsupported, "");
+}
+
+size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    switch(dParam) {
+        case ZSTD_d_windowLogMax:
+            if (value == 0) value = ZSTD_WINDOWLOG_LIMIT_DEFAULT;
+            CHECK_DBOUNDS(ZSTD_d_windowLogMax, value);
+            dctx->maxWindowSize = ((size_t)1) << value;
+            return 0;
+        case ZSTD_d_format:
+            CHECK_DBOUNDS(ZSTD_d_format, value);
+            dctx->format = (ZSTD_format_e)value;
+            return 0;
+        case ZSTD_d_stableOutBuffer:
+            CHECK_DBOUNDS(ZSTD_d_stableOutBuffer, value);
+            dctx->outBufferMode = (ZSTD_bufferMode_e)value;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            CHECK_DBOUNDS(ZSTD_d_forceIgnoreChecksum, value);
+            dctx->forceIgnoreChecksum = (ZSTD_forceIgnoreChecksum_e)value;
+            return 0;
+        case ZSTD_d_refMultipleDDicts:
+            CHECK_DBOUNDS(ZSTD_d_refMultipleDDicts, value);
+            if (dctx->staticSize != 0) {
+                RETURN_ERROR(parameter_unsupported, "Static dctx does not support multiple DDicts!");
+            }
+            dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
+            return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            CHECK_DBOUNDS(ZSTD_d_disableHuffmanAssembly, value);
+            dctx->disableHufAsm = value != 0;
+            return 0;
+        default:;
+    }
+    RETURN_ERROR(parameter_unsupported, "");
+}
+
+size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
+{
+    if ( (reset == ZSTD_reset_session_only)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        dctx->streamStage = zdss_init;
+        dctx->noForwardProgress = 0;
+    }
+    if ( (reset == ZSTD_reset_parameters)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+        ZSTD_clearDict(dctx);
+        ZSTD_DCtx_resetParameters(dctx);
+    }
+    return 0;
+}
+
+
+size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
+{
+    return ZSTD_sizeof_DCtx(dctx);
+}
+
+size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
+{
+    size_t const blockSize = (size_t) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+    /* space is needed to store the litbuffer after the output of a given block without stomping the extDict of a previous run, as well as to cover both windows against wildcopy*/
+    unsigned long long const neededRBSize = windowSize + blockSize + ZSTD_BLOCKSIZE_MAX + (WILDCOPY_OVERLENGTH * 2);
+    unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
+    size_t const minRBSize = (size_t) neededSize;
+    RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,
+                    frameParameter_windowTooLarge, "");
+    return minRBSize;
+}
+
+size_t ZSTD_estimateDStreamSize(size_t windowSize)
+{
+    size_t const blockSize = MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+    size_t const inBuffSize = blockSize;  /* no block can be larger */
+    size_t const outBuffSize = ZSTD_decodingBufferSize_min(windowSize, ZSTD_CONTENTSIZE_UNKNOWN);
+    return ZSTD_estimateDCtxSize() + inBuffSize + outBuffSize;
+}
+
+size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize)
+{
+    U32 const windowSizeMax = 1U << ZSTD_WINDOWLOG_MAX;   /* note : should be user-selectable, but requires an additional parameter (or a dctx) */
+    ZSTD_frameHeader zfh;
+    size_t const err = ZSTD_getFrameHeader(&zfh, src, srcSize);
+    if (ZSTD_isError(err)) return err;
+    RETURN_ERROR_IF(err>0, srcSize_wrong, "");
+    RETURN_ERROR_IF(zfh.windowSize > windowSizeMax,
+                    frameParameter_windowTooLarge, "");
+    return ZSTD_estimateDStreamSize((size_t)zfh.windowSize);
+}
+
+
+/* *****   Decompression   ***** */
+
+static int ZSTD_DCtx_isOverflow(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
+{
+    return (zds->inBuffSize + zds->outBuffSize) >= (neededInBuffSize + neededOutBuffSize) * ZSTD_WORKSPACETOOLARGE_FACTOR;
+}
+
+static void ZSTD_DCtx_updateOversizedDuration(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
+{
+    if (ZSTD_DCtx_isOverflow(zds, neededInBuffSize, neededOutBuffSize))
+        zds->oversizedDuration++;
+    else
+        zds->oversizedDuration = 0;
+}
+
+static int ZSTD_DCtx_isOversizedTooLong(ZSTD_DStream* zds)
+{
+    return zds->oversizedDuration >= ZSTD_WORKSPACETOOLARGE_MAXDURATION;
+}
+
+/* Checks that the output buffer hasn't changed if ZSTD_obm_stable is used. */
+static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const* output)
+{
+    ZSTD_outBuffer const expect = zds->expectedOutBuffer;
+    /* No requirement when ZSTD_obm_stable is not enabled. */
+    if (zds->outBufferMode != ZSTD_bm_stable)
+        return 0;
+    /* Any buffer is allowed in zdss_init, this must be the same for every other call until
+     * the context is reset.
+     */
+    if (zds->streamStage == zdss_init)
+        return 0;
+    /* The buffer must match our expectation exactly. */
+    if (expect.dst == output->dst && expect.pos == output->pos && expect.size == output->size)
+        return 0;
+    RETURN_ERROR(dstBuffer_wrong, "ZSTD_d_stableOutBuffer enabled but output differs!");
+}
+
+/* Calls ZSTD_decompressContinue() with the right parameters for ZSTD_decompressStream()
+ * and updates the stage and the output buffer state. This call is extracted so it can be
+ * used both when reading directly from the ZSTD_inBuffer, and in buffered input mode.
+ * NOTE: You must break after calling this function since the streamStage is modified.
+ */
+static size_t ZSTD_decompressContinueStream(
+            ZSTD_DStream* zds, char** op, char* oend,
+            void const* src, size_t srcSize) {
+    int const isSkipFrame = ZSTD_isSkipFrame(zds);
+    if (zds->outBufferMode == ZSTD_bm_buffered) {
+        size_t const dstSize = isSkipFrame ? 0 : zds->outBuffSize - zds->outStart;
+        size_t const decodedSize = ZSTD_decompressContinue(zds,
+                zds->outBuff + zds->outStart, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        if (!decodedSize && !isSkipFrame) {
+            zds->streamStage = zdss_read;
+        } else {
+            zds->outEnd = zds->outStart + decodedSize;
+            zds->streamStage = zdss_flush;
+        }
+    } else {
+        /* Write directly into the output buffer */
+        size_t const dstSize = isSkipFrame ? 0 : (size_t)(oend - *op);
+        size_t const decodedSize = ZSTD_decompressContinue(zds, *op, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        *op += decodedSize;
+        /* Flushing is not needed. */
+        zds->streamStage = zdss_read;
+        assert(*op <= oend);
+        assert(zds->outBufferMode == ZSTD_bm_stable);
+    }
+    return 0;
+}
+
+size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    const char* const src = (const char*)input->src;
+    const char* const istart = input->pos != 0 ? src + input->pos : src;
+    const char* const iend = input->size != 0 ? src + input->size : src;
+    const char* ip = istart;
+    char* const dst = (char*)output->dst;
+    char* const ostart = output->pos != 0 ? dst + output->pos : dst;
+    char* const oend = output->size != 0 ? dst + output->size : dst;
+    char* op = ostart;
+    U32 someMoreWork = 1;
+
+    DEBUGLOG(5, "ZSTD_decompressStream");
+    RETURN_ERROR_IF(
+        input->pos > input->size,
+        srcSize_wrong,
+        "forbidden. in: pos: %u   vs size: %u",
+        (U32)input->pos, (U32)input->size);
+    RETURN_ERROR_IF(
+        output->pos > output->size,
+        dstSize_tooSmall,
+        "forbidden. out: pos: %u   vs size: %u",
+        (U32)output->pos, (U32)output->size);
+    DEBUGLOG(5, "input size : %u", (U32)(input->size - input->pos));
+    FORWARD_IF_ERROR(ZSTD_checkOutBuffer(zds, output), "");
+
+    while (someMoreWork) {
+        switch(zds->streamStage)
+        {
+        case zdss_init :
+            DEBUGLOG(5, "stage zdss_init => transparent reset ");
+            zds->streamStage = zdss_loadHeader;
+            zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+            zds->legacyVersion = 0;
+#endif
+            zds->hostageByte = 0;
+            zds->expectedOutBuffer = *output;
+            ZSTD_FALLTHROUGH;
+
+        case zdss_loadHeader :
+            DEBUGLOG(5, "stage zdss_loadHeader (srcSize : %u)", (U32)(iend - ip));
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+            if (zds->legacyVersion) {
+                RETURN_ERROR_IF(zds->staticSize, memory_allocation,
+                    "legacy support is incompatible with static dctx");
+                {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
+                    if (hint==0) zds->streamStage = zdss_init;
+                    return hint;
+            }   }
+#endif
+            {   size_t const hSize = ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
+                if (zds->refMultipleDDicts && zds->ddictSet) {
+                    ZSTD_DCtx_selectFrameDDict(zds);
+                }
+                if (ZSTD_isError(hSize)) {
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+                    U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
+                    if (legacyVersion) {
+                        ZSTD_DDict const* const ddict = ZSTD_getDDict(zds);
+                        const void* const dict = ddict ? ZSTD_DDict_dictContent(ddict) : NULL;
+                        size_t const dictSize = ddict ? ZSTD_DDict_dictSize(ddict) : 0;
+                        DEBUGLOG(5, "ZSTD_decompressStream: detected legacy version v0.%u", legacyVersion);
+                        RETURN_ERROR_IF(zds->staticSize, memory_allocation,
+                            "legacy support is incompatible with static dctx");
+                        FORWARD_IF_ERROR(ZSTD_initLegacyStream(&zds->legacyContext,
+                                    zds->previousLegacyVersion, legacyVersion,
+                                    dict, dictSize), "");
+                        zds->legacyVersion = zds->previousLegacyVersion = legacyVersion;
+                        {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, legacyVersion, output, input);
+                            if (hint==0) zds->streamStage = zdss_init;   /* or stay in stage zdss_loadHeader */
+                            return hint;
+                    }   }
+#endif
+                    return hSize;   /* error */
+                }
+                if (hSize != 0) {   /* need more input */
+                    size_t const toLoad = hSize - zds->lhSize;   /* if hSize!=0, hSize > zds->lhSize */
+                    size_t const remainingInput = (size_t)(iend-ip);
+                    assert(iend >= ip);
+                    if (toLoad > remainingInput) {   /* not enough input to load full header */
+                        if (remainingInput > 0) {
+                            ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
+                            zds->lhSize += remainingInput;
+                        }
+                        input->pos = input->size;
+                        /* check first few bytes */
+                        FORWARD_IF_ERROR(
+                            ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format),
+                            "First few bytes detected incorrect" );
+                        /* return hint input size */
+                        return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    assert(ip != NULL);
+                    ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* check for single-pass mode opportunity */
+            if (zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && zds->fParams.frameType != ZSTD_skippableFrame
+                && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
+                size_t const cSize = ZSTD_findFrameCompressedSize(istart, (size_t)(iend-istart));
+                if (cSize <= (size_t)(iend-istart)) {
+                    /* shortcut : using single-pass mode */
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, (size_t)(oend-op), istart, cSize, ZSTD_getDDict(zds));
+                    if (ZSTD_isError(decompressedSize)) return decompressedSize;
+                    DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
+                    assert(istart != NULL);
+                    ip = istart + cSize;
+                    op = op ? op + decompressedSize : op; /* can occur if frameContentSize = 0 (empty frame) */
+                    zds->expected = 0;
+                    zds->streamStage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+            }   }
+
+            /* Check output buffer is large enough for ZSTD_odm_stable. */
+            if (zds->outBufferMode == ZSTD_bm_stable
+                && zds->fParams.frameType != ZSTD_skippableFrame
+                && zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && (U64)(size_t)(oend-op) < zds->fParams.frameContentSize) {
+                RETURN_ERROR(dstSize_tooSmall, "ZSTD_obm_stable passed but ZSTD_outBuffer is too small");
+            }
+
+            /* Consume header (see ZSTDds_decodeFrameHeader) */
+            DEBUGLOG(4, "Consume header");
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(zds, ZSTD_getDDict(zds)), "");
+
+            if ((MEM_readLE32(zds->headerBuffer) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
+                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_FRAMEIDSIZE);
+                zds->stage = ZSTDds_skipFrame;
+            } else {
+                FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize), "");
+                zds->expected = ZSTD_blockHeaderSize;
+                zds->stage = ZSTDds_decodeBlockHeader;
+            }
+
+            /* control buffer memory usage */
+            DEBUGLOG(4, "Control max memory usage (%u KB <= max %u KB)",
+                        (U32)(zds->fParams.windowSize >>10),
+                        (U32)(zds->maxWindowSize >> 10) );
+            zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
+            RETURN_ERROR_IF(zds->fParams.windowSize > zds->maxWindowSize,
+                            frameParameter_windowTooLarge, "");
+
+            /* Adapt buffer sizes to frame header instructions */
+            {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
+                size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_bm_buffered
+                        ? ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize)
+                        : 0;
+
+                ZSTD_DCtx_updateOversizedDuration(zds, neededInBuffSize, neededOutBuffSize);
+
+                {   int const tooSmall = (zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize);
+                    int const tooLarge = ZSTD_DCtx_isOversizedTooLong(zds);
+
+                    if (tooSmall || tooLarge) {
+                        size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
+                        DEBUGLOG(4, "inBuff  : from %u to %u",
+                                    (U32)zds->inBuffSize, (U32)neededInBuffSize);
+                        DEBUGLOG(4, "outBuff : from %u to %u",
+                                    (U32)zds->outBuffSize, (U32)neededOutBuffSize);
+                        if (zds->staticSize) {  /* static DCtx */
+                            DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
+                            assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
+                            RETURN_ERROR_IF(
+                                bufferSize > zds->staticSize - sizeof(ZSTD_DCtx),
+                                memory_allocation, "");
+                        } else {
+                            ZSTD_customFree(zds->inBuff, zds->customMem);
+                            zds->inBuffSize = 0;
+                            zds->outBuffSize = 0;
+                            zds->inBuff = (char*)ZSTD_customMalloc(bufferSize, zds->customMem);
+                            RETURN_ERROR_IF(zds->inBuff == NULL, memory_allocation, "");
+                        }
+                        zds->inBuffSize = neededInBuffSize;
+                        zds->outBuff = zds->inBuff + zds->inBuffSize;
+                        zds->outBuffSize = neededOutBuffSize;
+            }   }   }
+            zds->streamStage = zdss_read;
+            ZSTD_FALLTHROUGH;
+
+        case zdss_read:
+            DEBUGLOG(5, "stage zdss_read");
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip));
+                DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
+                if (neededInSize==0) {  /* end of frame */
+                    zds->streamStage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, ip, neededInSize), "");
+                    assert(ip != NULL);
+                    ip += neededInSize;
+                    /* Function modifies the stage so we must break */
+                    break;
+            }   }
+            if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
+            zds->streamStage = zdss_load;
+            ZSTD_FALLTHROUGH;
+
+        case zdss_load:
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+                size_t const toLoad = neededInSize - zds->inPos;
+                int const isSkipFrame = ZSTD_isSkipFrame(zds);
+                size_t loadedSize;
+                /* At this point we shouldn't be decompressing a block that we can stream. */
+                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip)));
+                if (isSkipFrame) {
+                    loadedSize = MIN(toLoad, (size_t)(iend-ip));
+                } else {
+                    RETURN_ERROR_IF(toLoad > zds->inBuffSize - zds->inPos,
+                                    corruption_detected,
+                                    "should never happen");
+                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
+                }
+                if (loadedSize != 0) {
+                    /* ip may be NULL */
+                    ip += loadedSize;
+                    zds->inPos += loadedSize;
+                }
+                if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                zds->inPos = 0;   /* input is consumed */
+                FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, zds->inBuff, neededInSize), "");
+                /* Function modifies the stage so we must break */
+                break;
+            }
+        case zdss_flush:
+            {
+                size_t const toFlushSize = zds->outEnd - zds->outStart;
+                size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
+
+                op = op ? op + flushedSize : op;
+
+                zds->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {  /* flush completed */
+                    zds->streamStage = zdss_read;
+                    if ( (zds->outBuffSize < zds->fParams.frameContentSize)
+                        && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
+                        DEBUGLOG(5, "restart filling outBuff from beginning (left:%i, needed:%u)",
+                                (int)(zds->outBuffSize - zds->outStart),
+                                (U32)zds->fParams.blockSizeMax);
+                        zds->outStart = zds->outEnd = 0;
+                    }
+                    break;
+            }   }
+            /* cannot complete flush */
+            someMoreWork = 0;
+            break;
+
+        default:
+            assert(0);    /* impossible */
+            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
+    }   }
+
+    /* result */
+    input->pos = (size_t)(ip - (const char*)(input->src));
+    output->pos = (size_t)(op - (char*)(output->dst));
+
+    /* Update the expected output buffer for ZSTD_obm_stable. */
+    zds->expectedOutBuffer = *output;
+
+    if ((ip==istart) && (op==ostart)) {  /* no forward progress */
+        zds->noForwardProgress ++;
+        if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
+            RETURN_ERROR_IF(op==oend, noForwardProgress_destFull, "");
+            RETURN_ERROR_IF(ip==iend, noForwardProgress_inputEmpty, "");
+            assert(0);
+        }
+    } else {
+        zds->noForwardProgress = 0;
+    }
+    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds);
+        if (!nextSrcSizeHint) {   /* frame fully decoded */
+            if (zds->outEnd == zds->outStart) {  /* output fully flushed */
+                if (zds->hostageByte) {
+                    if (input->pos >= input->size) {
+                        /* can't release hostage (not present) */
+                        zds->streamStage = zdss_read;
+                        return 1;
+                    }
+                    input->pos++;  /* release hostage */
+                }   /* zds->hostageByte */
+                return 0;
+            }  /* zds->outEnd == zds->outStart */
+            if (!zds->hostageByte) { /* output not fully flushed; keep last byte as hostage; will be released when all output is flushed */
+                input->pos--;   /* note : pos > 0, otherwise, impossible to finish reading last block */
+                zds->hostageByte=1;
+            }
+            return 1;
+        }  /* nextSrcSizeHint==0 */
+        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds) == ZSTDnit_block);   /* preload header of next block */
+        assert(zds->inPos <= nextSrcSizeHint);
+        nextSrcSizeHint -= zds->inPos;   /* part already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+size_t ZSTD_decompressStream_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos)
+{
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
+    {   size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_block.c b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_block.c
new file mode 100644
index 0000000..09896a9
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_block.c
@@ -0,0 +1,2192 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* zstd_decompress_block :
+ * this module takes care of decompressing _compressed_ block */
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/compiler.h"    /* prefetch */
+#include "../common/cpu.h"         /* bmi2 */
+#include "../common/mem.h"         /* low level memory routines */
+#define FSE_STATIC_LINKING_ONLY
+#include "../common/fse.h"
+#include "../common/huf.h"
+#include "../common/zstd_internal.h"
+#include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
+#include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
+#include "zstd_decompress_block.h"
+#include "../common/bits.h"  /* ZSTD_highbit32 */
+
+/*_*******************************************************
+*  Macros
+**********************************************************/
+
+/* These two optional macros force the use one way or another of the two
+ * ZSTD_decompressSequences implementations. You can't force in both directions
+ * at the same time.
+ */
+#if defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+#error "Cannot force the use of the short and the long ZSTD_decompressSequences variants!"
+#endif
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTD_copy4(void* dst, const void* src) { ZSTD_memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+ *   Block decoding
+ ***************************************************************/
+
+/*! ZSTD_getcBlockSize() :
+ *  Provides the size of compressed block from block header `src` */
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr)
+{
+    RETURN_ERROR_IF(srcSize < ZSTD_blockHeaderSize, srcSize_wrong, "");
+
+    {   U32 const cBlockHeader = MEM_readLE24(src);
+        U32 const cSize = cBlockHeader >> 3;
+        bpPtr->lastBlock = cBlockHeader & 1;
+        bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3);
+        bpPtr->origSize = cSize;   /* only useful for RLE */
+        if (bpPtr->blockType == bt_rle) return 1;
+        RETURN_ERROR_IF(bpPtr->blockType == bt_reserved, corruption_detected, "");
+        return cSize;
+    }
+}
+
+/* Allocate buffer for literals, either overlapping current dst, or split between dst and litExtraBuffer, or stored entirely within litExtraBuffer */
+static void ZSTD_allocateLiteralsBuffer(ZSTD_DCtx* dctx, void* const dst, const size_t dstCapacity, const size_t litSize,
+    const streaming_operation streaming, const size_t expectedWriteSize, const unsigned splitImmediately)
+{
+    if (streaming == not_streaming && dstCapacity > ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH + litSize + WILDCOPY_OVERLENGTH)
+    {
+        /* room for litbuffer to fit without read faulting */
+        dctx->litBuffer = (BYTE*)dst + ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH;
+        dctx->litBufferEnd = dctx->litBuffer + litSize;
+        dctx->litBufferLocation = ZSTD_in_dst;
+    }
+    else if (litSize > ZSTD_LITBUFFEREXTRASIZE)
+    {
+        /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
+        if (splitImmediately) {
+            /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
+            dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
+            dctx->litBufferEnd = dctx->litBuffer + litSize - ZSTD_LITBUFFEREXTRASIZE;
+        }
+        else {
+            /* initially this will be stored entirely in dst during huffman decoding, it will partially be shifted to litExtraBuffer after */
+            dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize;
+            dctx->litBufferEnd = (BYTE*)dst + expectedWriteSize;
+        }
+        dctx->litBufferLocation = ZSTD_split;
+    }
+    else
+    {
+        /* fits entirely within litExtraBuffer, so no split is necessary */
+        dctx->litBuffer = dctx->litExtraBuffer;
+        dctx->litBufferEnd = dctx->litBuffer + litSize;
+        dctx->litBufferLocation = ZSTD_not_in_dst;
+    }
+}
+
+/* Hidden declaration for fullbench */
+size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize,
+                          void* dst, size_t dstCapacity, const streaming_operation streaming);
+/*! ZSTD_decodeLiteralsBlock() :
+ * Where it is possible to do so without being stomped by the output during decompression, the literals block will be stored
+ * in the dstBuffer.  If there is room to do so, it will be stored in full in the excess dst space after where the current
+ * block will be output.  Otherwise it will be stored at the end of the current dst blockspace, with a small portion being
+ * stored in dctx->litExtraBuffer to help keep it "ahead" of the current output write.
+ *
+ * @return : nb of bytes read from src (< srcSize )
+ *  note : symbol not declared but exposed for fullbench */
+size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize,   /* note : srcSize < BLOCKSIZE */
+                          void* dst, size_t dstCapacity, const streaming_operation streaming)
+{
+    DEBUGLOG(5, "ZSTD_decodeLiteralsBlock");
+    RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected, "");
+
+    {   const BYTE* const istart = (const BYTE*) src;
+        symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3);
+
+        switch(litEncType)
+        {
+        case set_repeat:
+            DEBUGLOG(5, "set_repeat flag : re-using stats from previous compressed literals block");
+            RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted, "");
+            ZSTD_FALLTHROUGH;
+
+        case set_compressed:
+            RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need up to 5 for case 3");
+            {   size_t lhSize, litSize, litCSize;
+                U32 singleStream=0;
+                U32 const lhlCode = (istart[0] >> 2) & 3;
+                U32 const lhc = MEM_readLE32(istart);
+                size_t hufSuccess;
+                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
+                int const flags = 0
+                    | (ZSTD_DCtx_get_bmi2(dctx) ? HUF_flags_bmi2 : 0)
+                    | (dctx->disableHufAsm ? HUF_flags_disableAsm : 0);
+                switch(lhlCode)
+                {
+                case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    /* 2 - 2 - 10 - 10 */
+                    singleStream = !lhlCode;
+                    lhSize = 3;
+                    litSize  = (lhc >> 4) & 0x3FF;
+                    litCSize = (lhc >> 14) & 0x3FF;
+                    break;
+                case 2:
+                    /* 2 - 2 - 14 - 14 */
+                    lhSize = 4;
+                    litSize  = (lhc >> 4) & 0x3FFF;
+                    litCSize = lhc >> 18;
+                    break;
+                case 3:
+                    /* 2 - 2 - 18 - 18 */
+                    lhSize = 5;
+                    litSize  = (lhc >> 4) & 0x3FFFF;
+                    litCSize = (lhc >> 22) + ((size_t)istart[4] << 10);
+                    break;
+                }
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
+                if (!singleStream)
+                    RETURN_ERROR_IF(litSize < MIN_LITERALS_FOR_4_STREAMS, literals_headerWrong,
+                        "Not enough literals (%zu) for the 4-streams mode (min %u)",
+                        litSize, MIN_LITERALS_FOR_4_STREAMS);
+                RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
+                RETURN_ERROR_IF(expectedWriteSize < litSize , dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 0);
+
+                /* prefetch huffman table if cold */
+                if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
+                    PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable));
+                }
+
+                if (litEncType==set_repeat) {
+                    if (singleStream) {
+                        hufSuccess = HUF_decompress1X_usingDTable(
+                            dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                            dctx->HUFptr, flags);
+                    } else {
+                        assert(litSize >= MIN_LITERALS_FOR_4_STREAMS);
+                        hufSuccess = HUF_decompress4X_usingDTable(
+                            dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                            dctx->HUFptr, flags);
+                    }
+                } else {
+                    if (singleStream) {
+#if defined(HUF_FORCE_DECOMPRESS_X2)
+                        hufSuccess = HUF_decompress1X_DCtx_wksp(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), flags);
+#else
+                        hufSuccess = HUF_decompress1X1_DCtx_wksp(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), flags);
+#endif
+                    } else {
+                        hufSuccess = HUF_decompress4X_hufOnly_wksp(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), flags);
+                    }
+                }
+                if (dctx->litBufferLocation == ZSTD_split)
+                {
+                    ZSTD_memcpy(dctx->litExtraBuffer, dctx->litBufferEnd - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
+                    ZSTD_memmove(dctx->litBuffer + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH, dctx->litBuffer, litSize - ZSTD_LITBUFFEREXTRASIZE);
+                    dctx->litBuffer += ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
+                    dctx->litBufferEnd -= WILDCOPY_OVERLENGTH;
+                }
+
+                RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected, "");
+
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                dctx->litEntropy = 1;
+                if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
+                return litCSize + lhSize;
+            }
+
+        case set_basic:
+            {   size_t litSize, lhSize;
+                U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize = 3");
+                    litSize = MEM_readLE24(istart) >> 4;
+                    break;
+                }
+
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
+                if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                    RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
+                    if (dctx->litBufferLocation == ZSTD_split)
+                    {
+                        ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize - ZSTD_LITBUFFEREXTRASIZE);
+                        ZSTD_memcpy(dctx->litExtraBuffer, istart + lhSize + litSize - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
+                    }
+                    else
+                    {
+                        ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize);
+                    }
+                    dctx->litPtr = dctx->litBuffer;
+                    dctx->litSize = litSize;
+                    return lhSize+litSize;
+                }
+                /* direct reference into compressed stream */
+                dctx->litPtr = istart+lhSize;
+                dctx->litSize = litSize;
+                dctx->litBufferEnd = dctx->litPtr + litSize;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                return lhSize+litSize;
+            }
+
+        case set_rle:
+            {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                size_t litSize, lhSize;
+                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 3");
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 4");
+                    litSize = MEM_readLE24(istart) >> 4;
+                    break;
+                }
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
+                RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
+                if (dctx->litBufferLocation == ZSTD_split)
+                {
+                    ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize - ZSTD_LITBUFFEREXTRASIZE);
+                    ZSTD_memset(dctx->litExtraBuffer, istart[lhSize], ZSTD_LITBUFFEREXTRASIZE);
+                }
+                else
+                {
+                    ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize);
+                }
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                return lhSize+1;
+            }
+        default:
+            RETURN_ERROR(corruption_detected, "impossible");
+        }
+    }
+}
+
+/* Default FSE distribution tables.
+ * These are pre-calculated FSE decoding tables using default distributions as defined in specification :
+ * https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#default-distributions
+ * They were generated programmatically with following method :
+ * - start from default distributions, present in /lib/common/zstd_internal.h
+ * - generate tables normally, using ZSTD_buildFSETable()
+ * - printout the content of tables
+ * - pretify output, report below, test with fuzzer to ensure it's correct */
+
+/* Default FSE distribution table for Literal Lengths */
+static const ZSTD_seqSymbol LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
+     {  1,  1,  1, LL_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+     /* nextState, nbAddBits, nbBits, baseVal */
+     {  0,  0,  4,    0},  { 16,  0,  4,    0},
+     { 32,  0,  5,    1},  {  0,  0,  5,    3},
+     {  0,  0,  5,    4},  {  0,  0,  5,    6},
+     {  0,  0,  5,    7},  {  0,  0,  5,    9},
+     {  0,  0,  5,   10},  {  0,  0,  5,   12},
+     {  0,  0,  6,   14},  {  0,  1,  5,   16},
+     {  0,  1,  5,   20},  {  0,  1,  5,   22},
+     {  0,  2,  5,   28},  {  0,  3,  5,   32},
+     {  0,  4,  5,   48},  { 32,  6,  5,   64},
+     {  0,  7,  5,  128},  {  0,  8,  6,  256},
+     {  0, 10,  6, 1024},  {  0, 12,  6, 4096},
+     { 32,  0,  4,    0},  {  0,  0,  4,    1},
+     {  0,  0,  5,    2},  { 32,  0,  5,    4},
+     {  0,  0,  5,    5},  { 32,  0,  5,    7},
+     {  0,  0,  5,    8},  { 32,  0,  5,   10},
+     {  0,  0,  5,   11},  {  0,  0,  6,   13},
+     { 32,  1,  5,   16},  {  0,  1,  5,   18},
+     { 32,  1,  5,   22},  {  0,  2,  5,   24},
+     { 32,  3,  5,   32},  {  0,  3,  5,   40},
+     {  0,  6,  4,   64},  { 16,  6,  4,   64},
+     { 32,  7,  5,  128},  {  0,  9,  6,  512},
+     {  0, 11,  6, 2048},  { 48,  0,  4,    0},
+     { 16,  0,  4,    1},  { 32,  0,  5,    2},
+     { 32,  0,  5,    3},  { 32,  0,  5,    5},
+     { 32,  0,  5,    6},  { 32,  0,  5,    8},
+     { 32,  0,  5,    9},  { 32,  0,  5,   11},
+     { 32,  0,  5,   12},  {  0,  0,  6,   15},
+     { 32,  1,  5,   18},  { 32,  1,  5,   20},
+     { 32,  2,  5,   24},  { 32,  2,  5,   28},
+     { 32,  3,  5,   40},  { 32,  4,  5,   48},
+     {  0, 16,  6,65536},  {  0, 15,  6,32768},
+     {  0, 14,  6,16384},  {  0, 13,  6, 8192},
+};   /* LL_defaultDTable */
+
+/* Default FSE distribution table for Offset Codes */
+static const ZSTD_seqSymbol OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, OF_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  5,    0},     {  0,  6,  4,   61},
+    {  0,  9,  5,  509},     {  0, 15,  5,32765},
+    {  0, 21,  5,2097149},   {  0,  3,  5,    5},
+    {  0,  7,  4,  125},     {  0, 12,  5, 4093},
+    {  0, 18,  5,262141},    {  0, 23,  5,8388605},
+    {  0,  5,  5,   29},     {  0,  8,  4,  253},
+    {  0, 14,  5,16381},     {  0, 20,  5,1048573},
+    {  0,  2,  5,    1},     { 16,  7,  4,  125},
+    {  0, 11,  5, 2045},     {  0, 17,  5,131069},
+    {  0, 22,  5,4194301},   {  0,  4,  5,   13},
+    { 16,  8,  4,  253},     {  0, 13,  5, 8189},
+    {  0, 19,  5,524285},    {  0,  1,  5,    1},
+    { 16,  6,  4,   61},     {  0, 10,  5, 1021},
+    {  0, 16,  5,65533},     {  0, 28,  5,268435453},
+    {  0, 27,  5,134217725}, {  0, 26,  5,67108861},
+    {  0, 25,  5,33554429},  {  0, 24,  5,16777213},
+};   /* OF_defaultDTable */
+
+
+/* Default FSE distribution table for Match Lengths */
+static const ZSTD_seqSymbol ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, ML_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  6,    3},  {  0,  0,  4,    4},
+    { 32,  0,  5,    5},  {  0,  0,  5,    6},
+    {  0,  0,  5,    8},  {  0,  0,  5,    9},
+    {  0,  0,  5,   11},  {  0,  0,  6,   13},
+    {  0,  0,  6,   16},  {  0,  0,  6,   19},
+    {  0,  0,  6,   22},  {  0,  0,  6,   25},
+    {  0,  0,  6,   28},  {  0,  0,  6,   31},
+    {  0,  0,  6,   34},  {  0,  1,  6,   37},
+    {  0,  1,  6,   41},  {  0,  2,  6,   47},
+    {  0,  3,  6,   59},  {  0,  4,  6,   83},
+    {  0,  7,  6,  131},  {  0,  9,  6,  515},
+    { 16,  0,  4,    4},  {  0,  0,  4,    5},
+    { 32,  0,  5,    6},  {  0,  0,  5,    7},
+    { 32,  0,  5,    9},  {  0,  0,  5,   10},
+    {  0,  0,  6,   12},  {  0,  0,  6,   15},
+    {  0,  0,  6,   18},  {  0,  0,  6,   21},
+    {  0,  0,  6,   24},  {  0,  0,  6,   27},
+    {  0,  0,  6,   30},  {  0,  0,  6,   33},
+    {  0,  1,  6,   35},  {  0,  1,  6,   39},
+    {  0,  2,  6,   43},  {  0,  3,  6,   51},
+    {  0,  4,  6,   67},  {  0,  5,  6,   99},
+    {  0,  8,  6,  259},  { 32,  0,  4,    4},
+    { 48,  0,  4,    4},  { 16,  0,  4,    5},
+    { 32,  0,  5,    7},  { 32,  0,  5,    8},
+    { 32,  0,  5,   10},  { 32,  0,  5,   11},
+    {  0,  0,  6,   14},  {  0,  0,  6,   17},
+    {  0,  0,  6,   20},  {  0,  0,  6,   23},
+    {  0,  0,  6,   26},  {  0,  0,  6,   29},
+    {  0,  0,  6,   32},  {  0, 16,  6,65539},
+    {  0, 15,  6,32771},  {  0, 14,  6,16387},
+    {  0, 13,  6, 8195},  {  0, 12,  6, 4099},
+    {  0, 11,  6, 2051},  {  0, 10,  6, 1027},
+};   /* ML_defaultDTable */
+
+
+static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U8 nbAddBits)
+{
+    void* ptr = dt;
+    ZSTD_seqSymbol_header* const DTableH = (ZSTD_seqSymbol_header*)ptr;
+    ZSTD_seqSymbol* const cell = dt + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->nbBits = 0;
+    cell->nextState = 0;
+    assert(nbAddBits < 255);
+    cell->nbAdditionalBits = nbAddBits;
+    cell->baseValue = baseValue;
+}
+
+
+/* ZSTD_buildFSETable() :
+ * generate FSE decoding table for one symbol (ll, ml or off)
+ * cannot fail if input is valid =>
+ * all inputs are presumed validated at this stage */
+FORCE_INLINE_TEMPLATE
+void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_seqSymbol* const tableDecode = dt+1;
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+
+    U16* symbolNext = (U16*)wksp;
+    BYTE* spread = (BYTE*)(symbolNext + MaxSeq + 1);
+    U32 highThreshold = tableSize - 1;
+
+
+    /* Sanity Checks */
+    assert(maxSymbolValue <= MaxSeq);
+    assert(tableLog <= MaxFSELog);
+    assert(wkspSize >= ZSTD_BUILD_FSE_TABLE_WKSP_SIZE);
+    (void)wkspSize;
+    /* Init, lay down lowprob symbols */
+    {   ZSTD_seqSymbol_header DTableH;
+        DTableH.tableLog = tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].baseValue = s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    assert(normalizedCounter[s]>=0);
+                    symbolNext[s] = (U16)normalizedCounter[s];
+        }   }   }
+        ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    assert(tableSize <= 512);
+    /* Specialized symbol spreading for the case when there are
+     * no low probability (-1 count) symbols. When compressing
+     * small blocks we avoid low probability symbols to hit this
+     * case, since header decoding speed matters more.
+     */
+    if (highThreshold == tableSize - 1) {
+        size_t const tableMask = tableSize-1;
+        size_t const step = FSE_TABLESTEP(tableSize);
+        /* First lay down the symbols in order.
+         * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
+         * misses since small blocks generally have small table logs, so nearly
+         * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
+         * our buffer to handle the over-write.
+         */
+        {
+            U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                assert(n>=0);
+                pos += (size_t)n;
+            }
+        }
+        /* Now we spread those positions across the table.
+         * The benefit of doing it in two stages is that we avoid the
+         * variable size inner loop, which caused lots of branch misses.
+         * Now we can run through all the positions without any branch misses.
+         * We unroll the loop twice, since that is what empirically worked best.
+         */
+        {
+            size_t position = 0;
+            size_t s;
+            size_t const unroll = 2;
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableDecode[uPosition].baseValue = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);
+        }
+    } else {
+        U32 const tableMask = tableSize-1;
+        U32 const step = FSE_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            int const n = normalizedCounter[s];
+            for (i=0; i<n; i++) {
+                tableDecode[position].baseValue = s;
+                position = (position + step) & tableMask;
+                while (UNLIKELY(position > highThreshold)) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+        assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {
+        U32 u;
+        for (u=0; u<tableSize; u++) {
+            U32 const symbol = tableDecode[u].baseValue;
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
+            tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+            assert(nbAdditionalBits[symbol] < 255);
+            tableDecode[u].nbAdditionalBits = nbAdditionalBits[symbol];
+            tableDecode[u].baseValue = baseValue[symbol];
+        }
+    }
+}
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static void ZSTD_buildFSETable_body_default(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+
+#if DYNAMIC_BMI2
+BMI2_TARGET_ATTRIBUTE static void ZSTD_buildFSETable_body_bmi2(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+#endif
+
+void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        ZSTD_buildFSETable_body_bmi2(dt, normalizedCounter, maxSymbolValue,
+                baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+        return;
+    }
+#endif
+    (void)bmi2;
+    ZSTD_buildFSETable_body_default(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+
+
+/*! ZSTD_buildSeqTable() :
+ * @return : nb bytes read from src,
+ *           or an error code if it fails */
+static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymbol** DTablePtr,
+                                 symbolEncodingType_e type, unsigned max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const U32* baseValue, const U8* nbAdditionalBits,
+                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
+                                 int ddictIsCold, int nbSeq, U32* wksp, size_t wkspSize,
+                                 int bmi2)
+{
+    switch(type)
+    {
+    case set_rle :
+        RETURN_ERROR_IF(!srcSize, srcSize_wrong, "");
+        RETURN_ERROR_IF((*(const BYTE*)src) > max, corruption_detected, "");
+        {   U32 const symbol = *(const BYTE*)src;
+            U32 const baseline = baseValue[symbol];
+            U8 const nbBits = nbAdditionalBits[symbol];
+            ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits);
+        }
+        *DTablePtr = DTableSpace;
+        return 1;
+    case set_basic :
+        *DTablePtr = defaultTable;
+        return 0;
+    case set_repeat:
+        RETURN_ERROR_IF(!flagRepeatTable, corruption_detected, "");
+        /* prefetch FSE table if used */
+        if (ddictIsCold && (nbSeq > 24 /* heuristic */)) {
+            const void* const pStart = *DTablePtr;
+            size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog));
+            PREFETCH_AREA(pStart, pSize);
+        }
+        return 0;
+    case set_compressed :
+        {   unsigned tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
+            RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected, "");
+            RETURN_ERROR_IF(tableLog > maxLog, corruption_detected, "");
+            ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog, wksp, wkspSize, bmi2);
+            *DTablePtr = DTableSpace;
+            return headerSize;
+        }
+    default :
+        assert(0);
+        RETURN_ERROR(GENERIC, "impossible");
+    }
+}
+
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+    int nbSeq;
+    DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
+
+    /* check */
+    RETURN_ERROR_IF(srcSize < MIN_SEQUENCES_SIZE, srcSize_wrong, "");
+
+    /* SeqHead */
+    nbSeq = *ip++;
+    if (!nbSeq) {
+        *nbSeqPtr=0;
+        RETURN_ERROR_IF(srcSize != 1, srcSize_wrong, "");
+        return 1;
+    }
+    if (nbSeq > 0x7F) {
+        if (nbSeq == 0xFF) {
+            RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong, "");
+            nbSeq = MEM_readLE16(ip) + LONGNBSEQ;
+            ip+=2;
+        } else {
+            RETURN_ERROR_IF(ip >= iend, srcSize_wrong, "");
+            nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+        }
+    }
+    *nbSeqPtr = nbSeq;
+
+    /* FSE table descriptors */
+    RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong, ""); /* minimum possible size: 1 byte for symbol encoding types */
+    {   symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6);
+        symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3);
+        symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3);
+        ip++;
+
+        /* Build DTables */
+        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
+                                                      LLtype, MaxLL, LLFSELog,
+                                                      ip, iend-ip,
+                                                      LL_base, LL_bits,
+                                                      LL_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      ZSTD_DCtx_get_bmi2(dctx));
+            RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += llhSize;
+        }
+
+        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
+                                                      OFtype, MaxOff, OffFSELog,
+                                                      ip, iend-ip,
+                                                      OF_base, OF_bits,
+                                                      OF_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      ZSTD_DCtx_get_bmi2(dctx));
+            RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += ofhSize;
+        }
+
+        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
+                                                      MLtype, MaxML, MLFSELog,
+                                                      ip, iend-ip,
+                                                      ML_base, ML_bits,
+                                                      ML_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      ZSTD_DCtx_get_bmi2(dctx));
+            RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += mlhSize;
+        }
+    }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    size_t state;
+    const ZSTD_seqSymbol* table;
+} ZSTD_fseState;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    ZSTD_fseState stateLL;
+    ZSTD_fseState stateOffb;
+    ZSTD_fseState stateML;
+    size_t prevOffset[ZSTD_REP_NUM];
+} seqState_t;
+
+/*! ZSTD_overlapCopy8() :
+ *  Copies 8 bytes from ip to op and updates op and ip where ip <= op.
+ *  If the offset is < 8 then the offset is spread to at least 8 bytes.
+ *
+ *  Precondition: *ip <= *op
+ *  Postcondition: *op - *op >= 8
+ */
+HINT_INLINE void ZSTD_overlapCopy8(BYTE** op, BYTE const** ip, size_t offset) {
+    assert(*ip <= *op);
+    if (offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[offset];
+        (*op)[0] = (*ip)[0];
+        (*op)[1] = (*ip)[1];
+        (*op)[2] = (*ip)[2];
+        (*op)[3] = (*ip)[3];
+        *ip += dec32table[offset];
+        ZSTD_copy4(*op+4, *ip);
+        *ip -= sub2;
+    } else {
+        ZSTD_copy8(*op, *ip);
+    }
+    *ip += 8;
+    *op += 8;
+    assert(*op - *ip >= 8);
+}
+
+/*! ZSTD_safecopy() :
+ *  Specialized version of memcpy() that is allowed to READ up to WILDCOPY_OVERLENGTH past the input buffer
+ *  and write up to 16 bytes past oend_w (op >= oend_w is allowed).
+ *  This function is only called in the uncommon case where the sequence is near the end of the block. It
+ *  should be fast for a single long sequence, but can be slow for several short sequences.
+ *
+ *  @param ovtype controls the overlap detection
+ *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
+ *         - ZSTD_overlap_src_before_dst: The src and dst may overlap and may be any distance apart.
+ *           The src buffer must be before the dst buffer.
+ */
+static void ZSTD_safecopy(BYTE* op, const BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
+    ptrdiff_t const diff = op - ip;
+    BYTE* const oend = op + length;
+
+    assert((ovtype == ZSTD_no_overlap && (diff <= -8 || diff >= 8 || op >= oend_w)) ||
+           (ovtype == ZSTD_overlap_src_before_dst && diff >= 0));
+
+    if (length < 8) {
+        /* Handle short lengths. */
+        while (op < oend) *op++ = *ip++;
+        return;
+    }
+    if (ovtype == ZSTD_overlap_src_before_dst) {
+        /* Copy 8 bytes and ensure the offset >= 8 when there can be overlap. */
+        assert(length >= 8);
+        ZSTD_overlapCopy8(&op, &ip, diff);
+        length -= 8;
+        assert(op - ip >= 8);
+        assert(op <= oend);
+    }
+
+    if (oend <= oend_w) {
+        /* No risk of overwrite. */
+        ZSTD_wildcopy(op, ip, length, ovtype);
+        return;
+    }
+    if (op <= oend_w) {
+        /* Wildcopy until we get close to the end. */
+        assert(oend > oend_w);
+        ZSTD_wildcopy(op, ip, oend_w - op, ovtype);
+        ip += oend_w - op;
+        op += oend_w - op;
+    }
+    /* Handle the leftovers. */
+    while (op < oend) *op++ = *ip++;
+}
+
+/* ZSTD_safecopyDstBeforeSrc():
+ * This version allows overlap with dst before src, or handles the non-overlap case with dst after src
+ * Kept separate from more common ZSTD_safecopy case to avoid performance impact to the safecopy common case */
+static void ZSTD_safecopyDstBeforeSrc(BYTE* op, BYTE const* ip, ptrdiff_t length) {
+    ptrdiff_t const diff = op - ip;
+    BYTE* const oend = op + length;
+
+    if (length < 8 || diff > -8) {
+        /* Handle short lengths, close overlaps, and dst not before src. */
+        while (op < oend) *op++ = *ip++;
+        return;
+    }
+
+    if (op <= oend - WILDCOPY_OVERLENGTH && diff < -WILDCOPY_VECLEN) {
+        ZSTD_wildcopy(op, ip, oend - WILDCOPY_OVERLENGTH - op, ZSTD_no_overlap);
+        ip += oend - WILDCOPY_OVERLENGTH - op;
+        op += oend - WILDCOPY_OVERLENGTH - op;
+    }
+
+    /* Handle the leftovers. */
+    while (op < oend) *op++ = *ip++;
+}
+
+/* ZSTD_execSequenceEnd():
+ * This version handles cases that are near the end of the output buffer. It requires
+ * more careful checks to make sure there is no overflow. By separating out these hard
+ * and unlikely cases, we can speed up the common cases.
+ *
+ * NOTE: This function needs to be fast for a single long sequence, but doesn't need
+ * to be optimized for many small sequences, since those fall into ZSTD_execSequence().
+ */
+FORCE_NOINLINE
+size_t ZSTD_execSequenceEnd(BYTE* op,
+    BYTE* const oend, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+
+    /* bounds checks : careful of address space overflow in 32-bit mode */
+    RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
+    RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
+    assert(op < op + sequenceLength);
+    assert(oLitEnd < op + sequenceLength);
+
+    /* copy literals */
+    ZSTD_safecopy(op, oend_w, *litPtr, sequence.litLength, ZSTD_no_overlap);
+    op = oLitEnd;
+    *litPtr = iLitEnd;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
+        match = dictEnd - (prefixStart - match);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+        ZSTD_memmove(oLitEnd, match, length1);
+        op = oLitEnd + length1;
+        sequence.matchLength -= length1;
+        match = prefixStart;
+        }
+    }
+    ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
+    return sequenceLength;
+}
+
+/* ZSTD_execSequenceEndSplitLitBuffer():
+ * This version is intended to be used during instances where the litBuffer is still split.  It is kept separate to avoid performance impact for the good case.
+ */
+FORCE_NOINLINE
+size_t ZSTD_execSequenceEndSplitLitBuffer(BYTE* op,
+    BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+
+    /* bounds checks : careful of address space overflow in 32-bit mode */
+    RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
+    RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
+    assert(op < op + sequenceLength);
+    assert(oLitEnd < op + sequenceLength);
+
+    /* copy literals */
+    RETURN_ERROR_IF(op > *litPtr && op < *litPtr + sequence.litLength, dstSize_tooSmall, "output should not catch up to and overwrite literal buffer");
+    ZSTD_safecopyDstBeforeSrc(op, *litPtr, sequence.litLength);
+    op = oLitEnd;
+    *litPtr = iLitEnd;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
+        match = dictEnd - (prefixStart - match);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+        ZSTD_memmove(oLitEnd, match, length1);
+        op = oLitEnd + length1;
+        sequence.matchLength -= length1;
+        match = prefixStart;
+        }
+    }
+    ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
+    return sequenceLength;
+}
+
+HINT_INLINE
+size_t ZSTD_execSequence(BYTE* op,
+    BYTE* const oend, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;   /* risk : address space underflow on oend=NULL */
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    assert(op != NULL /* Precondition */);
+    assert(oend_w < oend /* No underflow */);
+
+#if defined(__aarch64__)
+    /* prefetch sequence starting from match that will be used for copy later */
+    PREFETCH_L1(match);
+#endif
+    /* Handle edge cases in a slow path:
+     *   - Read beyond end of literals
+     *   - Match end is within WILDCOPY_OVERLIMIT of oend
+     *   - 32-bit mode and the match length overflows
+     */
+    if (UNLIKELY(
+        iLitEnd > litLimit ||
+        oMatchEnd > oend_w ||
+        (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
+        return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+
+    /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
+    assert(op <= oLitEnd /* No overflow */);
+    assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
+    assert(oMatchEnd <= oend /* No underflow */);
+    assert(iLitEnd <= litLimit /* Literal length is in bounds */);
+    assert(oLitEnd <= oend_w /* Can wildcopy literals */);
+    assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
+
+    /* Copy Literals:
+     * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
+     * We likely don't need the full 32-byte wildcopy.
+     */
+    assert(WILDCOPY_OVERLENGTH >= 16);
+    ZSTD_copy16(op, (*litPtr));
+    if (UNLIKELY(sequence.litLength > 16)) {
+        ZSTD_wildcopy(op + 16, (*litPtr) + 16, sequence.litLength - 16, ZSTD_no_overlap);
+    }
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* Copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix -> go into extDict */
+        RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
+        match = dictEnd + (match - prefixStart);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+        ZSTD_memmove(oLitEnd, match, length1);
+        op = oLitEnd + length1;
+        sequence.matchLength -= length1;
+        match = prefixStart;
+        }
+    }
+    /* Match within prefix of 1 or more bytes */
+    assert(op <= oMatchEnd);
+    assert(oMatchEnd <= oend_w);
+    assert(match >= prefixStart);
+    assert(sequence.matchLength >= 1);
+
+    /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
+     * without overlap checking.
+     */
+    if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
+        /* We bet on a full wildcopy for matches, since we expect matches to be
+         * longer than literals (in general). In silesia, ~10% of matches are longer
+         * than 16 bytes.
+         */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
+        return sequenceLength;
+    }
+    assert(sequence.offset < WILDCOPY_VECLEN);
+
+    /* Copy 8 bytes and spread the offset to be >= 8. */
+    ZSTD_overlapCopy8(&op, &match, sequence.offset);
+
+    /* If the match length is > 8 bytes, then continue with the wildcopy. */
+    if (sequence.matchLength > 8) {
+        assert(op < oMatchEnd);
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength - 8, ZSTD_overlap_src_before_dst);
+    }
+    return sequenceLength;
+}
+
+HINT_INLINE
+size_t ZSTD_execSequenceSplitLitBuffer(BYTE* op,
+    BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    assert(op != NULL /* Precondition */);
+    assert(oend_w < oend /* No underflow */);
+    /* Handle edge cases in a slow path:
+     *   - Read beyond end of literals
+     *   - Match end is within WILDCOPY_OVERLIMIT of oend
+     *   - 32-bit mode and the match length overflows
+     */
+    if (UNLIKELY(
+            iLitEnd > litLimit ||
+            oMatchEnd > oend_w ||
+            (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
+        return ZSTD_execSequenceEndSplitLitBuffer(op, oend, oend_w, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+
+    /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
+    assert(op <= oLitEnd /* No overflow */);
+    assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
+    assert(oMatchEnd <= oend /* No underflow */);
+    assert(iLitEnd <= litLimit /* Literal length is in bounds */);
+    assert(oLitEnd <= oend_w /* Can wildcopy literals */);
+    assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
+
+    /* Copy Literals:
+     * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
+     * We likely don't need the full 32-byte wildcopy.
+     */
+    assert(WILDCOPY_OVERLENGTH >= 16);
+    ZSTD_copy16(op, (*litPtr));
+    if (UNLIKELY(sequence.litLength > 16)) {
+        ZSTD_wildcopy(op+16, (*litPtr)+16, sequence.litLength-16, ZSTD_no_overlap);
+    }
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* Copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix -> go into extDict */
+        RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
+        match = dictEnd + (match - prefixStart);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            ZSTD_memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+    }   }
+    /* Match within prefix of 1 or more bytes */
+    assert(op <= oMatchEnd);
+    assert(oMatchEnd <= oend_w);
+    assert(match >= prefixStart);
+    assert(sequence.matchLength >= 1);
+
+    /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
+     * without overlap checking.
+     */
+    if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
+        /* We bet on a full wildcopy for matches, since we expect matches to be
+         * longer than literals (in general). In silesia, ~10% of matches are longer
+         * than 16 bytes.
+         */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
+        return sequenceLength;
+    }
+    assert(sequence.offset < WILDCOPY_VECLEN);
+
+    /* Copy 8 bytes and spread the offset to be >= 8. */
+    ZSTD_overlapCopy8(&op, &match, sequence.offset);
+
+    /* If the match length is > 8 bytes, then continue with the wildcopy. */
+    if (sequence.matchLength > 8) {
+        assert(op < oMatchEnd);
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst);
+    }
+    return sequenceLength;
+}
+
+
+static void
+ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
+{
+    const void* ptr = dt;
+    const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits",
+                (U32)DStatePtr->state, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+FORCE_INLINE_TEMPLATE void
+ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16 nextState, U32 nbBits)
+{
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = nextState + lowBits;
+}
+
+/* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
+ * offset bits. But we can only read at most STREAM_ACCUMULATOR_MIN_32
+ * bits before reloading. This value is the maximum number of bytes we read
+ * after reloading when we are decoding long offsets.
+ */
+#define LONG_OFFSETS_MAX_EXTRA_BITS_32                       \
+    (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32       \
+        ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32  \
+        : 0)
+
+typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
+
+FORCE_INLINE_TEMPLATE seq_t
+ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
+{
+    seq_t seq;
+    /*
+     * ZSTD_seqSymbol is a structure with a total of 64 bits wide. So it can be
+     * loaded in one operation and extracted its fields by simply shifting or
+     * bit-extracting on aarch64.
+     * GCC doesn't recognize this and generates more unnecessary ldr/ldrb/ldrh
+     * operations that cause performance drop. This can be avoided by using this
+     * ZSTD_memcpy hack.
+     */
+#if defined(__aarch64__) && (defined(__GNUC__) && !defined(__clang__))
+    ZSTD_seqSymbol llDInfoS, mlDInfoS, ofDInfoS;
+    ZSTD_seqSymbol* const llDInfo = &llDInfoS;
+    ZSTD_seqSymbol* const mlDInfo = &mlDInfoS;
+    ZSTD_seqSymbol* const ofDInfo = &ofDInfoS;
+    ZSTD_memcpy(llDInfo, seqState->stateLL.table + seqState->stateLL.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(mlDInfo, seqState->stateML.table + seqState->stateML.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(ofDInfo, seqState->stateOffb.table + seqState->stateOffb.state, sizeof(ZSTD_seqSymbol));
+#else
+    const ZSTD_seqSymbol* const llDInfo = seqState->stateLL.table + seqState->stateLL.state;
+    const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
+    const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
+#endif
+    seq.matchLength = mlDInfo->baseValue;
+    seq.litLength = llDInfo->baseValue;
+    {   U32 const ofBase = ofDInfo->baseValue;
+        BYTE const llBits = llDInfo->nbAdditionalBits;
+        BYTE const mlBits = mlDInfo->nbAdditionalBits;
+        BYTE const ofBits = ofDInfo->nbAdditionalBits;
+        BYTE const totalBits = llBits+mlBits+ofBits;
+
+        U16 const llNext = llDInfo->nextState;
+        U16 const mlNext = mlDInfo->nextState;
+        U16 const ofNext = ofDInfo->nextState;
+        U32 const llnbBits = llDInfo->nbBits;
+        U32 const mlnbBits = mlDInfo->nbBits;
+        U32 const ofnbBits = ofDInfo->nbBits;
+
+        assert(llBits <= MaxLLBits);
+        assert(mlBits <= MaxMLBits);
+        assert(ofBits <= MaxOff);
+        /*
+         * As gcc has better branch and block analyzers, sometimes it is only
+         * valuable to mark likeliness for clang, it gives around 3-4% of
+         * performance.
+         */
+
+        /* sequence */
+        {   size_t offset;
+            if (ofBits > 1) {
+                ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+                ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 > LONG_OFFSETS_MAX_EXTRA_BITS_32);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 - LONG_OFFSETS_MAX_EXTRA_BITS_32 >= MaxMLBits);
+                if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
+                    /* Always read extra bits, this keeps the logic simple,
+                     * avoids branches, and avoids accidentally reading 0 bits.
+                     */
+                    U32 const extraBits = LONG_OFFSETS_MAX_EXTRA_BITS_32;
+                    offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                    BIT_reloadDStream(&seqState->DStream);
+                    offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+                } else {
+                    offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+                }
+                seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset;
+            } else {
+                U32 const ll0 = (llDInfo->baseValue == 0);
+                if (LIKELY((ofBits == 0))) {
+                    offset = seqState->prevOffset[ll0];
+                    seqState->prevOffset[1] = seqState->prevOffset[!ll0];
+                    seqState->prevOffset[0] = offset;
+                } else {
+                    offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
+                    {   size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                        temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                        if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                        seqState->prevOffset[1] = seqState->prevOffset[0];
+                        seqState->prevOffset[0] = offset = temp;
+            }   }   }
+            seq.offset = offset;
+        }
+
+        if (mlBits > 0)
+            seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
+
+        if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+            BIT_reloadDStream(&seqState->DStream);
+        if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+            BIT_reloadDStream(&seqState->DStream);
+        /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
+        ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+        if (llBits > 0)
+            seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
+
+        if (MEM_32bits())
+            BIT_reloadDStream(&seqState->DStream);
+
+        DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
+                    (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+
+        ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits);    /* <=  9 bits */
+        ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits);    /* <=  9 bits */
+        if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+        ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits);  /* <=  8 bits */
+    }
+
+    return seq;
+}
+
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+MEM_STATIC int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
+{
+    size_t const windowSize = dctx->fParams.windowSize;
+    /* No dictionary used. */
+    if (dctx->dictContentEndForFuzzing == NULL) return 0;
+    /* Dictionary is our prefix. */
+    if (prefixStart == dctx->dictContentBeginForFuzzing) return 1;
+    /* Dictionary is not our ext-dict. */
+    if (dctx->dictEnd != dctx->dictContentEndForFuzzing) return 0;
+    /* Dictionary is not within our window size. */
+    if ((size_t)(oLitEnd - prefixStart) >= windowSize) return 0;
+    /* Dictionary is active. */
+    return 1;
+}
+
+MEM_STATIC void ZSTD_assertValidSequence(
+        ZSTD_DCtx const* dctx,
+        BYTE const* op, BYTE const* oend,
+        seq_t const seq,
+        BYTE const* prefixStart, BYTE const* virtualStart)
+{
+#if DEBUGLEVEL >= 1
+    size_t const windowSize = dctx->fParams.windowSize;
+    size_t const sequenceSize = seq.litLength + seq.matchLength;
+    BYTE const* const oLitEnd = op + seq.litLength;
+    DEBUGLOG(6, "Checking sequence: litL=%u matchL=%u offset=%u",
+            (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+    assert(op <= oend);
+    assert((size_t)(oend - op) >= sequenceSize);
+    assert(sequenceSize <= ZSTD_BLOCKSIZE_MAX);
+    if (ZSTD_dictionaryIsActive(dctx, prefixStart, oLitEnd)) {
+        size_t const dictSize = (size_t)((char const*)dctx->dictContentEndForFuzzing - (char const*)dctx->dictContentBeginForFuzzing);
+        /* Offset must be within the dictionary. */
+        assert(seq.offset <= (size_t)(oLitEnd - virtualStart));
+        assert(seq.offset <= windowSize + dictSize);
+    } else {
+        /* Offset must be within our window. */
+        assert(seq.offset <= windowSize);
+    }
+#else
+    (void)dctx, (void)op, (void)oend, (void)seq, (void)prefixStart, (void)virtualStart;
+#endif
+}
+#endif
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+
+
+FORCE_INLINE_TEMPLATE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset,
+                         const int frame)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* litBufferEnd = dctx->litBufferEnd;
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer");
+    (void)frame;
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        assert(dst != NULL);
+
+        ZSTD_STATIC_ASSERT(
+                BIT_DStream_unfinished < BIT_DStream_completed &&
+                BIT_DStream_endOfBuffer < BIT_DStream_completed &&
+                BIT_DStream_completed < BIT_DStream_overflow);
+
+        /* decompress without overrunning litPtr begins */
+        {
+            seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            /* Align the decompression loop to 32 + 16 bytes.
+                *
+                * zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
+                * speed swings based on the alignment of the decompression loop. This
+                * performance swing is caused by parts of the decompression loop falling
+                * out of the DSB. The entire decompression loop should fit in the DSB,
+                * when it can't we get much worse performance. You can measure if you've
+                * hit the good case or the bad case with this perf command for some
+                * compressed file test.zst:
+                *
+                *   perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
+                *             -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
+                *
+                * If you see most cycles served out of the MITE you've hit the bad case.
+                * If you see most cycles served out of the DSB you've hit the good case.
+                * If it is pretty even then you may be in an okay case.
+                *
+                * This issue has been reproduced on the following CPUs:
+                *   - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
+                *               Use Instruments->Counters to get DSB/MITE cycles.
+                *               I never got performance swings, but I was able to
+                *               go from the good case of mostly DSB to half of the
+                *               cycles served from MITE.
+                *   - Coffeelake: Intel i9-9900k
+                *   - Coffeelake: Intel i7-9700k
+                *
+                * I haven't been able to reproduce the instability or DSB misses on any
+                * of the following CPUS:
+                *   - Haswell
+                *   - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
+                *   - Skylake
+                *
+                * Alignment is done for each of the three major decompression loops:
+                *   - ZSTD_decompressSequences_bodySplitLitBuffer - presplit section of the literal buffer
+                *   - ZSTD_decompressSequences_bodySplitLitBuffer - postsplit section of the literal buffer
+                *   - ZSTD_decompressSequences_body
+                * Alignment choices are made to minimize large swings on bad cases and influence on performance
+                * from changes external to this code, rather than to overoptimize on the current commit.
+                *
+                * If you are seeing performance stability this script can help test.
+                * It tests on 4 commits in zstd where I saw performance change.
+                *
+                *   https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
+                */
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+#  if __GNUC__ >= 7
+	    /* good for gcc-7, gcc-9, and gcc-11 */
+            __asm__("nop");
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 4");
+#    if __GNUC__ == 8 || __GNUC__ == 10
+	    /* good for gcc-8 and gcc-10 */
+            __asm__("nop");
+            __asm__(".p2align 3");
+#    endif
+#  endif
+#endif
+
+            /* Handle the initial state where litBuffer is currently split between dst and litExtraBuffer */
+            for (; litPtr + sequence.litLength <= dctx->litBufferEnd; ) {
+                size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                    return oneSeqSize;
+                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                op += oneSeqSize;
+                if (UNLIKELY(!--nbSeq))
+                    break;
+                BIT_reloadDStream(&(seqState.DStream));
+                sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            }
+
+            /* If there are more sequences, they will need to read literals from litExtraBuffer; copy over the remainder from dst and update litPtr and litEnd */
+            if (nbSeq > 0) {
+                const size_t leftoverLit = dctx->litBufferEnd - litPtr;
+                if (leftoverLit)
+                {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequence.litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                {
+                    size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                    assert(!ZSTD_isError(oneSeqSize));
+                    if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                    if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                        return oneSeqSize;
+                    DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                    op += oneSeqSize;
+                    if (--nbSeq)
+                        BIT_reloadDStream(&(seqState.DStream));
+                }
+            }
+        }
+
+        if (nbSeq > 0) /* there is remaining lit from extra buffer */
+        {
+
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+            __asm__("nop");
+#  if __GNUC__ != 7
+            /* worse for gcc-7 better for gcc-8, gcc-9, and gcc-10 and clang */
+            __asm__(".p2align 4");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  elif __GNUC__ >= 11
+            __asm__(".p2align 3");
+#  else
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  endif
+#endif
+
+            for (; ; ) {
+                seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                    return oneSeqSize;
+                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                op += oneSeqSize;
+                if (UNLIKELY(!--nbSeq))
+                    break;
+                BIT_reloadDStream(&(seqState.DStream));
+            }
+        }
+
+        /* check if reached exact end */
+        DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer: after decode loop, remaining nbSeq : %i", nbSeq);
+        RETURN_ERROR_IF(nbSeq, corruption_detected, "");
+        RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    if (dctx->litBufferLocation == ZSTD_split)  /* split hasn't been reached yet, first get dst then copy litExtraBuffer */
+    {
+        size_t const lastLLSize = litBufferEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+        litPtr = dctx->litExtraBuffer;
+        litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+        dctx->litBufferLocation = ZSTD_not_in_dst;
+    }
+    {   size_t const lastLLSize = litBufferEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
+    void* dst, size_t maxDstSize,
+    const void* seqStart, size_t seqSize, int nbSeq,
+    const ZSTD_longOffset_e isLongOffset,
+    const int frame)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = dctx->litBufferLocation == ZSTD_not_in_dst ? ostart + maxDstSize : dctx->litBuffer;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const prefixStart = (const BYTE*)(dctx->prefixStart);
+    const BYTE* const vBase = (const BYTE*)(dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*)(dctx->dictEnd);
+    DEBUGLOG(5, "ZSTD_decompressSequences_body: nbSeq = %d", nbSeq);
+    (void)frame;
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i = 0; i < ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend - ip)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        assert(dst != NULL);
+
+        ZSTD_STATIC_ASSERT(
+            BIT_DStream_unfinished < BIT_DStream_completed &&
+            BIT_DStream_endOfBuffer < BIT_DStream_completed &&
+            BIT_DStream_completed < BIT_DStream_overflow);
+
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+            __asm__("nop");
+#  if __GNUC__ >= 7
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  else
+            __asm__(".p2align 4");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  endif
+#endif
+
+        for ( ; ; ) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+            assert(!ZSTD_isError(oneSeqSize));
+            if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+            if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                return oneSeqSize;
+            DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+            op += oneSeqSize;
+            if (UNLIKELY(!--nbSeq))
+                break;
+            BIT_reloadDStream(&(seqState.DStream));
+        }
+
+        /* check if reached exact end */
+        DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq);
+        RETURN_ERROR_IF(nbSeq, corruption_detected, "");
+        RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+static size_t
+ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+
+static size_t
+ZSTD_decompressSequencesSplitLitBuffer_default(ZSTD_DCtx* dctx,
+                                               void* dst, size_t maxDstSize,
+                                         const void* seqStart, size_t seqSize, int nbSeq,
+                                         const ZSTD_longOffset_e isLongOffset,
+                                         const int frame)
+{
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_prefetchMatch(size_t prefetchPos, seq_t const sequence,
+                   const BYTE* const prefixStart, const BYTE* const dictEnd)
+{
+    prefetchPos += sequence.litLength;
+    {   const BYTE* const matchBase = (sequence.offset > prefetchPos) ? dictEnd : prefixStart;
+        const BYTE* const match = matchBase + prefetchPos - sequence.offset; /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
+                                                                              * No consequence though : memory address is only used for prefetching, not for dereferencing */
+        PREFETCH_L1(match); PREFETCH_L1(match+CACHELINE_SIZE);   /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
+    }
+    return prefetchPos + sequence.matchLength;
+}
+
+/* This decoding function employs prefetching
+ * to reduce latency impact of cache misses.
+ * It's generally employed when block contains a significant portion of long-distance matches
+ * or when coupled with a "cold" dictionary */
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_decompressSequencesLong_body(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset,
+                         const int frame)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = dctx->litBufferLocation == ZSTD_in_dst ? dctx->litBuffer : ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* litBufferEnd = dctx->litBufferEnd;
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    (void)frame;
+
+    /* Regen sequences */
+    if (nbSeq) {
+#define STORED_SEQS 8
+#define STORED_SEQS_MASK (STORED_SEQS-1)
+#define ADVANCED_SEQS STORED_SEQS
+        seq_t sequences[STORED_SEQS];
+        int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
+        seqState_t seqState;
+        int seqNb;
+        size_t prefetchPos = (size_t)(op-prefixStart); /* track position relative to prefixStart */
+
+        dctx->fseEntropy = 1;
+        { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        assert(dst != NULL);
+        assert(iend >= ip);
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+
+        /* prepare in advance */
+        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+            sequences[seqNb] = sequence;
+        }
+        RETURN_ERROR_IF(seqNb<seqAdvance, corruption_detected, "");
+
+        /* decompress without stomping litBuffer */
+        for (; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb < nbSeq); seqNb++) {
+            seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            size_t oneSeqSize;
+
+            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength > dctx->litBufferEnd)
+            {
+                /* lit buffer is reaching split point, empty out the first buffer and transition to litExtraBuffer */
+                const size_t leftoverLit = dctx->litBufferEnd - litPtr;
+                if (leftoverLit)
+                {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+
+                prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+                sequences[seqNb & STORED_SEQS_MASK] = sequence;
+                op += oneSeqSize;
+            }
+            else
+            {
+                /* lit buffer is either wholly contained in first or second split, or not split at all*/
+                oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
+                    ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength - WILDCOPY_OVERLENGTH, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
+                    ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+
+                prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+                sequences[seqNb & STORED_SEQS_MASK] = sequence;
+                op += oneSeqSize;
+            }
+        }
+        RETURN_ERROR_IF(seqNb<nbSeq, corruption_detected, "");
+
+        /* finish queue */
+        seqNb -= seqAdvance;
+        for ( ; seqNb<nbSeq ; seqNb++) {
+            seq_t *sequence = &(sequences[seqNb&STORED_SEQS_MASK]);
+            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequence->litLength > dctx->litBufferEnd)
+            {
+                const size_t leftoverLit = dctx->litBufferEnd - litPtr;
+                if (leftoverLit)
+                {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequence->litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                {
+                    size_t const oneSeqSize = ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                    assert(!ZSTD_isError(oneSeqSize));
+                    if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                    if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                    op += oneSeqSize;
+                }
+            }
+            else
+            {
+                size_t const oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
+                    ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence->litLength - WILDCOPY_OVERLENGTH, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
+                    ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+            }
+        }
+
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    if (dctx->litBufferLocation == ZSTD_split)  /* first deplete literal buffer in dst, then copy litExtraBuffer */
+    {
+        size_t const lastLLSize = litBufferEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+        litPtr = dctx->litExtraBuffer;
+        litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+    }
+    {   size_t const lastLLSize = litBufferEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+static size_t
+ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+
+
+#if DYNAMIC_BMI2
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+static BMI2_TARGET_ATTRIBUTE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+static BMI2_TARGET_ATTRIBUTE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequencesSplitLitBuffer_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+static BMI2_TARGET_ATTRIBUTE size_t
+ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+#endif /* DYNAMIC_BMI2 */
+
+typedef size_t (*ZSTD_decompressSequences_t)(
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t maxDstSize,
+                            const void* seqStart, size_t seqSize, int nbSeq,
+                            const ZSTD_longOffset_e isLongOffset,
+                            const int frame);
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+static size_t
+ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                   const void* seqStart, size_t seqSize, int nbSeq,
+                   const ZSTD_longOffset_e isLongOffset,
+                   const int frame)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequences");
+#if DYNAMIC_BMI2
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
+        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    }
+#endif
+    return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+static size_t
+ZSTD_decompressSequencesSplitLitBuffer(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                                 const void* seqStart, size_t seqSize, int nbSeq,
+                                 const ZSTD_longOffset_e isLongOffset,
+                                 const int frame)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesSplitLitBuffer");
+#if DYNAMIC_BMI2
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
+        return ZSTD_decompressSequencesSplitLitBuffer_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    }
+#endif
+    return ZSTD_decompressSequencesSplitLitBuffer_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+/* ZSTD_decompressSequencesLong() :
+ * decompression function triggered when a minimum share of offsets is considered "long",
+ * aka out of cache.
+ * note : "long" definition seems overloaded here, sometimes meaning "wider than bitstream register", and sometimes meaning "farther than memory cache distance".
+ * This function will try to mitigate main memory latency through the use of prefetching */
+static size_t
+ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
+                             void* dst, size_t maxDstSize,
+                             const void* seqStart, size_t seqSize, int nbSeq,
+                             const ZSTD_longOffset_e isLongOffset,
+                             const int frame)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesLong");
+#if DYNAMIC_BMI2
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
+        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    }
+#endif
+  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+
+/**
+ * @returns The total size of the history referenceable by zstd, including
+ * both the prefix and the extDict. At @p op any offset larger than this
+ * is invalid.
+ */
+static size_t ZSTD_totalHistorySize(BYTE* op, BYTE const* virtualStart)
+{
+    return (size_t)(op - virtualStart);
+}
+
+typedef struct {
+    unsigned longOffsetShare;
+    unsigned maxNbAdditionalBits;
+} ZSTD_OffsetInfo;
+
+/* ZSTD_getOffsetInfo() :
+ * condition : offTable must be valid
+ * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
+ *           compared to maximum possible of (1<<OffFSELog),
+ *           as well as the maximum number additional bits required.
+ */
+static ZSTD_OffsetInfo
+ZSTD_getOffsetInfo(const ZSTD_seqSymbol* offTable, int nbSeq)
+{
+    ZSTD_OffsetInfo info = {0, 0};
+    /* If nbSeq == 0, then the offTable is uninitialized, but we have
+     * no sequences, so both values should be 0.
+     */
+    if (nbSeq != 0) {
+        const void* ptr = offTable;
+        U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
+        const ZSTD_seqSymbol* table = offTable + 1;
+        U32 const max = 1 << tableLog;
+        U32 u;
+        DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
+
+        assert(max <= (1 << OffFSELog));  /* max not too large */
+        for (u=0; u<max; u++) {
+            info.maxNbAdditionalBits = MAX(info.maxNbAdditionalBits, table[u].nbAdditionalBits);
+            if (table[u].nbAdditionalBits > 22) info.longOffsetShare += 1;
+        }
+
+        assert(tableLog <= OffFSELog);
+        info.longOffsetShare <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
+    }
+
+    return info;
+}
+
+/**
+ * @returns The maximum offset we can decode in one read of our bitstream, without
+ * reloading more bits in the middle of the offset bits read. Any offsets larger
+ * than this must use the long offset decoder.
+ */
+static size_t ZSTD_maxShortOffset(void)
+{
+    if (MEM_64bits()) {
+        /* We can decode any offset without reloading bits.
+         * This might change if the max window size grows.
+         */
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        return (size_t)-1;
+    } else {
+        /* The maximum offBase is (1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1.
+         * This offBase would require STREAM_ACCUMULATOR_MIN extra bits.
+         * Then we have to subtract ZSTD_REP_NUM to get the maximum possible offset.
+         */
+        size_t const maxOffbase = ((size_t)1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1;
+        size_t const maxOffset = maxOffbase - ZSTD_REP_NUM;
+        assert(ZSTD_highbit32((U32)maxOffbase) == STREAM_ACCUMULATOR_MIN);
+        return maxOffset;
+    }
+}
+
+size_t
+ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize, const int frame, const streaming_operation streaming)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
+
+    /* Note : the wording of the specification
+     * allows compressed block to be sized exactly ZSTD_BLOCKSIZE_MAX.
+     * This generally does not happen, as it makes little sense,
+     * since an uncompressed block would feature same size and have no decompression cost.
+     * Also, note that decoder from reference libzstd before < v1.5.4
+     * would consider this edge case as an error.
+     * As a consequence, avoid generating compressed blocks of size ZSTD_BLOCKSIZE_MAX
+     * for broader compatibility with the deployed ecosystem of zstd decoders */
+    RETURN_ERROR_IF(srcSize > ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
+
+    /* Decode literals section */
+    {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, streaming);
+        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : cSize=%u, nbLiterals=%zu", (U32)litCSize, dctx->litSize);
+        if (ZSTD_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+
+    /* Build Decoding Tables */
+    {
+        /* Compute the maximum block size, which must also work when !frame and fParams are unset.
+         * Additionally, take the min with dstCapacity to ensure that the totalHistorySize fits in a size_t.
+         */
+        size_t const blockSizeMax = MIN(dstCapacity, (frame ? dctx->fParams.blockSizeMax : ZSTD_BLOCKSIZE_MAX));
+        size_t const totalHistorySize = ZSTD_totalHistorySize((BYTE*)dst + blockSizeMax, (BYTE const*)dctx->virtualStart);
+        /* isLongOffset must be true if there are long offsets.
+         * Offsets are long if they are larger than ZSTD_maxShortOffset().
+         * We don't expect that to be the case in 64-bit mode.
+         *
+         * We check here to see if our history is large enough to allow long offsets.
+         * If it isn't, then we can't possible have (valid) long offsets. If the offset
+         * is invalid, then it is okay to read it incorrectly.
+         *
+         * If isLongOffsets is true, then we will later check our decoding table to see
+         * if it is even possible to generate long offsets.
+         */
+        ZSTD_longOffset_e isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (totalHistorySize > ZSTD_maxShortOffset()));
+        /* These macros control at build-time which decompressor implementation
+         * we use. If neither is defined, we do some inspection and dispatch at
+         * runtime.
+         */
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+        int usePrefetchDecoder = dctx->ddictIsCold;
+#else
+        /* Set to 1 to avoid computing offset info if we don't need to.
+         * Otherwise this value is ignored.
+         */
+        int usePrefetchDecoder = 1;
+#endif
+        int nbSeq;
+        size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
+        if (ZSTD_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        srcSize -= seqHSize;
+
+        RETURN_ERROR_IF((dst == NULL || dstCapacity == 0) && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
+        RETURN_ERROR_IF(MEM_64bits() && sizeof(size_t) == sizeof(void*) && (size_t)(-1) - (size_t)dst < (size_t)(1 << 20), dstSize_tooSmall,
+                "invalid dst");
+
+        /* If we could potentially have long offsets, or we might want to use the prefetch decoder,
+         * compute information about the share of long offsets, and the maximum nbAdditionalBits.
+         * NOTE: could probably use a larger nbSeq limit
+         */
+        if (isLongOffset || (!usePrefetchDecoder && (totalHistorySize > (1u << 24)) && (nbSeq > 8))) {
+            ZSTD_OffsetInfo const info = ZSTD_getOffsetInfo(dctx->OFTptr, nbSeq);
+            if (isLongOffset && info.maxNbAdditionalBits <= STREAM_ACCUMULATOR_MIN) {
+                /* If isLongOffset, but the maximum number of additional bits that we see in our table is small
+                 * enough, then we know it is impossible to have too long an offset in this block, so we can
+                 * use the regular offset decoder.
+                 */
+                isLongOffset = ZSTD_lo_isRegularOffset;
+            }
+            if (!usePrefetchDecoder) {
+                U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
+                usePrefetchDecoder = (info.longOffsetShare >= minShare);
+            }
+        }
+
+        dctx->ddictIsCold = 0;
+
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+        if (usePrefetchDecoder) {
+#else
+        (void)usePrefetchDecoder;
+        {
+#endif
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+            return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+#endif
+        }
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+        /* else */
+        if (dctx->litBufferLocation == ZSTD_split)
+            return ZSTD_decompressSequencesSplitLitBuffer(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+        else
+            return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+#endif
+    }
+}
+
+
+void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize)
+{
+    if (dst != dctx->previousDstEnd && dstSize > 0) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->virtualStart = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+        dctx->prefixStart = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+size_t ZSTD_decompressBlock_deprecated(ZSTD_DCtx* dctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    size_t dSize;
+    ZSTD_checkContinuity(dctx, dst, dstCapacity);
+    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0, not_streaming);
+    dctx->previousDstEnd = (char*)dst + dSize;
+    return dSize;
+}
+
+
+/* NOTE: Must just wrap ZSTD_decompressBlock_deprecated() */
+size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    return ZSTD_decompressBlock_deprecated(dctx, dst, dstCapacity, src, srcSize);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_block.h b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_block.h
new file mode 100644
index 0000000..9d13188
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_block.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#ifndef ZSTD_DEC_BLOCK_H
+#define ZSTD_DEC_BLOCK_H
+
+/*-*******************************************************
+ *  Dependencies
+ *********************************************************/
+#include "../common/zstd_deps.h"   /* size_t */
+#include "../zstd.h"    /* DCtx, and some public functions */
+#include "../common/zstd_internal.h"  /* blockProperties_t, and some public functions */
+#include "zstd_decompress_internal.h"  /* ZSTD_seqSymbol */
+
+
+/* ===   Prototypes   === */
+
+/* note: prototypes already published within `zstd.h` :
+ * ZSTD_decompressBlock()
+ */
+
+/* note: prototypes already published within `zstd_internal.h` :
+ * ZSTD_getcBlockSize()
+ * ZSTD_decodeSeqHeaders()
+ */
+
+
+ /* Streaming state is used to inform allocation of the literal buffer */
+typedef enum {
+    not_streaming = 0,
+    is_streaming = 1
+} streaming_operation;
+
+/* ZSTD_decompressBlock_internal() :
+ * decompress block, starting at `src`,
+ * into destination buffer `dst`.
+ * @return : decompressed block size,
+ *           or an error code (which can be tested using ZSTD_isError())
+ */
+size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize, const int frame, const streaming_operation streaming);
+
+/* ZSTD_buildFSETable() :
+ * generate FSE decoding table for one symbol (ll, ml or off)
+ * this function must be called with valid parameters only
+ * (dt is large enough, normalizedCounter distribution total is a power of 2, max is within range, etc.)
+ * in which case it cannot fail.
+ * The workspace must be 4-byte aligned and at least ZSTD_BUILD_FSE_TABLE_WKSP_SIZE bytes, which is
+ * defined in zstd_decompress_internal.h.
+ * Internal use only.
+ */
+void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+             const short* normalizedCounter, unsigned maxSymbolValue,
+             const U32* baseValue, const U8* nbAdditionalBits,
+                   unsigned tableLog, void* wksp, size_t wkspSize,
+                   int bmi2);
+
+/* Internal definition of ZSTD_decompressBlock() to avoid deprecation warnings. */
+size_t ZSTD_decompressBlock_deprecated(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize);
+
+
+#endif /* ZSTD_DEC_BLOCK_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_internal.h b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_internal.h
new file mode 100644
index 0000000..c2ec5d9
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/decompress/zstd_decompress_internal.h
@@ -0,0 +1,238 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* zstd_decompress_internal:
+ * objects and definitions shared within lib/decompress modules */
+
+ #ifndef ZSTD_DECOMPRESS_INTERNAL_H
+ #define ZSTD_DECOMPRESS_INTERNAL_H
+
+
+/*-*******************************************************
+ *  Dependencies
+ *********************************************************/
+#include "../common/mem.h"             /* BYTE, U16, U32 */
+#include "../common/zstd_internal.h"   /* constants : MaxLL, MaxML, MaxOff, LLFSELog, etc. */
+
+
+
+/*-*******************************************************
+ *  Constants
+ *********************************************************/
+static UNUSED_ATTR const U32 LL_base[MaxLL+1] = {
+                 0,    1,    2,     3,     4,     5,     6,      7,
+                 8,    9,   10,    11,    12,    13,    14,     15,
+                16,   18,   20,    22,    24,    28,    32,     40,
+                48,   64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                0x2000, 0x4000, 0x8000, 0x10000 };
+
+static UNUSED_ATTR const U32 OF_base[MaxOff+1] = {
+                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD, 0x1FFFFFFD, 0x3FFFFFFD, 0x7FFFFFFD };
+
+static UNUSED_ATTR const U8 OF_bits[MaxOff+1] = {
+                     0,  1,  2,  3,  4,  5,  6,  7,
+                     8,  9, 10, 11, 12, 13, 14, 15,
+                    16, 17, 18, 19, 20, 21, 22, 23,
+                    24, 25, 26, 27, 28, 29, 30, 31 };
+
+static UNUSED_ATTR const U32 ML_base[MaxML+1] = {
+                     3,  4,  5,    6,     7,     8,     9,    10,
+                    11, 12, 13,   14,    15,    16,    17,    18,
+                    19, 20, 21,   22,    23,    24,    25,    26,
+                    27, 28, 29,   30,    31,    32,    33,    34,
+                    35, 37, 39,   41,    43,    47,    51,    59,
+                    67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                    0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+
+/*-*******************************************************
+ *  Decompression types
+ *********************************************************/
+ typedef struct {
+     U32 fastMode;
+     U32 tableLog;
+ } ZSTD_seqSymbol_header;
+
+ typedef struct {
+     U16  nextState;
+     BYTE nbAdditionalBits;
+     BYTE nbBits;
+     U32  baseValue;
+ } ZSTD_seqSymbol;
+
+ #define SEQSYMBOL_TABLE_SIZE(log)   (1 + (1 << (log)))
+
+#define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE (sizeof(S16) * (MaxSeq + 1) + (1u << MaxFSELog) + sizeof(U64))
+#define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32 ((ZSTD_BUILD_FSE_TABLE_WKSP_SIZE + sizeof(U32) - 1) / sizeof(U32))
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
+
+typedef struct {
+    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];    /* Note : Space reserved for FSE Tables */
+    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];   /* is also used as temporary workspace while building hufTable during DDict creation */
+    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];    /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
+    HUF_DTable hufTable[HUF_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];  /* can accommodate HUF_decompress4X */
+    U32 rep[ZSTD_REP_NUM];
+    U32 workspace[ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32];
+} ZSTD_entropyDTables_t;
+
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock,
+               ZSTDds_decompressLastBlock, ZSTDds_checkChecksum,
+               ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTD_dStage;
+
+typedef enum { zdss_init=0, zdss_loadHeader,
+               zdss_read, zdss_load, zdss_flush } ZSTD_dStreamStage;
+
+typedef enum {
+    ZSTD_use_indefinitely = -1,  /* Use the dictionary indefinitely */
+    ZSTD_dont_use = 0,           /* Do not use the dictionary (if one exists free it) */
+    ZSTD_use_once = 1            /* Use the dictionary once and set to ZSTD_dont_use */
+} ZSTD_dictUses_e;
+
+/* Hashset for storing references to multiple ZSTD_DDict within ZSTD_DCtx */
+typedef struct {
+    const ZSTD_DDict** ddictPtrTable;
+    size_t ddictPtrTableSize;
+    size_t ddictPtrCount;
+} ZSTD_DDictHashSet;
+
+#ifndef ZSTD_DECODER_INTERNAL_BUFFER
+#  define ZSTD_DECODER_INTERNAL_BUFFER  (1 << 16)
+#endif
+
+#define ZSTD_LBMIN 64
+#define ZSTD_LBMAX (128 << 10)
+
+/* extra buffer, compensates when dst is not large enough to store litBuffer */
+#define ZSTD_LITBUFFEREXTRASIZE  BOUNDED(ZSTD_LBMIN, ZSTD_DECODER_INTERNAL_BUFFER, ZSTD_LBMAX)
+
+typedef enum {
+    ZSTD_not_in_dst = 0,  /* Stored entirely within litExtraBuffer */
+    ZSTD_in_dst = 1,           /* Stored entirely within dst (in memory after current output write) */
+    ZSTD_split = 2            /* Split between litExtraBuffer and dst */
+} ZSTD_litLocation_e;
+
+struct ZSTD_DCtx_s
+{
+    const ZSTD_seqSymbol* LLTptr;
+    const ZSTD_seqSymbol* MLTptr;
+    const ZSTD_seqSymbol* OFTptr;
+    const HUF_DTable* HUFptr;
+    ZSTD_entropyDTables_t entropy;
+    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];   /* space needed when building huffman tables */
+    const void* previousDstEnd;   /* detect continuity */
+    const void* prefixStart;      /* start of current segment */
+    const void* virtualStart;     /* virtual start of previous segment if it was just before current one */
+    const void* dictEnd;          /* end of previous segment */
+    size_t expected;
+    ZSTD_frameHeader fParams;
+    U64 processedCSize;
+    U64 decodedSize;
+    blockType_e bType;            /* used in ZSTD_decompressContinue(), store blockType between block header decoding and block decompression stages */
+    ZSTD_dStage stage;
+    U32 litEntropy;
+    U32 fseEntropy;
+    XXH64_state_t xxhState;
+    size_t headerSize;
+    ZSTD_format_e format;
+    ZSTD_forceIgnoreChecksum_e forceIgnoreChecksum;   /* User specified: if == 1, will ignore checksums in compressed frame. Default == 0 */
+    U32 validateChecksum;         /* if == 1, will validate checksum. Is == 1 if (fParams.checksumFlag == 1) and (forceIgnoreChecksum == 0). */
+    const BYTE* litPtr;
+    ZSTD_customMem customMem;
+    size_t litSize;
+    size_t rleSize;
+    size_t staticSize;
+#if DYNAMIC_BMI2 != 0
+    int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+#endif
+
+    /* dictionary */
+    ZSTD_DDict* ddictLocal;
+    const ZSTD_DDict* ddict;     /* set by ZSTD_initDStream_usingDDict(), or ZSTD_DCtx_refDDict() */
+    U32 dictID;
+    int ddictIsCold;             /* if == 1 : dictionary is "new" for working context, and presumed "cold" (not in cpu cache) */
+    ZSTD_dictUses_e dictUses;
+    ZSTD_DDictHashSet* ddictSet;                    /* Hash set for multiple ddicts */
+    ZSTD_refMultipleDDicts_e refMultipleDDicts;     /* User specified: if == 1, will allow references to multiple DDicts. Default == 0 (disabled) */
+    int disableHufAsm;
+
+    /* streaming */
+    ZSTD_dStreamStage streamStage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    size_t maxWindowSize;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t lhSize;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+    void* legacyContext;
+    U32 previousLegacyVersion;
+    U32 legacyVersion;
+#endif
+    U32 hostageByte;
+    int noForwardProgress;
+    ZSTD_bufferMode_e outBufferMode;
+    ZSTD_outBuffer expectedOutBuffer;
+
+    /* workspace */
+    BYTE* litBuffer;
+    const BYTE* litBufferEnd;
+    ZSTD_litLocation_e litBufferLocation;
+    BYTE litExtraBuffer[ZSTD_LITBUFFEREXTRASIZE + WILDCOPY_OVERLENGTH]; /* literal buffer can be split between storage within dst and within this scratch buffer */
+    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
+
+    size_t oversizedDuration;
+
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    void const* dictContentBeginForFuzzing;
+    void const* dictContentEndForFuzzing;
+#endif
+
+    /* Tracing */
+#if ZSTD_TRACE
+    ZSTD_TraceCtx traceCtx;
+#endif
+};  /* typedef'd to ZSTD_DCtx within "zstd.h" */
+
+MEM_STATIC int ZSTD_DCtx_get_bmi2(const struct ZSTD_DCtx_s *dctx) {
+#if DYNAMIC_BMI2 != 0
+	return dctx->bmi2;
+#else
+    (void)dctx;
+	return 0;
+#endif
+}
+
+/*-*******************************************************
+ *  Shared internal functions
+ *********************************************************/
+
+/*! ZSTD_loadDEntropy() :
+ *  dict : must point at beginning of a valid zstd dictionary.
+ * @return : size of dictionary header (size of magic number + dict ID + entropy tables) */
+size_t ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
+                   const void* const dict, size_t const dictSize);
+
+/*! ZSTD_checkContinuity() :
+ *  check if next `dst` follows previous position, where decompression ended.
+ *  If yes, do nothing (continue on current segment).
+ *  If not, classify previous segment as "external dictionary", and start a new segment.
+ *  This function cannot fail. */
+void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize);
+
+
+#endif /* ZSTD_DECOMPRESS_INTERNAL_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/cover.c b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/cover.c
new file mode 100644
index 0000000..9e5e7d5
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/cover.c
@@ -0,0 +1,1257 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* *****************************************************************************
+ * Constructs a dictionary using a heuristic based on the following paper:
+ *
+ * Liao, Petri, Moffat, Wirth
+ * Effective Construction of Relative Lempel-Ziv Dictionaries
+ * Published in WWW 2016.
+ *
+ * Adapted from code originally written by @ot (Giuseppe Ottaviano).
+ ******************************************************************************/
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#  define ZDICT_STATIC_LINKING_ONLY
+#endif
+
+#include "../common/mem.h" /* read */
+#include "../common/pool.h"
+#include "../common/threading.h"
+#include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../common/bits.h" /* ZSTD_highbit32 */
+#include "../zdict.h"
+#include "cover.h"
+
+/*-*************************************
+*  Constants
+***************************************/
+/**
+* There are 32bit indexes used to ref samples, so limit samples size to 4GB
+* on 64bit builds.
+* For 32bit builds we choose 1 GB.
+* Most 32bit platforms have 2GB user-mode addressable space and we allocate a large
+* contiguous buffer, so 1GB is already a high limit.
+*/
+#define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
+#define COVER_DEFAULT_SPLITPOINT 1.0
+
+/*-*************************************
+*  Console display
+***************************************/
+#ifndef LOCALDISPLAYLEVEL
+static int g_displayLevel = 0;
+#endif
+#undef  DISPLAY
+#define DISPLAY(...)                                                           \
+  {                                                                            \
+    fprintf(stderr, __VA_ARGS__);                                              \
+    fflush(stderr);                                                            \
+  }
+#undef  LOCALDISPLAYLEVEL
+#define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
+  if (displayLevel >= l) {                                                     \
+    DISPLAY(__VA_ARGS__);                                                      \
+  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#undef  DISPLAYLEVEL
+#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
+
+#ifndef LOCALDISPLAYUPDATE
+static const clock_t g_refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+#endif
+#undef  LOCALDISPLAYUPDATE
+#define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
+  if (displayLevel >= l) {                                                     \
+    if ((clock() - g_time > g_refreshRate) || (displayLevel >= 4)) {             \
+      g_time = clock();                                                        \
+      DISPLAY(__VA_ARGS__);                                                    \
+    }                                                                          \
+  }
+#undef  DISPLAYUPDATE
+#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
+
+/*-*************************************
+* Hash table
+***************************************
+* A small specialized hash map for storing activeDmers.
+* The map does not resize, so if it becomes full it will loop forever.
+* Thus, the map must be large enough to store every value.
+* The map implements linear probing and keeps its load less than 0.5.
+*/
+
+#define MAP_EMPTY_VALUE ((U32)-1)
+typedef struct COVER_map_pair_t_s {
+  U32 key;
+  U32 value;
+} COVER_map_pair_t;
+
+typedef struct COVER_map_s {
+  COVER_map_pair_t *data;
+  U32 sizeLog;
+  U32 size;
+  U32 sizeMask;
+} COVER_map_t;
+
+/**
+ * Clear the map.
+ */
+static void COVER_map_clear(COVER_map_t *map) {
+  memset(map->data, MAP_EMPTY_VALUE, map->size * sizeof(COVER_map_pair_t));
+}
+
+/**
+ * Initializes a map of the given size.
+ * Returns 1 on success and 0 on failure.
+ * The map must be destroyed with COVER_map_destroy().
+ * The map is only guaranteed to be large enough to hold size elements.
+ */
+static int COVER_map_init(COVER_map_t *map, U32 size) {
+  map->sizeLog = ZSTD_highbit32(size) + 2;
+  map->size = (U32)1 << map->sizeLog;
+  map->sizeMask = map->size - 1;
+  map->data = (COVER_map_pair_t *)malloc(map->size * sizeof(COVER_map_pair_t));
+  if (!map->data) {
+    map->sizeLog = 0;
+    map->size = 0;
+    return 0;
+  }
+  COVER_map_clear(map);
+  return 1;
+}
+
+/**
+ * Internal hash function
+ */
+static const U32 COVER_prime4bytes = 2654435761U;
+static U32 COVER_map_hash(COVER_map_t *map, U32 key) {
+  return (key * COVER_prime4bytes) >> (32 - map->sizeLog);
+}
+
+/**
+ * Helper function that returns the index that a key should be placed into.
+ */
+static U32 COVER_map_index(COVER_map_t *map, U32 key) {
+  const U32 hash = COVER_map_hash(map, key);
+  U32 i;
+  for (i = hash;; i = (i + 1) & map->sizeMask) {
+    COVER_map_pair_t *pos = &map->data[i];
+    if (pos->value == MAP_EMPTY_VALUE) {
+      return i;
+    }
+    if (pos->key == key) {
+      return i;
+    }
+  }
+}
+
+/**
+ * Returns the pointer to the value for key.
+ * If key is not in the map, it is inserted and the value is set to 0.
+ * The map must not be full.
+ */
+static U32 *COVER_map_at(COVER_map_t *map, U32 key) {
+  COVER_map_pair_t *pos = &map->data[COVER_map_index(map, key)];
+  if (pos->value == MAP_EMPTY_VALUE) {
+    pos->key = key;
+    pos->value = 0;
+  }
+  return &pos->value;
+}
+
+/**
+ * Deletes key from the map if present.
+ */
+static void COVER_map_remove(COVER_map_t *map, U32 key) {
+  U32 i = COVER_map_index(map, key);
+  COVER_map_pair_t *del = &map->data[i];
+  U32 shift = 1;
+  if (del->value == MAP_EMPTY_VALUE) {
+    return;
+  }
+  for (i = (i + 1) & map->sizeMask;; i = (i + 1) & map->sizeMask) {
+    COVER_map_pair_t *const pos = &map->data[i];
+    /* If the position is empty we are done */
+    if (pos->value == MAP_EMPTY_VALUE) {
+      del->value = MAP_EMPTY_VALUE;
+      return;
+    }
+    /* If pos can be moved to del do so */
+    if (((i - COVER_map_hash(map, pos->key)) & map->sizeMask) >= shift) {
+      del->key = pos->key;
+      del->value = pos->value;
+      del = pos;
+      shift = 1;
+    } else {
+      ++shift;
+    }
+  }
+}
+
+/**
+ * Destroys a map that is inited with COVER_map_init().
+ */
+static void COVER_map_destroy(COVER_map_t *map) {
+  if (map->data) {
+    free(map->data);
+  }
+  map->data = NULL;
+  map->size = 0;
+}
+
+/*-*************************************
+* Context
+***************************************/
+
+typedef struct {
+  const BYTE *samples;
+  size_t *offsets;
+  const size_t *samplesSizes;
+  size_t nbSamples;
+  size_t nbTrainSamples;
+  size_t nbTestSamples;
+  U32 *suffix;
+  size_t suffixSize;
+  U32 *freqs;
+  U32 *dmerAt;
+  unsigned d;
+} COVER_ctx_t;
+
+/* We need a global context for qsort... */
+static COVER_ctx_t *g_coverCtx = NULL;
+
+/*-*************************************
+*  Helper functions
+***************************************/
+
+/**
+ * Returns the sum of the sample sizes.
+ */
+size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
+  size_t sum = 0;
+  unsigned i;
+  for (i = 0; i < nbSamples; ++i) {
+    sum += samplesSizes[i];
+  }
+  return sum;
+}
+
+/**
+ * Returns -1 if the dmer at lp is less than the dmer at rp.
+ * Return 0 if the dmers at lp and rp are equal.
+ * Returns 1 if the dmer at lp is greater than the dmer at rp.
+ */
+static int COVER_cmp(COVER_ctx_t *ctx, const void *lp, const void *rp) {
+  U32 const lhs = *(U32 const *)lp;
+  U32 const rhs = *(U32 const *)rp;
+  return memcmp(ctx->samples + lhs, ctx->samples + rhs, ctx->d);
+}
+/**
+ * Faster version for d <= 8.
+ */
+static int COVER_cmp8(COVER_ctx_t *ctx, const void *lp, const void *rp) {
+  U64 const mask = (ctx->d == 8) ? (U64)-1 : (((U64)1 << (8 * ctx->d)) - 1);
+  U64 const lhs = MEM_readLE64(ctx->samples + *(U32 const *)lp) & mask;
+  U64 const rhs = MEM_readLE64(ctx->samples + *(U32 const *)rp) & mask;
+  if (lhs < rhs) {
+    return -1;
+  }
+  return (lhs > rhs);
+}
+
+/**
+ * Same as COVER_cmp() except ties are broken by pointer value
+ * NOTE: g_coverCtx must be set to call this function.  A global is required because
+ * qsort doesn't take an opaque pointer.
+ */
+static int WIN_CDECL COVER_strict_cmp(const void *lp, const void *rp) {
+  int result = COVER_cmp(g_coverCtx, lp, rp);
+  if (result == 0) {
+    result = lp < rp ? -1 : 1;
+  }
+  return result;
+}
+/**
+ * Faster version for d <= 8.
+ */
+static int WIN_CDECL COVER_strict_cmp8(const void *lp, const void *rp) {
+  int result = COVER_cmp8(g_coverCtx, lp, rp);
+  if (result == 0) {
+    result = lp < rp ? -1 : 1;
+  }
+  return result;
+}
+
+/**
+ * Returns the first pointer in [first, last) whose element does not compare
+ * less than value.  If no such element exists it returns last.
+ */
+static const size_t *COVER_lower_bound(const size_t *first, const size_t *last,
+                                       size_t value) {
+  size_t count = last - first;
+  while (count != 0) {
+    size_t step = count / 2;
+    const size_t *ptr = first;
+    ptr += step;
+    if (*ptr < value) {
+      first = ++ptr;
+      count -= step + 1;
+    } else {
+      count = step;
+    }
+  }
+  return first;
+}
+
+/**
+ * Generic groupBy function.
+ * Groups an array sorted by cmp into groups with equivalent values.
+ * Calls grp for each group.
+ */
+static void
+COVER_groupBy(const void *data, size_t count, size_t size, COVER_ctx_t *ctx,
+              int (*cmp)(COVER_ctx_t *, const void *, const void *),
+              void (*grp)(COVER_ctx_t *, const void *, const void *)) {
+  const BYTE *ptr = (const BYTE *)data;
+  size_t num = 0;
+  while (num < count) {
+    const BYTE *grpEnd = ptr + size;
+    ++num;
+    while (num < count && cmp(ctx, ptr, grpEnd) == 0) {
+      grpEnd += size;
+      ++num;
+    }
+    grp(ctx, ptr, grpEnd);
+    ptr = grpEnd;
+  }
+}
+
+/*-*************************************
+*  Cover functions
+***************************************/
+
+/**
+ * Called on each group of positions with the same dmer.
+ * Counts the frequency of each dmer and saves it in the suffix array.
+ * Fills `ctx->dmerAt`.
+ */
+static void COVER_group(COVER_ctx_t *ctx, const void *group,
+                        const void *groupEnd) {
+  /* The group consists of all the positions with the same first d bytes. */
+  const U32 *grpPtr = (const U32 *)group;
+  const U32 *grpEnd = (const U32 *)groupEnd;
+  /* The dmerId is how we will reference this dmer.
+   * This allows us to map the whole dmer space to a much smaller space, the
+   * size of the suffix array.
+   */
+  const U32 dmerId = (U32)(grpPtr - ctx->suffix);
+  /* Count the number of samples this dmer shows up in */
+  U32 freq = 0;
+  /* Details */
+  const size_t *curOffsetPtr = ctx->offsets;
+  const size_t *offsetsEnd = ctx->offsets + ctx->nbSamples;
+  /* Once *grpPtr >= curSampleEnd this occurrence of the dmer is in a
+   * different sample than the last.
+   */
+  size_t curSampleEnd = ctx->offsets[0];
+  for (; grpPtr != grpEnd; ++grpPtr) {
+    /* Save the dmerId for this position so we can get back to it. */
+    ctx->dmerAt[*grpPtr] = dmerId;
+    /* Dictionaries only help for the first reference to the dmer.
+     * After that zstd can reference the match from the previous reference.
+     * So only count each dmer once for each sample it is in.
+     */
+    if (*grpPtr < curSampleEnd) {
+      continue;
+    }
+    freq += 1;
+    /* Binary search to find the end of the sample *grpPtr is in.
+     * In the common case that grpPtr + 1 == grpEnd we can skip the binary
+     * search because the loop is over.
+     */
+    if (grpPtr + 1 != grpEnd) {
+      const size_t *sampleEndPtr =
+          COVER_lower_bound(curOffsetPtr, offsetsEnd, *grpPtr);
+      curSampleEnd = *sampleEndPtr;
+      curOffsetPtr = sampleEndPtr + 1;
+    }
+  }
+  /* At this point we are never going to look at this segment of the suffix
+   * array again.  We take advantage of this fact to save memory.
+   * We store the frequency of the dmer in the first position of the group,
+   * which is dmerId.
+   */
+  ctx->suffix[dmerId] = freq;
+}
+
+
+/**
+ * Selects the best segment in an epoch.
+ * Segments of are scored according to the function:
+ *
+ * Let F(d) be the frequency of dmer d.
+ * Let S_i be the dmer at position i of segment S which has length k.
+ *
+ *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
+ *
+ * Once the dmer d is in the dictionary we set F(d) = 0.
+ */
+static COVER_segment_t COVER_selectSegment(const COVER_ctx_t *ctx, U32 *freqs,
+                                           COVER_map_t *activeDmers, U32 begin,
+                                           U32 end,
+                                           ZDICT_cover_params_t parameters) {
+  /* Constants */
+  const U32 k = parameters.k;
+  const U32 d = parameters.d;
+  const U32 dmersInK = k - d + 1;
+  /* Try each segment (activeSegment) and save the best (bestSegment) */
+  COVER_segment_t bestSegment = {0, 0, 0};
+  COVER_segment_t activeSegment;
+  /* Reset the activeDmers in the segment */
+  COVER_map_clear(activeDmers);
+  /* The activeSegment starts at the beginning of the epoch. */
+  activeSegment.begin = begin;
+  activeSegment.end = begin;
+  activeSegment.score = 0;
+  /* Slide the activeSegment through the whole epoch.
+   * Save the best segment in bestSegment.
+   */
+  while (activeSegment.end < end) {
+    /* The dmerId for the dmer at the next position */
+    U32 newDmer = ctx->dmerAt[activeSegment.end];
+    /* The entry in activeDmers for this dmerId */
+    U32 *newDmerOcc = COVER_map_at(activeDmers, newDmer);
+    /* If the dmer isn't already present in the segment add its score. */
+    if (*newDmerOcc == 0) {
+      /* The paper suggest using the L-0.5 norm, but experiments show that it
+       * doesn't help.
+       */
+      activeSegment.score += freqs[newDmer];
+    }
+    /* Add the dmer to the segment */
+    activeSegment.end += 1;
+    *newDmerOcc += 1;
+
+    /* If the window is now too large, drop the first position */
+    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
+      U32 delDmer = ctx->dmerAt[activeSegment.begin];
+      U32 *delDmerOcc = COVER_map_at(activeDmers, delDmer);
+      activeSegment.begin += 1;
+      *delDmerOcc -= 1;
+      /* If this is the last occurrence of the dmer, subtract its score */
+      if (*delDmerOcc == 0) {
+        COVER_map_remove(activeDmers, delDmer);
+        activeSegment.score -= freqs[delDmer];
+      }
+    }
+
+    /* If this segment is the best so far save it */
+    if (activeSegment.score > bestSegment.score) {
+      bestSegment = activeSegment;
+    }
+  }
+  {
+    /* Trim off the zero frequency head and tail from the segment. */
+    U32 newBegin = bestSegment.end;
+    U32 newEnd = bestSegment.begin;
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      U32 freq = freqs[ctx->dmerAt[pos]];
+      if (freq != 0) {
+        newBegin = MIN(newBegin, pos);
+        newEnd = pos + 1;
+      }
+    }
+    bestSegment.begin = newBegin;
+    bestSegment.end = newEnd;
+  }
+  {
+    /* Zero out the frequency of each dmer covered by the chosen segment. */
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      freqs[ctx->dmerAt[pos]] = 0;
+    }
+  }
+  return bestSegment;
+}
+
+/**
+ * Check the validity of the parameters.
+ * Returns non-zero if the parameters are valid and 0 otherwise.
+ */
+static int COVER_checkParameters(ZDICT_cover_params_t parameters,
+                                 size_t maxDictSize) {
+  /* k and d are required parameters */
+  if (parameters.d == 0 || parameters.k == 0) {
+    return 0;
+  }
+  /* k <= maxDictSize */
+  if (parameters.k > maxDictSize) {
+    return 0;
+  }
+  /* d <= k */
+  if (parameters.d > parameters.k) {
+    return 0;
+  }
+  /* 0 < splitPoint <= 1 */
+  if (parameters.splitPoint <= 0 || parameters.splitPoint > 1){
+    return 0;
+  }
+  return 1;
+}
+
+/**
+ * Clean up a context initialized with `COVER_ctx_init()`.
+ */
+static void COVER_ctx_destroy(COVER_ctx_t *ctx) {
+  if (!ctx) {
+    return;
+  }
+  if (ctx->suffix) {
+    free(ctx->suffix);
+    ctx->suffix = NULL;
+  }
+  if (ctx->freqs) {
+    free(ctx->freqs);
+    ctx->freqs = NULL;
+  }
+  if (ctx->dmerAt) {
+    free(ctx->dmerAt);
+    ctx->dmerAt = NULL;
+  }
+  if (ctx->offsets) {
+    free(ctx->offsets);
+    ctx->offsets = NULL;
+  }
+}
+
+/**
+ * Prepare a context for dictionary building.
+ * The context is only dependent on the parameter `d` and can be used multiple
+ * times.
+ * Returns 0 on success or error code on error.
+ * The context must be destroyed with `COVER_ctx_destroy()`.
+ */
+static size_t COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
+                          const size_t *samplesSizes, unsigned nbSamples,
+                          unsigned d, double splitPoint) {
+  const BYTE *const samples = (const BYTE *)samplesBuffer;
+  const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+  /* Split samples into testing and training sets */
+  const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
+  const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
+  const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
+  const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
+  /* Checks */
+  if (totalSamplesSize < MAX(d, sizeof(U64)) ||
+      totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
+    DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
+                 (unsigned)(totalSamplesSize>>20), (COVER_MAX_SAMPLES_SIZE >> 20));
+    return ERROR(srcSize_wrong);
+  }
+  /* Check if there are at least 5 training samples */
+  if (nbTrainSamples < 5) {
+    DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid.", nbTrainSamples);
+    return ERROR(srcSize_wrong);
+  }
+  /* Check if there's testing sample */
+  if (nbTestSamples < 1) {
+    DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.", nbTestSamples);
+    return ERROR(srcSize_wrong);
+  }
+  /* Zero the context */
+  memset(ctx, 0, sizeof(*ctx));
+  DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
+               (unsigned)trainingSamplesSize);
+  DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
+               (unsigned)testSamplesSize);
+  ctx->samples = samples;
+  ctx->samplesSizes = samplesSizes;
+  ctx->nbSamples = nbSamples;
+  ctx->nbTrainSamples = nbTrainSamples;
+  ctx->nbTestSamples = nbTestSamples;
+  /* Partial suffix array */
+  ctx->suffixSize = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
+  ctx->suffix = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  /* Maps index to the dmerID */
+  ctx->dmerAt = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  /* The offsets of each file */
+  ctx->offsets = (size_t *)malloc((nbSamples + 1) * sizeof(size_t));
+  if (!ctx->suffix || !ctx->dmerAt || !ctx->offsets) {
+    DISPLAYLEVEL(1, "Failed to allocate scratch buffers\n");
+    COVER_ctx_destroy(ctx);
+    return ERROR(memory_allocation);
+  }
+  ctx->freqs = NULL;
+  ctx->d = d;
+
+  /* Fill offsets from the samplesSizes */
+  {
+    U32 i;
+    ctx->offsets[0] = 0;
+    for (i = 1; i <= nbSamples; ++i) {
+      ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
+    }
+  }
+  DISPLAYLEVEL(2, "Constructing partial suffix array\n");
+  {
+    /* suffix is a partial suffix array.
+     * It only sorts suffixes by their first parameters.d bytes.
+     * The sort is stable, so each dmer group is sorted by position in input.
+     */
+    U32 i;
+    for (i = 0; i < ctx->suffixSize; ++i) {
+      ctx->suffix[i] = i;
+    }
+    /* qsort doesn't take an opaque pointer, so pass as a global.
+     * On OpenBSD qsort() is not guaranteed to be stable, their mergesort() is.
+     */
+    g_coverCtx = ctx;
+#if defined(__OpenBSD__)
+    mergesort(ctx->suffix, ctx->suffixSize, sizeof(U32),
+          (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
+#else
+    qsort(ctx->suffix, ctx->suffixSize, sizeof(U32),
+          (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
+#endif
+  }
+  DISPLAYLEVEL(2, "Computing frequencies\n");
+  /* For each dmer group (group of positions with the same first d bytes):
+   * 1. For each position we set dmerAt[position] = dmerID.  The dmerID is
+   *    (groupBeginPtr - suffix).  This allows us to go from position to
+   *    dmerID so we can look up values in freq.
+   * 2. We calculate how many samples the dmer occurs in and save it in
+   *    freqs[dmerId].
+   */
+  COVER_groupBy(ctx->suffix, ctx->suffixSize, sizeof(U32), ctx,
+                (ctx->d <= 8 ? &COVER_cmp8 : &COVER_cmp), &COVER_group);
+  ctx->freqs = ctx->suffix;
+  ctx->suffix = NULL;
+  return 0;
+}
+
+void COVER_warnOnSmallCorpus(size_t maxDictSize, size_t nbDmers, int displayLevel)
+{
+  const double ratio = (double)nbDmers / (double)maxDictSize;
+  if (ratio >= 10) {
+      return;
+  }
+  LOCALDISPLAYLEVEL(displayLevel, 1,
+                    "WARNING: The maximum dictionary size %u is too large "
+                    "compared to the source size %u! "
+                    "size(source)/size(dictionary) = %f, but it should be >= "
+                    "10! This may lead to a subpar dictionary! We recommend "
+                    "training on sources at least 10x, and preferably 100x "
+                    "the size of the dictionary! \n", (U32)maxDictSize,
+                    (U32)nbDmers, ratio);
+}
+
+COVER_epoch_info_t COVER_computeEpochs(U32 maxDictSize,
+                                       U32 nbDmers, U32 k, U32 passes)
+{
+  const U32 minEpochSize = k * 10;
+  COVER_epoch_info_t epochs;
+  epochs.num = MAX(1, maxDictSize / k / passes);
+  epochs.size = nbDmers / epochs.num;
+  if (epochs.size >= minEpochSize) {
+      assert(epochs.size * epochs.num <= nbDmers);
+      return epochs;
+  }
+  epochs.size = MIN(minEpochSize, nbDmers);
+  epochs.num = nbDmers / epochs.size;
+  assert(epochs.size * epochs.num <= nbDmers);
+  return epochs;
+}
+
+/**
+ * Given the prepared context build the dictionary.
+ */
+static size_t COVER_buildDictionary(const COVER_ctx_t *ctx, U32 *freqs,
+                                    COVER_map_t *activeDmers, void *dictBuffer,
+                                    size_t dictBufferCapacity,
+                                    ZDICT_cover_params_t parameters) {
+  BYTE *const dict = (BYTE *)dictBuffer;
+  size_t tail = dictBufferCapacity;
+  /* Divide the data into epochs. We will select one segment from each epoch. */
+  const COVER_epoch_info_t epochs = COVER_computeEpochs(
+      (U32)dictBufferCapacity, (U32)ctx->suffixSize, parameters.k, 4);
+  const size_t maxZeroScoreRun = MAX(10, MIN(100, epochs.num >> 3));
+  size_t zeroScoreRun = 0;
+  size_t epoch;
+  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n",
+                (U32)epochs.num, (U32)epochs.size);
+  /* Loop through the epochs until there are no more segments or the dictionary
+   * is full.
+   */
+  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs.num) {
+    const U32 epochBegin = (U32)(epoch * epochs.size);
+    const U32 epochEnd = epochBegin + epochs.size;
+    size_t segmentSize;
+    /* Select a segment */
+    COVER_segment_t segment = COVER_selectSegment(
+        ctx, freqs, activeDmers, epochBegin, epochEnd, parameters);
+    /* If the segment covers no dmers, then we are out of content.
+     * There may be new content in other epochs, for continue for some time.
+     */
+    if (segment.score == 0) {
+      if (++zeroScoreRun >= maxZeroScoreRun) {
+          break;
+      }
+      continue;
+    }
+    zeroScoreRun = 0;
+    /* Trim the segment if necessary and if it is too small then we are done */
+    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
+    if (segmentSize < parameters.d) {
+      break;
+    }
+    /* We fill the dictionary from the back to allow the best segments to be
+     * referenced with the smallest offsets.
+     */
+    tail -= segmentSize;
+    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
+    DISPLAYUPDATE(
+        2, "\r%u%%       ",
+        (unsigned)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
+  }
+  DISPLAYLEVEL(2, "\r%79s\r", "");
+  return tail;
+}
+
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
+    void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_cover_params_t parameters)
+{
+  BYTE* const dict = (BYTE*)dictBuffer;
+  COVER_ctx_t ctx;
+  COVER_map_t activeDmers;
+  parameters.splitPoint = 1.0;
+  /* Initialize global data */
+  g_displayLevel = (int)parameters.zParams.notificationLevel;
+  /* Checks */
+  if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
+    DISPLAYLEVEL(1, "Cover parameters incorrect\n");
+    return ERROR(parameter_outOfBound);
+  }
+  if (nbSamples == 0) {
+    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
+    return ERROR(srcSize_wrong);
+  }
+  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                 ZDICT_DICTSIZE_MIN);
+    return ERROR(dstSize_tooSmall);
+  }
+  /* Initialize context and activeDmers */
+  {
+    size_t const initVal = COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                      parameters.d, parameters.splitPoint);
+    if (ZSTD_isError(initVal)) {
+      return initVal;
+    }
+  }
+  COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.suffixSize, g_displayLevel);
+  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
+    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
+    COVER_ctx_destroy(&ctx);
+    return ERROR(memory_allocation);
+  }
+
+  DISPLAYLEVEL(2, "Building dictionary\n");
+  {
+    const size_t tail =
+        COVER_buildDictionary(&ctx, ctx.freqs, &activeDmers, dictBuffer,
+                              dictBufferCapacity, parameters);
+    const size_t dictionarySize = ZDICT_finalizeDictionary(
+        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+        samplesBuffer, samplesSizes, nbSamples, parameters.zParams);
+    if (!ZSTD_isError(dictionarySize)) {
+      DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
+                   (unsigned)dictionarySize);
+    }
+    COVER_ctx_destroy(&ctx);
+    COVER_map_destroy(&activeDmers);
+    return dictionarySize;
+  }
+}
+
+
+
+size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
+                                    const size_t *samplesSizes, const BYTE *samples,
+                                    size_t *offsets,
+                                    size_t nbTrainSamples, size_t nbSamples,
+                                    BYTE *const dict, size_t dictBufferCapacity) {
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Pointers */
+  ZSTD_CCtx *cctx;
+  ZSTD_CDict *cdict;
+  void *dst;
+  /* Local variables */
+  size_t dstCapacity;
+  size_t i;
+  /* Allocate dst with enough space to compress the maximum sized sample */
+  {
+    size_t maxSampleSize = 0;
+    i = parameters.splitPoint < 1.0 ? nbTrainSamples : 0;
+    for (; i < nbSamples; ++i) {
+      maxSampleSize = MAX(samplesSizes[i], maxSampleSize);
+    }
+    dstCapacity = ZSTD_compressBound(maxSampleSize);
+    dst = malloc(dstCapacity);
+  }
+  /* Create the cctx and cdict */
+  cctx = ZSTD_createCCtx();
+  cdict = ZSTD_createCDict(dict, dictBufferCapacity,
+                           parameters.zParams.compressionLevel);
+  if (!dst || !cctx || !cdict) {
+    goto _compressCleanup;
+  }
+  /* Compress each sample and sum their sizes (or error) */
+  totalCompressedSize = dictBufferCapacity;
+  i = parameters.splitPoint < 1.0 ? nbTrainSamples : 0;
+  for (; i < nbSamples; ++i) {
+    const size_t size = ZSTD_compress_usingCDict(
+        cctx, dst, dstCapacity, samples + offsets[i],
+        samplesSizes[i], cdict);
+    if (ZSTD_isError(size)) {
+      totalCompressedSize = size;
+      goto _compressCleanup;
+    }
+    totalCompressedSize += size;
+  }
+_compressCleanup:
+  ZSTD_freeCCtx(cctx);
+  ZSTD_freeCDict(cdict);
+  if (dst) {
+    free(dst);
+  }
+  return totalCompressedSize;
+}
+
+
+/**
+ * Initialize the `COVER_best_t`.
+ */
+void COVER_best_init(COVER_best_t *best) {
+  if (best==NULL) return; /* compatible with init on NULL */
+  (void)ZSTD_pthread_mutex_init(&best->mutex, NULL);
+  (void)ZSTD_pthread_cond_init(&best->cond, NULL);
+  best->liveJobs = 0;
+  best->dict = NULL;
+  best->dictSize = 0;
+  best->compressedSize = (size_t)-1;
+  memset(&best->parameters, 0, sizeof(best->parameters));
+}
+
+/**
+ * Wait until liveJobs == 0.
+ */
+void COVER_best_wait(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  ZSTD_pthread_mutex_lock(&best->mutex);
+  while (best->liveJobs != 0) {
+    ZSTD_pthread_cond_wait(&best->cond, &best->mutex);
+  }
+  ZSTD_pthread_mutex_unlock(&best->mutex);
+}
+
+/**
+ * Call COVER_best_wait() and then destroy the COVER_best_t.
+ */
+void COVER_best_destroy(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  COVER_best_wait(best);
+  if (best->dict) {
+    free(best->dict);
+  }
+  ZSTD_pthread_mutex_destroy(&best->mutex);
+  ZSTD_pthread_cond_destroy(&best->cond);
+}
+
+/**
+ * Called when a thread is about to be launched.
+ * Increments liveJobs.
+ */
+void COVER_best_start(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  ZSTD_pthread_mutex_lock(&best->mutex);
+  ++best->liveJobs;
+  ZSTD_pthread_mutex_unlock(&best->mutex);
+}
+
+/**
+ * Called when a thread finishes executing, both on error or success.
+ * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
+ * If this dictionary is the best so far save it and its parameters.
+ */
+void COVER_best_finish(COVER_best_t *best, ZDICT_cover_params_t parameters,
+                              COVER_dictSelection_t selection) {
+  void* dict = selection.dictContent;
+  size_t compressedSize = selection.totalCompressedSize;
+  size_t dictSize = selection.dictSize;
+  if (!best) {
+    return;
+  }
+  {
+    size_t liveJobs;
+    ZSTD_pthread_mutex_lock(&best->mutex);
+    --best->liveJobs;
+    liveJobs = best->liveJobs;
+    /* If the new dictionary is better */
+    if (compressedSize < best->compressedSize) {
+      /* Allocate space if necessary */
+      if (!best->dict || best->dictSize < dictSize) {
+        if (best->dict) {
+          free(best->dict);
+        }
+        best->dict = malloc(dictSize);
+        if (!best->dict) {
+          best->compressedSize = ERROR(GENERIC);
+          best->dictSize = 0;
+          ZSTD_pthread_cond_signal(&best->cond);
+          ZSTD_pthread_mutex_unlock(&best->mutex);
+          return;
+        }
+      }
+      /* Save the dictionary, parameters, and size */
+      if (dict) {
+        memcpy(best->dict, dict, dictSize);
+        best->dictSize = dictSize;
+        best->parameters = parameters;
+        best->compressedSize = compressedSize;
+      }
+    }
+    if (liveJobs == 0) {
+      ZSTD_pthread_cond_broadcast(&best->cond);
+    }
+    ZSTD_pthread_mutex_unlock(&best->mutex);
+  }
+}
+
+static COVER_dictSelection_t setDictSelection(BYTE* buf, size_t s, size_t csz)
+{
+    COVER_dictSelection_t ds;
+    ds.dictContent = buf;
+    ds.dictSize = s;
+    ds.totalCompressedSize = csz;
+    return ds;
+}
+
+COVER_dictSelection_t COVER_dictSelectionError(size_t error) {
+    return setDictSelection(NULL, 0, error);
+}
+
+unsigned COVER_dictSelectionIsError(COVER_dictSelection_t selection) {
+  return (ZSTD_isError(selection.totalCompressedSize) || !selection.dictContent);
+}
+
+void COVER_dictSelectionFree(COVER_dictSelection_t selection){
+  free(selection.dictContent);
+}
+
+COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBufferCapacity,
+        size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
+        size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize) {
+
+  size_t largestDict = 0;
+  size_t largestCompressed = 0;
+  BYTE* customDictContentEnd = customDictContent + dictContentSize;
+
+  BYTE * largestDictbuffer = (BYTE *)malloc(dictBufferCapacity);
+  BYTE * candidateDictBuffer = (BYTE *)malloc(dictBufferCapacity);
+  double regressionTolerance = ((double)params.shrinkDictMaxRegression / 100.0) + 1.00;
+
+  if (!largestDictbuffer || !candidateDictBuffer) {
+    free(largestDictbuffer);
+    free(candidateDictBuffer);
+    return COVER_dictSelectionError(dictContentSize);
+  }
+
+  /* Initial dictionary size and compressed size */
+  memcpy(largestDictbuffer, customDictContent, dictContentSize);
+  dictContentSize = ZDICT_finalizeDictionary(
+    largestDictbuffer, dictBufferCapacity, customDictContent, dictContentSize,
+    samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
+
+  if (ZDICT_isError(dictContentSize)) {
+    free(largestDictbuffer);
+    free(candidateDictBuffer);
+    return COVER_dictSelectionError(dictContentSize);
+  }
+
+  totalCompressedSize = COVER_checkTotalCompressedSize(params, samplesSizes,
+                                                       samplesBuffer, offsets,
+                                                       nbCheckSamples, nbSamples,
+                                                       largestDictbuffer, dictContentSize);
+
+  if (ZSTD_isError(totalCompressedSize)) {
+    free(largestDictbuffer);
+    free(candidateDictBuffer);
+    return COVER_dictSelectionError(totalCompressedSize);
+  }
+
+  if (params.shrinkDict == 0) {
+    free(candidateDictBuffer);
+    return setDictSelection(largestDictbuffer, dictContentSize, totalCompressedSize);
+  }
+
+  largestDict = dictContentSize;
+  largestCompressed = totalCompressedSize;
+  dictContentSize = ZDICT_DICTSIZE_MIN;
+
+  /* Largest dict is initially at least ZDICT_DICTSIZE_MIN */
+  while (dictContentSize < largestDict) {
+    memcpy(candidateDictBuffer, largestDictbuffer, largestDict);
+    dictContentSize = ZDICT_finalizeDictionary(
+      candidateDictBuffer, dictBufferCapacity, customDictContentEnd - dictContentSize, dictContentSize,
+      samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
+
+    if (ZDICT_isError(dictContentSize)) {
+      free(largestDictbuffer);
+      free(candidateDictBuffer);
+      return COVER_dictSelectionError(dictContentSize);
+
+    }
+
+    totalCompressedSize = COVER_checkTotalCompressedSize(params, samplesSizes,
+                                                         samplesBuffer, offsets,
+                                                         nbCheckSamples, nbSamples,
+                                                         candidateDictBuffer, dictContentSize);
+
+    if (ZSTD_isError(totalCompressedSize)) {
+      free(largestDictbuffer);
+      free(candidateDictBuffer);
+      return COVER_dictSelectionError(totalCompressedSize);
+    }
+
+    if ((double)totalCompressedSize <= (double)largestCompressed * regressionTolerance) {
+      free(largestDictbuffer);
+      return setDictSelection( candidateDictBuffer, dictContentSize, totalCompressedSize );
+    }
+    dictContentSize *= 2;
+  }
+  dictContentSize = largestDict;
+  totalCompressedSize = largestCompressed;
+  free(candidateDictBuffer);
+  return setDictSelection( largestDictbuffer, dictContentSize, totalCompressedSize );
+}
+
+/**
+ * Parameters for COVER_tryParameters().
+ */
+typedef struct COVER_tryParameters_data_s {
+  const COVER_ctx_t *ctx;
+  COVER_best_t *best;
+  size_t dictBufferCapacity;
+  ZDICT_cover_params_t parameters;
+} COVER_tryParameters_data_t;
+
+/**
+ * Tries a set of parameters and updates the COVER_best_t with the results.
+ * This function is thread safe if zstd is compiled with multithreaded support.
+ * It takes its parameters as an *OWNING* opaque pointer to support threading.
+ */
+static void COVER_tryParameters(void *opaque)
+{
+  /* Save parameters as local variables */
+  COVER_tryParameters_data_t *const data = (COVER_tryParameters_data_t*)opaque;
+  const COVER_ctx_t *const ctx = data->ctx;
+  const ZDICT_cover_params_t parameters = data->parameters;
+  size_t dictBufferCapacity = data->dictBufferCapacity;
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Allocate space for hash table, dict, and freqs */
+  COVER_map_t activeDmers;
+  BYTE* const dict = (BYTE*)malloc(dictBufferCapacity);
+  COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
+  U32* const freqs = (U32*)malloc(ctx->suffixSize * sizeof(U32));
+  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
+    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
+    goto _cleanup;
+  }
+  if (!dict || !freqs) {
+    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
+    goto _cleanup;
+  }
+  /* Copy the frequencies because we need to modify them */
+  memcpy(freqs, ctx->freqs, ctx->suffixSize * sizeof(U32));
+  /* Build the dictionary */
+  {
+    const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
+                                              dictBufferCapacity, parameters);
+    selection = COVER_selectDict(dict + tail, dictBufferCapacity, dictBufferCapacity - tail,
+        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbTrainSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
+        totalCompressedSize);
+
+    if (COVER_dictSelectionIsError(selection)) {
+      DISPLAYLEVEL(1, "Failed to select dictionary\n");
+      goto _cleanup;
+    }
+  }
+_cleanup:
+  free(dict);
+  COVER_best_finish(data->best, parameters, selection);
+  free(data);
+  COVER_map_destroy(&activeDmers);
+  COVER_dictSelectionFree(selection);
+  free(freqs);
+}
+
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
+    void* dictBuffer, size_t dictBufferCapacity, const void* samplesBuffer,
+    const size_t* samplesSizes, unsigned nbSamples,
+    ZDICT_cover_params_t* parameters)
+{
+  /* constants */
+  const unsigned nbThreads = parameters->nbThreads;
+  const double splitPoint =
+      parameters->splitPoint <= 0.0 ? COVER_DEFAULT_SPLITPOINT : parameters->splitPoint;
+  const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
+  const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+  const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+  const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+  const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
+  const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
+  const unsigned kIterations =
+      (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+  const unsigned shrinkDict = 0;
+  /* Local variables */
+  const int displayLevel = parameters->zParams.notificationLevel;
+  unsigned iteration = 1;
+  unsigned d;
+  unsigned k;
+  COVER_best_t best;
+  POOL_ctx *pool = NULL;
+  int warned = 0;
+
+  /* Checks */
+  if (splitPoint <= 0 || splitPoint > 1) {
+    LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
+    return ERROR(parameter_outOfBound);
+  }
+  if (kMinK < kMaxD || kMaxK < kMinK) {
+    LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
+    return ERROR(parameter_outOfBound);
+  }
+  if (nbSamples == 0) {
+    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
+    return ERROR(srcSize_wrong);
+  }
+  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                 ZDICT_DICTSIZE_MIN);
+    return ERROR(dstSize_tooSmall);
+  }
+  if (nbThreads > 1) {
+    pool = POOL_create(nbThreads, 1);
+    if (!pool) {
+      return ERROR(memory_allocation);
+    }
+  }
+  /* Initialization */
+  COVER_best_init(&best);
+  /* Turn down global display level to clean up display at level 2 and below */
+  g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
+  /* Loop through d first because each new value needs a new context */
+  LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
+                    kIterations);
+  for (d = kMinD; d <= kMaxD; d += 2) {
+    /* Initialize the context for this value of d */
+    COVER_ctx_t ctx;
+    LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
+    {
+      const size_t initVal = COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint);
+      if (ZSTD_isError(initVal)) {
+        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return initVal;
+      }
+    }
+    if (!warned) {
+      COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.suffixSize, displayLevel);
+      warned = 1;
+    }
+    /* Loop through k reusing the same context */
+    for (k = kMinK; k <= kMaxK; k += kStepSize) {
+      /* Prepare the arguments */
+      COVER_tryParameters_data_t *data = (COVER_tryParameters_data_t *)malloc(
+          sizeof(COVER_tryParameters_data_t));
+      LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
+      if (!data) {
+        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
+        COVER_best_destroy(&best);
+        COVER_ctx_destroy(&ctx);
+        POOL_free(pool);
+        return ERROR(memory_allocation);
+      }
+      data->ctx = &ctx;
+      data->best = &best;
+      data->dictBufferCapacity = dictBufferCapacity;
+      data->parameters = *parameters;
+      data->parameters.k = k;
+      data->parameters.d = d;
+      data->parameters.splitPoint = splitPoint;
+      data->parameters.steps = kSteps;
+      data->parameters.shrinkDict = shrinkDict;
+      data->parameters.zParams.notificationLevel = g_displayLevel;
+      /* Check the parameters */
+      if (!COVER_checkParameters(data->parameters, dictBufferCapacity)) {
+        DISPLAYLEVEL(1, "Cover parameters incorrect\n");
+        free(data);
+        continue;
+      }
+      /* Call the function and pass ownership of data to it */
+      COVER_best_start(&best);
+      if (pool) {
+        POOL_add(pool, &COVER_tryParameters, data);
+      } else {
+        COVER_tryParameters(data);
+      }
+      /* Print status */
+      LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
+                         (unsigned)((iteration * 100) / kIterations));
+      ++iteration;
+    }
+    COVER_best_wait(&best);
+    COVER_ctx_destroy(&ctx);
+  }
+  LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
+  /* Fill the output buffer and parameters with output of the best parameters */
+  {
+    const size_t dictSize = best.dictSize;
+    if (ZSTD_isError(best.compressedSize)) {
+      const size_t compressedSize = best.compressedSize;
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return compressedSize;
+    }
+    *parameters = best.parameters;
+    memcpy(dictBuffer, best.dict, dictSize);
+    COVER_best_destroy(&best);
+    POOL_free(pool);
+    return dictSize;
+  }
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/cover.h b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/cover.h
new file mode 100644
index 0000000..252624b
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/cover.h
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#  define ZDICT_STATIC_LINKING_ONLY
+#endif
+
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+#include "../common/mem.h" /* read */
+#include "../common/pool.h"
+#include "../common/threading.h"
+#include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../zdict.h"
+
+/**
+ * COVER_best_t is used for two purposes:
+ * 1. Synchronizing threads.
+ * 2. Saving the best parameters and dictionary.
+ *
+ * All of the methods except COVER_best_init() are thread safe if zstd is
+ * compiled with multithreaded support.
+ */
+typedef struct COVER_best_s {
+  ZSTD_pthread_mutex_t mutex;
+  ZSTD_pthread_cond_t cond;
+  size_t liveJobs;
+  void *dict;
+  size_t dictSize;
+  ZDICT_cover_params_t parameters;
+  size_t compressedSize;
+} COVER_best_t;
+
+/**
+ * A segment is a range in the source as well as the score of the segment.
+ */
+typedef struct {
+  U32 begin;
+  U32 end;
+  U32 score;
+} COVER_segment_t;
+
+/**
+ *Number of epochs and size of each epoch.
+ */
+typedef struct {
+  U32 num;
+  U32 size;
+} COVER_epoch_info_t;
+
+/**
+ * Struct used for the dictionary selection function.
+ */
+typedef struct COVER_dictSelection {
+  BYTE* dictContent;
+  size_t dictSize;
+  size_t totalCompressedSize;
+} COVER_dictSelection_t;
+
+/**
+ * Computes the number of epochs and the size of each epoch.
+ * We will make sure that each epoch gets at least 10 * k bytes.
+ *
+ * The COVER algorithms divide the data up into epochs of equal size and
+ * select one segment from each epoch.
+ *
+ * @param maxDictSize The maximum allowed dictionary size.
+ * @param nbDmers     The number of dmers we are training on.
+ * @param k           The parameter k (segment size).
+ * @param passes      The target number of passes over the dmer corpus.
+ *                    More passes means a better dictionary.
+ */
+COVER_epoch_info_t COVER_computeEpochs(U32 maxDictSize, U32 nbDmers,
+                                       U32 k, U32 passes);
+
+/**
+ * Warns the user when their corpus is too small.
+ */
+void COVER_warnOnSmallCorpus(size_t maxDictSize, size_t nbDmers, int displayLevel);
+
+/**
+ *  Checks total compressed size of a dictionary
+ */
+size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
+                                      const size_t *samplesSizes, const BYTE *samples,
+                                      size_t *offsets,
+                                      size_t nbTrainSamples, size_t nbSamples,
+                                      BYTE *const dict, size_t dictBufferCapacity);
+
+/**
+ * Returns the sum of the sample sizes.
+ */
+size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) ;
+
+/**
+ * Initialize the `COVER_best_t`.
+ */
+void COVER_best_init(COVER_best_t *best);
+
+/**
+ * Wait until liveJobs == 0.
+ */
+void COVER_best_wait(COVER_best_t *best);
+
+/**
+ * Call COVER_best_wait() and then destroy the COVER_best_t.
+ */
+void COVER_best_destroy(COVER_best_t *best);
+
+/**
+ * Called when a thread is about to be launched.
+ * Increments liveJobs.
+ */
+void COVER_best_start(COVER_best_t *best);
+
+/**
+ * Called when a thread finishes executing, both on error or success.
+ * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
+ * If this dictionary is the best so far save it and its parameters.
+ */
+void COVER_best_finish(COVER_best_t *best, ZDICT_cover_params_t parameters,
+                       COVER_dictSelection_t selection);
+/**
+ * Error function for COVER_selectDict function. Checks if the return
+ * value is an error.
+ */
+unsigned COVER_dictSelectionIsError(COVER_dictSelection_t selection);
+
+ /**
+  * Error function for COVER_selectDict function. Returns a struct where
+  * return.totalCompressedSize is a ZSTD error.
+  */
+COVER_dictSelection_t COVER_dictSelectionError(size_t error);
+
+/**
+ * Always call after selectDict is called to free up used memory from
+ * newly created dictionary.
+ */
+void COVER_dictSelectionFree(COVER_dictSelection_t selection);
+
+/**
+ * Called to finalize the dictionary and select one based on whether or not
+ * the shrink-dict flag was enabled. If enabled the dictionary used is the
+ * smallest dictionary within a specified regression of the compressed size
+ * from the largest dictionary.
+ */
+ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBufferCapacity,
+                       size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
+                       size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize);
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/divsufsort.c b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/divsufsort.c
new file mode 100644
index 0000000..a2870fb
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/divsufsort.c
@@ -0,0 +1,1913 @@
+/*
+ * divsufsort.c for libdivsufsort-lite
+ * Copyright (c) 2003-2008 Yuta Mori All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/*- Compiler specifics -*/
+#ifdef __clang__
+#pragma clang diagnostic ignored "-Wshorten-64-to-32"
+#endif
+
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4244)
+#  pragma warning(disable : 4127)    /* C4127 : Condition expression is constant */
+#endif
+
+
+/*- Dependencies -*/
+#include <assert.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "divsufsort.h"
+
+/*- Constants -*/
+#if defined(INLINE)
+# undef INLINE
+#endif
+#if !defined(INLINE)
+# define INLINE __inline
+#endif
+#if defined(ALPHABET_SIZE) && (ALPHABET_SIZE < 1)
+# undef ALPHABET_SIZE
+#endif
+#if !defined(ALPHABET_SIZE)
+# define ALPHABET_SIZE (256)
+#endif
+#define BUCKET_A_SIZE (ALPHABET_SIZE)
+#define BUCKET_B_SIZE (ALPHABET_SIZE * ALPHABET_SIZE)
+#if defined(SS_INSERTIONSORT_THRESHOLD)
+# if SS_INSERTIONSORT_THRESHOLD < 1
+#  undef SS_INSERTIONSORT_THRESHOLD
+#  define SS_INSERTIONSORT_THRESHOLD (1)
+# endif
+#else
+# define SS_INSERTIONSORT_THRESHOLD (8)
+#endif
+#if defined(SS_BLOCKSIZE)
+# if SS_BLOCKSIZE < 0
+#  undef SS_BLOCKSIZE
+#  define SS_BLOCKSIZE (0)
+# elif 32768 <= SS_BLOCKSIZE
+#  undef SS_BLOCKSIZE
+#  define SS_BLOCKSIZE (32767)
+# endif
+#else
+# define SS_BLOCKSIZE (1024)
+#endif
+/* minstacksize = log(SS_BLOCKSIZE) / log(3) * 2 */
+#if SS_BLOCKSIZE == 0
+# define SS_MISORT_STACKSIZE (96)
+#elif SS_BLOCKSIZE <= 4096
+# define SS_MISORT_STACKSIZE (16)
+#else
+# define SS_MISORT_STACKSIZE (24)
+#endif
+#define SS_SMERGE_STACKSIZE (32)
+#define TR_INSERTIONSORT_THRESHOLD (8)
+#define TR_STACKSIZE (64)
+
+
+/*- Macros -*/
+#ifndef SWAP
+# define SWAP(_a, _b) do { t = (_a); (_a) = (_b); (_b) = t; } while(0)
+#endif /* SWAP */
+#ifndef MIN
+# define MIN(_a, _b) (((_a) < (_b)) ? (_a) : (_b))
+#endif /* MIN */
+#ifndef MAX
+# define MAX(_a, _b) (((_a) > (_b)) ? (_a) : (_b))
+#endif /* MAX */
+#define STACK_PUSH(_a, _b, _c, _d)\
+  do {\
+    assert(ssize < STACK_SIZE);\
+    stack[ssize].a = (_a), stack[ssize].b = (_b),\
+    stack[ssize].c = (_c), stack[ssize++].d = (_d);\
+  } while(0)
+#define STACK_PUSH5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(ssize < STACK_SIZE);\
+    stack[ssize].a = (_a), stack[ssize].b = (_b),\
+    stack[ssize].c = (_c), stack[ssize].d = (_d), stack[ssize++].e = (_e);\
+  } while(0)
+#define STACK_POP(_a, _b, _c, _d)\
+  do {\
+    assert(0 <= ssize);\
+    if(ssize == 0) { return; }\
+    (_a) = stack[--ssize].a, (_b) = stack[ssize].b,\
+    (_c) = stack[ssize].c, (_d) = stack[ssize].d;\
+  } while(0)
+#define STACK_POP5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(0 <= ssize);\
+    if(ssize == 0) { return; }\
+    (_a) = stack[--ssize].a, (_b) = stack[ssize].b,\
+    (_c) = stack[ssize].c, (_d) = stack[ssize].d, (_e) = stack[ssize].e;\
+  } while(0)
+#define BUCKET_A(_c0) bucket_A[(_c0)]
+#if ALPHABET_SIZE == 256
+#define BUCKET_B(_c0, _c1) (bucket_B[((_c1) << 8) | (_c0)])
+#define BUCKET_BSTAR(_c0, _c1) (bucket_B[((_c0) << 8) | (_c1)])
+#else
+#define BUCKET_B(_c0, _c1) (bucket_B[(_c1) * ALPHABET_SIZE + (_c0)])
+#define BUCKET_BSTAR(_c0, _c1) (bucket_B[(_c0) * ALPHABET_SIZE + (_c1)])
+#endif
+
+
+/*- Private Functions -*/
+
+static const int lg_table[256]= {
+ -1,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,
+  5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7
+};
+
+#if (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE)
+
+static INLINE
+int
+ss_ilg(int n) {
+#if SS_BLOCKSIZE == 0
+  return (n & 0xffff0000) ?
+          ((n & 0xff000000) ?
+            24 + lg_table[(n >> 24) & 0xff] :
+            16 + lg_table[(n >> 16) & 0xff]) :
+          ((n & 0x0000ff00) ?
+             8 + lg_table[(n >>  8) & 0xff] :
+             0 + lg_table[(n >>  0) & 0xff]);
+#elif SS_BLOCKSIZE < 256
+  return lg_table[n];
+#else
+  return (n & 0xff00) ?
+          8 + lg_table[(n >> 8) & 0xff] :
+          0 + lg_table[(n >> 0) & 0xff];
+#endif
+}
+
+#endif /* (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE) */
+
+#if SS_BLOCKSIZE != 0
+
+static const int sqq_table[256] = {
+  0,  16,  22,  27,  32,  35,  39,  42,  45,  48,  50,  53,  55,  57,  59,  61,
+ 64,  65,  67,  69,  71,  73,  75,  76,  78,  80,  81,  83,  84,  86,  87,  89,
+ 90,  91,  93,  94,  96,  97,  98,  99, 101, 102, 103, 104, 106, 107, 108, 109,
+110, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126,
+128, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142,
+143, 144, 144, 145, 146, 147, 148, 149, 150, 150, 151, 152, 153, 154, 155, 155,
+156, 157, 158, 159, 160, 160, 161, 162, 163, 163, 164, 165, 166, 167, 167, 168,
+169, 170, 170, 171, 172, 173, 173, 174, 175, 176, 176, 177, 178, 178, 179, 180,
+181, 181, 182, 183, 183, 184, 185, 185, 186, 187, 187, 188, 189, 189, 190, 191,
+192, 192, 193, 193, 194, 195, 195, 196, 197, 197, 198, 199, 199, 200, 201, 201,
+202, 203, 203, 204, 204, 205, 206, 206, 207, 208, 208, 209, 209, 210, 211, 211,
+212, 212, 213, 214, 214, 215, 215, 216, 217, 217, 218, 218, 219, 219, 220, 221,
+221, 222, 222, 223, 224, 224, 225, 225, 226, 226, 227, 227, 228, 229, 229, 230,
+230, 231, 231, 232, 232, 233, 234, 234, 235, 235, 236, 236, 237, 237, 238, 238,
+239, 240, 240, 241, 241, 242, 242, 243, 243, 244, 244, 245, 245, 246, 246, 247,
+247, 248, 248, 249, 249, 250, 250, 251, 251, 252, 252, 253, 253, 254, 254, 255
+};
+
+static INLINE
+int
+ss_isqrt(int x) {
+  int y, e;
+
+  if(x >= (SS_BLOCKSIZE * SS_BLOCKSIZE)) { return SS_BLOCKSIZE; }
+  e = (x & 0xffff0000) ?
+        ((x & 0xff000000) ?
+          24 + lg_table[(x >> 24) & 0xff] :
+          16 + lg_table[(x >> 16) & 0xff]) :
+        ((x & 0x0000ff00) ?
+           8 + lg_table[(x >>  8) & 0xff] :
+           0 + lg_table[(x >>  0) & 0xff]);
+
+  if(e >= 16) {
+    y = sqq_table[x >> ((e - 6) - (e & 1))] << ((e >> 1) - 7);
+    if(e >= 24) { y = (y + 1 + x / y) >> 1; }
+    y = (y + 1 + x / y) >> 1;
+  } else if(e >= 8) {
+    y = (sqq_table[x >> ((e - 6) - (e & 1))] >> (7 - (e >> 1))) + 1;
+  } else {
+    return sqq_table[x] >> 4;
+  }
+
+  return (x < (y * y)) ? y - 1 : y;
+}
+
+#endif /* SS_BLOCKSIZE != 0 */
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Compares two suffixes. */
+static INLINE
+int
+ss_compare(const unsigned char *T,
+           const int *p1, const int *p2,
+           int depth) {
+  const unsigned char *U1, *U2, *U1n, *U2n;
+
+  for(U1 = T + depth + *p1,
+      U2 = T + depth + *p2,
+      U1n = T + *(p1 + 1) + 2,
+      U2n = T + *(p2 + 1) + 2;
+      (U1 < U1n) && (U2 < U2n) && (*U1 == *U2);
+      ++U1, ++U2) {
+  }
+
+  return U1 < U1n ?
+        (U2 < U2n ? *U1 - *U2 : 1) :
+        (U2 < U2n ? -1 : 0);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+#if (SS_BLOCKSIZE != 1) && (SS_INSERTIONSORT_THRESHOLD != 1)
+
+/* Insertionsort for small size groups */
+static
+void
+ss_insertionsort(const unsigned char *T, const int *PA,
+                 int *first, int *last, int depth) {
+  int *i, *j;
+  int t;
+  int r;
+
+  for(i = last - 2; first <= i; --i) {
+    for(t = *i, j = i + 1; 0 < (r = ss_compare(T, PA + t, PA + *j, depth));) {
+      do { *(j - 1) = *j; } while((++j < last) && (*j < 0));
+      if(last <= j) { break; }
+    }
+    if(r == 0) { *j = ~*j; }
+    *(j - 1) = t;
+  }
+}
+
+#endif /* (SS_BLOCKSIZE != 1) && (SS_INSERTIONSORT_THRESHOLD != 1) */
+
+
+/*---------------------------------------------------------------------------*/
+
+#if (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE)
+
+static INLINE
+void
+ss_fixdown(const unsigned char *Td, const int *PA,
+           int *SA, int i, int size) {
+  int j, k;
+  int v;
+  int c, d, e;
+
+  for(v = SA[i], c = Td[PA[v]]; (j = 2 * i + 1) < size; SA[i] = SA[k], i = k) {
+    d = Td[PA[SA[k = j++]]];
+    if(d < (e = Td[PA[SA[j]]])) { k = j; d = e; }
+    if(d <= c) { break; }
+  }
+  SA[i] = v;
+}
+
+/* Simple top-down heapsort. */
+static
+void
+ss_heapsort(const unsigned char *Td, const int *PA, int *SA, int size) {
+  int i, m;
+  int t;
+
+  m = size;
+  if((size % 2) == 0) {
+    m--;
+    if(Td[PA[SA[m / 2]]] < Td[PA[SA[m]]]) { SWAP(SA[m], SA[m / 2]); }
+  }
+
+  for(i = m / 2 - 1; 0 <= i; --i) { ss_fixdown(Td, PA, SA, i, m); }
+  if((size % 2) == 0) { SWAP(SA[0], SA[m]); ss_fixdown(Td, PA, SA, 0, m); }
+  for(i = m - 1; 0 < i; --i) {
+    t = SA[0], SA[0] = SA[i];
+    ss_fixdown(Td, PA, SA, 0, i);
+    SA[i] = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Returns the median of three elements. */
+static INLINE
+int *
+ss_median3(const unsigned char *Td, const int *PA,
+           int *v1, int *v2, int *v3) {
+  int *t;
+  if(Td[PA[*v1]] > Td[PA[*v2]]) { SWAP(v1, v2); }
+  if(Td[PA[*v2]] > Td[PA[*v3]]) {
+    if(Td[PA[*v1]] > Td[PA[*v3]]) { return v1; }
+    else { return v3; }
+  }
+  return v2;
+}
+
+/* Returns the median of five elements. */
+static INLINE
+int *
+ss_median5(const unsigned char *Td, const int *PA,
+           int *v1, int *v2, int *v3, int *v4, int *v5) {
+  int *t;
+  if(Td[PA[*v2]] > Td[PA[*v3]]) { SWAP(v2, v3); }
+  if(Td[PA[*v4]] > Td[PA[*v5]]) { SWAP(v4, v5); }
+  if(Td[PA[*v2]] > Td[PA[*v4]]) { SWAP(v2, v4); SWAP(v3, v5); }
+  if(Td[PA[*v1]] > Td[PA[*v3]]) { SWAP(v1, v3); }
+  if(Td[PA[*v1]] > Td[PA[*v4]]) { SWAP(v1, v4); SWAP(v3, v5); }
+  if(Td[PA[*v3]] > Td[PA[*v4]]) { return v4; }
+  return v3;
+}
+
+/* Returns the pivot element. */
+static INLINE
+int *
+ss_pivot(const unsigned char *Td, const int *PA, int *first, int *last) {
+  int *middle;
+  int t;
+
+  t = last - first;
+  middle = first + t / 2;
+
+  if(t <= 512) {
+    if(t <= 32) {
+      return ss_median3(Td, PA, first, middle, last - 1);
+    } else {
+      t >>= 2;
+      return ss_median5(Td, PA, first, first + t, middle, last - 1 - t, last - 1);
+    }
+  }
+  t >>= 3;
+  first  = ss_median3(Td, PA, first, first + t, first + (t << 1));
+  middle = ss_median3(Td, PA, middle - t, middle, middle + t);
+  last   = ss_median3(Td, PA, last - 1 - (t << 1), last - 1 - t, last - 1);
+  return ss_median3(Td, PA, first, middle, last);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Binary partition for substrings. */
+static INLINE
+int *
+ss_partition(const int *PA,
+                    int *first, int *last, int depth) {
+  int *a, *b;
+  int t;
+  for(a = first - 1, b = last;;) {
+    for(; (++a < b) && ((PA[*a] + depth) >= (PA[*a + 1] + 1));) { *a = ~*a; }
+    for(; (a < --b) && ((PA[*b] + depth) <  (PA[*b + 1] + 1));) { }
+    if(b <= a) { break; }
+    t = ~*b;
+    *b = *a;
+    *a = t;
+  }
+  if(first < a) { *first = ~*first; }
+  return a;
+}
+
+/* Multikey introsort for medium size groups. */
+static
+void
+ss_mintrosort(const unsigned char *T, const int *PA,
+              int *first, int *last,
+              int depth) {
+#define STACK_SIZE SS_MISORT_STACKSIZE
+  struct { int *a, *b, c; int d; } stack[STACK_SIZE];
+  const unsigned char *Td;
+  int *a, *b, *c, *d, *e, *f;
+  int s, t;
+  int ssize;
+  int limit;
+  int v, x = 0;
+
+  for(ssize = 0, limit = ss_ilg(last - first);;) {
+
+    if((last - first) <= SS_INSERTIONSORT_THRESHOLD) {
+#if 1 < SS_INSERTIONSORT_THRESHOLD
+      if(1 < (last - first)) { ss_insertionsort(T, PA, first, last, depth); }
+#endif
+      STACK_POP(first, last, depth, limit);
+      continue;
+    }
+
+    Td = T + depth;
+    if(limit-- == 0) { ss_heapsort(Td, PA, first, last - first); }
+    if(limit < 0) {
+      for(a = first + 1, v = Td[PA[*first]]; a < last; ++a) {
+        if((x = Td[PA[*a]]) != v) {
+          if(1 < (a - first)) { break; }
+          v = x;
+          first = a;
+        }
+      }
+      if(Td[PA[*first] - 1] < v) {
+        first = ss_partition(PA, first, a, depth);
+      }
+      if((a - first) <= (last - a)) {
+        if(1 < (a - first)) {
+          STACK_PUSH(a, last, depth, -1);
+          last = a, depth += 1, limit = ss_ilg(a - first);
+        } else {
+          first = a, limit = -1;
+        }
+      } else {
+        if(1 < (last - a)) {
+          STACK_PUSH(first, a, depth + 1, ss_ilg(a - first));
+          first = a, limit = -1;
+        } else {
+          last = a, depth += 1, limit = ss_ilg(a - first);
+        }
+      }
+      continue;
+    }
+
+    /* choose pivot */
+    a = ss_pivot(Td, PA, first, last);
+    v = Td[PA[*a]];
+    SWAP(*first, *a);
+
+    /* partition */
+    for(b = first; (++b < last) && ((x = Td[PA[*b]]) == v);) { }
+    if(((a = b) < last) && (x < v)) {
+      for(; (++b < last) && ((x = Td[PA[*b]]) <= v);) {
+        if(x == v) { SWAP(*b, *a); ++a; }
+      }
+    }
+    for(c = last; (b < --c) && ((x = Td[PA[*c]]) == v);) { }
+    if((b < (d = c)) && (x > v)) {
+      for(; (b < --c) && ((x = Td[PA[*c]]) >= v);) {
+        if(x == v) { SWAP(*c, *d); --d; }
+      }
+    }
+    for(; b < c;) {
+      SWAP(*b, *c);
+      for(; (++b < c) && ((x = Td[PA[*b]]) <= v);) {
+        if(x == v) { SWAP(*b, *a); ++a; }
+      }
+      for(; (b < --c) && ((x = Td[PA[*c]]) >= v);) {
+        if(x == v) { SWAP(*c, *d); --d; }
+      }
+    }
+
+    if(a <= d) {
+      c = b - 1;
+
+      if((s = a - first) > (t = b - a)) { s = t; }
+      for(e = first, f = b - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+      if((s = d - c) > (t = last - d - 1)) { s = t; }
+      for(e = b, f = last - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+
+      a = first + (b - a), c = last - (d - c);
+      b = (v <= Td[PA[*a] - 1]) ? a : ss_partition(PA, a, c, depth);
+
+      if((a - first) <= (last - c)) {
+        if((last - c) <= (c - b)) {
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          STACK_PUSH(c, last, depth, limit);
+          last = a;
+        } else if((a - first) <= (c - b)) {
+          STACK_PUSH(c, last, depth, limit);
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          last = a;
+        } else {
+          STACK_PUSH(c, last, depth, limit);
+          STACK_PUSH(first, a, depth, limit);
+          first = b, last = c, depth += 1, limit = ss_ilg(c - b);
+        }
+      } else {
+        if((a - first) <= (c - b)) {
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          STACK_PUSH(first, a, depth, limit);
+          first = c;
+        } else if((last - c) <= (c - b)) {
+          STACK_PUSH(first, a, depth, limit);
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          first = c;
+        } else {
+          STACK_PUSH(first, a, depth, limit);
+          STACK_PUSH(c, last, depth, limit);
+          first = b, last = c, depth += 1, limit = ss_ilg(c - b);
+        }
+      }
+    } else {
+      limit += 1;
+      if(Td[PA[*first] - 1] < v) {
+        first = ss_partition(PA, first, last, depth);
+        limit = ss_ilg(last - first);
+      }
+      depth += 1;
+    }
+  }
+#undef STACK_SIZE
+}
+
+#endif /* (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE) */
+
+
+/*---------------------------------------------------------------------------*/
+
+#if SS_BLOCKSIZE != 0
+
+static INLINE
+void
+ss_blockswap(int *a, int *b, int n) {
+  int t;
+  for(; 0 < n; --n, ++a, ++b) {
+    t = *a, *a = *b, *b = t;
+  }
+}
+
+static INLINE
+void
+ss_rotate(int *first, int *middle, int *last) {
+  int *a, *b, t;
+  int l, r;
+  l = middle - first, r = last - middle;
+  for(; (0 < l) && (0 < r);) {
+    if(l == r) { ss_blockswap(first, middle, l); break; }
+    if(l < r) {
+      a = last - 1, b = middle - 1;
+      t = *a;
+      do {
+        *a-- = *b, *b-- = *a;
+        if(b < first) {
+          *a = t;
+          last = a;
+          if((r -= l + 1) <= l) { break; }
+          a -= 1, b = middle - 1;
+          t = *a;
+        }
+      } while(1);
+    } else {
+      a = first, b = middle;
+      t = *a;
+      do {
+        *a++ = *b, *b++ = *a;
+        if(last <= b) {
+          *a = t;
+          first = a + 1;
+          if((l -= r + 1) <= r) { break; }
+          a += 1, b = middle;
+          t = *a;
+        }
+      } while(1);
+    }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static
+void
+ss_inplacemerge(const unsigned char *T, const int *PA,
+                int *first, int *middle, int *last,
+                int depth) {
+  const int *p;
+  int *a, *b;
+  int len, half;
+  int q, r;
+  int x;
+
+  for(;;) {
+    if(*(last - 1) < 0) { x = 1; p = PA + ~*(last - 1); }
+    else                { x = 0; p = PA +  *(last - 1); }
+    for(a = first, len = middle - first, half = len >> 1, r = -1;
+        0 < len;
+        len = half, half >>= 1) {
+      b = a + half;
+      q = ss_compare(T, PA + ((0 <= *b) ? *b : ~*b), p, depth);
+      if(q < 0) {
+        a = b + 1;
+        half -= (len & 1) ^ 1;
+      } else {
+        r = q;
+      }
+    }
+    if(a < middle) {
+      if(r == 0) { *a = ~*a; }
+      ss_rotate(a, middle, last);
+      last -= middle - a;
+      middle = a;
+      if(first == middle) { break; }
+    }
+    --last;
+    if(x != 0) { while(*--last < 0) { } }
+    if(middle == last) { break; }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Merge-forward with internal buffer. */
+static
+void
+ss_mergeforward(const unsigned char *T, const int *PA,
+                int *first, int *middle, int *last,
+                int *buf, int depth) {
+  int *a, *b, *c, *bufend;
+  int t;
+  int r;
+
+  bufend = buf + (middle - first) - 1;
+  ss_blockswap(buf, first, middle - first);
+
+  for(t = *(a = first), b = buf, c = middle;;) {
+    r = ss_compare(T, PA + *b, PA + *c, depth);
+    if(r < 0) {
+      do {
+        *a++ = *b;
+        if(bufend <= b) { *bufend = t; return; }
+        *b++ = *a;
+      } while(*b < 0);
+    } else if(r > 0) {
+      do {
+        *a++ = *c, *c++ = *a;
+        if(last <= c) {
+          while(b < bufend) { *a++ = *b, *b++ = *a; }
+          *a = *b, *b = t;
+          return;
+        }
+      } while(*c < 0);
+    } else {
+      *c = ~*c;
+      do {
+        *a++ = *b;
+        if(bufend <= b) { *bufend = t; return; }
+        *b++ = *a;
+      } while(*b < 0);
+
+      do {
+        *a++ = *c, *c++ = *a;
+        if(last <= c) {
+          while(b < bufend) { *a++ = *b, *b++ = *a; }
+          *a = *b, *b = t;
+          return;
+        }
+      } while(*c < 0);
+    }
+  }
+}
+
+/* Merge-backward with internal buffer. */
+static
+void
+ss_mergebackward(const unsigned char *T, const int *PA,
+                 int *first, int *middle, int *last,
+                 int *buf, int depth) {
+  const int *p1, *p2;
+  int *a, *b, *c, *bufend;
+  int t;
+  int r;
+  int x;
+
+  bufend = buf + (last - middle) - 1;
+  ss_blockswap(buf, middle, last - middle);
+
+  x = 0;
+  if(*bufend < 0)       { p1 = PA + ~*bufend; x |= 1; }
+  else                  { p1 = PA +  *bufend; }
+  if(*(middle - 1) < 0) { p2 = PA + ~*(middle - 1); x |= 2; }
+  else                  { p2 = PA +  *(middle - 1); }
+  for(t = *(a = last - 1), b = bufend, c = middle - 1;;) {
+    r = ss_compare(T, p1, p2, depth);
+    if(0 < r) {
+      if(x & 1) { do { *a-- = *b, *b-- = *a; } while(*b < 0); x ^= 1; }
+      *a-- = *b;
+      if(b <= buf) { *buf = t; break; }
+      *b-- = *a;
+      if(*b < 0) { p1 = PA + ~*b; x |= 1; }
+      else       { p1 = PA +  *b; }
+    } else if(r < 0) {
+      if(x & 2) { do { *a-- = *c, *c-- = *a; } while(*c < 0); x ^= 2; }
+      *a-- = *c, *c-- = *a;
+      if(c < first) {
+        while(buf < b) { *a-- = *b, *b-- = *a; }
+        *a = *b, *b = t;
+        break;
+      }
+      if(*c < 0) { p2 = PA + ~*c; x |= 2; }
+      else       { p2 = PA +  *c; }
+    } else {
+      if(x & 1) { do { *a-- = *b, *b-- = *a; } while(*b < 0); x ^= 1; }
+      *a-- = ~*b;
+      if(b <= buf) { *buf = t; break; }
+      *b-- = *a;
+      if(x & 2) { do { *a-- = *c, *c-- = *a; } while(*c < 0); x ^= 2; }
+      *a-- = *c, *c-- = *a;
+      if(c < first) {
+        while(buf < b) { *a-- = *b, *b-- = *a; }
+        *a = *b, *b = t;
+        break;
+      }
+      if(*b < 0) { p1 = PA + ~*b; x |= 1; }
+      else       { p1 = PA +  *b; }
+      if(*c < 0) { p2 = PA + ~*c; x |= 2; }
+      else       { p2 = PA +  *c; }
+    }
+  }
+}
+
+/* D&C based merge. */
+static
+void
+ss_swapmerge(const unsigned char *T, const int *PA,
+             int *first, int *middle, int *last,
+             int *buf, int bufsize, int depth) {
+#define STACK_SIZE SS_SMERGE_STACKSIZE
+#define GETIDX(a) ((0 <= (a)) ? (a) : (~(a)))
+#define MERGE_CHECK(a, b, c)\
+  do {\
+    if(((c) & 1) ||\
+       (((c) & 2) && (ss_compare(T, PA + GETIDX(*((a) - 1)), PA + *(a), depth) == 0))) {\
+      *(a) = ~*(a);\
+    }\
+    if(((c) & 4) && ((ss_compare(T, PA + GETIDX(*((b) - 1)), PA + *(b), depth) == 0))) {\
+      *(b) = ~*(b);\
+    }\
+  } while(0)
+  struct { int *a, *b, *c; int d; } stack[STACK_SIZE];
+  int *l, *r, *lm, *rm;
+  int m, len, half;
+  int ssize;
+  int check, next;
+
+  for(check = 0, ssize = 0;;) {
+    if((last - middle) <= bufsize) {
+      if((first < middle) && (middle < last)) {
+        ss_mergebackward(T, PA, first, middle, last, buf, depth);
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+      continue;
+    }
+
+    if((middle - first) <= bufsize) {
+      if(first < middle) {
+        ss_mergeforward(T, PA, first, middle, last, buf, depth);
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+      continue;
+    }
+
+    for(m = 0, len = MIN(middle - first, last - middle), half = len >> 1;
+        0 < len;
+        len = half, half >>= 1) {
+      if(ss_compare(T, PA + GETIDX(*(middle + m + half)),
+                       PA + GETIDX(*(middle - m - half - 1)), depth) < 0) {
+        m += half + 1;
+        half -= (len & 1) ^ 1;
+      }
+    }
+
+    if(0 < m) {
+      lm = middle - m, rm = middle + m;
+      ss_blockswap(lm, middle, m);
+      l = r = middle, next = 0;
+      if(rm < last) {
+        if(*rm < 0) {
+          *rm = ~*rm;
+          if(first < lm) { for(; *--l < 0;) { } next |= 4; }
+          next |= 1;
+        } else if(first < lm) {
+          for(; *r < 0; ++r) { }
+          next |= 2;
+        }
+      }
+
+      if((l - first) <= (last - r)) {
+        STACK_PUSH(r, rm, last, (next & 3) | (check & 4));
+        middle = lm, last = l, check = (check & 3) | (next & 4);
+      } else {
+        if((next & 2) && (r == middle)) { next ^= 6; }
+        STACK_PUSH(first, lm, l, (check & 3) | (next & 4));
+        first = r, middle = rm, check = (next & 3) | (check & 4);
+      }
+    } else {
+      if(ss_compare(T, PA + GETIDX(*(middle - 1)), PA + *middle, depth) == 0) {
+        *middle = ~*middle;
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+    }
+  }
+#undef STACK_SIZE
+}
+
+#endif /* SS_BLOCKSIZE != 0 */
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Substring sort */
+static
+void
+sssort(const unsigned char *T, const int *PA,
+       int *first, int *last,
+       int *buf, int bufsize,
+       int depth, int n, int lastsuffix) {
+  int *a;
+#if SS_BLOCKSIZE != 0
+  int *b, *middle, *curbuf;
+  int j, k, curbufsize, limit;
+#endif
+  int i;
+
+  if(lastsuffix != 0) { ++first; }
+
+#if SS_BLOCKSIZE == 0
+  ss_mintrosort(T, PA, first, last, depth);
+#else
+  if((bufsize < SS_BLOCKSIZE) &&
+      (bufsize < (last - first)) &&
+      (bufsize < (limit = ss_isqrt(last - first)))) {
+    if(SS_BLOCKSIZE < limit) { limit = SS_BLOCKSIZE; }
+    buf = middle = last - limit, bufsize = limit;
+  } else {
+    middle = last, limit = 0;
+  }
+  for(a = first, i = 0; SS_BLOCKSIZE < (middle - a); a += SS_BLOCKSIZE, ++i) {
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+    ss_mintrosort(T, PA, a, a + SS_BLOCKSIZE, depth);
+#elif 1 < SS_BLOCKSIZE
+    ss_insertionsort(T, PA, a, a + SS_BLOCKSIZE, depth);
+#endif
+    curbufsize = last - (a + SS_BLOCKSIZE);
+    curbuf = a + SS_BLOCKSIZE;
+    if(curbufsize <= bufsize) { curbufsize = bufsize, curbuf = buf; }
+    for(b = a, k = SS_BLOCKSIZE, j = i; j & 1; b -= k, k <<= 1, j >>= 1) {
+      ss_swapmerge(T, PA, b - k, b, b + k, curbuf, curbufsize, depth);
+    }
+  }
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+  ss_mintrosort(T, PA, a, middle, depth);
+#elif 1 < SS_BLOCKSIZE
+  ss_insertionsort(T, PA, a, middle, depth);
+#endif
+  for(k = SS_BLOCKSIZE; i != 0; k <<= 1, i >>= 1) {
+    if(i & 1) {
+      ss_swapmerge(T, PA, a - k, a, middle, buf, bufsize, depth);
+      a -= k;
+    }
+  }
+  if(limit != 0) {
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+    ss_mintrosort(T, PA, middle, last, depth);
+#elif 1 < SS_BLOCKSIZE
+    ss_insertionsort(T, PA, middle, last, depth);
+#endif
+    ss_inplacemerge(T, PA, first, middle, last, depth);
+  }
+#endif
+
+  if(lastsuffix != 0) {
+    /* Insert last type B* suffix. */
+    int PAi[2]; PAi[0] = PA[*(first - 1)], PAi[1] = n - 2;
+    for(a = first, i = *(first - 1);
+        (a < last) && ((*a < 0) || (0 < ss_compare(T, &(PAi[0]), PA + *a, depth)));
+        ++a) {
+      *(a - 1) = *a;
+    }
+    *(a - 1) = i;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+int
+tr_ilg(int n) {
+  return (n & 0xffff0000) ?
+          ((n & 0xff000000) ?
+            24 + lg_table[(n >> 24) & 0xff] :
+            16 + lg_table[(n >> 16) & 0xff]) :
+          ((n & 0x0000ff00) ?
+             8 + lg_table[(n >>  8) & 0xff] :
+             0 + lg_table[(n >>  0) & 0xff]);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Simple insertionsort for small size groups. */
+static
+void
+tr_insertionsort(const int *ISAd, int *first, int *last) {
+  int *a, *b;
+  int t, r;
+
+  for(a = first + 1; a < last; ++a) {
+    for(t = *a, b = a - 1; 0 > (r = ISAd[t] - ISAd[*b]);) {
+      do { *(b + 1) = *b; } while((first <= --b) && (*b < 0));
+      if(b < first) { break; }
+    }
+    if(r == 0) { *b = ~*b; }
+    *(b + 1) = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+void
+tr_fixdown(const int *ISAd, int *SA, int i, int size) {
+  int j, k;
+  int v;
+  int c, d, e;
+
+  for(v = SA[i], c = ISAd[v]; (j = 2 * i + 1) < size; SA[i] = SA[k], i = k) {
+    d = ISAd[SA[k = j++]];
+    if(d < (e = ISAd[SA[j]])) { k = j; d = e; }
+    if(d <= c) { break; }
+  }
+  SA[i] = v;
+}
+
+/* Simple top-down heapsort. */
+static
+void
+tr_heapsort(const int *ISAd, int *SA, int size) {
+  int i, m;
+  int t;
+
+  m = size;
+  if((size % 2) == 0) {
+    m--;
+    if(ISAd[SA[m / 2]] < ISAd[SA[m]]) { SWAP(SA[m], SA[m / 2]); }
+  }
+
+  for(i = m / 2 - 1; 0 <= i; --i) { tr_fixdown(ISAd, SA, i, m); }
+  if((size % 2) == 0) { SWAP(SA[0], SA[m]); tr_fixdown(ISAd, SA, 0, m); }
+  for(i = m - 1; 0 < i; --i) {
+    t = SA[0], SA[0] = SA[i];
+    tr_fixdown(ISAd, SA, 0, i);
+    SA[i] = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Returns the median of three elements. */
+static INLINE
+int *
+tr_median3(const int *ISAd, int *v1, int *v2, int *v3) {
+  int *t;
+  if(ISAd[*v1] > ISAd[*v2]) { SWAP(v1, v2); }
+  if(ISAd[*v2] > ISAd[*v3]) {
+    if(ISAd[*v1] > ISAd[*v3]) { return v1; }
+    else { return v3; }
+  }
+  return v2;
+}
+
+/* Returns the median of five elements. */
+static INLINE
+int *
+tr_median5(const int *ISAd,
+           int *v1, int *v2, int *v3, int *v4, int *v5) {
+  int *t;
+  if(ISAd[*v2] > ISAd[*v3]) { SWAP(v2, v3); }
+  if(ISAd[*v4] > ISAd[*v5]) { SWAP(v4, v5); }
+  if(ISAd[*v2] > ISAd[*v4]) { SWAP(v2, v4); SWAP(v3, v5); }
+  if(ISAd[*v1] > ISAd[*v3]) { SWAP(v1, v3); }
+  if(ISAd[*v1] > ISAd[*v4]) { SWAP(v1, v4); SWAP(v3, v5); }
+  if(ISAd[*v3] > ISAd[*v4]) { return v4; }
+  return v3;
+}
+
+/* Returns the pivot element. */
+static INLINE
+int *
+tr_pivot(const int *ISAd, int *first, int *last) {
+  int *middle;
+  int t;
+
+  t = last - first;
+  middle = first + t / 2;
+
+  if(t <= 512) {
+    if(t <= 32) {
+      return tr_median3(ISAd, first, middle, last - 1);
+    } else {
+      t >>= 2;
+      return tr_median5(ISAd, first, first + t, middle, last - 1 - t, last - 1);
+    }
+  }
+  t >>= 3;
+  first  = tr_median3(ISAd, first, first + t, first + (t << 1));
+  middle = tr_median3(ISAd, middle - t, middle, middle + t);
+  last   = tr_median3(ISAd, last - 1 - (t << 1), last - 1 - t, last - 1);
+  return tr_median3(ISAd, first, middle, last);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+typedef struct _trbudget_t trbudget_t;
+struct _trbudget_t {
+  int chance;
+  int remain;
+  int incval;
+  int count;
+};
+
+static INLINE
+void
+trbudget_init(trbudget_t *budget, int chance, int incval) {
+  budget->chance = chance;
+  budget->remain = budget->incval = incval;
+}
+
+static INLINE
+int
+trbudget_check(trbudget_t *budget, int size) {
+  if(size <= budget->remain) { budget->remain -= size; return 1; }
+  if(budget->chance == 0) { budget->count += size; return 0; }
+  budget->remain += budget->incval - size;
+  budget->chance -= 1;
+  return 1;
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+void
+tr_partition(const int *ISAd,
+             int *first, int *middle, int *last,
+             int **pa, int **pb, int v) {
+  int *a, *b, *c, *d, *e, *f;
+  int t, s;
+  int x = 0;
+
+  for(b = middle - 1; (++b < last) && ((x = ISAd[*b]) == v);) { }
+  if(((a = b) < last) && (x < v)) {
+    for(; (++b < last) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+  }
+  for(c = last; (b < --c) && ((x = ISAd[*c]) == v);) { }
+  if((b < (d = c)) && (x > v)) {
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+  for(; b < c;) {
+    SWAP(*b, *c);
+    for(; (++b < c) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+
+  if(a <= d) {
+    c = b - 1;
+    if((s = a - first) > (t = b - a)) { s = t; }
+    for(e = first, f = b - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    if((s = d - c) > (t = last - d - 1)) { s = t; }
+    for(e = b, f = last - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    first += (b - a), last -= (d - c);
+  }
+  *pa = first, *pb = last;
+}
+
+static
+void
+tr_copy(int *ISA, const int *SA,
+        int *first, int *a, int *b, int *last,
+        int depth) {
+  /* sort suffixes of middle partition
+     by using sorted order of suffixes of left and right partition. */
+  int *c, *d, *e;
+  int s, v;
+
+  v = b - SA - 1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      ISA[s] = d - SA;
+    }
+  }
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      ISA[s] = d - SA;
+    }
+  }
+}
+
+static
+void
+tr_partialcopy(int *ISA, const int *SA,
+               int *first, int *a, int *b, int *last,
+               int depth) {
+  int *c, *d, *e;
+  int s, v;
+  int rank, lastrank, newrank = -1;
+
+  v = b - SA - 1;
+  lastrank = -1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+
+  lastrank = -1;
+  for(e = d; first <= e; --e) {
+    rank = ISA[*e];
+    if(lastrank != rank) { lastrank = rank; newrank = e - SA; }
+    if(newrank != rank) { ISA[*e] = newrank; }
+  }
+
+  lastrank = -1;
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+}
+
+static
+void
+tr_introsort(int *ISA, const int *ISAd,
+             int *SA, int *first, int *last,
+             trbudget_t *budget) {
+#define STACK_SIZE TR_STACKSIZE
+  struct { const int *a; int *b, *c; int d, e; }stack[STACK_SIZE];
+  int *a, *b, *c;
+  int t;
+  int v, x = 0;
+  int incr = ISAd - ISA;
+  int limit, next;
+  int ssize, trlink = -1;
+
+  for(ssize = 0, limit = tr_ilg(last - first);;) {
+
+    if(limit < 0) {
+      if(limit == -1) {
+        /* tandem repeat partition */
+        tr_partition(ISAd - incr, first, first, last, &a, &b, last - SA - 1);
+
+        /* update ranks */
+        if(a < last) {
+          for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+        }
+        if(b < last) {
+          for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; }
+        }
+
+        /* push */
+        if(1 < (b - a)) {
+          STACK_PUSH5(NULL, a, b, 0, 0);
+          STACK_PUSH5(ISAd - incr, first, last, -2, trlink);
+          trlink = ssize - 2;
+        }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, tr_ilg(last - b), trlink);
+            last = a, limit = tr_ilg(a - first);
+          } else if(1 < (last - b)) {
+            first = b, limit = tr_ilg(last - b);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, tr_ilg(a - first), trlink);
+            first = b, limit = tr_ilg(last - b);
+          } else if(1 < (a - first)) {
+            last = a, limit = tr_ilg(a - first);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      } else if(limit == -2) {
+        /* tandem repeat copy */
+        a = stack[--ssize].b, b = stack[ssize].c;
+        if(stack[ssize].d == 0) {
+          tr_copy(ISA, SA, first, a, b, last, ISAd - ISA);
+        } else {
+          if(0 <= trlink) { stack[trlink].d = -1; }
+          tr_partialcopy(ISA, SA, first, a, b, last, ISAd - ISA);
+        }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      } else {
+        /* sorted partition */
+        if(0 <= *first) {
+          a = first;
+          do { ISA[*a] = a - SA; } while((++a < last) && (0 <= *a));
+          first = a;
+        }
+        if(first < last) {
+          a = first; do { *a = ~*a; } while(*++a < 0);
+          next = (ISA[*a] != ISAd[*a]) ? tr_ilg(a - first + 1) : -1;
+          if(++a < last) { for(b = first, v = a - SA - 1; b < a; ++b) { ISA[*b] = v; } }
+
+          /* push */
+          if(trbudget_check(budget, a - first)) {
+            if((a - first) <= (last - a)) {
+              STACK_PUSH5(ISAd, a, last, -3, trlink);
+              ISAd += incr, last = a, limit = next;
+            } else {
+              if(1 < (last - a)) {
+                STACK_PUSH5(ISAd + incr, first, a, next, trlink);
+                first = a, limit = -3;
+              } else {
+                ISAd += incr, last = a, limit = next;
+              }
+            }
+          } else {
+            if(0 <= trlink) { stack[trlink].d = -1; }
+            if(1 < (last - a)) {
+              first = a, limit = -3;
+            } else {
+              STACK_POP5(ISAd, first, last, limit, trlink);
+            }
+          }
+        } else {
+          STACK_POP5(ISAd, first, last, limit, trlink);
+        }
+      }
+      continue;
+    }
+
+    if((last - first) <= TR_INSERTIONSORT_THRESHOLD) {
+      tr_insertionsort(ISAd, first, last);
+      limit = -3;
+      continue;
+    }
+
+    if(limit-- == 0) {
+      tr_heapsort(ISAd, first, last - first);
+      for(a = last - 1; first < a; a = b) {
+        for(x = ISAd[*a], b = a - 1; (first <= b) && (ISAd[*b] == x); --b) { *b = ~*b; }
+      }
+      limit = -3;
+      continue;
+    }
+
+    /* choose pivot */
+    a = tr_pivot(ISAd, first, last);
+    SWAP(*first, *a);
+    v = ISAd[*first];
+
+    /* partition */
+    tr_partition(ISAd, first, first + 1, last, &a, &b, v);
+    if((last - first) != (b - a)) {
+      next = (ISA[*a] != v) ? tr_ilg(b - a) : -1;
+
+      /* update ranks */
+      for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+      if(b < last) { for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; } }
+
+      /* push */
+      if((1 < (b - a)) && (trbudget_check(budget, b - a))) {
+        if((a - first) <= (last - b)) {
+          if((last - b) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              last = a;
+            } else if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((a - first) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        } else {
+          if((a - first) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              first = b;
+            } else if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((last - b) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        }
+      } else {
+        if((1 < (b - a)) && (0 <= trlink)) { stack[trlink].d = -1; }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            last = a;
+          } else if(1 < (last - b)) {
+            first = b;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            first = b;
+          } else if(1 < (a - first)) {
+            last = a;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      }
+    } else {
+      if(trbudget_check(budget, last - first)) {
+        limit = tr_ilg(last - first), ISAd += incr;
+      } else {
+        if(0 <= trlink) { stack[trlink].d = -1; }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      }
+    }
+  }
+#undef STACK_SIZE
+}
+
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Tandem repeat sort */
+static
+void
+trsort(int *ISA, int *SA, int n, int depth) {
+  int *ISAd;
+  int *first, *last;
+  trbudget_t budget;
+  int t, skip, unsorted;
+
+  trbudget_init(&budget, tr_ilg(n) * 2 / 3, n);
+/*  trbudget_init(&budget, tr_ilg(n) * 3 / 4, n); */
+  for(ISAd = ISA + depth; -n < *SA; ISAd += ISAd - ISA) {
+    first = SA;
+    skip = 0;
+    unsorted = 0;
+    do {
+      if((t = *first) < 0) { first -= t; skip += t; }
+      else {
+        if(skip != 0) { *(first + skip) = skip; skip = 0; }
+        last = SA + ISA[t] + 1;
+        if(1 < (last - first)) {
+          budget.count = 0;
+          tr_introsort(ISA, ISAd, SA, first, last, &budget);
+          if(budget.count != 0) { unsorted += budget.count; }
+          else { skip = first - last; }
+        } else if((last - first) == 1) {
+          skip = -1;
+        }
+        first = last;
+      }
+    } while(first < (SA + n));
+    if(skip != 0) { *(first + skip) = skip; }
+    if(unsorted == 0) { break; }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Sorts suffixes of type B*. */
+static
+int
+sort_typeBstar(const unsigned char *T, int *SA,
+               int *bucket_A, int *bucket_B,
+               int n, int openMP) {
+  int *PAb, *ISAb, *buf;
+#ifdef LIBBSC_OPENMP
+  int *curbuf;
+  int l;
+#endif
+  int i, j, k, t, m, bufsize;
+  int c0, c1;
+#ifdef LIBBSC_OPENMP
+  int d0, d1;
+#endif
+  (void)openMP;
+
+  /* Initialize bucket arrays. */
+  for(i = 0; i < BUCKET_A_SIZE; ++i) { bucket_A[i] = 0; }
+  for(i = 0; i < BUCKET_B_SIZE; ++i) { bucket_B[i] = 0; }
+
+  /* Count the number of occurrences of the first one or two characters of each
+     type A, B and B* suffix. Moreover, store the beginning position of all
+     type B* suffixes into the array SA. */
+  for(i = n - 1, m = n, c0 = T[n - 1]; 0 <= i;) {
+    /* type A suffix. */
+    do { ++BUCKET_A(c1 = c0); } while((0 <= --i) && ((c0 = T[i]) >= c1));
+    if(0 <= i) {
+      /* type B* suffix. */
+      ++BUCKET_BSTAR(c0, c1);
+      SA[--m] = i;
+      /* type B suffix. */
+      for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) <= c1); --i, c1 = c0) {
+        ++BUCKET_B(c0, c1);
+      }
+    }
+  }
+  m = n - m;
+/*
+note:
+  A type B* suffix is lexicographically smaller than a type B suffix that
+  begins with the same first two characters.
+*/
+
+  /* Calculate the index of start/end point of each bucket. */
+  for(c0 = 0, i = 0, j = 0; c0 < ALPHABET_SIZE; ++c0) {
+    t = i + BUCKET_A(c0);
+    BUCKET_A(c0) = i + j; /* start point */
+    i = t + BUCKET_B(c0, c0);
+    for(c1 = c0 + 1; c1 < ALPHABET_SIZE; ++c1) {
+      j += BUCKET_BSTAR(c0, c1);
+      BUCKET_BSTAR(c0, c1) = j; /* end point */
+      i += BUCKET_B(c0, c1);
+    }
+  }
+
+  if(0 < m) {
+    /* Sort the type B* suffixes by their first two characters. */
+    PAb = SA + n - m; ISAb = SA + m;
+    for(i = m - 2; 0 <= i; --i) {
+      t = PAb[i], c0 = T[t], c1 = T[t + 1];
+      SA[--BUCKET_BSTAR(c0, c1)] = i;
+    }
+    t = PAb[m - 1], c0 = T[t], c1 = T[t + 1];
+    SA[--BUCKET_BSTAR(c0, c1)] = m - 1;
+
+    /* Sort the type B* substrings using sssort. */
+#ifdef LIBBSC_OPENMP
+    if (openMP)
+    {
+        buf = SA + m;
+        c0 = ALPHABET_SIZE - 2, c1 = ALPHABET_SIZE - 1, j = m;
+#pragma omp parallel default(shared) private(bufsize, curbuf, k, l, d0, d1)
+        {
+          bufsize = (n - (2 * m)) / omp_get_num_threads();
+          curbuf = buf + omp_get_thread_num() * bufsize;
+          k = 0;
+          for(;;) {
+            #pragma omp critical(sssort_lock)
+            {
+              if(0 < (l = j)) {
+                d0 = c0, d1 = c1;
+                do {
+                  k = BUCKET_BSTAR(d0, d1);
+                  if(--d1 <= d0) {
+                    d1 = ALPHABET_SIZE - 1;
+                    if(--d0 < 0) { break; }
+                  }
+                } while(((l - k) <= 1) && (0 < (l = k)));
+                c0 = d0, c1 = d1, j = k;
+              }
+            }
+            if(l == 0) { break; }
+            sssort(T, PAb, SA + k, SA + l,
+                   curbuf, bufsize, 2, n, *(SA + k) == (m - 1));
+          }
+        }
+    }
+    else
+    {
+        buf = SA + m, bufsize = n - (2 * m);
+        for(c0 = ALPHABET_SIZE - 2, j = m; 0 < j; --c0) {
+          for(c1 = ALPHABET_SIZE - 1; c0 < c1; j = i, --c1) {
+            i = BUCKET_BSTAR(c0, c1);
+            if(1 < (j - i)) {
+              sssort(T, PAb, SA + i, SA + j,
+                     buf, bufsize, 2, n, *(SA + i) == (m - 1));
+            }
+          }
+        }
+    }
+#else
+    buf = SA + m, bufsize = n - (2 * m);
+    for(c0 = ALPHABET_SIZE - 2, j = m; 0 < j; --c0) {
+      for(c1 = ALPHABET_SIZE - 1; c0 < c1; j = i, --c1) {
+        i = BUCKET_BSTAR(c0, c1);
+        if(1 < (j - i)) {
+          sssort(T, PAb, SA + i, SA + j,
+                 buf, bufsize, 2, n, *(SA + i) == (m - 1));
+        }
+      }
+    }
+#endif
+
+    /* Compute ranks of type B* substrings. */
+    for(i = m - 1; 0 <= i; --i) {
+      if(0 <= SA[i]) {
+        j = i;
+        do { ISAb[SA[i]] = i; } while((0 <= --i) && (0 <= SA[i]));
+        SA[i + 1] = i - j;
+        if(i <= 0) { break; }
+      }
+      j = i;
+      do { ISAb[SA[i] = ~SA[i]] = j; } while(SA[--i] < 0);
+      ISAb[SA[i]] = j;
+    }
+
+    /* Construct the inverse suffix array of type B* suffixes using trsort. */
+    trsort(ISAb, SA, m, 1);
+
+    /* Set the sorted order of type B* suffixes. */
+    for(i = n - 1, j = m, c0 = T[n - 1]; 0 <= i;) {
+      for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) >= c1); --i, c1 = c0) { }
+      if(0 <= i) {
+        t = i;
+        for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) <= c1); --i, c1 = c0) { }
+        SA[ISAb[--j]] = ((t == 0) || (1 < (t - i))) ? t : ~t;
+      }
+    }
+
+    /* Calculate the index of start/end point of each bucket. */
+    BUCKET_B(ALPHABET_SIZE - 1, ALPHABET_SIZE - 1) = n; /* end point */
+    for(c0 = ALPHABET_SIZE - 2, k = m - 1; 0 <= c0; --c0) {
+      i = BUCKET_A(c0 + 1) - 1;
+      for(c1 = ALPHABET_SIZE - 1; c0 < c1; --c1) {
+        t = i - BUCKET_B(c0, c1);
+        BUCKET_B(c0, c1) = i; /* end point */
+
+        /* Move all type B* suffixes to the correct position. */
+        for(i = t, j = BUCKET_BSTAR(c0, c1);
+            j <= k;
+            --i, --k) { SA[i] = SA[k]; }
+      }
+      BUCKET_BSTAR(c0, c0 + 1) = i - BUCKET_B(c0, c0) + 1; /* start point */
+      BUCKET_B(c0, c0) = i; /* end point */
+    }
+  }
+
+  return m;
+}
+
+/* Constructs the suffix array by using the sorted order of type B* suffixes. */
+static
+void
+construct_SA(const unsigned char *T, int *SA,
+             int *bucket_A, int *bucket_B,
+             int n, int m) {
+  int *i, *j, *k;
+  int s;
+  int c0, c1, c2;
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+          *j = ~s;
+          c0 = T[--s];
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j); assert(k != NULL);
+          *k-- = s;
+        } else {
+          assert(((s == 0) && (T[s] == c1)) || (s < 0));
+          *j = ~s;
+        }
+      }
+    }
+  }
+
+  /* Construct the suffix array by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  *k++ = (T[n - 2] < c2) ? ~(n - 1) : (n - 1);
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+      c0 = T[--s];
+      if((s == 0) || (T[s - 1] < c0)) { s = ~s; }
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      *k++ = s;
+    } else {
+      assert(s < 0);
+      *i = ~s;
+    }
+  }
+}
+
+/* Constructs the burrows-wheeler transformed string directly
+   by using the sorted order of type B* suffixes. */
+static
+int
+construct_BWT(const unsigned char *T, int *SA,
+              int *bucket_A, int *bucket_B,
+              int n, int m) {
+  int *i, *j, *k, *orig;
+  int s;
+  int c0, c1, c2;
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+          c0 = T[--s];
+          *j = ~((int)c0);
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j); assert(k != NULL);
+          *k-- = s;
+        } else if(s != 0) {
+          *j = ~s;
+#ifndef NDEBUG
+        } else {
+          assert(T[s] == c1);
+#endif
+        }
+      }
+    }
+  }
+
+  /* Construct the BWTed string by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  *k++ = (T[n - 2] < c2) ? ~((int)T[n - 2]) : (n - 1);
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n, orig = SA; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+      c0 = T[--s];
+      *i = c0;
+      if((0 < s) && (T[s - 1] < c0)) { s = ~((int)T[s - 1]); }
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      *k++ = s;
+    } else if(s != 0) {
+      *i = ~s;
+    } else {
+      orig = i;
+    }
+  }
+
+  return orig - SA;
+}
+
+/* Constructs the burrows-wheeler transformed string directly
+   by using the sorted order of type B* suffixes. */
+static
+int
+construct_BWT_indexes(const unsigned char *T, int *SA,
+                      int *bucket_A, int *bucket_B,
+                      int n, int m,
+                      unsigned char * num_indexes, int * indexes) {
+  int *i, *j, *k, *orig;
+  int s;
+  int c0, c1, c2;
+
+  int mod = n / 8;
+  {
+      mod |= mod >> 1;  mod |= mod >> 2;
+      mod |= mod >> 4;  mod |= mod >> 8;
+      mod |= mod >> 16; mod >>= 1;
+
+      *num_indexes = (unsigned char)((n - 1) / (mod + 1));
+  }
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+
+          if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = j - SA;
+
+          c0 = T[--s];
+          *j = ~((int)c0);
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j); assert(k != NULL);
+          *k-- = s;
+        } else if(s != 0) {
+          *j = ~s;
+#ifndef NDEBUG
+        } else {
+          assert(T[s] == c1);
+#endif
+        }
+      }
+    }
+  }
+
+  /* Construct the BWTed string by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  if (T[n - 2] < c2) {
+    if (((n - 1) & mod) == 0) indexes[(n - 1) / (mod + 1) - 1] = k - SA;
+    *k++ = ~((int)T[n - 2]);
+  }
+  else {
+    *k++ = n - 1;
+  }
+
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n, orig = SA; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+
+      if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = i - SA;
+
+      c0 = T[--s];
+      *i = c0;
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      if((0 < s) && (T[s - 1] < c0)) {
+          if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = k - SA;
+          *k++ = ~((int)T[s - 1]);
+      } else
+        *k++ = s;
+    } else if(s != 0) {
+      *i = ~s;
+    } else {
+      orig = i;
+    }
+  }
+
+  return orig - SA;
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/*- Function -*/
+
+int
+divsufsort(const unsigned char *T, int *SA, int n, int openMP) {
+  int *bucket_A, *bucket_B;
+  int m;
+  int err = 0;
+
+  /* Check arguments. */
+  if((T == NULL) || (SA == NULL) || (n < 0)) { return -1; }
+  else if(n == 0) { return 0; }
+  else if(n == 1) { SA[0] = 0; return 0; }
+  else if(n == 2) { m = (T[0] < T[1]); SA[m ^ 1] = 0, SA[m] = 1; return 0; }
+
+  bucket_A = (int *)malloc(BUCKET_A_SIZE * sizeof(int));
+  bucket_B = (int *)malloc(BUCKET_B_SIZE * sizeof(int));
+
+  /* Suffixsort. */
+  if((bucket_A != NULL) && (bucket_B != NULL)) {
+    m = sort_typeBstar(T, SA, bucket_A, bucket_B, n, openMP);
+    construct_SA(T, SA, bucket_A, bucket_B, n, m);
+  } else {
+    err = -2;
+  }
+
+  free(bucket_B);
+  free(bucket_A);
+
+  return err;
+}
+
+int
+divbwt(const unsigned char *T, unsigned char *U, int *A, int n, unsigned char * num_indexes, int * indexes, int openMP) {
+  int *B;
+  int *bucket_A, *bucket_B;
+  int m, pidx, i;
+
+  /* Check arguments. */
+  if((T == NULL) || (U == NULL) || (n < 0)) { return -1; }
+  else if(n <= 1) { if(n == 1) { U[0] = T[0]; } return n; }
+
+  if((B = A) == NULL) { B = (int *)malloc((size_t)(n + 1) * sizeof(int)); }
+  bucket_A = (int *)malloc(BUCKET_A_SIZE * sizeof(int));
+  bucket_B = (int *)malloc(BUCKET_B_SIZE * sizeof(int));
+
+  /* Burrows-Wheeler Transform. */
+  if((B != NULL) && (bucket_A != NULL) && (bucket_B != NULL)) {
+    m = sort_typeBstar(T, B, bucket_A, bucket_B, n, openMP);
+
+    if (num_indexes == NULL || indexes == NULL) {
+        pidx = construct_BWT(T, B, bucket_A, bucket_B, n, m);
+    } else {
+        pidx = construct_BWT_indexes(T, B, bucket_A, bucket_B, n, m, num_indexes, indexes);
+    }
+
+    /* Copy to output string. */
+    U[0] = T[n - 1];
+    for(i = 0; i < pidx; ++i) { U[i + 1] = (unsigned char)B[i]; }
+    for(i += 1; i < n; ++i) { U[i] = (unsigned char)B[i]; }
+    pidx += 1;
+  } else {
+    pidx = -2;
+  }
+
+  free(bucket_B);
+  free(bucket_A);
+  if(A == NULL) { free(B); }
+
+  return pidx;
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/divsufsort.h b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/divsufsort.h
new file mode 100644
index 0000000..5440994
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/divsufsort.h
@@ -0,0 +1,67 @@
+/*
+ * divsufsort.h for libdivsufsort-lite
+ * Copyright (c) 2003-2008 Yuta Mori All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef _DIVSUFSORT_H
+#define _DIVSUFSORT_H 1
+
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+
+
+/*- Prototypes -*/
+
+/**
+ * Constructs the suffix array of a given string.
+ * @param T [0..n-1] The input string.
+ * @param SA [0..n-1] The output array of suffixes.
+ * @param n The length of the given string.
+ * @param openMP enables OpenMP optimization.
+ * @return 0 if no error occurred, -1 or -2 otherwise.
+ */
+int
+divsufsort(const unsigned char *T, int *SA, int n, int openMP);
+
+/**
+ * Constructs the burrows-wheeler transformed string of a given string.
+ * @param T [0..n-1] The input string.
+ * @param U [0..n-1] The output string. (can be T)
+ * @param A [0..n-1] The temporary array. (can be NULL)
+ * @param n The length of the given string.
+ * @param num_indexes The length of secondary indexes array. (can be NULL)
+ * @param indexes The secondary indexes array. (can be NULL)
+ * @param openMP enables OpenMP optimization.
+ * @return The primary index if no error occurred, -1 or -2 otherwise.
+ */
+int
+divbwt(const unsigned char *T, unsigned char *U, int *A, int n, unsigned char * num_indexes, int * indexes, int openMP);
+
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif /* __cplusplus */
+
+#endif /* _DIVSUFSORT_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/fastcover.c b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/fastcover.c
new file mode 100644
index 0000000..46bba01
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/fastcover.c
@@ -0,0 +1,766 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#  define ZDICT_STATIC_LINKING_ONLY
+#endif
+
+#include "../common/mem.h" /* read */
+#include "../common/pool.h"
+#include "../common/threading.h"
+#include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../compress/zstd_compress_internal.h" /* ZSTD_hash*() */
+#include "../zdict.h"
+#include "cover.h"
+
+
+/*-*************************************
+*  Constants
+***************************************/
+/**
+* There are 32bit indexes used to ref samples, so limit samples size to 4GB
+* on 64bit builds.
+* For 32bit builds we choose 1 GB.
+* Most 32bit platforms have 2GB user-mode addressable space and we allocate a large
+* contiguous buffer, so 1GB is already a high limit.
+*/
+#define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
+#define FASTCOVER_MAX_F 31
+#define FASTCOVER_MAX_ACCEL 10
+#define FASTCOVER_DEFAULT_SPLITPOINT 0.75
+#define DEFAULT_F 20
+#define DEFAULT_ACCEL 1
+
+
+/*-*************************************
+*  Console display
+***************************************/
+#ifndef LOCALDISPLAYLEVEL
+static int g_displayLevel = 0;
+#endif
+#undef  DISPLAY
+#define DISPLAY(...)                                                           \
+  {                                                                            \
+    fprintf(stderr, __VA_ARGS__);                                              \
+    fflush(stderr);                                                            \
+  }
+#undef  LOCALDISPLAYLEVEL
+#define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
+  if (displayLevel >= l) {                                                     \
+    DISPLAY(__VA_ARGS__);                                                      \
+  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#undef  DISPLAYLEVEL
+#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
+
+#ifndef LOCALDISPLAYUPDATE
+static const clock_t g_refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+#endif
+#undef  LOCALDISPLAYUPDATE
+#define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
+  if (displayLevel >= l) {                                                     \
+    if ((clock() - g_time > g_refreshRate) || (displayLevel >= 4)) {             \
+      g_time = clock();                                                        \
+      DISPLAY(__VA_ARGS__);                                                    \
+    }                                                                          \
+  }
+#undef  DISPLAYUPDATE
+#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
+
+
+/*-*************************************
+* Hash Functions
+***************************************/
+/**
+ * Hash the d-byte value pointed to by p and mod 2^f into the frequency vector
+ */
+static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 f, unsigned d) {
+  if (d == 6) {
+    return ZSTD_hash6Ptr(p, f);
+  }
+  return ZSTD_hash8Ptr(p, f);
+}
+
+
+/*-*************************************
+* Acceleration
+***************************************/
+typedef struct {
+  unsigned finalize;    /* Percentage of training samples used for ZDICT_finalizeDictionary */
+  unsigned skip;        /* Number of dmer skipped between each dmer counted in computeFrequency */
+} FASTCOVER_accel_t;
+
+
+static const FASTCOVER_accel_t FASTCOVER_defaultAccelParameters[FASTCOVER_MAX_ACCEL+1] = {
+  { 100, 0 },   /* accel = 0, should not happen because accel = 0 defaults to accel = 1 */
+  { 100, 0 },   /* accel = 1 */
+  { 50, 1 },   /* accel = 2 */
+  { 34, 2 },   /* accel = 3 */
+  { 25, 3 },   /* accel = 4 */
+  { 20, 4 },   /* accel = 5 */
+  { 17, 5 },   /* accel = 6 */
+  { 14, 6 },   /* accel = 7 */
+  { 13, 7 },   /* accel = 8 */
+  { 11, 8 },   /* accel = 9 */
+  { 10, 9 },   /* accel = 10 */
+};
+
+
+/*-*************************************
+* Context
+***************************************/
+typedef struct {
+  const BYTE *samples;
+  size_t *offsets;
+  const size_t *samplesSizes;
+  size_t nbSamples;
+  size_t nbTrainSamples;
+  size_t nbTestSamples;
+  size_t nbDmers;
+  U32 *freqs;
+  unsigned d;
+  unsigned f;
+  FASTCOVER_accel_t accelParams;
+} FASTCOVER_ctx_t;
+
+
+/*-*************************************
+*  Helper functions
+***************************************/
+/**
+ * Selects the best segment in an epoch.
+ * Segments of are scored according to the function:
+ *
+ * Let F(d) be the frequency of all dmers with hash value d.
+ * Let S_i be hash value of the dmer at position i of segment S which has length k.
+ *
+ *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
+ *
+ * Once the dmer with hash value d is in the dictionary we set F(d) = 0.
+ */
+static COVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx,
+                                              U32 *freqs, U32 begin, U32 end,
+                                              ZDICT_cover_params_t parameters,
+                                              U16* segmentFreqs) {
+  /* Constants */
+  const U32 k = parameters.k;
+  const U32 d = parameters.d;
+  const U32 f = ctx->f;
+  const U32 dmersInK = k - d + 1;
+
+  /* Try each segment (activeSegment) and save the best (bestSegment) */
+  COVER_segment_t bestSegment = {0, 0, 0};
+  COVER_segment_t activeSegment;
+
+  /* Reset the activeDmers in the segment */
+  /* The activeSegment starts at the beginning of the epoch. */
+  activeSegment.begin = begin;
+  activeSegment.end = begin;
+  activeSegment.score = 0;
+
+  /* Slide the activeSegment through the whole epoch.
+   * Save the best segment in bestSegment.
+   */
+  while (activeSegment.end < end) {
+    /* Get hash value of current dmer */
+    const size_t idx = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.end, f, d);
+
+    /* Add frequency of this index to score if this is the first occurrence of index in active segment */
+    if (segmentFreqs[idx] == 0) {
+      activeSegment.score += freqs[idx];
+    }
+    /* Increment end of segment and segmentFreqs*/
+    activeSegment.end += 1;
+    segmentFreqs[idx] += 1;
+    /* If the window is now too large, drop the first position */
+    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
+      /* Get hash value of the dmer to be eliminated from active segment */
+      const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+      segmentFreqs[delIndex] -= 1;
+      /* Subtract frequency of this index from score if this is the last occurrence of this index in active segment */
+      if (segmentFreqs[delIndex] == 0) {
+        activeSegment.score -= freqs[delIndex];
+      }
+      /* Increment start of segment */
+      activeSegment.begin += 1;
+    }
+
+    /* If this segment is the best so far save it */
+    if (activeSegment.score > bestSegment.score) {
+      bestSegment = activeSegment;
+    }
+  }
+
+  /* Zero out rest of segmentFreqs array */
+  while (activeSegment.begin < end) {
+    const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+    segmentFreqs[delIndex] -= 1;
+    activeSegment.begin += 1;
+  }
+
+  {
+    /*  Zero the frequency of hash value of each dmer covered by the chosen segment. */
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, f, d);
+      freqs[i] = 0;
+    }
+  }
+
+  return bestSegment;
+}
+
+
+static int FASTCOVER_checkParameters(ZDICT_cover_params_t parameters,
+                                     size_t maxDictSize, unsigned f,
+                                     unsigned accel) {
+  /* k, d, and f are required parameters */
+  if (parameters.d == 0 || parameters.k == 0) {
+    return 0;
+  }
+  /* d has to be 6 or 8 */
+  if (parameters.d != 6 && parameters.d != 8) {
+    return 0;
+  }
+  /* k <= maxDictSize */
+  if (parameters.k > maxDictSize) {
+    return 0;
+  }
+  /* d <= k */
+  if (parameters.d > parameters.k) {
+    return 0;
+  }
+  /* 0 < f <= FASTCOVER_MAX_F*/
+  if (f > FASTCOVER_MAX_F || f == 0) {
+    return 0;
+  }
+  /* 0 < splitPoint <= 1 */
+  if (parameters.splitPoint <= 0 || parameters.splitPoint > 1) {
+    return 0;
+  }
+  /* 0 < accel <= 10 */
+  if (accel > 10 || accel == 0) {
+    return 0;
+  }
+  return 1;
+}
+
+
+/**
+ * Clean up a context initialized with `FASTCOVER_ctx_init()`.
+ */
+static void
+FASTCOVER_ctx_destroy(FASTCOVER_ctx_t* ctx)
+{
+    if (!ctx) return;
+
+    free(ctx->freqs);
+    ctx->freqs = NULL;
+
+    free(ctx->offsets);
+    ctx->offsets = NULL;
+}
+
+
+/**
+ * Calculate for frequency of hash value of each dmer in ctx->samples
+ */
+static void
+FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
+{
+    const unsigned f = ctx->f;
+    const unsigned d = ctx->d;
+    const unsigned skip = ctx->accelParams.skip;
+    const unsigned readLength = MAX(d, 8);
+    size_t i;
+    assert(ctx->nbTrainSamples >= 5);
+    assert(ctx->nbTrainSamples <= ctx->nbSamples);
+    for (i = 0; i < ctx->nbTrainSamples; i++) {
+        size_t start = ctx->offsets[i];  /* start of current dmer */
+        size_t const currSampleEnd = ctx->offsets[i+1];
+        while (start + readLength <= currSampleEnd) {
+            const size_t dmerIndex = FASTCOVER_hashPtrToIndex(ctx->samples + start, f, d);
+            freqs[dmerIndex]++;
+            start = start + skip + 1;
+        }
+    }
+}
+
+
+/**
+ * Prepare a context for dictionary building.
+ * The context is only dependent on the parameter `d` and can be used multiple
+ * times.
+ * Returns 0 on success or error code on error.
+ * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
+ */
+static size_t
+FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
+                   const void* samplesBuffer,
+                   const size_t* samplesSizes, unsigned nbSamples,
+                   unsigned d, double splitPoint, unsigned f,
+                   FASTCOVER_accel_t accelParams)
+{
+    const BYTE* const samples = (const BYTE*)samplesBuffer;
+    const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+    /* Split samples into testing and training sets */
+    const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
+    const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
+    const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
+    const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
+
+    /* Checks */
+    if (totalSamplesSize < MAX(d, sizeof(U64)) ||
+        totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
+        DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
+                    (unsigned)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
+        return ERROR(srcSize_wrong);
+    }
+
+    /* Check if there are at least 5 training samples */
+    if (nbTrainSamples < 5) {
+        DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
+        return ERROR(srcSize_wrong);
+    }
+
+    /* Check if there's testing sample */
+    if (nbTestSamples < 1) {
+        DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
+        return ERROR(srcSize_wrong);
+    }
+
+    /* Zero the context */
+    memset(ctx, 0, sizeof(*ctx));
+    DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
+                    (unsigned)trainingSamplesSize);
+    DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
+                    (unsigned)testSamplesSize);
+
+    ctx->samples = samples;
+    ctx->samplesSizes = samplesSizes;
+    ctx->nbSamples = nbSamples;
+    ctx->nbTrainSamples = nbTrainSamples;
+    ctx->nbTestSamples = nbTestSamples;
+    ctx->nbDmers = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
+    ctx->d = d;
+    ctx->f = f;
+    ctx->accelParams = accelParams;
+
+    /* The offsets of each file */
+    ctx->offsets = (size_t*)calloc((nbSamples + 1), sizeof(size_t));
+    if (ctx->offsets == NULL) {
+        DISPLAYLEVEL(1, "Failed to allocate scratch buffers \n");
+        FASTCOVER_ctx_destroy(ctx);
+        return ERROR(memory_allocation);
+    }
+
+    /* Fill offsets from the samplesSizes */
+    {   U32 i;
+        ctx->offsets[0] = 0;
+        assert(nbSamples >= 5);
+        for (i = 1; i <= nbSamples; ++i) {
+            ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
+        }
+    }
+
+    /* Initialize frequency array of size 2^f */
+    ctx->freqs = (U32*)calloc(((U64)1 << f), sizeof(U32));
+    if (ctx->freqs == NULL) {
+        DISPLAYLEVEL(1, "Failed to allocate frequency table \n");
+        FASTCOVER_ctx_destroy(ctx);
+        return ERROR(memory_allocation);
+    }
+
+    DISPLAYLEVEL(2, "Computing frequencies\n");
+    FASTCOVER_computeFrequency(ctx->freqs, ctx);
+
+    return 0;
+}
+
+
+/**
+ * Given the prepared context build the dictionary.
+ */
+static size_t
+FASTCOVER_buildDictionary(const FASTCOVER_ctx_t* ctx,
+                          U32* freqs,
+                          void* dictBuffer, size_t dictBufferCapacity,
+                          ZDICT_cover_params_t parameters,
+                          U16* segmentFreqs)
+{
+  BYTE *const dict = (BYTE *)dictBuffer;
+  size_t tail = dictBufferCapacity;
+  /* Divide the data into epochs. We will select one segment from each epoch. */
+  const COVER_epoch_info_t epochs = COVER_computeEpochs(
+      (U32)dictBufferCapacity, (U32)ctx->nbDmers, parameters.k, 1);
+  const size_t maxZeroScoreRun = 10;
+  size_t zeroScoreRun = 0;
+  size_t epoch;
+  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n",
+                (U32)epochs.num, (U32)epochs.size);
+  /* Loop through the epochs until there are no more segments or the dictionary
+   * is full.
+   */
+  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs.num) {
+    const U32 epochBegin = (U32)(epoch * epochs.size);
+    const U32 epochEnd = epochBegin + epochs.size;
+    size_t segmentSize;
+    /* Select a segment */
+    COVER_segment_t segment = FASTCOVER_selectSegment(
+        ctx, freqs, epochBegin, epochEnd, parameters, segmentFreqs);
+
+    /* If the segment covers no dmers, then we are out of content.
+     * There may be new content in other epochs, for continue for some time.
+     */
+    if (segment.score == 0) {
+      if (++zeroScoreRun >= maxZeroScoreRun) {
+          break;
+      }
+      continue;
+    }
+    zeroScoreRun = 0;
+
+    /* Trim the segment if necessary and if it is too small then we are done */
+    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
+    if (segmentSize < parameters.d) {
+      break;
+    }
+
+    /* We fill the dictionary from the back to allow the best segments to be
+     * referenced with the smallest offsets.
+     */
+    tail -= segmentSize;
+    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
+    DISPLAYUPDATE(
+        2, "\r%u%%       ",
+        (unsigned)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
+  }
+  DISPLAYLEVEL(2, "\r%79s\r", "");
+  return tail;
+}
+
+/**
+ * Parameters for FASTCOVER_tryParameters().
+ */
+typedef struct FASTCOVER_tryParameters_data_s {
+    const FASTCOVER_ctx_t* ctx;
+    COVER_best_t* best;
+    size_t dictBufferCapacity;
+    ZDICT_cover_params_t parameters;
+} FASTCOVER_tryParameters_data_t;
+
+
+/**
+ * Tries a set of parameters and updates the COVER_best_t with the results.
+ * This function is thread safe if zstd is compiled with multithreaded support.
+ * It takes its parameters as an *OWNING* opaque pointer to support threading.
+ */
+static void FASTCOVER_tryParameters(void* opaque)
+{
+  /* Save parameters as local variables */
+  FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t*)opaque;
+  const FASTCOVER_ctx_t *const ctx = data->ctx;
+  const ZDICT_cover_params_t parameters = data->parameters;
+  size_t dictBufferCapacity = data->dictBufferCapacity;
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Initialize array to keep track of frequency of dmer within activeSegment */
+  U16* segmentFreqs = (U16*)calloc(((U64)1 << ctx->f), sizeof(U16));
+  /* Allocate space for hash table, dict, and freqs */
+  BYTE *const dict = (BYTE*)malloc(dictBufferCapacity);
+  COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
+  U32* freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
+  if (!segmentFreqs || !dict || !freqs) {
+    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
+    goto _cleanup;
+  }
+  /* Copy the frequencies because we need to modify them */
+  memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
+  /* Build the dictionary */
+  { const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
+                                                    parameters, segmentFreqs);
+
+    const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
+    selection = COVER_selectDict(dict + tail, dictBufferCapacity, dictBufferCapacity - tail,
+         ctx->samples, ctx->samplesSizes, nbFinalizeSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
+         totalCompressedSize);
+
+    if (COVER_dictSelectionIsError(selection)) {
+      DISPLAYLEVEL(1, "Failed to select dictionary\n");
+      goto _cleanup;
+    }
+  }
+_cleanup:
+  free(dict);
+  COVER_best_finish(data->best, parameters, selection);
+  free(data);
+  free(segmentFreqs);
+  COVER_dictSelectionFree(selection);
+  free(freqs);
+}
+
+
+static void
+FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
+                               ZDICT_cover_params_t* coverParams)
+{
+    coverParams->k = fastCoverParams.k;
+    coverParams->d = fastCoverParams.d;
+    coverParams->steps = fastCoverParams.steps;
+    coverParams->nbThreads = fastCoverParams.nbThreads;
+    coverParams->splitPoint = fastCoverParams.splitPoint;
+    coverParams->zParams = fastCoverParams.zParams;
+    coverParams->shrinkDict = fastCoverParams.shrinkDict;
+}
+
+
+static void
+FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
+                                   ZDICT_fastCover_params_t* fastCoverParams,
+                                   unsigned f, unsigned accel)
+{
+    fastCoverParams->k = coverParams.k;
+    fastCoverParams->d = coverParams.d;
+    fastCoverParams->steps = coverParams.steps;
+    fastCoverParams->nbThreads = coverParams.nbThreads;
+    fastCoverParams->splitPoint = coverParams.splitPoint;
+    fastCoverParams->f = f;
+    fastCoverParams->accel = accel;
+    fastCoverParams->zParams = coverParams.zParams;
+    fastCoverParams->shrinkDict = coverParams.shrinkDict;
+}
+
+
+ZDICTLIB_API size_t
+ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
+                                const void* samplesBuffer,
+                                const size_t* samplesSizes, unsigned nbSamples,
+                                ZDICT_fastCover_params_t parameters)
+{
+    BYTE* const dict = (BYTE*)dictBuffer;
+    FASTCOVER_ctx_t ctx;
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    /* Initialize global data */
+    g_displayLevel = (int)parameters.zParams.notificationLevel;
+    /* Assign splitPoint and f if not provided */
+    parameters.splitPoint = 1.0;
+    parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
+    parameters.accel = parameters.accel == 0 ? DEFAULT_ACCEL : parameters.accel;
+    /* Convert to cover parameter */
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(parameters, &coverParams);
+    /* Checks */
+    if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
+                                   parameters.accel)) {
+      DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (nbSamples == 0) {
+      DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
+      return ERROR(srcSize_wrong);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    /* Assign corresponding FASTCOVER_accel_t to accelParams*/
+    accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
+    /* Initialize context */
+    {
+      size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                            coverParams.d, parameters.splitPoint, parameters.f,
+                            accelParams);
+      if (ZSTD_isError(initVal)) {
+        DISPLAYLEVEL(1, "Failed to initialize context\n");
+        return initVal;
+      }
+    }
+    COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, g_displayLevel);
+    /* Build the dictionary */
+    DISPLAYLEVEL(2, "Building dictionary\n");
+    {
+      /* Initialize array to keep track of frequency of dmer within activeSegment */
+      U16* segmentFreqs = (U16 *)calloc(((U64)1 << parameters.f), sizeof(U16));
+      const size_t tail = FASTCOVER_buildDictionary(&ctx, ctx.freqs, dictBuffer,
+                                                dictBufferCapacity, coverParams, segmentFreqs);
+      const unsigned nbFinalizeSamples = (unsigned)(ctx.nbTrainSamples * ctx.accelParams.finalize / 100);
+      const size_t dictionarySize = ZDICT_finalizeDictionary(
+          dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+          samplesBuffer, samplesSizes, nbFinalizeSamples, coverParams.zParams);
+      if (!ZSTD_isError(dictionarySize)) {
+          DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
+                      (unsigned)dictionarySize);
+      }
+      FASTCOVER_ctx_destroy(&ctx);
+      free(segmentFreqs);
+      return dictionarySize;
+    }
+}
+
+
+ZDICTLIB_API size_t
+ZDICT_optimizeTrainFromBuffer_fastCover(
+                    void* dictBuffer, size_t dictBufferCapacity,
+                    const void* samplesBuffer,
+                    const size_t* samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t* parameters)
+{
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    /* constants */
+    const unsigned nbThreads = parameters->nbThreads;
+    const double splitPoint =
+        parameters->splitPoint <= 0.0 ? FASTCOVER_DEFAULT_SPLITPOINT : parameters->splitPoint;
+    const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
+    const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+    const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+    const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+    const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
+    const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
+    const unsigned kIterations =
+        (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+    const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
+    const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
+    const unsigned shrinkDict = 0;
+    /* Local variables */
+    const int displayLevel = (int)parameters->zParams.notificationLevel;
+    unsigned iteration = 1;
+    unsigned d;
+    unsigned k;
+    COVER_best_t best;
+    POOL_ctx *pool = NULL;
+    int warned = 0;
+    /* Checks */
+    if (splitPoint <= 0 || splitPoint > 1) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect splitPoint\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect accel\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (kMinK < kMaxD || kMaxK < kMinK) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect k\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (nbSamples == 0) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "FASTCOVER must have at least one input file\n");
+      return ERROR(srcSize_wrong);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    if (nbThreads > 1) {
+      pool = POOL_create(nbThreads, 1);
+      if (!pool) {
+        return ERROR(memory_allocation);
+      }
+    }
+    /* Initialization */
+    COVER_best_init(&best);
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(*parameters, &coverParams);
+    accelParams = FASTCOVER_defaultAccelParameters[accel];
+    /* Turn down global display level to clean up display at level 2 and below */
+    g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
+    /* Loop through d first because each new value needs a new context */
+    LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
+                      kIterations);
+    for (d = kMinD; d <= kMaxD; d += 2) {
+      /* Initialize the context for this value of d */
+      FASTCOVER_ctx_t ctx;
+      LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
+      {
+        size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams);
+        if (ZSTD_isError(initVal)) {
+          LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+          COVER_best_destroy(&best);
+          POOL_free(pool);
+          return initVal;
+        }
+      }
+      if (!warned) {
+        COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, displayLevel);
+        warned = 1;
+      }
+      /* Loop through k reusing the same context */
+      for (k = kMinK; k <= kMaxK; k += kStepSize) {
+        /* Prepare the arguments */
+        FASTCOVER_tryParameters_data_t *data = (FASTCOVER_tryParameters_data_t *)malloc(
+            sizeof(FASTCOVER_tryParameters_data_t));
+        LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
+        if (!data) {
+          LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
+          COVER_best_destroy(&best);
+          FASTCOVER_ctx_destroy(&ctx);
+          POOL_free(pool);
+          return ERROR(memory_allocation);
+        }
+        data->ctx = &ctx;
+        data->best = &best;
+        data->dictBufferCapacity = dictBufferCapacity;
+        data->parameters = coverParams;
+        data->parameters.k = k;
+        data->parameters.d = d;
+        data->parameters.splitPoint = splitPoint;
+        data->parameters.steps = kSteps;
+        data->parameters.shrinkDict = shrinkDict;
+        data->parameters.zParams.notificationLevel = (unsigned)g_displayLevel;
+        /* Check the parameters */
+        if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
+                                       data->ctx->f, accel)) {
+          DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+          free(data);
+          continue;
+        }
+        /* Call the function and pass ownership of data to it */
+        COVER_best_start(&best);
+        if (pool) {
+          POOL_add(pool, &FASTCOVER_tryParameters, data);
+        } else {
+          FASTCOVER_tryParameters(data);
+        }
+        /* Print status */
+        LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
+                           (unsigned)((iteration * 100) / kIterations));
+        ++iteration;
+      }
+      COVER_best_wait(&best);
+      FASTCOVER_ctx_destroy(&ctx);
+    }
+    LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
+    /* Fill the output buffer and parameters with output of the best parameters */
+    {
+      const size_t dictSize = best.dictSize;
+      if (ZSTD_isError(best.compressedSize)) {
+        const size_t compressedSize = best.compressedSize;
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return compressedSize;
+      }
+      FASTCOVER_convertToFastCoverParams(best.parameters, parameters, f, accel);
+      memcpy(dictBuffer, best.dict, dictSize);
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return dictSize;
+    }
+
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/zdict.c b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/zdict.c
new file mode 100644
index 0000000..58290f4
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/dictBuilder/zdict.c
@@ -0,0 +1,1127 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*-**************************************
+*  Tuning parameters
+****************************************/
+#define MINRATIO 4   /* minimum nb of apparition to be selected in dictionary */
+#define ZDICT_MAX_SAMPLES_SIZE (2000U << 20)
+#define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO)
+
+
+/*-**************************************
+*  Compiler Options
+****************************************/
+/* Unix Large Files support (>4GB) */
+#define _FILE_OFFSET_BITS 64
+#if (defined(__sun__) && (!defined(__LP64__)))   /* Sun Solaris 32-bits requires specific definitions */
+#  ifndef _LARGEFILE_SOURCE
+#  define _LARGEFILE_SOURCE
+#  endif
+#elif ! defined(__LP64__)                        /* No point defining Large file for 64 bit */
+#  ifndef _LARGEFILE64_SOURCE
+#  define _LARGEFILE64_SOURCE
+#  endif
+#endif
+
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdlib.h>        /* malloc, free */
+#include <string.h>        /* memset */
+#include <stdio.h>         /* fprintf, fopen, ftello64 */
+#include <time.h>          /* clock */
+
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#  define ZDICT_STATIC_LINKING_ONLY
+#endif
+
+#include "../common/mem.h"           /* read */
+#include "../common/fse.h"           /* FSE_normalizeCount, FSE_writeNCount */
+#include "../common/huf.h"           /* HUF_buildCTable, HUF_writeCTable */
+#include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../common/xxhash.h"        /* XXH64 */
+#include "../compress/zstd_compress_internal.h" /* ZSTD_loadCEntropy() */
+#include "../zdict.h"
+#include "divsufsort.h"
+#include "../common/bits.h"          /* ZSTD_NbCommonBytes */
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define DICTLISTSIZE_DEFAULT 10000
+
+#define NOISELENGTH 32
+
+static const U32 g_selectivity_default = 9;
+
+
+/*-*************************************
+*  Console display
+***************************************/
+#undef  DISPLAY
+#define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush( stderr ); }
+#undef  DISPLAYLEVEL
+#define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); }    /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+
+static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
+
+static void ZDICT_printHex(const void* ptr, size_t length)
+{
+    const BYTE* const b = (const BYTE*)ptr;
+    size_t u;
+    for (u=0; u<length; u++) {
+        BYTE c = b[u];
+        if (c<32 || c>126) c = '.';   /* non-printable char */
+        DISPLAY("%c", c);
+    }
+}
+
+
+/*-********************************************************
+*  Helper functions
+**********************************************************/
+unsigned ZDICT_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZDICT_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize)
+{
+    if (dictSize < 8) return 0;
+    if (MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return 0;
+    return MEM_readLE32((const char*)dictBuffer + 4);
+}
+
+size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize)
+{
+    size_t headerSize;
+    if (dictSize <= 8 || MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return ERROR(dictionary_corrupted);
+
+    {   ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t));
+        U32* wksp = (U32*)malloc(HUF_WORKSPACE_SIZE);
+        if (!bs || !wksp) {
+            headerSize = ERROR(memory_allocation);
+        } else {
+            ZSTD_reset_compressedBlockState(bs);
+            headerSize = ZSTD_loadCEntropy(bs, wksp, dictBuffer, dictSize);
+        }
+
+        free(bs);
+        free(wksp);
+    }
+
+    return headerSize;
+}
+
+/*-********************************************************
+*  Dictionary training functions
+**********************************************************/
+/*! ZDICT_count() :
+    Count the nb of common bytes between 2 pointers.
+    Note : this function presumes end of buffer followed by noisy guard band.
+*/
+static size_t ZDICT_count(const void* pIn, const void* pMatch)
+{
+    const char* const pStart = (const char*)pIn;
+    for (;;) {
+        size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+        if (!diff) {
+            pIn = (const char*)pIn+sizeof(size_t);
+            pMatch = (const char*)pMatch+sizeof(size_t);
+            continue;
+        }
+        pIn = (const char*)pIn+ZSTD_NbCommonBytes(diff);
+        return (size_t)((const char*)pIn - pStart);
+    }
+}
+
+
+typedef struct {
+    U32 pos;
+    U32 length;
+    U32 savings;
+} dictItem;
+
+static void ZDICT_initDictItem(dictItem* d)
+{
+    d->pos = 1;
+    d->length = 0;
+    d->savings = (U32)(-1);
+}
+
+
+#define LLIMIT 64          /* heuristic determined experimentally */
+#define MINMATCHLENGTH 7   /* heuristic determined experimentally */
+static dictItem ZDICT_analyzePos(
+                       BYTE* doneMarks,
+                       const int* suffix, U32 start,
+                       const void* buffer, U32 minRatio, U32 notificationLevel)
+{
+    U32 lengthList[LLIMIT] = {0};
+    U32 cumulLength[LLIMIT] = {0};
+    U32 savings[LLIMIT] = {0};
+    const BYTE* b = (const BYTE*)buffer;
+    size_t maxLength = LLIMIT;
+    size_t pos = (size_t)suffix[start];
+    U32 end = start;
+    dictItem solution;
+
+    /* init */
+    memset(&solution, 0, sizeof(solution));
+    doneMarks[pos] = 1;
+
+    /* trivial repetition cases */
+    if ( (MEM_read16(b+pos+0) == MEM_read16(b+pos+2))
+       ||(MEM_read16(b+pos+1) == MEM_read16(b+pos+3))
+       ||(MEM_read16(b+pos+2) == MEM_read16(b+pos+4)) ) {
+        /* skip and mark segment */
+        U16 const pattern16 = MEM_read16(b+pos+4);
+        U32 u, patternEnd = 6;
+        while (MEM_read16(b+pos+patternEnd) == pattern16) patternEnd+=2 ;
+        if (b[pos+patternEnd] == b[pos+patternEnd-1]) patternEnd++;
+        for (u=1; u<patternEnd; u++)
+            doneMarks[pos+u] = 1;
+        return solution;
+    }
+
+    /* look forward */
+    {   size_t length;
+        do {
+            end++;
+            length = ZDICT_count(b + pos, b + suffix[end]);
+        } while (length >= MINMATCHLENGTH);
+    }
+
+    /* look backward */
+    {   size_t length;
+        do {
+            length = ZDICT_count(b + pos, b + *(suffix+start-1));
+            if (length >=MINMATCHLENGTH) start--;
+        } while(length >= MINMATCHLENGTH);
+    }
+
+    /* exit if not found a minimum nb of repetitions */
+    if (end-start < minRatio) {
+        U32 idx;
+        for(idx=start; idx<end; idx++)
+            doneMarks[suffix[idx]] = 1;
+        return solution;
+    }
+
+    {   int i;
+        U32 mml;
+        U32 refinedStart = start;
+        U32 refinedEnd = end;
+
+        DISPLAYLEVEL(4, "\n");
+        DISPLAYLEVEL(4, "found %3u matches of length >= %i at pos %7u  ", (unsigned)(end-start), MINMATCHLENGTH, (unsigned)pos);
+        DISPLAYLEVEL(4, "\n");
+
+        for (mml = MINMATCHLENGTH ; ; mml++) {
+            BYTE currentChar = 0;
+            U32 currentCount = 0;
+            U32 currentID = refinedStart;
+            U32 id;
+            U32 selectedCount = 0;
+            U32 selectedID = currentID;
+            for (id =refinedStart; id < refinedEnd; id++) {
+                if (b[suffix[id] + mml] != currentChar) {
+                    if (currentCount > selectedCount) {
+                        selectedCount = currentCount;
+                        selectedID = currentID;
+                    }
+                    currentID = id;
+                    currentChar = b[ suffix[id] + mml];
+                    currentCount = 0;
+                }
+                currentCount ++;
+            }
+            if (currentCount > selectedCount) {  /* for last */
+                selectedCount = currentCount;
+                selectedID = currentID;
+            }
+
+            if (selectedCount < minRatio)
+                break;
+            refinedStart = selectedID;
+            refinedEnd = refinedStart + selectedCount;
+        }
+
+        /* evaluate gain based on new dict */
+        start = refinedStart;
+        pos = suffix[refinedStart];
+        end = start;
+        memset(lengthList, 0, sizeof(lengthList));
+
+        /* look forward */
+        {   size_t length;
+            do {
+                end++;
+                length = ZDICT_count(b + pos, b + suffix[end]);
+                if (length >= LLIMIT) length = LLIMIT-1;
+                lengthList[length]++;
+            } while (length >=MINMATCHLENGTH);
+        }
+
+        /* look backward */
+        {   size_t length = MINMATCHLENGTH;
+            while ((length >= MINMATCHLENGTH) & (start > 0)) {
+                length = ZDICT_count(b + pos, b + suffix[start - 1]);
+                if (length >= LLIMIT) length = LLIMIT - 1;
+                lengthList[length]++;
+                if (length >= MINMATCHLENGTH) start--;
+            }
+        }
+
+        /* largest useful length */
+        memset(cumulLength, 0, sizeof(cumulLength));
+        cumulLength[maxLength-1] = lengthList[maxLength-1];
+        for (i=(int)(maxLength-2); i>=0; i--)
+            cumulLength[i] = cumulLength[i+1] + lengthList[i];
+
+        for (i=LLIMIT-1; i>=MINMATCHLENGTH; i--) if (cumulLength[i]>=minRatio) break;
+        maxLength = i;
+
+        /* reduce maxLength in case of final into repetitive data */
+        {   U32 l = (U32)maxLength;
+            BYTE const c = b[pos + maxLength-1];
+            while (b[pos+l-2]==c) l--;
+            maxLength = l;
+        }
+        if (maxLength < MINMATCHLENGTH) return solution;   /* skip : no long-enough solution */
+
+        /* calculate savings */
+        savings[5] = 0;
+        for (i=MINMATCHLENGTH; i<=(int)maxLength; i++)
+            savings[i] = savings[i-1] + (lengthList[i] * (i-3));
+
+        DISPLAYLEVEL(4, "Selected dict at position %u, of length %u : saves %u (ratio: %.2f)  \n",
+                     (unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / (double)maxLength);
+
+        solution.pos = (U32)pos;
+        solution.length = (U32)maxLength;
+        solution.savings = savings[maxLength];
+
+        /* mark positions done */
+        {   U32 id;
+            for (id=start; id<end; id++) {
+                U32 p, pEnd, length;
+                U32 const testedPos = (U32)suffix[id];
+                if (testedPos == pos)
+                    length = solution.length;
+                else {
+                    length = (U32)ZDICT_count(b+pos, b+testedPos);
+                    if (length > solution.length) length = solution.length;
+                }
+                pEnd = (U32)(testedPos + length);
+                for (p=testedPos; p<pEnd; p++)
+                    doneMarks[p] = 1;
+    }   }   }
+
+    return solution;
+}
+
+
+static int isIncluded(const void* in, const void* container, size_t length)
+{
+    const char* const ip = (const char*) in;
+    const char* const into = (const char*) container;
+    size_t u;
+
+    for (u=0; u<length; u++) {  /* works because end of buffer is a noisy guard band */
+        if (ip[u] != into[u]) break;
+    }
+
+    return u==length;
+}
+
+/*! ZDICT_tryMerge() :
+    check if dictItem can be merged, do it if possible
+    @return : id of destination elt, 0 if not merged
+*/
+static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const void* buffer)
+{
+    const U32 tableSize = table->pos;
+    const U32 eltEnd = elt.pos + elt.length;
+    const char* const buf = (const char*) buffer;
+
+    /* tail overlap */
+    U32 u; for (u=1; u<tableSize; u++) {
+        if (u==eltNbToSkip) continue;
+        if ((table[u].pos > elt.pos) && (table[u].pos <= eltEnd)) {  /* overlap, existing > new */
+            /* append */
+            U32 const addedLength = table[u].pos - elt.pos;
+            table[u].length += addedLength;
+            table[u].pos = elt.pos;
+            table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
+            elt = table[u];
+            /* sort : improve rank */
+            while ((u>1) && (table[u-1].savings < elt.savings))
+                table[u] = table[u-1], u--;
+            table[u] = elt;
+            return u;
+    }   }
+
+    /* front overlap */
+    for (u=1; u<tableSize; u++) {
+        if (u==eltNbToSkip) continue;
+
+        if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
+            /* append */
+            int const addedLength = (int)eltEnd - (int)(table[u].pos + table[u].length);
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
+            if (addedLength > 0) {   /* otherwise, elt fully included into existing */
+                table[u].length += addedLength;
+                table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
+            }
+            /* sort : improve rank */
+            elt = table[u];
+            while ((u>1) && (table[u-1].savings < elt.savings))
+                table[u] = table[u-1], u--;
+            table[u] = elt;
+            return u;
+        }
+
+        if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) {
+            if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) {
+                size_t const addedLength = MAX( (int)elt.length - (int)table[u].length , 1 );
+                table[u].pos = elt.pos;
+                table[u].savings += (U32)(elt.savings * addedLength / elt.length);
+                table[u].length = MIN(elt.length, table[u].length + 1);
+                return u;
+            }
+        }
+    }
+
+    return 0;
+}
+
+
+static void ZDICT_removeDictItem(dictItem* table, U32 id)
+{
+    /* convention : table[0].pos stores nb of elts */
+    U32 const max = table[0].pos;
+    U32 u;
+    if (!id) return;   /* protection, should never happen */
+    for (u=id; u<max-1; u++)
+        table[u] = table[u+1];
+    table->pos--;
+}
+
+
+static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer)
+{
+    /* merge if possible */
+    U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer);
+    if (mergeId) {
+        U32 newMerge = 1;
+        while (newMerge) {
+            newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer);
+            if (newMerge) ZDICT_removeDictItem(table, mergeId);
+            mergeId = newMerge;
+        }
+        return;
+    }
+
+    /* insert */
+    {   U32 current;
+        U32 nextElt = table->pos;
+        if (nextElt >= maxSize) nextElt = maxSize-1;
+        current = nextElt-1;
+        while (table[current].savings < elt.savings) {
+            table[current+1] = table[current];
+            current--;
+        }
+        table[current+1] = elt;
+        table->pos = nextElt+1;
+    }
+}
+
+
+static U32 ZDICT_dictSize(const dictItem* dictList)
+{
+    U32 u, dictSize = 0;
+    for (u=1; u<dictList[0].pos; u++)
+        dictSize += dictList[u].length;
+    return dictSize;
+}
+
+
+static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
+                            const void* const buffer, size_t bufferSize,   /* buffer must end with noisy guard band */
+                            const size_t* fileSizes, unsigned nbFiles,
+                            unsigned minRatio, U32 notificationLevel)
+{
+    int* const suffix0 = (int*)malloc((bufferSize+2)*sizeof(*suffix0));
+    int* const suffix = suffix0+1;
+    U32* reverseSuffix = (U32*)malloc((bufferSize)*sizeof(*reverseSuffix));
+    BYTE* doneMarks = (BYTE*)malloc((bufferSize+16)*sizeof(*doneMarks));   /* +16 for overflow security */
+    U32* filePos = (U32*)malloc(nbFiles * sizeof(*filePos));
+    size_t result = 0;
+    clock_t displayClock = 0;
+    clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
+
+#   undef  DISPLAYUPDATE
+#   define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
+            if (ZDICT_clockSpan(displayClock) > refreshRate)  \
+            { displayClock = clock(); DISPLAY(__VA_ARGS__); \
+            if (notificationLevel>=4) fflush(stderr); } }
+
+    /* init */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    if (!suffix0 || !reverseSuffix || !doneMarks || !filePos) {
+        result = ERROR(memory_allocation);
+        goto _cleanup;
+    }
+    if (minRatio < MINRATIO) minRatio = MINRATIO;
+    memset(doneMarks, 0, bufferSize+16);
+
+    /* limit sample set size (divsufsort limitation)*/
+    if (bufferSize > ZDICT_MAX_SAMPLES_SIZE) DISPLAYLEVEL(3, "sample set too large : reduced to %u MB ...\n", (unsigned)(ZDICT_MAX_SAMPLES_SIZE>>20));
+    while (bufferSize > ZDICT_MAX_SAMPLES_SIZE) bufferSize -= fileSizes[--nbFiles];
+
+    /* sort */
+    DISPLAYLEVEL(2, "sorting %u files of total size %u MB ...\n", nbFiles, (unsigned)(bufferSize>>20));
+    {   int const divSuftSortResult = divsufsort((const unsigned char*)buffer, suffix, (int)bufferSize, 0);
+        if (divSuftSortResult != 0) { result = ERROR(GENERIC); goto _cleanup; }
+    }
+    suffix[bufferSize] = (int)bufferSize;   /* leads into noise */
+    suffix0[0] = (int)bufferSize;           /* leads into noise */
+    /* build reverse suffix sort */
+    {   size_t pos;
+        for (pos=0; pos < bufferSize; pos++)
+            reverseSuffix[suffix[pos]] = (U32)pos;
+        /* note filePos tracks borders between samples.
+           It's not used at this stage, but planned to become useful in a later update */
+        filePos[0] = 0;
+        for (pos=1; pos<nbFiles; pos++)
+            filePos[pos] = (U32)(filePos[pos-1] + fileSizes[pos-1]);
+    }
+
+    DISPLAYLEVEL(2, "finding patterns ... \n");
+    DISPLAYLEVEL(3, "minimum ratio : %u \n", minRatio);
+
+    {   U32 cursor; for (cursor=0; cursor < bufferSize; ) {
+            dictItem solution;
+            if (doneMarks[cursor]) { cursor++; continue; }
+            solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio, notificationLevel);
+            if (solution.length==0) { cursor++; continue; }
+            ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
+            cursor += solution.length;
+            DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / (double)bufferSize * 100.0);
+    }   }
+
+_cleanup:
+    free(suffix0);
+    free(reverseSuffix);
+    free(doneMarks);
+    free(filePos);
+    return result;
+}
+
+
+static void ZDICT_fillNoise(void* buffer, size_t length)
+{
+    unsigned const prime1 = 2654435761U;
+    unsigned const prime2 = 2246822519U;
+    unsigned acc = prime1;
+    size_t p=0;
+    for (p=0; p<length; p++) {
+        acc *= prime2;
+        ((unsigned char*)buffer)[p] = (unsigned char)(acc >> 21);
+    }
+}
+
+
+typedef struct
+{
+    ZSTD_CDict* dict;    /* dictionary */
+    ZSTD_CCtx* zc;     /* working context */
+    void* workPlace;   /* must be ZSTD_BLOCKSIZE_MAX allocated */
+} EStats_ress_t;
+
+#define MAXREPOFFSET 1024
+
+static void ZDICT_countEStats(EStats_ress_t esr, const ZSTD_parameters* params,
+                              unsigned* countLit, unsigned* offsetcodeCount, unsigned* matchlengthCount, unsigned* litlengthCount, U32* repOffsets,
+                              const void* src, size_t srcSize,
+                              U32 notificationLevel)
+{
+    size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params->cParams.windowLog);
+    size_t cSize;
+
+    if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
+    {   size_t const errorCode = ZSTD_compressBegin_usingCDict_deprecated(esr.zc, esr.dict);
+        if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_compressBegin_usingCDict failed \n"); return; }
+
+    }
+    cSize = ZSTD_compressBlock_deprecated(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize);
+    if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (unsigned)srcSize); return; }
+
+    if (cSize) {  /* if == 0; block is not compressible */
+        const seqStore_t* const seqStorePtr = ZSTD_getSeqStore(esr.zc);
+
+        /* literals stats */
+        {   const BYTE* bytePtr;
+            for(bytePtr = seqStorePtr->litStart; bytePtr < seqStorePtr->lit; bytePtr++)
+                countLit[*bytePtr]++;
+        }
+
+        /* seqStats */
+        {   U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+            ZSTD_seqToCodes(seqStorePtr);
+
+            {   const BYTE* codePtr = seqStorePtr->ofCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) offsetcodeCount[codePtr[u]]++;
+            }
+
+            {   const BYTE* codePtr = seqStorePtr->mlCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) matchlengthCount[codePtr[u]]++;
+            }
+
+            {   const BYTE* codePtr = seqStorePtr->llCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) litlengthCount[codePtr[u]]++;
+            }
+
+            if (nbSeq >= 2) { /* rep offsets */
+                const seqDef* const seq = seqStorePtr->sequencesStart;
+                U32 offset1 = seq[0].offBase - ZSTD_REP_NUM;
+                U32 offset2 = seq[1].offBase - ZSTD_REP_NUM;
+                if (offset1 >= MAXREPOFFSET) offset1 = 0;
+                if (offset2 >= MAXREPOFFSET) offset2 = 0;
+                repOffsets[offset1] += 3;
+                repOffsets[offset2] += 1;
+    }   }   }
+}
+
+static size_t ZDICT_totalSampleSize(const size_t* fileSizes, unsigned nbFiles)
+{
+    size_t total=0;
+    unsigned u;
+    for (u=0; u<nbFiles; u++) total += fileSizes[u];
+    return total;
+}
+
+typedef struct { U32 offset; U32 count; } offsetCount_t;
+
+static void ZDICT_insertSortCount(offsetCount_t table[ZSTD_REP_NUM+1], U32 val, U32 count)
+{
+    U32 u;
+    table[ZSTD_REP_NUM].offset = val;
+    table[ZSTD_REP_NUM].count = count;
+    for (u=ZSTD_REP_NUM; u>0; u--) {
+        offsetCount_t tmp;
+        if (table[u-1].count >= table[u].count) break;
+        tmp = table[u-1];
+        table[u-1] = table[u];
+        table[u] = tmp;
+    }
+}
+
+/* ZDICT_flatLit() :
+ * rewrite `countLit` to contain a mostly flat but still compressible distribution of literals.
+ * necessary to avoid generating a non-compressible distribution that HUF_writeCTable() cannot encode.
+ */
+static void ZDICT_flatLit(unsigned* countLit)
+{
+    int u;
+    for (u=1; u<256; u++) countLit[u] = 2;
+    countLit[0]   = 4;
+    countLit[253] = 1;
+    countLit[254] = 1;
+}
+
+#define OFFCODE_MAX 30  /* only applicable to first block */
+static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
+                                   int compressionLevel,
+                             const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
+                             const void* dictBuffer, size_t  dictBufferSize,
+                                   unsigned notificationLevel)
+{
+    unsigned countLit[256];
+    HUF_CREATE_STATIC_CTABLE(hufTable, 255);
+    unsigned offcodeCount[OFFCODE_MAX+1];
+    short offcodeNCount[OFFCODE_MAX+1];
+    U32 offcodeMax = ZSTD_highbit32((U32)(dictBufferSize + 128 KB));
+    unsigned matchLengthCount[MaxML+1];
+    short matchLengthNCount[MaxML+1];
+    unsigned litLengthCount[MaxLL+1];
+    short litLengthNCount[MaxLL+1];
+    U32 repOffset[MAXREPOFFSET];
+    offsetCount_t bestRepOffset[ZSTD_REP_NUM+1];
+    EStats_ress_t esr = { NULL, NULL, NULL };
+    ZSTD_parameters params;
+    U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total;
+    size_t pos = 0, errorCode;
+    size_t eSize = 0;
+    size_t const totalSrcSize = ZDICT_totalSampleSize(fileSizes, nbFiles);
+    size_t const averageSampleSize = totalSrcSize / (nbFiles + !nbFiles);
+    BYTE* dstPtr = (BYTE*)dstBuffer;
+    U32 wksp[HUF_CTABLE_WORKSPACE_SIZE_U32];
+
+    /* init */
+    DEBUGLOG(4, "ZDICT_analyzeEntropy");
+    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
+    for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
+    for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
+    for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;
+    for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1;
+    memset(repOffset, 0, sizeof(repOffset));
+    repOffset[1] = repOffset[4] = repOffset[8] = 1;
+    memset(bestRepOffset, 0, sizeof(bestRepOffset));
+    if (compressionLevel==0) compressionLevel = ZSTD_CLEVEL_DEFAULT;
+    params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
+
+    esr.dict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, params.cParams, ZSTD_defaultCMem);
+    esr.zc = ZSTD_createCCtx();
+    esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
+    if (!esr.dict || !esr.zc || !esr.workPlace) {
+        eSize = ERROR(memory_allocation);
+        DISPLAYLEVEL(1, "Not enough memory \n");
+        goto _cleanup;
+    }
+
+    /* collect stats on all samples */
+    for (u=0; u<nbFiles; u++) {
+        ZDICT_countEStats(esr, &params,
+                          countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
+                         (const char*)srcBuffer + pos, fileSizes[u],
+                          notificationLevel);
+        pos += fileSizes[u];
+    }
+
+    if (notificationLevel >= 4) {
+        /* writeStats */
+        DISPLAYLEVEL(4, "Offset Code Frequencies : \n");
+        for (u=0; u<=offcodeMax; u++) {
+            DISPLAYLEVEL(4, "%2u :%7u \n", u, offcodeCount[u]);
+    }   }
+
+    /* analyze, build stats, starting with literals */
+    {   size_t maxNbBits = HUF_buildCTable_wksp(hufTable, countLit, 255, huffLog, wksp, sizeof(wksp));
+        if (HUF_isError(maxNbBits)) {
+            eSize = maxNbBits;
+            DISPLAYLEVEL(1, " HUF_buildCTable error \n");
+            goto _cleanup;
+        }
+        if (maxNbBits==8) {  /* not compressible : will fail on HUF_writeCTable() */
+            DISPLAYLEVEL(2, "warning : pathological dataset : literals are not compressible : samples are noisy or too regular \n");
+            ZDICT_flatLit(countLit);  /* replace distribution by a fake "mostly flat but still compressible" distribution, that HUF_writeCTable() can encode */
+            maxNbBits = HUF_buildCTable_wksp(hufTable, countLit, 255, huffLog, wksp, sizeof(wksp));
+            assert(maxNbBits==9);
+        }
+        huffLog = (U32)maxNbBits;
+    }
+
+    /* looking for most common first offsets */
+    {   U32 offset;
+        for (offset=1; offset<MAXREPOFFSET; offset++)
+            ZDICT_insertSortCount(bestRepOffset, offset, repOffset[offset]);
+    }
+    /* note : the result of this phase should be used to better appreciate the impact on statistics */
+
+    total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
+    errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax, /* useLowProbCount */ 1);
+    if (FSE_isError(errorCode)) {
+        eSize = errorCode;
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
+        goto _cleanup;
+    }
+    Offlog = (U32)errorCode;
+
+    total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
+    errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML, /* useLowProbCount */ 1);
+    if (FSE_isError(errorCode)) {
+        eSize = errorCode;
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
+        goto _cleanup;
+    }
+    mlLog = (U32)errorCode;
+
+    total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
+    errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL, /* useLowProbCount */ 1);
+    if (FSE_isError(errorCode)) {
+        eSize = errorCode;
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
+        goto _cleanup;
+    }
+    llLog = (U32)errorCode;
+
+    /* write result to buffer */
+    {   size_t const hhSize = HUF_writeCTable_wksp(dstPtr, maxDstSize, hufTable, 255, huffLog, wksp, sizeof(wksp));
+        if (HUF_isError(hhSize)) {
+            eSize = hhSize;
+            DISPLAYLEVEL(1, "HUF_writeCTable error \n");
+            goto _cleanup;
+        }
+        dstPtr += hhSize;
+        maxDstSize -= hhSize;
+        eSize += hhSize;
+    }
+
+    {   size_t const ohSize = FSE_writeNCount(dstPtr, maxDstSize, offcodeNCount, OFFCODE_MAX, Offlog);
+        if (FSE_isError(ohSize)) {
+            eSize = ohSize;
+            DISPLAYLEVEL(1, "FSE_writeNCount error with offcodeNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += ohSize;
+        maxDstSize -= ohSize;
+        eSize += ohSize;
+    }
+
+    {   size_t const mhSize = FSE_writeNCount(dstPtr, maxDstSize, matchLengthNCount, MaxML, mlLog);
+        if (FSE_isError(mhSize)) {
+            eSize = mhSize;
+            DISPLAYLEVEL(1, "FSE_writeNCount error with matchLengthNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += mhSize;
+        maxDstSize -= mhSize;
+        eSize += mhSize;
+    }
+
+    {   size_t const lhSize = FSE_writeNCount(dstPtr, maxDstSize, litLengthNCount, MaxLL, llLog);
+        if (FSE_isError(lhSize)) {
+            eSize = lhSize;
+            DISPLAYLEVEL(1, "FSE_writeNCount error with litlengthNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += lhSize;
+        maxDstSize -= lhSize;
+        eSize += lhSize;
+    }
+
+    if (maxDstSize<12) {
+        eSize = ERROR(dstSize_tooSmall);
+        DISPLAYLEVEL(1, "not enough space to write RepOffsets \n");
+        goto _cleanup;
+    }
+# if 0
+    MEM_writeLE32(dstPtr+0, bestRepOffset[0].offset);
+    MEM_writeLE32(dstPtr+4, bestRepOffset[1].offset);
+    MEM_writeLE32(dstPtr+8, bestRepOffset[2].offset);
+#else
+    /* at this stage, we don't use the result of "most common first offset",
+     * as the impact of statistics is not properly evaluated */
+    MEM_writeLE32(dstPtr+0, repStartValue[0]);
+    MEM_writeLE32(dstPtr+4, repStartValue[1]);
+    MEM_writeLE32(dstPtr+8, repStartValue[2]);
+#endif
+    eSize += 12;
+
+_cleanup:
+    ZSTD_freeCDict(esr.dict);
+    ZSTD_freeCCtx(esr.zc);
+    free(esr.workPlace);
+
+    return eSize;
+}
+
+
+/**
+ * @returns the maximum repcode value
+ */
+static U32 ZDICT_maxRep(U32 const reps[ZSTD_REP_NUM])
+{
+    U32 maxRep = reps[0];
+    int r;
+    for (r = 1; r < ZSTD_REP_NUM; ++r)
+        maxRep = MAX(maxRep, reps[r]);
+    return maxRep;
+}
+
+size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
+                          const void* customDictContent, size_t dictContentSize,
+                          const void* samplesBuffer, const size_t* samplesSizes,
+                          unsigned nbSamples, ZDICT_params_t params)
+{
+    size_t hSize;
+#define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
+    BYTE header[HBUFFSIZE];
+    int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+    /* The final dictionary content must be at least as large as the largest repcode */
+    size_t const minContentSize = (size_t)ZDICT_maxRep(repStartValue);
+    size_t paddingSize;
+
+    /* check conditions */
+    DEBUGLOG(4, "ZDICT_finalizeDictionary");
+    if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
+
+    /* dictionary header */
+    MEM_writeLE32(header, ZSTD_MAGIC_DICTIONARY);
+    {   U64 const randomID = XXH64(customDictContent, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32(header+4, dictID);
+    }
+    hSize = 8;
+
+    /* entropy tables */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    DISPLAYLEVEL(2, "statistics ... \n");
+    {   size_t const eSize = ZDICT_analyzeEntropy(header+hSize, HBUFFSIZE-hSize,
+                                  compressionLevel,
+                                  samplesBuffer, samplesSizes, nbSamples,
+                                  customDictContent, dictContentSize,
+                                  notificationLevel);
+        if (ZDICT_isError(eSize)) return eSize;
+        hSize += eSize;
+    }
+
+    /* Shrink the content size if it doesn't fit in the buffer */
+    if (hSize + dictContentSize > dictBufferCapacity) {
+        dictContentSize = dictBufferCapacity - hSize;
+    }
+
+    /* Pad the dictionary content with zeros if it is too small */
+    if (dictContentSize < minContentSize) {
+        RETURN_ERROR_IF(hSize + minContentSize > dictBufferCapacity, dstSize_tooSmall,
+                        "dictBufferCapacity too small to fit max repcode");
+        paddingSize = minContentSize - dictContentSize;
+    } else {
+        paddingSize = 0;
+    }
+
+    {
+        size_t const dictSize = hSize + paddingSize + dictContentSize;
+
+        /* The dictionary consists of the header, optional padding, and the content.
+         * The padding comes before the content because the "best" position in the
+         * dictionary is the last byte.
+         */
+        BYTE* const outDictHeader = (BYTE*)dictBuffer;
+        BYTE* const outDictPadding = outDictHeader + hSize;
+        BYTE* const outDictContent = outDictPadding + paddingSize;
+
+        assert(dictSize <= dictBufferCapacity);
+        assert(outDictContent + dictContentSize == (BYTE*)dictBuffer + dictSize);
+
+        /* First copy the customDictContent into its final location.
+         * `customDictContent` and `dictBuffer` may overlap, so we must
+         * do this before any other writes into the output buffer.
+         * Then copy the header & padding into the output buffer.
+         */
+        memmove(outDictContent, customDictContent, dictContentSize);
+        memcpy(outDictHeader, header, hSize);
+        memset(outDictPadding, 0, paddingSize);
+
+        return dictSize;
+    }
+}
+
+
+static size_t ZDICT_addEntropyTablesFromBuffer_advanced(
+        void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+        const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+        ZDICT_params_t params)
+{
+    int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+    size_t hSize = 8;
+
+    /* calculate entropy tables */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    DISPLAYLEVEL(2, "statistics ... \n");
+    {   size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
+                                  compressionLevel,
+                                  samplesBuffer, samplesSizes, nbSamples,
+                                  (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize,
+                                  notificationLevel);
+        if (ZDICT_isError(eSize)) return eSize;
+        hSize += eSize;
+    }
+
+    /* add dictionary header (after entropy tables) */
+    MEM_writeLE32(dictBuffer, ZSTD_MAGIC_DICTIONARY);
+    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32((char*)dictBuffer+4, dictID);
+    }
+
+    if (hSize + dictContentSize < dictBufferCapacity)
+        memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
+    return MIN(dictBufferCapacity, hSize+dictContentSize);
+}
+
+/*! ZDICT_trainFromBuffer_unsafe_legacy() :
+*   Warning : `samplesBuffer` must be followed by noisy guard band !!!
+*   @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
+*/
+static size_t ZDICT_trainFromBuffer_unsafe_legacy(
+                            void* dictBuffer, size_t maxDictSize,
+                            const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                            ZDICT_legacy_params_t params)
+{
+    U32 const dictListSize = MAX(MAX(DICTLISTSIZE_DEFAULT, nbSamples), (U32)(maxDictSize/16));
+    dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList));
+    unsigned const selectivity = params.selectivityLevel == 0 ? g_selectivity_default : params.selectivityLevel;
+    unsigned const minRep = (selectivity > 30) ? MINRATIO : nbSamples >> selectivity;
+    size_t const targetDictSize = maxDictSize;
+    size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
+    size_t dictSize = 0;
+    U32 const notificationLevel = params.zParams.notificationLevel;
+
+    /* checks */
+    if (!dictList) return ERROR(memory_allocation);
+    if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); }   /* requested dictionary size is too small */
+    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* not enough source to create dictionary */
+
+    /* init */
+    ZDICT_initDictItem(dictList);
+
+    /* build dictionary */
+    ZDICT_trainBuffer_legacy(dictList, dictListSize,
+                       samplesBuffer, samplesBuffSize,
+                       samplesSizes, nbSamples,
+                       minRep, notificationLevel);
+
+    /* display best matches */
+    if (params.zParams.notificationLevel>= 3) {
+        unsigned const nb = MIN(25, dictList[0].pos);
+        unsigned const dictContentSize = ZDICT_dictSize(dictList);
+        unsigned u;
+        DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", (unsigned)dictList[0].pos-1, dictContentSize);
+        DISPLAYLEVEL(3, "list %u best segments \n", nb-1);
+        for (u=1; u<nb; u++) {
+            unsigned const pos = dictList[u].pos;
+            unsigned const length = dictList[u].length;
+            U32 const printedLength = MIN(40, length);
+            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize)) {
+                free(dictList);
+                return ERROR(GENERIC);   /* should never happen */
+            }
+            DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
+                         u, length, pos, (unsigned)dictList[u].savings);
+            ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
+            DISPLAYLEVEL(3, "| \n");
+    }   }
+
+
+    /* create dictionary */
+    {   unsigned dictContentSize = ZDICT_dictSize(dictList);
+        if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* dictionary content too small */
+        if (dictContentSize < targetDictSize/4) {
+            DISPLAYLEVEL(2, "!  warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (unsigned)maxDictSize);
+            if (samplesBuffSize < 10 * targetDictSize)
+                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (unsigned)(samplesBuffSize>>20));
+            if (minRep > MINRATIO) {
+                DISPLAYLEVEL(2, "!  consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1);
+                DISPLAYLEVEL(2, "!  note : larger dictionaries are not necessarily better, test its efficiency on samples \n");
+            }
+        }
+
+        if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
+            unsigned proposedSelectivity = selectivity-1;
+            while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; }
+            DISPLAYLEVEL(2, "!  note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (unsigned)maxDictSize);
+            DISPLAYLEVEL(2, "!  consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity);
+            DISPLAYLEVEL(2, "!  always test dictionary efficiency on real samples \n");
+        }
+
+        /* limit dictionary size */
+        {   U32 const max = dictList->pos;   /* convention : nb of useful elts within dictList */
+            U32 currentSize = 0;
+            U32 n; for (n=1; n<max; n++) {
+                currentSize += dictList[n].length;
+                if (currentSize > targetDictSize) { currentSize -= dictList[n].length; break; }
+            }
+            dictList->pos = n;
+            dictContentSize = currentSize;
+        }
+
+        /* build dict content */
+        {   U32 u;
+            BYTE* ptr = (BYTE*)dictBuffer + maxDictSize;
+            for (u=1; u<dictList->pos; u++) {
+                U32 l = dictList[u].length;
+                ptr -= l;
+                if (ptr<(BYTE*)dictBuffer) { free(dictList); return ERROR(GENERIC); }   /* should not happen */
+                memcpy(ptr, (const char*)samplesBuffer+dictList[u].pos, l);
+        }   }
+
+        dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize,
+                                                             samplesBuffer, samplesSizes, nbSamples,
+                                                             params.zParams);
+    }
+
+    /* clean up */
+    free(dictList);
+    return dictSize;
+}
+
+
+/* ZDICT_trainFromBuffer_legacy() :
+ * issue : samplesBuffer need to be followed by a noisy guard band.
+ * work around : duplicate the buffer, and add the noise */
+size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
+                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                              ZDICT_legacy_params_t params)
+{
+    size_t result;
+    void* newBuff;
+    size_t const sBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
+    if (sBuffSize < ZDICT_MIN_SAMPLES_SIZE) return 0;   /* not enough content => no dictionary */
+
+    newBuff = malloc(sBuffSize + NOISELENGTH);
+    if (!newBuff) return ERROR(memory_allocation);
+
+    memcpy(newBuff, samplesBuffer, sBuffSize);
+    ZDICT_fillNoise((char*)newBuff + sBuffSize, NOISELENGTH);   /* guard band, for end of buffer condition */
+
+    result =
+        ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, dictBufferCapacity, newBuff,
+                                            samplesSizes, nbSamples, params);
+    free(newBuff);
+    return result;
+}
+
+
+size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
+                             const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
+{
+    ZDICT_fastCover_params_t params;
+    DEBUGLOG(3, "ZDICT_trainFromBuffer");
+    memset(&params, 0, sizeof(params));
+    params.d = 8;
+    params.steps = 4;
+    /* Use default level since no compression level information is available */
+    params.zParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1)
+    params.zParams.notificationLevel = DEBUGLEVEL;
+#endif
+    return ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, dictBufferCapacity,
+                                               samplesBuffer, samplesSizes, nbSamples,
+                                               &params);
+}
+
+size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
+{
+    ZDICT_params_t params;
+    memset(&params, 0, sizeof(params));
+    return ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, dictBufferCapacity,
+                                                     samplesBuffer, samplesSizes, nbSamples,
+                                                     params);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_legacy.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_legacy.h
new file mode 100644
index 0000000..dd17325
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_legacy.h
@@ -0,0 +1,422 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LEGACY_H
+#define ZSTD_LEGACY_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include "../common/mem.h"            /* MEM_STATIC */
+#include "../common/error_private.h"  /* ERROR */
+#include "../common/zstd_internal.h"  /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTD_frameSizeInfo */
+
+#if !defined (ZSTD_LEGACY_SUPPORT) || (ZSTD_LEGACY_SUPPORT == 0)
+#  undef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 8
+#endif
+
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+#  include "zstd_v01.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+#  include "zstd_v02.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+#  include "zstd_v03.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+#  include "zstd_v04.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+#  include "zstd_v05.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+#  include "zstd_v06.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+#  include "zstd_v07.h"
+#endif
+
+/** ZSTD_isLegacy() :
+    @return : > 0 if supported by legacy decoder. 0 otherwise.
+              return value is the version.
+*/
+MEM_STATIC unsigned ZSTD_isLegacy(const void* src, size_t srcSize)
+{
+    U32 magicNumberLE;
+    if (srcSize<4) return 0;
+    magicNumberLE = MEM_readLE32(src);
+    switch(magicNumberLE)
+    {
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+        case ZSTDv01_magicNumberLE:return 1;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+        case ZSTDv02_magicNumber : return 2;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+        case ZSTDv03_magicNumber : return 3;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case ZSTDv04_magicNumber : return 4;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case ZSTDv05_MAGICNUMBER : return 5;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case ZSTDv06_MAGICNUMBER : return 6;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case ZSTDv07_MAGICNUMBER : return 7;
+#endif
+        default : return 0;
+    }
+}
+
+
+MEM_STATIC unsigned long long ZSTD_getDecompressedSize_legacy(const void* src, size_t srcSize)
+{
+    U32 const version = ZSTD_isLegacy(src, srcSize);
+    if (version < 5) return 0;  /* no decompressed size in frame header, or not a legacy format */
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+    if (version==5) {
+        ZSTDv05_parameters fParams;
+        size_t const frResult = ZSTDv05_getFrameParams(&fParams, src, srcSize);
+        if (frResult != 0) return 0;
+        return fParams.srcSize;
+    }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+    if (version==6) {
+        ZSTDv06_frameParams fParams;
+        size_t const frResult = ZSTDv06_getFrameParams(&fParams, src, srcSize);
+        if (frResult != 0) return 0;
+        return fParams.frameContentSize;
+    }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+    if (version==7) {
+        ZSTDv07_frameParams fParams;
+        size_t const frResult = ZSTDv07_getFrameParams(&fParams, src, srcSize);
+        if (frResult != 0) return 0;
+        return fParams.frameContentSize;
+    }
+#endif
+    return 0;   /* should not be possible */
+}
+
+
+MEM_STATIC size_t ZSTD_decompressLegacy(
+                     void* dst, size_t dstCapacity,
+               const void* src, size_t compressedSize,
+               const void* dict,size_t dictSize)
+{
+    U32 const version = ZSTD_isLegacy(src, compressedSize);
+    (void)dst; (void)dstCapacity; (void)dict; (void)dictSize;  /* unused when ZSTD_LEGACY_SUPPORT >= 8 */
+    switch(version)
+    {
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+        case 1 :
+            return ZSTDv01_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+        case 2 :
+            return ZSTDv02_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+        case 3 :
+            return ZSTDv03_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+            return ZSTDv04_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+            {   size_t result;
+                ZSTDv05_DCtx* const zd = ZSTDv05_createDCtx();
+                if (zd==NULL) return ERROR(memory_allocation);
+                result = ZSTDv05_decompress_usingDict(zd, dst, dstCapacity, src, compressedSize, dict, dictSize);
+                ZSTDv05_freeDCtx(zd);
+                return result;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+            {   size_t result;
+                ZSTDv06_DCtx* const zd = ZSTDv06_createDCtx();
+                if (zd==NULL) return ERROR(memory_allocation);
+                result = ZSTDv06_decompress_usingDict(zd, dst, dstCapacity, src, compressedSize, dict, dictSize);
+                ZSTDv06_freeDCtx(zd);
+                return result;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+            {   size_t result;
+                ZSTDv07_DCtx* const zd = ZSTDv07_createDCtx();
+                if (zd==NULL) return ERROR(memory_allocation);
+                result = ZSTDv07_decompress_usingDict(zd, dst, dstCapacity, src, compressedSize, dict, dictSize);
+                ZSTDv07_freeDCtx(zd);
+                return result;
+            }
+#endif
+        default :
+            return ERROR(prefix_unknown);
+    }
+}
+
+MEM_STATIC ZSTD_frameSizeInfo ZSTD_findFrameSizeInfoLegacy(const void *src, size_t srcSize)
+{
+    ZSTD_frameSizeInfo frameSizeInfo;
+    U32 const version = ZSTD_isLegacy(src, srcSize);
+    switch(version)
+    {
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+        case 1 :
+            ZSTDv01_findFrameSizeInfoLegacy(src, srcSize,
+                &frameSizeInfo.compressedSize,
+                &frameSizeInfo.decompressedBound);
+            break;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+        case 2 :
+            ZSTDv02_findFrameSizeInfoLegacy(src, srcSize,
+                &frameSizeInfo.compressedSize,
+                &frameSizeInfo.decompressedBound);
+            break;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+        case 3 :
+            ZSTDv03_findFrameSizeInfoLegacy(src, srcSize,
+                &frameSizeInfo.compressedSize,
+                &frameSizeInfo.decompressedBound);
+            break;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+            ZSTDv04_findFrameSizeInfoLegacy(src, srcSize,
+                &frameSizeInfo.compressedSize,
+                &frameSizeInfo.decompressedBound);
+            break;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+            ZSTDv05_findFrameSizeInfoLegacy(src, srcSize,
+                &frameSizeInfo.compressedSize,
+                &frameSizeInfo.decompressedBound);
+            break;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+            ZSTDv06_findFrameSizeInfoLegacy(src, srcSize,
+                &frameSizeInfo.compressedSize,
+                &frameSizeInfo.decompressedBound);
+            break;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+            ZSTDv07_findFrameSizeInfoLegacy(src, srcSize,
+                &frameSizeInfo.compressedSize,
+                &frameSizeInfo.decompressedBound);
+            break;
+#endif
+        default :
+            frameSizeInfo.compressedSize = ERROR(prefix_unknown);
+            frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
+            break;
+    }
+    if (!ZSTD_isError(frameSizeInfo.compressedSize) && frameSizeInfo.compressedSize > srcSize) {
+        frameSizeInfo.compressedSize = ERROR(srcSize_wrong);
+        frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
+    }
+    /* In all cases, decompressedBound == nbBlocks * ZSTD_BLOCKSIZE_MAX.
+     * So we can compute nbBlocks without having to change every function.
+     */
+    if (frameSizeInfo.decompressedBound != ZSTD_CONTENTSIZE_ERROR) {
+        assert((frameSizeInfo.decompressedBound & (ZSTD_BLOCKSIZE_MAX - 1)) == 0);
+        frameSizeInfo.nbBlocks = (size_t)(frameSizeInfo.decompressedBound / ZSTD_BLOCKSIZE_MAX);
+    }
+    return frameSizeInfo;
+}
+
+MEM_STATIC size_t ZSTD_findFrameCompressedSizeLegacy(const void *src, size_t srcSize)
+{
+    ZSTD_frameSizeInfo frameSizeInfo = ZSTD_findFrameSizeInfoLegacy(src, srcSize);
+    return frameSizeInfo.compressedSize;
+}
+
+MEM_STATIC size_t ZSTD_freeLegacyStreamContext(void* legacyContext, U32 version)
+{
+    switch(version)
+    {
+        default :
+        case 1 :
+        case 2 :
+        case 3 :
+            (void)legacyContext;
+            return ERROR(version_unsupported);
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 : return ZBUFFv04_freeDCtx((ZBUFFv04_DCtx*)legacyContext);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 : return ZBUFFv05_freeDCtx((ZBUFFv05_DCtx*)legacyContext);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 : return ZBUFFv06_freeDCtx((ZBUFFv06_DCtx*)legacyContext);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 : return ZBUFFv07_freeDCtx((ZBUFFv07_DCtx*)legacyContext);
+#endif
+    }
+}
+
+
+MEM_STATIC size_t ZSTD_initLegacyStream(void** legacyContext, U32 prevVersion, U32 newVersion,
+                                        const void* dict, size_t dictSize)
+{
+    DEBUGLOG(5, "ZSTD_initLegacyStream for v0.%u", newVersion);
+    if (prevVersion != newVersion) ZSTD_freeLegacyStreamContext(*legacyContext, prevVersion);
+    switch(newVersion)
+    {
+        default :
+        case 1 :
+        case 2 :
+        case 3 :
+            (void)dict; (void)dictSize;
+            return 0;
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+        {
+            ZBUFFv04_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv04_createDCtx() : (ZBUFFv04_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv04_decompressInit(dctx);
+            ZBUFFv04_decompressWithDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+        {
+            ZBUFFv05_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv05_createDCtx() : (ZBUFFv05_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv05_decompressInitDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+        {
+            ZBUFFv06_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv06_createDCtx() : (ZBUFFv06_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv06_decompressInitDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+        {
+            ZBUFFv07_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv07_createDCtx() : (ZBUFFv07_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv07_decompressInitDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+    }
+}
+
+
+
+MEM_STATIC size_t ZSTD_decompressLegacyStream(void* legacyContext, U32 version,
+                                              ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    DEBUGLOG(5, "ZSTD_decompressLegacyStream for v0.%u", version);
+    switch(version)
+    {
+        default :
+        case 1 :
+        case 2 :
+        case 3 :
+            (void)legacyContext; (void)output; (void)input;
+            return ERROR(version_unsupported);
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+            {
+                ZBUFFv04_DCtx* dctx = (ZBUFFv04_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv04_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+            {
+                ZBUFFv05_DCtx* dctx = (ZBUFFv05_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv05_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+            {
+                ZBUFFv06_DCtx* dctx = (ZBUFFv06_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv06_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+            {
+                ZBUFFv07_DCtx* dctx = (ZBUFFv07_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv07_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+    }
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* ZSTD_LEGACY_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v01.c b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v01.c
new file mode 100644
index 0000000..1a3aad0
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v01.c
@@ -0,0 +1,2125 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v01.h"
+#include "../common/error_private.h"
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+/* You can statically allocate Huff0 DTable as a table of unsigned short using below macro */
+#define HUF_DTABLE_SIZE_U16(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUF_CREATE_STATIC_DTABLE(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE_U16(maxTableLog)] = { maxTableLog }
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define FSE_LIST_ERRORS(ITEM) \
+        ITEM(FSE_OK_NoError) ITEM(FSE_ERROR_GENERIC) \
+        ITEM(FSE_ERROR_tableLog_tooLarge) ITEM(FSE_ERROR_maxSymbolValue_tooLarge) ITEM(FSE_ERROR_maxSymbolValue_tooSmall) \
+        ITEM(FSE_ERROR_dstSize_tooSmall) ITEM(FSE_ERROR_srcSize_wrong)\
+        ITEM(FSE_ERROR_corruptionDetected) \
+        ITEM(FSE_ERROR_maxCode)
+
+#define FSE_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { FSE_LIST_ERRORS(FSE_GENERATE_ENUM) } FSE_errorCodes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+
+/******************************************
+*  FSE symbol compression API
+******************************************/
+/*
+   This API consists of small unitary functions, which highly benefit from being inlined.
+   You will want to enable link-time-optimization to ensure these functions are properly inlined in your binary.
+   Visual seems to do it automatically.
+   For gcc or clang, you'll need to add -flto flag at compilation and linking stages.
+   If none of these solutions is applicable, include "fse.c" directly.
+*/
+
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+typedef struct
+{
+    size_t bitContainer;
+    int    bitPos;
+    char*  startPtr;
+    char*  ptr;
+    char*  endPtr;
+} FSE_CStream_t;
+
+typedef struct
+{
+    ptrdiff_t   value;
+    const void* stateTable;
+    const void* symbolTT;
+    unsigned    stateLog;
+} FSE_CState_t;
+
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} FSE_DStream_t;
+
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+typedef enum { FSE_DStream_unfinished = 0,
+               FSE_DStream_endOfBuffer = 1,
+               FSE_DStream_completed = 2,
+               FSE_DStream_tooFar = 3 } FSE_DStream_status;  /* result of FSE_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... ?! */
+
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+#ifndef MEM_ACCESS_MODULE
+#define MEM_ACCESS_MODULE
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+# include <stdint.h>
+typedef  uint8_t BYTE;
+typedef uint16_t U16;
+typedef  int16_t S16;
+typedef uint32_t U32;
+typedef  int32_t S32;
+typedef uint64_t U64;
+typedef  int64_t S64;
+#else
+typedef unsigned char       BYTE;
+typedef unsigned short      U16;
+typedef   signed short      S16;
+typedef unsigned int        U32;
+typedef   signed int        S32;
+typedef unsigned long long  U64;
+typedef   signed long long  S64;
+#endif
+
+#endif   /* MEM_ACCESS_MODULE */
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+
+static unsigned FSE_32bits(void)
+{
+    return sizeof(void*)==4;
+}
+
+static unsigned FSE_isLittleEndian(void)
+{
+    const union { U32 i; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+static U16 FSE_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U32 FSE_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U64 FSE_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U16 FSE_readLE16(const void* memPtr)
+{
+    if (FSE_isLittleEndian())
+        return FSE_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+static U32 FSE_readLE32(const void* memPtr)
+{
+    if (FSE_isLittleEndian())
+        return FSE_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+static U64 FSE_readLE64(const void* memPtr)
+{
+    if (FSE_isLittleEndian())
+        return FSE_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+static size_t FSE_readLEST(const void* memPtr)
+{
+    if (FSE_32bits())
+        return (size_t)FSE_readLE32(memPtr);
+    else
+        return (size_t)FSE_readLE64(memPtr);
+}
+
+
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef struct
+{
+    int deltaFindState;
+    U32 deltaNbBits;
+} FSE_symbolCompressionTransform; /* total 8 bytes */
+
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+/****************************************************************
+*  Internal functions
+****************************************************************/
+FORCE_INLINE unsigned FSE_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (GCC_VERSION >= 304)   /* GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)(ptr) + 1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return (size_t)-FSE_ERROR_maxSymbolValue_tooLarge;
+    if (tableLog > FSE_MAX_TABLELOG) return (size_t)-FSE_ERROR_tableLog_tooLarge;
+
+    /* Init, lay down lowprob symbols */
+    DTableH[0].tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return (size_t)-FSE_ERROR_GENERIC;   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - FSE_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH->fastMode = (U16)noLarge;
+    return 0;
+}
+
+
+/******************************************
+*  FSE byte symbol
+******************************************/
+#ifndef FSE_COMMONDEFS_ONLY
+
+static unsigned FSE_isError(size_t code) { return (code > (size_t)(-FSE_ERROR_maxCode)); }
+
+static short FSE_abs(short a)
+{
+    return a<0? -a : a;
+}
+
+
+/****************************************************************
+*  Header bitstream management
+****************************************************************/
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return (size_t)-FSE_ERROR_srcSize_wrong;
+    bitStream = FSE_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return (size_t)-FSE_ERROR_tableLog_tooLarge;
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = FSE_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return (size_t)-FSE_ERROR_maxSymbolValue_tooSmall;
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = FSE_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = FSE_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return (size_t)-FSE_ERROR_GENERIC;
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return (size_t)-FSE_ERROR_GENERIC;             /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+
+/* FSE_initDStream
+ * Initialize a FSE_DStream_t.
+ * srcBuffer must point at the beginning of an FSE block.
+ * The function result is the size of the FSE_block (== srcSize).
+ * If srcSize is too small, the function will return an errorCode;
+ */
+static size_t FSE_initDStream(FSE_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) return (size_t)-FSE_ERROR_srcSize_wrong;
+
+    if (srcSize >=  sizeof(size_t))
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = FSE_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return (size_t)-FSE_ERROR_GENERIC;   /* stop bit not present */
+        bitD->bitsConsumed = 8 - FSE_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return (size_t)-FSE_ERROR_GENERIC;   /* stop bit not present */
+        bitD->bitsConsumed = 8 - FSE_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+/*!FSE_lookBits
+ * Provides next n bits from the bitContainer.
+ * bitContainer is not modified (bits are still present for next read/look)
+ * On 32-bits, maxNbBits==25
+ * On 64-bits, maxNbBits==57
+ * return : value extracted.
+ */
+static size_t FSE_lookBits(FSE_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+static size_t FSE_lookBitsFast(FSE_DStream_t* bitD, U32 nbBits)   /* only if nbBits >= 1 !! */
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+static void FSE_skipBits(FSE_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+
+/*!FSE_readBits
+ * Read next n bits from the bitContainer.
+ * On 32-bits, don't read more than maxNbBits==25
+ * On 64-bits, don't read more than maxNbBits==57
+ * Use the fast variant *only* if n >= 1.
+ * return : value extracted.
+ */
+static size_t FSE_readBits(FSE_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = FSE_lookBits(bitD, nbBits);
+    FSE_skipBits(bitD, nbBits);
+    return value;
+}
+
+static size_t FSE_readBitsFast(FSE_DStream_t* bitD, U32 nbBits)   /* only if nbBits >= 1 !! */
+{
+    size_t value = FSE_lookBitsFast(bitD, nbBits);
+    FSE_skipBits(bitD, nbBits);
+    return value;
+}
+
+static unsigned FSE_reloadDStream(FSE_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return FSE_DStream_tooFar;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = FSE_readLEST(bitD->ptr);
+        return FSE_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return FSE_DStream_endOfBuffer;
+        return FSE_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        U32 result = FSE_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = FSE_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = FSE_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+
+static void FSE_initDState(FSE_DState_t* DStatePtr, FSE_DStream_t* bitD, const FSE_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSE_DTableHeader* const DTableH = (const FSE_DTableHeader*)ptr;
+    DStatePtr->state = FSE_readBits(bitD, DTableH->tableLog);
+    FSE_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+static BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, FSE_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = FSE_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+static BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, FSE_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = FSE_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/* FSE_endOfDStream
+   Tells if bitD has reached end of bitStream or not */
+
+static unsigned FSE_endOfDStream(const FSE_DStream_t* bitD)
+{
+    return ((bitD->ptr == bitD->start) && (bitD->bitsConsumed == sizeof(bitD->bitContainer)*8));
+}
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    FSE_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = FSE_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (FSE_reloadDStream(&bitD)==FSE_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            FSE_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (FSE_reloadDStream(&bitD) > FSE_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            FSE_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : FSE_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly FSE_DStream_completed */
+    while (1)
+    {
+        if ( (FSE_reloadDStream(&bitD)>FSE_DStream_completed) || (op==omax) || (FSE_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (FSE_reloadDStream(&bitD)>FSE_DStream_completed) || (op==omax) || (FSE_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (FSE_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall;   /* dst buffer is full, but cSrc unfinished */
+
+    return (size_t)-FSE_ERROR_corruptionDetected;
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));   /* memcpy() into local variable, to avoid strict aliasing warning */
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return (size_t)-FSE_ERROR_srcSize_wrong;   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+/* *******************************************************
+*  Huff0 : Huffman block compression
+*********************************************************/
+#define HUF_MAX_SYMBOL_VALUE 255
+#define HUF_DEFAULT_TABLELOG  12       /* used by default, when not specified */
+#define HUF_MAX_TABLELOG  12           /* max possible tableLog; for allocation purpose; can be modified */
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+typedef struct HUF_CElt_s {
+  U16  val;
+  BYTE nbBits;
+} HUF_CElt ;
+
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+
+/* *******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct {
+    BYTE byte;
+    BYTE nbBits;
+} HUF_DElt;
+
+static size_t HUF_readDTable (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];  /* large enough for values from 0 to 16 */
+    U32 weightTotal;
+    U32 maxBits;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DElt* const dt = (HUF_DElt*)ptr;
+
+    if (!srcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+    iSize = ip[0];
+
+    FSE_STATIC_ASSERT(sizeof(HUF_DElt) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* should not be necessary, but some analyzer complain ... */
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, sizeof(huffWeight));
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+        oSize = FSE_decompress(huffWeight, HUF_MAX_SYMBOL_VALUE, ip+1, iSize);   /* max 255 values decoded, last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankVal, 0, sizeof(rankVal));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return (size_t)-FSE_ERROR_corruptionDetected;
+        rankVal[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return (size_t)-FSE_ERROR_corruptionDetected;
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    maxBits = FSE_highbit32(weightTotal) + 1;
+    if (maxBits > DTable[0]) return (size_t)-FSE_ERROR_tableLog_tooLarge;   /* DTable is too small */
+    DTable[0] = (U16)maxBits;
+    {
+        U32 total = 1 << maxBits;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << FSE_highbit32(rest);
+        U32 lastWeight = FSE_highbit32(rest) + 1;
+        if (verif != rest) return (size_t)-FSE_ERROR_corruptionDetected;    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankVal[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankVal[1] < 2) || (rankVal[1] & 1)) return (size_t)-FSE_ERROR_corruptionDetected;   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=maxBits; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<=oSize; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DElt D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(maxBits + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize+1;
+}
+
+
+static BYTE HUF_decodeSymbol(FSE_DStream_t* Dstream, const HUF_DElt* dt, const U32 dtLog)
+{
+        const size_t val = FSE_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        FSE_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+static size_t HUF_decompress_usingDTable(   /* -3% slower when non static */
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 6) return (size_t)-FSE_ERROR_srcSize_wrong;
+    {
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* op = ostart;
+        BYTE* const omax = op + maxDstSize;
+        BYTE* const olimit = maxDstSize < 15 ? op : omax-15;
+
+        const void* ptr = DTable;
+        const HUF_DElt* const dt = (const HUF_DElt*)(ptr)+1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+        U32 reloadStatus;
+
+        /* Init */
+
+        const U16* jumpTable = (const U16*)cSrc;
+        const size_t length1 = FSE_readLE16(jumpTable);
+        const size_t length2 = FSE_readLE16(jumpTable+1);
+        const size_t length3 = FSE_readLE16(jumpTable+2);
+        const size_t length4 = cSrcSize - 6 - length1 - length2 - length3;   /* check coherency !! */
+        const char* const start1 = (const char*)(cSrc) + 6;
+        const char* const start2 = start1 + length1;
+        const char* const start3 = start2 + length2;
+        const char* const start4 = start3 + length3;
+        FSE_DStream_t bitD1, bitD2, bitD3, bitD4;
+
+        if (length1+length2+length3+6 >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+
+        errorCode = FSE_initDStream(&bitD1, start1, length1);
+        if (FSE_isError(errorCode)) return errorCode;
+        errorCode = FSE_initDStream(&bitD2, start2, length2);
+        if (FSE_isError(errorCode)) return errorCode;
+        errorCode = FSE_initDStream(&bitD3, start3, length3);
+        if (FSE_isError(errorCode)) return errorCode;
+        errorCode = FSE_initDStream(&bitD4, start4, length4);
+        if (FSE_isError(errorCode)) return errorCode;
+
+        reloadStatus=FSE_reloadDStream(&bitD2);
+
+        /* 16 symbols per loop */
+        for ( ; (reloadStatus<FSE_DStream_completed) && (op<olimit);  /* D2-3-4 are supposed to be synchronized and finish together */
+            op+=16, reloadStatus = FSE_reloadDStream(&bitD2) | FSE_reloadDStream(&bitD3) | FSE_reloadDStream(&bitD4), FSE_reloadDStream(&bitD1))
+        {
+    #define HUF_DECODE_SYMBOL_0(n, Dstream) \
+            op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog);
+
+    #define HUF_DECODE_SYMBOL_1(n, Dstream) \
+            op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
+            if (FSE_32bits() && (HUF_MAX_TABLELOG>12)) FSE_reloadDStream(&Dstream)
+
+    #define HUF_DECODE_SYMBOL_2(n, Dstream) \
+            op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
+            if (FSE_32bits()) FSE_reloadDStream(&Dstream)
+
+            HUF_DECODE_SYMBOL_1( 0, bitD1);
+            HUF_DECODE_SYMBOL_1( 1, bitD2);
+            HUF_DECODE_SYMBOL_1( 2, bitD3);
+            HUF_DECODE_SYMBOL_1( 3, bitD4);
+            HUF_DECODE_SYMBOL_2( 4, bitD1);
+            HUF_DECODE_SYMBOL_2( 5, bitD2);
+            HUF_DECODE_SYMBOL_2( 6, bitD3);
+            HUF_DECODE_SYMBOL_2( 7, bitD4);
+            HUF_DECODE_SYMBOL_1( 8, bitD1);
+            HUF_DECODE_SYMBOL_1( 9, bitD2);
+            HUF_DECODE_SYMBOL_1(10, bitD3);
+            HUF_DECODE_SYMBOL_1(11, bitD4);
+            HUF_DECODE_SYMBOL_0(12, bitD1);
+            HUF_DECODE_SYMBOL_0(13, bitD2);
+            HUF_DECODE_SYMBOL_0(14, bitD3);
+            HUF_DECODE_SYMBOL_0(15, bitD4);
+        }
+
+        if (reloadStatus!=FSE_DStream_completed)   /* not complete : some bitStream might be FSE_DStream_unfinished */
+            return (size_t)-FSE_ERROR_corruptionDetected;
+
+        /* tail */
+        {
+            /* bitTail = bitD1; */   /* *much* slower : -20% !??! */
+            FSE_DStream_t bitTail;
+            bitTail.ptr = bitD1.ptr;
+            bitTail.bitsConsumed = bitD1.bitsConsumed;
+            bitTail.bitContainer = bitD1.bitContainer;   /* required in case of FSE_DStream_endOfBuffer */
+            bitTail.start = start1;
+            for ( ; (FSE_reloadDStream(&bitTail) < FSE_DStream_completed) && (op<omax) ; op++)
+            {
+                HUF_DECODE_SYMBOL_0(0, bitTail);
+            }
+
+            if (FSE_endOfDStream(&bitTail))
+                return op-ostart;
+        }
+
+        if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall;   /* dst buffer is full, but cSrc unfinished */
+
+        return (size_t)-FSE_ERROR_corruptionDetected;
+    }
+}
+
+
+static size_t HUF_decompress (void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLE(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTable (DTable, cSrc, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, DTable);
+}
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/****************************************************************
+*  Tuning parameters
+*****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect */
+#define ZSTD_MEMORY_USAGE 17
+
+
+/**************************************
+   CPU Feature Detection
+**************************************/
+/*
+ * Automated efficient unaligned memory access detection
+ * Based on known hardware architectures
+ * This list will be updated thanks to feedbacks
+ */
+#if defined(CPU_HAS_EFFICIENT_UNALIGNED_MEMORY_ACCESS) \
+    || defined(__ARM_FEATURE_UNALIGNED) \
+    || defined(__i386__) || defined(__x86_64__) \
+    || defined(_M_IX86) || defined(_M_X64) \
+    || defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_8__) \
+    || (defined(_M_ARM) && (_M_ARM >= 7))
+#  define ZSTD_UNALIGNED_ACCESS 1
+#else
+#  define ZSTD_UNALIGNED_ACCESS 0
+#endif
+
+
+/********************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/********************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+#ifndef MEM_ACCESS_MODULE
+#define MEM_ACCESS_MODULE
+/********************************************************
+*  Basic Types
+*********************************************************/
+#if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+typedef  uint8_t BYTE;
+typedef uint16_t U16;
+typedef  int16_t S16;
+typedef uint32_t U32;
+typedef  int32_t S32;
+typedef uint64_t U64;
+#else
+typedef unsigned char       BYTE;
+typedef unsigned short      U16;
+typedef   signed short      S16;
+typedef unsigned int        U32;
+typedef   signed int        S32;
+typedef unsigned long long  U64;
+#endif
+
+#endif   /* MEM_ACCESS_MODULE */
+
+
+/********************************************************
+*  Constants
+*********************************************************/
+static const U32 ZSTD_magicNumber = 0xFD2FB51E;   /* 3rd version : seqNb header */
+
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff ((1<<Offbits)-1)
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/********************************************************
+*  Memory operations
+*********************************************************/
+static unsigned ZSTD_32bits(void) { return sizeof(void*)==4; }
+
+static unsigned ZSTD_isLittleEndian(void)
+{
+    const union { U32 i; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+static U16    ZSTD_read16(const void* p) { U16 r; memcpy(&r, p, sizeof(r)); return r; }
+
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s)    { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    while (op < oend) COPY8(op, ip);
+}
+
+static U16 ZSTD_readLE16(const void* memPtr)
+{
+    if (ZSTD_isLittleEndian()) return ZSTD_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)((U16)p[0] + ((U16)p[1]<<8));
+    }
+}
+
+static U32 ZSTD_readLE24(const void* memPtr)
+{
+    return ZSTD_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+static U32 ZSTD_readBE32(const void* memPtr)
+{
+    const BYTE* p = (const BYTE*)memPtr;
+    return (U32)(((U32)p[0]<<24) + ((U32)p[1]<<16) + ((U32)p[2]<<8) + ((U32)p[3]<<0));
+}
+
+
+/**************************************
+*  Local structures
+***************************************/
+typedef struct ZSTD_Cctx_s ZSTD_Cctx;
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} seqStore_t;
+
+
+typedef struct ZSTD_Cctx_s
+{
+    const BYTE* base;
+    U32 current;
+    U32 nextUpdate;
+    seqStore_t seqStore;
+#ifdef __AVX2__
+    __m256i hashTable[HASH_TABLESIZE>>3];
+#else
+    U32 hashTable[HASH_TABLESIZE];
+#endif
+    BYTE buffer[WORKPLACESIZE];
+} cctxi_t;
+
+
+
+
+/**************************************
+*  Error Management
+**************************************/
+/* published entry point */
+unsigned ZSTDv01_isError(size_t code) { return ERR_isError(code); }
+
+
+/**************************************
+*  Tool functions
+**************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    1    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  3    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+/**************************************************************
+*   Decompression code
+**************************************************************/
+
+static size_t ZSTDv01_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+static size_t ZSTD_decompressLiterals(void* ctx,
+                                      void* dst, size_t maxDstSize,
+                                const void* src, size_t srcSize)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + maxDstSize;
+    const BYTE* ip = (const BYTE*)src;
+    size_t errorCode;
+    size_t litSize;
+
+    /* check : minimum 2, for litSize, +1, for content */
+    if (srcSize <= 3) return ERROR(corruption_detected);
+
+    litSize = ip[1] + (ip[0]<<8);
+    litSize += ((ip[-3] >> 3) & 7) << 16;   /* mmmmh.... */
+    op = oend - litSize;
+
+    (void)ctx;
+    if (litSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    errorCode = HUF_decompress(op, litSize, ip+2, srcSize-2);
+    if (FSE_isError(errorCode)) return ERROR(GENERIC);
+    return litSize;
+}
+
+
+static size_t ZSTDv01_decodeLiteralsBlock(void* ctx,
+                                void* dst, size_t maxDstSize,
+                          const BYTE** litStart, size_t* litSize,
+                          const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    blockProperties_t litbp;
+
+    size_t litcSize = ZSTDv01_getcBlockSize(src, srcSize, &litbp);
+    if (ZSTDv01_isError(litcSize)) return litcSize;
+    if (litcSize > srcSize - ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    ip += ZSTD_blockHeaderSize;
+
+    switch(litbp.blockType)
+    {
+    case bt_raw:
+        *litStart = ip;
+        ip += litcSize;
+        *litSize = litcSize;
+        break;
+    case bt_rle:
+        {
+            size_t rleSize = litbp.origSize;
+            if (rleSize>maxDstSize) return ERROR(dstSize_tooSmall);
+            if (!srcSize) return ERROR(srcSize_wrong);
+            if (rleSize > 0) {
+                memset(oend - rleSize, *ip, rleSize);
+            }
+            *litStart = oend - rleSize;
+            *litSize = rleSize;
+            ip++;
+            break;
+        }
+    case bt_compressed:
+        {
+            size_t decodedLitSize = ZSTD_decompressLiterals(ctx, dst, maxDstSize, ip, litcSize);
+            if (ZSTDv01_isError(decodedLitSize)) return decodedLitSize;
+            *litStart = oend - decodedLitSize;
+            *litSize = decodedLitSize;
+            ip += litcSize;
+            break;
+        }
+    case bt_end:
+    default:
+        return ERROR(GENERIC);
+    }
+
+    return ip-istart;
+}
+
+
+static size_t ZSTDv01_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = ZSTD_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++); break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    FSE_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else
+        {
+            if (dumps<=(de-3))
+            {
+                litLength = ZSTD_readLE24(dumps);
+                dumps += 3;
+            }
+        }
+    }
+
+    /* Offset */
+    {
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));
+        if (ZSTD_32bits()) FSE_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = ((size_t)1 << (nbBits & ((sizeof(offset)*8)-1))) + FSE_readBits(&(seqState->DStream), nbBits);
+        if (ZSTD_32bits()) FSE_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else
+        {
+            if (dumps<=(de-3))
+            {
+                matchLength = ZSTD_readLE24(dumps);
+                dumps += 3;
+            }
+        }
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* subtracted */
+    const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
+    const size_t litLength = sequence.litLength;
+    BYTE* const endMatch = op + litLength + sequence.matchLength;    /* risk : address space overflow (32-bits) */
+    const BYTE* const litEnd = *litPtr + litLength;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
+    if (endMatch > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+    if (sequence.matchLength > (size_t)(*litPtr-op)) return ERROR(dstSize_tooSmall);  /* overwrite literal segment */
+
+    /* copy Literals */
+    ZSTD_memmove(op, *litPtr, sequence.litLength);   /* note : v0.1 seems to allow scenarios where output or input are close to end of buffer */
+
+    op += litLength;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* check : last match must be at a minimum distance of 8 from end of dest buffer */
+    if (oend-op < 8) return ERROR(dstSize_tooSmall);
+
+    /* copy Match */
+    {
+        const U32 overlapRisk = (((size_t)(litEnd - endMatch)) < 12);
+        const BYTE* match = op - sequence.offset;            /* possible underflow at op - offset ? */
+        size_t qutt = 12;
+        U64 saved[2];
+
+        /* check */
+        if (match < base) return ERROR(corruption_detected);
+        if (sequence.offset > (size_t)base) return ERROR(corruption_detected);
+
+        /* save beginning of literal sequence, in case of write overlap */
+        if (overlapRisk)
+        {
+            if ((endMatch + qutt) > oend) qutt = oend-endMatch;
+            memcpy(saved, endMatch, qutt);
+        }
+
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        } else { ZSTD_copy8(op, match); }
+        op += 8; match += 8;
+
+        if (endMatch > oend-(16-MINMATCH))
+        {
+            if (op < oend-8)
+            {
+                ZSTD_wildcopy(op, match, (oend-8) - op);
+                match += (oend-8) - op;
+                op = oend-8;
+            }
+            while (op<endMatch) *op++ = *match++;
+        }
+        else
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+
+        /* restore, in case of overlap */
+        if (overlapRisk) memcpy(endMatch, saved, qutt);
+    }
+
+    return endMatch-ostart;
+}
+
+typedef struct ZSTDv01_Dctx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+} dctx_t;
+
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize,
+                         const BYTE* litStart, size_t litSize)
+{
+    dctx_t* dctx = (dctx_t*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = litStart;
+    const BYTE* const litEnd = litStart + litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTDv01_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTDv01_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 1;
+        errorCode = FSE_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (FSE_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (FSE_reloadDStream(&(seqState.DStream)) <= FSE_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTDv01_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !FSE_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (lastLLSize > 0) {
+                if (op != litPtr) memmove(op, litPtr, lastLLSize);
+                op += lastLLSize;
+            }
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed, srcSize is trusted */
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* litPtr = NULL;
+    size_t litSize = 0;
+    size_t errorCode;
+
+    /* Decode literals sub-block */
+    errorCode = ZSTDv01_decodeLiteralsBlock(ctx, dst, maxDstSize, &litPtr, &litSize, src, srcSize);
+    if (ZSTDv01_isError(errorCode)) return errorCode;
+    ip += errorCode;
+    srcSize -= errorCode;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize, litPtr, litSize);
+}
+
+
+size_t ZSTDv01_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    size_t errorCode=0;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = ZSTD_readBE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t blockSize = ZSTDv01_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv01_isError(blockSize)) return blockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (blockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            errorCode = ZSTD_decompressBlock(ctx, op, oend-op, ip, blockSize);
+            break;
+        case bt_raw :
+            errorCode = ZSTD_copyUncompressedBlock(op, oend-op, ip, blockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        if (blockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv01_isError(errorCode)) return errorCode;
+        op += errorCode;
+        ip += blockSize;
+        remainingSize -= blockSize;
+    }
+
+    return op-ostart;
+}
+
+size_t ZSTDv01_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    dctx_t ctx;
+    ctx.base = dst;
+    return ZSTDv01_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv01_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    magicNumber = ZSTD_readBE32(src);
+    if (magicNumber != ZSTD_magicNumber) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t blockSize = ZSTDv01_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv01_isError(blockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, blockSize);
+            return;
+        }
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (blockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (blockSize == 0) break;   /* bt_end */
+
+        ip += blockSize;
+        remainingSize -= blockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+size_t ZSTDv01_resetDCtx(ZSTDv01_Dctx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+ZSTDv01_Dctx* ZSTDv01_createDCtx(void)
+{
+    ZSTDv01_Dctx* dctx = (ZSTDv01_Dctx*)malloc(sizeof(ZSTDv01_Dctx));
+    if (dctx==NULL) return NULL;
+    ZSTDv01_resetDCtx(dctx);
+    return dctx;
+}
+
+size_t ZSTDv01_freeDCtx(ZSTDv01_Dctx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+size_t ZSTDv01_nextSrcSizeToDecompress(ZSTDv01_Dctx* dctx)
+{
+    return ((dctx_t*)dctx)->expected;
+}
+
+size_t ZSTDv01_decompressContinue(ZSTDv01_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    dctx_t* ctx = (dctx_t*)dctx;
+
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = ZSTD_readBE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTDv01_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTDv01_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v01.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v01.h
new file mode 100644
index 0000000..6ac8769
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v01.h
@@ -0,0 +1,94 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V01_H_28739879432
+#define ZSTD_V01_H_28739879432
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv01_decompress() : decompress ZSTD frames compliant with v0.1.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv01_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+ /**
+ ZSTDv01_findFrameSizeInfoLegacy() : get the source length and decompressed bound of a ZSTD frame compliant with v0.1.x format
+     srcSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+     cSize (output parameter)  : the number of bytes that would be read to decompress this frame
+                                 or an error code if it fails (which can be tested using ZSTDv01_isError())
+     dBound (output parameter) : an upper-bound for the decompressed size of the data in the frame
+                                 or ZSTD_CONTENTSIZE_ERROR if an error occurs
+
+     note : assumes `cSize` and `dBound` are _not_ NULL.
+ */
+void ZSTDv01_findFrameSizeInfoLegacy(const void *src, size_t srcSize,
+                                     size_t* cSize, unsigned long long* dBound);
+
+/**
+ZSTDv01_isError() : tells if the result of ZSTDv01_decompress() is an error
+*/
+unsigned ZSTDv01_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv01_Dctx_s ZSTDv01_Dctx;
+ZSTDv01_Dctx* ZSTDv01_createDCtx(void);
+size_t ZSTDv01_freeDCtx(ZSTDv01_Dctx* dctx);
+
+size_t ZSTDv01_decompressDCtx(void* ctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+/* *************************************
+*  Streaming functions
+***************************************/
+size_t ZSTDv01_resetDCtx(ZSTDv01_Dctx* dctx);
+
+size_t ZSTDv01_nextSrcSizeToDecompress(ZSTDv01_Dctx* dctx);
+size_t ZSTDv01_decompressContinue(ZSTDv01_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv01_magicNumber   0xFD2FB51E   /* Big Endian version */
+#define ZSTDv01_magicNumberLE 0x1EB52FFD   /* Little Endian version */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V01_H_28739879432 */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v02.c b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v02.c
new file mode 100644
index 0000000..e09bb4a
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v02.c
@@ -0,0 +1,3477 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v02.h"
+#include "../common/error_private.h"
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   Error codes and messages
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#elif defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define PREFIX(name) ZSTD_error_##name
+
+#define ERROR(name) (size_t)-PREFIX(name)
+
+#define ERROR_LIST(ITEM) \
+        ITEM(PREFIX(No_Error)) ITEM(PREFIX(GENERIC)) \
+        ITEM(PREFIX(dstSize_tooSmall)) ITEM(PREFIX(srcSize_wrong)) \
+        ITEM(PREFIX(prefix_unknown)) ITEM(PREFIX(corruption_detected)) \
+        ITEM(PREFIX(tableLog_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooSmall)) \
+        ITEM(PREFIX(maxCode))
+
+#define ERROR_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { ERROR_LIST(ERROR_GENERATE_ENUM) } ERR_codes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+#define ERROR_CONVERTTOSTRING(STRING) #STRING,
+#define ERROR_GENERATE_STRING(EXPR) ERROR_CONVERTTOSTRING(EXPR)
+static const char* ERR_strings[] = { ERROR_LIST(ERROR_GENERATE_STRING) };
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    static const char* codeError = "Unspecified error code";
+    if (ERR_isError(code)) return ERR_strings[-(int)(code)];
+    return codeError;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
+/*
+Constructor and Destructor of type FSE_CTable
+    Note that its size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/******************************************
+*  FSE advanced API
+******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+/******************************************
+*  FSE symbol decompression API
+******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/******************************************
+*  FSE unsafe API
+******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/******************************************
+*  Implementation of inline functions
+******************************************/
+
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Static allocation macros
+******************************************/
+/* Huff0 buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/******************************************
+*  Advanced functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+static size_t HUF_decompress4X6 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* quad-symbols decoder */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+/*
+    zstd - standard compression library
+    Header File
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Version
+***************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    2    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  2    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
+
+#if defined (__cplusplus)
+}
+#endif
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Streaming functions
+***************************************/
+
+typedef struct ZSTD_DCtx_s ZSTD_DCtx;
+
+/*
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTD_magicNumber 0xFD2FB522   /* v0.2 (current)*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt+1;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)ptr;
+    FSE_DTableHeader DTableH;
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));   /* memcpy(), to avoid strict aliasing warnings */
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return (short)(a<0 ? -a : a);
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/******************************************
+*  Helper functions
+******************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*********************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)ptr;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    void* ptr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)ptr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        {if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* quad-symbol decoding           */
+/**********************************/
+typedef struct { BYTE nbBits; BYTE nbBytes; } HUF_DDescX6;
+typedef union { BYTE byte[4]; U32 sequence; } HUF_DSeqX6;
+
+/* recursive, up to level 3; may benefit from <template>-like strategy to nest each level inline */
+static void HUF_fillDTableX6LevelN(HUF_DDescX6* DDescription, HUF_DSeqX6* DSequence, int sizeLog,
+                           const rankVal_t rankValOrigin, const U32 consumed, const int minWeight, const U32 maxWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize, const U32* rankStart,
+                           const U32 nbBitsBaseline, HUF_DSeqX6 baseSeq, HUF_DDescX6 DDesc)
+{
+    const int scaleLog = nbBitsBaseline - sizeLog;   /* note : targetLog >= (nbBitsBaseline-1), hence scaleLog <= 1 */
+    const int minBits  = nbBitsBaseline - maxWeight;
+    const U32 level = DDesc.nbBytes;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 symbolStartPos, s;
+
+    /* local rankVal, will be modified */
+    memcpy(rankVal, rankValOrigin[consumed], sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i;
+        const U32 skipSize = rankVal[minWeight];
+        for (i = 0; i < skipSize; i++)
+        {
+            DSequence[i] = baseSeq;
+            DDescription[i] = DDesc;
+        }
+    }
+
+    /* fill DTable */
+    DDesc.nbBytes++;
+    symbolStartPos = rankStart[minWeight];
+    for (s=symbolStartPos; s<sortedListSize; s++)
+    {
+        const BYTE symbol = sortedSymbols[s].symbol;
+        const U32  weight = sortedSymbols[s].weight;   /* >= 1 (sorted) */
+        const int  nbBits = nbBitsBaseline - weight;   /* >= 1 (by construction) */
+        const int  totalBits = consumed+nbBits;
+        const U32  start  = rankVal[weight];
+        const U32  length = 1 << (sizeLog-nbBits);
+        baseSeq.byte[level] = symbol;
+        DDesc.nbBits = (BYTE)totalBits;
+
+        if ((level<3) && (sizeLog-totalBits >= minBits))   /* enough room for another symbol */
+        {
+            int nextMinWeight = totalBits + scaleLog;
+            if (nextMinWeight < 1) nextMinWeight = 1;
+            HUF_fillDTableX6LevelN(DDescription+start, DSequence+start, sizeLog-nbBits,
+                           rankValOrigin, totalBits, nextMinWeight, maxWeight,
+                           sortedSymbols, sortedListSize, rankStart,
+                           nbBitsBaseline, baseSeq, DDesc);   /* recursive (max : level 3) */
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            for (i = start; i < end; i++)
+            {
+                DDescription[i] = DDesc;
+                DSequence[i] = baseSeq;
+            }
+        }
+        rankVal[weight] += length;
+    }
+}
+
+
+/* note : same preparation as X4 */
+static size_t HUF_readDTableX6 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    rankVal_t rankVal;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+
+    /* fill tables */
+    {
+        void* ptr = DTable+1;
+        HUF_DDescX6* DDescription = (HUF_DDescX6*)(ptr);
+        void* dSeqStart = DTable + 1 + ((size_t)1<<(memLog-1));
+        HUF_DSeqX6* DSequence = (HUF_DSeqX6*)(dSeqStart);
+        HUF_DSeqX6 DSeq;
+        HUF_DDescX6 DDesc;
+        DSeq.sequence = 0;
+        DDesc.nbBits = 0;
+        DDesc.nbBytes = 0;
+        HUF_fillDTableX6LevelN(DDescription, DSequence, memLog,
+                       (const U32 (*)[HUF_ABSOLUTEMAX_TABLELOG + 1])rankVal, 0, 1, maxW,
+                       sortedSymbol, sizeOfSort, rankStart0,
+                       tableLog+1, DSeq, DDesc);
+    }
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX6(void* op, BIT_DStream_t* DStream, const HUF_DDescX6* dd, const HUF_DSeqX6* ds, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, ds+val, sizeof(HUF_DSeqX6));
+    BIT_skipBits(DStream, dd[val].nbBits);
+    return dd[val].nbBytes;
+}
+
+static U32 HUF_decodeLastSymbolsX6(void* op, const U32 maxL, BIT_DStream_t* DStream,
+                                  const HUF_DDescX6* dd, const HUF_DSeqX6* ds, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    U32 length = dd[val].nbBytes;
+    if (length <= maxL)
+    {
+        memcpy(op, ds+val, length);
+        BIT_skipBits(DStream, dd[val].nbBits);
+        return length;
+    }
+    memcpy(op, ds+val, maxL);
+    if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+    {
+        BIT_skipBits(DStream, dd[val].nbBits);
+        if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+            DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }
+    return maxL;
+}
+
+
+#define HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX6(ptr, DStreamPtr, dd, ds, dtLog)
+
+#define HUF_DECODE_SYMBOLX6_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX6_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX6(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const U32* DTable, const U32 dtLog)
+{
+    const void* ddPtr = DTable+1;
+    const HUF_DDescX6* dd = (const HUF_DDescX6*)(ddPtr);
+    const void* dsPtr = DTable + 1 + ((size_t)1<<(dtLog-1));
+    const HUF_DSeqX6* ds = (const HUF_DSeqX6*)(dsPtr);
+    BYTE* const pStart = p;
+
+    /* up to 16 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-16))
+    {
+        HUF_DECODE_SYMBOLX6_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);
+    }
+
+    /* closer to the end, up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);
+
+    while (p <= pEnd-4)
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    while (p < pEnd)
+        p += HUF_decodeLastSymbolsX6(p, (U32)(pEnd-p), bitDPtr, dd, ds, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X6_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const U32 dtLog = DTable[0];
+        const void* ddPtr = DTable+1;
+        const HUF_DDescX6* dd = (const HUF_DDescX6*)(ddPtr);
+        const void* dsPtr = DTable + 1 + ((size_t)1<<(dtLog-1));
+        const HUF_DSeqX6* ds = (const HUF_DSeqX6*)(dsPtr);
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-64 symbols per loop (4-16 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (op3 <= opStart4) && (endSignal==BIT_DStream_unfinished) && (op4<=(oend-16)) ; )
+        {
+            HUF_DECODE_SYMBOLX6_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX6(op1, &bitD1, opStart2, DTable, dtLog);
+        HUF_decodeStreamX6(op2, &bitD2, opStart3, DTable, dtLog);
+        HUF_decodeStreamX6(op3, &bitD3, opStart4, DTable, dtLog);
+        HUF_decodeStreamX6(op4, &bitD4, oend,     DTable, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X6 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX6(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX6 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X6_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, HUF_decompress4X6 };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+    if (Dtime[2] < Dtime[algoNb]) algoNb = 2;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+*  MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*/
+#define ZSTD_MEMORY_USAGE 17
+
+/*!
+ * HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0, fastest), or in memory heap (1, requires malloc())
+ * Note that compression context is fairly large, as a consequence heap memory is recommended.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif /* ZSTD_HEAPMODE */
+
+/*!
+*  LEGACY_SUPPORT :
+*  decompressor can decode older formats (starting from Zstd 0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/* *******************************************************
+*  Constants
+*********************************************************/
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff   31
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do COPY8(op, ip) while (op < oend);
+}
+
+
+/* **************************************
+*  Local structures
+****************************************/
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} seqStore_t;
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+struct ZSTD_DCtx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+};   /* typedef'd to ZSTD_Dctx within "zstd_static.h" */
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize )*/
+static size_t ZSTD_decodeLiteralsBlock(void* ctx,
+                          const void* src, size_t srcSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* const istart = (const BYTE* const)src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    default:
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;
+        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else if (dumps + 3 <= de)
+        {
+            litLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const size_t offsetPrefix[MaxOff+1] = {  /* note : size_t faster than U32 */
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 3 <= de)
+        {
+            matchLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* subtracted */
+    const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
+    BYTE* const oMatchEnd = op + sequence.litLength + sequence.matchLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use the pointer check */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    {
+        const BYTE* match = op - sequence.offset;
+
+        /* check */
+        if (sequence.offset > (size_t)op) return ERROR(corruption_detected);   /* address space overflow test (this test seems kept by clang optimizer) */
+        //if (match > op) return ERROR(corruption_detected);   /* address space overflow test (is clang optimizer removing this test ?) */
+        if (match < base) return ERROR(corruption_detected);
+
+        /* close range match, overlap */
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        }
+        else
+        {
+            ZSTD_copy8(op, match);
+        }
+        op += 8; match += 8;
+
+        if (oMatchEnd > oend-(16-MINMATCH))
+        {
+            if (op < oend_8)
+            {
+                ZSTD_wildcopy(op, match, oend_8 - op);
+                match += oend_8 - op;
+                op = oend_8;
+            }
+            while (op < oMatchEnd) *op++ = *match++;
+        }
+        else
+        {
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        }
+    }
+
+    return oMatchEnd - ostart;
+}
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 1;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (lastLLSize > 0) {
+                if (op != litPtr) memmove(op, litPtr, lastLLSize);
+                op += lastLLSize;
+            }
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    /* Decode literals sub-block */
+    size_t litCSize = ZSTD_decodeLiteralsBlock(ctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyUncompressedBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t ZSTD_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    ZSTD_DCtx ctx;
+    ctx.base = dst;
+    return ZSTD_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv02_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = MEM_readLE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTD_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
+
+
+/* wrapper layer */
+
+unsigned ZSTDv02_isError(size_t code)
+{
+    return ZSTD_isError(code);
+}
+
+size_t ZSTDv02_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize)
+{
+    return ZSTD_decompress(dst, maxOriginalSize, src, compressedSize);
+}
+
+ZSTDv02_Dctx* ZSTDv02_createDCtx(void)
+{
+    return (ZSTDv02_Dctx*)ZSTD_createDCtx();
+}
+
+size_t ZSTDv02_freeDCtx(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_freeDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_resetDCtx(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_resetDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_nextSrcSizeToDecompress(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_decompressContinue(ZSTDv02_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue((ZSTD_DCtx*)dctx, dst, maxDstSize, src, srcSize);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v02.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v02.h
new file mode 100644
index 0000000..dab0260
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v02.h
@@ -0,0 +1,93 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V02_H_4174539423
+#define ZSTD_V02_H_4174539423
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv02_decompress() : decompress ZSTD frames compliant with v0.2.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv02_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+ /**
+ ZSTDv02_findFrameSizeInfoLegacy() : get the source length and decompressed bound of a ZSTD frame compliant with v0.2.x format
+     srcSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+     cSize (output parameter)  : the number of bytes that would be read to decompress this frame
+                                 or an error code if it fails (which can be tested using ZSTDv01_isError())
+     dBound (output parameter) : an upper-bound for the decompressed size of the data in the frame
+                                 or ZSTD_CONTENTSIZE_ERROR if an error occurs
+
+    note : assumes `cSize` and `dBound` are _not_ NULL.
+ */
+void ZSTDv02_findFrameSizeInfoLegacy(const void *src, size_t srcSize,
+                                     size_t* cSize, unsigned long long* dBound);
+
+/**
+ZSTDv02_isError() : tells if the result of ZSTDv02_decompress() is an error
+*/
+unsigned ZSTDv02_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv02_Dctx_s ZSTDv02_Dctx;
+ZSTDv02_Dctx* ZSTDv02_createDCtx(void);
+size_t ZSTDv02_freeDCtx(ZSTDv02_Dctx* dctx);
+
+size_t ZSTDv02_decompressDCtx(void* ctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+/* *************************************
+*  Streaming functions
+***************************************/
+size_t ZSTDv02_resetDCtx(ZSTDv02_Dctx* dctx);
+
+size_t ZSTDv02_nextSrcSizeToDecompress(ZSTDv02_Dctx* dctx);
+size_t ZSTDv02_decompressContinue(ZSTDv02_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv02_magicNumber 0xFD2FB522   /* v0.2 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V02_H_4174539423 */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v03.c b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v03.c
new file mode 100644
index 0000000..b0d7f52
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v03.c
@@ -0,0 +1,3117 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v03.h"
+#include "../common/error_private.h"
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   Error codes and messages
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#elif defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define PREFIX(name) ZSTD_error_##name
+
+#define ERROR(name) (size_t)-PREFIX(name)
+
+#define ERROR_LIST(ITEM) \
+        ITEM(PREFIX(No_Error)) ITEM(PREFIX(GENERIC)) \
+        ITEM(PREFIX(dstSize_tooSmall)) ITEM(PREFIX(srcSize_wrong)) \
+        ITEM(PREFIX(prefix_unknown)) ITEM(PREFIX(corruption_detected)) \
+        ITEM(PREFIX(tableLog_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooSmall)) \
+        ITEM(PREFIX(maxCode))
+
+#define ERROR_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { ERROR_LIST(ERROR_GENERATE_ENUM) } ERR_codes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+#define ERROR_CONVERTTOSTRING(STRING) #STRING,
+#define ERROR_GENERATE_STRING(EXPR) ERROR_CONVERTTOSTRING(EXPR)
+static const char* ERR_strings[] = { ERROR_LIST(ERROR_GENERATE_STRING) };
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    static const char* codeError = "Unspecified error code";
+    if (ERR_isError(code)) return ERR_strings[-(int)(code)];
+    return codeError;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
+/*
+Constructor and Destructor of type FSE_CTable
+    Note that its size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/******************************************
+*  FSE advanced API
+******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+/******************************************
+*  FSE symbol decompression API
+******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/******************************************
+*  FSE unsafe API
+******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/******************************************
+*  Implementation of inline functions
+******************************************/
+
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Static allocation macros
+******************************************/
+/* Huff0 buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/******************************************
+*  Advanced functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+/*
+    zstd - standard compression library
+    Header File
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Version
+***************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    2    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  2    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
+
+#if defined (__cplusplus)
+}
+#endif
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Streaming functions
+***************************************/
+
+typedef struct ZSTD_DCtx_s ZSTD_DCtx;
+
+/*
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTD_magicNumber 0xFD2FB523   /* v0.3 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt+1;
+    FSE_DTableHeader DTableH;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)ptr;
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return a<0 ? -a : a;
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/******************************************
+*  Helper functions
+******************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*********************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)(ptr);
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    void* ptr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)ptr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+*  MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*/
+#define ZSTD_MEMORY_USAGE 17
+
+/*!
+ * HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0, fastest), or in memory heap (1, requires malloc())
+ * Note that compression context is fairly large, as a consequence heap memory is recommended.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif /* ZSTD_HEAPMODE */
+
+/*!
+*  LEGACY_SUPPORT :
+*  decompressor can decode older formats (starting from Zstd 0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#else
+#  define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#endif
+
+
+/* *******************************************************
+*  Constants
+*********************************************************/
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff   31
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do COPY8(op, ip) while (op < oend);
+}
+
+
+/* **************************************
+*  Local structures
+****************************************/
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} seqStore_t;
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+struct ZSTD_DCtx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+};   /* typedef'd to ZSTD_Dctx within "zstd_static.h" */
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize )*/
+static size_t ZSTD_decodeLiteralsBlock(void* ctx,
+                          const void* src, size_t srcSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* const istart = (const BYTE* const)src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    default:
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;
+        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else if (dumps + 3 <= de)
+        {
+            litLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const size_t offsetPrefix[MaxOff+1] = {  /* note : size_t faster than U32 */
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 3 <= de)
+        {
+            matchLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* subtracted */
+    const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
+    BYTE* const oMatchEnd = op + sequence.litLength + sequence.matchLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    {   const BYTE* match = op - sequence.offset;
+
+        /* check */
+        if (sequence.offset > (size_t)op) return ERROR(corruption_detected);   /* address space overflow test (this test seems kept by clang optimizer) */
+        //if (match > op) return ERROR(corruption_detected);   /* address space overflow test (is clang optimizer removing this test ?) */
+        if (match < base) return ERROR(corruption_detected);
+
+        /* close range match, overlap */
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        }
+        else
+        {
+            ZSTD_copy8(op, match);
+        }
+        op += 8; match += 8;
+
+        if (oMatchEnd > oend-(16-MINMATCH))
+        {
+            if (op < oend_8)
+            {
+                ZSTD_wildcopy(op, match, oend_8 - op);
+                match += oend_8 - op;
+                op = oend_8;
+            }
+            while (op < oMatchEnd) *op++ = *match++;
+        }
+        else
+        {
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        }
+    }
+
+    return oMatchEnd - ostart;
+}
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = sequence.offset = 4;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (lastLLSize > 0) {
+                if (op != litPtr) memmove(op, litPtr, lastLLSize);
+                op += lastLLSize;
+            }
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    /* Decode literals sub-block */
+    size_t litCSize = ZSTD_decodeLiteralsBlock(ctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyUncompressedBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t ZSTD_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    ZSTD_DCtx ctx;
+    ctx.base = dst;
+    return ZSTD_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+MEM_STATIC void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv03_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = MEM_readLE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTD_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
+
+
+/* wrapper layer */
+
+unsigned ZSTDv03_isError(size_t code)
+{
+    return ZSTD_isError(code);
+}
+
+size_t ZSTDv03_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize)
+{
+    return ZSTD_decompress(dst, maxOriginalSize, src, compressedSize);
+}
+
+ZSTDv03_Dctx* ZSTDv03_createDCtx(void)
+{
+    return (ZSTDv03_Dctx*)ZSTD_createDCtx();
+}
+
+size_t ZSTDv03_freeDCtx(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_freeDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_resetDCtx(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_resetDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_nextSrcSizeToDecompress(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_decompressContinue(ZSTDv03_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue((ZSTD_DCtx*)dctx, dst, maxDstSize, src, srcSize);
+}
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v03.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v03.h
new file mode 100644
index 0000000..9bf3cce
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v03.h
@@ -0,0 +1,93 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V03_H_298734209782
+#define ZSTD_V03_H_298734209782
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv03_decompress() : decompress ZSTD frames compliant with v0.3.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv03_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+ /**
+ ZSTDv03_findFrameSizeInfoLegacy() : get the source length and decompressed bound of a ZSTD frame compliant with v0.3.x format
+     srcSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+     cSize (output parameter)  : the number of bytes that would be read to decompress this frame
+                                 or an error code if it fails (which can be tested using ZSTDv01_isError())
+     dBound (output parameter) : an upper-bound for the decompressed size of the data in the frame
+                                 or ZSTD_CONTENTSIZE_ERROR if an error occurs
+
+    note : assumes `cSize` and `dBound` are _not_ NULL.
+ */
+ void ZSTDv03_findFrameSizeInfoLegacy(const void *src, size_t srcSize,
+                                      size_t* cSize, unsigned long long* dBound);
+
+    /**
+ZSTDv03_isError() : tells if the result of ZSTDv03_decompress() is an error
+*/
+unsigned ZSTDv03_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv03_Dctx_s ZSTDv03_Dctx;
+ZSTDv03_Dctx* ZSTDv03_createDCtx(void);
+size_t ZSTDv03_freeDCtx(ZSTDv03_Dctx* dctx);
+
+size_t ZSTDv03_decompressDCtx(void* ctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+/* *************************************
+*  Streaming functions
+***************************************/
+size_t ZSTDv03_resetDCtx(ZSTDv03_Dctx* dctx);
+
+size_t ZSTDv03_nextSrcSizeToDecompress(ZSTDv03_Dctx* dctx);
+size_t ZSTDv03_decompressContinue(ZSTDv03_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv03_magicNumber 0xFD2FB523   /* v0.3 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V03_H_298734209782 */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v04.c b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v04.c
new file mode 100644
index 0000000..57be832
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v04.c
@@ -0,0 +1,3605 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+ /******************************************
+ *  Includes
+ ******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+#include "zstd_v04.h"
+#include "../common/error_private.h"
+
+
+/* ******************************************************************
+ *   mem.h
+ *******************************************************************/
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-*************************************
+*  Debug
+***************************************/
+#include "../common/debug.h"
+#ifndef assert
+#  define assert(condition) ((void)0)
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+*/
+#ifndef ZSTD_STATIC_H
+#define ZSTD_STATIC_H
+
+
+/* *************************************
+*  Types
+***************************************/
+#define ZSTD_WINDOWLOG_ABSOLUTEMIN 11
+
+/** from faster to stronger */
+typedef enum { ZSTD_fast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2 } ZSTD_strategy;
+
+typedef struct
+{
+    U64 srcSize;       /* optional : tells how much bytes are present in the frame. Use 0 if not known. */
+    U32 windowLog;     /* largest match distance : larger == more compression, more memory needed during decompression */
+    U32 contentLog;    /* full search segment : larger == more compression, slower, more memory (useless for fast) */
+    U32 hashLog;       /* dispatch table : larger == more memory, faster */
+    U32 searchLog;     /* nb of searches : larger == more compression, slower */
+    U32 searchLength;  /* size of matches : larger == faster decompression, sometimes less compression */
+    ZSTD_strategy strategy;
+} ZSTD_parameters;
+
+typedef ZSTDv04_Dctx ZSTD_DCtx;
+
+/* *************************************
+*  Advanced functions
+***************************************/
+/** ZSTD_decompress_usingDict
+*   Same as ZSTD_decompressDCtx, using a Dictionary content as prefix
+*   Note : dict can be NULL, in which case, it's equivalent to ZSTD_decompressDCtx() */
+static size_t ZSTD_decompress_usingDict(ZSTD_DCtx* ctx,
+                                             void* dst, size_t maxDstSize,
+                                       const void* src, size_t srcSize,
+                                       const void* dict,size_t dictSize);
+
+
+/* **************************************
+*  Streaming functions (direct mode)
+****************************************/
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx);
+static size_t ZSTD_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize);
+static void   ZSTD_decompress_insertDictionary(ZSTD_DCtx* ctx, const void* src, size_t srcSize);
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+
+/**
+  Streaming decompression, bufferless mode
+
+  A ZSTD_DCtx object is required to track streaming operations.
+  Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
+  A ZSTD_DCtx object can be re-used multiple times. Use ZSTD_resetDCtx() to return to fresh status.
+
+  First operation is to retrieve frame parameters, using ZSTD_getFrameParams().
+  This function doesn't consume its input. It needs enough input data to properly decode the frame header.
+  Objective is to retrieve *params.windowlog, to know minimum amount of memory required during decoding.
+  Result : 0 when successful, it means the ZSTD_parameters structure has been filled.
+           >0 : means there is not enough data into src. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTD_isError() (For example, if it's not a ZSTD header)
+
+  Then, you can optionally insert a dictionary.
+  This operation must mimic the compressor behavior, otherwise decompression will fail or be corrupted.
+
+  Then it's possible to start decompression.
+  Use ZSTD_nextSrcSizeToDecompress() and ZSTD_decompressContinue() alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTD_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTD_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+
+
+#endif  /* ZSTD_STATIC_H */
+
+
+/*
+    zstd_internal - common functions to include
+    Header File for include
+*/
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+/* *************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/* *************************************
+*  Common constants
+***************************************/
+#define ZSTD_MAGICNUMBER 0xFD2FB524   /* v0.4 */
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize_min = 5;
+#define ZSTD_frameHeaderSize_max 5         /* define, for static allocation */
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define MINMATCH 4
+#define REPCODE_STARTVALUE 4
+
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits) - 1)
+#define MaxLL  ((1<<LLbits) - 1)
+#define MaxOff ((1<<Offbits)- 1)
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+
+/* ******************************************
+*  Shared functions to include for inlining
+********************************************/
+static void ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file
+****************************************************************** */
+#ifndef FSE_H
+#define FSE_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+
+
+/* *****************************************
+*  FSE simple functions
+******************************************/
+static size_t FSE_decompress(void* dst,  size_t maxDstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+FSE_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'maxDstSize'.
+    return : size of regenerated data (<= maxDstSize)
+             or an error code, which can be tested using FSE_isError()
+
+    ** Important ** : FSE_decompress() doesn't decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+
+
+/* *****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+static unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+
+
+
+/* *****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSE_compress() does the following:
+1. count symbol occurrence from source[] into table count[]
+2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
+3. save normalized counters to memory buffer using writeNCount()
+4. build encoding table 'CTable' from normalized counters
+5. encode the data stream using encoding table 'CTable'
+
+FSE_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*!
+FSE_readNCount():
+   Read compactly saved 'normalizedCounter' from 'rBuffer'.
+   return : size read from 'rBuffer'
+            or an errorCode, which can be tested using FSE_isError()
+            maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+static  size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*!
+Constructor and Destructor of type FSE_DTable
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+/*!
+FSE_buildDTable():
+   Builds 'dt', which must be already allocated, using FSE_createDTable()
+   return : 0,
+            or an errorCode, which can be tested using FSE_isError() */
+static size_t FSE_buildDTable ( FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*!
+FSE_decompress_usingDTable():
+   Decompress compressed source 'cSrc' of size 'cSrcSize' using 'dt'
+   into 'dst' which must be already allocated.
+   return : size of regenerated data (necessarily <= maxDstSize)
+            or an errorCode, which can be tested using FSE_isError() */
+static  size_t FSE_decompress_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.
+This is performed by the function FSE_buildDTable().
+The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+'FSE_DTable' can then be used to decompress 'cSrc', with FSE_decompress_usingDTable().
+'cSrcSize' must be strictly correct, otherwise decompression will fail.
+FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=maxDstSize).
+If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
+*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSE_H */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSE_STATIC_H
+#define FSE_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSE_STATIC_H */
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/*!FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Dependencies
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/*-**************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+
+static size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSE_DTableHeader DTableH;
+    void* const tdPtr = dt+1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSE_DECODE_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return a<0 ? -a : a;
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const cell = (FSE_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    U32 fastMode;
+
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    fastMode = DTableH.fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef HUFF0_H
+#define HUFF0_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Dependency
+******************************************/
+#include <stddef.h>    /* size_t */
+
+
+/* ****************************************
+*  Huff0 simple functions
+******************************************/
+static size_t HUF_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+HUF_decompress():
+    Decompress Huff0 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    'dstSize' must be the exact size of original (uncompressed) data.
+    Note : in contrast with FSE, HUF_decompress can regenerate RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data, because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUF_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+static unsigned    HUF_isError(size_t code);        /* tells if a return value is an error code */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFF0_H */
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef HUFF0_STATIC_H
+#define HUFF0_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Static allocation macros
+******************************************/
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+/* ****************************************
+*  Huff0 detailed API
+******************************************/
+/*!
+HUF_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUF_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUF_decompressSXn_usingDTable
+
+*/
+static size_t HUF_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+static size_t HUF_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+static size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+static size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUFF0_STATIC_H */
+
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+
+/*-*******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)dtPtr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+
+
+
+#endif   /* ZSTD_CCOMMON_H_MODULE */
+
+
+/*
+    zstd - decompression module fo v0.4 legacy format
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTD_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/* *************************************
+*  Local types
+***************************************/
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/* *************************************
+*  Error Management
+***************************************/
+
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+/* *************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock } ZSTD_dStage;
+
+struct ZSTDv04_Dctx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTD_parameters params;
+    blockType_t bType;
+    ZSTD_dStage stage;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+    BYTE headerBuffer[ZSTD_frameHeaderSize_max];
+};  /* typedef'd to ZSTD_DCtx within "zstd_static.h" */
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+/** ZSTD_decodeFrameHeader_Part1
+*   decode the 1st part of the Frame Header, which tells Frame Header size.
+*   srcSize must be == ZSTD_frameHeaderSize_min
+*   @return : the full size of the Frame Header */
+static size_t ZSTD_decodeFrameHeader_Part1(ZSTD_DCtx* zc, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize != ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_MAGICNUMBER) return ERROR(prefix_unknown);
+    zc->headerSize = ZSTD_frameHeaderSize_min;
+    return zc->headerSize;
+}
+
+
+static size_t ZSTD_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize < ZSTD_frameHeaderSize_min) return ZSTD_frameHeaderSize_max;
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_MAGICNUMBER) return ERROR(prefix_unknown);
+    memset(params, 0, sizeof(*params));
+    params->windowLog = (((const BYTE*)src)[4] & 15) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
+    if ((((const BYTE*)src)[4] >> 4) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits */
+    return 0;
+}
+
+/** ZSTD_decodeFrameHeader_Part2
+*   decode the full Frame Header
+*   srcSize must be the size provided by ZSTD_decodeFrameHeader_Part1
+*   @return : 0, or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_decodeFrameHeader_Part2(ZSTD_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t result;
+    if (srcSize != zc->headerSize) return ERROR(srcSize_wrong);
+    result = ZSTD_getFrameParams(&(zc->params), src, srcSize);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    /* compressed */
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    default:
+        return ERROR(corruption_detected);   /* forbidden nominal case */
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL >= MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    if (litLength == MaxLL) {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else if (dumps + 3 <= de) {
+            litLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {   static const U32 offsetPrefix[MaxOff+1] = {
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+        if (offsetCode | !litLength) seqState->prevOffset = seq->offset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML) {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 3 <= de){
+            matchLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    static const int dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+    static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+    BYTE* const oLitEnd = op + sequence.litLength;
+    const size_t sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base))
+    {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd)
+        {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {
+            size_t length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+        }
+    }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        const int sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTD_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTD_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH))
+    {
+        if (op < oend_8)
+        {
+            ZSTD_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    }
+    else
+    {
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8, but must be signed */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTD_decompressSequences(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = 4;
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 4;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* DStream should be entirely and exactly consumed; otherwise data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (lastLLSize > 0) {
+                if (op != litPtr) memcpy(op, litPtr, lastLLSize);
+                op += lastLLSize;
+            }
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd)   /* not contiguous */
+    {
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    size_t litCSize;
+
+    if (srcSize > BLOCKSIZE) return ERROR(corruption_detected);
+
+    /* Decode literals sub-block */
+    litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(dctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompress_usingDict(ZSTD_DCtx* ctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+
+    /* init */
+    ZSTD_resetDCtx(ctx);
+    if (dict)
+    {
+        ZSTD_decompress_insertDictionary(ctx, dict, dictSize);
+        ctx->dictEnd = ctx->previousDstEnd;
+        ctx->vBase = (const char*)dst - ((const char*)(ctx->previousDstEnd) - (const char*)(ctx->base));
+        ctx->base = dst;
+    }
+    else
+    {
+        ctx->vBase = ctx->base = ctx->dictEnd = dst;
+    }
+
+    /* Frame Header */
+    {
+        size_t frameHeaderSize;
+        if (srcSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+        frameHeaderSize = ZSTD_decodeFrameHeader_Part1(ctx, src, ZSTD_frameHeaderSize_min);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+        frameHeaderSize = ZSTD_decodeFrameHeader_Part2(ctx, src, frameHeaderSize);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock_internal(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv04_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize_min) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    if (MEM_readLE32(src) != ZSTD_MAGICNUMBER) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTD_frameHeaderSize_min; remainingSize -= ZSTD_frameHeaderSize_min;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+/* ******************************
+*  Streaming Decompression API
+********************************/
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    ZSTD_checkContinuity(ctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (ctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        /* get frame header size */
+        if (srcSize != ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        ctx->headerSize = ZSTD_decodeFrameHeader_Part1(ctx, src, ZSTD_frameHeaderSize_min);
+        if (ZSTD_isError(ctx->headerSize)) return ctx->headerSize;
+        memcpy(ctx->headerBuffer, src, ZSTD_frameHeaderSize_min);
+        if (ctx->headerSize > ZSTD_frameHeaderSize_min) return ERROR(GENERIC);   /* impossible */
+        ctx->expected = 0;   /* not necessary to copy more */
+        /* fallthrough */
+    case ZSTDds_decodeFrameHeader:
+        /* get frame header */
+        {   size_t const result = ZSTD_decodeFrameHeader_Part2(ctx, ctx->headerBuffer, ctx->headerSize);
+            if (ZSTD_isError(result)) return result;
+            ctx->expected = ZSTD_blockHeaderSize;
+            ctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        /* Decode block header */
+        {   blockProperties_t bp;
+            size_t const blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+            if (ZSTD_isError(blockSize)) return blockSize;
+            if (bp.blockType == bt_end)
+            {
+                ctx->expected = 0;
+                ctx->stage = ZSTDds_getFrameHeaderSize;
+            }
+            else
+            {
+                ctx->expected = blockSize;
+                ctx->bType = bp.blockType;
+                ctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {
+            /* Decompress : block content */
+            size_t rSize;
+            switch(ctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTD_decompressBlock_internal(ctx, dst, maxDstSize, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTD_copyRawBlock(dst, maxDstSize, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);
+            }
+            ctx->stage = ZSTDds_decodeBlockHeader;
+            ctx->expected = ZSTD_blockHeaderSize;
+            ctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTD_decompress_insertDictionary(ZSTD_DCtx* ctx, const void* dict, size_t dictSize)
+{
+    ctx->dictEnd = ctx->previousDstEnd;
+    ctx->vBase = (const char*)dict - ((const char*)(ctx->previousDstEnd) - (const char*)(ctx->base));
+    ctx->base = dict;
+    ctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stdlib.h>
+
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFF_DCtx object is required to track streaming operation.
+*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
+*  Use ZBUFF_decompressInit() to start a new decompression operation.
+*  ZBUFF_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFF_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst .
+*  return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  output : 128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_readHeader, ZBUFFds_loadHeader, ZBUFFds_decodeHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFF_dStage;
+
+/* *** Resource management *** */
+
+#define ZSTD_frameHeaderSize_max 5   /* too magical, should come from reference */
+struct ZBUFFv04_DCtx_s {
+    ZSTD_DCtx* zc;
+    ZSTD_parameters params;
+    char* inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char* outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t hPos;
+    const char* dict;
+    size_t dictSize;
+    ZBUFF_dStage stage;
+    unsigned char headerBuffer[ZSTD_frameHeaderSize_max];
+};   /* typedef'd to ZBUFF_DCtx within "zstd_buffered.h" */
+
+typedef ZBUFFv04_DCtx ZBUFF_DCtx;
+
+
+static ZBUFF_DCtx* ZBUFF_createDCtx(void)
+{
+    ZBUFF_DCtx* zbc = (ZBUFF_DCtx*)malloc(sizeof(ZBUFF_DCtx));
+    if (zbc==NULL) return NULL;
+    memset(zbc, 0, sizeof(*zbc));
+    zbc->zc = ZSTD_createDCtx();
+    zbc->stage = ZBUFFds_init;
+    return zbc;
+}
+
+static size_t ZBUFF_freeDCtx(ZBUFF_DCtx* zbc)
+{
+    if (zbc==NULL) return 0;   /* support free on null */
+    ZSTD_freeDCtx(zbc->zc);
+    free(zbc->inBuff);
+    free(zbc->outBuff);
+    free(zbc);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+static size_t ZBUFF_decompressInit(ZBUFF_DCtx* zbc)
+{
+    zbc->stage = ZBUFFds_readHeader;
+    zbc->hPos = zbc->inPos = zbc->outStart = zbc->outEnd = zbc->dictSize = 0;
+    return ZSTD_resetDCtx(zbc->zc);
+}
+
+
+static size_t ZBUFF_decompressWithDictionary(ZBUFF_DCtx* zbc, const void* src, size_t srcSize)
+{
+    zbc->dict = (const char*)src;
+    zbc->dictSize = srcSize;
+    return 0;
+}
+
+static size_t ZBUFF_limitCopy(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    size_t length = MIN(maxDstSize, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+/* *** Decompression *** */
+
+static size_t ZBUFF_decompressContinue(ZBUFF_DCtx* zbc, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* ip = istart;
+    const char* const iend = istart + *srcSizePtr;
+    char* const ostart = (char*)dst;
+    char* op = ostart;
+    char* const oend = ostart + *maxDstSizePtr;
+    U32 notDone = 1;
+
+    DEBUGLOG(5, "ZBUFF_decompressContinue");
+    while (notDone)
+    {
+        switch(zbc->stage)
+        {
+
+        case ZBUFFds_init :
+            DEBUGLOG(5, "ZBUFF_decompressContinue: stage==ZBUFFds_init => ERROR(init_missing)");
+            return ERROR(init_missing);
+
+        case ZBUFFds_readHeader :
+            /* read header from src */
+            {   size_t const headerSize = ZSTD_getFrameParams(&(zbc->params), src, *srcSizePtr);
+                if (ZSTD_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    memcpy(zbc->headerBuffer+zbc->hPos, src, *srcSizePtr);
+                    zbc->hPos += *srcSizePtr;
+                    *maxDstSizePtr = 0;
+                    zbc->stage = ZBUFFds_loadHeader;
+                    return headerSize - zbc->hPos;
+                }
+                zbc->stage = ZBUFFds_decodeHeader;
+                break;
+            }
+
+        case ZBUFFds_loadHeader:
+            /* complete header from src */
+            {   size_t headerSize = ZBUFF_limitCopy(
+                    zbc->headerBuffer + zbc->hPos, ZSTD_frameHeaderSize_max - zbc->hPos,
+                    src, *srcSizePtr);
+                zbc->hPos += headerSize;
+                ip += headerSize;
+                headerSize = ZSTD_getFrameParams(&(zbc->params), zbc->headerBuffer, zbc->hPos);
+                if (ZSTD_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    *maxDstSizePtr = 0;
+                    return headerSize - zbc->hPos;
+            }   }
+            /* intentional fallthrough */
+
+        case ZBUFFds_decodeHeader:
+                /* apply header to create / resize buffers */
+                {   size_t const neededOutSize = (size_t)1 << zbc->params.windowLog;
+                    size_t const neededInSize = BLOCKSIZE;   /* a block is never > BLOCKSIZE */
+                    if (zbc->inBuffSize < neededInSize) {
+                        free(zbc->inBuff);
+                        zbc->inBuffSize = neededInSize;
+                        zbc->inBuff = (char*)malloc(neededInSize);
+                        if (zbc->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    if (zbc->outBuffSize < neededOutSize) {
+                        free(zbc->outBuff);
+                        zbc->outBuffSize = neededOutSize;
+                        zbc->outBuff = (char*)malloc(neededOutSize);
+                        if (zbc->outBuff == NULL) return ERROR(memory_allocation);
+                }   }
+                if (zbc->dictSize)
+                    ZSTD_decompress_insertDictionary(zbc->zc, zbc->dict, zbc->dictSize);
+                if (zbc->hPos) {
+                    /* some data already loaded into headerBuffer : transfer into inBuff */
+                    memcpy(zbc->inBuff, zbc->headerBuffer, zbc->hPos);
+                    zbc->inPos = zbc->hPos;
+                    zbc->hPos = 0;
+                    zbc->stage = ZBUFFds_load;
+                    break;
+                }
+                zbc->stage = ZBUFFds_read;
+		/* fall-through */
+        case ZBUFFds_read:
+            {
+                size_t neededInSize = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+                if (neededInSize==0)   /* end of frame */
+                {
+                    zbc->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize)
+                {
+                    /* directly decode from src */
+                    size_t decodedSize = ZSTD_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        ip, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbc->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {
+                size_t neededInSize = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+                size_t toLoad = neededInSize - zbc->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbc->inBuffSize - zbc->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFF_limitCopy(zbc->inBuff + zbc->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbc->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+                {
+                    size_t decodedSize = ZSTD_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        zbc->inBuff, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    zbc->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbc->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFds_flush;
+                    /* ZBUFFds_flush follows */
+                }
+            }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {
+                size_t toFlushSize = zbc->outEnd - zbc->outStart;
+                size_t flushedSize = ZBUFF_limitCopy(op, oend-op, zbc->outBuff + zbc->outStart, toFlushSize);
+                op += flushedSize;
+                zbc->outStart += flushedSize;
+                if (flushedSize == toFlushSize)
+                {
+                    zbc->stage = ZBUFFds_read;
+                    if (zbc->outStart + BLOCKSIZE > zbc->outBuffSize)
+                        zbc->outStart = zbc->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+        }
+    }
+
+    *srcSizePtr = ip-istart;
+    *maxDstSizePtr = op-ostart;
+
+    {
+        size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+        if (nextSrcSizeHint > 3) nextSrcSizeHint+= 3;   /* get the next block header while at it */
+        nextSrcSizeHint -= zbc->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv04_isError(size_t errorCode) { return ERR_isError(errorCode); }
+const char* ZBUFFv04_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+size_t ZBUFFv04_recommendedDInSize()  { return BLOCKSIZE + 3; }
+size_t ZBUFFv04_recommendedDOutSize() { return BLOCKSIZE; }
+
+
+
+/*- ========================================================================= -*/
+
+/* final wrapping stage */
+
+size_t ZSTDv04_decompressDCtx(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompress_usingDict(dctx, dst, maxDstSize, src, srcSize, NULL, 0);
+}
+
+size_t ZSTDv04_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE==1)
+    size_t regenSize;
+    ZSTD_DCtx* dctx = ZSTD_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv04_decompressDCtx(dctx, dst, maxDstSize, src, srcSize);
+    ZSTD_freeDCtx(dctx);
+    return regenSize;
+#else
+    ZSTD_DCtx dctx;
+    return ZSTDv04_decompressDCtx(&dctx, dst, maxDstSize, src, srcSize);
+#endif
+}
+
+size_t ZSTDv04_resetDCtx(ZSTDv04_Dctx* dctx) { return ZSTD_resetDCtx(dctx); }
+
+size_t ZSTDv04_nextSrcSizeToDecompress(ZSTDv04_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress(dctx);
+}
+
+size_t ZSTDv04_decompressContinue(ZSTDv04_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+
+ZBUFFv04_DCtx* ZBUFFv04_createDCtx(void) { return ZBUFF_createDCtx(); }
+size_t ZBUFFv04_freeDCtx(ZBUFFv04_DCtx* dctx) { return ZBUFF_freeDCtx(dctx); }
+
+size_t ZBUFFv04_decompressInit(ZBUFFv04_DCtx* dctx) { return ZBUFF_decompressInit(dctx); }
+size_t ZBUFFv04_decompressWithDictionary(ZBUFFv04_DCtx* dctx, const void* src, size_t srcSize)
+{ return ZBUFF_decompressWithDictionary(dctx, src, srcSize); }
+
+size_t ZBUFFv04_decompressContinue(ZBUFFv04_DCtx* dctx, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    DEBUGLOG(5, "ZBUFFv04_decompressContinue");
+    return ZBUFF_decompressContinue(dctx, dst, maxDstSizePtr, src, srcSizePtr);
+}
+
+ZSTD_DCtx* ZSTDv04_createDCtx(void) { return ZSTD_createDCtx(); }
+size_t ZSTDv04_freeDCtx(ZSTD_DCtx* dctx) { return ZSTD_freeDCtx(dctx); }
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v04.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v04.h
new file mode 100644
index 0000000..640240d
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v04.h
@@ -0,0 +1,142 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V04_H_91868324769238
+#define ZSTD_V04_H_91868324769238
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv04_decompress() : decompress ZSTD frames compliant with v0.4.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv04_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+ /**
+ ZSTDv04_findFrameSizeInfoLegacy() : get the source length and decompressed bound of a ZSTD frame compliant with v0.4.x format
+     srcSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+     cSize (output parameter)  : the number of bytes that would be read to decompress this frame
+                                 or an error code if it fails (which can be tested using ZSTDv01_isError())
+     dBound (output parameter) : an upper-bound for the decompressed size of the data in the frame
+                                 or ZSTD_CONTENTSIZE_ERROR if an error occurs
+
+    note : assumes `cSize` and `dBound` are _not_ NULL.
+ */
+ void ZSTDv04_findFrameSizeInfoLegacy(const void *src, size_t srcSize,
+                                      size_t* cSize, unsigned long long* dBound);
+
+/**
+ZSTDv04_isError() : tells if the result of ZSTDv04_decompress() is an error
+*/
+unsigned ZSTDv04_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv04_Dctx_s ZSTDv04_Dctx;
+ZSTDv04_Dctx* ZSTDv04_createDCtx(void);
+size_t ZSTDv04_freeDCtx(ZSTDv04_Dctx* dctx);
+
+size_t ZSTDv04_decompressDCtx(ZSTDv04_Dctx* dctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+
+/* *************************************
+*  Direct Streaming
+***************************************/
+size_t ZSTDv04_resetDCtx(ZSTDv04_Dctx* dctx);
+
+size_t ZSTDv04_nextSrcSizeToDecompress(ZSTDv04_Dctx* dctx);
+size_t ZSTDv04_decompressContinue(ZSTDv04_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+
+/* *************************************
+*  Buffered Streaming
+***************************************/
+typedef struct ZBUFFv04_DCtx_s ZBUFFv04_DCtx;
+ZBUFFv04_DCtx* ZBUFFv04_createDCtx(void);
+size_t         ZBUFFv04_freeDCtx(ZBUFFv04_DCtx* dctx);
+
+size_t ZBUFFv04_decompressInit(ZBUFFv04_DCtx* dctx);
+size_t ZBUFFv04_decompressWithDictionary(ZBUFFv04_DCtx* dctx, const void* dict, size_t dictSize);
+
+size_t ZBUFFv04_decompressContinue(ZBUFFv04_DCtx* dctx, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr);
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFF_DCtx object is required to track streaming operation.
+*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
+*  Use ZBUFF_decompressInit() to start a new decompression operation.
+*  ZBUFF_DCtx objects can be reused multiple times.
+*
+*  Optionally, a reference to a static dictionary can be set, using ZBUFF_decompressWithDictionary()
+*  It must be the same content as the one set during compression phase.
+*  Dictionary content must remain accessible during the decompression process.
+*
+*  Use ZBUFF_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFF_recommendedDInSize / ZBUFF_recommendedDOutSize
+*  output : ZBUFF_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : ZBUFF_recommendedDInSize==128Kb+3; just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+unsigned ZBUFFv04_isError(size_t errorCode);
+const char* ZBUFFv04_getErrorName(size_t errorCode);
+
+
+/** The below functions provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are not compulsory, they just tend to offer better latency */
+size_t ZBUFFv04_recommendedDInSize(void);
+size_t ZBUFFv04_recommendedDOutSize(void);
+
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv04_magicNumber 0xFD2FB524   /* v0.4 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V04_H_91868324769238 */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v05.c b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v05.c
new file mode 100644
index 0000000..93a1169
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v05.c
@@ -0,0 +1,4004 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include "zstd_v05.h"
+#include "../common/error_private.h"
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write32(void* memPtr, U32 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write64(void* memPtr, U64 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+#ifndef ZSTD_STATIC_H
+#define ZSTD_STATIC_H
+
+/* The prototypes defined within this file are considered experimental.
+ * They should not be used in the context DLL as they may change in the future.
+ * Prefer static linking if you need them, to control breaking version changes issues.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-*************************************
+*  Types
+***************************************/
+#define ZSTDv05_WINDOWLOG_ABSOLUTEMIN 11
+
+
+/*-*************************************
+*  Advanced functions
+***************************************/
+/*- Advanced Decompression functions -*/
+
+/*! ZSTDv05_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv05_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv05_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference, which will not be modified, and 1 to run the decompression operation */
+size_t ZSTDv05_decompress_usingPreparedDCtx(
+                                             ZSTDv05_DCtx* dctx, const ZSTDv05_DCtx* preparedDCtx,
+                                             void* dst, size_t dstCapacity,
+                                       const void* src, size_t srcSize);
+
+
+/* **************************************
+*  Streaming functions (direct mode)
+****************************************/
+size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx);
+
+/*
+  Streaming decompression, direct mode (bufferless)
+
+  A ZSTDv05_DCtx object is required to track streaming operations.
+  Use ZSTDv05_createDCtx() / ZSTDv05_freeDCtx() to manage it.
+  A ZSTDv05_DCtx object can be re-used multiple times.
+
+  First typical operation is to retrieve frame parameters, using ZSTDv05_getFrameParams().
+  This operation is independent, and just needs enough input data to properly decode the frame header.
+  Objective is to retrieve *params.windowlog, to know minimum amount of memory required during decoding.
+  Result : 0 when successful, it means the ZSTDv05_parameters structure has been filled.
+           >0 : means there is not enough data into src. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv05_isError()
+
+  Start decompression, with ZSTDv05_decompressBegin() or ZSTDv05_decompressBegin_usingDict()
+  Alternatively, you can copy a prepared context, using ZSTDv05_copyDCtx()
+
+  Then use ZSTDv05_nextSrcSizeToDecompress() and ZSTDv05_decompressContinue() alternatively.
+  ZSTDv05_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv05_decompressContinue().
+  ZSTDv05_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTDv05_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTDv05_decompressContinue() is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTDv05_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTDv05_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    User will have to take in charge required information to regenerate data, such as block sizes.
+
+    A few rules to respect :
+    - Uncompressed block size must be <= 128 KB
+    - Compressing or decompressing requires a context structure
+      + Use ZSTDv05_createCCtx() and ZSTDv05_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv05_compressBegin()
+      + decompression : ZSTDv05_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - When a block is considered not compressible enough, ZSTDv05_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv05_decompressBlock() doesn't accept uncompressed data as input !!
+*/
+
+size_t ZSTDv05_decompressBlock(ZSTDv05_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv05_STATIC_H */
+
+
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+*/
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv05_DICT_MAGIC  0xEC30A435
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+static const size_t ZSTDv05_blockHeaderSize = 3;
+static const size_t ZSTDv05_frameHeaderSize_min = 5;
+#define ZSTDv05_frameHeaderSize_max 5         /* define, for static allocation */
+
+#define BITv057 128
+#define BITv056  64
+#define BITv055  32
+#define BITv054  16
+#define BITv051   2
+#define BITv050   1
+
+#define IS_HUFv05 0
+#define IS_PCH 1
+#define IS_RAW 2
+#define IS_RLE 3
+
+#define MINMATCH 4
+#define REPCODE_STARTVALUE 1
+
+#define Litbits  8
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  ((1<<MLbits) - 1)
+#define MaxLL  ((1<<LLbits) - 1)
+#define MaxOff ((1<<Offbits)- 1)
+#define MLFSEv05Log   10
+#define LLFSEv05Log   10
+#define OffFSEv05Log   9
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define FSEv05_ENCODING_RAW     0
+#define FSEv05_ENCODING_RLE     1
+#define FSEv05_ENCODING_STATIC  2
+#define FSEv05_ENCODING_DYNAMIC 3
+
+
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define WILDCOPY_OVERLENGTH 8
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv05_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTDv05_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv05_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+MEM_STATIC void ZSTDv05_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+    /* opt */
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+} seqStore_t;
+
+
+
+#endif   /* ZSTDv05_CCOMMON_H_MODULE */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   header file
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv05_H
+#define FSEv05_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+
+
+/*-****************************************
+*  FSEv05 simple functions
+******************************************/
+size_t FSEv05_decompress(void* dst,  size_t maxDstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+FSEv05_decompress():
+    Decompress FSEv05 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'maxDstSize'.
+    return : size of regenerated data (<= maxDstSize)
+             or an error code, which can be tested using FSEv05_isError()
+
+    ** Important ** : FSEv05_decompress() doesn't decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+
+
+/* *****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+unsigned    FSEv05_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv05_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+
+
+/* *****************************************
+*  FSEv05 detailed API
+******************************************/
+/* *** DECOMPRESSION *** */
+
+/*!
+FSEv05_readNCount():
+   Read compactly saved 'normalizedCounter' from 'rBuffer'.
+   return : size read from 'rBuffer'
+            or an errorCode, which can be tested using FSEv05_isError()
+            maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv05_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*!
+Constructor and Destructor of type FSEv05_DTable
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv05_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv05_DTable* FSEv05_createDTable(unsigned tableLog);
+void        FSEv05_freeDTable(FSEv05_DTable* dt);
+
+/*!
+FSEv05_buildDTable():
+   Builds 'dt', which must be already allocated, using FSEv05_createDTable()
+   @return : 0,
+             or an errorCode, which can be tested using FSEv05_isError() */
+size_t FSEv05_buildDTable (FSEv05_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*!
+FSEv05_decompress_usingDTable():
+   Decompress compressed source @cSrc of size @cSrcSize using `dt`
+   into `dst` which must be already allocated.
+   @return : size of regenerated data (necessarily <= @dstCapacity)
+             or an errorCode, which can be tested using FSEv05_isError() */
+size_t FSEv05_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv05_DTable* dt);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv05_H */
+/* ******************************************************************
+   bitstream
+   Part of FSEv05 library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITv05STREAM_H_MODULE
+#define BITv05STREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv05_DStream_t;
+
+typedef enum { BITv05_DStream_unfinished = 0,
+               BITv05_DStream_endOfBuffer = 1,
+               BITv05_DStream_completed = 2,
+               BITv05_DStream_overflow = 3 } BITv05_DStream_status;  /* result of BITv05_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv05_initDStream(BITv05_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv05_readBits(BITv05_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv05_DStream_status BITv05_reloadDStream(BITv05_DStream_t* bitD);
+MEM_STATIC unsigned BITv05_endOfDStream(const BITv05_DStream_t* bitD);
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BITv05_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*!BITv05_initDStream
+*  Initialize a BITv05_DStream_t.
+*  @bitD : a pointer to an already allocated BITv05_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv05_initDStream(BITv05_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t)) {  /* normal case */
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BITv05_highbit32(contain32);
+    } else {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BITv05_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BITv05_lookBits(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv05_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv05_lookBitsFast(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv05_skipBits(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv05_readBits(BITv05_DStream_t* bitD, unsigned nbBits)
+{
+    size_t value = BITv05_lookBits(bitD, nbBits);
+    BITv05_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BITv05_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits)
+{
+    size_t value = BITv05_lookBitsFast(bitD, nbBits);
+    BITv05_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv05_DStream_status BITv05_reloadDStream(BITv05_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BITv05_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv05_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv05_DStream_endOfBuffer;
+        return BITv05_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv05_DStream_status result = BITv05_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv05_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv05_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BITv05_endOfDStream(const BITv05_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITv05STREAM_H_MODULE */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv05_STATIC_H
+#define FSEv05_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* It is possible to statically allocate FSEv05 CTable/DTable as a table of unsigned using below macros */
+#define FSEv05_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSEv05 advanced API
+*******************************************/
+size_t FSEv05_buildDTable_raw (FSEv05_DTable* dt, unsigned nbBits);
+/* build a fake FSEv05_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv05_buildDTable_rle (FSEv05_DTable* dt, unsigned char symbolValue);
+/* build a fake FSEv05_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSEv05 symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv05_DState_t;
+
+
+static void     FSEv05_initDState(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD, const FSEv05_DTable* dt);
+
+static unsigned char FSEv05_decodeSymbol(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD);
+
+static unsigned FSEv05_endOfDState(const FSEv05_DState_t* DStatePtr);
+
+
+
+/* *****************************************
+*  FSEv05 unsafe API
+*******************************************/
+static unsigned char FSEv05_decodeSymbolFast(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv05_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv05_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv05_initDState(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD, const FSEv05_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv05_DTableHeader* const DTableH = (const FSEv05_DTableHeader*)ptr;
+    DStatePtr->state = BITv05_readBits(bitD, DTableH->tableLog);
+    BITv05_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv05_peakSymbol(FSEv05_DState_t* DStatePtr)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC BYTE FSEv05_decodeSymbol(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BITv05_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSEv05_decodeSymbolFast(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BITv05_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSEv05_endOfDState(const FSEv05_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv05_STATIC_H */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSEv05_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv05_MAX_MEMORY_USAGE 14
+#define FSEv05_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv05_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv05_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv05_FUNCTION_TYPE BYTE
+#define FSEv05_FUNCTION_EXTENSION
+#define FSEv05_DECODE_TYPE FSEv05_decode_t
+
+
+#endif   /* !FSEv05_COMMONDEFS_ONLY */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv05_MAX_TABLELOG  (FSEv05_MAX_MEMORY_USAGE-2)
+#define FSEv05_MAX_TABLESIZE (1U<<FSEv05_MAX_TABLELOG)
+#define FSEv05_MAXTABLESIZE_MASK (FSEv05_MAX_TABLESIZE-1)
+#define FSEv05_DEFAULT_TABLELOG (FSEv05_DEFAULT_MEMORY_USAGE-2)
+#define FSEv05_MIN_TABLELOG 5
+
+#define FSEv05_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv05_MAX_TABLELOG > FSEv05_TABLELOG_ABSOLUTE_MAX
+#error "FSEv05_MAX_TABLELOG > FSEv05_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv05_STATIC_ASSERT(c) { enum { FSEv05_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef unsigned DTable_max_t[FSEv05_DTABLE_SIZE_U32(FSEv05_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv05_FUNCTION_EXTENSION
+#  error "FSEv05_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv05_FUNCTION_TYPE
+#  error "FSEv05_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv05_CAT(X,Y) X##Y
+#define FSEv05_FUNCTION_NAME(X,Y) FSEv05_CAT(X,Y)
+#define FSEv05_TYPE_NAME(X,Y) FSEv05_CAT(X,Y)
+
+
+/* Function templates */
+static U32 FSEv05_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+
+
+FSEv05_DTable* FSEv05_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv05_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv05_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv05_DTable*)malloc( FSEv05_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv05_freeDTable (FSEv05_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv05_buildDTable(FSEv05_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSEv05_DTableHeader DTableH;
+    void* const tdPtr = dt+1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv05_DECODE_TYPE* const tableDecode = (FSEv05_DECODE_TYPE*) (tdPtr);
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSEv05_tableStep(tableSize);
+    U16 symbolNext[FSEv05_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv05_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv05_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSEv05_FUNCTION_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++) {
+        if (normalizedCounter[s]==-1) {
+            tableDecode[highThreshold--].symbol = (FSEv05_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        } else {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+    }   }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++) {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++) {
+            tableDecode[position].symbol = (FSEv05_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+    }   }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++) {
+            FSEv05_FUNCTION_TYPE symbol = (FSEv05_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BITv05_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+    }   }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSEv05_COMMONDEFS_ONLY
+/*-****************************************
+*  FSEv05 helper functions
+******************************************/
+unsigned FSEv05_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSEv05 NCount encoding-decoding
+****************************************************************/
+static short FSEv05_abs(short a) { return a<0 ? -a : a; }
+
+
+size_t FSEv05_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv05_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv05_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv05_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv05_buildDTable_rle (FSEv05_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv05_DTableHeader* const DTableH = (FSEv05_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv05_decode_t* const cell = (FSEv05_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv05_buildDTable_raw (FSEv05_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv05_DTableHeader* const DTableH = (FSEv05_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv05_decode_t* const dinfo = (FSEv05_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv05_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv05_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv05_DStream_t bitD;
+    FSEv05_DState_t state1;
+    FSEv05_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BITv05_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSEv05_isError(errorCode)) return errorCode;
+
+    FSEv05_initDState(&state1, &bitD, dt);
+    FSEv05_initDState(&state2, &bitD, dt);
+
+#define FSEv05_GETSYMBOL(statePtr) fast ? FSEv05_decodeSymbolFast(statePtr, &bitD) : FSEv05_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv05_reloadDStream(&bitD)==BITv05_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv05_GETSYMBOL(&state1);
+
+        if (FSEv05_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv05_reloadDStream(&bitD);
+
+        op[1] = FSEv05_GETSYMBOL(&state2);
+
+        if (FSEv05_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv05_reloadDStream(&bitD) > BITv05_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv05_GETSYMBOL(&state1);
+
+        if (FSEv05_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv05_reloadDStream(&bitD);
+
+        op[3] = FSEv05_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv05_reloadDStream(&bitD) >= FSEv05_DStream_partiallyFilled; Ends at exactly BITv05_DStream_completed */
+    while (1) {
+        if ( (BITv05_reloadDStream(&bitD)>BITv05_DStream_completed) || (op==omax) || (BITv05_endOfDStream(&bitD) && (fast || FSEv05_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSEv05_GETSYMBOL(&state1);
+
+        if ( (BITv05_reloadDStream(&bitD)>BITv05_DStream_completed) || (op==omax) || (BITv05_endOfDStream(&bitD) && (fast || FSEv05_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSEv05_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BITv05_endOfDStream(&bitD) && FSEv05_endOfDState(&state1) && FSEv05_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+size_t FSEv05_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv05_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv05_DTableHeader* DTableH = (const FSEv05_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv05_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv05_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv05_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv05_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv05_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSEv05 decoding mode */
+    errorCode = FSEv05_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSEv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSEv05_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSEv05_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSEv05_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSEv05_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFF0_H
+#define HUFF0_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Huff0 simple functions
+******************************************/
+size_t HUFv05_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+HUFv05_decompress():
+    Decompress Huff0 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    @dstSize : must be the **exact** size of original (uncompressed) data.
+    Note : in contrast with FSEv05, HUFv05_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUFv05_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+unsigned    HUFv05_isError(size_t code);        /* tells if a return value is an error code */
+const char* HUFv05_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUF0_H */
+/* ******************************************************************
+   Huff0 : Huffman codec, part of New Generation Entropy library
+   header file, for static linking only
+   Copyright (C) 2013-2016, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUF0_STATIC_H
+#define HUF0_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* static allocation of Huff0's DTable */
+#define HUFv05_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUFv05_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUFv05_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv05_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUFv05_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv05_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUFv05_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv05_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv05_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+/* ****************************************
+*  Huff0 detailed API
+******************************************/
+/*!
+HUFv05_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv05_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv05_decompressSXn_usingDTable
+*/
+size_t HUFv05_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+size_t HUFv05_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+size_t HUFv05_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv05_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+/* single stream variants */
+
+size_t HUFv05_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv05_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv05_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv05_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUF0_STATIC_H */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUFv05_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUFv05_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv05_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUFv05_ABSOLUTEMAX_TABLELOG */
+#define HUFv05_DEFAULT_TABLELOG  HUFv05_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUFv05_MAX_SYMBOL_VALUE 255
+#if (HUFv05_MAX_TABLELOG > HUFv05_ABSOLUTEMAX_TABLELOG)
+#  error "HUFv05_MAX_TABLELOG is too large !"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+unsigned HUFv05_isError(size_t code) { return ERR_isError(code); }
+const char* HUFv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+#define HUFv05_STATIC_ASSERT(c) { enum { HUFv05_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* *******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUFv05_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv05_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUFv05_readStats
+    Read compact Huffman tree, saved by HUFv05_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUFv05_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* memset(huffWeight, 0, hwSize); */   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2) {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }
+    else  {   /* header compressed with FSEv05 (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv05_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv05_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv05_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++) {
+        if (huffWeight[n] >= HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BITv05_highbit32(weightTotal) + 1;
+    if (tableLog > HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {   /* determine last weight */
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BITv05_highbit32(rest);
+        U32 lastWeight = BITv05_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+size_t HUFv05_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv05_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUFv05_DEltX2* const dt = (HUFv05_DEltX2*)dtPtr;
+
+    HUFv05_STATIC_ASSERT(sizeof(HUFv05_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv05_readStats(huffWeight, HUFv05_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv05_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof allocated DTable, from used size of DTable, in case of re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++) {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++) {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUFv05_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUFv05_decodeSymbolX2(BITv05_DStream_t* Dstream, const HUFv05_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BITv05_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BITv05_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv05_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv05_MAX_TABLELOG<=12)) \
+        HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv05_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv05_decodeStreamX2(BYTE* p, BITv05_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv05_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv05_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p < pEnd))
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+size_t HUFv05_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const U32 dtLog = DTable[0];
+    const void* dtPtr = DTable;
+    const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr)+1;
+    BITv05_DStream_t bitD;
+
+    if (dstSize <= cSrcSize) return ERROR(dstSize_tooSmall);
+    { size_t const errorCode = BITv05_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv05_isError(errorCode)) return errorCode; }
+
+    HUFv05_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv05_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv05_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX2(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUFv05_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv05_decompress1X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+size_t HUFv05_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv05_DStream_t bitD1;
+        BITv05_DStream_t bitD2;
+        BITv05_DStream_t bitD3;
+        BITv05_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv05_initDStream(&bitD1, istart1, length1);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD2, istart2, length2);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD3, istart3, length3);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD4, istart4, length4);
+        if (HUFv05_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv05_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv05_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv05_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv05_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv05_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX2(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUFv05_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv05_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+static void HUFv05_fillDTableX4Level2(HUFv05_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv05_DEltX4 DElt;
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUFv05_ABSOLUTEMAX_TABLELOG][HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUFv05_fillDTableX4(HUFv05_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv05_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            U32 i;
+            const U32 end = start + length;
+            HUFv05_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv05_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv05_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUFv05_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUFv05_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUFv05_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUFv05_DEltX4* const dt = ((HUFv05_DEltX4*)dtPtr) + 1;
+
+    HUFv05_STATIC_ASSERT(sizeof(HUFv05_DEltX4) == sizeof(unsigned));   /* if compilation fails here, assertion is false */
+    if (memLog > HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    /* memset(weightList, 0, sizeof(weightList)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv05_readStats(weightList, HUFv05_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv05_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++) {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++) {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++) {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }
+
+    HUFv05_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUFv05_decodeSymbolX4(void* op, BITv05_DStream_t* DStream, const HUFv05_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv05_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv05_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv05_decodeLastSymbolX4(void* op, BITv05_DStream_t* DStream, const HUFv05_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv05_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv05_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv05_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv05_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv05_MAX_TABLELOG<=12)) \
+        ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv05_decodeStreamX4(BYTE* p, BITv05_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv05_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv05_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p <= pEnd-2))
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv05_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+size_t HUFv05_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const unsigned* DTable)
+{
+    const BYTE* const istart = (const BYTE*) cSrc;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+
+    const U32 dtLog = DTable[0];
+    const void* const dtPtr = DTable;
+    const HUFv05_DEltX4* const dt = ((const HUFv05_DEltX4*)dtPtr) +1;
+    size_t errorCode;
+
+    /* Init */
+    BITv05_DStream_t bitD;
+    errorCode = BITv05_initDStream(&bitD, istart, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+
+    /* finish bitStreams one by one */
+    HUFv05_decodeStreamX4(ostart, &bitD, oend,     dt, dtLog);
+
+    /* check */
+    if (!BITv05_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv05_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX4(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv05_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv05_decompress1X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+size_t HUFv05_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const unsigned* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv05_DEltX4* const dt = ((const HUFv05_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv05_DStream_t bitD1;
+        BITv05_DStream_t bitD2;
+        BITv05_DStream_t bitD3;
+        BITv05_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv05_initDStream(&bitD1, istart1, length1);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD2, istart2, length2);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD3, istart3, length3);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD4, istart4, length4);
+        if (HUFv05_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv05_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv05_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv05_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv05_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv05_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv05_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX4(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv05_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv05_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv05_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUFv05_decompress4X2, HUFv05_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize >= dstSize) return ERROR(corruption_detected);   /* invalid, or not compressed, but not compressed already dealt with */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    /* return HUFv05_decompress4X2(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams single-symbol decoding */
+    /* return HUFv05_decompress4X4(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams double-symbols decoding */
+    /* return HUFv05_decompress4X6(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv05_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv05_HEAPMODE
+#  define ZSTDv05_HEAPMODE 1
+#endif
+
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug only : printf */
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/*-*************************************
+*  Local types
+***************************************/
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv05_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTDv05_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv05_isError(size_t code) { return ERR_isError(code); }
+
+
+/*! ZSTDv05_getErrorName() :
+*   provides error code string (useful for debugging) */
+const char* ZSTDv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* *************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDv05ds_getFrameHeaderSize, ZSTDv05ds_decodeFrameHeader,
+               ZSTDv05ds_decodeBlockHeader, ZSTDv05ds_decompressBlock } ZSTDv05_dStage;
+
+struct ZSTDv05_DCtx_s
+{
+    FSEv05_DTable LLTable[FSEv05_DTABLE_SIZE_U32(LLFSEv05Log)];
+    FSEv05_DTable OffTable[FSEv05_DTABLE_SIZE_U32(OffFSEv05Log)];
+    FSEv05_DTable MLTable[FSEv05_DTABLE_SIZE_U32(MLFSEv05Log)];
+    unsigned   hufTableX4[HUFv05_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTDv05_parameters params;
+    blockType_t bType;   /* used in ZSTDv05_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv05_dStage stage;
+    U32 flagStaticTables;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv05_frameHeaderSize_max];
+};  /* typedef'd to ZSTDv05_DCtx within "zstd_static.h" */
+
+size_t ZSTDv05_sizeofDCtx (void); /* Hidden declaration */
+size_t ZSTDv05_sizeofDCtx (void) { return sizeof(ZSTDv05_DCtx); }
+
+size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx)
+{
+    dctx->expected = ZSTDv05_frameHeaderSize_min;
+    dctx->stage = ZSTDv05ds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTableX4[0] = ZSTD_HUFFDTABLE_CAPACITY_LOG;
+    dctx->flagStaticTables = 0;
+    return 0;
+}
+
+ZSTDv05_DCtx* ZSTDv05_createDCtx(void)
+{
+    ZSTDv05_DCtx* dctx = (ZSTDv05_DCtx*)malloc(sizeof(ZSTDv05_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTDv05_decompressBegin(dctx);
+    return dctx;
+}
+
+size_t ZSTDv05_freeDCtx(ZSTDv05_DCtx* dctx)
+{
+    free(dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv05_copyDCtx(ZSTDv05_DCtx* dstDCtx, const ZSTDv05_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv05_DCtx) - (BLOCKSIZE+WILDCOPY_OVERLENGTH + ZSTDv05_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv05_MAGICNUMBER (defined within zstd_internal.h)
+      - 1 byte  - Window Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+/* Block format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+
+/** ZSTDv05_decodeFrameHeader_Part1() :
+*   decode the 1st part of the Frame Header, which tells Frame Header size.
+*   srcSize must be == ZSTDv05_frameHeaderSize_min.
+*   @return : the full size of the Frame Header */
+static size_t ZSTDv05_decodeFrameHeader_Part1(ZSTDv05_DCtx* zc, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize != ZSTDv05_frameHeaderSize_min)
+        return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTDv05_MAGICNUMBER) return ERROR(prefix_unknown);
+    zc->headerSize = ZSTDv05_frameHeaderSize_min;
+    return zc->headerSize;
+}
+
+
+size_t ZSTDv05_getFrameParams(ZSTDv05_parameters* params, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize < ZSTDv05_frameHeaderSize_min) return ZSTDv05_frameHeaderSize_max;
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTDv05_MAGICNUMBER) return ERROR(prefix_unknown);
+    memset(params, 0, sizeof(*params));
+    params->windowLog = (((const BYTE*)src)[4] & 15) + ZSTDv05_WINDOWLOG_ABSOLUTEMIN;
+    if ((((const BYTE*)src)[4] >> 4) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits */
+    return 0;
+}
+
+/** ZSTDv05_decodeFrameHeader_Part2() :
+*   decode the full Frame Header.
+*   srcSize must be the size provided by ZSTDv05_decodeFrameHeader_Part1().
+*   @return : 0, or an error code, which can be tested using ZSTDv05_isError() */
+static size_t ZSTDv05_decodeFrameHeader_Part2(ZSTDv05_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t result;
+    if (srcSize != zc->headerSize)
+        return ERROR(srcSize_wrong);
+    result = ZSTDv05_getFrameParams(&(zc->params), src, srcSize);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+static size_t ZSTDv05_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE*)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3)
+        return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv05_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/*! ZSTDv05_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTDv05_decodeLiteralsBlock(ZSTDv05_DCtx* dctx,
+                                    const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(istart[0]>> 6)
+    {
+    case IS_HUFv05:
+        {
+            size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3 */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv05_isError(singleStream ?
+                            HUFv05_decompress1X2(dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv05_decompress   (dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_PCH:
+        {
+            size_t errorCode;
+            size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (!dctx->flagStaticTables)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            errorCode = HUFv05_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTableX4);
+            if (HUFv05_isError(errorCode)) return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_RAW:
+        {
+            size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case IS_RLE:
+        {
+            size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+static size_t ZSTDv05_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSEv05_DTable* DTableLL, FSEv05_DTable* DTableML, FSEv05_DTable* DTableOffb,
+                         const void* src, size_t srcSize, U32 flagStaticTable)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    unsigned LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE)
+        return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = *ip++;
+    if (*nbSeq==0) return 1;
+    if (*nbSeq >= 128) {
+        if (ip >= iend) return ERROR(srcSize_wrong);
+        *nbSeq = ((nbSeq[0]-128)<<8) + *ip++;
+    }
+
+    if (ip >= iend) return ERROR(srcSize_wrong);
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2) {
+        if (ip+3 > iend) return ERROR(srcSize_wrong);
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    } else {
+        if (ip+2 > iend) return ERROR(srcSize_wrong);
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL >= MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            LLlog = 0;
+            FSEv05_buildDTable_rle(DTableLL, *ip++);
+            break;
+        case FSEv05_ENCODING_RAW :
+            LLlog = LLbits;
+            FSEv05_buildDTable_raw(DTableLL, LLbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   unsigned max = MaxLL;
+                headerSize = FSEv05_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSEv05_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case FSEv05_ENCODING_RAW :
+            Offlog = Offbits;
+            FSEv05_buildDTable_raw(DTableOffb, Offbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   unsigned max = MaxOff;
+                headerSize = FSEv05_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSEv05_buildDTable_rle(DTableML, *ip++);
+            break;
+        case FSEv05_ENCODING_RAW :
+            MLlog = MLbits;
+            FSEv05_buildDTable_raw(DTableML, MLbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   unsigned max = MaxML;
+                headerSize = FSEv05_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv05_DStream_t DStream;
+    FSEv05_DState_t stateLL;
+    FSEv05_DState_t stateOffb;
+    FSEv05_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+
+static void ZSTDv05_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSEv05_peakSymbol(&(seqState->stateLL));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    if (litLength == MaxLL) {
+        const U32 add = *dumps++;
+        if (add < 255) litLength += add;
+        else if (dumps + 2 <= de) {
+            litLength = MEM_readLE16(dumps);
+            dumps += 2;
+            if ((litLength & 1) && dumps < de) {
+                litLength += *dumps << 16;
+                dumps += 1;
+            }
+            litLength>>=1;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const U32 offsetPrefix[MaxOff+1] = {
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode = FSEv05_peakSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+        U32 nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BITv05_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BITv05_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* repcode, cmove */
+        if (offsetCode | !litLength) seqState->prevOffset = seq->offset;   /* cmove */
+        FSEv05_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));    /* update */
+    }
+
+    /* Literal length update */
+    FSEv05_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));   /* update */
+    if (MEM_32bits()) BITv05_reloadDStream(&(seqState->DStream));
+
+    /* MatchLength */
+    matchLength = FSEv05_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML) {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 2 <= de) {
+            matchLength = MEM_readLE16(dumps);
+            dumps += 2;
+            if ((matchLength & 1) && dumps < de) {
+                matchLength += *dumps << 16;
+                dumps += 1;
+            }
+            matchLength >>= 1;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+
+#if 0   /* debug */
+    {
+        static U64 totalDecoded = 0;
+        printf("pos %6u : %3u literals & match %3u bytes at distance %6u \n",
+           (U32)(totalDecoded), (U32)litLength, (U32)matchLength, (U32)offset);
+        totalDecoded += litLength + matchLength;
+    }
+#endif
+}
+
+
+static size_t ZSTDv05_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    static const int dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+    static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+    BYTE* const oLitEnd = op + sequence.litLength;
+    const size_t sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTDv05_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {
+            size_t length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        const int sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv05_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv05_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_8) {
+            ZSTDv05_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd)
+            *op++ = *match++;
+    } else {
+        ZSTDv05_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv05_decompressSequences(
+                               ZSTDv05_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength=0;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq=0;
+    const BYTE* dumps = NULL;
+    unsigned* DTableLL = dctx->LLTable;
+    unsigned* DTableML = dctx->MLTable;
+    unsigned* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTDv05_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, seqSize, dctx->flagStaticTables);
+    if (ZSTDv05_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = REPCODE_STARTVALUE;
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = REPCODE_STARTVALUE;
+        errorCode = BITv05_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSEv05_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv05_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv05_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv05_reloadDStream(&(seqState.DStream)) <= BITv05_DStream_completed) && nbSeq ; ) {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTDv05_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTDv05_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTDv05_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+    }
+
+    /* last literal segment */
+    {
+        size_t lastLLSize = litEnd - litPtr;
+        if (litPtr > litEnd) return ERROR(corruption_detected);   /* too many literals already used */
+        if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+        if (lastLLSize > 0) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv05_checkContinuity(ZSTDv05_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv05_decompressBlock_internal(ZSTDv05_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    size_t litCSize;
+
+    if (srcSize >= BLOCKSIZE) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    litCSize = ZSTDv05_decodeLiteralsBlock(dctx, src, srcSize);
+    if (ZSTDv05_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTDv05_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv05_decompressBlock(ZSTDv05_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+/*! ZSTDv05_decompress_continueDCtx
+*   dctx must have been properly initialized */
+static size_t ZSTDv05_decompress_continueDCtx(ZSTDv05_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+    memset(&blockProperties, 0, sizeof(blockProperties));
+
+    /* Frame Header */
+    {   size_t frameHeaderSize;
+        if (srcSize < ZSTDv05_frameHeaderSize_min+ZSTDv05_blockHeaderSize) return ERROR(srcSize_wrong);
+        frameHeaderSize = ZSTDv05_decodeFrameHeader_Part1(dctx, src, ZSTDv05_frameHeaderSize_min);
+        if (ZSTDv05_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv05_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+        frameHeaderSize = ZSTDv05_decodeFrameHeader_Part2(dctx, src, frameHeaderSize);
+        if (ZSTDv05_isError(frameHeaderSize)) return frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTDv05_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv05_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv05_blockHeaderSize;
+        remainingSize -= ZSTDv05_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv05_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv05_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv05_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+size_t ZSTDv05_decompress_usingPreparedDCtx(ZSTDv05_DCtx* dctx, const ZSTDv05_DCtx* refDCtx,
+                                         void* dst, size_t maxDstSize,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv05_copyDCtx(dctx, refDCtx);
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompress_continueDCtx(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+size_t ZSTDv05_decompress_usingDict(ZSTDv05_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv05_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompress_continueDCtx(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+size_t ZSTDv05_decompressDCtx(ZSTDv05_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTDv05_decompress_usingDict(dctx, dst, maxDstSize, src, srcSize, NULL, 0);
+}
+
+size_t ZSTDv05_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv05_HEAPMODE) && (ZSTDv05_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv05_DCtx* dctx = ZSTDv05_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv05_decompressDCtx(dctx, dst, maxDstSize, src, srcSize);
+    ZSTDv05_freeDCtx(dctx);
+    return regenSize;
+#else
+    ZSTDv05_DCtx dctx;
+    return ZSTDv05_decompressDCtx(&dctx, dst, maxDstSize, src, srcSize);
+#endif
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv05_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTDv05_frameHeaderSize_min) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    if (MEM_readLE32(src) != ZSTDv05_MAGICNUMBER) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTDv05_frameHeaderSize_min; remainingSize -= ZSTDv05_frameHeaderSize_min;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTDv05_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv05_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTDv05_blockHeaderSize;
+        remainingSize -= ZSTDv05_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+/* ******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv05_nextSrcSizeToDecompress(ZSTDv05_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+size_t ZSTDv05_decompressContinue(ZSTDv05_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    ZSTDv05_checkContinuity(dctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (dctx->stage)
+    {
+    case ZSTDv05ds_getFrameHeaderSize :
+        /* get frame header size */
+        if (srcSize != ZSTDv05_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        dctx->headerSize = ZSTDv05_decodeFrameHeader_Part1(dctx, src, ZSTDv05_frameHeaderSize_min);
+        if (ZSTDv05_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv05_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv05_frameHeaderSize_min) return ERROR(GENERIC); /* should never happen */
+        dctx->expected = 0;   /* not necessary to copy more */
+        /* fallthrough */
+    case ZSTDv05ds_decodeFrameHeader:
+        /* get frame header */
+        {   size_t const result = ZSTDv05_decodeFrameHeader_Part2(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv05_isError(result)) return result;
+            dctx->expected = ZSTDv05_blockHeaderSize;
+            dctx->stage = ZSTDv05ds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDv05ds_decodeBlockHeader:
+        {
+            /* Decode block header */
+            blockProperties_t bp;
+            size_t blockSize = ZSTDv05_getcBlockSize(src, ZSTDv05_blockHeaderSize, &bp);
+            if (ZSTDv05_isError(blockSize)) return blockSize;
+            if (bp.blockType == bt_end) {
+                dctx->expected = 0;
+                dctx->stage = ZSTDv05ds_getFrameHeaderSize;
+            }
+            else {
+                dctx->expected = blockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDv05ds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDv05ds_decompressBlock:
+        {
+            /* Decompress : block content */
+            size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv05_decompressBlock_internal(dctx, dst, maxDstSize, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv05_copyRawBlock(dst, maxDstSize, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDv05ds_decodeBlockHeader;
+            dctx->expected = ZSTDv05_blockHeaderSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTDv05_refDictContent(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+static size_t ZSTDv05_loadEntropy(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t hSize, offcodeHeaderSize, matchlengthHeaderSize, errorCode, litlengthHeaderSize;
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue=MaxOff, offcodeLog;
+    short matchlengthNCount[MaxML+1];
+    unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+    short litlengthNCount[MaxLL+1];
+    unsigned litlengthMaxValue = MaxLL, litlengthLog;
+
+    hSize = HUFv05_readDTableX4(dctx->hufTableX4, dict, dictSize);
+    if (HUFv05_isError(hSize)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + hSize;
+    dictSize -= hSize;
+
+    offcodeHeaderSize = FSEv05_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dict, dictSize);
+    if (FSEv05_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+    if (offcodeLog > OffFSEv05Log) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + offcodeHeaderSize;
+    dictSize -= offcodeHeaderSize;
+
+    matchlengthHeaderSize = FSEv05_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dict, dictSize);
+    if (FSEv05_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+    if (matchlengthLog > MLFSEv05Log) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + matchlengthHeaderSize;
+    dictSize -= matchlengthHeaderSize;
+
+    litlengthHeaderSize = FSEv05_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dict, dictSize);
+    if (litlengthLog > LLFSEv05Log) return ERROR(dictionary_corrupted);
+    if (FSEv05_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+
+    dctx->flagStaticTables = 1;
+    return hSize + offcodeHeaderSize + matchlengthHeaderSize + litlengthHeaderSize;
+}
+
+static size_t ZSTDv05_decompress_insertDictionary(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t eSize;
+    U32 magic = MEM_readLE32(dict);
+    if (magic != ZSTDv05_DICT_MAGIC) {
+        /* pure content mode */
+        ZSTDv05_refDictContent(dctx, dict, dictSize);
+        return 0;
+    }
+    /* load entropy tables */
+    dict = (const char*)dict + 4;
+    dictSize -= 4;
+    eSize = ZSTDv05_loadEntropy(dctx, dict, dictSize);
+    if (ZSTDv05_isError(eSize)) return ERROR(dictionary_corrupted);
+
+    /* reference dictionary content */
+    dict = (const char*)dict + eSize;
+    dictSize -= eSize;
+    ZSTDv05_refDictContent(dctx, dict, dictSize);
+
+    return 0;
+}
+
+
+size_t ZSTDv05_decompressBegin_usingDict(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t errorCode;
+    errorCode = ZSTDv05_decompressBegin(dctx);
+    if (ZSTDv05_isError(errorCode)) return errorCode;
+
+    if (dict && dictSize) {
+        errorCode = ZSTDv05_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+
+
+/* *************************************
+*  Constants
+***************************************/
+static size_t ZBUFFv05_blockHeaderSize = 3;
+
+
+
+/* *** Compression *** */
+
+static size_t ZBUFFv05_limitCopy(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    size_t length = MIN(maxDstSize, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+
+
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFFv05_DCtx object is required to track streaming operation.
+*  Use ZBUFFv05_createDCtx() and ZBUFFv05_freeDCtx() to create/release resources.
+*  Use ZBUFFv05_decompressInit() to start a new decompression operation.
+*  ZBUFFv05_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFFv05_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst .
+*  return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFFv05_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  output : 128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : just follow indications from ZBUFFv05_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+
+typedef enum { ZBUFFv05ds_init, ZBUFFv05ds_readHeader, ZBUFFv05ds_loadHeader, ZBUFFv05ds_decodeHeader,
+               ZBUFFv05ds_read, ZBUFFv05ds_load, ZBUFFv05ds_flush } ZBUFFv05_dStage;
+
+/* *** Resource management *** */
+
+#define ZSTDv05_frameHeaderSize_max 5   /* too magical, should come from reference */
+struct ZBUFFv05_DCtx_s {
+    ZSTDv05_DCtx* zc;
+    ZSTDv05_parameters params;
+    char* inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char* outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t hPos;
+    ZBUFFv05_dStage stage;
+    unsigned char headerBuffer[ZSTDv05_frameHeaderSize_max];
+};   /* typedef'd to ZBUFFv05_DCtx within "zstd_buffered.h" */
+
+
+ZBUFFv05_DCtx* ZBUFFv05_createDCtx(void)
+{
+    ZBUFFv05_DCtx* zbc = (ZBUFFv05_DCtx*)malloc(sizeof(ZBUFFv05_DCtx));
+    if (zbc==NULL) return NULL;
+    memset(zbc, 0, sizeof(*zbc));
+    zbc->zc = ZSTDv05_createDCtx();
+    zbc->stage = ZBUFFv05ds_init;
+    return zbc;
+}
+
+size_t ZBUFFv05_freeDCtx(ZBUFFv05_DCtx* zbc)
+{
+    if (zbc==NULL) return 0;   /* support free on null */
+    ZSTDv05_freeDCtx(zbc->zc);
+    free(zbc->inBuff);
+    free(zbc->outBuff);
+    free(zbc);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv05_decompressInitDictionary(ZBUFFv05_DCtx* zbc, const void* dict, size_t dictSize)
+{
+    zbc->stage = ZBUFFv05ds_readHeader;
+    zbc->hPos = zbc->inPos = zbc->outStart = zbc->outEnd = 0;
+    return ZSTDv05_decompressBegin_usingDict(zbc->zc, dict, dictSize);
+}
+
+size_t ZBUFFv05_decompressInit(ZBUFFv05_DCtx* zbc)
+{
+    return ZBUFFv05_decompressInitDictionary(zbc, NULL, 0);
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv05_decompressContinue(ZBUFFv05_DCtx* zbc, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* ip = istart;
+    const char* const iend = istart + *srcSizePtr;
+    char* const ostart = (char*)dst;
+    char* op = ostart;
+    char* const oend = ostart + *maxDstSizePtr;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbc->stage)
+        {
+        case ZBUFFv05ds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFv05ds_readHeader :
+            /* read header from src */
+            {
+                size_t headerSize = ZSTDv05_getFrameParams(&(zbc->params), src, *srcSizePtr);
+                if (ZSTDv05_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    memcpy(zbc->headerBuffer+zbc->hPos, src, *srcSizePtr);
+                    zbc->hPos += *srcSizePtr;
+                    *maxDstSizePtr = 0;
+                    zbc->stage = ZBUFFv05ds_loadHeader;
+                    return headerSize - zbc->hPos;
+                }
+                zbc->stage = ZBUFFv05ds_decodeHeader;
+                break;
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_loadHeader:
+            /* complete header from src */
+            {
+                size_t headerSize = ZBUFFv05_limitCopy(
+                    zbc->headerBuffer + zbc->hPos, ZSTDv05_frameHeaderSize_max - zbc->hPos,
+                    src, *srcSizePtr);
+                zbc->hPos += headerSize;
+                ip += headerSize;
+                headerSize = ZSTDv05_getFrameParams(&(zbc->params), zbc->headerBuffer, zbc->hPos);
+                if (ZSTDv05_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    *maxDstSizePtr = 0;
+                    return headerSize - zbc->hPos;
+                }
+                /* zbc->stage = ZBUFFv05ds_decodeHeader; break; */   /* useless : stage follows */
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_decodeHeader:
+                /* apply header to create / resize buffers */
+                {
+                    size_t neededOutSize = (size_t)1 << zbc->params.windowLog;
+                    size_t neededInSize = BLOCKSIZE;   /* a block is never > BLOCKSIZE */
+                    if (zbc->inBuffSize < neededInSize) {
+                        free(zbc->inBuff);
+                        zbc->inBuffSize = neededInSize;
+                        zbc->inBuff = (char*)malloc(neededInSize);
+                        if (zbc->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    if (zbc->outBuffSize < neededOutSize) {
+                        free(zbc->outBuff);
+                        zbc->outBuffSize = neededOutSize;
+                        zbc->outBuff = (char*)malloc(neededOutSize);
+                        if (zbc->outBuff == NULL) return ERROR(memory_allocation);
+                }   }
+                if (zbc->hPos) {
+                    /* some data already loaded into headerBuffer : transfer into inBuff */
+                    memcpy(zbc->inBuff, zbc->headerBuffer, zbc->hPos);
+                    zbc->inPos = zbc->hPos;
+                    zbc->hPos = 0;
+                    zbc->stage = ZBUFFv05ds_load;
+                    break;
+                }
+                zbc->stage = ZBUFFv05ds_read;
+		/* fall-through */
+        case ZBUFFv05ds_read:
+            {
+                size_t neededInSize = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+                if (neededInSize==0) {  /* end of frame */
+                    zbc->stage = ZBUFFv05ds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {
+                    /* directly decode from src */
+                    size_t decodedSize = ZSTDv05_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        ip, neededInSize);
+                    if (ZSTDv05_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFv05ds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbc->stage = ZBUFFv05ds_load;
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_load:
+            {
+                size_t neededInSize = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+                size_t toLoad = neededInSize - zbc->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbc->inBuffSize - zbc->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv05_limitCopy(zbc->inBuff + zbc->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbc->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+                {
+                    size_t decodedSize = ZSTDv05_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        zbc->inBuff, neededInSize);
+                    if (ZSTDv05_isError(decodedSize)) return decodedSize;
+                    zbc->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbc->stage = ZBUFFv05ds_read; break; }   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFv05ds_flush;
+                    /* break; */  /* ZBUFFv05ds_flush follows */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFv05ds_flush:
+            {
+                size_t toFlushSize = zbc->outEnd - zbc->outStart;
+                size_t flushedSize = ZBUFFv05_limitCopy(op, oend-op, zbc->outBuff + zbc->outStart, toFlushSize);
+                op += flushedSize;
+                zbc->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbc->stage = ZBUFFv05ds_read;
+                    if (zbc->outStart + BLOCKSIZE > zbc->outBuffSize)
+                        zbc->outStart = zbc->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    *srcSizePtr = ip-istart;
+    *maxDstSizePtr = op-ostart;
+
+    {   size_t nextSrcSizeHint = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+        if (nextSrcSizeHint > ZBUFFv05_blockHeaderSize) nextSrcSizeHint+= ZBUFFv05_blockHeaderSize;   /* get next block header too */
+        nextSrcSizeHint -= zbc->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv05_isError(size_t errorCode) { return ERR_isError(errorCode); }
+const char* ZBUFFv05_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+size_t ZBUFFv05_recommendedDInSize(void)  { return BLOCKSIZE + ZBUFFv05_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv05_recommendedDOutSize(void) { return BLOCKSIZE; }
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v05.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v05.h
new file mode 100644
index 0000000..2dcffc9
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v05.h
@@ -0,0 +1,162 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTDv05_H
+#define ZSTDv05_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stddef.h>   /* size_t */
+#include "../common/mem.h"      /* U64, U32 */
+
+
+/* *************************************
+*  Simple functions
+***************************************/
+/*! ZSTDv05_decompress() :
+    `compressedSize` : is the _exact_ size of the compressed blob, otherwise decompression will fail.
+    `dstCapacity` must be large enough, equal or larger than originalSize.
+    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+              or an errorCode if it fails (which can be tested using ZSTDv05_isError()) */
+size_t ZSTDv05_decompress( void* dst, size_t dstCapacity,
+                     const void* src, size_t compressedSize);
+
+ /**
+ ZSTDv05_findFrameSizeInfoLegacy() : get the source length and decompressed bound of a ZSTD frame compliant with v0.5.x format
+     srcSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+     cSize (output parameter)  : the number of bytes that would be read to decompress this frame
+                                 or an error code if it fails (which can be tested using ZSTDv01_isError())
+     dBound (output parameter) : an upper-bound for the decompressed size of the data in the frame
+                                 or ZSTD_CONTENTSIZE_ERROR if an error occurs
+
+    note : assumes `cSize` and `dBound` are _not_ NULL.
+ */
+void ZSTDv05_findFrameSizeInfoLegacy(const void *src, size_t srcSize,
+                                     size_t* cSize, unsigned long long* dBound);
+
+/* *************************************
+*  Helper functions
+***************************************/
+/* Error Management */
+unsigned    ZSTDv05_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+const char* ZSTDv05_getErrorName(size_t code);     /*!< provides readable string for an error code */
+
+
+/* *************************************
+*  Explicit memory management
+***************************************/
+/** Decompression context */
+typedef struct ZSTDv05_DCtx_s ZSTDv05_DCtx;
+ZSTDv05_DCtx* ZSTDv05_createDCtx(void);
+size_t ZSTDv05_freeDCtx(ZSTDv05_DCtx* dctx);      /*!< @return : errorCode */
+
+/** ZSTDv05_decompressDCtx() :
+*   Same as ZSTDv05_decompress(), but requires an already allocated ZSTDv05_DCtx (see ZSTDv05_createDCtx()) */
+size_t ZSTDv05_decompressDCtx(ZSTDv05_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-***********************
+*  Simple Dictionary API
+*************************/
+/*! ZSTDv05_decompress_usingDict() :
+*   Decompression using a pre-defined Dictionary content (see dictBuilder).
+*   Dictionary must be identical to the one used during compression, otherwise regenerated data will be corrupted.
+*   Note : dict can be NULL, in which case, it's equivalent to ZSTDv05_decompressDCtx() */
+size_t ZSTDv05_decompress_usingDict(ZSTDv05_DCtx* dctx,
+                                            void* dst, size_t dstCapacity,
+                                      const void* src, size_t srcSize,
+                                      const void* dict,size_t dictSize);
+
+/*-************************
+*  Advanced Streaming API
+***************************/
+typedef enum { ZSTDv05_fast, ZSTDv05_greedy, ZSTDv05_lazy, ZSTDv05_lazy2, ZSTDv05_btlazy2, ZSTDv05_opt, ZSTDv05_btopt } ZSTDv05_strategy;
+typedef struct {
+    U64 srcSize;
+    U32 windowLog;     /* the only useful information to retrieve */
+    U32 contentLog; U32 hashLog; U32 searchLog; U32 searchLength; U32 targetLength; ZSTDv05_strategy strategy;
+} ZSTDv05_parameters;
+size_t ZSTDv05_getFrameParams(ZSTDv05_parameters* params, const void* src, size_t srcSize);
+
+size_t ZSTDv05_decompressBegin_usingDict(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize);
+void   ZSTDv05_copyDCtx(ZSTDv05_DCtx* dstDCtx, const ZSTDv05_DCtx* srcDCtx);
+size_t ZSTDv05_nextSrcSizeToDecompress(ZSTDv05_DCtx* dctx);
+size_t ZSTDv05_decompressContinue(ZSTDv05_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-***********************
+*  ZBUFF API
+*************************/
+typedef struct ZBUFFv05_DCtx_s ZBUFFv05_DCtx;
+ZBUFFv05_DCtx* ZBUFFv05_createDCtx(void);
+size_t         ZBUFFv05_freeDCtx(ZBUFFv05_DCtx* dctx);
+
+size_t ZBUFFv05_decompressInit(ZBUFFv05_DCtx* dctx);
+size_t ZBUFFv05_decompressInitDictionary(ZBUFFv05_DCtx* dctx, const void* dict, size_t dictSize);
+
+size_t ZBUFFv05_decompressContinue(ZBUFFv05_DCtx* dctx,
+                                            void* dst, size_t* dstCapacityPtr,
+                                      const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression
+*
+*  A ZBUFFv05_DCtx object is required to track streaming operations.
+*  Use ZBUFFv05_createDCtx() and ZBUFFv05_freeDCtx() to create/release resources.
+*  Use ZBUFFv05_decompressInit() to start a new decompression operation,
+*   or ZBUFFv05_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv05_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFFv05_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFFv05_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv05_recommendedDInSize() / ZBUFFv05_recommendedDOutSize()
+*  output : ZBUFFv05_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv05_recommendedDInSize==128Kb+3; just follow indications from ZBUFFv05_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv05_isError(size_t errorCode);
+const char* ZBUFFv05_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, and tend to offer better latency */
+size_t ZBUFFv05_recommendedDInSize(void);
+size_t ZBUFFv05_recommendedDOutSize(void);
+
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define ZSTDv05_MAGICNUMBER 0xFD2FB525   /* v0.5 */
+
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv0505_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v06.c b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v06.c
new file mode 100644
index 0000000..175f7cc
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v06.c
@@ -0,0 +1,4113 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include "zstd_v06.h"
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+#include <stdlib.h>    /* malloc, free, qsort */
+#include "../common/error_private.h"
+
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap64(in);
+#else
+    return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+#endif
+}
+
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+#ifndef ZSTDv06_STATIC_H
+#define ZSTDv06_STATIC_H
+
+/* The prototypes defined within this file are considered experimental.
+ * They should not be used in the context DLL as they may change in the future.
+ * Prefer static linking if you need them, to control breaking version changes issues.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*- Advanced Decompression functions -*/
+
+/*! ZSTDv06_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv06_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv06_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
+ZSTDLIBv06_API size_t ZSTDv06_decompress_usingPreparedDCtx(
+                                           ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* preparedDCtx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize);
+
+
+
+#define ZSTDv06_FRAMEHEADERSIZE_MAX 13    /* for static allocation */
+static const size_t ZSTDv06_frameHeaderSize_min = 5;
+static const size_t ZSTDv06_frameHeaderSize_max = ZSTDv06_FRAMEHEADERSIZE_MAX;
+
+ZSTDLIBv06_API size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx);
+
+/*
+  Streaming decompression, direct mode (bufferless)
+
+  A ZSTDv06_DCtx object is required to track streaming operations.
+  Use ZSTDv06_createDCtx() / ZSTDv06_freeDCtx() to manage it.
+  A ZSTDv06_DCtx object can be re-used multiple times.
+
+  First optional operation is to retrieve frame parameters, using ZSTDv06_getFrameParams(), which doesn't consume the input.
+  It can provide the minimum size of rolling buffer required to properly decompress data,
+  and optionally the final size of uncompressed content.
+  (Note : content size is an optional info that may not be present. 0 means : content size unknown)
+  Frame parameters are extracted from the beginning of compressed frame.
+  The amount of data to read is variable, from ZSTDv06_frameHeaderSize_min to ZSTDv06_frameHeaderSize_max (so if `srcSize` >= ZSTDv06_frameHeaderSize_max, it will always work)
+  If `srcSize` is too small for operation to succeed, function will return the minimum size it requires to produce a result.
+  Result : 0 when successful, it means the ZSTDv06_frameParams structure has been filled.
+          >0 : means there is not enough data into `src`. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv06_isError()
+
+  Start decompression, with ZSTDv06_decompressBegin() or ZSTDv06_decompressBegin_usingDict().
+  Alternatively, you can copy a prepared context, using ZSTDv06_copyDCtx().
+
+  Then use ZSTDv06_nextSrcSizeToDecompress() and ZSTDv06_decompressContinue() alternatively.
+  ZSTDv06_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv06_decompressContinue().
+  ZSTDv06_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTDv06_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTDv06_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity)
+  It can be zero, which is not an error; it just means ZSTDv06_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTDv06_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    User will have to take in charge required information to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Uncompressed block size must be <= ZSTDv06_BLOCKSIZE_MAX (128 KB)
+    - Compressing or decompressing requires a context structure
+      + Use ZSTDv06_createCCtx() and ZSTDv06_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv06_compressBegin()
+      + decompression : ZSTDv06_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - When a block is considered not compressible enough, ZSTDv06_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv06_decompressBlock() doesn't accept uncompressed data as input !!
+*/
+
+#define ZSTDv06_BLOCKSIZE_MAX (128 * 1024)   /* define, for static allocation */
+ZSTDLIBv06_API size_t ZSTDv06_decompressBlock(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv06_STATIC_H */
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://www.zstd.net
+*/
+#ifndef ZSTDv06_CCOMMON_H_MODULE
+#define ZSTDv06_CCOMMON_H_MODULE
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv06_DICT_MAGIC  0xEC30A436
+
+#define ZSTDv06_REP_NUM    3
+#define ZSTDv06_REP_INIT   ZSTDv06_REP_NUM
+#define ZSTDv06_REP_MOVE   (ZSTDv06_REP_NUM-1)
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTDv06_WINDOWLOG_ABSOLUTEMIN 12
+static const size_t ZSTDv06_fcs_fieldSize[4] = { 0, 1, 2, 8 };
+
+#define ZSTDv06_BLOCKHEADERSIZE 3   /* because C standard does not allow a static const value to be defined using another static const value .... :( */
+static const size_t ZSTDv06_blockHeaderSize = ZSTDv06_BLOCKHEADERSIZE;
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
+
+#define IS_HUF 0
+#define IS_PCH 1
+#define IS_RAW 2
+#define IS_RLE 3
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+#define EQUAL_READ32 4
+#define REPCODE_STARTVALUE 1
+
+#define Litbits  8
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  52
+#define MaxLL  35
+#define MaxOff 28
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+
+#define FSEv06_ENCODING_RAW     0
+#define FSEv06_ENCODING_RLE     1
+#define FSEv06_ENCODING_STATIC  2
+#define FSEv06_ENCODING_DYNAMIC 3
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+                                     13,14,15,16 };
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+                                            -1,-1,-1,-1 };
+static const U32 LL_defaultNormLog = 6;
+
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+                                     12,13,14,15,16 };
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+                                            -1,-1,-1,-1,-1 };
+static const U32 ML_defaultNormLog = 6;
+
+static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const U32 OF_defaultNormLog = 5;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv06_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+#define COPY8(d,s) { ZSTDv06_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv06_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+#define WILDCOPY_OVERLENGTH 8
+MEM_STATIC void ZSTDv06_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTDv06_match_t;
+
+typedef struct {
+    U32 price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTDv06_REP_INIT];
+} ZSTDv06_optimal_t;
+
+typedef struct { U32  unused; } ZSTDv06_stats_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* litStart;
+    BYTE* lit;
+    U16*  litLengthStart;
+    U16*  litLength;
+    BYTE* llCodeStart;
+    U16*  matchLengthStart;
+    U16*  matchLength;
+    BYTE* mlCodeStart;
+    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
+    U32   longLengthPos;
+    /* opt */
+    ZSTDv06_optimal_t* priceTable;
+    ZSTDv06_match_t* matchTable;
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  matchSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+    U32  log2matchLengthSum;
+    U32  log2matchSum;
+    U32  log2litLengthSum;
+    U32  log2litSum;
+    U32  log2offCodeSum;
+    U32  factor;
+    U32  cachedPrice;
+    U32  cachedLitLength;
+    const BYTE* cachedLiterals;
+    ZSTDv06_stats_t stats;
+} seqStore_t;
+
+void ZSTDv06_seqToCodes(const seqStore_t* seqStorePtr, size_t const nbSeq);
+
+
+#endif   /* ZSTDv06_CCOMMON_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy codec
+   Public Prototypes declaration
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef FSEv06_H
+#define FSEv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-****************************************
+*  FSE simple functions
+******************************************/
+/*! FSEv06_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstCapacity'.
+    @return : size of regenerated data (<= maxDstSize),
+              or an error code, which can be tested using FSEv06_isError() .
+
+    ** Important ** : FSEv06_decompress() does not decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+size_t FSEv06_decompress(void* dst,  size_t dstCapacity,
+                const void* cSrc, size_t cSrcSize);
+
+
+/*-*****************************************
+*  Tool functions
+******************************************/
+size_t FSEv06_compressBound(size_t size);       /* maximum compressed size */
+
+/* Error Management */
+unsigned    FSEv06_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv06_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+
+FSEv06_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSEv06_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSEv06_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv06_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*! Constructor and Destructor of FSEv06_DTable.
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv06_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv06_DTable* FSEv06_createDTable(unsigned tableLog);
+void        FSEv06_freeDTable(FSEv06_DTable* dt);
+
+/*! FSEv06_buildDTable():
+    Builds 'dt', which must be already allocated, using FSEv06_createDTable().
+    return : 0, or an errorCode, which can be tested using FSEv06_isError() */
+size_t FSEv06_buildDTable (FSEv06_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSEv06_decompress_usingDTable():
+    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
+    into `dst` which must be already allocated.
+    @return : size of regenerated data (necessarily <= `dstCapacity`),
+              or an errorCode, which can be tested using FSEv06_isError() */
+size_t FSEv06_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv06_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSEv06_readNCount() if it was saved using FSEv06_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSEv06_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSEv06_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError().
+
+The next step is to build the decompression tables 'FSEv06_DTable' from 'normalizedCounter'.
+This is performed by the function FSEv06_buildDTable().
+The space required by 'FSEv06_DTable' must be already allocated using FSEv06_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError().
+
+`FSEv06_DTable` can then be used to decompress `cSrc`, with FSEv06_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSEv06_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError(). (ex: dst buffer too small)
+*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv06_H */
+/* ******************************************************************
+   bitstream
+   Part of FSE library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#if defined(__BMI__) && defined(__GNUC__)
+#  include <immintrin.h>   /* support for bextr (experimental) */
+#endif
+
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv06_DStream_t;
+
+typedef enum { BITv06_DStream_unfinished = 0,
+               BITv06_DStream_endOfBuffer = 1,
+               BITv06_DStream_completed = 2,
+               BITv06_DStream_overflow = 3 } BITv06_DStream_status;  /* result of BITv06_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv06_initDStream(BITv06_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv06_readBits(BITv06_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv06_DStream_status BITv06_reloadDStream(BITv06_DStream_t* bitD);
+MEM_STATIC unsigned BITv06_endOfDStream(const BITv06_DStream_t* bitD);
+
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Internal functions
+****************************************************************/
+MEM_STATIC unsigned BITv06_highbit32 ( U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*! BITv06_initDStream() :
+*   Initialize a BITv06_DStream_t.
+*   `bitD` : a pointer to an already allocated BITv06_DStream_t structure.
+*   `srcSize` must be the *exact* size of the bitStream, in bytes.
+*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv06_initDStream(BITv06_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          if (lastByte == 0) return ERROR(GENERIC);   /* endMark not present */
+          bitD->bitsConsumed = 8 - BITv06_highbit32(lastByte); }
+    } else {
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          if (lastByte == 0) return ERROR(GENERIC);   /* endMark not present */
+          bitD->bitsConsumed = 8 - BITv06_highbit32(lastByte); }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+ MEM_STATIC size_t BITv06_lookBits(const BITv06_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv06_lookBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv06_lookBitsFast(const BITv06_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv06_skipBits(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv06_readBits(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv06_lookBits(bitD, nbBits);
+    BITv06_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BITv06_readBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv06_lookBitsFast(bitD, nbBits);
+    BITv06_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv06_DStream_status BITv06_reloadDStream(BITv06_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BITv06_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv06_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv06_DStream_endOfBuffer;
+        return BITv06_DStream_completed;
+    }
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv06_DStream_status result = BITv06_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv06_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv06_endOfDStream() :
+*   @return Tells if DStream has exactly reached its end (all bits consumed).
+*/
+MEM_STATIC unsigned BITv06_endOfDStream(const BITv06_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv06_STATIC_H
+#define FSEv06_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSEv06_NCOUNTBOUND 512
+#define FSEv06_BLOCKBOUND(size) (size + (size>>7))
+#define FSEv06_COMPRESSBOUND(size) (FSEv06_NCOUNTBOUND + FSEv06_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSEv06_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+size_t FSEv06_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+/* same as FSEv06_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr  */
+
+size_t FSEv06_buildDTable_raw (FSEv06_DTable* dt, unsigned nbBits);
+/* build a fake FSEv06_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv06_buildDTable_rle (FSEv06_DTable* dt, unsigned char symbolValue);
+/* build a fake FSEv06_DTable, designed to always generate the same symbolValue */
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv06_DState_t;
+
+
+static void     FSEv06_initDState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD, const FSEv06_DTable* dt);
+
+static unsigned char FSEv06_decodeSymbol(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD);
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSEv06_decodeSymbolFast(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv06_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv06_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv06_initDState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD, const FSEv06_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv06_DTableHeader* const DTableH = (const FSEv06_DTableHeader*)ptr;
+    DStatePtr->state = BITv06_readBits(bitD, DTableH->tableLog);
+    BITv06_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv06_peekSymbol(const FSEv06_DState_t* DStatePtr)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSEv06_updateState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BITv06_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSEv06_decodeSymbol(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv06_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSEv06_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSEv06_decodeSymbolFast(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv06_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+
+
+#ifndef FSEv06_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv06_MAX_MEMORY_USAGE 14
+#define FSEv06_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv06_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv06_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv06_FUNCTION_TYPE BYTE
+#define FSEv06_FUNCTION_EXTENSION
+#define FSEv06_DECODE_TYPE FSEv06_decode_t
+
+
+#endif   /* !FSEv06_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv06_MAX_TABLELOG  (FSEv06_MAX_MEMORY_USAGE-2)
+#define FSEv06_MAX_TABLESIZE (1U<<FSEv06_MAX_TABLELOG)
+#define FSEv06_MAXTABLESIZE_MASK (FSEv06_MAX_TABLESIZE-1)
+#define FSEv06_DEFAULT_TABLELOG (FSEv06_DEFAULT_MEMORY_USAGE-2)
+#define FSEv06_MIN_TABLELOG 5
+
+#define FSEv06_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv06_MAX_TABLELOG > FSEv06_TABLELOG_ABSOLUTE_MAX
+#error "FSEv06_MAX_TABLELOG > FSEv06_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSEv06_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv06_STATIC_H */
+/*
+   Common functions of New Generation Entropy library
+   Copyright (C) 2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+*************************************************************************** */
+
+
+/*-****************************************
+*  FSE Error Management
+******************************************/
+unsigned FSEv06_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  HUF Error Management
+****************************************************************/
+static unsigned HUFv06_isError(size_t code) { return ERR_isError(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSEv06_abs(short a) { return a<0 ? -a : a; }
+
+size_t FSEv06_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv06_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv06_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {   short const max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv06_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }   /* while ((remaining>1) && (charnum<=*maxSVPtr)) */
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+/* ******************************************************************
+   FSE : Finite State Entropy decoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv06_isError ERR_isError
+#define FSEv06_STATIC_ASSERT(c) { enum { FSEv06_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSEv06_DTABLE_SIZE_U32(FSEv06_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv06_FUNCTION_EXTENSION
+#  error "FSEv06_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv06_FUNCTION_TYPE
+#  error "FSEv06_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv06_CAT(X,Y) X##Y
+#define FSEv06_FUNCTION_NAME(X,Y) FSEv06_CAT(X,Y)
+#define FSEv06_TYPE_NAME(X,Y) FSEv06_CAT(X,Y)
+
+
+/* Function templates */
+FSEv06_DTable* FSEv06_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv06_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv06_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv06_DTable*)malloc( FSEv06_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv06_freeDTable (FSEv06_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv06_buildDTable(FSEv06_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv06_DECODE_TYPE* const tableDecode = (FSEv06_DECODE_TYPE*) (tdPtr);
+    U16 symbolNext[FSEv06_MAX_SYMBOL_VALUE+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv06_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv06_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSEv06_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSEv06_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSEv06_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSEv06_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSEv06_FUNCTION_TYPE const symbol = (FSEv06_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BITv06_highbit32 ((U32)nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+
+
+#ifndef FSEv06_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv06_buildDTable_rle (FSEv06_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv06_DTableHeader* const DTableH = (FSEv06_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv06_decode_t* const cell = (FSEv06_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv06_buildDTable_raw (FSEv06_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv06_DTableHeader* const DTableH = (FSEv06_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv06_decode_t* const dinfo = (FSEv06_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSV1 = tableMask+1;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<maxSV1; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv06_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv06_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv06_DStream_t bitD;
+    FSEv06_DState_t state1;
+    FSEv06_DState_t state2;
+
+    /* Init */
+    { size_t const errorCode = BITv06_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+      if (FSEv06_isError(errorCode)) return errorCode; }
+
+    FSEv06_initDState(&state1, &bitD, dt);
+    FSEv06_initDState(&state2, &bitD, dt);
+
+#define FSEv06_GETSYMBOL(statePtr) fast ? FSEv06_decodeSymbolFast(statePtr, &bitD) : FSEv06_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv06_reloadDStream(&bitD)==BITv06_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv06_GETSYMBOL(&state1);
+
+        if (FSEv06_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv06_reloadDStream(&bitD);
+
+        op[1] = FSEv06_GETSYMBOL(&state2);
+
+        if (FSEv06_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv06_reloadDStream(&bitD) > BITv06_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv06_GETSYMBOL(&state1);
+
+        if (FSEv06_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv06_reloadDStream(&bitD);
+
+        op[3] = FSEv06_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv06_reloadDStream(&bitD) >= FSEv06_DStream_partiallyFilled; Ends at exactly BITv06_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv06_GETSYMBOL(&state1);
+
+        if (BITv06_reloadDStream(&bitD)==BITv06_DStream_overflow) {
+            *op++ = FSEv06_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv06_GETSYMBOL(&state2);
+
+        if (BITv06_reloadDStream(&bitD)==BITv06_DStream_overflow) {
+            *op++ = FSEv06_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+
+size_t FSEv06_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv06_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv06_DTableHeader* DTableH = (const FSEv06_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv06_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv06_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv06_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv06_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv06_MAX_SYMBOL_VALUE;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    {   size_t const NCountLength = FSEv06_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+        if (FSEv06_isError(NCountLength)) return NCountLength;
+        if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+        ip += NCountLength;
+        cSrcSize -= NCountLength;
+    }
+
+    { size_t const errorCode = FSEv06_buildDTable (dt, counting, maxSymbolValue, tableLog);
+      if (FSEv06_isError(errorCode)) return errorCode; }
+
+    return FSEv06_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);   /* always return, even if it is an error code */
+}
+
+
+
+#endif   /* FSEv06_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv06_H
+#define HUFv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  HUF simple functions
+******************************************/
+size_t HUFv06_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*
+HUFv06_decompress() :
+    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    `dstSize` : must be the **exact** size of original (uncompressed) data.
+    Note : in contrast with FSE, HUFv06_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUFv06_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+size_t HUFv06_compressBound(size_t size);       /**< maximum compressed size */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFv06_H */
+/* ******************************************************************
+   Huffman codec, part of New Generation Entropy library
+   header file, for static linking only
+   Copyright (C) 2013-2016, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv06_STATIC_H
+#define HUFv06_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUFv06_CTABLEBOUND 129
+#define HUFv06_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUFv06_COMPRESSBOUND(size) (HUFv06_CTABLEBOUND + HUFv06_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of HUF's DTable */
+#define HUFv06_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUFv06_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUFv06_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv06_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUFv06_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv06_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUFv06_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv06_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv06_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+
+/*!
+HUFv06_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv06_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv06_decompressSXn_usingDTable
+*/
+size_t HUFv06_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+size_t HUFv06_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+size_t HUFv06_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv06_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+/* single stream variants */
+size_t HUFv06_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv06_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv06_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv06_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUFv06_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUFv06_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv06_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUFv06_ABSOLUTEMAX_TABLELOG */
+#define HUFv06_DEFAULT_TABLELOG  HUFv06_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUFv06_MAX_SYMBOL_VALUE 255
+#if (HUFv06_MAX_TABLELOG > HUFv06_ABSOLUTEMAX_TABLELOG)
+#  error "HUFv06_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*! HUFv06_readStats() :
+    Read compact Huffman tree, saved by HUFv06_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src`
+*/
+MEM_STATIC size_t HUFv06_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* memset(huffWeight, 0, hwSize); */   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static U32 l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            {   U32 n;
+                for (n=0; n<oSize; n+=2) {
+                    huffWeight[n]   = ip[n/2] >> 4;
+                    huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv06_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv06_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv06_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] >= HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = BITv06_highbit32(weightTotal) + 1;
+        if (tableLog > HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << BITv06_highbit32(rest);
+            U32 const lastWeight = BITv06_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUFv06_STATIC_H */
+/* ******************************************************************
+   Huffman decoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUFv06_STATIC_ASSERT(c) { enum { HUFv06_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+
+/* *******************************************************
+*  HUF : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUFv06_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv06_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+size_t HUFv06_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv06_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUFv06_DEltX2* const dt = (HUFv06_DEltX2*)dtPtr;
+
+    HUFv06_STATIC_ASSERT(sizeof(HUFv06_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv06_readStats(huffWeight, HUFv06_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv06_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof allocated DTable, from used size of DTable, in case of re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<tableLog+1; n++) {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++) {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUFv06_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+
+static BYTE HUFv06_decodeSymbolX2(BITv06_DStream_t* Dstream, const HUFv06_DEltX2* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    const BYTE c = dt[val].byte;
+    BITv06_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv06_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv06_MAX_TABLELOG<=12)) \
+        HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv06_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv06_decodeStreamX2(BYTE* p, BITv06_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv06_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv06_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p < pEnd))
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+size_t HUFv06_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const U32 dtLog = DTable[0];
+    const void* dtPtr = DTable;
+    const HUFv06_DEltX2* const dt = ((const HUFv06_DEltX2*)dtPtr)+1;
+    BITv06_DStream_t bitD;
+
+    { size_t const errorCode = BITv06_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv06_isError(errorCode)) return errorCode; }
+
+    HUFv06_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv06_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv06_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX2(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const errorCode = HUFv06_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv06_decompress1X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+size_t HUFv06_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv06_DEltX2* const dt = ((const HUFv06_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv06_DStream_t bitD1;
+        BITv06_DStream_t bitD2;
+        BITv06_DStream_t bitD3;
+        BITv06_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv06_initDStream(&bitD1, istart1, length1);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD2, istart2, length2);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD3, istart3, length3);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD4, istart4, length4);
+        if (HUFv06_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv06_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv06_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv06_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv06_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv06_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv06_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv06_endOfDStream(&bitD1) & BITv06_endOfDStream(&bitD2) & BITv06_endOfDStream(&bitD3) & BITv06_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv06_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX2(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const errorCode = HUFv06_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv06_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+static void HUFv06_fillDTableX4Level2(HUFv06_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv06_DEltX4 DElt;
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    { U32 s; for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }}
+}
+
+typedef U32 rankVal_t[HUFv06_ABSOLUTEMAX_TABLELOG][HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUFv06_fillDTableX4(HUFv06_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv06_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            HUFv06_DEltX4 DElt;
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits = (BYTE)(nbBits);
+            DElt.length = 1;
+            {   U32 u;
+                const U32 end = start + length;
+                for (u = start; u < end; u++) DTable[u] = DElt;
+        }   }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv06_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv06_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUFv06_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUFv06_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUFv06_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUFv06_DEltX4* const dt = ((HUFv06_DEltX4*)dtPtr) + 1;
+
+    HUFv06_STATIC_ASSERT(sizeof(HUFv06_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    /* memset(weightList, 0, sizeof(weightList)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv06_readStats(weightList, HUFv06_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv06_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = weightList[s];
+            U32 const r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = rankVal[0];
+        {   int const rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 current = nextRankVal;
+                nextRankVal += rankStats[w] << (w+rescale);
+                rankVal0[w] = current;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < memLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUFv06_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUFv06_decodeSymbolX4(void* op, BITv06_DStream_t* DStream, const HUFv06_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv06_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv06_decodeLastSymbolX4(void* op, BITv06_DStream_t* DStream, const HUFv06_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv06_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv06_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv06_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv06_MAX_TABLELOG<=12)) \
+        ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv06_decodeStreamX4(BYTE* p, BITv06_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv06_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv06_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p <= pEnd-2))
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv06_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+size_t HUFv06_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    const BYTE* const istart = (const BYTE*) cSrc;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+
+    const U32 dtLog = DTable[0];
+    const void* const dtPtr = DTable;
+    const HUFv06_DEltX4* const dt = ((const HUFv06_DEltX4*)dtPtr) +1;
+
+    /* Init */
+    BITv06_DStream_t bitD;
+    { size_t const errorCode = BITv06_initDStream(&bitD, istart, cSrcSize);
+      if (HUFv06_isError(errorCode)) return errorCode; }
+
+    /* decode */
+    HUFv06_decodeStreamX4(ostart, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv06_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv06_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX4(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv06_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv06_decompress1X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+size_t HUFv06_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv06_DEltX4* const dt = ((const HUFv06_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv06_DStream_t bitD1;
+        BITv06_DStream_t bitD2;
+        BITv06_DStream_t bitD3;
+        BITv06_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv06_initDStream(&bitD1, istart1, length1);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD2, istart2, length2);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD3, istart3, length3);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD4, istart4, length4);
+        if (HUFv06_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv06_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv06_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv06_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv06_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv06_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv06_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv06_endOfDStream(&bitD1) & BITv06_endOfDStream(&bitD2) & BITv06_endOfDStream(&bitD3) & BITv06_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv06_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX4(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv06_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv06_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv06_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUFv06_decompress4X2, HUFv06_decompress4X4, NULL };
+    U32 Dtime[3];   /* decompression time estimation */
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    {   U32 const Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 n; for (n=0; n<3; n++)
+            Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+    }
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    {   U32 algoNb = 0;
+        if (Dtime[1] < Dtime[0]) algoNb = 1;
+        /* if (Dtime[2] < Dtime[algoNb]) algoNb = 2; */   /* current speed of HUFv06_decompress4X6 is not good */
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+    }
+
+    /* return HUFv06_decompress4X2(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams single-symbol decoding */
+    /* return HUFv06_decompress4X4(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams double-symbols decoding */
+    /* return HUFv06_decompress4X6(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams quad-symbols decoding */
+}
+/*
+    Common functions of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+/*-****************************************
+*  Version
+******************************************/
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+/*! ZSTDv06_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv06_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTDv06_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZSTDv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  ZBUFF Error Management
+****************************************************************/
+unsigned ZBUFFv06_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZBUFFv06_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv06_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv06_HEAPMODE
+#  define ZSTDv06_HEAPMODE 1
+#endif
+
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/*-*************************************
+*  Macros
+***************************************/
+#define ZSTDv06_isError ERR_isError   /* for inlining */
+#define FSEv06_isError  ERR_isError
+#define HUFv06_isError  ERR_isError
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv06_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock } ZSTDv06_dStage;
+
+struct ZSTDv06_DCtx_s
+{
+    FSEv06_DTable LLTable[FSEv06_DTABLE_SIZE_U32(LLFSELog)];
+    FSEv06_DTable OffTable[FSEv06_DTABLE_SIZE_U32(OffFSELog)];
+    FSEv06_DTable MLTable[FSEv06_DTABLE_SIZE_U32(MLFSELog)];
+    unsigned   hufTableX4[HUFv06_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTDv06_frameParams fParams;
+    blockType_t bType;   /* used in ZSTDv06_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv06_dStage stage;
+    U32 flagRepeatTable;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[ZSTDv06_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv06_FRAMEHEADERSIZE_MAX];
+};  /* typedef'd to ZSTDv06_DCtx within "zstd_static.h" */
+
+size_t ZSTDv06_sizeofDCtx (void); /* Hidden declaration */
+size_t ZSTDv06_sizeofDCtx (void) { return sizeof(ZSTDv06_DCtx); }
+
+size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx)
+{
+    dctx->expected = ZSTDv06_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTableX4[0] = ZSTD_HUFFDTABLE_CAPACITY_LOG;
+    dctx->flagRepeatTable = 0;
+    return 0;
+}
+
+ZSTDv06_DCtx* ZSTDv06_createDCtx(void)
+{
+    ZSTDv06_DCtx* dctx = (ZSTDv06_DCtx*)malloc(sizeof(ZSTDv06_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTDv06_decompressBegin(dctx);
+    return dctx;
+}
+
+size_t ZSTDv06_freeDCtx(ZSTDv06_DCtx* dctx)
+{
+    free(dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv06_copyDCtx(ZSTDv06_DCtx* dstDCtx, const ZSTDv06_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv06_DCtx) - (ZSTDv06_BLOCKSIZE_MAX+WILDCOPY_OVERLENGTH + ZSTDv06_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/*-*************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv06_MAGICNUMBER (defined within zstd_static.h)
+      - 1 byte  - Frame Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+
+/* Frame descriptor
+
+   1 byte, using :
+   bit 0-3 : windowLog - ZSTDv06_WINDOWLOG_ABSOLUTEMIN   (see zstd_internal.h)
+   bit 4   : minmatch 4(0) or 3(1)
+   bit 5   : reserved (must be zero)
+   bit 6-7 : Frame content size : unknown, 1 byte, 2 bytes, 8 bytes
+
+   Optional : content size (0, 1, 2 or 8 bytes)
+   0 : unknown
+   1 : 0-255 bytes
+   2 : 256 - 65535+256
+   8 : up to 16 exa
+*/
+
+
+/* Compressed Block, format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+/** ZSTDv06_frameHeaderSize() :
+*   srcSize must be >= ZSTDv06_frameHeaderSize_min.
+*   @return : size of the Frame Header */
+static size_t ZSTDv06_frameHeaderSize(const void* src, size_t srcSize)
+{
+    if (srcSize < ZSTDv06_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    { U32 const fcsId = (((const BYTE*)src)[4]) >> 6;
+      return ZSTDv06_frameHeaderSize_min + ZSTDv06_fcs_fieldSize[fcsId]; }
+}
+
+
+/** ZSTDv06_getFrameParams() :
+*   decode Frame Header, or provide expected `srcSize`.
+*   @return : 0, `fparamsPtr` is correctly filled,
+*            >0, `srcSize` is too small, result is expected `srcSize`,
+*             or an error code, which can be tested using ZSTDv06_isError() */
+size_t ZSTDv06_getFrameParams(ZSTDv06_frameParams* fparamsPtr, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize < ZSTDv06_frameHeaderSize_min) return ZSTDv06_frameHeaderSize_min;
+    if (MEM_readLE32(src) != ZSTDv06_MAGICNUMBER) return ERROR(prefix_unknown);
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    { size_t const fhsize = ZSTDv06_frameHeaderSize(src, srcSize);
+      if (srcSize < fhsize) return fhsize; }
+
+    memset(fparamsPtr, 0, sizeof(*fparamsPtr));
+    {   BYTE const frameDesc = ip[4];
+        fparamsPtr->windowLog = (frameDesc & 0xF) + ZSTDv06_WINDOWLOG_ABSOLUTEMIN;
+        if ((frameDesc & 0x20) != 0) return ERROR(frameParameter_unsupported);   /* reserved 1 bit */
+        switch(frameDesc >> 6)  /* fcsId */
+        {
+            default:   /* impossible */
+            case 0 : fparamsPtr->frameContentSize = 0; break;
+            case 1 : fparamsPtr->frameContentSize = ip[5]; break;
+            case 2 : fparamsPtr->frameContentSize = MEM_readLE16(ip+5)+256; break;
+            case 3 : fparamsPtr->frameContentSize = MEM_readLE64(ip+5); break;
+    }   }
+    return 0;
+}
+
+
+/** ZSTDv06_decodeFrameHeader() :
+*   `srcSize` must be the size provided by ZSTDv06_frameHeaderSize().
+*   @return : 0 if success, or an error code, which can be tested using ZSTDv06_isError() */
+static size_t ZSTDv06_decodeFrameHeader(ZSTDv06_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t const result = ZSTDv06_getFrameParams(&(zc->fParams), src, srcSize);
+    if ((MEM_32bits()) && (zc->fParams.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTDv06_getcBlockSize() :
+*   Provides the size of compressed block from block header `src` */
+static size_t ZSTDv06_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE*)src;
+    U32 cSize;
+
+    if (srcSize < ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    bpPtr->blockType = (blockType_t)((*in) >> 6);
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv06_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
+    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/*! ZSTDv06_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTDv06_decodeLiteralsBlock(ZSTDv06_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(istart[0]>> 6)
+    {
+    case IS_HUF:
+        {   size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for lhSize, + cSize (+nbSeq) */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > ZSTDv06_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv06_isError(singleStream ?
+                            HUFv06_decompress1X2(dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv06_decompress   (dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_PCH:
+        {   size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (!dctx->flagRepeatTable)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            {   size_t const errorCode = HUFv06_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTableX4);
+                if (HUFv06_isError(errorCode)) return ERROR(corruption_detected);
+            }
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_RAW:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case IS_RLE:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > ZSTDv06_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+/*! ZSTDv06_buildSeqTable() :
+    @return : nb bytes read from src,
+              or an error code if it fails, testable with ZSTDv06_isError()
+*/
+static size_t ZSTDv06_buildSeqTable(FSEv06_DTable* DTable, U32 type, U32 max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
+{
+    switch(type)
+    {
+    case FSEv06_ENCODING_RLE :
+        if (!srcSize) return ERROR(srcSize_wrong);
+        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
+        FSEv06_buildDTable_rle(DTable, *(const BYTE*)src);   /* if *src > max, data is corrupted */
+        return 1;
+    case FSEv06_ENCODING_RAW :
+        FSEv06_buildDTable(DTable, defaultNorm, max, defaultLog);
+        return 0;
+    case FSEv06_ENCODING_STATIC:
+        if (!flagRepeatTable) return ERROR(corruption_detected);
+        return 0;
+    default :   /* impossible */
+    case FSEv06_ENCODING_DYNAMIC :
+        {   U32 tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSEv06_readNCount(norm, &max, &tableLog, src, srcSize);
+            if (FSEv06_isError(headerSize)) return ERROR(corruption_detected);
+            if (tableLog > maxLog) return ERROR(corruption_detected);
+            FSEv06_buildDTable(DTable, norm, max, tableLog);
+            return headerSize;
+    }   }
+}
+
+
+static size_t ZSTDv06_decodeSeqHeaders(int* nbSeqPtr,
+                             FSEv06_DTable* DTableLL, FSEv06_DTable* DTableML, FSEv06_DTable* DTableOffb, U32 flagRepeatTable,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    {   int nbSeq = *ip++;
+        if (!nbSeq) { *nbSeqPtr=0; return 1; }
+        if (nbSeq > 0x7F) {
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
+                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
+                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
+        }
+        *nbSeqPtr = nbSeq;
+    }
+
+    /* FSE table descriptors */
+    if (ip + 4 > iend) return ERROR(srcSize_wrong); /* min : header byte + all 3 are "raw", hence no header, but at least xxLog bits per type */
+    {   U32 const LLtype  = *ip >> 6;
+        U32 const Offtype = (*ip >> 4) & 3;
+        U32 const MLtype  = (*ip >> 2) & 3;
+        ip++;
+
+        /* Build DTables */
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+        }
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableOffb, Offtype, MaxOff, OffFSELog, ip, iend-ip, OF_defaultNorm, OF_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+        }
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableML, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_defaultNorm, ML_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+    }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv06_DStream_t DStream;
+    FSEv06_DState_t stateLL;
+    FSEv06_DState_t stateOffb;
+    FSEv06_DState_t stateML;
+    size_t prevOffset[ZSTDv06_REP_INIT];
+} seqState_t;
+
+
+
+static void ZSTDv06_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    /* Literal length */
+    U32 const llCode = FSEv06_peekSymbol(&(seqState->stateLL));
+    U32 const mlCode = FSEv06_peekSymbol(&(seqState->stateML));
+    U32 const ofCode = FSEv06_peekSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+
+    U32 const llBits = LL_bits[llCode];
+    U32 const mlBits = ML_bits[mlCode];
+    U32 const ofBits = ofCode;
+    U32 const totalBits = llBits+mlBits+ofBits;
+
+    static const U32 LL_base[MaxLL+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
+                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                            0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 ML_base[MaxML+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10,   11,    12,    13,    14,    15,
+                            16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,   27,    28,    29,    30,    31,
+                            32, 34, 36, 38, 40, 44, 48, 56, 64, 80, 96, 0x80, 0x100, 0x200, 0x400, 0x800,
+                            0x1000, 0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       3,       7,     0xF,     0x1F,     0x3F,     0x7F,
+                 0xFF,   0x1FF,   0x3FF,   0x7FF,   0xFFF,   0x1FFF,   0x3FFF,   0x7FFF,
+                 0xFFFF, 0x1FFFF, 0x3FFFF, 0x7FFFF, 0xFFFFF, 0x1FFFFF, 0x3FFFFF, 0x7FFFFF,
+                 0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF, /*fake*/ 1, 1 };
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            offset = OF_base[ofCode] + BITv06_readBits(&(seqState->DStream), ofBits);   /* <=  26 bits */
+            if (MEM_32bits()) BITv06_reloadDStream(&(seqState->DStream));
+        }
+
+        if (offset < ZSTDv06_REP_NUM) {
+            if (llCode == 0 && offset <= 1) offset = 1-offset;
+
+            if (offset != 0) {
+                size_t temp = seqState->prevOffset[offset];
+                if (offset != 1) {
+                    seqState->prevOffset[2] = seqState->prevOffset[1];
+                }
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            offset -= ZSTDv06_REP_MOVE;
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq->offset = offset;
+    }
+
+    seq->matchLength = ML_base[mlCode] + MINMATCH + ((mlCode>31) ? BITv06_readBits(&(seqState->DStream), mlBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits>24)) BITv06_reloadDStream(&(seqState->DStream));
+
+    seq->litLength = LL_base[llCode] + ((llCode>15) ? BITv06_readBits(&(seqState->DStream), llBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() ||
+       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BITv06_reloadDStream(&(seqState->DStream));
+
+    /* ANS state update */
+    FSEv06_updateState(&(seqState->stateLL), &(seqState->DStream));   /* <=  9 bits */
+    FSEv06_updateState(&(seqState->stateML), &(seqState->DStream));   /* <=  9 bits */
+    if (MEM_32bits()) BITv06_reloadDStream(&(seqState->DStream));     /* <= 18 bits */
+    FSEv06_updateState(&(seqState->stateOffb), &(seqState->DStream)); /* <=  8 bits */
+}
+
+
+static size_t ZSTDv06_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTDv06_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv06_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv06_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_8) {
+            ZSTDv06_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTDv06_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv06_decompressSequences(
+                               ZSTDv06_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    FSEv06_DTable* DTableLL = dctx->LLTable;
+    FSEv06_DTable* DTableML = dctx->MLTable;
+    FSEv06_DTable* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    int nbSeq;
+
+    /* Build Decoding Tables */
+    {   size_t const seqHSize = ZSTDv06_decodeSeqHeaders(&nbSeq, DTableLL, DTableML, DTableOffb, dctx->flagRepeatTable, ip, seqSize);
+        if (ZSTDv06_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        dctx->flagRepeatTable = 0;
+    }
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = REPCODE_STARTVALUE;
+        { U32 i; for (i=0; i<ZSTDv06_REP_INIT; i++) seqState.prevOffset[i] = REPCODE_STARTVALUE; }
+        { size_t const errorCode = BITv06_initDStream(&(seqState.DStream), ip, iend-ip);
+          if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
+        FSEv06_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv06_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv06_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv06_reloadDStream(&(seqState.DStream)) <= BITv06_DStream_completed) && nbSeq ; ) {
+            nbSeq--;
+            ZSTDv06_decodeSequence(&sequence, &seqState);
+
+#if 0  /* debug */
+            static BYTE* start = NULL;
+            if (start==NULL) start = op;
+            size_t pos = (size_t)(op-start);
+            if ((pos >= 5810037) && (pos < 5810400))
+                printf("Dpos %6u :%5u literals & match %3u bytes at distance %6u \n",
+                       pos, (U32)sequence.litLength, (U32)sequence.matchLength, (U32)sequence.offset);
+#endif
+
+            {   size_t const oneSeqSize = ZSTDv06_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                if (ZSTDv06_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+        }   }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        if (litPtr > litEnd) return ERROR(corruption_detected);   /* too many literals already used */
+        if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+        if (lastLLSize > 0) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv06_checkContinuity(ZSTDv06_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv06_decompressBlock_internal(ZSTDv06_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize >= ZSTDv06_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    {   size_t const litCSize = ZSTDv06_decodeLiteralsBlock(dctx, src, srcSize);
+        if (ZSTDv06_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+    return ZSTDv06_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv06_decompressBlock(ZSTDv06_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+/*! ZSTDv06_decompressFrame() :
+*   `dctx` must be properly initialized */
+static size_t ZSTDv06_decompressFrame(ZSTDv06_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstCapacity;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties = { bt_compressed, 0 };
+
+    /* check */
+    if (srcSize < ZSTDv06_frameHeaderSize_min+ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+        if (ZSTDv06_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+        if (ZSTDv06_decodeFrameHeader(dctx, src, frameHeaderSize)) return ERROR(corruption_detected);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t decodedSize=0;
+        size_t const cBlockSize = ZSTDv06_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv06_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv06_blockHeaderSize;
+        remainingSize -= ZSTDv06_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv06_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv06_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv06_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+size_t ZSTDv06_decompress_usingPreparedDCtx(ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* refDCtx,
+                                         void* dst, size_t dstCapacity,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv06_copyDCtx(dctx, refDCtx);
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv06_decompress_usingDict(ZSTDv06_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv06_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv06_decompressDCtx(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTDv06_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+}
+
+
+size_t ZSTDv06_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv06_HEAPMODE) && (ZSTDv06_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv06_DCtx* dctx = ZSTDv06_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv06_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTDv06_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTDv06_DCtx dctx;
+    return ZSTDv06_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv06_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    blockProperties_t blockProperties = { bt_compressed, 0 };
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, srcSize);
+        if (ZSTDv06_isError(frameHeaderSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, frameHeaderSize);
+            return;
+        }
+        if (MEM_readLE32(src) != ZSTDv06_MAGICNUMBER) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+            return;
+        }
+        if (srcSize < frameHeaderSize+ZSTDv06_blockHeaderSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t const cBlockSize = ZSTDv06_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv06_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTDv06_blockHeaderSize;
+        remainingSize -= ZSTDv06_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * ZSTDv06_BLOCKSIZE_MAX;
+}
+
+/*_******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv06_nextSrcSizeToDecompress(ZSTDv06_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+size_t ZSTDv06_decompressContinue(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    if (dstCapacity) ZSTDv06_checkContinuity(dctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        if (srcSize != ZSTDv06_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        dctx->headerSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+        if (ZSTDv06_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv06_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv06_frameHeaderSize_min) {
+            dctx->expected = dctx->headerSize - ZSTDv06_frameHeaderSize_min;
+            dctx->stage = ZSTDds_decodeFrameHeader;
+            return 0;
+        }
+        dctx->expected = 0;   /* not necessary to copy more */
+	/* fall-through */
+    case ZSTDds_decodeFrameHeader:
+        {   size_t result;
+            memcpy(dctx->headerBuffer + ZSTDv06_frameHeaderSize_min, src, dctx->expected);
+            result = ZSTDv06_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv06_isError(result)) return result;
+            dctx->expected = ZSTDv06_blockHeaderSize;
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTDv06_getcBlockSize(src, ZSTDv06_blockHeaderSize, &bp);
+            if (ZSTDv06_isError(cBlockSize)) return cBlockSize;
+            if (bp.blockType == bt_end) {
+                dctx->expected = 0;
+                dctx->stage = ZSTDds_getFrameHeaderSize;
+            } else {
+                dctx->expected = cBlockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv06_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv06_copyRawBlock(dst, dstCapacity, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            dctx->expected = ZSTDv06_blockHeaderSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTDv06_refDictContent(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+static size_t ZSTDv06_loadEntropy(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t hSize, offcodeHeaderSize, matchlengthHeaderSize, litlengthHeaderSize;
+
+    hSize = HUFv06_readDTableX4(dctx->hufTableX4, dict, dictSize);
+    if (HUFv06_isError(hSize)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + hSize;
+    dictSize -= hSize;
+
+    {   short offcodeNCount[MaxOff+1];
+        U32 offcodeMaxValue=MaxOff, offcodeLog;
+        offcodeHeaderSize = FSEv06_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dict, dictSize);
+        if (FSEv06_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dict = (const char*)dict + offcodeHeaderSize;
+        dictSize -= offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        matchlengthHeaderSize = FSEv06_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dict, dictSize);
+        if (FSEv06_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dict = (const char*)dict + matchlengthHeaderSize;
+        dictSize -= matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        litlengthHeaderSize = FSEv06_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dict, dictSize);
+        if (FSEv06_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+    }
+
+    dctx->flagRepeatTable = 1;
+    return hSize + offcodeHeaderSize + matchlengthHeaderSize + litlengthHeaderSize;
+}
+
+static size_t ZSTDv06_decompress_insertDictionary(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t eSize;
+    U32 const magic = MEM_readLE32(dict);
+    if (magic != ZSTDv06_DICT_MAGIC) {
+        /* pure content mode */
+        ZSTDv06_refDictContent(dctx, dict, dictSize);
+        return 0;
+    }
+    /* load entropy tables */
+    dict = (const char*)dict + 4;
+    dictSize -= 4;
+    eSize = ZSTDv06_loadEntropy(dctx, dict, dictSize);
+    if (ZSTDv06_isError(eSize)) return ERROR(dictionary_corrupted);
+
+    /* reference dictionary content */
+    dict = (const char*)dict + eSize;
+    dictSize -= eSize;
+    ZSTDv06_refDictContent(dctx, dict, dictSize);
+
+    return 0;
+}
+
+
+size_t ZSTDv06_decompressBegin_usingDict(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    { size_t const errorCode = ZSTDv06_decompressBegin(dctx);
+      if (ZSTDv06_isError(errorCode)) return errorCode; }
+
+    if (dict && dictSize) {
+        size_t const errorCode = ZSTDv06_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv06_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv06_DCtx object is required to track streaming operations.
+*  Use ZBUFFv06_createDCtx() and ZBUFFv06_freeDCtx() to create/release resources.
+*  Use ZBUFFv06_decompressInit() to start a new decompression operation,
+*   or ZBUFFv06_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv06_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv06_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv06_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv06_recommendedDInSize() and ZBUFFv06_recommendedDOutSize()
+*  output : ZBUFFv06_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv06_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv06_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_loadHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFFv06_dStage;
+
+/* *** Resource management *** */
+struct ZBUFFv06_DCtx_s {
+    ZSTDv06_DCtx* zd;
+    ZSTDv06_frameParams fParams;
+    ZBUFFv06_dStage stage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t blockSize;
+    BYTE headerBuffer[ZSTDv06_FRAMEHEADERSIZE_MAX];
+    size_t lhSize;
+};   /* typedef'd to ZBUFFv06_DCtx within "zstd_buffered.h" */
+
+
+ZBUFFv06_DCtx* ZBUFFv06_createDCtx(void)
+{
+    ZBUFFv06_DCtx* zbd = (ZBUFFv06_DCtx*)malloc(sizeof(ZBUFFv06_DCtx));
+    if (zbd==NULL) return NULL;
+    memset(zbd, 0, sizeof(*zbd));
+    zbd->zd = ZSTDv06_createDCtx();
+    zbd->stage = ZBUFFds_init;
+    return zbd;
+}
+
+size_t ZBUFFv06_freeDCtx(ZBUFFv06_DCtx* zbd)
+{
+    if (zbd==NULL) return 0;   /* support free on null */
+    ZSTDv06_freeDCtx(zbd->zd);
+    free(zbd->inBuff);
+    free(zbd->outBuff);
+    free(zbd);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv06_decompressInitDictionary(ZBUFFv06_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    zbd->stage = ZBUFFds_loadHeader;
+    zbd->lhSize = zbd->inPos = zbd->outStart = zbd->outEnd = 0;
+    return ZSTDv06_decompressBegin_usingDict(zbd->zd, dict, dictSize);
+}
+
+size_t ZBUFFv06_decompressInit(ZBUFFv06_DCtx* zbd)
+{
+    return ZBUFFv06_decompressInitDictionary(zbd, NULL, 0);
+}
+
+
+
+MEM_STATIC size_t ZBUFFv06_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t length = MIN(dstCapacity, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* const iend = istart + *srcSizePtr;
+    const char* ip = istart;
+    char* const ostart = (char*)dst;
+    char* const oend = ostart + *dstCapacityPtr;
+    char* op = ostart;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbd->stage)
+        {
+        case ZBUFFds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFds_loadHeader :
+            {   size_t const hSize = ZSTDv06_getFrameParams(&(zbd->fParams), zbd->headerBuffer, zbd->lhSize);
+                if (hSize != 0) {
+                    size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
+                    if (ZSTDv06_isError(hSize)) return hSize;
+                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
+                        if (ip != NULL)
+                            memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        zbd->lhSize += iend-ip;
+                        *dstCapacityPtr = 0;
+                        return (hSize - zbd->lhSize) + ZSTDv06_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    memcpy(zbd->headerBuffer + zbd->lhSize, ip, toLoad); zbd->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* Consume header */
+            {   size_t const h1Size = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);  /* == ZSTDv06_frameHeaderSize_min */
+                size_t const h1Result = ZSTDv06_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer, h1Size);
+                if (ZSTDv06_isError(h1Result)) return h1Result;
+                if (h1Size < zbd->lhSize) {   /* long header */
+                    size_t const h2Size = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                    size_t const h2Result = ZSTDv06_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer+h1Size, h2Size);
+                    if (ZSTDv06_isError(h2Result)) return h2Result;
+            }   }
+
+            /* Frame header instruct buffer sizes */
+            {   size_t const blockSize = MIN(1 << zbd->fParams.windowLog, ZSTDv06_BLOCKSIZE_MAX);
+                zbd->blockSize = blockSize;
+                if (zbd->inBuffSize < blockSize) {
+                    free(zbd->inBuff);
+                    zbd->inBuffSize = blockSize;
+                    zbd->inBuff = (char*)malloc(blockSize);
+                    if (zbd->inBuff == NULL) return ERROR(memory_allocation);
+                }
+                {   size_t const neededOutSize = ((size_t)1 << zbd->fParams.windowLog) + blockSize + WILDCOPY_OVERLENGTH * 2;
+                    if (zbd->outBuffSize < neededOutSize) {
+                        free(zbd->outBuff);
+                        zbd->outBuffSize = neededOutSize;
+                        zbd->outBuff = (char*)malloc(neededOutSize);
+                        if (zbd->outBuff == NULL) return ERROR(memory_allocation);
+            }   }   }
+            zbd->stage = ZBUFFds_read;
+	    /* fall-through */
+        case ZBUFFds_read:
+            {   size_t const neededInSize = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                if (neededInSize==0) {  /* end of frame */
+                    zbd->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    size_t const decodedSize = ZSTDv06_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        ip, neededInSize);
+                    if (ZSTDv06_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbd->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {   size_t const neededInSize = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                size_t const toLoad = neededInSize - zbd->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbd->inBuffSize - zbd->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv06_limitCopy(zbd->inBuff + zbd->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbd->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                {   size_t const decodedSize = ZSTDv06_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        zbd->inBuff, neededInSize);
+                    if (ZSTDv06_isError(decodedSize)) return decodedSize;
+                    zbd->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbd->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    /* break; */ /* ZBUFFds_flush follows */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {   size_t const toFlushSize = zbd->outEnd - zbd->outStart;
+                size_t const flushedSize = ZBUFFv06_limitCopy(op, oend-op, zbd->outBuff + zbd->outStart, toFlushSize);
+                op += flushedSize;
+                zbd->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbd->stage = ZBUFFds_read;
+                    if (zbd->outStart + zbd->blockSize > zbd->outBuffSize)
+                        zbd->outStart = zbd->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    /* result */
+    *srcSizePtr = ip-istart;
+    *dstCapacityPtr = op-ostart;
+    {   size_t nextSrcSizeHint = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+        if (nextSrcSizeHint > ZSTDv06_blockHeaderSize) nextSrcSizeHint+= ZSTDv06_blockHeaderSize;   /* get following block header too */
+        nextSrcSizeHint -= zbd->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFFv06_recommendedDInSize(void)  { return ZSTDv06_BLOCKSIZE_MAX + ZSTDv06_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv06_recommendedDOutSize(void) { return ZSTDv06_BLOCKSIZE_MAX; }
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v06.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v06.h
new file mode 100644
index 0000000..6338910
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v06.h
@@ -0,0 +1,172 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTDv06_H
+#define ZSTDv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*======  Dependency  ======*/
+#include <stddef.h>   /* size_t */
+
+
+/*======  Export for Windows  ======*/
+/*!
+*  ZSTDv06_DLL_EXPORT :
+*  Enable exporting of functions when building a Windows DLL
+*/
+#if defined(_WIN32) && defined(ZSTDv06_DLL_EXPORT) && (ZSTDv06_DLL_EXPORT==1)
+#  define ZSTDLIBv06_API __declspec(dllexport)
+#else
+#  define ZSTDLIBv06_API
+#endif
+
+
+/* *************************************
+*  Simple functions
+***************************************/
+/*! ZSTDv06_decompress() :
+    `compressedSize` : is the _exact_ size of the compressed blob, otherwise decompression will fail.
+    `dstCapacity` must be large enough, equal or larger than originalSize.
+    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+              or an errorCode if it fails (which can be tested using ZSTDv06_isError()) */
+ZSTDLIBv06_API size_t ZSTDv06_decompress( void* dst, size_t dstCapacity,
+                                    const void* src, size_t compressedSize);
+
+/**
+ZSTDv06_findFrameSizeInfoLegacy() : get the source length and decompressed bound of a ZSTD frame compliant with v0.6.x format
+    srcSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    cSize (output parameter)  : the number of bytes that would be read to decompress this frame
+                                or an error code if it fails (which can be tested using ZSTDv01_isError())
+    dBound (output parameter) : an upper-bound for the decompressed size of the data in the frame
+                                or ZSTD_CONTENTSIZE_ERROR if an error occurs
+
+    note : assumes `cSize` and `dBound` are _not_ NULL.
+*/
+void ZSTDv06_findFrameSizeInfoLegacy(const void *src, size_t srcSize,
+                                     size_t* cSize, unsigned long long* dBound);
+
+/* *************************************
+*  Helper functions
+***************************************/
+ZSTDLIBv06_API size_t      ZSTDv06_compressBound(size_t srcSize); /*!< maximum compressed size (worst case scenario) */
+
+/* Error Management */
+ZSTDLIBv06_API unsigned    ZSTDv06_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+ZSTDLIBv06_API const char* ZSTDv06_getErrorName(size_t code);     /*!< provides readable string for an error code */
+
+
+/* *************************************
+*  Explicit memory management
+***************************************/
+/** Decompression context */
+typedef struct ZSTDv06_DCtx_s ZSTDv06_DCtx;
+ZSTDLIBv06_API ZSTDv06_DCtx* ZSTDv06_createDCtx(void);
+ZSTDLIBv06_API size_t     ZSTDv06_freeDCtx(ZSTDv06_DCtx* dctx);      /*!< @return : errorCode */
+
+/** ZSTDv06_decompressDCtx() :
+*   Same as ZSTDv06_decompress(), but requires an already allocated ZSTDv06_DCtx (see ZSTDv06_createDCtx()) */
+ZSTDLIBv06_API size_t ZSTDv06_decompressDCtx(ZSTDv06_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-***********************
+*  Dictionary API
+*************************/
+/*! ZSTDv06_decompress_usingDict() :
+*   Decompression using a pre-defined Dictionary content (see dictBuilder).
+*   Dictionary must be identical to the one used during compression, otherwise regenerated data will be corrupted.
+*   Note : dict can be NULL, in which case, it's equivalent to ZSTDv06_decompressDCtx() */
+ZSTDLIBv06_API size_t ZSTDv06_decompress_usingDict(ZSTDv06_DCtx* dctx,
+                                                   void* dst, size_t dstCapacity,
+                                             const void* src, size_t srcSize,
+                                             const void* dict,size_t dictSize);
+
+
+/*-************************
+*  Advanced Streaming API
+***************************/
+struct ZSTDv06_frameParams_s { unsigned long long frameContentSize; unsigned windowLog; };
+typedef struct ZSTDv06_frameParams_s ZSTDv06_frameParams;
+
+ZSTDLIBv06_API size_t ZSTDv06_getFrameParams(ZSTDv06_frameParams* fparamsPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+ZSTDLIBv06_API size_t ZSTDv06_decompressBegin_usingDict(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIBv06_API void   ZSTDv06_copyDCtx(ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* preparedDCtx);
+
+ZSTDLIBv06_API size_t ZSTDv06_nextSrcSizeToDecompress(ZSTDv06_DCtx* dctx);
+ZSTDLIBv06_API size_t ZSTDv06_decompressContinue(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+/* *************************************
+*  ZBUFF API
+***************************************/
+
+typedef struct ZBUFFv06_DCtx_s ZBUFFv06_DCtx;
+ZSTDLIBv06_API ZBUFFv06_DCtx* ZBUFFv06_createDCtx(void);
+ZSTDLIBv06_API size_t         ZBUFFv06_freeDCtx(ZBUFFv06_DCtx* dctx);
+
+ZSTDLIBv06_API size_t ZBUFFv06_decompressInit(ZBUFFv06_DCtx* dctx);
+ZSTDLIBv06_API size_t ZBUFFv06_decompressInitDictionary(ZBUFFv06_DCtx* dctx, const void* dict, size_t dictSize);
+
+ZSTDLIBv06_API size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* dctx,
+                                                  void* dst, size_t* dstCapacityPtr,
+                                            const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv06_DCtx object is required to track streaming operations.
+*  Use ZBUFFv06_createDCtx() and ZBUFFv06_freeDCtx() to create/release resources.
+*  Use ZBUFFv06_decompressInit() to start a new decompression operation,
+*   or ZBUFFv06_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv06_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv06_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change `dst`.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv06_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv06_recommendedDInSize() and ZBUFFv06_recommendedDOutSize()
+*  output : ZBUFFv06_recommendedDOutSize== 128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv06_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv06_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+ZSTDLIBv06_API unsigned ZBUFFv06_isError(size_t errorCode);
+ZSTDLIBv06_API const char* ZBUFFv06_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, they tend to offer better latency */
+ZSTDLIBv06_API size_t ZBUFFv06_recommendedDInSize(void);
+ZSTDLIBv06_API size_t ZBUFFv06_recommendedDOutSize(void);
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define ZSTDv06_MAGICNUMBER 0xFD2FB526   /* v0.6 */
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv06_BUFFERED_H */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v07.c b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v07.c
new file mode 100644
index 0000000..15dc3ef
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v07.c
@@ -0,0 +1,4498 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include <stddef.h>     /* size_t, ptrdiff_t */
+#include <string.h>     /* memcpy */
+#include <stdlib.h>     /* malloc, free, qsort */
+
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY    /* XXH64_state_t */
+#endif
+#include "../common/xxhash.h"                  /* XXH64_* */
+#include "zstd_v07.h"
+
+#define FSEv07_STATIC_LINKING_ONLY   /* FSEv07_MIN_TABLELOG */
+#define HUFv07_STATIC_LINKING_ONLY   /* HUFv07_TABLELOG_ABSOLUTEMAX */
+#define ZSTDv07_STATIC_LINKING_ONLY
+
+#include "../common/error_private.h"
+
+
+#ifdef ZSTDv07_STATIC_LINKING_ONLY
+
+/* ====================================================================================
+ * The definitions in this section are considered experimental.
+ * They should never be used with a dynamic library, as they may change in the future.
+ * They are provided for advanced usages.
+ * Use them only in association with static linking.
+ * ==================================================================================== */
+
+/*--- Constants ---*/
+#define ZSTDv07_MAGIC_SKIPPABLE_START  0x184D2A50U
+
+#define ZSTDv07_WINDOWLOG_MAX_32  25
+#define ZSTDv07_WINDOWLOG_MAX_64  27
+#define ZSTDv07_WINDOWLOG_MAX    ((U32)(MEM_32bits() ? ZSTDv07_WINDOWLOG_MAX_32 : ZSTDv07_WINDOWLOG_MAX_64))
+#define ZSTDv07_WINDOWLOG_MIN     18
+#define ZSTDv07_CHAINLOG_MAX     (ZSTDv07_WINDOWLOG_MAX+1)
+#define ZSTDv07_CHAINLOG_MIN       4
+#define ZSTDv07_HASHLOG_MAX       ZSTDv07_WINDOWLOG_MAX
+#define ZSTDv07_HASHLOG_MIN       12
+#define ZSTDv07_HASHLOG3_MAX      17
+#define ZSTDv07_SEARCHLOG_MAX    (ZSTDv07_WINDOWLOG_MAX-1)
+#define ZSTDv07_SEARCHLOG_MIN      1
+#define ZSTDv07_SEARCHLENGTH_MAX   7
+#define ZSTDv07_SEARCHLENGTH_MIN   3
+#define ZSTDv07_TARGETLENGTH_MIN   4
+#define ZSTDv07_TARGETLENGTH_MAX 999
+
+#define ZSTDv07_FRAMEHEADERSIZE_MAX 18    /* for static allocation */
+static const size_t ZSTDv07_frameHeaderSize_min = 5;
+static const size_t ZSTDv07_frameHeaderSize_max = ZSTDv07_FRAMEHEADERSIZE_MAX;
+static const size_t ZSTDv07_skippableHeaderSize = 8;  /* magic number + skippable frame length */
+
+
+/* custom memory allocation functions */
+typedef void* (*ZSTDv07_allocFunction) (void* opaque, size_t size);
+typedef void  (*ZSTDv07_freeFunction) (void* opaque, void* address);
+typedef struct { ZSTDv07_allocFunction customAlloc; ZSTDv07_freeFunction customFree; void* opaque; } ZSTDv07_customMem;
+
+
+/*--- Advanced Decompression functions ---*/
+
+/*! ZSTDv07_estimateDCtxSize() :
+ *  Gives the potential amount of memory allocated to create a ZSTDv07_DCtx */
+ZSTDLIBv07_API size_t ZSTDv07_estimateDCtxSize(void);
+
+/*! ZSTDv07_createDCtx_advanced() :
+ *  Create a ZSTD decompression context using external alloc and free functions */
+ZSTDLIBv07_API ZSTDv07_DCtx* ZSTDv07_createDCtx_advanced(ZSTDv07_customMem customMem);
+
+/*! ZSTDv07_sizeofDCtx() :
+ *  Gives the amount of memory used by a given ZSTDv07_DCtx */
+ZSTDLIBv07_API size_t ZSTDv07_sizeofDCtx(const ZSTDv07_DCtx* dctx);
+
+
+/* ******************************************************************
+*  Buffer-less streaming functions (synchronous mode)
+********************************************************************/
+
+ZSTDLIBv07_API size_t ZSTDv07_decompressBegin(ZSTDv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZSTDv07_decompressBegin_usingDict(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIBv07_API void   ZSTDv07_copyDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* preparedDCtx);
+
+ZSTDLIBv07_API size_t ZSTDv07_nextSrcSizeToDecompress(ZSTDv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZSTDv07_decompressContinue(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/*
+  Buffer-less streaming decompression (synchronous mode)
+
+  A ZSTDv07_DCtx object is required to track streaming operations.
+  Use ZSTDv07_createDCtx() / ZSTDv07_freeDCtx() to manage it.
+  A ZSTDv07_DCtx object can be re-used multiple times.
+
+  First optional operation is to retrieve frame parameters, using ZSTDv07_getFrameParams(), which doesn't consume the input.
+  It can provide the minimum size of rolling buffer required to properly decompress data (`windowSize`),
+  and optionally the final size of uncompressed content.
+  (Note : content size is an optional info that may not be present. 0 means : content size unknown)
+  Frame parameters are extracted from the beginning of compressed frame.
+  The amount of data to read is variable, from ZSTDv07_frameHeaderSize_min to ZSTDv07_frameHeaderSize_max (so if `srcSize` >= ZSTDv07_frameHeaderSize_max, it will always work)
+  If `srcSize` is too small for operation to succeed, function will return the minimum size it requires to produce a result.
+  Result : 0 when successful, it means the ZSTDv07_frameParams structure has been filled.
+          >0 : means there is not enough data into `src`. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv07_isError()
+
+  Start decompression, with ZSTDv07_decompressBegin() or ZSTDv07_decompressBegin_usingDict().
+  Alternatively, you can copy a prepared context, using ZSTDv07_copyDCtx().
+
+  Then use ZSTDv07_nextSrcSizeToDecompress() and ZSTDv07_decompressContinue() alternatively.
+  ZSTDv07_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv07_decompressContinue().
+  ZSTDv07_decompressContinue() requires this exact amount of bytes, or it will fail.
+
+  @result of ZSTDv07_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  It can be zero, which is not an error; it just means ZSTDv07_decompressContinue() has decoded some header.
+
+  ZSTDv07_decompressContinue() needs previous data blocks during decompression, up to `windowSize`.
+  They should preferably be located contiguously, prior to current block.
+  Alternatively, a round buffer of sufficient size is also possible. Sufficient size is determined by frame parameters.
+  ZSTDv07_decompressContinue() is very sensitive to contiguity,
+  if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
+    or that previous contiguous segment is large enough to properly handle maximum back-reference.
+
+  A frame is fully decoded when ZSTDv07_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+
+
+  == Special case : skippable frames ==
+
+  Skippable frames allow the integration of user-defined data into a flow of concatenated frames.
+  Skippable frames will be ignored (skipped) by a decompressor. The format of skippable frame is following:
+  a) Skippable frame ID - 4 Bytes, Little endian format, any value from 0x184D2A50 to 0x184D2A5F
+  b) Frame Size - 4 Bytes, Little endian format, unsigned 32-bits
+  c) Frame Content - any content (User Data) of length equal to Frame Size
+  For skippable frames ZSTDv07_decompressContinue() always returns 0.
+  For skippable frames ZSTDv07_getFrameParams() returns fparamsPtr->windowLog==0 what means that a frame is skippable.
+  It also returns Frame Size as fparamsPtr->frameContentSize.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    Frame metadata cost is typically ~18 bytes, which can be non-negligible for very small blocks (< 100 bytes).
+    User will have to take in charge required information to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Compressing and decompressing require a context structure
+      + Use ZSTDv07_createCCtx() and ZSTDv07_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv07_compressBegin()
+      + decompression : ZSTDv07_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - Block size is limited, it must be <= ZSTDv07_getBlockSizeMax()
+      + If you need to compress more, cut data into multiple blocks
+      + Consider using the regular ZSTDv07_compress() instead, as frame metadata costs become negligible when source size is large.
+    - When a block is considered not compressible enough, ZSTDv07_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv07_decompressBlock() doesn't accept uncompressed data as input !!!
+      + In case of multiple successive blocks, decoder must be informed of uncompressed block existence to follow proper history.
+        Use ZSTDv07_insertBlock() in such a case.
+*/
+
+#define ZSTDv07_BLOCKSIZE_ABSOLUTEMAX (128 * 1024)   /* define, for static allocation */
+ZSTDLIBv07_API size_t ZSTDv07_decompressBlock(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert block into `dctx` history. Useful for uncompressed blocks */
+
+
+#endif   /* ZSTDv07_STATIC_LINKING_ONLY */
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap64(in);
+#else
+    return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+#endif
+}
+
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+/* ******************************************************************
+   bitstream
+   Part of FSE library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#if defined(__BMI__) && defined(__GNUC__)
+#  include <immintrin.h>   /* support for bextr (experimental) */
+#endif
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv07_DStream_t;
+
+typedef enum { BITv07_DStream_unfinished = 0,
+               BITv07_DStream_endOfBuffer = 1,
+               BITv07_DStream_completed = 2,
+               BITv07_DStream_overflow = 3 } BITv07_DStream_status;  /* result of BITv07_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv07_initDStream(BITv07_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv07_readBits(BITv07_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv07_DStream_status BITv07_reloadDStream(BITv07_DStream_t* bitD);
+MEM_STATIC unsigned BITv07_endOfDStream(const BITv07_DStream_t* bitD);
+
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Internal functions
+****************************************************************/
+MEM_STATIC unsigned BITv07_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*! BITv07_initDStream() :
+*   Initialize a BITv07_DStream_t.
+*   `bitD` : a pointer to an already allocated BITv07_DStream_t structure.
+*   `srcSize` must be the *exact* size of the bitStream, in bytes.
+*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv07_initDStream(BITv07_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - BITv07_highbit32(lastByte) : 0;
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+    } else {
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - BITv07_highbit32(lastByte) : 0;
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+ MEM_STATIC size_t BITv07_lookBits(const BITv07_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv07_lookBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv07_lookBitsFast(const BITv07_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv07_skipBits(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv07_readBits(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv07_lookBits(bitD, nbBits);
+    BITv07_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BITv07_readBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv07_lookBitsFast(bitD, nbBits);
+    BITv07_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv07_DStream_status BITv07_reloadDStream(BITv07_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should not happen => corruption detected */
+        return BITv07_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv07_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv07_DStream_endOfBuffer;
+        return BITv07_DStream_completed;
+    }
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv07_DStream_status result = BITv07_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv07_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv07_endOfDStream() :
+*   @return Tells if DStream has exactly reached its end (all bits consumed).
+*/
+MEM_STATIC unsigned BITv07_endOfDStream(const BITv07_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy codec
+   Public Prototypes declaration
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef FSEv07_H
+#define FSEv07_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-****************************************
+*  FSE simple functions
+******************************************/
+
+/*! FSEv07_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstCapacity'.
+    @return : size of regenerated data (<= maxDstSize),
+              or an error code, which can be tested using FSEv07_isError() .
+
+    ** Important ** : FSEv07_decompress() does not decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+size_t FSEv07_decompress(void* dst,  size_t dstCapacity,
+                const void* cSrc, size_t cSrcSize);
+
+
+/* Error Management */
+unsigned    FSEv07_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv07_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSEv07_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSEv07_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSEv07_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv07_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*! Constructor and Destructor of FSEv07_DTable.
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv07_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv07_DTable* FSEv07_createDTable(unsigned tableLog);
+void        FSEv07_freeDTable(FSEv07_DTable* dt);
+
+/*! FSEv07_buildDTable():
+    Builds 'dt', which must be already allocated, using FSEv07_createDTable().
+    return : 0, or an errorCode, which can be tested using FSEv07_isError() */
+size_t FSEv07_buildDTable (FSEv07_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSEv07_decompress_usingDTable():
+    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
+    into `dst` which must be already allocated.
+    @return : size of regenerated data (necessarily <= `dstCapacity`),
+              or an errorCode, which can be tested using FSEv07_isError() */
+size_t FSEv07_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv07_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSEv07_readNCount() if it was saved using FSEv07_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSEv07_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSEv07_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError().
+
+The next step is to build the decompression tables 'FSEv07_DTable' from 'normalizedCounter'.
+This is performed by the function FSEv07_buildDTable().
+The space required by 'FSEv07_DTable' must be already allocated using FSEv07_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError().
+
+`FSEv07_DTable` can then be used to decompress `cSrc`, with FSEv07_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSEv07_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError(). (ex: dst buffer too small)
+*/
+
+
+#ifdef FSEv07_STATIC_LINKING_ONLY
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSEv07_NCOUNTBOUND 512
+#define FSEv07_BLOCKBOUND(size) (size + (size>>7))
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSEv07_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+size_t FSEv07_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+/**< same as FSEv07_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr  */
+
+unsigned FSEv07_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
+/**< same as FSEv07_optimalTableLog(), which used `minus==2` */
+
+size_t FSEv07_buildDTable_raw (FSEv07_DTable* dt, unsigned nbBits);
+/**< build a fake FSEv07_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv07_buildDTable_rle (FSEv07_DTable* dt, unsigned char symbolValue);
+/**< build a fake FSEv07_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv07_DState_t;
+
+
+static void     FSEv07_initDState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD, const FSEv07_DTable* dt);
+
+static unsigned char FSEv07_decodeSymbol(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD);
+
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSEv07_decodeSymbolFast(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv07_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv07_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv07_initDState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD, const FSEv07_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv07_DTableHeader* const DTableH = (const FSEv07_DTableHeader*)ptr;
+    DStatePtr->state = BITv07_readBits(bitD, DTableH->tableLog);
+    BITv07_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv07_peekSymbol(const FSEv07_DState_t* DStatePtr)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSEv07_updateState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BITv07_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSEv07_decodeSymbol(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv07_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSEv07_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSEv07_decodeSymbolFast(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv07_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+
+
+#ifndef FSEv07_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv07_MAX_MEMORY_USAGE 14
+#define FSEv07_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv07_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv07_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv07_FUNCTION_TYPE BYTE
+#define FSEv07_FUNCTION_EXTENSION
+#define FSEv07_DECODE_TYPE FSEv07_decode_t
+
+
+#endif   /* !FSEv07_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv07_MAX_TABLELOG  (FSEv07_MAX_MEMORY_USAGE-2)
+#define FSEv07_MAX_TABLESIZE (1U<<FSEv07_MAX_TABLELOG)
+#define FSEv07_MAXTABLESIZE_MASK (FSEv07_MAX_TABLESIZE-1)
+#define FSEv07_DEFAULT_TABLELOG (FSEv07_DEFAULT_MEMORY_USAGE-2)
+#define FSEv07_MIN_TABLELOG 5
+
+#define FSEv07_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv07_MAX_TABLELOG > FSEv07_TABLELOG_ABSOLUTE_MAX
+#  error "FSEv07_MAX_TABLELOG > FSEv07_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSEv07_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+
+
+#endif /* FSEv07_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv07_H */
+/* ******************************************************************
+   Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv07_H_298734234
+#define HUFv07_H_298734234
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* *** simple functions *** */
+/**
+HUFv07_decompress() :
+    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated buffer 'dst', of minimum size 'dstSize'.
+    `dstSize` : **must** be the ***exact*** size of original (uncompressed) data.
+    Note : in contrast with FSE, HUFv07_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize),
+              or an error code, which can be tested using HUFv07_isError()
+*/
+size_t HUFv07_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+#define HUFv07_BLOCKSIZE_MAX (128 * 1024)
+
+/* Error Management */
+unsigned    HUFv07_isError(size_t code);        /**< tells if a return value is an error code */
+const char* HUFv07_getErrorName(size_t code);   /**< provides error code string (useful for debugging) */
+
+
+/* *** Advanced function *** */
+
+
+#ifdef HUFv07_STATIC_LINKING_ONLY
+
+
+/* *** Constants *** */
+#define HUFv07_TABLELOG_ABSOLUTEMAX  16   /* absolute limit of HUFv07_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv07_TABLELOG_MAX  12           /* max configured tableLog (for static allocation); can be modified up to HUFv07_ABSOLUTEMAX_TABLELOG */
+#define HUFv07_TABLELOG_DEFAULT  11       /* tableLog by default, when not specified */
+#define HUFv07_SYMBOLVALUE_MAX 255
+#if (HUFv07_TABLELOG_MAX > HUFv07_TABLELOG_ABSOLUTEMAX)
+#  error "HUFv07_TABLELOG_MAX is too large !"
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUFv07_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+
+/* static allocation of HUF's DTable */
+typedef U32 HUFv07_DTable;
+#define HUFv07_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
+#define HUFv07_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        HUFv07_DTable DTable[HUFv07_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1)*0x1000001) }
+#define HUFv07_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        HUFv07_DTable DTable[HUFv07_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog)*0x1000001) }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv07_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUFv07_decompress4X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
+size_t HUFv07_decompress4X_hufOnly(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
+size_t HUFv07_decompress4X2_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress4X4_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+size_t HUFv07_decompress1X2_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress1X4_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+
+/* ****************************************
+*  HUF detailed API
+******************************************/
+/*!
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and regenerate 'CTable' using external methods.
+*/
+/* FSEv07_count() : find it within "fse.h" */
+
+/*! HUFv07_readStats() :
+    Read compact Huffman tree, saved by HUFv07_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUFv07_readCTable() and HUFv07_readDTableXn() . */
+size_t HUFv07_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize);
+
+
+/*
+HUFv07_decompress() does the following:
+1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv07_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv07_decompressSXn_usingDTable
+*/
+
+/** HUFv07_selectDecoder() :
+*   Tells which decoder is likely to decode faster,
+*   based on a set of pre-determined metrics.
+*   @return : 0==HUFv07_decompress4X2, 1==HUFv07_decompress4X4 .
+*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+U32 HUFv07_selectDecoder (size_t dstSize, size_t cSrcSize);
+
+size_t HUFv07_readDTableX2 (HUFv07_DTable* DTable, const void* src, size_t srcSize);
+size_t HUFv07_readDTableX4 (HUFv07_DTable* DTable, const void* src, size_t srcSize);
+
+size_t HUFv07_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+
+
+/* single stream variants */
+size_t HUFv07_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv07_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv07_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+
+
+#endif /* HUFv07_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFv07_H_298734234 */
+/*
+   Common functions of New Generation Entropy library
+   Copyright (C) 2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+*************************************************************************** */
+
+
+
+/*-****************************************
+*  FSE Error Management
+******************************************/
+unsigned FSEv07_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  HUF Error Management
+****************************************************************/
+unsigned HUFv07_isError(size_t code) { return ERR_isError(code); }
+
+const char* HUFv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSEv07_abs(short a) { return (short)(a<0 ? -a : a); }
+
+size_t FSEv07_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv07_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv07_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {   short const max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv07_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }   /* while ((remaining>1) && (charnum<=*maxSVPtr)) */
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*! HUFv07_readStats() :
+    Read compact Huffman tree, saved by HUFv07_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUFv07_readCTable() and HUFv07_readDTableXn() .
+*/
+size_t HUFv07_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* memset(huffWeight, 0, hwSize); */   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static U32 l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            {   U32 n;
+                for (n=0; n<oSize; n+=2) {
+                    huffWeight[n]   = ip[n/2] >> 4;
+                    huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv07_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv07_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv07_TABLELOG_ABSOLUTEMAX + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] >= HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = BITv07_highbit32(weightTotal) + 1;
+        if (tableLog > HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << BITv07_highbit32(rest);
+            U32 const lastWeight = BITv07_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
+/* ******************************************************************
+   FSE : Finite State Entropy decoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv07_isError ERR_isError
+#define FSEv07_STATIC_ASSERT(c) { enum { FSEv07_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSEv07_DTABLE_SIZE_U32(FSEv07_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv07_FUNCTION_EXTENSION
+#  error "FSEv07_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv07_FUNCTION_TYPE
+#  error "FSEv07_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv07_CAT(X,Y) X##Y
+#define FSEv07_FUNCTION_NAME(X,Y) FSEv07_CAT(X,Y)
+#define FSEv07_TYPE_NAME(X,Y) FSEv07_CAT(X,Y)
+
+
+/* Function templates */
+FSEv07_DTable* FSEv07_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv07_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv07_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv07_DTable*)malloc( FSEv07_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv07_freeDTable (FSEv07_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv07_buildDTable(FSEv07_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv07_DECODE_TYPE* const tableDecode = (FSEv07_DECODE_TYPE*) (tdPtr);
+    U16 symbolNext[FSEv07_MAX_SYMBOL_VALUE+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv07_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv07_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSEv07_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSEv07_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSEv07_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSEv07_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSEv07_FUNCTION_TYPE const symbol = (FSEv07_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BITv07_highbit32 ((U32)nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+
+
+#ifndef FSEv07_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv07_buildDTable_rle (FSEv07_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv07_DTableHeader* const DTableH = (FSEv07_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv07_decode_t* const cell = (FSEv07_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv07_buildDTable_raw (FSEv07_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv07_DTableHeader* const DTableH = (FSEv07_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv07_decode_t* const dinfo = (FSEv07_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSV1 = tableMask+1;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<maxSV1; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv07_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv07_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv07_DStream_t bitD;
+    FSEv07_DState_t state1;
+    FSEv07_DState_t state2;
+
+    /* Init */
+    { size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+      if (FSEv07_isError(errorCode)) return errorCode; }
+
+    FSEv07_initDState(&state1, &bitD, dt);
+    FSEv07_initDState(&state2, &bitD, dt);
+
+#define FSEv07_GETSYMBOL(statePtr) fast ? FSEv07_decodeSymbolFast(statePtr, &bitD) : FSEv07_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv07_reloadDStream(&bitD)==BITv07_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv07_GETSYMBOL(&state1);
+
+        if (FSEv07_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv07_reloadDStream(&bitD);
+
+        op[1] = FSEv07_GETSYMBOL(&state2);
+
+        if (FSEv07_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv07_reloadDStream(&bitD) > BITv07_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv07_GETSYMBOL(&state1);
+
+        if (FSEv07_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv07_reloadDStream(&bitD);
+
+        op[3] = FSEv07_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv07_reloadDStream(&bitD) >= FSEv07_DStream_partiallyFilled; Ends at exactly BITv07_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv07_GETSYMBOL(&state1);
+
+        if (BITv07_reloadDStream(&bitD)==BITv07_DStream_overflow) {
+            *op++ = FSEv07_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv07_GETSYMBOL(&state2);
+
+        if (BITv07_reloadDStream(&bitD)==BITv07_DStream_overflow) {
+            *op++ = FSEv07_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+
+size_t FSEv07_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv07_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv07_DTableHeader* DTableH = (const FSEv07_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv07_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv07_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv07_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv07_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv07_MAX_SYMBOL_VALUE;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    {   size_t const NCountLength = FSEv07_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+        if (FSEv07_isError(NCountLength)) return NCountLength;
+        if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+        ip += NCountLength;
+        cSrcSize -= NCountLength;
+    }
+
+    { size_t const errorCode = FSEv07_buildDTable (dt, counting, maxSymbolValue, tableLog);
+      if (FSEv07_isError(errorCode)) return errorCode; }
+
+    return FSEv07_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);   /* always return, even if it is an error code */
+}
+
+
+
+#endif   /* FSEv07_COMMONDEFS_ONLY */
+
+/* ******************************************************************
+   Huffman decoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUFv07_STATIC_ASSERT(c) { enum { HUFv07_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/*-***************************/
+/*  generic DTableDesc       */
+/*-***************************/
+
+typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
+
+static DTableDesc HUFv07_getDTableDesc(const HUFv07_DTable* table)
+{
+    DTableDesc dtd;
+    memcpy(&dtd, table, sizeof(dtd));
+    return dtd;
+}
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+typedef struct { BYTE byte; BYTE nbBits; } HUFv07_DEltX2;   /* single-symbol decoding */
+
+size_t HUFv07_readDTableX2 (HUFv07_DTable* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv07_SYMBOLVALUE_MAX + 1];
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    size_t iSize;
+    void* const dtPtr = DTable + 1;
+    HUFv07_DEltX2* const dt = (HUFv07_DEltX2*)dtPtr;
+
+    HUFv07_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUFv07_DTable));
+    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv07_readStats(huffWeight, HUFv07_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv07_isError(iSize)) return iSize;
+
+    /* Table header */
+    {   DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, huffman tree cannot fit in */
+        dtd.tableType = 0;
+        dtd.tableLog = (BYTE)tableLog;
+        memcpy(DTable, &dtd, sizeof(dtd));
+    }
+
+    /* Prepare ranks */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<tableLog+1; n++) {
+            U32 current = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = current;
+    }   }
+
+    /* fill DTable */
+    {   U32 n;
+        for (n=0; n<nbSymbols; n++) {
+            U32 const w = huffWeight[n];
+            U32 const length = (1 << w) >> 1;
+            U32 i;
+            HUFv07_DEltX2 D;
+            D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+            for (i = rankVal[w]; i < rankVal[w] + length; i++)
+                dt[i] = D;
+            rankVal[w] += length;
+    }   }
+
+    return iSize;
+}
+
+
+static BYTE HUFv07_decodeSymbolX2(BITv07_DStream_t* Dstream, const HUFv07_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BITv07_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    BYTE const c = dt[val].byte;
+    BITv07_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv07_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv07_TABLELOG_MAX<=12)) \
+        HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv07_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv07_decodeStreamX2(BYTE* p, BITv07_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv07_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv07_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p < pEnd))
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+static size_t HUFv07_decompress1X2_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const void* dtPtr = DTable + 1;
+    const HUFv07_DEltX2* const dt = (const HUFv07_DEltX2*)dtPtr;
+    BITv07_DStream_t bitD;
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    U32 const dtLog = dtd.tableLog;
+
+    { size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv07_isError(errorCode)) return errorCode; }
+
+    HUFv07_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv07_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv07_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUFv07_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress1X2_DCtx (HUFv07_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX2 (DCtx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress1X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
+}
+
+size_t HUFv07_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX2(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+static size_t HUFv07_decompress4X2_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable + 1;
+        const HUFv07_DEltX2* const dt = (const HUFv07_DEltX2*)dtPtr;
+
+        /* Init */
+        BITv07_DStream_t bitD1;
+        BITv07_DStream_t bitD2;
+        BITv07_DStream_t bitD3;
+        BITv07_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        { size_t const errorCode = BITv07_initDStream(&bitD1, istart1, length1);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD2, istart2, length2);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD3, istart3, length3);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD4, istart4, length4);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv07_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv07_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv07_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv07_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv07_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv07_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv07_endOfDStream(&bitD1) & BITv07_endOfDStream(&bitD2) & BITv07_endOfDStream(&bitD3) & BITv07_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv07_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUFv07_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+
+size_t HUFv07_decompress4X2_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX2 (dctx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress4X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, dctx);
+}
+
+size_t HUFv07_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX2(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv07_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+static void HUFv07_fillDTableX4Level2(HUFv07_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv07_DEltX4 DElt;
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    { U32 s; for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }}
+}
+
+typedef U32 rankVal_t[HUFv07_TABLELOG_ABSOLUTEMAX][HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+
+static void HUFv07_fillDTableX4(HUFv07_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv07_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            HUFv07_DEltX4 DElt;
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits = (BYTE)(nbBits);
+            DElt.length = 1;
+            {   U32 u;
+                const U32 end = start + length;
+                for (u = start; u < end; u++) DTable[u] = DElt;
+        }   }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv07_readDTableX4 (HUFv07_DTable* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv07_SYMBOLVALUE_MAX + 1];
+    sortedSymbol_t sortedSymbol[HUFv07_SYMBOLVALUE_MAX + 1];
+    U32 rankStats[HUFv07_TABLELOG_ABSOLUTEMAX + 1] = { 0 };
+    U32 rankStart0[HUFv07_TABLELOG_ABSOLUTEMAX + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    U32 const maxTableLog = dtd.maxTableLog;
+    size_t iSize;
+    void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
+    HUFv07_DEltX4* const dt = (HUFv07_DEltX4*)dtPtr;
+
+    HUFv07_STATIC_ASSERT(sizeof(HUFv07_DEltX4) == sizeof(HUFv07_DTable));   /* if compilation fails here, assertion is false */
+    if (maxTableLog > HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(tableLog_tooLarge);
+    /* memset(weightList, 0, sizeof(weightList)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv07_readStats(weightList, HUFv07_SYMBOLVALUE_MAX + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv07_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = weightList[s];
+            U32 const r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = rankVal[0];
+        {   int const rescale = (maxTableLog-tableLog) - 1;   /* tableLog <= maxTableLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 current = nextRankVal;
+                nextRankVal += rankStats[w] << (w+rescale);
+                rankVal0[w] = current;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < maxTableLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUFv07_fillDTableX4(dt, maxTableLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    dtd.tableLog = (BYTE)maxTableLog;
+    dtd.tableType = 1;
+    memcpy(DTable, &dtd, sizeof(dtd));
+    return iSize;
+}
+
+
+static U32 HUFv07_decodeSymbolX4(void* op, BITv07_DStream_t* DStream, const HUFv07_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv07_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv07_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv07_decodeLastSymbolX4(void* op, BITv07_DStream_t* DStream, const HUFv07_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv07_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv07_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv07_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv07_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv07_TABLELOG_MAX<=12)) \
+        ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv07_decodeStreamX4(BYTE* p, BITv07_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv07_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv07_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p <= pEnd-2))
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv07_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+static size_t HUFv07_decompress1X4_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    BITv07_DStream_t bitD;
+
+    /* Init */
+    {   size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);
+        if (HUFv07_isError(errorCode)) return errorCode;
+    }
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUFv07_DEltX4* const dt = (const HUFv07_DEltX4*)dtPtr;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        HUFv07_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BITv07_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv07_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUFv07_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress1X4_DCtx (HUFv07_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX4 (DCtx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress1X4_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
+}
+
+size_t HUFv07_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX4(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress1X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+static size_t HUFv07_decompress4X4_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;
+        const HUFv07_DEltX4* const dt = (const HUFv07_DEltX4*)dtPtr;
+
+        /* Init */
+        BITv07_DStream_t bitD1;
+        BITv07_DStream_t bitD2;
+        BITv07_DStream_t bitD3;
+        BITv07_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        { size_t const errorCode = BITv07_initDStream(&bitD1, istart1, length1);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD2, istart2, length2);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD3, istart3, length3);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD4, istart4, length4);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv07_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv07_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv07_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv07_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv07_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv07_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BITv07_endOfDStream(&bitD1) & BITv07_endOfDStream(&bitD2) & BITv07_endOfDStream(&bitD3) & BITv07_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv07_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUFv07_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+
+size_t HUFv07_decompress4X4_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv07_readDTableX4 (dctx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx);
+}
+
+size_t HUFv07_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX4(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress4X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+size_t HUFv07_decompress1X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUFv07_DTable* DTable)
+{
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    return dtd.tableType ? HUFv07_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
+                           HUFv07_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress4X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUFv07_DTable* DTable)
+{
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    return dtd.tableType ? HUFv07_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
+                           HUFv07_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+}
+
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+/** HUFv07_selectDecoder() :
+*   Tells which decoder is likely to decode faster,
+*   based on a set of pre-determined metrics.
+*   @return : 0==HUFv07_decompress4X2, 1==HUFv07_decompress4X4 .
+*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+U32 HUFv07_selectDecoder (size_t dstSize, size_t cSrcSize)
+{
+    /* decoder timing evaluation */
+    U32 const Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    U32 const D256 = (U32)(dstSize >> 8);
+    U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+    U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+    DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, for cache eviction */
+
+    return DTime1 < DTime0;
+}
+
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv07_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[2] = { HUFv07_decompress4X2, HUFv07_decompress4X4 };
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+    }
+
+    /* return HUFv07_decompress4X2(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams single-symbol decoding */
+    /* return HUFv07_decompress4X4(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams double-symbols decoding */
+}
+
+size_t HUFv07_decompress4X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+
+size_t HUFv07_decompress4X_hufOnly (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if ((cSrcSize >= dstSize) || (cSrcSize <= 1)) return ERROR(corruption_detected);   /* invalid */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+
+size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress1X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress1X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+/*
+    Common functions of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+/*! ZSTDv07_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv07_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTDv07_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZSTDv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+
+/* **************************************************************
+*  ZBUFF Error Management
+****************************************************************/
+unsigned ZBUFFv07_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZBUFFv07_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+
+
+static void* ZSTDv07_defaultAllocFunction(void* opaque, size_t size)
+{
+    void* address = malloc(size);
+    (void)opaque;
+    /* printf("alloc %p, %d opaque=%p \n", address, (int)size, opaque); */
+    return address;
+}
+
+static void ZSTDv07_defaultFreeFunction(void* opaque, void* address)
+{
+    (void)opaque;
+    /* if (address) printf("free %p opaque=%p \n", address, opaque); */
+    free(address);
+}
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://www.zstd.net
+*/
+#ifndef ZSTDv07_CCOMMON_H_MODULE
+#define ZSTDv07_CCOMMON_H_MODULE
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv07_OPT_NUM    (1<<12)
+#define ZSTDv07_DICT_MAGIC  0xEC30A437   /* v0.7 */
+
+#define ZSTDv07_REP_NUM    3
+#define ZSTDv07_REP_INIT   ZSTDv07_REP_NUM
+#define ZSTDv07_REP_MOVE   (ZSTDv07_REP_NUM-1)
+static const U32 repStartValue[ZSTDv07_REP_NUM] = { 1, 4, 8 };
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTDv07_WINDOWLOG_ABSOLUTEMIN 10
+static const size_t ZSTDv07_fcs_fieldSize[4] = { 0, 2, 4, 8 };
+static const size_t ZSTDv07_did_fieldSize[4] = { 0, 1, 2, 4 };
+
+#define ZSTDv07_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
+static const size_t ZSTDv07_blockHeaderSize = ZSTDv07_BLOCKHEADERSIZE;
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
+typedef enum { lbt_huffman, lbt_repeat, lbt_raw, lbt_rle } litBlockType_t;
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+#define EQUAL_READ32 4
+
+#define Litbits  8
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  52
+#define MaxLL  35
+#define MaxOff 28
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+
+#define FSEv07_ENCODING_RAW     0
+#define FSEv07_ENCODING_RLE     1
+#define FSEv07_ENCODING_STATIC  2
+#define FSEv07_ENCODING_DYNAMIC 3
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+                                     13,14,15,16 };
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+                                            -1,-1,-1,-1 };
+static const U32 LL_defaultNormLog = 6;
+
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+                                     12,13,14,15,16 };
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+                                            -1,-1,-1,-1,-1 };
+static const U32 ML_defaultNormLog = 6;
+
+static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const U32 OF_defaultNormLog = 5;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv07_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+#define COPY8(d,s) { ZSTDv07_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv07_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+#define WILDCOPY_OVERLENGTH 8
+MEM_STATIC void ZSTDv07_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct ZSTDv07_stats_s ZSTDv07_stats_t;
+
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTDv07_match_t;
+
+typedef struct {
+    U32 price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTDv07_REP_INIT];
+} ZSTDv07_optimal_t;
+
+struct ZSTDv07_stats_s { U32 unused; };
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* litStart;
+    BYTE* lit;
+    U16*  litLengthStart;
+    U16*  litLength;
+    BYTE* llCodeStart;
+    U16*  matchLengthStart;
+    U16*  matchLength;
+    BYTE* mlCodeStart;
+    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
+    U32   longLengthPos;
+    /* opt */
+    ZSTDv07_optimal_t* priceTable;
+    ZSTDv07_match_t* matchTable;
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  matchSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+    U32  log2matchLengthSum;
+    U32  log2matchSum;
+    U32  log2litLengthSum;
+    U32  log2litSum;
+    U32  log2offCodeSum;
+    U32  factor;
+    U32  cachedPrice;
+    U32  cachedLitLength;
+    const BYTE* cachedLiterals;
+    ZSTDv07_stats_t stats;
+} seqStore_t;
+
+void ZSTDv07_seqToCodes(const seqStore_t* seqStorePtr, size_t const nbSeq);
+
+/* custom memory allocation functions */
+static const ZSTDv07_customMem defaultCustomMem = { ZSTDv07_defaultAllocFunction, ZSTDv07_defaultFreeFunction, NULL };
+
+#endif   /* ZSTDv07_CCOMMON_H_MODULE */
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv07_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv07_HEAPMODE
+#  define ZSTDv07_HEAPMODE 1
+#endif
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#endif
+
+
+/*-*************************************
+*  Macros
+***************************************/
+#define ZSTDv07_isError ERR_isError   /* for inlining */
+#define FSEv07_isError  ERR_isError
+#define HUFv07_isError  ERR_isError
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv07_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock,
+               ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTDv07_dStage;
+
+struct ZSTDv07_DCtx_s
+{
+    FSEv07_DTable LLTable[FSEv07_DTABLE_SIZE_U32(LLFSELog)];
+    FSEv07_DTable OffTable[FSEv07_DTABLE_SIZE_U32(OffFSELog)];
+    FSEv07_DTable MLTable[FSEv07_DTABLE_SIZE_U32(MLFSELog)];
+    HUFv07_DTable hufTable[HUFv07_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];  /* can accommodate HUFv07_decompress4X */
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    U32 rep[3];
+    ZSTDv07_frameParams fParams;
+    blockType_t bType;   /* used in ZSTDv07_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv07_dStage stage;
+    U32 litEntropy;
+    U32 fseEntropy;
+    XXH64_state_t xxhState;
+    size_t headerSize;
+    U32 dictID;
+    const BYTE* litPtr;
+    ZSTDv07_customMem customMem;
+    size_t litSize;
+    BYTE litBuffer[ZSTDv07_BLOCKSIZE_ABSOLUTEMAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv07_FRAMEHEADERSIZE_MAX];
+};  /* typedef'd to ZSTDv07_DCtx within "zstd_static.h" */
+
+int ZSTDv07_isSkipFrame(ZSTDv07_DCtx* dctx);
+
+size_t ZSTDv07_sizeofDCtx (const ZSTDv07_DCtx* dctx) { return sizeof(*dctx); }
+
+size_t ZSTDv07_estimateDCtxSize(void) { return sizeof(ZSTDv07_DCtx); }
+
+size_t ZSTDv07_decompressBegin(ZSTDv07_DCtx* dctx)
+{
+    dctx->expected = ZSTDv07_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTable[0] = (HUFv07_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    { int i; for (i=0; i<ZSTDv07_REP_NUM; i++) dctx->rep[i] = repStartValue[i]; }
+    return 0;
+}
+
+ZSTDv07_DCtx* ZSTDv07_createDCtx_advanced(ZSTDv07_customMem customMem)
+{
+    ZSTDv07_DCtx* dctx;
+
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    dctx = (ZSTDv07_DCtx*) customMem.customAlloc(customMem.opaque, sizeof(ZSTDv07_DCtx));
+    if (!dctx) return NULL;
+    memcpy(&dctx->customMem, &customMem, sizeof(ZSTDv07_customMem));
+    ZSTDv07_decompressBegin(dctx);
+    return dctx;
+}
+
+ZSTDv07_DCtx* ZSTDv07_createDCtx(void)
+{
+    return ZSTDv07_createDCtx_advanced(defaultCustomMem);
+}
+
+size_t ZSTDv07_freeDCtx(ZSTDv07_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support free on NULL */
+    dctx->customMem.customFree(dctx->customMem.opaque, dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv07_copyDCtx(ZSTDv07_DCtx* dstDCtx, const ZSTDv07_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv07_DCtx) - (ZSTDv07_BLOCKSIZE_ABSOLUTEMAX+WILDCOPY_OVERLENGTH + ZSTDv07_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/*-*************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv07_MAGICNUMBER (defined within zstd.h)
+      - 1 byte  - Frame Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+
+/* Frame Header :
+
+   1 byte - FrameHeaderDescription :
+   bit 0-1 : dictID (0, 1, 2 or 4 bytes)
+   bit 2   : checksumFlag
+   bit 3   : reserved (must be zero)
+   bit 4   : reserved (unused, can be any value)
+   bit 5   : Single Segment (if 1, WindowLog byte is not present)
+   bit 6-7 : FrameContentFieldSize (0, 2, 4, or 8)
+             if (SkippedWindowLog && !FrameContentFieldsize) FrameContentFieldsize=1;
+
+   Optional : WindowLog (0 or 1 byte)
+   bit 0-2 : octal Fractional (1/8th)
+   bit 3-7 : Power of 2, with 0 = 1 KB (up to 2 TB)
+
+   Optional : dictID (0, 1, 2 or 4 bytes)
+   Automatic adaptation
+   0 : no dictID
+   1 : 1 - 255
+   2 : 256 - 65535
+   4 : all other values
+
+   Optional : content size (0, 1, 2, 4 or 8 bytes)
+   0 : unknown          (fcfs==0 and swl==0)
+   1 : 0-255 bytes      (fcfs==0 and swl==1)
+   2 : 256 - 65535+256  (fcfs==1)
+   4 : 0 - 4GB-1        (fcfs==2)
+   8 : 0 - 16EB-1       (fcfs==3)
+*/
+
+
+/* Compressed Block, format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+/** ZSTDv07_frameHeaderSize() :
+*   srcSize must be >= ZSTDv07_frameHeaderSize_min.
+*   @return : size of the Frame Header */
+static size_t ZSTDv07_frameHeaderSize(const void* src, size_t srcSize)
+{
+    if (srcSize < ZSTDv07_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    {   BYTE const fhd = ((const BYTE*)src)[4];
+        U32 const dictID= fhd & 3;
+        U32 const directMode = (fhd >> 5) & 1;
+        U32 const fcsId = fhd >> 6;
+        return ZSTDv07_frameHeaderSize_min + !directMode + ZSTDv07_did_fieldSize[dictID] + ZSTDv07_fcs_fieldSize[fcsId]
+                + (directMode && !ZSTDv07_fcs_fieldSize[fcsId]);
+    }
+}
+
+
+/** ZSTDv07_getFrameParams() :
+*   decode Frame Header, or require larger `srcSize`.
+*   @return : 0, `fparamsPtr` is correctly filled,
+*            >0, `srcSize` is too small, result is expected `srcSize`,
+*             or an error code, which can be tested using ZSTDv07_isError() */
+size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize < ZSTDv07_frameHeaderSize_min) return ZSTDv07_frameHeaderSize_min;
+    memset(fparamsPtr, 0, sizeof(*fparamsPtr));
+    if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) {
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
+            if (srcSize < ZSTDv07_skippableHeaderSize) return ZSTDv07_skippableHeaderSize; /* magic number + skippable frame length */
+            fparamsPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
+            fparamsPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
+            return 0;
+        }
+        return ERROR(prefix_unknown);
+    }
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    { size_t const fhsize = ZSTDv07_frameHeaderSize(src, srcSize);
+      if (srcSize < fhsize) return fhsize; }
+
+    {   BYTE const fhdByte = ip[4];
+        size_t pos = 5;
+        U32 const dictIDSizeCode = fhdByte&3;
+        U32 const checksumFlag = (fhdByte>>2)&1;
+        U32 const directMode = (fhdByte>>5)&1;
+        U32 const fcsID = fhdByte>>6;
+        U32 const windowSizeMax = 1U << ZSTDv07_WINDOWLOG_MAX;
+        U32 windowSize = 0;
+        U32 dictID = 0;
+        U64 frameContentSize = 0;
+        if ((fhdByte & 0x08) != 0)   /* reserved bits, which must be zero */
+            return ERROR(frameParameter_unsupported);
+        if (!directMode) {
+            BYTE const wlByte = ip[pos++];
+            U32 const windowLog = (wlByte >> 3) + ZSTDv07_WINDOWLOG_ABSOLUTEMIN;
+            if (windowLog > ZSTDv07_WINDOWLOG_MAX)
+                return ERROR(frameParameter_unsupported);
+            windowSize = (1U << windowLog);
+            windowSize += (windowSize >> 3) * (wlByte&7);
+        }
+
+        switch(dictIDSizeCode)
+        {
+            default:   /* impossible */
+            case 0 : break;
+            case 1 : dictID = ip[pos]; pos++; break;
+            case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
+            case 3 : dictID = MEM_readLE32(ip+pos); pos+=4; break;
+        }
+        switch(fcsID)
+        {
+            default:   /* impossible */
+            case 0 : if (directMode) frameContentSize = ip[pos]; break;
+            case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
+            case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
+            case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
+        }
+        if (!windowSize) windowSize = (U32)frameContentSize;
+        if (windowSize > windowSizeMax)
+            return ERROR(frameParameter_unsupported);
+        fparamsPtr->frameContentSize = frameContentSize;
+        fparamsPtr->windowSize = windowSize;
+        fparamsPtr->dictID = dictID;
+        fparamsPtr->checksumFlag = checksumFlag;
+    }
+    return 0;
+}
+
+
+/** ZSTDv07_getDecompressedSize() :
+*   compatible with legacy mode
+*   @return : decompressed size if known, 0 otherwise
+              note : 0 can mean any of the following :
+                   - decompressed size is not provided within frame header
+                   - frame header unknown / not supported
+                   - frame header not completely provided (`srcSize` too small) */
+unsigned long long ZSTDv07_getDecompressedSize(const void* src, size_t srcSize)
+{
+    ZSTDv07_frameParams fparams;
+    size_t const frResult = ZSTDv07_getFrameParams(&fparams, src, srcSize);
+    if (frResult!=0) return 0;
+    return fparams.frameContentSize;
+}
+
+
+/** ZSTDv07_decodeFrameHeader() :
+*   `srcSize` must be the size provided by ZSTDv07_frameHeaderSize().
+*   @return : 0 if success, or an error code, which can be tested using ZSTDv07_isError() */
+static size_t ZSTDv07_decodeFrameHeader(ZSTDv07_DCtx* dctx, const void* src, size_t srcSize)
+{
+    size_t const result = ZSTDv07_getFrameParams(&(dctx->fParams), src, srcSize);
+    if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID)) return ERROR(dictionary_wrong);
+    if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
+    return result;
+}
+
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTDv07_getcBlockSize() :
+*   Provides the size of compressed block from block header `src` */
+static size_t ZSTDv07_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE*)src;
+    U32 cSize;
+
+    if (srcSize < ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    bpPtr->blockType = (blockType_t)((*in) >> 6);
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv07_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/*! ZSTDv07_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTDv07_decodeLiteralsBlock(ZSTDv07_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch((litBlockType_t)(istart[0]>> 6))
+    {
+    case lbt_huffman:
+        {   size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = (istart[0] >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for lhSize, + cSize (+nbSeq) */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv07_isError(singleStream ?
+                            HUFv07_decompress1X2_DCtx(dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv07_decompress4X_hufOnly (dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            dctx->litEntropy = 1;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case lbt_repeat:
+        {   size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (dctx->litEntropy==0)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            {   size_t const errorCode = HUFv07_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTable);
+                if (HUFv07_isError(errorCode)) return ERROR(corruption_detected);
+            }
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case lbt_raw:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case lbt_rle:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+/*! ZSTDv07_buildSeqTable() :
+    @return : nb bytes read from src,
+              or an error code if it fails, testable with ZSTDv07_isError()
+*/
+static size_t ZSTDv07_buildSeqTable(FSEv07_DTable* DTable, U32 type, U32 max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
+{
+    switch(type)
+    {
+    case FSEv07_ENCODING_RLE :
+        if (!srcSize) return ERROR(srcSize_wrong);
+        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
+        FSEv07_buildDTable_rle(DTable, *(const BYTE*)src);   /* if *src > max, data is corrupted */
+        return 1;
+    case FSEv07_ENCODING_RAW :
+        FSEv07_buildDTable(DTable, defaultNorm, max, defaultLog);
+        return 0;
+    case FSEv07_ENCODING_STATIC:
+        if (!flagRepeatTable) return ERROR(corruption_detected);
+        return 0;
+    default :   /* impossible */
+    case FSEv07_ENCODING_DYNAMIC :
+        {   U32 tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSEv07_readNCount(norm, &max, &tableLog, src, srcSize);
+            if (FSEv07_isError(headerSize)) return ERROR(corruption_detected);
+            if (tableLog > maxLog) return ERROR(corruption_detected);
+            FSEv07_buildDTable(DTable, norm, max, tableLog);
+            return headerSize;
+    }   }
+}
+
+
+static size_t ZSTDv07_decodeSeqHeaders(int* nbSeqPtr,
+                             FSEv07_DTable* DTableLL, FSEv07_DTable* DTableML, FSEv07_DTable* DTableOffb, U32 flagRepeatTable,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    {   int nbSeq = *ip++;
+        if (!nbSeq) { *nbSeqPtr=0; return 1; }
+        if (nbSeq > 0x7F) {
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
+                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
+                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
+        }
+        *nbSeqPtr = nbSeq;
+    }
+
+    /* FSE table descriptors */
+    if (ip + 4 > iend) return ERROR(srcSize_wrong); /* min : header byte + all 3 are "raw", hence no header, but at least xxLog bits per type */
+    {   U32 const LLtype  = *ip >> 6;
+        U32 const OFtype = (*ip >> 4) & 3;
+        U32 const MLtype  = (*ip >> 2) & 3;
+        ip++;
+
+        /* Build DTables */
+        {   size_t const llhSize = ZSTDv07_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(llhSize)) return ERROR(corruption_detected);
+            ip += llhSize;
+        }
+        {   size_t const ofhSize = ZSTDv07_buildSeqTable(DTableOffb, OFtype, MaxOff, OffFSELog, ip, iend-ip, OF_defaultNorm, OF_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(ofhSize)) return ERROR(corruption_detected);
+            ip += ofhSize;
+        }
+        {   size_t const mlhSize = ZSTDv07_buildSeqTable(DTableML, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_defaultNorm, ML_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(mlhSize)) return ERROR(corruption_detected);
+            ip += mlhSize;
+    }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv07_DStream_t DStream;
+    FSEv07_DState_t stateLL;
+    FSEv07_DState_t stateOffb;
+    FSEv07_DState_t stateML;
+    size_t prevOffset[ZSTDv07_REP_INIT];
+} seqState_t;
+
+
+static seq_t ZSTDv07_decodeSequence(seqState_t* seqState)
+{
+    seq_t seq;
+
+    U32 const llCode = FSEv07_peekSymbol(&(seqState->stateLL));
+    U32 const mlCode = FSEv07_peekSymbol(&(seqState->stateML));
+    U32 const ofCode = FSEv07_peekSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+
+    U32 const llBits = LL_bits[llCode];
+    U32 const mlBits = ML_bits[mlCode];
+    U32 const ofBits = ofCode;
+    U32 const totalBits = llBits+mlBits+ofBits;
+
+    static const U32 LL_base[MaxLL+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
+                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                            0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 ML_base[MaxML+1] = {
+                             3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13,   14,    15,    16,    17,    18,
+                            19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,   30,    31,    32,    33,    34,
+                            35, 37, 39, 41, 43, 47, 51, 59, 67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+    static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            offset = OF_base[ofCode] + BITv07_readBits(&(seqState->DStream), ofBits);   /* <=  (ZSTDv07_WINDOWLOG_MAX-1) bits */
+            if (MEM_32bits()) BITv07_reloadDStream(&(seqState->DStream));
+        }
+
+        if (ofCode <= 1) {
+            if ((llCode == 0) & (offset <= 1)) offset = 1-offset;
+            if (offset) {
+                size_t const temp = seqState->prevOffset[offset];
+                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BITv07_readBits(&(seqState->DStream), mlBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits>24)) BITv07_reloadDStream(&(seqState->DStream));
+
+    seq.litLength = LL_base[llCode] + ((llCode>15) ? BITv07_readBits(&(seqState->DStream), llBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() ||
+       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BITv07_reloadDStream(&(seqState->DStream));
+
+    /* ANS state update */
+    FSEv07_updateState(&(seqState->stateLL), &(seqState->DStream));   /* <=  9 bits */
+    FSEv07_updateState(&(seqState->stateML), &(seqState->DStream));   /* <=  9 bits */
+    if (MEM_32bits()) BITv07_reloadDStream(&(seqState->DStream));     /* <= 18 bits */
+    FSEv07_updateState(&(seqState->stateOffb), &(seqState->DStream)); /* <=  8 bits */
+
+    return seq;
+}
+
+
+static
+size_t ZSTDv07_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend-WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    assert(oend >= op);
+    if (sequence.litLength + WILDCOPY_OVERLENGTH > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequenceLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    assert(litLimit >= *litPtr);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);;
+
+    /* copy Literals */
+    ZSTDv07_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = (size_t)(dictEnd - match);
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_w */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv07_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv07_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_w) {
+            ZSTDv07_wildcopy(op, match, oend_w - op);
+            match += oend_w - op;
+            op = oend_w;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTDv07_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv07_decompressSequences(
+                               ZSTDv07_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    FSEv07_DTable* DTableLL = dctx->LLTable;
+    FSEv07_DTable* DTableML = dctx->MLTable;
+    FSEv07_DTable* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    int nbSeq;
+
+    /* Build Decoding Tables */
+    {   size_t const seqHSize = ZSTDv07_decodeSeqHeaders(&nbSeq, DTableLL, DTableML, DTableOffb, dctx->fseEntropy, ip, seqSize);
+        if (ZSTDv07_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+    }
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTDv07_REP_INIT; i++) seqState.prevOffset[i] = dctx->rep[i]; }
+        { size_t const errorCode = BITv07_initDStream(&(seqState.DStream), ip, iend-ip);
+          if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
+        FSEv07_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv07_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv07_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv07_reloadDStream(&(seqState.DStream)) <= BITv07_DStream_completed) && nbSeq ; ) {
+            nbSeq--;
+            {   seq_t const sequence = ZSTDv07_decodeSequence(&seqState);
+                size_t const oneSeqSize = ZSTDv07_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                if (ZSTDv07_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+        }   }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTDv07_REP_INIT; i++) dctx->rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        /* if (litPtr > litEnd) return ERROR(corruption_detected); */   /* too many literals already used */
+        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
+        if (lastLLSize > 0) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv07_checkContinuity(ZSTDv07_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv07_decompressBlock_internal(ZSTDv07_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize >= ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    {   size_t const litCSize = ZSTDv07_decodeLiteralsBlock(dctx, src, srcSize);
+        if (ZSTDv07_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+    return ZSTDv07_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv07_decompressBlock(ZSTDv07_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    size_t dSize;
+    ZSTDv07_checkContinuity(dctx, dst);
+    dSize = ZSTDv07_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+    dctx->previousDstEnd = (char*)dst + dSize;
+    return dSize;
+}
+
+
+/** ZSTDv07_insertBlock() :
+    insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockStart, size_t blockSize)
+{
+    ZSTDv07_checkContinuity(dctx, blockStart);
+    dctx->previousDstEnd = (const char*)blockStart + blockSize;
+    return blockSize;
+}
+
+
+static size_t ZSTDv07_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+{
+    if (length > dstCapacity) return ERROR(dstSize_tooSmall);
+    if (length > 0) {
+        memset(dst, byte, length);
+    }
+    return length;
+}
+
+
+/*! ZSTDv07_decompressFrame() :
+*   `dctx` must be properly initialized */
+static size_t ZSTDv07_decompressFrame(ZSTDv07_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t remainingSize = srcSize;
+
+    /* check */
+    if (srcSize < ZSTDv07_frameHeaderSize_min+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+        if (ZSTDv07_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+        if (ZSTDv07_decodeFrameHeader(dctx, src, frameHeaderSize)) return ERROR(corruption_detected);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t decodedSize;
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTDv07_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv07_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv07_blockHeaderSize;
+        remainingSize -= ZSTDv07_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv07_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv07_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            decodedSize = ZSTDv07_generateNxBytes(op, oend-op, *ip, blockProperties.origSize);
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            decodedSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (blockProperties.blockType == bt_end) break;   /* bt_end */
+
+        if (ZSTDv07_isError(decodedSize)) return decodedSize;
+        if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, op, decodedSize);
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+/*! ZSTDv07_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv07_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv07_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
+static size_t ZSTDv07_decompress_usingPreparedDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* refDCtx,
+                                         void* dst, size_t dstCapacity,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv07_copyDCtx(dctx, refDCtx);
+    ZSTDv07_checkContinuity(dctx, dst);
+    return ZSTDv07_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv07_decompress_usingDict(ZSTDv07_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv07_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv07_checkContinuity(dctx, dst);
+    return ZSTDv07_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv07_decompressDCtx(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTDv07_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+}
+
+
+size_t ZSTDv07_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv07_HEAPMODE) && (ZSTDv07_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv07_DCtx* const dctx = ZSTDv07_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv07_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTDv07_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTDv07_DCtx dctx;
+    return ZSTDv07_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv07_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+
+    /* check */
+    if (srcSize < ZSTDv07_frameHeaderSize_min+ZSTDv07_blockHeaderSize) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, srcSize);
+        if (ZSTDv07_isError(frameHeaderSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, frameHeaderSize);
+            return;
+        }
+        if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+            return;
+        }
+        if (srcSize < frameHeaderSize+ZSTDv07_blockHeaderSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTDv07_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv07_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTDv07_blockHeaderSize;
+        remainingSize -= ZSTDv07_blockHeaderSize;
+
+        if (blockProperties.blockType == bt_end) break;
+
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * ZSTDv07_BLOCKSIZE_ABSOLUTEMAX;
+}
+
+/*_******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv07_nextSrcSizeToDecompress(ZSTDv07_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+int ZSTDv07_isSkipFrame(ZSTDv07_DCtx* dctx)
+{
+    return dctx->stage == ZSTDds_skipFrame;
+}
+
+/** ZSTDv07_decompressContinue() :
+*   @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
+*             or an error code, which can be tested using ZSTDv07_isError() */
+size_t ZSTDv07_decompressContinue(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    if (dstCapacity) ZSTDv07_checkContinuity(dctx, dst);
+
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        if (srcSize != ZSTDv07_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
+            memcpy(dctx->headerBuffer, src, ZSTDv07_frameHeaderSize_min);
+            dctx->expected = ZSTDv07_skippableHeaderSize - ZSTDv07_frameHeaderSize_min; /* magic number + skippable frame length */
+            dctx->stage = ZSTDds_decodeSkippableHeader;
+            return 0;
+        }
+        dctx->headerSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+        if (ZSTDv07_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv07_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv07_frameHeaderSize_min) {
+            dctx->expected = dctx->headerSize - ZSTDv07_frameHeaderSize_min;
+            dctx->stage = ZSTDds_decodeFrameHeader;
+            return 0;
+        }
+        dctx->expected = 0;   /* not necessary to copy more */
+	/* fall-through */
+    case ZSTDds_decodeFrameHeader:
+        {   size_t result;
+            memcpy(dctx->headerBuffer + ZSTDv07_frameHeaderSize_min, src, dctx->expected);
+            result = ZSTDv07_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv07_isError(result)) return result;
+            dctx->expected = ZSTDv07_blockHeaderSize;
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTDv07_getcBlockSize(src, ZSTDv07_blockHeaderSize, &bp);
+            if (ZSTDv07_isError(cBlockSize)) return cBlockSize;
+            if (bp.blockType == bt_end) {
+                if (dctx->fParams.checksumFlag) {
+                    U64 const h64 = XXH64_digest(&dctx->xxhState);
+                    U32 const h32 = (U32)(h64>>11) & ((1<<22)-1);
+                    const BYTE* const ip = (const BYTE*)src;
+                    U32 const check32 = ip[2] + (ip[1] << 8) + ((ip[0] & 0x3F) << 16);
+                    if (check32 != h32) return ERROR(checksum_wrong);
+                }
+                dctx->expected = 0;
+                dctx->stage = ZSTDds_getFrameHeaderSize;
+            } else {
+                dctx->expected = cBlockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv07_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv07_copyRawBlock(dst, dstCapacity, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            dctx->expected = ZSTDv07_blockHeaderSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            if (ZSTDv07_isError(rSize)) return rSize;
+            if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
+            return rSize;
+        }
+    case ZSTDds_decodeSkippableHeader:
+        {   memcpy(dctx->headerBuffer + ZSTDv07_frameHeaderSize_min, src, dctx->expected);
+            dctx->expected = MEM_readLE32(dctx->headerBuffer + 4);
+            dctx->stage = ZSTDds_skipFrame;
+            return 0;
+        }
+    case ZSTDds_skipFrame:
+        {   dctx->expected = 0;
+            dctx->stage = ZSTDds_getFrameHeaderSize;
+            return 0;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static size_t ZSTDv07_refDictContent(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+    return 0;
+}
+
+static size_t ZSTDv07_loadEntropy(ZSTDv07_DCtx* dctx, const void* const dict, size_t const dictSize)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+
+    {   size_t const hSize = HUFv07_readDTableX4(dctx->hufTable, dict, dictSize);
+        if (HUFv07_isError(hSize)) return ERROR(dictionary_corrupted);
+        dictPtr += hSize;
+    }
+
+    {   short offcodeNCount[MaxOff+1];
+        U32 offcodeMaxValue=MaxOff, offcodeLog;
+        size_t const offcodeHeaderSize = FSEv07_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSEv07_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSEv07_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += litlengthHeaderSize;
+    }
+
+    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
+    dctx->rep[0] = MEM_readLE32(dictPtr+0); if (dctx->rep[0] == 0 || dctx->rep[0] >= dictSize) return ERROR(dictionary_corrupted);
+    dctx->rep[1] = MEM_readLE32(dictPtr+4); if (dctx->rep[1] == 0 || dctx->rep[1] >= dictSize) return ERROR(dictionary_corrupted);
+    dctx->rep[2] = MEM_readLE32(dictPtr+8); if (dctx->rep[2] == 0 || dctx->rep[2] >= dictSize) return ERROR(dictionary_corrupted);
+    dictPtr += 12;
+
+    dctx->litEntropy = dctx->fseEntropy = 1;
+    return dictPtr - (const BYTE*)dict;
+}
+
+static size_t ZSTDv07_decompress_insertDictionary(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return ZSTDv07_refDictContent(dctx, dict, dictSize);
+    {   U32 const magic = MEM_readLE32(dict);
+        if (magic != ZSTDv07_DICT_MAGIC) {
+            return ZSTDv07_refDictContent(dctx, dict, dictSize);   /* pure content mode */
+    }   }
+    dctx->dictID = MEM_readLE32((const char*)dict + 4);
+
+    /* load entropy tables */
+    dict = (const char*)dict + 8;
+    dictSize -= 8;
+    {   size_t const eSize = ZSTDv07_loadEntropy(dctx, dict, dictSize);
+        if (ZSTDv07_isError(eSize)) return ERROR(dictionary_corrupted);
+        dict = (const char*)dict + eSize;
+        dictSize -= eSize;
+    }
+
+    /* reference dictionary content */
+    return ZSTDv07_refDictContent(dctx, dict, dictSize);
+}
+
+
+size_t ZSTDv07_decompressBegin_usingDict(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    { size_t const errorCode = ZSTDv07_decompressBegin(dctx);
+      if (ZSTDv07_isError(errorCode)) return errorCode; }
+
+    if (dict && dictSize) {
+        size_t const errorCode = ZSTDv07_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv07_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+
+struct ZSTDv07_DDict_s {
+    void* dict;
+    size_t dictSize;
+    ZSTDv07_DCtx* refContext;
+};  /* typedef'd tp ZSTDv07_CDict within zstd.h */
+
+static ZSTDv07_DDict* ZSTDv07_createDDict_advanced(const void* dict, size_t dictSize, ZSTDv07_customMem customMem)
+{
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    {   ZSTDv07_DDict* const ddict = (ZSTDv07_DDict*) customMem.customAlloc(customMem.opaque, sizeof(*ddict));
+        void* const dictContent = customMem.customAlloc(customMem.opaque, dictSize);
+        ZSTDv07_DCtx* const dctx = ZSTDv07_createDCtx_advanced(customMem);
+
+        if (!dictContent || !ddict || !dctx) {
+            customMem.customFree(customMem.opaque, dictContent);
+            customMem.customFree(customMem.opaque, ddict);
+            customMem.customFree(customMem.opaque, dctx);
+            return NULL;
+        }
+
+        memcpy(dictContent, dict, dictSize);
+        {   size_t const errorCode = ZSTDv07_decompressBegin_usingDict(dctx, dictContent, dictSize);
+            if (ZSTDv07_isError(errorCode)) {
+                customMem.customFree(customMem.opaque, dictContent);
+                customMem.customFree(customMem.opaque, ddict);
+                customMem.customFree(customMem.opaque, dctx);
+                return NULL;
+        }   }
+
+        ddict->dict = dictContent;
+        ddict->dictSize = dictSize;
+        ddict->refContext = dctx;
+        return ddict;
+    }
+}
+
+/*! ZSTDv07_createDDict() :
+*   Create a digested dictionary, ready to start decompression without startup delay.
+*   `dict` can be released after `ZSTDv07_DDict` creation */
+ZSTDv07_DDict* ZSTDv07_createDDict(const void* dict, size_t dictSize)
+{
+    ZSTDv07_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTDv07_createDDict_advanced(dict, dictSize, allocator);
+}
+
+size_t ZSTDv07_freeDDict(ZSTDv07_DDict* ddict)
+{
+    ZSTDv07_freeFunction const cFree = ddict->refContext->customMem.customFree;
+    void* const opaque = ddict->refContext->customMem.opaque;
+    ZSTDv07_freeDCtx(ddict->refContext);
+    cFree(opaque, ddict->dict);
+    cFree(opaque, ddict);
+    return 0;
+}
+
+/*! ZSTDv07_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Use dictionary without significant overhead. */
+ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDDict(ZSTDv07_DCtx* dctx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     const ZSTDv07_DDict* ddict)
+{
+    return ZSTDv07_decompress_usingPreparedDCtx(dctx, ddict->refContext,
+                                           dst, dstCapacity,
+                                           src, srcSize);
+}
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv07_DCtx object is required to track streaming operations.
+*  Use ZBUFFv07_createDCtx() and ZBUFFv07_freeDCtx() to create/release resources.
+*  Use ZBUFFv07_decompressInit() to start a new decompression operation,
+*   or ZBUFFv07_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv07_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv07_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv07_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv07_recommendedDInSize() and ZBUFFv07_recommendedDOutSize()
+*  output : ZBUFFv07_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv07_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv07_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_loadHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFFv07_dStage;
+
+/* *** Resource management *** */
+struct ZBUFFv07_DCtx_s {
+    ZSTDv07_DCtx* zd;
+    ZSTDv07_frameParams fParams;
+    ZBUFFv07_dStage stage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t blockSize;
+    BYTE headerBuffer[ZSTDv07_FRAMEHEADERSIZE_MAX];
+    size_t lhSize;
+    ZSTDv07_customMem customMem;
+};   /* typedef'd to ZBUFFv07_DCtx within "zstd_buffered.h" */
+
+ZSTDLIBv07_API ZBUFFv07_DCtx* ZBUFFv07_createDCtx_advanced(ZSTDv07_customMem customMem);
+
+ZBUFFv07_DCtx* ZBUFFv07_createDCtx(void)
+{
+    return ZBUFFv07_createDCtx_advanced(defaultCustomMem);
+}
+
+ZBUFFv07_DCtx* ZBUFFv07_createDCtx_advanced(ZSTDv07_customMem customMem)
+{
+    ZBUFFv07_DCtx* zbd;
+
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    zbd = (ZBUFFv07_DCtx*)customMem.customAlloc(customMem.opaque, sizeof(ZBUFFv07_DCtx));
+    if (zbd==NULL) return NULL;
+    memset(zbd, 0, sizeof(ZBUFFv07_DCtx));
+    memcpy(&zbd->customMem, &customMem, sizeof(ZSTDv07_customMem));
+    zbd->zd = ZSTDv07_createDCtx_advanced(customMem);
+    if (zbd->zd == NULL) { ZBUFFv07_freeDCtx(zbd); return NULL; }
+    zbd->stage = ZBUFFds_init;
+    return zbd;
+}
+
+size_t ZBUFFv07_freeDCtx(ZBUFFv07_DCtx* zbd)
+{
+    if (zbd==NULL) return 0;   /* support free on null */
+    ZSTDv07_freeDCtx(zbd->zd);
+    if (zbd->inBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
+    if (zbd->outBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
+    zbd->customMem.customFree(zbd->customMem.opaque, zbd);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv07_decompressInitDictionary(ZBUFFv07_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    zbd->stage = ZBUFFds_loadHeader;
+    zbd->lhSize = zbd->inPos = zbd->outStart = zbd->outEnd = 0;
+    return ZSTDv07_decompressBegin_usingDict(zbd->zd, dict, dictSize);
+}
+
+size_t ZBUFFv07_decompressInit(ZBUFFv07_DCtx* zbd)
+{
+    return ZBUFFv07_decompressInitDictionary(zbd, NULL, 0);
+}
+
+
+/* internal util function */
+MEM_STATIC size_t ZBUFFv07_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const length = MIN(dstCapacity, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv07_decompressContinue(ZBUFFv07_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* const iend = istart + *srcSizePtr;
+    const char* ip = istart;
+    char* const ostart = (char*)dst;
+    char* const oend = ostart + *dstCapacityPtr;
+    char* op = ostart;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbd->stage)
+        {
+        case ZBUFFds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFds_loadHeader :
+            {   size_t const hSize = ZSTDv07_getFrameParams(&(zbd->fParams), zbd->headerBuffer, zbd->lhSize);
+                if (ZSTDv07_isError(hSize)) return hSize;
+                if (hSize != 0) {
+                    size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
+                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
+                        if (ip != NULL)
+                            memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        zbd->lhSize += iend-ip;
+                        *dstCapacityPtr = 0;
+                        return (hSize - zbd->lhSize) + ZSTDv07_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    memcpy(zbd->headerBuffer + zbd->lhSize, ip, toLoad); zbd->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* Consume header */
+            {   size_t const h1Size = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);  /* == ZSTDv07_frameHeaderSize_min */
+                size_t const h1Result = ZSTDv07_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer, h1Size);
+                if (ZSTDv07_isError(h1Result)) return h1Result;
+                if (h1Size < zbd->lhSize) {   /* long header */
+                    size_t const h2Size = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                    size_t const h2Result = ZSTDv07_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer+h1Size, h2Size);
+                    if (ZSTDv07_isError(h2Result)) return h2Result;
+            }   }
+
+            zbd->fParams.windowSize = MAX(zbd->fParams.windowSize, 1U << ZSTDv07_WINDOWLOG_ABSOLUTEMIN);
+
+            /* Frame header instruct buffer sizes */
+            {   size_t const blockSize = MIN(zbd->fParams.windowSize, ZSTDv07_BLOCKSIZE_ABSOLUTEMAX);
+                zbd->blockSize = blockSize;
+                if (zbd->inBuffSize < blockSize) {
+                    zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
+                    zbd->inBuffSize = blockSize;
+                    zbd->inBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, blockSize);
+                    if (zbd->inBuff == NULL) return ERROR(memory_allocation);
+                }
+                {   size_t const neededOutSize = zbd->fParams.windowSize + blockSize + WILDCOPY_OVERLENGTH * 2;
+                    if (zbd->outBuffSize < neededOutSize) {
+                        zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
+                        zbd->outBuffSize = neededOutSize;
+                        zbd->outBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, neededOutSize);
+                        if (zbd->outBuff == NULL) return ERROR(memory_allocation);
+            }   }   }
+            zbd->stage = ZBUFFds_read;
+            /* pass-through */
+	    /* fall-through */
+        case ZBUFFds_read:
+            {   size_t const neededInSize = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                if (neededInSize==0) {  /* end of frame */
+                    zbd->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    const int isSkipFrame = ZSTDv07_isSkipFrame(zbd->zd);
+                    size_t const decodedSize = ZSTDv07_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, (isSkipFrame ? 0 : zbd->outBuffSize - zbd->outStart),
+                        ip, neededInSize);
+                    if (ZSTDv07_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbd->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {   size_t const neededInSize = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                size_t const toLoad = neededInSize - zbd->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbd->inBuffSize - zbd->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv07_limitCopy(zbd->inBuff + zbd->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbd->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                {  const int isSkipFrame = ZSTDv07_isSkipFrame(zbd->zd);
+                   size_t const decodedSize = ZSTDv07_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        zbd->inBuff, neededInSize);
+                    if (ZSTDv07_isError(decodedSize)) return decodedSize;
+                    zbd->inPos = 0;   /* input is consumed */
+                    if (!decodedSize && !isSkipFrame) { zbd->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    /* break; */
+                    /* pass-through */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {   size_t const toFlushSize = zbd->outEnd - zbd->outStart;
+                size_t const flushedSize = ZBUFFv07_limitCopy(op, oend-op, zbd->outBuff + zbd->outStart, toFlushSize);
+                op += flushedSize;
+                zbd->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbd->stage = ZBUFFds_read;
+                    if (zbd->outStart + zbd->blockSize > zbd->outBuffSize)
+                        zbd->outStart = zbd->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    /* result */
+    *srcSizePtr = ip-istart;
+    *dstCapacityPtr = op-ostart;
+    {   size_t nextSrcSizeHint = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+        nextSrcSizeHint -= zbd->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFFv07_recommendedDInSize(void)  { return ZSTDv07_BLOCKSIZE_ABSOLUTEMAX + ZSTDv07_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv07_recommendedDOutSize(void) { return ZSTDv07_BLOCKSIZE_ABSOLUTEMAX; }
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v07.h b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v07.h
new file mode 100644
index 0000000..1ff3904
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/legacy/zstd_v07.h
@@ -0,0 +1,187 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTDv07_H_235446
+#define ZSTDv07_H_235446
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*======  Dependency  ======*/
+#include <stddef.h>   /* size_t */
+
+
+/*======  Export for Windows  ======*/
+/*!
+*  ZSTDv07_DLL_EXPORT :
+*  Enable exporting of functions when building a Windows DLL
+*/
+#if defined(_WIN32) && defined(ZSTDv07_DLL_EXPORT) && (ZSTDv07_DLL_EXPORT==1)
+#  define ZSTDLIBv07_API __declspec(dllexport)
+#else
+#  define ZSTDLIBv07_API
+#endif
+
+
+/* *************************************
+*  Simple API
+***************************************/
+/*! ZSTDv07_getDecompressedSize() :
+*   @return : decompressed size if known, 0 otherwise.
+       note 1 : if `0`, follow up with ZSTDv07_getFrameParams() to know precise failure cause.
+       note 2 : decompressed size could be wrong or intentionally modified !
+                always ensure results fit within application's authorized limits */
+unsigned long long ZSTDv07_getDecompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTDv07_decompress() :
+    `compressedSize` : must be _exact_ size of compressed input, otherwise decompression will fail.
+    `dstCapacity` must be equal or larger than originalSize.
+    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+              or an errorCode if it fails (which can be tested using ZSTDv07_isError()) */
+ZSTDLIBv07_API size_t ZSTDv07_decompress( void* dst, size_t dstCapacity,
+                                    const void* src, size_t compressedSize);
+
+/**
+ZSTDv07_findFrameSizeInfoLegacy() : get the source length and decompressed bound of a ZSTD frame compliant with v0.7.x format
+    srcSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    cSize (output parameter)  : the number of bytes that would be read to decompress this frame
+                                or an error code if it fails (which can be tested using ZSTDv01_isError())
+    dBound (output parameter) : an upper-bound for the decompressed size of the data in the frame
+                                or ZSTD_CONTENTSIZE_ERROR if an error occurs
+
+    note : assumes `cSize` and `dBound` are _not_ NULL.
+*/
+void ZSTDv07_findFrameSizeInfoLegacy(const void *src, size_t srcSize,
+                                     size_t* cSize, unsigned long long* dBound);
+
+/*======  Helper functions  ======*/
+ZSTDLIBv07_API unsigned    ZSTDv07_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+ZSTDLIBv07_API const char* ZSTDv07_getErrorName(size_t code);     /*!< provides readable string from an error code */
+
+
+/*-*************************************
+*  Explicit memory management
+***************************************/
+/** Decompression context */
+typedef struct ZSTDv07_DCtx_s ZSTDv07_DCtx;
+ZSTDLIBv07_API ZSTDv07_DCtx* ZSTDv07_createDCtx(void);
+ZSTDLIBv07_API size_t     ZSTDv07_freeDCtx(ZSTDv07_DCtx* dctx);      /*!< @return : errorCode */
+
+/** ZSTDv07_decompressDCtx() :
+*   Same as ZSTDv07_decompress(), requires an allocated ZSTDv07_DCtx (see ZSTDv07_createDCtx()) */
+ZSTDLIBv07_API size_t ZSTDv07_decompressDCtx(ZSTDv07_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-************************
+*  Simple dictionary API
+***************************/
+/*! ZSTDv07_decompress_usingDict() :
+*   Decompression using a pre-defined Dictionary content (see dictBuilder).
+*   Dictionary must be identical to the one used during compression.
+*   Note : This function load the dictionary, resulting in a significant startup time */
+ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDict(ZSTDv07_DCtx* dctx,
+                                                   void* dst, size_t dstCapacity,
+                                             const void* src, size_t srcSize,
+                                             const void* dict,size_t dictSize);
+
+
+/*-**************************
+*  Advanced Dictionary API
+****************************/
+/*! ZSTDv07_createDDict() :
+*   Create a digested dictionary, ready to start decompression operation without startup delay.
+*   `dict` can be released after creation */
+typedef struct ZSTDv07_DDict_s ZSTDv07_DDict;
+ZSTDLIBv07_API ZSTDv07_DDict* ZSTDv07_createDDict(const void* dict, size_t dictSize);
+ZSTDLIBv07_API size_t      ZSTDv07_freeDDict(ZSTDv07_DDict* ddict);
+
+/*! ZSTDv07_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Faster startup than ZSTDv07_decompress_usingDict(), recommended when same dictionary is used multiple times. */
+ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDDict(ZSTDv07_DCtx* dctx,
+                                                    void* dst, size_t dstCapacity,
+                                              const void* src, size_t srcSize,
+                                              const ZSTDv07_DDict* ddict);
+
+typedef struct {
+    unsigned long long frameContentSize;
+    unsigned windowSize;
+    unsigned dictID;
+    unsigned checksumFlag;
+} ZSTDv07_frameParams;
+
+ZSTDLIBv07_API size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+
+
+
+
+/* *************************************
+*  Streaming functions
+***************************************/
+typedef struct ZBUFFv07_DCtx_s ZBUFFv07_DCtx;
+ZSTDLIBv07_API ZBUFFv07_DCtx* ZBUFFv07_createDCtx(void);
+ZSTDLIBv07_API size_t      ZBUFFv07_freeDCtx(ZBUFFv07_DCtx* dctx);
+
+ZSTDLIBv07_API size_t ZBUFFv07_decompressInit(ZBUFFv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZBUFFv07_decompressInitDictionary(ZBUFFv07_DCtx* dctx, const void* dict, size_t dictSize);
+
+ZSTDLIBv07_API size_t ZBUFFv07_decompressContinue(ZBUFFv07_DCtx* dctx,
+                                            void* dst, size_t* dstCapacityPtr,
+                                      const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv07_DCtx object is required to track streaming operations.
+*  Use ZBUFFv07_createDCtx() and ZBUFFv07_freeDCtx() to create/release resources.
+*  Use ZBUFFv07_decompressInit() to start a new decompression operation,
+*   or ZBUFFv07_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv07_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv07_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change `dst`.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv07_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv07_recommendedDInSize() and ZBUFFv07_recommendedDOutSize()
+*  output : ZBUFFv07_recommendedDOutSize== 128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv07_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv07_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+ZSTDLIBv07_API unsigned ZBUFFv07_isError(size_t errorCode);
+ZSTDLIBv07_API const char* ZBUFFv07_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, they tend to offer better latency */
+ZSTDLIBv07_API size_t ZBUFFv07_recommendedDInSize(void);
+ZSTDLIBv07_API size_t ZBUFFv07_recommendedDOutSize(void);
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define ZSTDv07_MAGICNUMBER            0xFD2FB527   /* v0.7 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv07_H_235446 */
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/zdict.h b/lib/clickhouse-cpp/contrib/zstd/zstd/zdict.h
new file mode 100644
index 0000000..2268f94
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/zdict.h
@@ -0,0 +1,474 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef ZSTD_ZDICT_H
+#define ZSTD_ZDICT_H
+
+/*======  Dependencies  ======*/
+#include <stddef.h>  /* size_t */
+
+
+/* =====   ZDICTLIB_API : control library symbols visibility   ===== */
+#ifndef ZDICTLIB_VISIBLE
+   /* Backwards compatibility with old macro name */
+#  ifdef ZDICTLIB_VISIBILITY
+#    define ZDICTLIB_VISIBLE ZDICTLIB_VISIBILITY
+#  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZDICTLIB_VISIBLE __attribute__ ((visibility ("default")))
+#  else
+#    define ZDICTLIB_VISIBLE
+#  endif
+#endif
+
+#ifndef ZDICTLIB_HIDDEN
+#  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZDICTLIB_HIDDEN __attribute__ ((visibility ("hidden")))
+#  else
+#    define ZDICTLIB_HIDDEN
+#  endif
+#endif
+
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBLE
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZDICTLIB_API ZDICTLIB_VISIBLE
+#endif
+
+/*******************************************************************************
+ * Zstd dictionary builder
+ *
+ * FAQ
+ * ===
+ * Why should I use a dictionary?
+ * ------------------------------
+ *
+ * Zstd can use dictionaries to improve compression ratio of small data.
+ * Traditionally small files don't compress well because there is very little
+ * repetition in a single sample, since it is small. But, if you are compressing
+ * many similar files, like a bunch of JSON records that share the same
+ * structure, you can train a dictionary on ahead of time on some samples of
+ * these files. Then, zstd can use the dictionary to find repetitions that are
+ * present across samples. This can vastly improve compression ratio.
+ *
+ * When is a dictionary useful?
+ * ----------------------------
+ *
+ * Dictionaries are useful when compressing many small files that are similar.
+ * The larger a file is, the less benefit a dictionary will have. Generally,
+ * we don't expect dictionary compression to be effective past 100KB. And the
+ * smaller a file is, the more we would expect the dictionary to help.
+ *
+ * How do I use a dictionary?
+ * --------------------------
+ *
+ * Simply pass the dictionary to the zstd compressor with
+ * `ZSTD_CCtx_loadDictionary()`. The same dictionary must then be passed to
+ * the decompressor, using `ZSTD_DCtx_loadDictionary()`. There are other
+ * more advanced functions that allow selecting some options, see zstd.h for
+ * complete documentation.
+ *
+ * What is a zstd dictionary?
+ * --------------------------
+ *
+ * A zstd dictionary has two pieces: Its header, and its content. The header
+ * contains a magic number, the dictionary ID, and entropy tables. These
+ * entropy tables allow zstd to save on header costs in the compressed file,
+ * which really matters for small data. The content is just bytes, which are
+ * repeated content that is common across many samples.
+ *
+ * What is a raw content dictionary?
+ * ---------------------------------
+ *
+ * A raw content dictionary is just bytes. It doesn't have a zstd dictionary
+ * header, a dictionary ID, or entropy tables. Any buffer is a valid raw
+ * content dictionary.
+ *
+ * How do I train a dictionary?
+ * ----------------------------
+ *
+ * Gather samples from your use case. These samples should be similar to each
+ * other. If you have several use cases, you could try to train one dictionary
+ * per use case.
+ *
+ * Pass those samples to `ZDICT_trainFromBuffer()` and that will train your
+ * dictionary. There are a few advanced versions of this function, but this
+ * is a great starting point. If you want to further tune your dictionary
+ * you could try `ZDICT_optimizeTrainFromBuffer_cover()`. If that is too slow
+ * you can try `ZDICT_optimizeTrainFromBuffer_fastCover()`.
+ *
+ * If the dictionary training function fails, that is likely because you
+ * either passed too few samples, or a dictionary would not be effective
+ * for your data. Look at the messages that the dictionary trainer printed,
+ * if it doesn't say too few samples, then a dictionary would not be effective.
+ *
+ * How large should my dictionary be?
+ * ----------------------------------
+ *
+ * A reasonable dictionary size, the `dictBufferCapacity`, is about 100KB.
+ * The zstd CLI defaults to a 110KB dictionary. You likely don't need a
+ * dictionary larger than that. But, most use cases can get away with a
+ * smaller dictionary. The advanced dictionary builders can automatically
+ * shrink the dictionary for you, and select the smallest size that doesn't
+ * hurt compression ratio too much. See the `shrinkDict` parameter.
+ * A smaller dictionary can save memory, and potentially speed up
+ * compression.
+ *
+ * How many samples should I provide to the dictionary builder?
+ * ------------------------------------------------------------
+ *
+ * We generally recommend passing ~100x the size of the dictionary
+ * in samples. A few thousand should suffice. Having too few samples
+ * can hurt the dictionaries effectiveness. Having more samples will
+ * only improve the dictionaries effectiveness. But having too many
+ * samples can slow down the dictionary builder.
+ *
+ * How do I determine if a dictionary will be effective?
+ * -----------------------------------------------------
+ *
+ * Simply train a dictionary and try it out. You can use zstd's built in
+ * benchmarking tool to test the dictionary effectiveness.
+ *
+ *   # Benchmark levels 1-3 without a dictionary
+ *   zstd -b1e3 -r /path/to/my/files
+ *   # Benchmark levels 1-3 with a dictionary
+ *   zstd -b1e3 -r /path/to/my/files -D /path/to/my/dictionary
+ *
+ * When should I retrain a dictionary?
+ * -----------------------------------
+ *
+ * You should retrain a dictionary when its effectiveness drops. Dictionary
+ * effectiveness drops as the data you are compressing changes. Generally, we do
+ * expect dictionaries to "decay" over time, as your data changes, but the rate
+ * at which they decay depends on your use case. Internally, we regularly
+ * retrain dictionaries, and if the new dictionary performs significantly
+ * better than the old dictionary, we will ship the new dictionary.
+ *
+ * I have a raw content dictionary, how do I turn it into a zstd dictionary?
+ * -------------------------------------------------------------------------
+ *
+ * If you have a raw content dictionary, e.g. by manually constructing it, or
+ * using a third-party dictionary builder, you can turn it into a zstd
+ * dictionary by using `ZDICT_finalizeDictionary()`. You'll also have to
+ * provide some samples of the data. It will add the zstd header to the
+ * raw content, which contains a dictionary ID and entropy tables, which
+ * will improve compression ratio, and allow zstd to write the dictionary ID
+ * into the frame, if you so choose.
+ *
+ * Do I have to use zstd's dictionary builder?
+ * -------------------------------------------
+ *
+ * No! You can construct dictionary content however you please, it is just
+ * bytes. It will always be valid as a raw content dictionary. If you want
+ * a zstd dictionary, which can improve compression ratio, use
+ * `ZDICT_finalizeDictionary()`.
+ *
+ * What is the attack surface of a zstd dictionary?
+ * ------------------------------------------------
+ *
+ * Zstd is heavily fuzz tested, including loading fuzzed dictionaries, so
+ * zstd should never crash, or access out-of-bounds memory no matter what
+ * the dictionary is. However, if an attacker can control the dictionary
+ * during decompression, they can cause zstd to generate arbitrary bytes,
+ * just like if they controlled the compressed data.
+ *
+ ******************************************************************************/
+
+
+/*! ZDICT_trainFromBuffer():
+ *  Train a dictionary from an array of samples.
+ *  Redirect towards ZDICT_optimizeTrainFromBuffer_fastCover() single-threaded, with d=8, steps=4,
+ *  f=20, and accel=1.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Note:  Dictionary training will fail if there are not enough samples to construct a
+ *         dictionary, or if most of the samples are too small (< 8 bytes being the lower limit).
+ *         If dictionary training fails, you should use zstd without a dictionary, as the dictionary
+ *         would've been ineffective anyways. If you believe your samples would benefit from a dictionary
+ *         please open an issue with details, and we can look into it.
+ *  Note: ZDICT_trainFromBuffer()'s memory usage is about 6 MB.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
+                                    const void* samplesBuffer,
+                                    const size_t* samplesSizes, unsigned nbSamples);
+
+typedef struct {
+    int      compressionLevel;   /**< optimize for a specific zstd compression level; 0 means default */
+    unsigned notificationLevel;  /**< Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
+    unsigned dictID;             /**< force dictID value; 0 means auto mode (32-bits random value)
+                                  *   NOTE: The zstd format reserves some dictionary IDs for future use.
+                                  *         You may use them in private settings, but be warned that they
+                                  *         may be used by zstd in a public dictionary registry in the future.
+                                  *         These dictionary IDs are:
+                                  *           - low range  : <= 32767
+                                  *           - high range : >= (2^31)
+                                  */
+} ZDICT_params_t;
+
+/*! ZDICT_finalizeDictionary():
+ * Given a custom content as a basis for dictionary, and a set of samples,
+ * finalize dictionary by adding headers and statistics according to the zstd
+ * dictionary format.
+ *
+ * Samples must be stored concatenated in a flat buffer `samplesBuffer`,
+ * supplied with an array of sizes `samplesSizes`, providing the size of each
+ * sample in order. The samples are used to construct the statistics, so they
+ * should be representative of what you will compress with this dictionary.
+ *
+ * The compression level can be set in `parameters`. You should pass the
+ * compression level you expect to use in production. The statistics for each
+ * compression level differ, so tuning the dictionary for the compression level
+ * can help quite a bit.
+ *
+ * You can set an explicit dictionary ID in `parameters`, or allow us to pick
+ * a random dictionary ID for you, but we can't guarantee no collisions.
+ *
+ * The dstDictBuffer and the dictContent may overlap, and the content will be
+ * appended to the end of the header. If the header + the content doesn't fit in
+ * maxDictSize the beginning of the content is truncated to make room, since it
+ * is presumed that the most profitable content is at the end of the dictionary,
+ * since that is the cheapest to reference.
+ *
+ * `maxDictSize` must be >= max(dictContentSize, ZSTD_DICTSIZE_MIN).
+ *
+ * @return: size of dictionary stored into `dstDictBuffer` (<= `maxDictSize`),
+ *          or an error code, which can be tested by ZDICT_isError().
+ * Note: ZDICT_finalizeDictionary() will push notifications into stderr if
+ *       instructed to, using notificationLevel>0.
+ * NOTE: This function currently may fail in several edge cases including:
+ *         * Not enough samples
+ *         * Samples are uncompressible
+ *         * Samples are all exactly the same
+ */
+ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dstDictBuffer, size_t maxDictSize,
+                                const void* dictContent, size_t dictContentSize,
+                                const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                                ZDICT_params_t parameters);
+
+
+/*======   Helper functions   ======*/
+ZDICTLIB_API unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize);  /**< extracts dictID; @return zero if error (not a valid dictionary) */
+ZDICTLIB_API size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize);  /* returns dict header size; returns a ZSTD error code on failure */
+ZDICTLIB_API unsigned ZDICT_isError(size_t errorCode);
+ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
+
+#endif   /* ZSTD_ZDICT_H */
+
+#if defined(ZDICT_STATIC_LINKING_ONLY) && !defined(ZSTD_ZDICT_H_STATIC)
+#define ZSTD_ZDICT_H_STATIC
+
+/* This can be overridden externally to hide static symbols. */
+#ifndef ZDICTLIB_STATIC_API
+#  if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#    define ZDICTLIB_STATIC_API __declspec(dllexport) ZDICTLIB_VISIBLE
+#  elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#    define ZDICTLIB_STATIC_API __declspec(dllimport) ZDICTLIB_VISIBLE
+#  else
+#    define ZDICTLIB_STATIC_API ZDICTLIB_VISIBLE
+#  endif
+#endif
+
+/* ====================================================================================
+ * The definitions in this section are considered experimental.
+ * They should never be used with a dynamic library, as they may change in the future.
+ * They are provided for advanced usages.
+ * Use them only in association with static linking.
+ * ==================================================================================== */
+
+#define ZDICT_DICTSIZE_MIN    256
+/* Deprecated: Remove in v1.6.0 */
+#define ZDICT_CONTENTSIZE_MIN 128
+
+/*! ZDICT_cover_params_t:
+ *  k and d are the only required parameters.
+ *  For others, value 0 means default.
+ */
+typedef struct {
+    unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
+    unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
+    unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
+    double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */
+    unsigned shrinkDict;         /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking  */
+    unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
+    ZDICT_params_t zParams;
+} ZDICT_cover_params_t;
+
+typedef struct {
+    unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
+    unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
+    unsigned f;                  /* log of size of frequency array : constraint: 0 < f <= 31 : 1 means default(20)*/
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
+    unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
+    double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (0.75), 1.0 when all samples are used for both training and testing */
+    unsigned accel;              /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */
+    unsigned shrinkDict;         /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking  */
+    unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
+
+    ZDICT_params_t zParams;
+} ZDICT_fastCover_params_t;
+
+/*! ZDICT_trainFromBuffer_cover():
+ *  Train a dictionary from an array of samples using the COVER algorithm.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *          See ZDICT_trainFromBuffer() for details on failure modes.
+ *  Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_cover(
+          void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+          ZDICT_cover_params_t parameters);
+
+/*! ZDICT_optimizeTrainFromBuffer_cover():
+ * The same requirements as above hold for all the parameters except `parameters`.
+ * This function tries many parameter combinations and picks the best parameters.
+ * `*parameters` is filled with the best parameters found,
+ * dictionary constructed with those parameters is stored in `dictBuffer`.
+ *
+ * All of the parameters d, k, steps are optional.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
+ * if steps is zero it defaults to its default value.
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *          On success `*parameters` contains the parameters selected.
+ *          See ZDICT_trainFromBuffer() for details on failure modes.
+ * Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
+ */
+ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_cover(
+          void* dictBuffer, size_t dictBufferCapacity,
+    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+          ZDICT_cover_params_t* parameters);
+
+/*! ZDICT_trainFromBuffer_fastCover():
+ *  Train a dictionary from an array of samples using a modified version of COVER algorithm.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  d and k are required.
+ *  All other parameters are optional, will use default values if not provided
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *          See ZDICT_trainFromBuffer() for details on failure modes.
+ *  Note: ZDICT_trainFromBuffer_fastCover() requires 6 * 2^f bytes of memory.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
+                    size_t dictBufferCapacity, const void *samplesBuffer,
+                    const size_t *samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t parameters);
+
+/*! ZDICT_optimizeTrainFromBuffer_fastCover():
+ * The same requirements as above hold for all the parameters except `parameters`.
+ * This function tries many parameter combinations (specifically, k and d combinations)
+ * and picks the best parameters. `*parameters` is filled with the best parameters found,
+ * dictionary constructed with those parameters is stored in `dictBuffer`.
+ * All of the parameters d, k, steps, f, and accel are optional.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
+ * if steps is zero it defaults to its default value.
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
+ * If f is zero, default value of 20 is used.
+ * If accel is zero, default value of 1 is used.
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *          On success `*parameters` contains the parameters selected.
+ *          See ZDICT_trainFromBuffer() for details on failure modes.
+ * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 6 * 2^f bytes of memory for each thread.
+ */
+ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer,
+                    size_t dictBufferCapacity, const void* samplesBuffer,
+                    const size_t* samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t* parameters);
+
+typedef struct {
+    unsigned selectivityLevel;   /* 0 means default; larger => select more => larger dictionary */
+    ZDICT_params_t zParams;
+} ZDICT_legacy_params_t;
+
+/*! ZDICT_trainFromBuffer_legacy():
+ *  Train a dictionary from an array of samples.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * `parameters` is optional and can be provided with values set to 0 to mean "default".
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *          See ZDICT_trainFromBuffer() for details on failure modes.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ *  Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
+ */
+ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_legacy(
+    void* dictBuffer, size_t dictBufferCapacity,
+    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+    ZDICT_legacy_params_t parameters);
+
+
+/* Deprecation warnings */
+/* It is generally possible to disable deprecation warnings from compiler,
+   for example with -Wno-deprecated-declarations for gcc
+   or _CRT_SECURE_NO_WARNINGS in Visual.
+   Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */
+#ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS
+#  define ZDICT_DEPRECATED(message) /* disable deprecation warnings */
+#else
+#  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define ZDICT_DEPRECATED(message) [[deprecated(message)]]
+#  elif defined(__clang__) || (ZDICT_GCC_VERSION >= 405)
+#    define ZDICT_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif (ZDICT_GCC_VERSION >= 301)
+#    define ZDICT_DEPRECATED(message) __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define ZDICT_DEPRECATED(message) __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler")
+#    define ZDICT_DEPRECATED(message)
+#  endif
+#endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */
+
+ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead")
+ZDICTLIB_STATIC_API
+size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+
+
+#endif   /* ZSTD_ZDICT_H_STATIC */
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/zstd.h b/lib/clickhouse-cpp/contrib/zstd/zstd/zstd.h
new file mode 100644
index 0000000..e5c3f8b
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/zstd.h
@@ -0,0 +1,3020 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef ZSTD_H_235446
+#define ZSTD_H_235446
+
+/* ======   Dependencies   ======*/
+#include <limits.h>   /* INT_MAX */
+#include <stddef.h>   /* size_t */
+
+
+/* =====   ZSTDLIB_API : control library symbols visibility   ===== */
+#ifndef ZSTDLIB_VISIBLE
+   /* Backwards compatibility with old macro name */
+#  ifdef ZSTDLIB_VISIBILITY
+#    define ZSTDLIB_VISIBLE ZSTDLIB_VISIBILITY
+#  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDLIB_VISIBLE __attribute__ ((visibility ("default")))
+#  else
+#    define ZSTDLIB_VISIBLE
+#  endif
+#endif
+
+#ifndef ZSTDLIB_HIDDEN
+#  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDLIB_HIDDEN __attribute__ ((visibility ("hidden")))
+#  else
+#    define ZSTDLIB_HIDDEN
+#  endif
+#endif
+
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZSTDLIB_API __declspec(dllexport) ZSTDLIB_VISIBLE
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZSTDLIB_API __declspec(dllimport) ZSTDLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZSTDLIB_API ZSTDLIB_VISIBLE
+#endif
+
+/* Deprecation warnings :
+ * Should these warnings be a problem, it is generally possible to disable them,
+ * typically with -Wno-deprecated-declarations for gcc or _CRT_SECURE_NO_WARNINGS in Visual.
+ * Otherwise, it's also possible to define ZSTD_DISABLE_DEPRECATE_WARNINGS.
+ */
+#ifdef ZSTD_DISABLE_DEPRECATE_WARNINGS
+#  define ZSTD_DEPRECATED(message) /* disable deprecation warnings */
+#else
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define ZSTD_DEPRECATED(message) [[deprecated(message)]]
+#  elif (defined(GNUC) && (GNUC > 4 || (GNUC == 4 && GNUC_MINOR >= 5))) || defined(__clang__)
+#    define ZSTD_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif defined(__GNUC__) && (__GNUC__ >= 3)
+#    define ZSTD_DEPRECATED(message) __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define ZSTD_DEPRECATED(message) __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement ZSTD_DEPRECATED for this compiler")
+#    define ZSTD_DEPRECATED(message)
+#  endif
+#endif /* ZSTD_DISABLE_DEPRECATE_WARNINGS */
+
+
+/*******************************************************************************
+  Introduction
+
+  zstd, short for Zstandard, is a fast lossless compression algorithm, targeting
+  real-time compression scenarios at zlib-level and better compression ratios.
+  The zstd compression library provides in-memory compression and decompression
+  functions.
+
+  The library supports regular compression levels from 1 up to ZSTD_maxCLevel(),
+  which is currently 22. Levels >= 20, labeled `--ultra`, should be used with
+  caution, as they require more memory. The library also offers negative
+  compression levels, which extend the range of speed vs. ratio preferences.
+  The lower the level, the faster the speed (at the cost of compression).
+
+  Compression can be done in:
+    - a single step (described as Simple API)
+    - a single step, reusing a context (described as Explicit context)
+    - unbounded multiple steps (described as Streaming compression)
+
+  The compression ratio achievable on small data can be highly improved using
+  a dictionary. Dictionary compression can be performed in:
+    - a single step (described as Simple dictionary API)
+    - a single step, reusing a dictionary (described as Bulk-processing
+      dictionary API)
+
+  Advanced experimental functions can be accessed using
+  `#define ZSTD_STATIC_LINKING_ONLY` before including zstd.h.
+
+  Advanced experimental APIs should never be used with a dynamically-linked
+  library. They are not "stable"; their definitions or signatures may change in
+  the future. Only static linking is allowed.
+*******************************************************************************/
+
+/*------   Version   ------*/
+#define ZSTD_VERSION_MAJOR    1
+#define ZSTD_VERSION_MINOR    5
+#define ZSTD_VERSION_RELEASE  5
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+/*! ZSTD_versionNumber() :
+ *  Return runtime library version, the value is (MAJOR*100*100 + MINOR*100 + RELEASE). */
+ZSTDLIB_API unsigned ZSTD_versionNumber(void);
+
+#define ZSTD_LIB_VERSION ZSTD_VERSION_MAJOR.ZSTD_VERSION_MINOR.ZSTD_VERSION_RELEASE
+#define ZSTD_QUOTE(str) #str
+#define ZSTD_EXPAND_AND_QUOTE(str) ZSTD_QUOTE(str)
+#define ZSTD_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LIB_VERSION)
+
+/*! ZSTD_versionString() :
+ *  Return runtime library version, like "1.4.5". Requires v1.3.0+. */
+ZSTDLIB_API const char* ZSTD_versionString(void);
+
+/* *************************************
+ *  Default constant
+ ***************************************/
+#ifndef ZSTD_CLEVEL_DEFAULT
+#  define ZSTD_CLEVEL_DEFAULT 3
+#endif
+
+/* *************************************
+ *  Constants
+ ***************************************/
+
+/* All magic numbers are supposed read/written to/from files/memory using little-endian convention */
+#define ZSTD_MAGICNUMBER            0xFD2FB528    /* valid since v0.8.0 */
+#define ZSTD_MAGIC_DICTIONARY       0xEC30A437    /* valid since v0.7.0 */
+#define ZSTD_MAGIC_SKIPPABLE_START  0x184D2A50    /* all 16 values, from 0x184D2A50 to 0x184D2A5F, signal the beginning of a skippable frame */
+#define ZSTD_MAGIC_SKIPPABLE_MASK   0xFFFFFFF0
+
+#define ZSTD_BLOCKSIZELOG_MAX  17
+#define ZSTD_BLOCKSIZE_MAX     (1<<ZSTD_BLOCKSIZELOG_MAX)
+
+
+/***************************************
+*  Simple API
+***************************************/
+/*! ZSTD_compress() :
+ *  Compresses `src` content as a single zstd compressed frame into already allocated `dst`.
+ *  NOTE: Providing `dstCapacity >= ZSTD_compressBound(srcSize)` guarantees that zstd will have
+ *        enough space to successfully compress the data.
+ *  @return : compressed size written into `dst` (<= `dstCapacity),
+ *            or an error code if it fails (which can be tested using ZSTD_isError()). */
+ZSTDLIB_API size_t ZSTD_compress( void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                                  int compressionLevel);
+
+/*! ZSTD_decompress() :
+ *  `compressedSize` : must be the _exact_ size of some number of compressed and/or skippable frames.
+ *  `dstCapacity` is an upper bound of originalSize to regenerate.
+ *  If user cannot imply a maximum upper bound, it's better to use streaming mode to decompress data.
+ *  @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+ *            or an errorCode if it fails (which can be tested using ZSTD_isError()). */
+ZSTDLIB_API size_t ZSTD_decompress( void* dst, size_t dstCapacity,
+                              const void* src, size_t compressedSize);
+
+/*! ZSTD_getFrameContentSize() : requires v1.3.0+
+ *  `src` should point to the start of a ZSTD encoded frame.
+ *  `srcSize` must be at least as large as the frame header.
+ *            hint : any size >= `ZSTD_frameHeaderSize_max` is large enough.
+ *  @return : - decompressed size of `src` frame content, if known
+ *            - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *            - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small)
+ *   note 1 : a 0 return value means the frame is valid but "empty".
+ *   note 2 : decompressed size is an optional field, it may not be present, typically in streaming mode.
+ *            When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *            Optionally, application can rely on some implicit limit,
+ *            as ZSTD_decompress() only needs an upper bound of decompressed size.
+ *            (For example, data could be necessarily cut into blocks <= 16 KB).
+ *   note 3 : decompressed size is always present when compression is completed using single-pass functions,
+ *            such as ZSTD_compress(), ZSTD_compressCCtx() ZSTD_compress_usingDict() or ZSTD_compress_usingCDict().
+ *   note 4 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 5 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure return value fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 6 : This function replaces ZSTD_getDecompressedSize() */
+#define ZSTD_CONTENTSIZE_UNKNOWN (0ULL - 1)
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize);
+
+/*! ZSTD_getDecompressedSize() :
+ *  NOTE: This function is now obsolete, in favor of ZSTD_getFrameContentSize().
+ *  Both functions work the same way, but ZSTD_getDecompressedSize() blends
+ *  "empty", "unknown" and "error" results to the same return value (0),
+ *  while ZSTD_getFrameContentSize() gives them separate return values.
+ * @return : decompressed size of `src` frame content _if known and not empty_, 0 otherwise. */
+ZSTD_DEPRECATED("Replaced by ZSTD_getFrameContentSize")
+ZSTDLIB_API
+unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTD_findFrameCompressedSize() : Requires v1.4.0+
+ * `src` should point to the start of a ZSTD frame or skippable frame.
+ * `srcSize` must be >= first frame size
+ * @return : the compressed size of the first frame starting at `src`,
+ *           suitable to pass as `srcSize` to `ZSTD_decompress` or similar,
+ *        or an error code if input is invalid */
+ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize);
+
+
+/*======  Helper functions  ======*/
+/* ZSTD_compressBound() :
+ * maximum compressed size in worst case single-pass scenario.
+ * When invoking `ZSTD_compress()` or any other one-pass compression function,
+ * it's recommended to provide @dstCapacity >= ZSTD_compressBound(srcSize)
+ * as it eliminates one potential failure scenario,
+ * aka not enough room in dst buffer to write the compressed frame.
+ * Note : ZSTD_compressBound() itself can fail, if @srcSize > ZSTD_MAX_INPUT_SIZE .
+ *        In which case, ZSTD_compressBound() will return an error code
+ *        which can be tested using ZSTD_isError().
+ *
+ * ZSTD_COMPRESSBOUND() :
+ * same as ZSTD_compressBound(), but as a macro.
+ * It can be used to produce constants, which can be useful for static allocation,
+ * for example to size a static array on stack.
+ * Will produce constant value 0 if srcSize too large.
+ */
+#define ZSTD_MAX_INPUT_SIZE ((sizeof(size_t)==8) ? 0xFF00FF00FF00FF00LLU : 0xFF00FF00U)
+#define ZSTD_COMPRESSBOUND(srcSize)   (((size_t)(srcSize) >= ZSTD_MAX_INPUT_SIZE) ? 0 : (srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
+ZSTDLIB_API size_t ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+/* ZSTD_isError() :
+ * Most ZSTD_* functions returning a size_t value can be tested for error,
+ * using ZSTD_isError().
+ * @return 1 if error, 0 otherwise
+ */
+ZSTDLIB_API unsigned    ZSTD_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+ZSTDLIB_API const char* ZSTD_getErrorName(size_t code);     /*!< provides readable string from an error code */
+ZSTDLIB_API int         ZSTD_minCLevel(void);               /*!< minimum negative compression level allowed, requires v1.4.0+ */
+ZSTDLIB_API int         ZSTD_maxCLevel(void);               /*!< maximum compression level available */
+ZSTDLIB_API int         ZSTD_defaultCLevel(void);           /*!< default compression level, specified by ZSTD_CLEVEL_DEFAULT, requires v1.5.0+ */
+
+
+/***************************************
+*  Explicit context
+***************************************/
+/*= Compression context
+ *  When compressing many times,
+ *  it is recommended to allocate a context just once,
+ *  and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Note : re-using context is just a speed / resource optimization.
+ *         It doesn't change the compression ratio, which remains identical.
+ *  Note 2 : In multi-threaded environments,
+ *         use one different context per thread for parallel execution.
+ */
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;
+ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx(void);
+ZSTDLIB_API size_t     ZSTD_freeCCtx(ZSTD_CCtx* cctx);  /* accept NULL pointer */
+
+/*! ZSTD_compressCCtx() :
+ *  Same as ZSTD_compress(), using an explicit ZSTD_CCtx.
+ *  Important : in order to behave similarly to `ZSTD_compress()`,
+ *  this function compresses at requested compression level,
+ *  __ignoring any other parameter__ .
+ *  If any advanced parameter was set using the advanced API,
+ *  they will all be reset. Only `compressionLevel` remains.
+ */
+ZSTDLIB_API size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     int compressionLevel);
+
+/*= Decompression context
+ *  When decompressing many times,
+ *  it is recommended to allocate a context only once,
+ *  and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Use one context per thread for parallel execution. */
+typedef struct ZSTD_DCtx_s ZSTD_DCtx;
+ZSTDLIB_API ZSTD_DCtx* ZSTD_createDCtx(void);
+ZSTDLIB_API size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);  /* accept NULL pointer */
+
+/*! ZSTD_decompressDCtx() :
+ *  Same as ZSTD_decompress(),
+ *  requires an allocated ZSTD_DCtx.
+ *  Compatible with sticky parameters.
+ */
+ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize);
+
+
+/*********************************************
+*  Advanced compression API (Requires v1.4.0+)
+**********************************************/
+
+/* API design :
+ *   Parameters are pushed one by one into an existing context,
+ *   using ZSTD_CCtx_set*() functions.
+ *   Pushed parameters are sticky : they are valid for next compressed frame, and any subsequent frame.
+ *   "sticky" parameters are applicable to `ZSTD_compress2()` and `ZSTD_compressStream*()` !
+ *   __They do not apply to "simple" one-shot variants such as ZSTD_compressCCtx()__ .
+ *
+ *   It's possible to reset all parameters to "default" using ZSTD_CCtx_reset().
+ *
+ *   This API supersedes all other "advanced" API entry points in the experimental section.
+ *   In the future, we expect to remove from experimental API entry points which are redundant with this API.
+ */
+
+
+/* Compression strategies, listed from fastest to strongest */
+typedef enum { ZSTD_fast=1,
+               ZSTD_dfast=2,
+               ZSTD_greedy=3,
+               ZSTD_lazy=4,
+               ZSTD_lazy2=5,
+               ZSTD_btlazy2=6,
+               ZSTD_btopt=7,
+               ZSTD_btultra=8,
+               ZSTD_btultra2=9
+               /* note : new strategies _might_ be added in the future.
+                         Only the order (from fast to strong) is guaranteed */
+} ZSTD_strategy;
+
+typedef enum {
+
+    /* compression parameters
+     * Note: When compressing with a ZSTD_CDict these parameters are superseded
+     * by the parameters used to construct the ZSTD_CDict.
+     * See ZSTD_CCtx_refCDict() for more info (superseded-by-cdict). */
+    ZSTD_c_compressionLevel=100, /* Set compression parameters according to pre-defined cLevel table.
+                              * Note that exact compression parameters are dynamically determined,
+                              * depending on both compression level and srcSize (when known).
+                              * Default level is ZSTD_CLEVEL_DEFAULT==3.
+                              * Special: value 0 means default, which is controlled by ZSTD_CLEVEL_DEFAULT.
+                              * Note 1 : it's possible to pass a negative compression level.
+                              * Note 2 : setting a level does not automatically set all other compression parameters
+                              *   to default. Setting this will however eventually dynamically impact the compression
+                              *   parameters which have not been manually set. The manually set
+                              *   ones will 'stick'. */
+    /* Advanced compression parameters :
+     * It's possible to pin down compression parameters to some specific values.
+     * In which case, these values are no longer dynamically selected by the compressor */
+    ZSTD_c_windowLog=101,    /* Maximum allowed back-reference distance, expressed as power of 2.
+                              * This will set a memory budget for streaming decompression,
+                              * with larger values requiring more memory
+                              * and typically compressing more.
+                              * Must be clamped between ZSTD_WINDOWLOG_MIN and ZSTD_WINDOWLOG_MAX.
+                              * Special: value 0 means "use default windowLog".
+                              * Note: Using a windowLog greater than ZSTD_WINDOWLOG_LIMIT_DEFAULT
+                              *       requires explicitly allowing such size at streaming decompression stage. */
+    ZSTD_c_hashLog=102,      /* Size of the initial probe table, as a power of 2.
+                              * Resulting memory usage is (1 << (hashLog+2)).
+                              * Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX.
+                              * Larger tables improve compression ratio of strategies <= dFast,
+                              * and improve speed of strategies > dFast.
+                              * Special: value 0 means "use default hashLog". */
+    ZSTD_c_chainLog=103,     /* Size of the multi-probe search table, as a power of 2.
+                              * Resulting memory usage is (1 << (chainLog+2)).
+                              * Must be clamped between ZSTD_CHAINLOG_MIN and ZSTD_CHAINLOG_MAX.
+                              * Larger tables result in better and slower compression.
+                              * This parameter is useless for "fast" strategy.
+                              * It's still useful when using "dfast" strategy,
+                              * in which case it defines a secondary probe table.
+                              * Special: value 0 means "use default chainLog". */
+    ZSTD_c_searchLog=104,    /* Number of search attempts, as a power of 2.
+                              * More attempts result in better and slower compression.
+                              * This parameter is useless for "fast" and "dFast" strategies.
+                              * Special: value 0 means "use default searchLog". */
+    ZSTD_c_minMatch=105,     /* Minimum size of searched matches.
+                              * Note that Zstandard can still find matches of smaller size,
+                              * it just tweaks its search algorithm to look for this size and larger.
+                              * Larger values increase compression and decompression speed, but decrease ratio.
+                              * Must be clamped between ZSTD_MINMATCH_MIN and ZSTD_MINMATCH_MAX.
+                              * Note that currently, for all strategies < btopt, effective minimum is 4.
+                              *                    , for all strategies > fast, effective maximum is 6.
+                              * Special: value 0 means "use default minMatchLength". */
+    ZSTD_c_targetLength=106, /* Impact of this field depends on strategy.
+                              * For strategies btopt, btultra & btultra2:
+                              *     Length of Match considered "good enough" to stop search.
+                              *     Larger values make compression stronger, and slower.
+                              * For strategy fast:
+                              *     Distance between match sampling.
+                              *     Larger values make compression faster, and weaker.
+                              * Special: value 0 means "use default targetLength". */
+    ZSTD_c_strategy=107,     /* See ZSTD_strategy enum definition.
+                              * The higher the value of selected strategy, the more complex it is,
+                              * resulting in stronger and slower compression.
+                              * Special: value 0 means "use default strategy". */
+    /* LDM mode parameters */
+    ZSTD_c_enableLongDistanceMatching=160, /* Enable long distance matching.
+                                     * This parameter is designed to improve compression ratio
+                                     * for large inputs, by finding large matches at long distance.
+                                     * It increases memory usage and window size.
+                                     * Note: enabling this parameter increases default ZSTD_c_windowLog to 128 MB
+                                     * except when expressly set to a different value.
+                                     * Note: will be enabled by default if ZSTD_c_windowLog >= 128 MB and
+                                     * compression strategy >= ZSTD_btopt (== compression level 16+) */
+    ZSTD_c_ldmHashLog=161,   /* Size of the table for long distance matching, as a power of 2.
+                              * Larger values increase memory usage and compression ratio,
+                              * but decrease compression speed.
+                              * Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX
+                              * default: windowlog - 7.
+                              * Special: value 0 means "automatically determine hashlog". */
+    ZSTD_c_ldmMinMatch=162,  /* Minimum match size for long distance matcher.
+                              * Larger/too small values usually decrease compression ratio.
+                              * Must be clamped between ZSTD_LDM_MINMATCH_MIN and ZSTD_LDM_MINMATCH_MAX.
+                              * Special: value 0 means "use default value" (default: 64). */
+    ZSTD_c_ldmBucketSizeLog=163, /* Log size of each bucket in the LDM hash table for collision resolution.
+                              * Larger values improve collision resolution but decrease compression speed.
+                              * The maximum value is ZSTD_LDM_BUCKETSIZELOG_MAX.
+                              * Special: value 0 means "use default value" (default: 3). */
+    ZSTD_c_ldmHashRateLog=164, /* Frequency of inserting/looking up entries into the LDM hash table.
+                              * Must be clamped between 0 and (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN).
+                              * Default is MAX(0, (windowLog - ldmHashLog)), optimizing hash table usage.
+                              * Larger values improve compression speed.
+                              * Deviating far from default value will likely result in a compression ratio decrease.
+                              * Special: value 0 means "automatically determine hashRateLog". */
+
+    /* frame parameters */
+    ZSTD_c_contentSizeFlag=200, /* Content size will be written into frame header _whenever known_ (default:1)
+                              * Content size must be known at the beginning of compression.
+                              * This is automatically the case when using ZSTD_compress2(),
+                              * For streaming scenarios, content size must be provided with ZSTD_CCtx_setPledgedSrcSize() */
+    ZSTD_c_checksumFlag=201, /* A 32-bits checksum of content is written at end of frame (default:0) */
+    ZSTD_c_dictIDFlag=202,   /* When applicable, dictionary's ID is written into frame header (default:1) */
+
+    /* multi-threading parameters */
+    /* These parameters are only active if multi-threading is enabled (compiled with build macro ZSTD_MULTITHREAD).
+     * Otherwise, trying to set any other value than default (0) will be a no-op and return an error.
+     * In a situation where it's unknown if the linked library supports multi-threading or not,
+     * setting ZSTD_c_nbWorkers to any value >= 1 and consulting the return value provides a quick way to check this property.
+     */
+    ZSTD_c_nbWorkers=400,    /* Select how many threads will be spawned to compress in parallel.
+                              * When nbWorkers >= 1, triggers asynchronous mode when invoking ZSTD_compressStream*() :
+                              * ZSTD_compressStream*() consumes input and flush output if possible, but immediately gives back control to caller,
+                              * while compression is performed in parallel, within worker thread(s).
+                              * (note : a strong exception to this rule is when first invocation of ZSTD_compressStream2() sets ZSTD_e_end :
+                              *  in which case, ZSTD_compressStream2() delegates to ZSTD_compress2(), which is always a blocking call).
+                              * More workers improve speed, but also increase memory usage.
+                              * Default value is `0`, aka "single-threaded mode" : no worker is spawned,
+                              * compression is performed inside Caller's thread, and all invocations are blocking */
+    ZSTD_c_jobSize=401,      /* Size of a compression job. This value is enforced only when nbWorkers >= 1.
+                              * Each compression job is completed in parallel, so this value can indirectly impact the nb of active threads.
+                              * 0 means default, which is dynamically determined based on compression parameters.
+                              * Job size must be a minimum of overlap size, or ZSTDMT_JOBSIZE_MIN (= 512 KB), whichever is largest.
+                              * The minimum size is automatically and transparently enforced. */
+    ZSTD_c_overlapLog=402,   /* Control the overlap size, as a fraction of window size.
+                              * The overlap size is an amount of data reloaded from previous job at the beginning of a new job.
+                              * It helps preserve compression ratio, while each job is compressed in parallel.
+                              * This value is enforced only when nbWorkers >= 1.
+                              * Larger values increase compression ratio, but decrease speed.
+                              * Possible values range from 0 to 9 :
+                              * - 0 means "default" : value will be determined by the library, depending on strategy
+                              * - 1 means "no overlap"
+                              * - 9 means "full overlap", using a full window size.
+                              * Each intermediate rank increases/decreases load size by a factor 2 :
+                              * 9: full window;  8: w/2;  7: w/4;  6: w/8;  5:w/16;  4: w/32;  3:w/64;  2:w/128;  1:no overlap;  0:default
+                              * default value varies between 6 and 9, depending on strategy */
+
+    /* note : additional experimental parameters are also available
+     * within the experimental section of the API.
+     * At the time of this writing, they include :
+     * ZSTD_c_rsyncable
+     * ZSTD_c_format
+     * ZSTD_c_forceMaxWindow
+     * ZSTD_c_forceAttachDict
+     * ZSTD_c_literalCompressionMode
+     * ZSTD_c_targetCBlockSize
+     * ZSTD_c_srcSizeHint
+     * ZSTD_c_enableDedicatedDictSearch
+     * ZSTD_c_stableInBuffer
+     * ZSTD_c_stableOutBuffer
+     * ZSTD_c_blockDelimiters
+     * ZSTD_c_validateSequences
+     * ZSTD_c_useBlockSplitter
+     * ZSTD_c_useRowMatchFinder
+     * ZSTD_c_prefetchCDictTables
+     * ZSTD_c_enableSeqProducerFallback
+     * ZSTD_c_maxBlockSize
+     * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
+     * note : never ever use experimentalParam? names directly;
+     *        also, the enums values themselves are unstable and can still change.
+     */
+     ZSTD_c_experimentalParam1=500,
+     ZSTD_c_experimentalParam2=10,
+     ZSTD_c_experimentalParam3=1000,
+     ZSTD_c_experimentalParam4=1001,
+     ZSTD_c_experimentalParam5=1002,
+     ZSTD_c_experimentalParam6=1003,
+     ZSTD_c_experimentalParam7=1004,
+     ZSTD_c_experimentalParam8=1005,
+     ZSTD_c_experimentalParam9=1006,
+     ZSTD_c_experimentalParam10=1007,
+     ZSTD_c_experimentalParam11=1008,
+     ZSTD_c_experimentalParam12=1009,
+     ZSTD_c_experimentalParam13=1010,
+     ZSTD_c_experimentalParam14=1011,
+     ZSTD_c_experimentalParam15=1012,
+     ZSTD_c_experimentalParam16=1013,
+     ZSTD_c_experimentalParam17=1014,
+     ZSTD_c_experimentalParam18=1015,
+     ZSTD_c_experimentalParam19=1016
+} ZSTD_cParameter;
+
+typedef struct {
+    size_t error;
+    int lowerBound;
+    int upperBound;
+} ZSTD_bounds;
+
+/*! ZSTD_cParam_getBounds() :
+ *  All parameters must belong to an interval with lower and upper bounds,
+ *  otherwise they will either trigger an error or be automatically clamped.
+ * @return : a structure, ZSTD_bounds, which contains
+ *         - an error status field, which must be tested using ZSTD_isError()
+ *         - lower and upper bounds, both inclusive
+ */
+ZSTDLIB_API ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter cParam);
+
+/*! ZSTD_CCtx_setParameter() :
+ *  Set one compression parameter, selected by enum ZSTD_cParameter.
+ *  All parameters have valid bounds. Bounds can be queried using ZSTD_cParam_getBounds().
+ *  Providing a value beyond bound will either clamp it, or trigger an error (depending on parameter).
+ *  Setting a parameter is generally only possible during frame initialization (before starting compression).
+ *  Exception : when using multi-threading mode (nbWorkers >= 1),
+ *              the following parameters can be updated _during_ compression (within same frame):
+ *              => compressionLevel, hashLog, chainLog, searchLog, minMatch, targetLength and strategy.
+ *              new parameters will be active for next job only (after a flush()).
+ * @return : an error code (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value);
+
+/*! ZSTD_CCtx_setPledgedSrcSize() :
+ *  Total input data size to be compressed as a single frame.
+ *  Value will be written in frame header, unless if explicitly forbidden using ZSTD_c_contentSizeFlag.
+ *  This value will also be controlled at end of frame, and trigger an error if not respected.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : pledgedSrcSize==0 actually means zero, aka an empty frame.
+ *           In order to mean "unknown content size", pass constant ZSTD_CONTENTSIZE_UNKNOWN.
+ *           ZSTD_CONTENTSIZE_UNKNOWN is default value for any new frame.
+ *  Note 2 : pledgedSrcSize is only valid once, for the next frame.
+ *           It's discarded at the end of the frame, and replaced by ZSTD_CONTENTSIZE_UNKNOWN.
+ *  Note 3 : Whenever all input data is provided and consumed in a single round,
+ *           for example with ZSTD_compress2(),
+ *           or invoking immediately ZSTD_compressStream2(,,,ZSTD_e_end),
+ *           this value is automatically overridden by srcSize instead.
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize);
+
+typedef enum {
+    ZSTD_reset_session_only = 1,
+    ZSTD_reset_parameters = 2,
+    ZSTD_reset_session_and_parameters = 3
+} ZSTD_ResetDirective;
+
+/*! ZSTD_CCtx_reset() :
+ *  There are 2 different things that can be reset, independently or jointly :
+ *  - The session : will stop compressing current frame, and make CCtx ready to start a new one.
+ *                  Useful after an error, or to interrupt any ongoing compression.
+ *                  Any internal data not yet flushed is cancelled.
+ *                  Compression parameters and dictionary remain unchanged.
+ *                  They will be used to compress next frame.
+ *                  Resetting session never fails.
+ *  - The parameters : changes all parameters back to "default".
+ *                  This also removes any reference to any dictionary or external sequence producer.
+ *                  Parameters can only be changed between 2 sessions (i.e. no compression is currently ongoing)
+ *                  otherwise the reset fails, and function returns an error value (which can be tested using ZSTD_isError())
+ *  - Both : similar to resetting the session, followed by resetting parameters.
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset);
+
+/*! ZSTD_compress2() :
+ *  Behave the same as ZSTD_compressCCtx(), but compression parameters are set using the advanced API.
+ *  ZSTD_compress2() always starts a new frame.
+ *  Should cctx hold data from a previously unfinished frame, everything about it is forgotten.
+ *  - Compression parameters are pushed into CCtx before starting compression, using ZSTD_CCtx_set*()
+ *  - The function is always blocking, returns when compression is completed.
+ *  NOTE: Providing `dstCapacity >= ZSTD_compressBound(srcSize)` guarantees that zstd will have
+ *        enough space to successfully compress the data, though it is possible it fails for other reasons.
+ * @return : compressed size written into `dst` (<= `dstCapacity),
+ *           or an error code if it fails (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_compress2( ZSTD_CCtx* cctx,
+                                   void* dst, size_t dstCapacity,
+                             const void* src, size_t srcSize);
+
+
+/***********************************************
+*  Advanced decompression API (Requires v1.4.0+)
+************************************************/
+
+/* The advanced API pushes parameters one by one into an existing DCtx context.
+ * Parameters are sticky, and remain valid for all following frames
+ * using the same DCtx context.
+ * It's possible to reset parameters to default values using ZSTD_DCtx_reset().
+ * Note : This API is compatible with existing ZSTD_decompressDCtx() and ZSTD_decompressStream().
+ *        Therefore, no new decompression function is necessary.
+ */
+
+typedef enum {
+
+    ZSTD_d_windowLogMax=100, /* Select a size limit (in power of 2) beyond which
+                              * the streaming API will refuse to allocate memory buffer
+                              * in order to protect the host from unreasonable memory requirements.
+                              * This parameter is only useful in streaming mode, since no internal buffer is allocated in single-pass mode.
+                              * By default, a decompression context accepts window sizes <= (1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT).
+                              * Special: value 0 means "use default maximum windowLog". */
+
+    /* note : additional experimental parameters are also available
+     * within the experimental section of the API.
+     * At the time of this writing, they include :
+     * ZSTD_d_format
+     * ZSTD_d_stableOutBuffer
+     * ZSTD_d_forceIgnoreChecksum
+     * ZSTD_d_refMultipleDDicts
+     * ZSTD_d_disableHuffmanAssembly
+     * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
+     * note : never ever use experimentalParam? names directly
+     */
+     ZSTD_d_experimentalParam1=1000,
+     ZSTD_d_experimentalParam2=1001,
+     ZSTD_d_experimentalParam3=1002,
+     ZSTD_d_experimentalParam4=1003,
+     ZSTD_d_experimentalParam5=1004
+
+} ZSTD_dParameter;
+
+/*! ZSTD_dParam_getBounds() :
+ *  All parameters must belong to an interval with lower and upper bounds,
+ *  otherwise they will either trigger an error or be automatically clamped.
+ * @return : a structure, ZSTD_bounds, which contains
+ *         - an error status field, which must be tested using ZSTD_isError()
+ *         - both lower and upper bounds, inclusive
+ */
+ZSTDLIB_API ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam);
+
+/*! ZSTD_DCtx_setParameter() :
+ *  Set one compression parameter, selected by enum ZSTD_dParameter.
+ *  All parameters have valid bounds. Bounds can be queried using ZSTD_dParam_getBounds().
+ *  Providing a value beyond bound will either clamp it, or trigger an error (depending on parameter).
+ *  Setting a parameter is only possible during frame initialization (before starting decompression).
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int value);
+
+/*! ZSTD_DCtx_reset() :
+ *  Return a DCtx to clean state.
+ *  Session and parameters can be reset jointly or separately.
+ *  Parameters can only be reset when no active frame is being decompressed.
+ * @return : 0, or an error code, which can be tested with ZSTD_isError()
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset);
+
+
+/****************************
+*  Streaming
+****************************/
+
+typedef struct ZSTD_inBuffer_s {
+  const void* src;    /**< start of input buffer */
+  size_t size;        /**< size of input buffer */
+  size_t pos;         /**< position where reading stopped. Will be updated. Necessarily 0 <= pos <= size */
+} ZSTD_inBuffer;
+
+typedef struct ZSTD_outBuffer_s {
+  void*  dst;         /**< start of output buffer */
+  size_t size;        /**< size of output buffer */
+  size_t pos;         /**< position where writing stopped. Will be updated. Necessarily 0 <= pos <= size */
+} ZSTD_outBuffer;
+
+
+
+/*-***********************************************************************
+*  Streaming compression - HowTo
+*
+*  A ZSTD_CStream object is required to track streaming operation.
+*  Use ZSTD_createCStream() and ZSTD_freeCStream() to create/release resources.
+*  ZSTD_CStream objects can be reused multiple times on consecutive compression operations.
+*  It is recommended to re-use ZSTD_CStream since it will play nicer with system's memory, by re-using already allocated memory.
+*
+*  For parallel execution, use one separate ZSTD_CStream per thread.
+*
+*  note : since v1.3.0, ZSTD_CStream and ZSTD_CCtx are the same thing.
+*
+*  Parameters are sticky : when starting a new compression on the same context,
+*  it will re-use the same sticky parameters as previous compression session.
+*  When in doubt, it's recommended to fully initialize the context before usage.
+*  Use ZSTD_CCtx_reset() to reset the context and ZSTD_CCtx_setParameter(),
+*  ZSTD_CCtx_setPledgedSrcSize(), or ZSTD_CCtx_loadDictionary() and friends to
+*  set more specific parameters, the pledged source size, or load a dictionary.
+*
+*  Use ZSTD_compressStream2() with ZSTD_e_continue as many times as necessary to
+*  consume input stream. The function will automatically update both `pos`
+*  fields within `input` and `output`.
+*  Note that the function may not consume the entire input, for example, because
+*  the output buffer is already full, in which case `input.pos < input.size`.
+*  The caller must check if input has been entirely consumed.
+*  If not, the caller must make some room to receive more compressed data,
+*  and then present again remaining input data.
+*  note: ZSTD_e_continue is guaranteed to make some forward progress when called,
+*        but doesn't guarantee maximal forward progress. This is especially relevant
+*        when compressing with multiple threads. The call won't block if it can
+*        consume some input, but if it can't it will wait for some, but not all,
+*        output to be flushed.
+* @return : provides a minimum amount of data remaining to be flushed from internal buffers
+*           or an error code, which can be tested using ZSTD_isError().
+*
+*  At any moment, it's possible to flush whatever data might remain stuck within internal buffer,
+*  using ZSTD_compressStream2() with ZSTD_e_flush. `output->pos` will be updated.
+*  Note that, if `output->size` is too small, a single invocation with ZSTD_e_flush might not be enough (return code > 0).
+*  In which case, make some room to receive more compressed data, and call again ZSTD_compressStream2() with ZSTD_e_flush.
+*  You must continue calling ZSTD_compressStream2() with ZSTD_e_flush until it returns 0, at which point you can change the
+*  operation.
+*  note: ZSTD_e_flush will flush as much output as possible, meaning when compressing with multiple threads, it will
+*        block until the flush is complete or the output buffer is full.
+*  @return : 0 if internal buffers are entirely flushed,
+*            >0 if some data still present within internal buffer (the value is minimal estimation of remaining size),
+*            or an error code, which can be tested using ZSTD_isError().
+*
+*  Calling ZSTD_compressStream2() with ZSTD_e_end instructs to finish a frame.
+*  It will perform a flush and write frame epilogue.
+*  The epilogue is required for decoders to consider a frame completed.
+*  flush operation is the same, and follows same rules as calling ZSTD_compressStream2() with ZSTD_e_flush.
+*  You must continue calling ZSTD_compressStream2() with ZSTD_e_end until it returns 0, at which point you are free to
+*  start a new frame.
+*  note: ZSTD_e_end will flush as much output as possible, meaning when compressing with multiple threads, it will
+*        block until the flush is complete or the output buffer is full.
+*  @return : 0 if frame fully completed and fully flushed,
+*            >0 if some data still present within internal buffer (the value is minimal estimation of remaining size),
+*            or an error code, which can be tested using ZSTD_isError().
+*
+* *******************************************************************/
+
+typedef ZSTD_CCtx ZSTD_CStream;  /**< CCtx and CStream are now effectively same object (>= v1.3.0) */
+                                 /* Continue to distinguish them for compatibility with older versions <= v1.2.0 */
+/*===== ZSTD_CStream management functions =====*/
+ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream(void);
+ZSTDLIB_API size_t ZSTD_freeCStream(ZSTD_CStream* zcs);  /* accept NULL pointer */
+
+/*===== Streaming compression functions =====*/
+typedef enum {
+    ZSTD_e_continue=0, /* collect more data, encoder decides when to output compressed result, for optimal compression ratio */
+    ZSTD_e_flush=1,    /* flush any data provided so far,
+                        * it creates (at least) one new block, that can be decoded immediately on reception;
+                        * frame will continue: any future data can still reference previously compressed data, improving compression.
+                        * note : multithreaded compression will block to flush as much output as possible. */
+    ZSTD_e_end=2       /* flush any remaining data _and_ close current frame.
+                        * note that frame is only closed after compressed data is fully flushed (return value == 0).
+                        * After that point, any additional data starts a new frame.
+                        * note : each frame is independent (does not reference any content from previous frame).
+                        : note : multithreaded compression will block to flush as much output as possible. */
+} ZSTD_EndDirective;
+
+/*! ZSTD_compressStream2() : Requires v1.4.0+
+ *  Behaves about the same as ZSTD_compressStream, with additional control on end directive.
+ *  - Compression parameters are pushed into CCtx before starting compression, using ZSTD_CCtx_set*()
+ *  - Compression parameters cannot be changed once compression is started (save a list of exceptions in multi-threading mode)
+ *  - output->pos must be <= dstCapacity, input->pos must be <= srcSize
+ *  - output->pos and input->pos will be updated. They are guaranteed to remain below their respective limit.
+ *  - endOp must be a valid directive
+ *  - When nbWorkers==0 (default), function is blocking : it completes its job before returning to caller.
+ *  - When nbWorkers>=1, function is non-blocking : it copies a portion of input, distributes jobs to internal worker threads, flush to output whatever is available,
+ *                                                  and then immediately returns, just indicating that there is some data remaining to be flushed.
+ *                                                  The function nonetheless guarantees forward progress : it will return only after it reads or write at least 1+ byte.
+ *  - Exception : if the first call requests a ZSTD_e_end directive and provides enough dstCapacity, the function delegates to ZSTD_compress2() which is always blocking.
+ *  - @return provides a minimum amount of data remaining to be flushed from internal buffers
+ *            or an error code, which can be tested using ZSTD_isError().
+ *            if @return != 0, flush is not fully completed, there is still some data left within internal buffers.
+ *            This is useful for ZSTD_e_flush, since in this case more flushes are necessary to empty all buffers.
+ *            For ZSTD_e_end, @return == 0 when internal buffers are fully flushed and frame is completed.
+ *  - after a ZSTD_e_end directive, if internal buffer is not fully flushed (@return != 0),
+ *            only ZSTD_e_end or ZSTD_e_flush operations are allowed.
+ *            Before starting a new compression job, or changing compression parameters,
+ *            it is required to fully flush internal buffers.
+ */
+ZSTDLIB_API size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
+                                         ZSTD_outBuffer* output,
+                                         ZSTD_inBuffer* input,
+                                         ZSTD_EndDirective endOp);
+
+
+/* These buffer sizes are softly recommended.
+ * They are not required : ZSTD_compressStream*() happily accepts any buffer size, for both input and output.
+ * Respecting the recommended size just makes it a bit easier for ZSTD_compressStream*(),
+ * reducing the amount of memory shuffling and buffering, resulting in minor performance savings.
+ *
+ * However, note that these recommendations are from the perspective of a C caller program.
+ * If the streaming interface is invoked from some other language,
+ * especially managed ones such as Java or Go, through a foreign function interface such as jni or cgo,
+ * a major performance rule is to reduce crossing such interface to an absolute minimum.
+ * It's not rare that performance ends being spent more into the interface, rather than compression itself.
+ * In which cases, prefer using large buffers, as large as practical,
+ * for both input and output, to reduce the nb of roundtrips.
+ */
+ZSTDLIB_API size_t ZSTD_CStreamInSize(void);    /**< recommended size for input buffer */
+ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output buffer. Guarantee to successfully flush at least one complete compressed block. */
+
+
+/* *****************************************************************************
+ * This following is a legacy streaming API, available since v1.0+ .
+ * It can be replaced by ZSTD_CCtx_reset() and ZSTD_compressStream2().
+ * It is redundant, but remains fully supported.
+ ******************************************************************************/
+
+/*!
+ * Equivalent to:
+ *
+ *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+ *     ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
+ *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
+ *
+ * Note that ZSTD_initCStream() clears any previously set dictionary. Use the new API
+ * to compress with a dictionary.
+ */
+ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
+/*!
+ * Alternative for ZSTD_compressStream2(zcs, output, input, ZSTD_e_continue).
+ * NOTE: The return value is different. ZSTD_compressStream() returns a hint for
+ * the next read size (if non-zero and not an error). ZSTD_compressStream2()
+ * returns the minimum nb of bytes left to flush (if non-zero and not an error).
+ */
+ZSTDLIB_API size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
+/*! Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_flush). */
+ZSTDLIB_API size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
+/*! Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_end). */
+ZSTDLIB_API size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
+
+
+/*-***************************************************************************
+*  Streaming decompression - HowTo
+*
+*  A ZSTD_DStream object is required to track streaming operations.
+*  Use ZSTD_createDStream() and ZSTD_freeDStream() to create/release resources.
+*  ZSTD_DStream objects can be re-used multiple times.
+*
+*  Use ZSTD_initDStream() to start a new decompression operation.
+* @return : recommended first input size
+*  Alternatively, use advanced API to set specific properties.
+*
+*  Use ZSTD_decompressStream() repetitively to consume your input.
+*  The function will update both `pos` fields.
+*  If `input.pos < input.size`, some input has not been consumed.
+*  It's up to the caller to present again remaining data.
+*  The function tries to flush all data decoded immediately, respecting output buffer size.
+*  If `output.pos < output.size`, decoder has flushed everything it could.
+*  But if `output.pos == output.size`, there might be some data left within internal buffers.,
+*  In which case, call ZSTD_decompressStream() again to flush whatever remains in the buffer.
+*  Note : with no additional input provided, amount of data flushed is necessarily <= ZSTD_BLOCKSIZE_MAX.
+* @return : 0 when a frame is completely decoded and fully flushed,
+*        or an error code, which can be tested using ZSTD_isError(),
+*        or any other value > 0, which means there is still some decoding or flushing to do to complete current frame :
+*                                the return value is a suggested next input size (just a hint for better latency)
+*                                that will never request more than the remaining frame size.
+* *******************************************************************************/
+
+typedef ZSTD_DCtx ZSTD_DStream;  /**< DCtx and DStream are now effectively same object (>= v1.3.0) */
+                                 /* For compatibility with versions <= v1.2.0, prefer differentiating them. */
+/*===== ZSTD_DStream management functions =====*/
+ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream(void);
+ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds);  /* accept NULL pointer */
+
+/*===== Streaming decompression functions =====*/
+
+/*! ZSTD_initDStream() :
+ * Initialize/reset DStream state for new decompression operation.
+ * Call before new decompression operation using same DStream.
+ *
+ * Note : This function is redundant with the advanced API and equivalent to:
+ *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
+ *     ZSTD_DCtx_refDDict(zds, NULL);
+ */
+ZSTDLIB_API size_t ZSTD_initDStream(ZSTD_DStream* zds);
+
+/*! ZSTD_decompressStream() :
+ * Streaming decompression function.
+ * Call repetitively to consume full input updating it as necessary.
+ * Function will update both input and output `pos` fields exposing current state via these fields:
+ * - `input.pos < input.size`, some input remaining and caller should provide remaining input
+ *   on the next call.
+ * - `output.pos < output.size`, decoder finished and flushed all remaining buffers.
+ * - `output.pos == output.size`, potentially uncflushed data present in the internal buffers,
+ *   call ZSTD_decompressStream() again to flush remaining data to output.
+ * Note : with no additional input, amount of data flushed <= ZSTD_BLOCKSIZE_MAX.
+ *
+ * @return : 0 when a frame is completely decoded and fully flushed,
+ *           or an error code, which can be tested using ZSTD_isError(),
+ *           or any other value > 0, which means there is some decoding or flushing to do to complete current frame.
+ */
+ZSTDLIB_API size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
+
+ZSTDLIB_API size_t ZSTD_DStreamInSize(void);    /*!< recommended size for input buffer */
+ZSTDLIB_API size_t ZSTD_DStreamOutSize(void);   /*!< recommended size for output buffer. Guarantee to successfully flush at least one complete block in all circumstances. */
+
+
+/**************************
+*  Simple dictionary API
+***************************/
+/*! ZSTD_compress_usingDict() :
+ *  Compression at an explicit compression level using a Dictionary.
+ *  A dictionary can be any arbitrary data segment (also called a prefix),
+ *  or a buffer with specified information (see zdict.h).
+ *  Note : This function loads the dictionary, resulting in significant startup delay.
+ *         It's intended for a dictionary used only once.
+ *  Note 2 : When `dict == NULL || dictSize < 8` no dictionary is used. */
+ZSTDLIB_API size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     const void* dict,size_t dictSize,
+                                           int compressionLevel);
+
+/*! ZSTD_decompress_usingDict() :
+ *  Decompression using a known Dictionary.
+ *  Dictionary must be identical to the one used during compression.
+ *  Note : This function loads the dictionary, resulting in significant startup delay.
+ *         It's intended for a dictionary used only once.
+ *  Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
+ZSTDLIB_API size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
+                                             void* dst, size_t dstCapacity,
+                                       const void* src, size_t srcSize,
+                                       const void* dict,size_t dictSize);
+
+
+/***********************************
+ *  Bulk processing dictionary API
+ **********************************/
+typedef struct ZSTD_CDict_s ZSTD_CDict;
+
+/*! ZSTD_createCDict() :
+ *  When compressing multiple messages or blocks using the same dictionary,
+ *  it's recommended to digest the dictionary only once, since it's a costly operation.
+ *  ZSTD_createCDict() will create a state from digesting a dictionary.
+ *  The resulting state can be used for future compression operations with very limited startup cost.
+ *  ZSTD_CDict can be created once and shared by multiple threads concurrently, since its usage is read-only.
+ * @dictBuffer can be released after ZSTD_CDict creation, because its content is copied within CDict.
+ *  Note 1 : Consider experimental function `ZSTD_createCDict_byReference()` if you prefer to not duplicate @dictBuffer content.
+ *  Note 2 : A ZSTD_CDict can be created from an empty @dictBuffer,
+ *      in which case the only thing that it transports is the @compressionLevel.
+ *      This can be useful in a pipeline featuring ZSTD_compress_usingCDict() exclusively,
+ *      expecting a ZSTD_CDict parameter with any data, including those without a known dictionary. */
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize,
+                                         int compressionLevel);
+
+/*! ZSTD_freeCDict() :
+ *  Function frees memory allocated by ZSTD_createCDict().
+ *  If a NULL pointer is passed, no operation is performed. */
+ZSTDLIB_API size_t      ZSTD_freeCDict(ZSTD_CDict* CDict);
+
+/*! ZSTD_compress_usingCDict() :
+ *  Compression using a digested Dictionary.
+ *  Recommended when same dictionary is used multiple times.
+ *  Note : compression level is _decided at dictionary creation time_,
+ *     and frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) */
+ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
+                                            void* dst, size_t dstCapacity,
+                                      const void* src, size_t srcSize,
+                                      const ZSTD_CDict* cdict);
+
+
+typedef struct ZSTD_DDict_s ZSTD_DDict;
+
+/*! ZSTD_createDDict() :
+ *  Create a digested dictionary, ready to start decompression operation without startup delay.
+ *  dictBuffer can be released after DDict creation, as its content is copied inside DDict. */
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict(const void* dictBuffer, size_t dictSize);
+
+/*! ZSTD_freeDDict() :
+ *  Function frees memory allocated with ZSTD_createDDict()
+ *  If a NULL pointer is passed, no operation is performed. */
+ZSTDLIB_API size_t      ZSTD_freeDDict(ZSTD_DDict* ddict);
+
+/*! ZSTD_decompress_usingDDict() :
+ *  Decompression using a digested Dictionary.
+ *  Recommended when same dictionary is used multiple times. */
+ZSTDLIB_API size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
+                                              void* dst, size_t dstCapacity,
+                                        const void* src, size_t srcSize,
+                                        const ZSTD_DDict* ddict);
+
+
+/********************************
+ *  Dictionary helper functions
+ *******************************/
+
+/*! ZSTD_getDictID_fromDict() : Requires v1.4.0+
+ *  Provides the dictID stored within dictionary.
+ *  if @return == 0, the dictionary is not conformant with Zstandard specification.
+ *  It can still be loaded, but as a content-only dictionary. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize);
+
+/*! ZSTD_getDictID_fromCDict() : Requires v1.5.0+
+ *  Provides the dictID of the dictionary loaded into `cdict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromCDict(const ZSTD_CDict* cdict);
+
+/*! ZSTD_getDictID_fromDDict() : Requires v1.4.0+
+ *  Provides the dictID of the dictionary loaded into `ddict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict);
+
+/*! ZSTD_getDictID_fromFrame() : Requires v1.4.0+
+ *  Provides the dictID required to decompressed the frame stored within `src`.
+ *  If @return == 0, the dictID could not be decoded.
+ *  This could for one of the following reasons :
+ *  - The frame does not require a dictionary to be decoded (most common case).
+ *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden piece of information.
+ *    Note : this use case also happens when using a non-conformant dictionary.
+ *  - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
+ *  - This is not a Zstandard frame.
+ *  When identifying the exact failure cause, it's possible to use ZSTD_getFrameHeader(), which will provide a more precise error code. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
+
+
+/*******************************************************************************
+ * Advanced dictionary and prefix API (Requires v1.4.0+)
+ *
+ * This API allows dictionaries to be used with ZSTD_compress2(),
+ * ZSTD_compressStream2(), and ZSTD_decompressDCtx().
+ * Dictionaries are sticky, they remain valid when same context is re-used,
+ * they only reset when the context is reset
+ * with ZSTD_reset_parameters or ZSTD_reset_session_and_parameters.
+ * In contrast, Prefixes are single-use.
+ ******************************************************************************/
+
+
+/*! ZSTD_CCtx_loadDictionary() : Requires v1.4.0+
+ *  Create an internal CDict from `dict` buffer.
+ *  Decompression will have to use same dictionary.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special: Loading a NULL (or 0-size) dictionary invalidates previous dictionary,
+ *           meaning "return to no-dictionary mode".
+ *  Note 1 : Dictionary is sticky, it will be used for all future compressed frames,
+ *           until parameters are reset, a new dictionary is loaded, or the dictionary
+ *           is explicitly invalidated by loading a NULL dictionary.
+ *  Note 2 : Loading a dictionary involves building tables.
+ *           It's also a CPU consuming operation, with non-negligible impact on latency.
+ *           Tables are dependent on compression parameters, and for this reason,
+ *           compression parameters can no longer be changed after loading a dictionary.
+ *  Note 3 :`dict` content will be copied internally.
+ *           Use experimental ZSTD_CCtx_loadDictionary_byReference() to reference content instead.
+ *           In such a case, dictionary buffer must outlive its users.
+ *  Note 4 : Use ZSTD_CCtx_loadDictionary_advanced()
+ *           to precisely select how dictionary content must be interpreted.
+ *  Note 5 : This method does not benefit from LDM (long distance mode).
+ *           If you want to employ LDM on some large dictionary content,
+ *           prefer employing ZSTD_CCtx_refPrefix() described below.
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+
+/*! ZSTD_CCtx_refCDict() : Requires v1.4.0+
+ *  Reference a prepared dictionary, to be used for all future compressed frames.
+ *  Note that compression parameters are enforced from within CDict,
+ *  and supersede any compression parameter previously set within CCtx.
+ *  The parameters ignored are labelled as "superseded-by-cdict" in the ZSTD_cParameter enum docs.
+ *  The ignored parameters will be used again if the CCtx is returned to no-dictionary mode.
+ *  The dictionary will remain valid for future compressed frames using same CCtx.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special : Referencing a NULL CDict means "return to no-dictionary mode".
+ *  Note 1 : Currently, only one dictionary can be managed.
+ *           Referencing a new dictionary effectively "discards" any previous one.
+ *  Note 2 : CDict is just referenced, its lifetime must outlive its usage within CCtx. */
+ZSTDLIB_API size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
+
+/*! ZSTD_CCtx_refPrefix() : Requires v1.4.0+
+ *  Reference a prefix (single-usage dictionary) for next compressed frame.
+ *  A prefix is **only used once**. Tables are discarded at end of frame (ZSTD_e_end).
+ *  Decompression will need same prefix to properly regenerate data.
+ *  Compressing with a prefix is similar in outcome as performing a diff and compressing it,
+ *  but performs much faster, especially during decompression (compression speed is tunable with compression level).
+ *  This method is compatible with LDM (long distance mode).
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
+ *  Note 1 : Prefix buffer is referenced. It **must** outlive compression.
+ *           Its content must remain unmodified during compression.
+ *  Note 2 : If the intention is to diff some large src data blob with some prior version of itself,
+ *           ensure that the window size is large enough to contain the entire source.
+ *           See ZSTD_c_windowLog.
+ *  Note 3 : Referencing a prefix involves building tables, which are dependent on compression parameters.
+ *           It's a CPU consuming operation, with non-negligible impact on latency.
+ *           If there is a need to use the same prefix multiple times, consider loadDictionary instead.
+ *  Note 4 : By default, the prefix is interpreted as raw content (ZSTD_dct_rawContent).
+ *           Use experimental ZSTD_CCtx_refPrefix_advanced() to alter dictionary interpretation. */
+ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx,
+                                 const void* prefix, size_t prefixSize);
+
+/*! ZSTD_DCtx_loadDictionary() : Requires v1.4.0+
+ *  Create an internal DDict from dict buffer, to be used to decompress all future frames.
+ *  The dictionary remains valid for all future frames, until explicitly invalidated, or
+ *  a new dictionary is loaded.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special : Adding a NULL (or 0-size) dictionary invalidates any previous dictionary,
+ *            meaning "return to no-dictionary mode".
+ *  Note 1 : Loading a dictionary involves building tables,
+ *           which has a non-negligible impact on CPU usage and latency.
+ *           It's recommended to "load once, use many times", to amortize the cost
+ *  Note 2 :`dict` content will be copied internally, so `dict` can be released after loading.
+ *           Use ZSTD_DCtx_loadDictionary_byReference() to reference dictionary content instead.
+ *  Note 3 : Use ZSTD_DCtx_loadDictionary_advanced() to take control of
+ *           how dictionary content is loaded and interpreted.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+
+/*! ZSTD_DCtx_refDDict() : Requires v1.4.0+
+ *  Reference a prepared dictionary, to be used to decompress next frames.
+ *  The dictionary remains active for decompression of future frames using same DCtx.
+ *
+ *  If called with ZSTD_d_refMultipleDDicts enabled, repeated calls of this function
+ *  will store the DDict references in a table, and the DDict used for decompression
+ *  will be determined at decompression time, as per the dict ID in the frame.
+ *  The memory for the table is allocated on the first call to refDDict, and can be
+ *  freed with ZSTD_freeDCtx().
+ *
+ *  If called with ZSTD_d_refMultipleDDicts disabled (the default), only one dictionary
+ *  will be managed, and referencing a dictionary effectively "discards" any previous one.
+ *
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special: referencing a NULL DDict means "return to no-dictionary mode".
+ *  Note 2 : DDict is just referenced, its lifetime must outlive its usage from DCtx.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
+/*! ZSTD_DCtx_refPrefix() : Requires v1.4.0+
+ *  Reference a prefix (single-usage dictionary) to decompress next frame.
+ *  This is the reverse operation of ZSTD_CCtx_refPrefix(),
+ *  and must use the same prefix as the one used during compression.
+ *  Prefix is **only used once**. Reference is discarded at end of frame.
+ *  End of frame is reached when ZSTD_decompressStream() returns 0.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : Adding any prefix (including NULL) invalidates any previously set prefix or dictionary
+ *  Note 2 : Prefix buffer is referenced. It **must** outlive decompression.
+ *           Prefix buffer must remain unmodified up to the end of frame,
+ *           reached when ZSTD_decompressStream() returns 0.
+ *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dct_rawContent).
+ *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode (Experimental section)
+ *  Note 4 : Referencing a raw content prefix has almost no cpu nor memory cost.
+ *           A full dictionary is more costly, as it requires building tables.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx,
+                                 const void* prefix, size_t prefixSize);
+
+/* ===   Memory management   === */
+
+/*! ZSTD_sizeof_*() : Requires v1.4.0+
+ *  These functions give the _current_ memory usage of selected object.
+ *  Note that object memory usage can evolve (increase or decrease) over time. */
+ZSTDLIB_API size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx);
+ZSTDLIB_API size_t ZSTD_sizeof_DCtx(const ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs);
+ZSTDLIB_API size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds);
+ZSTDLIB_API size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict);
+ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
+
+#endif  /* ZSTD_H_235446 */
+
+
+/* **************************************************************************************
+ *   ADVANCED AND EXPERIMENTAL FUNCTIONS
+ ****************************************************************************************
+ * The definitions in the following section are considered experimental.
+ * They are provided for advanced scenarios.
+ * They should never be used with a dynamic library, as prototypes may change in the future.
+ * Use them only in association with static linking.
+ * ***************************************************************************************/
+
+#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
+#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
+
+/* This can be overridden externally to hide static symbols. */
+#ifndef ZSTDLIB_STATIC_API
+#  if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#    define ZSTDLIB_STATIC_API __declspec(dllexport) ZSTDLIB_VISIBLE
+#  elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#    define ZSTDLIB_STATIC_API __declspec(dllimport) ZSTDLIB_VISIBLE
+#  else
+#    define ZSTDLIB_STATIC_API ZSTDLIB_VISIBLE
+#  endif
+#endif
+
+/****************************************************************************************
+ *   experimental API (static linking only)
+ ****************************************************************************************
+ * The following symbols and constants
+ * are not planned to join "stable API" status in the near future.
+ * They can still change in future versions.
+ * Some of them are planned to remain in the static_only section indefinitely.
+ * Some of them might be removed in the future (especially when redundant with existing stable functions)
+ * ***************************************************************************************/
+
+#define ZSTD_FRAMEHEADERSIZE_PREFIX(format) ((format) == ZSTD_f_zstd1 ? 5 : 1)   /* minimum input size required to query frame header size */
+#define ZSTD_FRAMEHEADERSIZE_MIN(format)    ((format) == ZSTD_f_zstd1 ? 6 : 2)
+#define ZSTD_FRAMEHEADERSIZE_MAX   18   /* can be useful for static allocation */
+#define ZSTD_SKIPPABLEHEADERSIZE    8
+
+/* compression parameter bounds */
+#define ZSTD_WINDOWLOG_MAX_32    30
+#define ZSTD_WINDOWLOG_MAX_64    31
+#define ZSTD_WINDOWLOG_MAX     ((int)(sizeof(size_t) == 4 ? ZSTD_WINDOWLOG_MAX_32 : ZSTD_WINDOWLOG_MAX_64))
+#define ZSTD_WINDOWLOG_MIN       10
+#define ZSTD_HASHLOG_MAX       ((ZSTD_WINDOWLOG_MAX < 30) ? ZSTD_WINDOWLOG_MAX : 30)
+#define ZSTD_HASHLOG_MIN          6
+#define ZSTD_CHAINLOG_MAX_32     29
+#define ZSTD_CHAINLOG_MAX_64     30
+#define ZSTD_CHAINLOG_MAX      ((int)(sizeof(size_t) == 4 ? ZSTD_CHAINLOG_MAX_32 : ZSTD_CHAINLOG_MAX_64))
+#define ZSTD_CHAINLOG_MIN        ZSTD_HASHLOG_MIN
+#define ZSTD_SEARCHLOG_MAX      (ZSTD_WINDOWLOG_MAX-1)
+#define ZSTD_SEARCHLOG_MIN        1
+#define ZSTD_MINMATCH_MAX         7   /* only for ZSTD_fast, other strategies are limited to 6 */
+#define ZSTD_MINMATCH_MIN         3   /* only for ZSTD_btopt+, faster strategies are limited to 4 */
+#define ZSTD_TARGETLENGTH_MAX    ZSTD_BLOCKSIZE_MAX
+#define ZSTD_TARGETLENGTH_MIN     0   /* note : comparing this constant to an unsigned results in a tautological test */
+#define ZSTD_STRATEGY_MIN        ZSTD_fast
+#define ZSTD_STRATEGY_MAX        ZSTD_btultra2
+#define ZSTD_BLOCKSIZE_MAX_MIN (1 << 10) /* The minimum valid max blocksize. Maximum blocksizes smaller than this make compressBound() inaccurate. */
+
+
+#define ZSTD_OVERLAPLOG_MIN       0
+#define ZSTD_OVERLAPLOG_MAX       9
+
+#define ZSTD_WINDOWLOG_LIMIT_DEFAULT 27   /* by default, the streaming decoder will refuse any frame
+                                           * requiring larger than (1<<ZSTD_WINDOWLOG_LIMIT_DEFAULT) window size,
+                                           * to preserve host's memory from unreasonable requirements.
+                                           * This limit can be overridden using ZSTD_DCtx_setParameter(,ZSTD_d_windowLogMax,).
+                                           * The limit does not apply for one-pass decoders (such as ZSTD_decompress()), since no additional memory is allocated */
+
+
+/* LDM parameter bounds */
+#define ZSTD_LDM_HASHLOG_MIN      ZSTD_HASHLOG_MIN
+#define ZSTD_LDM_HASHLOG_MAX      ZSTD_HASHLOG_MAX
+#define ZSTD_LDM_MINMATCH_MIN        4
+#define ZSTD_LDM_MINMATCH_MAX     4096
+#define ZSTD_LDM_BUCKETSIZELOG_MIN   1
+#define ZSTD_LDM_BUCKETSIZELOG_MAX   8
+#define ZSTD_LDM_HASHRATELOG_MIN     0
+#define ZSTD_LDM_HASHRATELOG_MAX (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN)
+
+/* Advanced parameter bounds */
+#define ZSTD_TARGETCBLOCKSIZE_MIN   64
+#define ZSTD_TARGETCBLOCKSIZE_MAX   ZSTD_BLOCKSIZE_MAX
+#define ZSTD_SRCSIZEHINT_MIN        0
+#define ZSTD_SRCSIZEHINT_MAX        INT_MAX
+
+
+/* ---  Advanced types  --- */
+
+typedef struct ZSTD_CCtx_params_s ZSTD_CCtx_params;
+
+typedef struct {
+    unsigned int offset;      /* The offset of the match. (NOT the same as the offset code)
+                               * If offset == 0 and matchLength == 0, this sequence represents the last
+                               * literals in the block of litLength size.
+                               */
+
+    unsigned int litLength;   /* Literal length of the sequence. */
+    unsigned int matchLength; /* Match length of the sequence. */
+
+                              /* Note: Users of this API may provide a sequence with matchLength == litLength == offset == 0.
+                               * In this case, we will treat the sequence as a marker for a block boundary.
+                               */
+
+    unsigned int rep;         /* Represents which repeat offset is represented by the field 'offset'.
+                               * Ranges from [0, 3].
+                               *
+                               * Repeat offsets are essentially previous offsets from previous sequences sorted in
+                               * recency order. For more detail, see doc/zstd_compression_format.md
+                               *
+                               * If rep == 0, then 'offset' does not contain a repeat offset.
+                               * If rep > 0:
+                               *  If litLength != 0:
+                               *      rep == 1 --> offset == repeat_offset_1
+                               *      rep == 2 --> offset == repeat_offset_2
+                               *      rep == 3 --> offset == repeat_offset_3
+                               *  If litLength == 0:
+                               *      rep == 1 --> offset == repeat_offset_2
+                               *      rep == 2 --> offset == repeat_offset_3
+                               *      rep == 3 --> offset == repeat_offset_1 - 1
+                               *
+                               * Note: This field is optional. ZSTD_generateSequences() will calculate the value of
+                               * 'rep', but repeat offsets do not necessarily need to be calculated from an external
+                               * sequence provider's perspective. For example, ZSTD_compressSequences() does not
+                               * use this 'rep' field at all (as of now).
+                               */
+} ZSTD_Sequence;
+
+typedef struct {
+    unsigned windowLog;       /**< largest match distance : larger == more compression, more memory needed during decompression */
+    unsigned chainLog;        /**< fully searched segment : larger == more compression, slower, more memory (useless for fast) */
+    unsigned hashLog;         /**< dispatch table : larger == faster, more memory */
+    unsigned searchLog;       /**< nb of searches : larger == more compression, slower */
+    unsigned minMatch;        /**< match length searched : larger == faster decompression, sometimes less compression */
+    unsigned targetLength;    /**< acceptable match size for optimal parser (only) : larger == more compression, slower */
+    ZSTD_strategy strategy;   /**< see ZSTD_strategy definition above */
+} ZSTD_compressionParameters;
+
+typedef struct {
+    int contentSizeFlag; /**< 1: content size will be in frame header (when known) */
+    int checksumFlag;    /**< 1: generate a 32-bits checksum using XXH64 algorithm at end of frame, for error detection */
+    int noDictIDFlag;    /**< 1: no dictID will be saved into frame header (dictID is only useful for dictionary compression) */
+} ZSTD_frameParameters;
+
+typedef struct {
+    ZSTD_compressionParameters cParams;
+    ZSTD_frameParameters fParams;
+} ZSTD_parameters;
+
+typedef enum {
+    ZSTD_dct_auto = 0,       /* dictionary is "full" when starting with ZSTD_MAGIC_DICTIONARY, otherwise it is "rawContent" */
+    ZSTD_dct_rawContent = 1, /* ensures dictionary is always loaded as rawContent, even if it starts with ZSTD_MAGIC_DICTIONARY */
+    ZSTD_dct_fullDict = 2    /* refuses to load a dictionary if it does not respect Zstandard's specification, starting with ZSTD_MAGIC_DICTIONARY */
+} ZSTD_dictContentType_e;
+
+typedef enum {
+    ZSTD_dlm_byCopy = 0,  /**< Copy dictionary content internally */
+    ZSTD_dlm_byRef = 1    /**< Reference dictionary content -- the dictionary buffer must outlive its users. */
+} ZSTD_dictLoadMethod_e;
+
+typedef enum {
+    ZSTD_f_zstd1 = 0,           /* zstd frame format, specified in zstd_compression_format.md (default) */
+    ZSTD_f_zstd1_magicless = 1  /* Variant of zstd frame format, without initial 4-bytes magic number.
+                                 * Useful to save 4 bytes per generated frame.
+                                 * Decoder cannot recognise automatically this format, requiring this instruction. */
+} ZSTD_format_e;
+
+typedef enum {
+    /* Note: this enum controls ZSTD_d_forceIgnoreChecksum */
+    ZSTD_d_validateChecksum = 0,
+    ZSTD_d_ignoreChecksum = 1
+} ZSTD_forceIgnoreChecksum_e;
+
+typedef enum {
+    /* Note: this enum controls ZSTD_d_refMultipleDDicts */
+    ZSTD_rmd_refSingleDDict = 0,
+    ZSTD_rmd_refMultipleDDicts = 1
+} ZSTD_refMultipleDDicts_e;
+
+typedef enum {
+    /* Note: this enum and the behavior it controls are effectively internal
+     * implementation details of the compressor. They are expected to continue
+     * to evolve and should be considered only in the context of extremely
+     * advanced performance tuning.
+     *
+     * Zstd currently supports the use of a CDict in three ways:
+     *
+     * - The contents of the CDict can be copied into the working context. This
+     *   means that the compression can search both the dictionary and input
+     *   while operating on a single set of internal tables. This makes
+     *   the compression faster per-byte of input. However, the initial copy of
+     *   the CDict's tables incurs a fixed cost at the beginning of the
+     *   compression. For small compressions (< 8 KB), that copy can dominate
+     *   the cost of the compression.
+     *
+     * - The CDict's tables can be used in-place. In this model, compression is
+     *   slower per input byte, because the compressor has to search two sets of
+     *   tables. However, this model incurs no start-up cost (as long as the
+     *   working context's tables can be reused). For small inputs, this can be
+     *   faster than copying the CDict's tables.
+     *
+     * - The CDict's tables are not used at all, and instead we use the working
+     *   context alone to reload the dictionary and use params based on the source
+     *   size. See ZSTD_compress_insertDictionary() and ZSTD_compress_usingDict().
+     *   This method is effective when the dictionary sizes are very small relative
+     *   to the input size, and the input size is fairly large to begin with.
+     *
+     * Zstd has a simple internal heuristic that selects which strategy to use
+     * at the beginning of a compression. However, if experimentation shows that
+     * Zstd is making poor choices, it is possible to override that choice with
+     * this enum.
+     */
+    ZSTD_dictDefaultAttach = 0, /* Use the default heuristic. */
+    ZSTD_dictForceAttach   = 1, /* Never copy the dictionary. */
+    ZSTD_dictForceCopy     = 2, /* Always copy the dictionary. */
+    ZSTD_dictForceLoad     = 3  /* Always reload the dictionary */
+} ZSTD_dictAttachPref_e;
+
+typedef enum {
+  ZSTD_lcm_auto = 0,          /**< Automatically determine the compression mode based on the compression level.
+                               *   Negative compression levels will be uncompressed, and positive compression
+                               *   levels will be compressed. */
+  ZSTD_lcm_huffman = 1,       /**< Always attempt Huffman compression. Uncompressed literals will still be
+                               *   emitted if Huffman compression is not profitable. */
+  ZSTD_lcm_uncompressed = 2   /**< Always emit uncompressed literals. */
+} ZSTD_literalCompressionMode_e;
+
+typedef enum {
+  /* Note: This enum controls features which are conditionally beneficial. Zstd typically will make a final
+   * decision on whether or not to enable the feature (ZSTD_ps_auto), but setting the switch to ZSTD_ps_enable
+   * or ZSTD_ps_disable allow for a force enable/disable the feature.
+   */
+  ZSTD_ps_auto = 0,         /* Let the library automatically determine whether the feature shall be enabled */
+  ZSTD_ps_enable = 1,       /* Force-enable the feature */
+  ZSTD_ps_disable = 2       /* Do not use the feature */
+} ZSTD_paramSwitch_e;
+
+/***************************************
+*  Frame header and size functions
+***************************************/
+
+/*! ZSTD_findDecompressedSize() :
+ *  `src` should point to the start of a series of ZSTD encoded and/or skippable frames
+ *  `srcSize` must be the _exact_ size of this series
+ *       (i.e. there should be a frame boundary at `src + srcSize`)
+ *  @return : - decompressed size of all data in all successive frames
+ *            - if the decompressed size cannot be determined: ZSTD_CONTENTSIZE_UNKNOWN
+ *            - if an error occurred: ZSTD_CONTENTSIZE_ERROR
+ *
+ *   note 1 : decompressed size is an optional field, that may not be present, especially in streaming mode.
+ *            When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 2 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 3 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 4 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure result fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 5 : ZSTD_findDecompressedSize handles multiple frames, and so it must traverse the input to
+ *            read each contained frame header.  This is fast as most of the data is skipped,
+ *            however it does mean that all frame data must be present and valid. */
+ZSTDLIB_STATIC_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTD_decompressBound() :
+ *  `src` should point to the start of a series of ZSTD encoded and/or skippable frames
+ *  `srcSize` must be the _exact_ size of this series
+ *       (i.e. there should be a frame boundary at `src + srcSize`)
+ *  @return : - upper-bound for the decompressed size of all data in all successive frames
+ *            - if an error occurred: ZSTD_CONTENTSIZE_ERROR
+ *
+ *  note 1  : an error can occur if `src` contains an invalid or incorrectly formatted frame.
+ *  note 2  : the upper-bound is exact when the decompressed size field is available in every ZSTD encoded frame of `src`.
+ *            in this case, `ZSTD_findDecompressedSize` and `ZSTD_decompressBound` return the same value.
+ *  note 3  : when the decompressed size field isn't available, the upper-bound for that frame is calculated by:
+ *              upper-bound = # blocks * min(128 KB, Window_Size)
+ */
+ZSTDLIB_STATIC_API unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize);
+
+/*! ZSTD_frameHeaderSize() :
+ *  srcSize must be >= ZSTD_FRAMEHEADERSIZE_PREFIX.
+ * @return : size of the Frame Header,
+ *           or an error code (if srcSize is too small) */
+ZSTDLIB_STATIC_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
+
+typedef enum { ZSTD_frame, ZSTD_skippableFrame } ZSTD_frameType_e;
+typedef struct {
+    unsigned long long frameContentSize; /* if == ZSTD_CONTENTSIZE_UNKNOWN, it means this field is not available. 0 means "empty" */
+    unsigned long long windowSize;       /* can be very large, up to <= frameContentSize */
+    unsigned blockSizeMax;
+    ZSTD_frameType_e frameType;          /* if == ZSTD_skippableFrame, frameContentSize is the size of skippable content */
+    unsigned headerSize;
+    unsigned dictID;
+    unsigned checksumFlag;
+    unsigned _reserved1;
+    unsigned _reserved2;
+} ZSTD_frameHeader;
+
+/*! ZSTD_getFrameHeader() :
+ *  decode Frame Header, or requires larger `srcSize`.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+/*! ZSTD_getFrameHeader_advanced() :
+ *  same as ZSTD_getFrameHeader(),
+ *  with added capability to select a format (like ZSTD_f_zstd1_magicless) */
+ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
+
+/*! ZSTD_decompressionMargin() :
+ * Zstd supports in-place decompression, where the input and output buffers overlap.
+ * In this case, the output buffer must be at least (Margin + Output_Size) bytes large,
+ * and the input buffer must be at the end of the output buffer.
+ *
+ *  _______________________ Output Buffer ________________________
+ * |                                                              |
+ * |                                        ____ Input Buffer ____|
+ * |                                       |                      |
+ * v                                       v                      v
+ * |---------------------------------------|-----------|----------|
+ * ^                                                   ^          ^
+ * |___________________ Output_Size ___________________|_ Margin _|
+ *
+ * NOTE: See also ZSTD_DECOMPRESSION_MARGIN().
+ * NOTE: This applies only to single-pass decompression through ZSTD_decompress() or
+ * ZSTD_decompressDCtx().
+ * NOTE: This function supports multi-frame input.
+ *
+ * @param src The compressed frame(s)
+ * @param srcSize The size of the compressed frame(s)
+ * @returns The decompression margin or an error that can be checked with ZSTD_isError().
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_decompressionMargin(const void* src, size_t srcSize);
+
+/*! ZSTD_DECOMPRESS_MARGIN() :
+ * Similar to ZSTD_decompressionMargin(), but instead of computing the margin from
+ * the compressed frame, compute it from the original size and the blockSizeLog.
+ * See ZSTD_decompressionMargin() for details.
+ *
+ * WARNING: This macro does not support multi-frame input, the input must be a single
+ * zstd frame. If you need that support use the function, or implement it yourself.
+ *
+ * @param originalSize The original uncompressed size of the data.
+ * @param blockSize    The block size == MIN(windowSize, ZSTD_BLOCKSIZE_MAX).
+ *                     Unless you explicitly set the windowLog smaller than
+ *                     ZSTD_BLOCKSIZELOG_MAX you can just use ZSTD_BLOCKSIZE_MAX.
+ */
+#define ZSTD_DECOMPRESSION_MARGIN(originalSize, blockSize) ((size_t)(                                              \
+        ZSTD_FRAMEHEADERSIZE_MAX                                                              /* Frame header */ + \
+        4                                                                                         /* checksum */ + \
+        ((originalSize) == 0 ? 0 : 3 * (((originalSize) + (blockSize) - 1) / blockSize)) /* 3 bytes per block */ + \
+        (blockSize)                                                                    /* One block of margin */   \
+    ))
+
+typedef enum {
+  ZSTD_sf_noBlockDelimiters = 0,         /* Representation of ZSTD_Sequence has no block delimiters, sequences only */
+  ZSTD_sf_explicitBlockDelimiters = 1    /* Representation of ZSTD_Sequence contains explicit block delimiters */
+} ZSTD_sequenceFormat_e;
+
+/*! ZSTD_sequenceBound() :
+ * `srcSize` : size of the input buffer
+ *  @return : upper-bound for the number of sequences that can be generated
+ *            from a buffer of srcSize bytes
+ *
+ *  note : returns number of sequences - to get bytes, multiply by sizeof(ZSTD_Sequence).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_sequenceBound(size_t srcSize);
+
+/*! ZSTD_generateSequences() :
+ * Generate sequences using ZSTD_compress2(), given a source buffer.
+ *
+ * Each block will end with a dummy sequence
+ * with offset == 0, matchLength == 0, and litLength == length of last literals.
+ * litLength may be == 0, and if so, then the sequence of (of: 0 ml: 0 ll: 0)
+ * simply acts as a block delimiter.
+ *
+ * @zc can be used to insert custom compression params.
+ * This function invokes ZSTD_compress2().
+ *
+ * The output of this function can be fed into ZSTD_compressSequences() with CCtx
+ * setting of ZSTD_c_blockDelimiters as ZSTD_sf_explicitBlockDelimiters
+ * @return : number of sequences generated
+ */
+
+ZSTDLIB_STATIC_API size_t
+ZSTD_generateSequences( ZSTD_CCtx* zc,
+                        ZSTD_Sequence* outSeqs, size_t outSeqsSize,
+                        const void* src, size_t srcSize);
+
+/*! ZSTD_mergeBlockDelimiters() :
+ * Given an array of ZSTD_Sequence, remove all sequences that represent block delimiters/last literals
+ * by merging them into the literals of the next sequence.
+ *
+ * As such, the final generated result has no explicit representation of block boundaries,
+ * and the final last literals segment is not represented in the sequences.
+ *
+ * The output of this function can be fed into ZSTD_compressSequences() with CCtx
+ * setting of ZSTD_c_blockDelimiters as ZSTD_sf_noBlockDelimiters
+ * @return : number of sequences left after merging
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
+
+/*! ZSTD_compressSequences() :
+ * Compress an array of ZSTD_Sequence, associated with @src buffer, into dst.
+ * @src contains the entire input (not just the literals).
+ * If @srcSize > sum(sequence.length), the remaining bytes are considered all literals
+ * If a dictionary is included, then the cctx should reference the dict. (see: ZSTD_CCtx_refCDict(), ZSTD_CCtx_loadDictionary(), etc.)
+ * The entire source is compressed into a single frame.
+ *
+ * The compression behavior changes based on cctx params. In particular:
+ *    If ZSTD_c_blockDelimiters == ZSTD_sf_noBlockDelimiters, the array of ZSTD_Sequence is expected to contain
+ *    no block delimiters (defined in ZSTD_Sequence). Block boundaries are roughly determined based on
+ *    the block size derived from the cctx, and sequences may be split. This is the default setting.
+ *
+ *    If ZSTD_c_blockDelimiters == ZSTD_sf_explicitBlockDelimiters, the array of ZSTD_Sequence is expected to contain
+ *    block delimiters (defined in ZSTD_Sequence). Behavior is undefined if no block delimiters are provided.
+ *
+ *    If ZSTD_c_validateSequences == 0, this function will blindly accept the sequences provided. Invalid sequences cause undefined
+ *    behavior. If ZSTD_c_validateSequences == 1, then if sequence is invalid (see doc/zstd_compression_format.md for
+ *    specifics regarding offset/matchlength requirements) then the function will bail out and return an error.
+ *
+ *    In addition to the two adjustable experimental params, there are other important cctx params.
+ *    - ZSTD_c_minMatch MUST be set as less than or equal to the smallest match generated by the match finder. It has a minimum value of ZSTD_MINMATCH_MIN.
+ *    - ZSTD_c_compressionLevel accordingly adjusts the strength of the entropy coder, as it would in typical compression.
+ *    - ZSTD_c_windowLog affects offset validation: this function will return an error at higher debug levels if a provided offset
+ *      is larger than what the spec allows for a given window log and dictionary (if present). See: doc/zstd_compression_format.md
+ *
+ * Note: Repcodes are, as of now, always re-calculated within this function, so ZSTD_Sequence::rep is unused.
+ * Note 2: Once we integrate ability to ingest repcodes, the explicit block delims mode must respect those repcodes exactly,
+ *         and cannot emit an RLE block that disagrees with the repcode history
+ * @return : final compressed size, or a ZSTD error code.
+ */
+ZSTDLIB_STATIC_API size_t
+ZSTD_compressSequences( ZSTD_CCtx* cctx, void* dst, size_t dstSize,
+                        const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                        const void* src, size_t srcSize);
+
+
+/*! ZSTD_writeSkippableFrame() :
+ * Generates a zstd skippable frame containing data given by src, and writes it to dst buffer.
+ *
+ * Skippable frames begin with a 4-byte magic number. There are 16 possible choices of magic number,
+ * ranging from ZSTD_MAGIC_SKIPPABLE_START to ZSTD_MAGIC_SKIPPABLE_START+15.
+ * As such, the parameter magicVariant controls the exact skippable frame magic number variant used, so
+ * the magic number used will be ZSTD_MAGIC_SKIPPABLE_START + magicVariant.
+ *
+ * Returns an error if destination buffer is not large enough, if the source size is not representable
+ * with a 4-byte unsigned int, or if the parameter magicVariant is greater than 15 (and therefore invalid).
+ *
+ * @return : number of bytes written or a ZSTD error.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
+                                            const void* src, size_t srcSize, unsigned magicVariant);
+
+/*! ZSTD_readSkippableFrame() :
+ * Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
+ *
+ * The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
+ * i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START.  This can be NULL if the caller is not interested
+ * in the magicVariant.
+ *
+ * Returns an error if destination buffer is not large enough, or if the frame is not skippable.
+ *
+ * @return : number of bytes written or a ZSTD error.
+ */
+ZSTDLIB_API size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
+                                            const void* src, size_t srcSize);
+
+/*! ZSTD_isSkippableFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
+ */
+ZSTDLIB_API unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size);
+
+
+
+/***************************************
+*  Memory management
+***************************************/
+
+/*! ZSTD_estimate*() :
+ *  These functions make it possible to estimate memory usage
+ *  of a future {D,C}Ctx, before its creation.
+ *
+ *  ZSTD_estimateCCtxSize() will provide a memory budget large enough
+ *  for any compression level up to selected one.
+ *  Note : Unlike ZSTD_estimateCStreamSize*(), this estimate
+ *         does not include space for a window buffer.
+ *         Therefore, the estimation is only guaranteed for single-shot compressions, not streaming.
+ *  The estimate will assume the input may be arbitrarily large,
+ *  which is the worst case.
+ *
+ *  When srcSize can be bound by a known and rather "small" value,
+ *  this fact can be used to provide a tighter estimation
+ *  because the CCtx compression context will need less memory.
+ *  This tighter estimation can be provided by more advanced functions
+ *  ZSTD_estimateCCtxSize_usingCParams(), which can be used in tandem with ZSTD_getCParams(),
+ *  and ZSTD_estimateCCtxSize_usingCCtxParams(), which can be used in tandem with ZSTD_CCtxParams_setParameter().
+ *  Both can be used to estimate memory using custom compression parameters and arbitrary srcSize limits.
+ *
+ *  Note : only single-threaded compression is supported.
+ *  ZSTD_estimateCCtxSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
+ *
+ *  Note 2 : ZSTD_estimateCCtxSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
+ *  Size estimates assume that no external sequence producer is registered.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize(int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDCtxSize(void);
+
+/*! ZSTD_estimateCStreamSize() :
+ *  ZSTD_estimateCStreamSize() will provide a budget large enough for any compression level up to selected one.
+ *  It will also consider src size to be arbitrarily "large", which is worst case.
+ *  If srcSize is known to always be small, ZSTD_estimateCStreamSize_usingCParams() can provide a tighter estimation.
+ *  ZSTD_estimateCStreamSize_usingCParams() can be used in tandem with ZSTD_getCParams() to create cParams from compressionLevel.
+ *  ZSTD_estimateCStreamSize_usingCCtxParams() can be used in tandem with ZSTD_CCtxParams_setParameter(). Only single-threaded compression is supported. This function will return an error code if ZSTD_c_nbWorkers is >= 1.
+ *  Note : CStream size estimation is only correct for single-threaded compression.
+ *  ZSTD_DStream memory budget depends on window Size.
+ *  This information can be passed manually, using ZSTD_estimateDStreamSize,
+ *  or deducted from a valid frame Header, using ZSTD_estimateDStreamSize_fromFrame();
+ *  Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
+ *         an internal ?Dict will be created, which additional size is not estimated here.
+ *         In this case, get total size by adding ZSTD_estimate?DictSize
+ *  Note 2 : only single-threaded compression is supported.
+ *  ZSTD_estimateCStreamSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
+ *  Note 3 : ZSTD_estimateCStreamSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
+ *  Size estimates assume that no external sequence producer is registered.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize(int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize(size_t windowSize);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
+
+/*! ZSTD_estimate?DictSize() :
+ *  ZSTD_estimateCDictSize() will bet that src size is relatively "small", and content is copied, like ZSTD_createCDict().
+ *  ZSTD_estimateCDictSize_advanced() makes it possible to control compression parameters precisely, like ZSTD_createCDict_advanced().
+ *  Note : dictionaries created by reference (`ZSTD_dlm_byRef`) are logically smaller.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
+
+/*! ZSTD_initStatic*() :
+ *  Initialize an object using a pre-allocated fixed-size buffer.
+ *  workspace: The memory area to emplace the object into.
+ *             Provided pointer *must be 8-bytes aligned*.
+ *             Buffer must outlive object.
+ *  workspaceSize: Use ZSTD_estimate*Size() to determine
+ *                 how large workspace must be to support target scenario.
+ * @return : pointer to object (same address as workspace, just different type),
+ *           or NULL if error (size too small, incorrect alignment, etc.)
+ *  Note : zstd will never resize nor malloc() when using a static buffer.
+ *         If the object requires more memory than available,
+ *         zstd will just error out (typically ZSTD_error_memory_allocation).
+ *  Note 2 : there is no corresponding "free" function.
+ *           Since workspace is allocated externally, it must be freed externally too.
+ *  Note 3 : cParams : use ZSTD_getCParams() to convert a compression level
+ *           into its associated cParams.
+ *  Limitation 1 : currently not compatible with internal dictionary creation, triggered by
+ *                 ZSTD_CCtx_loadDictionary(), ZSTD_initCStream_usingDict() or ZSTD_initDStream_usingDict().
+ *  Limitation 2 : static cctx currently not compatible with multi-threading.
+ *  Limitation 3 : static dctx is incompatible with legacy support.
+ */
+ZSTDLIB_STATIC_API ZSTD_CCtx*    ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_STATIC_API ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticCCtx() */
+
+ZSTDLIB_STATIC_API ZSTD_DCtx*    ZSTD_initStaticDCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_STATIC_API ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticDCtx() */
+
+ZSTDLIB_STATIC_API const ZSTD_CDict* ZSTD_initStaticCDict(
+                                        void* workspace, size_t workspaceSize,
+                                        const void* dict, size_t dictSize,
+                                        ZSTD_dictLoadMethod_e dictLoadMethod,
+                                        ZSTD_dictContentType_e dictContentType,
+                                        ZSTD_compressionParameters cParams);
+
+ZSTDLIB_STATIC_API const ZSTD_DDict* ZSTD_initStaticDDict(
+                                        void* workspace, size_t workspaceSize,
+                                        const void* dict, size_t dictSize,
+                                        ZSTD_dictLoadMethod_e dictLoadMethod,
+                                        ZSTD_dictContentType_e dictContentType);
+
+
+/*! Custom memory allocation :
+ *  These prototypes make it possible to pass your own allocation/free functions.
+ *  ZSTD_customMem is provided at creation time, using ZSTD_create*_advanced() variants listed below.
+ *  All allocation/free operations will be completed using these custom variants instead of regular <stdlib.h> ones.
+ */
+typedef void* (*ZSTD_allocFunction) (void* opaque, size_t size);
+typedef void  (*ZSTD_freeFunction) (void* opaque, void* address);
+typedef struct { ZSTD_allocFunction customAlloc; ZSTD_freeFunction customFree; void* opaque; } ZSTD_customMem;
+static
+#ifdef __GNUC__
+__attribute__((__unused__))
+#endif
+ZSTD_customMem const ZSTD_defaultCMem = { NULL, NULL, NULL };  /**< this constant defers to stdlib's functions */
+
+ZSTDLIB_STATIC_API ZSTD_CCtx*    ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_STATIC_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
+ZSTDLIB_STATIC_API ZSTD_DCtx*    ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_STATIC_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
+
+ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize,
+                                                  ZSTD_dictLoadMethod_e dictLoadMethod,
+                                                  ZSTD_dictContentType_e dictContentType,
+                                                  ZSTD_compressionParameters cParams,
+                                                  ZSTD_customMem customMem);
+
+/*! Thread pool :
+ *  These prototypes make it possible to share a thread pool among multiple compression contexts.
+ *  This can limit resources for applications with multiple threads where each one uses
+ *  a threaded compression mode (via ZSTD_c_nbWorkers parameter).
+ *  ZSTD_createThreadPool creates a new thread pool with a given number of threads.
+ *  Note that the lifetime of such pool must exist while being used.
+ *  ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
+ *  to use an internal thread pool).
+ *  ZSTD_freeThreadPool frees a thread pool, accepts NULL pointer.
+ */
+typedef struct POOL_ctx_s ZSTD_threadPool;
+ZSTDLIB_STATIC_API ZSTD_threadPool* ZSTD_createThreadPool(size_t numThreads);
+ZSTDLIB_STATIC_API void ZSTD_freeThreadPool (ZSTD_threadPool* pool);  /* accept NULL pointer */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool);
+
+
+/*
+ * This API is temporary and is expected to change or disappear in the future!
+ */
+ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_advanced2(
+    const void* dict, size_t dictSize,
+    ZSTD_dictLoadMethod_e dictLoadMethod,
+    ZSTD_dictContentType_e dictContentType,
+    const ZSTD_CCtx_params* cctxParams,
+    ZSTD_customMem customMem);
+
+ZSTDLIB_STATIC_API ZSTD_DDict* ZSTD_createDDict_advanced(
+    const void* dict, size_t dictSize,
+    ZSTD_dictLoadMethod_e dictLoadMethod,
+    ZSTD_dictContentType_e dictContentType,
+    ZSTD_customMem customMem);
+
+
+/***************************************
+*  Advanced compression functions
+***************************************/
+
+/*! ZSTD_createCDict_byReference() :
+ *  Create a digested dictionary for compression
+ *  Dictionary content is just referenced, not duplicated.
+ *  As a consequence, `dictBuffer` **must** outlive CDict,
+ *  and its content must remain unmodified throughout the lifetime of CDict.
+ *  note: equivalent to ZSTD_createCDict_advanced(), with dictLoadMethod==ZSTD_dlm_byRef */
+ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
+
+/*! ZSTD_getCParams() :
+ * @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
+ * `estimatedSrcSize` value is optional, select 0 if not known */
+ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+
+/*! ZSTD_getParams() :
+ *  same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of sub-component `ZSTD_compressionParameters`.
+ *  All fields of `ZSTD_frameParameters` are set to default : contentSize=1, checksum=0, noDictID=0 */
+ZSTDLIB_STATIC_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+
+/*! ZSTD_checkCParams() :
+ *  Ensure param values remain within authorized range.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()) */
+ZSTDLIB_STATIC_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
+
+/*! ZSTD_adjustCParams() :
+ *  optimize params for a given `srcSize` and `dictSize`.
+ * `srcSize` can be unknown, in which case use ZSTD_CONTENTSIZE_UNKNOWN.
+ * `dictSize` must be `0` when there is no dictionary.
+ *  cPar can be invalid : all parameters will be clamped within valid range in the @return struct.
+ *  This function never fails (wide contract) */
+ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
+
+/*! ZSTD_CCtx_setCParams() :
+ *  Set all parameters provided within @p cparams into the working @p cctx.
+ *  Note : if modifying parameters during compression (MT mode only),
+ *         note that changes to the .windowLog parameter will be ignored.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()).
+ *         On failure, no parameters are updated.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams);
+
+/*! ZSTD_CCtx_setFParams() :
+ *  Set all parameters provided within @p fparams into the working @p cctx.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setFParams(ZSTD_CCtx* cctx, ZSTD_frameParameters fparams);
+
+/*! ZSTD_CCtx_setParams() :
+ *  Set all parameters provided within @p params into the working @p cctx.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setParams(ZSTD_CCtx* cctx, ZSTD_parameters params);
+
+/*! ZSTD_compress_advanced() :
+ *  Note : this function is now DEPRECATED.
+ *         It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_setParameter() and other parameter setters.
+ *  This prototype will generate compilation warnings. */
+ZSTD_DEPRECATED("use ZSTD_compress2")
+ZSTDLIB_STATIC_API
+size_t ZSTD_compress_advanced(ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                        const void* dict,size_t dictSize,
+                              ZSTD_parameters params);
+
+/*! ZSTD_compress_usingCDict_advanced() :
+ *  Note : this function is now DEPRECATED.
+ *         It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_loadDictionary() and other parameter setters.
+ *  This prototype will generate compilation warnings. */
+ZSTD_DEPRECATED("use ZSTD_compress2 with ZSTD_CCtx_loadDictionary")
+ZSTDLIB_STATIC_API
+size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                              void* dst, size_t dstCapacity,
+                                        const void* src, size_t srcSize,
+                                        const ZSTD_CDict* cdict,
+                                              ZSTD_frameParameters fParams);
+
+
+/*! ZSTD_CCtx_loadDictionary_byReference() :
+ *  Same as ZSTD_CCtx_loadDictionary(), but dictionary content is referenced, instead of being copied into CCtx.
+ *  It saves some memory, but also requires that `dict` outlives its usage within `cctx` */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+
+/*! ZSTD_CCtx_loadDictionary_advanced() :
+ *  Same as ZSTD_CCtx_loadDictionary(), but gives finer control over
+ *  how to load the dictionary (by copy ? by reference ?)
+ *  and how to interpret it (automatic ? force raw mode ? full mode only ?) */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
+/*! ZSTD_CCtx_refPrefix_advanced() :
+ *  Same as ZSTD_CCtx_refPrefix(), but gives finer control over
+ *  how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
+/* ===   experimental parameters   === */
+/* these parameters can be used with ZSTD_setParameter()
+ * they are not guaranteed to remain supported in the future */
+
+ /* Enables rsyncable mode,
+  * which makes compressed files more rsync friendly
+  * by adding periodic synchronization points to the compressed data.
+  * The target average block size is ZSTD_c_jobSize / 2.
+  * It's possible to modify the job size to increase or decrease
+  * the granularity of the synchronization point.
+  * Once the jobSize is smaller than the window size,
+  * it will result in compression ratio degradation.
+  * NOTE 1: rsyncable mode only works when multithreading is enabled.
+  * NOTE 2: rsyncable performs poorly in combination with long range mode,
+  * since it will decrease the effectiveness of synchronization points,
+  * though mileage may vary.
+  * NOTE 3: Rsyncable mode limits maximum compression speed to ~400 MB/s.
+  * If the selected compression level is already running significantly slower,
+  * the overall speed won't be significantly impacted.
+  */
+ #define ZSTD_c_rsyncable ZSTD_c_experimentalParam1
+
+/* Select a compression format.
+ * The value must be of type ZSTD_format_e.
+ * See ZSTD_format_e enum definition for details */
+#define ZSTD_c_format ZSTD_c_experimentalParam2
+
+/* Force back-reference distances to remain < windowSize,
+ * even when referencing into Dictionary content (default:0) */
+#define ZSTD_c_forceMaxWindow ZSTD_c_experimentalParam3
+
+/* Controls whether the contents of a CDict
+ * are used in place, or copied into the working context.
+ * Accepts values from the ZSTD_dictAttachPref_e enum.
+ * See the comments on that enum for an explanation of the feature. */
+#define ZSTD_c_forceAttachDict ZSTD_c_experimentalParam4
+
+/* Controlled with ZSTD_paramSwitch_e enum.
+ * Default is ZSTD_ps_auto.
+ * Set to ZSTD_ps_disable to never compress literals.
+ * Set to ZSTD_ps_enable to always compress literals. (Note: uncompressed literals
+ * may still be emitted if huffman is not beneficial to use.)
+ *
+ * By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
+ * literals compression based on the compression parameters - specifically,
+ * negative compression levels do not use literal compression.
+ */
+#define ZSTD_c_literalCompressionMode ZSTD_c_experimentalParam5
+
+/* Tries to fit compressed block size to be around targetCBlockSize.
+ * No target when targetCBlockSize == 0.
+ * There is no guarantee on compressed block size (default:0) */
+#define ZSTD_c_targetCBlockSize ZSTD_c_experimentalParam6
+
+/* User's best guess of source size.
+ * Hint is not valid when srcSizeHint == 0.
+ * There is no guarantee that hint is close to actual source size,
+ * but compression ratio may regress significantly if guess considerably underestimates */
+#define ZSTD_c_srcSizeHint ZSTD_c_experimentalParam7
+
+/* Controls whether the new and experimental "dedicated dictionary search
+ * structure" can be used. This feature is still rough around the edges, be
+ * prepared for surprising behavior!
+ *
+ * How to use it:
+ *
+ * When using a CDict, whether to use this feature or not is controlled at
+ * CDict creation, and it must be set in a CCtxParams set passed into that
+ * construction (via ZSTD_createCDict_advanced2()). A compression will then
+ * use the feature or not based on how the CDict was constructed; the value of
+ * this param, set in the CCtx, will have no effect.
+ *
+ * However, when a dictionary buffer is passed into a CCtx, such as via
+ * ZSTD_CCtx_loadDictionary(), this param can be set on the CCtx to control
+ * whether the CDict that is created internally can use the feature or not.
+ *
+ * What it does:
+ *
+ * Normally, the internal data structures of the CDict are analogous to what
+ * would be stored in a CCtx after compressing the contents of a dictionary.
+ * To an approximation, a compression using a dictionary can then use those
+ * data structures to simply continue what is effectively a streaming
+ * compression where the simulated compression of the dictionary left off.
+ * Which is to say, the search structures in the CDict are normally the same
+ * format as in the CCtx.
+ *
+ * It is possible to do better, since the CDict is not like a CCtx: the search
+ * structures are written once during CDict creation, and then are only read
+ * after that, while the search structures in the CCtx are both read and
+ * written as the compression goes along. This means we can choose a search
+ * structure for the dictionary that is read-optimized.
+ *
+ * This feature enables the use of that different structure.
+ *
+ * Note that some of the members of the ZSTD_compressionParameters struct have
+ * different semantics and constraints in the dedicated search structure. It is
+ * highly recommended that you simply set a compression level in the CCtxParams
+ * you pass into the CDict creation call, and avoid messing with the cParams
+ * directly.
+ *
+ * Effects:
+ *
+ * This will only have any effect when the selected ZSTD_strategy
+ * implementation supports this feature. Currently, that's limited to
+ * ZSTD_greedy, ZSTD_lazy, and ZSTD_lazy2.
+ *
+ * Note that this means that the CDict tables can no longer be copied into the
+ * CCtx, so the dict attachment mode ZSTD_dictForceCopy will no longer be
+ * usable. The dictionary can only be attached or reloaded.
+ *
+ * In general, you should expect compression to be faster--sometimes very much
+ * so--and CDict creation to be slightly slower. Eventually, we will probably
+ * make this mode the default.
+ */
+#define ZSTD_c_enableDedicatedDictSearch ZSTD_c_experimentalParam8
+
+/* ZSTD_c_stableInBuffer
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable.
+ *
+ * Tells the compressor that input data presented with ZSTD_inBuffer
+ * will ALWAYS be the same between calls.
+ * Technically, the @src pointer must never be changed,
+ * and the @pos field can only be updated by zstd.
+ * However, it's possible to increase the @size field,
+ * allowing scenarios where more data can be appended after compressions starts.
+ * These conditions are checked by the compressor,
+ * and compression will fail if they are not respected.
+ * Also, data in the ZSTD_inBuffer within the range [src, src + pos)
+ * MUST not be modified during compression or it will result in data corruption.
+ *
+ * When this flag is enabled zstd won't allocate an input window buffer,
+ * because the user guarantees it can reference the ZSTD_inBuffer until
+ * the frame is complete. But, it will still allocate an output buffer
+ * large enough to fit a block (see ZSTD_c_stableOutBuffer). This will also
+ * avoid the memcpy() from the input buffer to the input window buffer.
+ *
+ * NOTE: So long as the ZSTD_inBuffer always points to valid memory, using
+ * this flag is ALWAYS memory safe, and will never access out-of-bounds
+ * memory. However, compression WILL fail if conditions are not respected.
+ *
+ * WARNING: The data in the ZSTD_inBuffer in the range [src, src + pos) MUST
+ * not be modified during compression or it will result in data corruption.
+ * This is because zstd needs to reference data in the ZSTD_inBuffer to find
+ * matches. Normally zstd maintains its own window buffer for this purpose,
+ * but passing this flag tells zstd to rely on user provided buffer instead.
+ */
+#define ZSTD_c_stableInBuffer ZSTD_c_experimentalParam9
+
+/* ZSTD_c_stableOutBuffer
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable.
+ *
+ * Tells he compressor that the ZSTD_outBuffer will not be resized between
+ * calls. Specifically: (out.size - out.pos) will never grow. This gives the
+ * compressor the freedom to say: If the compressed data doesn't fit in the
+ * output buffer then return ZSTD_error_dstSizeTooSmall. This allows us to
+ * always decompress directly into the output buffer, instead of decompressing
+ * into an internal buffer and copying to the output buffer.
+ *
+ * When this flag is enabled zstd won't allocate an output buffer, because
+ * it can write directly to the ZSTD_outBuffer. It will still allocate the
+ * input window buffer (see ZSTD_c_stableInBuffer).
+ *
+ * Zstd will check that (out.size - out.pos) never grows and return an error
+ * if it does. While not strictly necessary, this should prevent surprises.
+ */
+#define ZSTD_c_stableOutBuffer ZSTD_c_experimentalParam10
+
+/* ZSTD_c_blockDelimiters
+ * Default is 0 == ZSTD_sf_noBlockDelimiters.
+ *
+ * For use with sequence compression API: ZSTD_compressSequences().
+ *
+ * Designates whether or not the given array of ZSTD_Sequence contains block delimiters
+ * and last literals, which are defined as sequences with offset == 0 and matchLength == 0.
+ * See the definition of ZSTD_Sequence for more specifics.
+ */
+#define ZSTD_c_blockDelimiters ZSTD_c_experimentalParam11
+
+/* ZSTD_c_validateSequences
+ * Default is 0 == disabled. Set to 1 to enable sequence validation.
+ *
+ * For use with sequence compression API: ZSTD_compressSequences().
+ * Designates whether or not we validate sequences provided to ZSTD_compressSequences()
+ * during function execution.
+ *
+ * Without validation, providing a sequence that does not conform to the zstd spec will cause
+ * undefined behavior, and may produce a corrupted block.
+ *
+ * With validation enabled, if sequence is invalid (see doc/zstd_compression_format.md for
+ * specifics regarding offset/matchlength requirements) then the function will bail out and
+ * return an error.
+ *
+ */
+#define ZSTD_c_validateSequences ZSTD_c_experimentalParam12
+
+/* ZSTD_c_useBlockSplitter
+ * Controlled with ZSTD_paramSwitch_e enum.
+ * Default is ZSTD_ps_auto.
+ * Set to ZSTD_ps_disable to never use block splitter.
+ * Set to ZSTD_ps_enable to always use block splitter.
+ *
+ * By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
+ * block splitting based on the compression parameters.
+ */
+#define ZSTD_c_useBlockSplitter ZSTD_c_experimentalParam13
+
+/* ZSTD_c_useRowMatchFinder
+ * Controlled with ZSTD_paramSwitch_e enum.
+ * Default is ZSTD_ps_auto.
+ * Set to ZSTD_ps_disable to never use row-based matchfinder.
+ * Set to ZSTD_ps_enable to force usage of row-based matchfinder.
+ *
+ * By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
+ * the row-based matchfinder based on support for SIMD instructions and the window log.
+ * Note that this only pertains to compression strategies: greedy, lazy, and lazy2
+ */
+#define ZSTD_c_useRowMatchFinder ZSTD_c_experimentalParam14
+
+/* ZSTD_c_deterministicRefPrefix
+ * Default is 0 == disabled. Set to 1 to enable.
+ *
+ * Zstd produces different results for prefix compression when the prefix is
+ * directly adjacent to the data about to be compressed vs. when it isn't.
+ * This is because zstd detects that the two buffers are contiguous and it can
+ * use a more efficient match finding algorithm. However, this produces different
+ * results than when the two buffers are non-contiguous. This flag forces zstd
+ * to always load the prefix in non-contiguous mode, even if it happens to be
+ * adjacent to the data, to guarantee determinism.
+ *
+ * If you really care about determinism when using a dictionary or prefix,
+ * like when doing delta compression, you should select this option. It comes
+ * at a speed penalty of about ~2.5% if the dictionary and data happened to be
+ * contiguous, and is free if they weren't contiguous. We don't expect that
+ * intentionally making the dictionary and data contiguous will be worth the
+ * cost to memcpy() the data.
+ */
+#define ZSTD_c_deterministicRefPrefix ZSTD_c_experimentalParam15
+
+/* ZSTD_c_prefetchCDictTables
+ * Controlled with ZSTD_paramSwitch_e enum. Default is ZSTD_ps_auto.
+ *
+ * In some situations, zstd uses CDict tables in-place rather than copying them
+ * into the working context. (See docs on ZSTD_dictAttachPref_e above for details).
+ * In such situations, compression speed is seriously impacted when CDict tables are
+ * "cold" (outside CPU cache). This parameter instructs zstd to prefetch CDict tables
+ * when they are used in-place.
+ *
+ * For sufficiently small inputs, the cost of the prefetch will outweigh the benefit.
+ * For sufficiently large inputs, zstd will by default memcpy() CDict tables
+ * into the working context, so there is no need to prefetch. This parameter is
+ * targeted at a middle range of input sizes, where a prefetch is cheap enough to be
+ * useful but memcpy() is too expensive. The exact range of input sizes where this
+ * makes sense is best determined by careful experimentation.
+ *
+ * Note: for this parameter, ZSTD_ps_auto is currently equivalent to ZSTD_ps_disable,
+ * but in the future zstd may conditionally enable this feature via an auto-detection
+ * heuristic for cold CDicts.
+ * Use ZSTD_ps_disable to opt out of prefetching under any circumstances.
+ */
+#define ZSTD_c_prefetchCDictTables ZSTD_c_experimentalParam16
+
+/* ZSTD_c_enableSeqProducerFallback
+ * Allowed values are 0 (disable) and 1 (enable). The default setting is 0.
+ *
+ * Controls whether zstd will fall back to an internal sequence producer if an
+ * external sequence producer is registered and returns an error code. This fallback
+ * is block-by-block: the internal sequence producer will only be called for blocks
+ * where the external sequence producer returns an error code. Fallback parsing will
+ * follow any other cParam settings, such as compression level, the same as in a
+ * normal (fully-internal) compression operation.
+ *
+ * The user is strongly encouraged to read the full Block-Level Sequence Producer API
+ * documentation (below) before setting this parameter. */
+#define ZSTD_c_enableSeqProducerFallback ZSTD_c_experimentalParam17
+
+/* ZSTD_c_maxBlockSize
+ * Allowed values are between 1KB and ZSTD_BLOCKSIZE_MAX (128KB).
+ * The default is ZSTD_BLOCKSIZE_MAX, and setting to 0 will set to the default.
+ *
+ * This parameter can be used to set an upper bound on the blocksize
+ * that overrides the default ZSTD_BLOCKSIZE_MAX. It cannot be used to set upper
+ * bounds greater than ZSTD_BLOCKSIZE_MAX or bounds lower than 1KB (will make
+ * compressBound() inaccurate). Only currently meant to be used for testing.
+ *
+ */
+#define ZSTD_c_maxBlockSize ZSTD_c_experimentalParam18
+
+/* ZSTD_c_searchForExternalRepcodes
+ * This parameter affects how zstd parses external sequences, such as sequences
+ * provided through the compressSequences() API or from an external block-level
+ * sequence producer.
+ *
+ * If set to ZSTD_ps_enable, the library will check for repeated offsets in
+ * external sequences, even if those repcodes are not explicitly indicated in
+ * the "rep" field. Note that this is the only way to exploit repcode matches
+ * while using compressSequences() or an external sequence producer, since zstd
+ * currently ignores the "rep" field of external sequences.
+ *
+ * If set to ZSTD_ps_disable, the library will not exploit repeated offsets in
+ * external sequences, regardless of whether the "rep" field has been set. This
+ * reduces sequence compression overhead by about 25% while sacrificing some
+ * compression ratio.
+ *
+ * The default value is ZSTD_ps_auto, for which the library will enable/disable
+ * based on compression level.
+ *
+ * Note: for now, this param only has an effect if ZSTD_c_blockDelimiters is
+ * set to ZSTD_sf_explicitBlockDelimiters. That may change in the future.
+ */
+#define ZSTD_c_searchForExternalRepcodes ZSTD_c_experimentalParam19
+
+/*! ZSTD_CCtx_getParameter() :
+ *  Get the requested compression parameter value, selected by enum ZSTD_cParameter,
+ *  and store it into int* value.
+ * @return : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_getParameter(const ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value);
+
+
+/*! ZSTD_CCtx_params :
+ *  Quick howto :
+ *  - ZSTD_createCCtxParams() : Create a ZSTD_CCtx_params structure
+ *  - ZSTD_CCtxParams_setParameter() : Push parameters one by one into
+ *                                     an existing ZSTD_CCtx_params structure.
+ *                                     This is similar to
+ *                                     ZSTD_CCtx_setParameter().
+ *  - ZSTD_CCtx_setParametersUsingCCtxParams() : Apply parameters to
+ *                                    an existing CCtx.
+ *                                    These parameters will be applied to
+ *                                    all subsequent frames.
+ *  - ZSTD_compressStream2() : Do compression using the CCtx.
+ *  - ZSTD_freeCCtxParams() : Free the memory, accept NULL pointer.
+ *
+ *  This can be used with ZSTD_estimateCCtxSize_advanced_usingCCtxParams()
+ *  for static allocation of CCtx for single-threaded compression.
+ */
+ZSTDLIB_STATIC_API ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
+ZSTDLIB_STATIC_API size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);  /* accept NULL pointer */
+
+/*! ZSTD_CCtxParams_reset() :
+ *  Reset params to default values.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
+
+/*! ZSTD_CCtxParams_init() :
+ *  Initializes the compression parameters of cctxParams according to
+ *  compression level. All other parameters are reset to their default values.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
+
+/*! ZSTD_CCtxParams_init_advanced() :
+ *  Initializes the compression and frame parameters of cctxParams according to
+ *  params. All other parameters are reset to their default values.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
+
+/*! ZSTD_CCtxParams_setParameter() : Requires v1.4.0+
+ *  Similar to ZSTD_CCtx_setParameter.
+ *  Set one compression parameter, selected by enum ZSTD_cParameter.
+ *  Parameters must be applied to a ZSTD_CCtx using
+ *  ZSTD_CCtx_setParametersUsingCCtxParams().
+ * @result : a code representing success or failure (which can be tested with
+ *           ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
+
+/*! ZSTD_CCtxParams_getParameter() :
+ * Similar to ZSTD_CCtx_getParameter.
+ * Get the requested value of one compression parameter, selected by enum ZSTD_cParameter.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_getParameter(const ZSTD_CCtx_params* params, ZSTD_cParameter param, int* value);
+
+/*! ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  Apply a set of ZSTD_CCtx_params to the compression context.
+ *  This can be done even after compression is started,
+ *    if nbWorkers==0, this will have no impact until a new compression is started.
+ *    if nbWorkers>=1, new parameters will be picked up at next job,
+ *       with a few restrictions (windowLog, pledgedSrcSize, nbWorkers, jobSize, and overlapLog are not updated).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+        ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params);
+
+/*! ZSTD_compressStream2_simpleArgs() :
+ *  Same as ZSTD_compressStream2(),
+ *  but using only integral types as arguments.
+ *  This variant might be helpful for binders from dynamic languages
+ *  which have troubles handling structures containing memory pointers.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_compressStream2_simpleArgs (
+                            ZSTD_CCtx* cctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos,
+                            ZSTD_EndDirective endOp);
+
+
+/***************************************
+*  Advanced decompression functions
+***************************************/
+
+/*! ZSTD_isFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
+ *  Note 3 : Skippable Frame Identifiers are considered valid. */
+ZSTDLIB_STATIC_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
+
+/*! ZSTD_createDDict_byReference() :
+ *  Create a digested dictionary, ready to start decompression operation without startup delay.
+ *  Dictionary content is referenced, and therefore stays in dictBuffer.
+ *  It is important that dictBuffer outlives DDict,
+ *  it must remain read accessible throughout the lifetime of DDict */
+ZSTDLIB_STATIC_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
+
+/*! ZSTD_DCtx_loadDictionary_byReference() :
+ *  Same as ZSTD_DCtx_loadDictionary(),
+ *  but references `dict` content instead of copying it into `dctx`.
+ *  This saves memory if `dict` remains around.,
+ *  However, it's imperative that `dict` remains accessible (and unmodified) while being used, so it must outlive decompression. */
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+
+/*! ZSTD_DCtx_loadDictionary_advanced() :
+ *  Same as ZSTD_DCtx_loadDictionary(),
+ *  but gives direct control over
+ *  how to load the dictionary (by copy ? by reference ?)
+ *  and how to interpret it (automatic ? force raw mode ? full mode only ?). */
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
+/*! ZSTD_DCtx_refPrefix_advanced() :
+ *  Same as ZSTD_DCtx_refPrefix(), but gives finer control over
+ *  how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) */
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
+/*! ZSTD_DCtx_setMaxWindowSize() :
+ *  Refuses allocating internal buffers for frames requiring a window size larger than provided limit.
+ *  This protects a decoder context from reserving too much memory for itself (potential attack scenario).
+ *  This parameter is only useful in streaming mode, since no internal buffer is allocated in single-pass mode.
+ *  By default, a decompression context accepts all window sizes <= (1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT)
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
+
+/*! ZSTD_DCtx_getParameter() :
+ *  Get the requested decompression parameter value, selected by enum ZSTD_dParameter,
+ *  and store it into int* value.
+ * @return : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
+
+/* ZSTD_d_format
+ * experimental parameter,
+ * allowing selection between ZSTD_format_e input compression formats
+ */
+#define ZSTD_d_format ZSTD_d_experimentalParam1
+/* ZSTD_d_stableOutBuffer
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable.
+ *
+ * Tells the decompressor that the ZSTD_outBuffer will ALWAYS be the same
+ * between calls, except for the modifications that zstd makes to pos (the
+ * caller must not modify pos). This is checked by the decompressor, and
+ * decompression will fail if it ever changes. Therefore the ZSTD_outBuffer
+ * MUST be large enough to fit the entire decompressed frame. This will be
+ * checked when the frame content size is known. The data in the ZSTD_outBuffer
+ * in the range [dst, dst + pos) MUST not be modified during decompression
+ * or you will get data corruption.
+ *
+ * When this flag is enabled zstd won't allocate an output buffer, because
+ * it can write directly to the ZSTD_outBuffer, but it will still allocate
+ * an input buffer large enough to fit any compressed block. This will also
+ * avoid the memcpy() from the internal output buffer to the ZSTD_outBuffer.
+ * If you need to avoid the input buffer allocation use the buffer-less
+ * streaming API.
+ *
+ * NOTE: So long as the ZSTD_outBuffer always points to valid memory, using
+ * this flag is ALWAYS memory safe, and will never access out-of-bounds
+ * memory. However, decompression WILL fail if you violate the preconditions.
+ *
+ * WARNING: The data in the ZSTD_outBuffer in the range [dst, dst + pos) MUST
+ * not be modified during decompression or you will get data corruption. This
+ * is because zstd needs to reference data in the ZSTD_outBuffer to regenerate
+ * matches. Normally zstd maintains its own buffer for this purpose, but passing
+ * this flag tells zstd to use the user provided buffer.
+ */
+#define ZSTD_d_stableOutBuffer ZSTD_d_experimentalParam2
+
+/* ZSTD_d_forceIgnoreChecksum
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable
+ *
+ * Tells the decompressor to skip checksum validation during decompression, regardless
+ * of whether checksumming was specified during compression. This offers some
+ * slight performance benefits, and may be useful for debugging.
+ * Param has values of type ZSTD_forceIgnoreChecksum_e
+ */
+#define ZSTD_d_forceIgnoreChecksum ZSTD_d_experimentalParam3
+
+/* ZSTD_d_refMultipleDDicts
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable
+ *
+ * If enabled and dctx is allocated on the heap, then additional memory will be allocated
+ * to store references to multiple ZSTD_DDict. That is, multiple calls of ZSTD_refDDict()
+ * using a given ZSTD_DCtx, rather than overwriting the previous DDict reference, will instead
+ * store all references. At decompression time, the appropriate dictID is selected
+ * from the set of DDicts based on the dictID in the frame.
+ *
+ * Usage is simply calling ZSTD_refDDict() on multiple dict buffers.
+ *
+ * Param has values of byte ZSTD_refMultipleDDicts_e
+ *
+ * WARNING: Enabling this parameter and calling ZSTD_DCtx_refDDict(), will trigger memory
+ * allocation for the hash table. ZSTD_freeDCtx() also frees this memory.
+ * Memory is allocated as per ZSTD_DCtx::customMem.
+ *
+ * Although this function allocates memory for the table, the user is still responsible for
+ * memory management of the underlying ZSTD_DDict* themselves.
+ */
+#define ZSTD_d_refMultipleDDicts ZSTD_d_experimentalParam4
+
+/* ZSTD_d_disableHuffmanAssembly
+ * Set to 1 to disable the Huffman assembly implementation.
+ * The default value is 0, which allows zstd to use the Huffman assembly
+ * implementation if available.
+ *
+ * This parameter can be used to disable Huffman assembly at runtime.
+ * If you want to disable it at compile time you can define the macro
+ * ZSTD_DISABLE_ASM.
+ */
+#define ZSTD_d_disableHuffmanAssembly ZSTD_d_experimentalParam5
+
+
+/*! ZSTD_DCtx_setFormat() :
+ *  This function is REDUNDANT. Prefer ZSTD_DCtx_setParameter().
+ *  Instruct the decoder context about what kind of data to decode next.
+ *  This instruction is mandatory to decode data without a fully-formed header,
+ *  such ZSTD_f_zstd1_magicless for example.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()). */
+ZSTD_DEPRECATED("use ZSTD_DCtx_setParameter() instead")
+ZSTDLIB_STATIC_API
+size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
+
+/*! ZSTD_decompressStream_simpleArgs() :
+ *  Same as ZSTD_decompressStream(),
+ *  but using only integral types as arguments.
+ *  This can be helpful for binders from dynamic languages
+ *  which have troubles handling structures containing memory pointers.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_decompressStream_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos);
+
+
+/********************************************************************
+*  Advanced streaming functions
+*  Warning : most of these functions are now redundant with the Advanced API.
+*  Once Advanced API reaches "stable" status,
+*  redundant functions will be deprecated, and then at some point removed.
+********************************************************************/
+
+/*=====   Advanced Streaming compression functions  =====*/
+
+/*! ZSTD_initCStream_srcSize() :
+ * This function is DEPRECATED, and equivalent to:
+ *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+ *     ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
+ *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
+ *     ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
+ *
+ * pledgedSrcSize must be correct. If it is not known at init time, use
+ * ZSTD_CONTENTSIZE_UNKNOWN. Note that, for compatibility with older programs,
+ * "0" also disables frame content size field. It may be enabled in the future.
+ * This prototype will generate compilation warnings.
+ */
+ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
+size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
+                         int compressionLevel,
+                         unsigned long long pledgedSrcSize);
+
+/*! ZSTD_initCStream_usingDict() :
+ * This function is DEPRECATED, and is equivalent to:
+ *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+ *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
+ *     ZSTD_CCtx_loadDictionary(zcs, dict, dictSize);
+ *
+ * Creates of an internal CDict (incompatible with static CCtx), except if
+ * dict == NULL or dictSize < 8, in which case no dict is used.
+ * Note: dict is loaded with ZSTD_dct_auto (treated as a full zstd dictionary if
+ * it begins with ZSTD_MAGIC_DICTIONARY, else as raw content) and ZSTD_dlm_byCopy.
+ * This prototype will generate compilation warnings.
+ */
+ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
+size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
+                     const void* dict, size_t dictSize,
+                           int compressionLevel);
+
+/*! ZSTD_initCStream_advanced() :
+ * This function is DEPRECATED, and is equivalent to:
+ *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+ *     ZSTD_CCtx_setParams(zcs, params);
+ *     ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
+ *     ZSTD_CCtx_loadDictionary(zcs, dict, dictSize);
+ *
+ * dict is loaded with ZSTD_dct_auto and ZSTD_dlm_byCopy.
+ * pledgedSrcSize must be correct.
+ * If srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
+ * This prototype will generate compilation warnings.
+ */
+ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
+size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
+                    const void* dict, size_t dictSize,
+                          ZSTD_parameters params,
+                          unsigned long long pledgedSrcSize);
+
+/*! ZSTD_initCStream_usingCDict() :
+ * This function is DEPRECATED, and equivalent to:
+ *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+ *     ZSTD_CCtx_refCDict(zcs, cdict);
+ *
+ * note : cdict will just be referenced, and must outlive compression session
+ * This prototype will generate compilation warnings.
+ */
+ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
+size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
+
+/*! ZSTD_initCStream_usingCDict_advanced() :
+ *   This function is DEPRECATED, and is equivalent to:
+ *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+ *     ZSTD_CCtx_setFParams(zcs, fParams);
+ *     ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
+ *     ZSTD_CCtx_refCDict(zcs, cdict);
+ *
+ * same as ZSTD_initCStream_usingCDict(), with control over frame parameters.
+ * pledgedSrcSize must be correct. If srcSize is not known at init time, use
+ * value ZSTD_CONTENTSIZE_UNKNOWN.
+ * This prototype will generate compilation warnings.
+ */
+ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
+size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
+                               const ZSTD_CDict* cdict,
+                                     ZSTD_frameParameters fParams,
+                                     unsigned long long pledgedSrcSize);
+
+/*! ZSTD_resetCStream() :
+ * This function is DEPRECATED, and is equivalent to:
+ *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+ *     ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
+ * Note: ZSTD_resetCStream() interprets pledgedSrcSize == 0 as ZSTD_CONTENTSIZE_UNKNOWN, but
+ *       ZSTD_CCtx_setPledgedSrcSize() does not do the same, so ZSTD_CONTENTSIZE_UNKNOWN must be
+ *       explicitly specified.
+ *
+ *  start a new frame, using same parameters from previous frame.
+ *  This is typically useful to skip dictionary loading stage, since it will re-use it in-place.
+ *  Note that zcs must be init at least once before using ZSTD_resetCStream().
+ *  If pledgedSrcSize is not known at reset time, use macro ZSTD_CONTENTSIZE_UNKNOWN.
+ *  If pledgedSrcSize > 0, its value must be correct, as it will be written in header, and controlled at the end.
+ *  For the time being, pledgedSrcSize==0 is interpreted as "srcSize unknown" for compatibility with older programs,
+ *  but it will change to mean "empty" in future version, so use macro ZSTD_CONTENTSIZE_UNKNOWN instead.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError())
+ *  This prototype will generate compilation warnings.
+ */
+ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
+
+
+typedef struct {
+    unsigned long long ingested;   /* nb input bytes read and buffered */
+    unsigned long long consumed;   /* nb input bytes actually compressed */
+    unsigned long long produced;   /* nb of compressed bytes generated and buffered */
+    unsigned long long flushed;    /* nb of compressed bytes flushed : not provided; can be tracked from caller side */
+    unsigned currentJobID;         /* MT only : latest started job nb */
+    unsigned nbActiveWorkers;      /* MT only : nb of workers actively compressing at probe time */
+} ZSTD_frameProgression;
+
+/* ZSTD_getFrameProgression() :
+ * tells how much data has been ingested (read from input)
+ * consumed (input actually compressed) and produced (output) for current frame.
+ * Note : (ingested - consumed) is amount of input data buffered internally, not yet compressed.
+ * Aggregates progression inside active worker threads.
+ */
+ZSTDLIB_STATIC_API ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
+
+/*! ZSTD_toFlushNow() :
+ *  Tell how many bytes are ready to be flushed immediately.
+ *  Useful for multithreading scenarios (nbWorkers >= 1).
+ *  Probe the oldest active job, defined as oldest job not yet entirely flushed,
+ *  and check its output buffer.
+ * @return : amount of data stored in oldest job and ready to be flushed immediately.
+ *  if @return == 0, it means either :
+ *  + there is no active job (could be checked with ZSTD_frameProgression()), or
+ *  + oldest job is still actively compressing data,
+ *    but everything it has produced has also been flushed so far,
+ *    therefore flush speed is limited by production speed of oldest job
+ *    irrespective of the speed of concurrent (and newer) jobs.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
+
+
+/*=====   Advanced Streaming decompression functions  =====*/
+
+/*!
+ * This function is deprecated, and is equivalent to:
+ *
+ *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
+ *     ZSTD_DCtx_loadDictionary(zds, dict, dictSize);
+ *
+ * note: no dictionary will be used if dict == NULL or dictSize < 8
+ */
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_loadDictionary, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
+
+/*!
+ * This function is deprecated, and is equivalent to:
+ *
+ *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
+ *     ZSTD_DCtx_refDDict(zds, ddict);
+ *
+ * note : ddict is referenced, it must outlive decompression session
+ */
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_refDDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
+
+/*!
+ * This function is deprecated, and is equivalent to:
+ *
+ *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
+ *
+ * re-use decompression parameters from previous init; saves dictionary loading
+ */
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
+
+
+/* ********************* BLOCK-LEVEL SEQUENCE PRODUCER API *********************
+ *
+ * *** OVERVIEW ***
+ * The Block-Level Sequence Producer API allows users to provide their own custom
+ * sequence producer which libzstd invokes to process each block. The produced list
+ * of sequences (literals and matches) is then post-processed by libzstd to produce
+ * valid compressed blocks.
+ *
+ * This block-level offload API is a more granular complement of the existing
+ * frame-level offload API compressSequences() (introduced in v1.5.1). It offers
+ * an easier migration story for applications already integrated with libzstd: the
+ * user application continues to invoke the same compression functions
+ * ZSTD_compress2() or ZSTD_compressStream2() as usual, and transparently benefits
+ * from the specific advantages of the external sequence producer. For example,
+ * the sequence producer could be tuned to take advantage of known characteristics
+ * of the input, to offer better speed / ratio, or could leverage hardware
+ * acceleration not available within libzstd itself.
+ *
+ * See contrib/externalSequenceProducer for an example program employing the
+ * Block-Level Sequence Producer API.
+ *
+ * *** USAGE ***
+ * The user is responsible for implementing a function of type
+ * ZSTD_sequenceProducer_F. For each block, zstd will pass the following
+ * arguments to the user-provided function:
+ *
+ *   - sequenceProducerState: a pointer to a user-managed state for the sequence
+ *     producer.
+ *
+ *   - outSeqs, outSeqsCapacity: an output buffer for the sequence producer.
+ *     outSeqsCapacity is guaranteed >= ZSTD_sequenceBound(srcSize). The memory
+ *     backing outSeqs is managed by the CCtx.
+ *
+ *   - src, srcSize: an input buffer for the sequence producer to parse.
+ *     srcSize is guaranteed to be <= ZSTD_BLOCKSIZE_MAX.
+ *
+ *   - dict, dictSize: a history buffer, which may be empty, which the sequence
+ *     producer may reference as it parses the src buffer. Currently, zstd will
+ *     always pass dictSize == 0 into external sequence producers, but this will
+ *     change in the future.
+ *
+ *   - compressionLevel: a signed integer representing the zstd compression level
+ *     set by the user for the current operation. The sequence producer may choose
+ *     to use this information to change its compression strategy and speed/ratio
+ *     tradeoff. Note: the compression level does not reflect zstd parameters set
+ *     through the advanced API.
+ *
+ *   - windowSize: a size_t representing the maximum allowed offset for external
+ *     sequences. Note that sequence offsets are sometimes allowed to exceed the
+ *     windowSize if a dictionary is present, see doc/zstd_compression_format.md
+ *     for details.
+ *
+ * The user-provided function shall return a size_t representing the number of
+ * sequences written to outSeqs. This return value will be treated as an error
+ * code if it is greater than outSeqsCapacity. The return value must be non-zero
+ * if srcSize is non-zero. The ZSTD_SEQUENCE_PRODUCER_ERROR macro is provided
+ * for convenience, but any value greater than outSeqsCapacity will be treated as
+ * an error code.
+ *
+ * If the user-provided function does not return an error code, the sequences
+ * written to outSeqs must be a valid parse of the src buffer. Data corruption may
+ * occur if the parse is not valid. A parse is defined to be valid if the
+ * following conditions hold:
+ *   - The sum of matchLengths and literalLengths must equal srcSize.
+ *   - All sequences in the parse, except for the final sequence, must have
+ *     matchLength >= ZSTD_MINMATCH_MIN. The final sequence must have
+ *     matchLength >= ZSTD_MINMATCH_MIN or matchLength == 0.
+ *   - All offsets must respect the windowSize parameter as specified in
+ *     doc/zstd_compression_format.md.
+ *   - If the final sequence has matchLength == 0, it must also have offset == 0.
+ *
+ * zstd will only validate these conditions (and fail compression if they do not
+ * hold) if the ZSTD_c_validateSequences cParam is enabled. Note that sequence
+ * validation has a performance cost.
+ *
+ * If the user-provided function returns an error, zstd will either fall back
+ * to an internal sequence producer or fail the compression operation. The user can
+ * choose between the two behaviors by setting the ZSTD_c_enableSeqProducerFallback
+ * cParam. Fallback compression will follow any other cParam settings, such as
+ * compression level, the same as in a normal compression operation.
+ *
+ * The user shall instruct zstd to use a particular ZSTD_sequenceProducer_F
+ * function by calling
+ *         ZSTD_registerSequenceProducer(cctx,
+ *                                       sequenceProducerState,
+ *                                       sequenceProducer)
+ * This setting will persist until the next parameter reset of the CCtx.
+ *
+ * The sequenceProducerState must be initialized by the user before calling
+ * ZSTD_registerSequenceProducer(). The user is responsible for destroying the
+ * sequenceProducerState.
+ *
+ * *** LIMITATIONS ***
+ * This API is compatible with all zstd compression APIs which respect advanced parameters.
+ * However, there are three limitations:
+ *
+ * First, the ZSTD_c_enableLongDistanceMatching cParam is not currently supported.
+ * COMPRESSION WILL FAIL if it is enabled and the user tries to compress with a block-level
+ * external sequence producer.
+ *   - Note that ZSTD_c_enableLongDistanceMatching is auto-enabled by default in some
+ *     cases (see its documentation for details). Users must explicitly set
+ *     ZSTD_c_enableLongDistanceMatching to ZSTD_ps_disable in such cases if an external
+ *     sequence producer is registered.
+ *   - As of this writing, ZSTD_c_enableLongDistanceMatching is disabled by default
+ *     whenever ZSTD_c_windowLog < 128MB, but that's subject to change. Users should
+ *     check the docs on ZSTD_c_enableLongDistanceMatching whenever the Block-Level Sequence
+ *     Producer API is used in conjunction with advanced settings (like ZSTD_c_windowLog).
+ *
+ * Second, history buffers are not currently supported. Concretely, zstd will always pass
+ * dictSize == 0 to the external sequence producer (for now). This has two implications:
+ *   - Dictionaries are not currently supported. Compression will *not* fail if the user
+ *     references a dictionary, but the dictionary won't have any effect.
+ *   - Stream history is not currently supported. All advanced compression APIs, including
+ *     streaming APIs, work with external sequence producers, but each block is treated as
+ *     an independent chunk without history from previous blocks.
+ *
+ * Third, multi-threading within a single compression is not currently supported. In other words,
+ * COMPRESSION WILL FAIL if ZSTD_c_nbWorkers > 0 and an external sequence producer is registered.
+ * Multi-threading across compressions is fine: simply create one CCtx per thread.
+ *
+ * Long-term, we plan to overcome all three limitations. There is no technical blocker to
+ * overcoming them. It is purely a question of engineering effort.
+ */
+
+#define ZSTD_SEQUENCE_PRODUCER_ERROR ((size_t)(-1))
+
+typedef size_t ZSTD_sequenceProducer_F (
+  void* sequenceProducerState,
+  ZSTD_Sequence* outSeqs, size_t outSeqsCapacity,
+  const void* src, size_t srcSize,
+  const void* dict, size_t dictSize,
+  int compressionLevel,
+  size_t windowSize
+);
+
+/*! ZSTD_registerSequenceProducer() :
+ * Instruct zstd to use a block-level external sequence producer function.
+ *
+ * The sequenceProducerState must be initialized by the caller, and the caller is
+ * responsible for managing its lifetime. This parameter is sticky across
+ * compressions. It will remain set until the user explicitly resets compression
+ * parameters.
+ *
+ * Sequence producer registration is considered to be an "advanced parameter",
+ * part of the "advanced API". This means it will only have an effect on compression
+ * APIs which respect advanced parameters, such as compress2() and compressStream2().
+ * Older compression APIs such as compressCCtx(), which predate the introduction of
+ * "advanced parameters", will ignore any external sequence producer setting.
+ *
+ * The sequence producer can be "cleared" by registering a NULL function pointer. This
+ * removes all limitations described above in the "LIMITATIONS" section of the API docs.
+ *
+ * The user is strongly encouraged to read the full API documentation (above) before
+ * calling this function. */
+ZSTDLIB_STATIC_API void
+ZSTD_registerSequenceProducer(
+  ZSTD_CCtx* cctx,
+  void* sequenceProducerState,
+  ZSTD_sequenceProducer_F* sequenceProducer
+);
+
+
+/*********************************************************************
+*  Buffer-less and synchronous inner streaming functions (DEPRECATED)
+*
+*  This API is deprecated, and will be removed in a future version.
+*  It allows streaming (de)compression with user allocated buffers.
+*  However, it is hard to use, and not as well tested as the rest of
+*  our API.
+*
+*  Please use the normal streaming API instead: ZSTD_compressStream2,
+*  and ZSTD_decompressStream.
+*  If there is functionality that you need, but it doesn't provide,
+*  please open an issue on our GitHub.
+********************************************************************* */
+
+/**
+  Buffer-less streaming compression (synchronous mode)
+
+  A ZSTD_CCtx object is required to track streaming operations.
+  Use ZSTD_createCCtx() / ZSTD_freeCCtx() to manage resource.
+  ZSTD_CCtx object can be re-used multiple times within successive compression operations.
+
+  Start by initializing a context.
+  Use ZSTD_compressBegin(), or ZSTD_compressBegin_usingDict() for dictionary compression.
+
+  Then, consume your input using ZSTD_compressContinue().
+  There are some important considerations to keep in mind when using this advanced function :
+  - ZSTD_compressContinue() has no internal buffer. It uses externally provided buffers only.
+  - Interface is synchronous : input is consumed entirely and produces 1+ compressed blocks.
+  - Caller must ensure there is enough space in `dst` to store compressed data under worst case scenario.
+    Worst case evaluation is provided by ZSTD_compressBound().
+    ZSTD_compressContinue() doesn't guarantee recover after a failed compression.
+  - ZSTD_compressContinue() presumes prior input ***is still accessible and unmodified*** (up to maximum distance size, see WindowLog).
+    It remembers all previous contiguous blocks, plus one separated memory segment (which can itself consists of multiple contiguous blocks)
+  - ZSTD_compressContinue() detects that prior input has been overwritten when `src` buffer overlaps.
+    In which case, it will "discard" the relevant memory section from its history.
+
+  Finish a frame with ZSTD_compressEnd(), which will write the last block(s) and optional checksum.
+  It's possible to use srcSize==0, in which case, it will write a final empty block to end the frame.
+  Without last block mark, frames are considered unfinished (hence corrupted) by compliant decoders.
+
+  `ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress again.
+*/
+
+/*=====   Buffer-less streaming compression functions  =====*/
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
+
+ZSTD_DEPRECATED("This function will likely be removed in a future release. It is misleading and has very limited utility.")
+ZSTDLIB_STATIC_API
+size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/* The ZSTD_compressBegin_advanced() and ZSTD_compressBegin_usingCDict_advanced() are now DEPRECATED and will generate a compiler warning */
+ZSTD_DEPRECATED("use advanced API to access custom parameters")
+ZSTDLIB_STATIC_API
+size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /**< pledgedSrcSize : If srcSize is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTD_DEPRECATED("use advanced API to access custom parameters")
+ZSTDLIB_STATIC_API
+size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize);   /* compression parameters are already set within cdict. pledgedSrcSize must be correct. If srcSize is not known, use macro ZSTD_CONTENTSIZE_UNKNOWN */
+/**
+  Buffer-less streaming decompression (synchronous mode)
+
+  A ZSTD_DCtx object is required to track streaming operations.
+  Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
+  A ZSTD_DCtx object can be re-used multiple times.
+
+  First typical operation is to retrieve frame parameters, using ZSTD_getFrameHeader().
+  Frame header is extracted from the beginning of compressed frame, so providing only the frame's beginning is enough.
+  Data fragment must be large enough to ensure successful decoding.
+ `ZSTD_frameHeaderSize_max` bytes is guaranteed to always be large enough.
+  result  : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
+           >0 : `srcSize` is too small, please provide at least result bytes on next attempt.
+           errorCode, which can be tested using ZSTD_isError().
+
+  It fills a ZSTD_frameHeader structure with important information to correctly decode the frame,
+  such as the dictionary ID, content size, or maximum back-reference distance (`windowSize`).
+  Note that these values could be wrong, either because of data corruption, or because a 3rd party deliberately spoofs false information.
+  As a consequence, check that values remain within valid application range.
+  For example, do not allocate memory blindly, check that `windowSize` is within expectation.
+  Each application can set its own limits, depending on local restrictions.
+  For extended interoperability, it is recommended to support `windowSize` of at least 8 MB.
+
+  ZSTD_decompressContinue() needs previous data blocks during decompression, up to `windowSize` bytes.
+  ZSTD_decompressContinue() is very sensitive to contiguity,
+  if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
+  or that previous contiguous segment is large enough to properly handle maximum back-reference distance.
+  There are multiple ways to guarantee this condition.
+
+  The most memory efficient way is to use a round buffer of sufficient size.
+  Sufficient size is determined by invoking ZSTD_decodingBufferSize_min(),
+  which can return an error code if required value is too large for current system (in 32-bits mode).
+  In a round buffer methodology, ZSTD_decompressContinue() decompresses each block next to previous one,
+  up to the moment there is not enough room left in the buffer to guarantee decoding another full block,
+  which maximum size is provided in `ZSTD_frameHeader` structure, field `blockSizeMax`.
+  At which point, decoding can resume from the beginning of the buffer.
+  Note that already decoded data stored in the buffer should be flushed before being overwritten.
+
+  There are alternatives possible, for example using two or more buffers of size `windowSize` each, though they consume more memory.
+
+  Finally, if you control the compression process, you can also ignore all buffer size rules,
+  as long as the encoder and decoder progress in "lock-step",
+  aka use exactly the same buffer sizes, break contiguity at the same place, etc.
+
+  Once buffers are setup, start decompression, with ZSTD_decompressBegin().
+  If decompression requires a dictionary, use ZSTD_decompressBegin_usingDict() or ZSTD_decompressBegin_usingDDict().
+
+  Then use ZSTD_nextSrcSizeToDecompress() and ZSTD_decompressContinue() alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how many bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() requires this _exact_ amount of bytes, or it will fail.
+
+  result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  It can be zero : it just means ZSTD_decompressContinue() has decoded some metadata item.
+  It can also be an error code, which can be tested with ZSTD_isError().
+
+  A frame is fully decoded when ZSTD_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+
+  Note : it's possible to know if next input to present is a header or a block, using ZSTD_nextInputType().
+  This information is not required to properly decode a frame.
+
+  == Special case : skippable frames ==
+
+  Skippable frames allow integration of user-defined data into a flow of concatenated frames.
+  Skippable frames will be ignored (skipped) by decompressor.
+  The format of skippable frames is as follows :
+  a) Skippable frame ID - 4 Bytes, Little endian format, any value from 0x184D2A50 to 0x184D2A5F
+  b) Frame Size - 4 Bytes, Little endian format, unsigned 32-bits
+  c) Frame Content - any content (User Data) of length equal to Frame Size
+  For skippable frames ZSTD_getFrameHeader() returns zfhPtr->frameType==ZSTD_skippableFrame.
+  For skippable frames ZSTD_decompressContinue() always returns 0 : it only skips the content.
+*/
+
+/*=====   Buffer-less streaming decompression functions  =====*/
+
+ZSTDLIB_STATIC_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
+
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
+ZSTDLIB_STATIC_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/* misc */
+ZSTD_DEPRECATED("This function will likely be removed in the next minor release. It is misleading and has very limited utility.")
+ZSTDLIB_STATIC_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
+typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
+ZSTDLIB_STATIC_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
+
+
+
+
+/* ========================================= */
+/**       Block level API (DEPRECATED)       */
+/* ========================================= */
+
+/*!
+
+    This API is deprecated in favor of the regular compression API.
+    You can get the frame header down to 2 bytes by setting:
+      - ZSTD_c_format = ZSTD_f_zstd1_magicless
+      - ZSTD_c_contentSizeFlag = 0
+      - ZSTD_c_checksumFlag = 0
+      - ZSTD_c_dictIDFlag = 0
+
+    This API is not as well tested as our normal API, so we recommend not using it.
+    We will be removing it in a future version. If the normal API doesn't provide
+    the functionality you need, please open a GitHub issue.
+
+    Block functions produce and decode raw zstd blocks, without frame metadata.
+    Frame metadata cost is typically ~12 bytes, which can be non-negligible for very small blocks (< 100 bytes).
+    But users will have to take in charge needed metadata to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Compressing and decompressing require a context structure
+      + Use ZSTD_createCCtx() and ZSTD_createDCtx()
+    - It is necessary to init context before starting
+      + compression : any ZSTD_compressBegin*() variant, including with dictionary
+      + decompression : any ZSTD_decompressBegin*() variant, including with dictionary
+    - Block size is limited, it must be <= ZSTD_getBlockSize() <= ZSTD_BLOCKSIZE_MAX == 128 KB
+      + If input is larger than a block size, it's necessary to split input data into multiple blocks
+      + For inputs larger than a single block, consider using regular ZSTD_compress() instead.
+        Frame metadata is not that costly, and quickly becomes negligible as source size grows larger than a block.
+    - When a block is considered not compressible enough, ZSTD_compressBlock() result will be 0 (zero) !
+      ===> In which case, nothing is produced into `dst` !
+      + User __must__ test for such outcome and deal directly with uncompressed data
+      + A block cannot be declared incompressible if ZSTD_compressBlock() return value was != 0.
+        Doing so would mess up with statistics history, leading to potential data corruption.
+      + ZSTD_decompressBlock() _doesn't accept uncompressed data as input_ !!
+      + In case of multiple successive blocks, should some of them be uncompressed,
+        decoder must be informed of their existence in order to follow proper history.
+        Use ZSTD_insertBlock() for such a case.
+*/
+
+/*=====   Raw zstd block functions  =====*/
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
+ZSTDLIB_STATIC_API size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
+
+#endif   /* ZSTD_H_ZSTD_STATIC_LINKING_ONLY */
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/clickhouse-cpp/contrib/zstd/zstd/zstd_errors.h b/lib/clickhouse-cpp/contrib/zstd/zstd/zstd_errors.h
new file mode 100644
index 0000000..dc75eee
--- /dev/null
+++ b/lib/clickhouse-cpp/contrib/zstd/zstd/zstd_errors.h
@@ -0,0 +1,114 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_ERRORS_H_398273423
+#define ZSTD_ERRORS_H_398273423
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*===== dependency =====*/
+#include <stddef.h>   /* size_t */
+
+
+/* =====   ZSTDERRORLIB_API : control library symbols visibility   ===== */
+#ifndef ZSTDERRORLIB_VISIBLE
+   /* Backwards compatibility with old macro name */
+#  ifdef ZSTDERRORLIB_VISIBILITY
+#    define ZSTDERRORLIB_VISIBLE ZSTDERRORLIB_VISIBILITY
+#  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDERRORLIB_VISIBLE __attribute__ ((visibility ("default")))
+#  else
+#    define ZSTDERRORLIB_VISIBLE
+#  endif
+#endif
+
+#ifndef ZSTDERRORLIB_HIDDEN
+#  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDERRORLIB_HIDDEN __attribute__ ((visibility ("hidden")))
+#  else
+#    define ZSTDERRORLIB_HIDDEN
+#  endif
+#endif
+
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZSTDERRORLIB_API __declspec(dllexport) ZSTDERRORLIB_VISIBLE
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZSTDERRORLIB_API __declspec(dllimport) ZSTDERRORLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBLE
+#endif
+
+/*-*********************************************
+ *  Error codes list
+ *-*********************************************
+ *  Error codes _values_ are pinned down since v1.3.1 only.
+ *  Therefore, don't rely on values if you may link to any version < v1.3.1.
+ *
+ *  Only values < 100 are considered stable.
+ *
+ *  note 1 : this API shall be used with static linking only.
+ *           dynamic linking is not yet officially supported.
+ *  note 2 : Prefer relying on the enum than on its value whenever possible
+ *           This is the only supported way to use the error list < v1.3.1
+ *  note 3 : ZSTD_isError() is always correct, whatever the library version.
+ **********************************************/
+typedef enum {
+  ZSTD_error_no_error = 0,
+  ZSTD_error_GENERIC  = 1,
+  ZSTD_error_prefix_unknown                = 10,
+  ZSTD_error_version_unsupported           = 12,
+  ZSTD_error_frameParameter_unsupported    = 14,
+  ZSTD_error_frameParameter_windowTooLarge = 16,
+  ZSTD_error_corruption_detected = 20,
+  ZSTD_error_checksum_wrong      = 22,
+  ZSTD_error_literals_headerWrong = 24,
+  ZSTD_error_dictionary_corrupted      = 30,
+  ZSTD_error_dictionary_wrong          = 32,
+  ZSTD_error_dictionaryCreation_failed = 34,
+  ZSTD_error_parameter_unsupported   = 40,
+  ZSTD_error_parameter_combination_unsupported = 41,
+  ZSTD_error_parameter_outOfBound    = 42,
+  ZSTD_error_tableLog_tooLarge       = 44,
+  ZSTD_error_maxSymbolValue_tooLarge = 46,
+  ZSTD_error_maxSymbolValue_tooSmall = 48,
+  ZSTD_error_stabilityCondition_notRespected = 50,
+  ZSTD_error_stage_wrong       = 60,
+  ZSTD_error_init_missing      = 62,
+  ZSTD_error_memory_allocation = 64,
+  ZSTD_error_workSpace_tooSmall= 66,
+  ZSTD_error_dstSize_tooSmall = 70,
+  ZSTD_error_srcSize_wrong    = 72,
+  ZSTD_error_dstBuffer_null   = 74,
+  ZSTD_error_noForwardProgress_destFull = 80,
+  ZSTD_error_noForwardProgress_inputEmpty = 82,
+  /* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
+  ZSTD_error_frameIndex_tooLarge = 100,
+  ZSTD_error_seekableIO          = 102,
+  ZSTD_error_dstBuffer_wrong     = 104,
+  ZSTD_error_srcBuffer_wrong     = 105,
+  ZSTD_error_sequenceProducer_failed = 106,
+  ZSTD_error_externalSequences_invalid = 107,
+  ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
+} ZSTD_ErrorCode;
+
+/*! ZSTD_getErrorCode() :
+    convert a `size_t` function result into a `ZSTD_ErrorCode` enum type,
+    which can be used to compare with enum list published above */
+ZSTDERRORLIB_API ZSTD_ErrorCode ZSTD_getErrorCode(size_t functionResult);
+ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code);   /**< Same as ZSTD_getErrorName, but using a `ZSTD_ErrorCode` enum argument */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_ERRORS_H_398273423 */
diff --git a/lib/clickhouse-cpp/tests/simple/BUCK b/lib/clickhouse-cpp/tests/simple/BUCK
deleted file mode 100644
index 37bf2ce..0000000
--- a/lib/clickhouse-cpp/tests/simple/BUCK
+++ /dev/null
@@ -1,12 +0,0 @@
-cxx_binary(
-  name = 'simple',
-  srcs = [
-    'main.cpp',
-  ],
-  compiler_flags = [
-    '-std=c++11',
-  ],
-  deps = [
-    '//:clickhouse-cpp',
-  ],
-)
diff --git a/lib/clickhouse-cpp/tests/simple/CMakeLists.txt b/lib/clickhouse-cpp/tests/simple/CMakeLists.txt
deleted file mode 100644
index e4a71e7..0000000
--- a/lib/clickhouse-cpp/tests/simple/CMakeLists.txt
+++ /dev/null
@@ -1,7 +0,0 @@
-ADD_EXECUTABLE (simple-test
-	main.cpp
-)
-
-TARGET_LINK_LIBRARIES (simple-test
-	clickhouse-cpp-lib
-)
\ No newline at end of file
diff --git a/lib/clickhouse-cpp/tests/simple/main.cpp b/lib/clickhouse-cpp/tests/simple/main.cpp
deleted file mode 100644
index 2c7b942..0000000
--- a/lib/clickhouse-cpp/tests/simple/main.cpp
+++ /dev/null
@@ -1,332 +0,0 @@
-#include <clickhouse/client.h>
-#include <clickhouse/error_codes.h>
-#include <clickhouse/types/type_parser.h>
-
-#include <iostream>
-
-#if defined(_MSC_VER)
-#   pragma warning(disable : 4996)
-#endif
-
-using namespace clickhouse;
-using namespace std;
-
-inline void PrintBlock(const Block& block) {
-    for (Block::Iterator bi(block); bi.IsValid(); bi.Next()) {
-        std::cout << bi.Name() << " ";
-    }
-    std::cout << std::endl;
-
-    for (size_t i = 0; i < block.GetRowCount(); ++i) {
-        std::cout << (*block[0]->As<ColumnUInt64>())[i] << " "
-                  << (*block[1]->As<ColumnString>())[i] << "\n";
-    }
-}
-
-inline void ArrayExample(Client& client) {
-    Block b;
-
-    /// Create a table.
-    client.Execute("CREATE TABLE IF NOT EXISTS test.array (arr Array(UInt64)) ENGINE = Memory");
-
-    auto arr = std::make_shared<ColumnArray>(std::make_shared<ColumnUInt64>());
-
-    auto id = std::make_shared<ColumnUInt64>();
-    id->Append(1);
-    arr->AppendAsColumn(id);
-
-    id->Append(3);
-    arr->AppendAsColumn(id);
-
-    id->Append(7);
-    arr->AppendAsColumn(id);
-
-    id->Append(9);
-    arr->AppendAsColumn(id);
-
-    b.AppendColumn("arr", arr);
-    client.Insert("test.array", b);
-
-
-    client.Select("SELECT arr FROM test.array", [](const Block& block)
-        {
-            for (size_t c = 0; c < block.GetRowCount(); ++c) {
-                auto col = block[0]->As<ColumnArray>()->GetAsColumn(c);
-                for (size_t i = 0; i < col->Size(); ++i) {
-                    std::cerr << (int)(*col->As<ColumnUInt64>())[i] << " ";
-                }
-                std::cerr << std::endl;
-            }
-        }
-    );
-
-    /// Delete table.
-    client.Execute("DROP TABLE test.array");
-}
-
-inline void DateExample(Client& client) {
-    Block b;
-
-    /// Create a table.
-    client.Execute("CREATE TABLE IF NOT EXISTS test.date (d DateTime) ENGINE = Memory");
-
-    auto d = std::make_shared<ColumnDateTime>();
-    d->Append(std::time(nullptr));
-    b.AppendColumn("d", d);
-    client.Insert("test.date", b);
-
-
-    client.Select("SELECT d FROM test.date", [](const Block& block)
-        {
-            for (size_t c = 0; c < block.GetRowCount(); ++c) {
-                auto col = block[0]->As<ColumnDateTime>();
-                std::time_t t = col->As<ColumnDateTime>()->At(c);
-                std::cerr << std::asctime(std::localtime(&t)) << " " << std::endl;
-            }
-        }
-    );
-
-    /// Delete table.
-    client.Execute("DROP TABLE test.date");
-}
-
-inline void GenericExample(Client& client) {
-    /// Create a table.
-    client.Execute("CREATE TABLE IF NOT EXISTS test.client (id UInt64, name String) ENGINE = Memory");
-
-    /// Insert some values.
-    {
-        Block block;
-
-        auto id = std::make_shared<ColumnUInt64>();
-        id->Append(1);
-        id->Append(7);
-
-        auto name = std::make_shared<ColumnString>();
-        name->Append("one");
-        name->Append("seven");
-
-        block.AppendColumn("id"  , id);
-        block.AppendColumn("name", name);
-
-        client.Insert("test.client", block);
-    }
-
-    /// Select values inserted in the previous step.
-    client.Select("SELECT id, name FROM test.client", [](const Block& block)
-        {
-            PrintBlock(block);
-        }
-    );
-
-    /// Delete table.
-    client.Execute("DROP TABLE test.client");
-}
-
-inline void NullableExample(Client& client) {
-    /// Create a table.
-    client.Execute("CREATE TABLE IF NOT EXISTS test.client (id Nullable(UInt64), date Nullable(Date)) ENGINE = Memory");
-
-    /// Insert some values.
-    {
-        Block block;
-
-        {
-            auto id = std::make_shared<ColumnUInt64>();
-            id->Append(1);
-            id->Append(2);
-
-            auto nulls = std::make_shared<ColumnUInt8>();
-            nulls->Append(0);
-            nulls->Append(0);
-
-            block.AppendColumn("id", std::make_shared<ColumnNullable>(id, nulls));
-        }
-
-        {
-            auto date = std::make_shared<ColumnDate>();
-            date->Append(std::time(nullptr));
-            date->Append(std::time(nullptr));
-
-            auto nulls = std::make_shared<ColumnUInt8>();
-            nulls->Append(0);
-            nulls->Append(1);
-
-            block.AppendColumn("date", std::make_shared<ColumnNullable>(date, nulls));
-        }
-
-        client.Insert("test.client", block);
-    }
-
-    /// Select values inserted in the previous step.
-    client.Select("SELECT id, date FROM test.client", [](const Block& block)
-        {
-            for (size_t c = 0; c < block.GetRowCount(); ++c) {
-                auto col_id   = block[0]->As<ColumnNullable>();
-                auto col_date = block[1]->As<ColumnNullable>();
-
-                if (col_id->IsNull(c)) {
-                    std::cerr << "\\N ";
-                } else {
-                    std::cerr << col_id->Nested()->As<ColumnUInt64>()->At(c)
-                              << " ";
-                }
-
-                if (col_date->IsNull(c)) {
-                    std::cerr << "\\N\n";
-                } else {
-                    std::time_t t = col_date->Nested()->As<ColumnDate>()->At(c);
-                    std::cerr << std::asctime(std::localtime(&t))
-                              << "\n";
-                }
-            }
-        }
-    );
-
-    /// Delete table.
-    client.Execute("DROP TABLE test.client");
-}
-
-inline void NumbersExample(Client& client) {
-    size_t num = 0;
-
-    client.Select("SELECT number, number FROM system.numbers LIMIT 100000", [&num](const Block& block)
-        {
-            if (Block::Iterator(block).IsValid()) {
-                auto col = block[0]->As<ColumnUInt64>();
-
-                for (size_t i = 0; i < col->Size(); ++i) {
-                    if (col->At(i) < num) {
-                        throw std::runtime_error("invalid sequence of numbers");
-                    }
-
-                    num = col->At(i);
-                }
-            }
-        }
-    );
-}
-
-inline void CancelableExample(Client& client) {
-    /// Create a table.
-    client.Execute("CREATE TABLE IF NOT EXISTS test.client (x UInt64) ENGINE = Memory");
-
-    /// Insert a few blocks.
-    for (unsigned j = 0; j < 10; j++) {
-        Block b;
-
-        auto x = std::make_shared<ColumnUInt64>();
-        for (uint64_t i = 0; i < 1000; i++) {
-            x->Append(i);
-        }
-
-        b.AppendColumn("x", x);
-        client.Insert("test.client", b);
-    }
-
-    /// Send a query which is canceled after receiving the first block (note:
-    /// due to the low number of rows in this test, this will not actually have
-    /// any effect, it just tests for errors)
-    client.SelectCancelable("SELECT * FROM test.client", [](const Block&)
-        {
-            return false;
-        }
-    );
-
-    /// Delete table.
-    client.Execute("DROP TABLE test.client");
-}
-
-inline void ExecptionExample(Client& client) {
-    /// Create a table.
-    client.Execute("CREATE TABLE IF NOT EXISTS test.exceptions (id UInt64, name String) ENGINE = Memory");
-    /// Expect failing on table creation.
-    try {
-        client.Execute("CREATE TABLE test.exceptions (id UInt64, name String) ENGINE = Memory");
-    } catch (const ServerException& e) {
-        if (e.GetCode() == ErrorCodes::TABLE_ALREADY_EXISTS) {
-            // OK
-        } else {
-            throw;
-        }
-    }
-
-    /// Delete table.
-    client.Execute("DROP TABLE test.exceptions");
-}
-
-inline void EnumExample(Client& client) {
-    /// Create a table.
-    client.Execute("CREATE TABLE IF NOT EXISTS test.enums (id UInt64, e Enum8('One' = 1, 'Two' = 2)) ENGINE = Memory");
-
-    /// Insert some values.
-    {
-        Block block;
-
-        auto id = std::make_shared<ColumnUInt64>();
-        id->Append(1);
-        id->Append(2);
-
-        auto e = std::make_shared<ColumnEnum8>(Type::CreateEnum8({{"One", 1}, {"Two", 2}}));
-        e->Append(1);
-        e->Append("Two");
-
-        block.AppendColumn("id", id);
-        block.AppendColumn("e", e);
-
-        client.Insert("test.enums", block);
-    }
-
-    /// Select values inserted in the previous step.
-    client.Select("SELECT id, e FROM test.enums", [](const Block& block)
-        {
-            for (Block::Iterator bi(block); bi.IsValid(); bi.Next()) {
-                std::cout << bi.Name() << " ";
-            }
-            std::cout << std::endl;
-
-            for (size_t i = 0; i < block.GetRowCount(); ++i) {
-                std::cout << (*block[0]->As<ColumnUInt64>())[i] << " "
-                          << (*block[1]->As<ColumnEnum8>()).NameAt(i) << "\n";
-            }
-        }
-    );
-
-
-    /// Delete table.
-    client.Execute("DROP TABLE test.enums");
-}
-
-static void RunTests(Client& client) {
-    ArrayExample(client);
-    DateExample(client);
-    GenericExample(client);
-    NullableExample(client);
-    NumbersExample(client);
-    CancelableExample(client);
-    ExecptionExample(client);
-    EnumExample(client);
-}
-
-int main() {
-    try {
-        {
-            Client client(ClientOptions()
-                            .SetHost("localhost")
-                            .SetPingBeforeQuery(true));
-            RunTests(client);
-        }
-
-        {
-            Client client(ClientOptions()
-                            .SetHost("localhost")
-                            .SetPingBeforeQuery(true)
-                            .SetCompressionMethod(CompressionMethod::LZ4));
-            RunTests(client);
-        }
-    } catch (const std::exception& e) {
-        std::cerr << "exception : " << e.what() << std::endl;
-    }
-
-    return 0;
-}
diff --git a/lib/clickhouse-cpp/ut/BUCK b/lib/clickhouse-cpp/ut/BUCK
deleted file mode 100644
index 136716c..0000000
--- a/lib/clickhouse-cpp/ut/BUCK
+++ /dev/null
@@ -1,12 +0,0 @@
-cxx_test(
-  name = 'ut',
-  srcs = glob([
-    '*.cpp',
-  ]),
-  compiler_flags = [
-    '-std=c++11',
-  ],
-  deps = [
-    '//:clickhouse-cpp',
-  ],
-)
diff --git a/lib/clickhouse-cpp/ut/CMakeLists.txt b/lib/clickhouse-cpp/ut/CMakeLists.txt
deleted file mode 100644
index 85c6670..0000000
--- a/lib/clickhouse-cpp/ut/CMakeLists.txt
+++ /dev/null
@@ -1,13 +0,0 @@
-ADD_EXECUTABLE (clickhouse-cpp-ut
-    main.cpp
-
-    columns_ut.cpp
-    types_ut.cpp
-    type_parser_ut.cpp
-    client_ut.cpp
-)
-
-TARGET_LINK_LIBRARIES (clickhouse-cpp-ut
-    clickhouse-cpp-lib
-    gtest-lib
-)
diff --git a/lib/clickhouse-cpp/ut/client_ut.cpp b/lib/clickhouse-cpp/ut/client_ut.cpp
deleted file mode 100644
index 2c7f0e7..0000000
--- a/lib/clickhouse-cpp/ut/client_ut.cpp
+++ /dev/null
@@ -1,372 +0,0 @@
-#include <clickhouse/client.h>
-#include <contrib/gtest/gtest.h>
-
-using namespace clickhouse;
-
-// Use value-parameterized tests to run same tests with different client
-// options.
-class ClientCase : public testing::TestWithParam<ClientOptions> {
-protected:
-    void SetUp() override {
-        client_ = new Client(GetParam());
-        client_->Execute("CREATE DATABASE IF NOT EXISTS test");
-    }
-
-    void TearDown() override {
-        client_->Execute("DROP DATABASE test");
-        delete client_;
-    }
-
-    Client* client_ = nullptr;
-};
-
-TEST_P(ClientCase, Array) {
-    Block b;
-
-    /// Create a table.
-    client_->Execute(
-            "CREATE TABLE IF NOT EXISTS test.array (arr Array(UInt64)) "
-            "ENGINE = Memory");
-
-    /// Insert some values.
-    {
-        auto arr = std::make_shared<ColumnArray>(std::make_shared<ColumnUInt64>());
-
-        auto id = std::make_shared<ColumnUInt64>();
-        id->Append(1);
-        arr->AppendAsColumn(id);
-
-        id->Append(3);
-        arr->AppendAsColumn(id);
-
-        id->Append(7);
-        arr->AppendAsColumn(id);
-
-        id->Append(9);
-        arr->AppendAsColumn(id);
-
-        b.AppendColumn("arr", arr);
-        client_->Insert("test.array", b);
-    }
-
-    const uint64_t ARR_SIZE[] = { 1, 2, 3, 4 };
-    const uint64_t VALUE[] = { 1, 3, 7, 9 };
-    size_t row = 0;
-    client_->Select("SELECT arr FROM test.array",
-            [ARR_SIZE, VALUE, &row](const Block& block)
-        {
-            if (block.GetRowCount() == 0) {
-                return;
-            }
-            EXPECT_EQ(1U, block.GetColumnCount());
-            for (size_t c = 0; c < block.GetRowCount(); ++c, ++row) {
-                auto col = block[0]->As<ColumnArray>()->GetAsColumn(c);
-                EXPECT_EQ(ARR_SIZE[row], col->Size());
-                for (size_t i = 0; i < col->Size(); ++i) {
-                    EXPECT_EQ(VALUE[i], (*col->As<ColumnUInt64>())[i]);
-                }
-            }
-        }
-    );
-
-    EXPECT_EQ(4U, row);
-}
-
-TEST_P(ClientCase, Date) {
-    Block b;
-
-    /// Create a table.
-    client_->Execute(
-            "CREATE TABLE IF NOT EXISTS test.date (d DateTime) "
-            "ENGINE = Memory");
-
-    auto d = std::make_shared<ColumnDateTime>();
-    auto const now = std::time(nullptr);
-    d->Append(now);
-    b.AppendColumn("d", d);
-    client_->Insert("test.date", b);
-
-    client_->Select("SELECT d FROM test.date", [&now](const Block& block)
-        {
-            if (block.GetRowCount() == 0) {
-                return;
-            }
-            EXPECT_EQ(1U, block.GetRowCount());
-            EXPECT_EQ(1U, block.GetColumnCount());
-            for (size_t c = 0; c < block.GetRowCount(); ++c) {
-                auto col = block[0]->As<ColumnDateTime>();
-                std::time_t t = col->As<ColumnDateTime>()->At(c);
-                EXPECT_EQ(now, t);
-            }
-        }
-    );
-}
-
-TEST_P(ClientCase, Generic) {
-    client_->Execute(
-            "CREATE TABLE IF NOT EXISTS test.client (id UInt64, name String) "
-            "ENGINE = Memory");
-
-    const struct {
-        uint64_t id;
-        std::string name;
-    } TEST_DATA[] = {
-        { 1, "id" },
-        { 3, "foo" },
-        { 5, "bar" },
-        { 7, "name" },
-    };
-
-    /// Insert some values.
-    {
-        Block block;
-
-        auto id = std::make_shared<ColumnUInt64>();
-        auto name = std::make_shared<ColumnString>();
-        for (auto const& td : TEST_DATA) {
-            id->Append(td.id);
-            name->Append(td.name);
-        }
-
-        block.AppendColumn("id"  , id);
-        block.AppendColumn("name", name);
-
-        client_->Insert("test.client", block);
-    }
-
-    /// Select values inserted in the previous step.
-    size_t row = 0;
-    client_->Select("SELECT id, name FROM test.client", [TEST_DATA, &row](const Block& block)
-        {
-            if (block.GetRowCount() == 0) {
-                return;
-            }
-            EXPECT_EQ("id", block.GetColumnName(0));
-            EXPECT_EQ("name", block.GetColumnName(1));
-            for (size_t c = 0; c < block.GetRowCount(); ++c, ++row) {
-                EXPECT_EQ(TEST_DATA[row].id, (*block[0]->As<ColumnUInt64>())[c]);
-                EXPECT_EQ(TEST_DATA[row].name, (*block[1]->As<ColumnString>())[c]);
-            }
-        }
-    );
-    EXPECT_EQ(sizeof(TEST_DATA)/sizeof(TEST_DATA[0]), row);
-}
-
-TEST_P(ClientCase, Nullable) {
-    /// Create a table.
-    client_->Execute(
-            "CREATE TABLE IF NOT EXISTS test.nullable (id Nullable(UInt64), date Nullable(Date)) "
-            "ENGINE = Memory");
-
-    // Round std::time_t to start of date.
-    const std::time_t cur_date = std::time(nullptr) / 86400 * 86400;
-    const struct {
-        uint64_t id;
-        uint8_t id_null;
-        std::time_t date;
-        uint8_t date_null;
-    } TEST_DATA[] = {
-        { 1, 0, cur_date - 2 * 86400, 0 },
-        { 2, 0, cur_date - 1 * 86400, 1 },
-        { 3, 1, cur_date + 1 * 86400, 0 },
-        { 4, 1, cur_date + 2 * 86400, 1 },
-    };
-
-    /// Insert some values.
-    {
-        Block block;
-
-        {
-            auto id = std::make_shared<ColumnUInt64>();
-            auto nulls = std::make_shared<ColumnUInt8>();
-            for (auto const& td : TEST_DATA) {
-                id->Append(td.id);
-                nulls->Append(td.id_null);
-            }
-            block.AppendColumn("id", std::make_shared<ColumnNullable>(id, nulls));
-        }
-        {
-            auto date = std::make_shared<ColumnDate>();
-            auto nulls = std::make_shared<ColumnUInt8>();
-            for (auto const& td : TEST_DATA) {
-                date->Append(td.date);
-                nulls->Append(td.date_null);
-            }
-            block.AppendColumn("date", std::make_shared<ColumnNullable>(date, nulls));
-        }
-
-        client_->Insert("test.nullable", block);
-    }
-
-    /// Select values inserted in the previous step.
-    size_t row = 0;
-    client_->Select("SELECT id, date FROM test.nullable",
-            [TEST_DATA, &row](const Block& block)
-        {
-            for (size_t c = 0; c < block.GetRowCount(); ++c, ++row) {
-                auto col_id   = block[0]->As<ColumnNullable>();
-                auto col_date = block[1]->As<ColumnNullable>();
-
-                EXPECT_EQ(static_cast<bool>(TEST_DATA[row].id_null),
-                        col_id->IsNull(c));
-                if (!col_id->IsNull(c)) {
-                    EXPECT_EQ(TEST_DATA[row].id,
-                            col_id->Nested()->As<ColumnUInt64>()->At(c));
-                }
-
-                EXPECT_EQ(static_cast<bool>(TEST_DATA[row].date_null),
-                        col_date->IsNull(c));
-                if (!col_date->IsNull(c)) {
-                    // Because date column type is Date instead of
-                    // DateTime, round to start second of date for test.
-                    EXPECT_EQ(TEST_DATA[row].date,
-                            col_date->Nested()->As<ColumnDate>()->At(c));
-                }
-            }
-        }
-    );
-
-    EXPECT_EQ(sizeof(TEST_DATA) / sizeof(TEST_DATA[0]), row);
-}
-
-TEST_P(ClientCase, Numbers) {
-    size_t num = 0;
-
-    client_->Select("SELECT number, number FROM system.numbers LIMIT 100000", [&num](const Block& block)
-        {
-            if (block.GetRowCount() == 0) {
-                return;
-            }
-            auto col = block[0]->As<ColumnUInt64>();
-
-            for (size_t i = 0; i < col->Size(); ++i, ++num) {
-                EXPECT_EQ(num, col->At(i));
-            }
-        }
-    );
-    EXPECT_EQ(100000U, num);
-}
-
-TEST_P(ClientCase, Cancelable) {
-    /// Create a table.
-    client_->Execute(
-            "CREATE TABLE IF NOT EXISTS test.cancel (x UInt64) "
-            "ENGINE = Memory");
-
-    /// Insert a few blocks. In order to make cancel have effect, we have to
-    /// insert a relative larget amount of data.
-    const int kBlock = 10;
-    const int kRowEachBlock = 1000000;
-    for (unsigned j = 0; j < kBlock; j++) {
-        Block b;
-
-        auto x = std::make_shared<ColumnUInt64>();
-        for (uint64_t i = 0; i < kRowEachBlock; i++) {
-            x->Append(i);
-        }
-
-        b.AppendColumn("x", x);
-        client_->Insert("test.cancel", b);
-    }
-
-    /// Send a query which is canceled after receiving the first blockr.
-    int row_cnt = 0;
-    EXPECT_NO_THROW(
-        client_->SelectCancelable("SELECT * FROM test.cancel",
-            [&row_cnt](const Block& block)
-            {
-                row_cnt += block.GetRowCount();
-                return false;
-            }
-        );
-    );
-    /// It's easier to get query cancelled for compress enabled client.
-    EXPECT_LE(row_cnt, kBlock * kRowEachBlock);
-}
-
-TEST_P(ClientCase, Exception) {
-    /// Create a table.
-    client_->Execute(
-            "CREATE TABLE IF NOT EXISTS test.exceptions (id UInt64, name String) "
-            "ENGINE = Memory");
-
-    /// Expect failing on table creation.
-    EXPECT_THROW(
-        client_->Execute(
-            "CREATE TABLE test.exceptions (id UInt64, name String) "
-            "ENGINE = Memory"),
-        ServerException);
-}
-
-TEST_P(ClientCase, Enum) {
-    /// Create a table.
-    client_->Execute(
-            "CREATE TABLE IF NOT EXISTS test.enums (id UInt64, e Enum8('One' = 1, 'Two' = 2)) "
-            "ENGINE = Memory");
-
-    const struct {
-        uint64_t id;
-        int8_t eval;
-        std::string ename;
-    } TEST_DATA[] = {
-        { 1, 1, "One" },
-        { 2, 2, "Two" },
-        { 3, 2, "Two" },
-        { 4, 1, "One", },
-    };
-
-    /// Insert some values.
-    {
-        Block block;
-
-        auto id = std::make_shared<ColumnUInt64>();
-        auto e = std::make_shared<ColumnEnum8>(Type::CreateEnum8({{"One", 1}, {"Two", 2}}));
-
-        int i = 0;
-        for (auto const& td : TEST_DATA) {
-            id->Append(td.id);
-            if (++i % 2) {
-                e->Append(td.eval);
-            } else {
-                e->Append(td.ename);
-            }
-        }
-
-        block.AppendColumn("id", id);
-        block.AppendColumn("e", e);
-
-        client_->Insert("test.enums", block);
-    }
-
-    /// Select values inserted in the previous step.
-    size_t row = 0;
-    client_->Select("SELECT id, e FROM test.enums", [&row, TEST_DATA](const Block& block)
-        {
-            if (block.GetRowCount() == 0) {
-                return;
-            }
-
-            EXPECT_EQ("id", block.GetColumnName(0));
-            EXPECT_EQ("e", block.GetColumnName(1));
-            for (size_t i = 0; i < block.GetRowCount(); ++i, ++row) {
-                EXPECT_EQ(TEST_DATA[row].id, (*block[0]->As<ColumnUInt64>())[i]);
-                EXPECT_EQ(TEST_DATA[row].eval, (*block[1]->As<ColumnEnum8>()).At(i));
-                EXPECT_EQ(TEST_DATA[row].ename, (*block[1]->As<ColumnEnum8>()).NameAt(i));
-            }
-        }
-    );
-    EXPECT_EQ(sizeof(TEST_DATA)/sizeof(TEST_DATA[0]), row);
-}
-
-INSTANTIATE_TEST_CASE_P(
-    Client, ClientCase,
-    ::testing::Values(
-        ClientOptions()
-            .SetHost("localhost")
-            .SetPingBeforeQuery(true),
-        ClientOptions()
-            .SetHost("localhost")
-            .SetPingBeforeQuery(false)
-            .SetCompressionMethod(CompressionMethod::LZ4)
-    ));
-
diff --git a/lib/clickhouse-cpp/ut/columns_ut.cpp b/lib/clickhouse-cpp/ut/columns_ut.cpp
deleted file mode 100644
index a874cf6..0000000
--- a/lib/clickhouse-cpp/ut/columns_ut.cpp
+++ /dev/null
@@ -1,163 +0,0 @@
-#include <clickhouse/columns/array.h>
-#include <clickhouse/columns/date.h>
-#include <clickhouse/columns/enum.h>
-#include <clickhouse/columns/nullable.h>
-#include <clickhouse/columns/numeric.h>
-#include <clickhouse/columns/string.h>
-#include <clickhouse/columns/uuid.h>
-
-#include <contrib/gtest/gtest.h>
-
-using namespace clickhouse;
-
-static std::vector<uint32_t> MakeNumbers() {
-    return std::vector<uint32_t>
-        {1, 2, 3, 7, 11, 13, 17, 19, 23, 29, 31};
-}
-
-static std::vector<uint8_t> MakeBools() {
-    return std::vector<uint8_t>
-        {1, 0, 0, 0, 1, 1, 0, 1, 1, 1, 0};
-}
-
-static std::vector<std::string> MakeFixedStrings() {
-    return std::vector<std::string>
-        {"aaa", "bbb", "ccc", "ddd"};
-}
-
-static std::vector<std::string> MakeStrings() {
-    return std::vector<std::string>
-        {"a", "ab", "abc", "abcd"};
-}
-
-static std::vector<uint64_t> MakeUUIDs() {
-    return std::vector<uint64_t>
-        {0xbb6a8c699ab2414cllu, 0x86697b7fd27f0825llu,
-         0x84b9f24bc26b49c6llu, 0xa03b4ab723341951llu,
-         0x3507213c178649f9llu, 0x9faf035d662f60aellu};
-}
-
-
-TEST(ColumnsCase, NumericInit) {
-    auto col = std::make_shared<ColumnUInt32>(MakeNumbers());
-
-    ASSERT_EQ(col->Size(), 11u);
-    ASSERT_EQ(col->At(3),   7u);
-    ASSERT_EQ(col->At(10), 31u);
-
-    auto sun = std::make_shared<ColumnUInt32>(MakeNumbers());
-}
-
-TEST(ColumnsCase, NumericSlice) {
-    auto col = std::make_shared<ColumnUInt32>(MakeNumbers());
-    auto sub = col->Slice(3, 3)->As<ColumnUInt32>();
-
-    ASSERT_EQ(sub->Size(), 3u);
-    ASSERT_EQ(sub->At(0),  7u);
-    ASSERT_EQ(sub->At(2), 13u);
-}
-
-
-TEST(ColumnsCase, FixedStringInit) {
-    auto col = std::make_shared<ColumnFixedString>(3);
-    for (const auto& s : MakeFixedStrings()) {
-        col->Append(s);
-    }
-
-    ASSERT_EQ(col->Size(), 4u);
-    ASSERT_EQ(col->At(1), "bbb");
-    ASSERT_EQ(col->At(3), "ddd");
-}
-
-TEST(ColumnsCase, StringInit) {
-    auto col = std::make_shared<ColumnString>(MakeStrings());
-
-    ASSERT_EQ(col->Size(), 4u);
-    ASSERT_EQ(col->At(1), "ab");
-    ASSERT_EQ(col->At(3), "abcd");
-}
-
-
-TEST(ColumnsCase, ArrayAppend) {
-    auto arr1 = std::make_shared<ColumnArray>(std::make_shared<ColumnUInt64>());
-    auto arr2 = std::make_shared<ColumnArray>(std::make_shared<ColumnUInt64>());
-
-    auto id = std::make_shared<ColumnUInt64>();
-    id->Append(1);
-    arr1->AppendAsColumn(id);
-
-    id->Append(3);
-    arr2->AppendAsColumn(id);
-
-    arr1->Append(arr2);
-
-    auto col = arr1->GetAsColumn(1);
-
-    ASSERT_EQ(arr1->Size(), 2u);
-    //ASSERT_EQ(col->As<ColumnUInt64>()->At(0), 1u);
-    //ASSERT_EQ(col->As<ColumnUInt64>()->At(1), 3u);
-}
-
-TEST(ColumnsCase, DateAppend) {
-    auto col1 = std::make_shared<ColumnDate>();
-    auto col2 = std::make_shared<ColumnDate>();
-    auto now  = std::time(nullptr);
-
-    col1->Append(now);
-    col2->Append(col1);
-
-    ASSERT_EQ(col2->Size(), 1u);
-    ASSERT_EQ(col2->At(0), (now / 86400) * 86400);
-}
-
-TEST(ColumnsCase, EnumTest) {
-    std::vector<Type::EnumItem> enum_items = {{"Hi", 1}, {"Hello", 2}};
-
-    auto col = std::make_shared<ColumnEnum8>(Type::CreateEnum8(enum_items));
-    ASSERT_TRUE(col->Type()->IsEqual(Type::CreateEnum8(enum_items)));
-
-    col->Append(1);
-    ASSERT_EQ(col->Size(), 1u);
-    ASSERT_EQ(col->At(0), 1);
-    ASSERT_EQ(col->NameAt(0), "Hi");
-
-    col->Append("Hello");
-    ASSERT_EQ(col->Size(), 2u);
-    ASSERT_EQ(col->At(1), 2);
-    ASSERT_EQ(col->NameAt(1), "Hello");
-
-    auto col16 = std::make_shared<ColumnEnum16>(Type::CreateEnum16(enum_items));
-    ASSERT_TRUE(col16->Type()->IsEqual(Type::CreateEnum16(enum_items)));
-}
-
-TEST(ColumnsCase, NullableSlice) {
-    auto data = std::make_shared<ColumnUInt32>(MakeNumbers());
-    auto nulls = std::make_shared<ColumnUInt8>(MakeBools());
-    auto col = std::make_shared<ColumnNullable>(data, nulls);
-    auto sub = col->Slice(3, 4)->As<ColumnNullable>();
-    auto subData = sub->Nested()->As<ColumnUInt32>();
-
-    ASSERT_EQ(sub->Size(), 4u);
-    ASSERT_FALSE(sub->IsNull(0));
-    ASSERT_EQ(subData->At(0),  7u);
-    ASSERT_TRUE(sub->IsNull(1));
-    ASSERT_FALSE(sub->IsNull(3));
-    ASSERT_EQ(subData->At(3), 17u);
-}
-
-TEST(ColumnsCase, UUIDInit) {
-    auto col = std::make_shared<ColumnUUID>(std::make_shared<ColumnUInt64>(MakeUUIDs()));
-
-    ASSERT_EQ(col->Size(), 3u);
-    ASSERT_EQ(col->At(0), UInt128(0xbb6a8c699ab2414cllu, 0x86697b7fd27f0825llu));
-    ASSERT_EQ(col->At(2), UInt128(0x3507213c178649f9llu, 0x9faf035d662f60aellu));
-}
-
-TEST(ColumnsCase, UUIDSlice) {
-    auto col = std::make_shared<ColumnUUID>(std::make_shared<ColumnUInt64>(MakeUUIDs()));
-    auto sub = col->Slice(1, 2)->As<ColumnUUID>();
-
-    ASSERT_EQ(sub->Size(), 2u);
-    ASSERT_EQ(sub->At(0), UInt128(0x84b9f24bc26b49c6llu, 0xa03b4ab723341951llu));
-    ASSERT_EQ(sub->At(1), UInt128(0x3507213c178649f9llu, 0x9faf035d662f60aellu));
-}
diff --git a/lib/clickhouse-cpp/ut/main.cpp b/lib/clickhouse-cpp/ut/main.cpp
deleted file mode 100644
index a500047..0000000
--- a/lib/clickhouse-cpp/ut/main.cpp
+++ /dev/null
@@ -1,6 +0,0 @@
-#include <contrib/gtest/gtest.h>
-
-int main(int argc, char* argv[]) {
-    ::testing::InitGoogleTest(&argc, argv);
-    return RUN_ALL_TESTS();
-}
diff --git a/lib/clickhouse-cpp/ut/type_parser_ut.cpp b/lib/clickhouse-cpp/ut/type_parser_ut.cpp
deleted file mode 100644
index 26b51b2..0000000
--- a/lib/clickhouse-cpp/ut/type_parser_ut.cpp
+++ /dev/null
@@ -1,67 +0,0 @@
-#include <clickhouse/types/type_parser.h>
-#include <contrib/gtest/gtest.h>
-
-using namespace clickhouse;
-
-TEST(TypeParserCase, ParseTerminals) {
-    TypeAst ast;
-    TypeParser("UInt8").Parse(&ast);
-
-    ASSERT_EQ(ast.meta, TypeAst::Terminal);
-    ASSERT_EQ(ast.name, "UInt8");
-    ASSERT_EQ(ast.code, Type::UInt8);
-}
-
-TEST(TypeParserCase, ParseFixedString) {
-    TypeAst ast;
-    TypeParser("FixedString(24)").Parse(&ast);
-
-    ASSERT_EQ(ast.meta, TypeAst::Terminal);
-    ASSERT_EQ(ast.name, "FixedString");
-    ASSERT_EQ(ast.code, Type::FixedString);
-    ASSERT_EQ(ast.elements.front().value, 24U);
-}
-
-TEST(TypeParserCase, ParseArray) {
-    TypeAst ast;
-    TypeParser("Array(Int32)").Parse(&ast);
-
-    ASSERT_EQ(ast.meta, TypeAst::Array);
-    ASSERT_EQ(ast.name, "Array");
-    ASSERT_EQ(ast.code, Type::Array);
-    ASSERT_EQ(ast.elements.front().meta, TypeAst::Terminal);
-    ASSERT_EQ(ast.elements.front().name, "Int32");
-}
-
-TEST(TypeParserCase, ParseNullable) {
-    TypeAst ast;
-    TypeParser("Nullable(Date)").Parse(&ast);
-
-    ASSERT_EQ(ast.meta, TypeAst::Nullable);
-    ASSERT_EQ(ast.name, "Nullable");
-    ASSERT_EQ(ast.code, Type::Nullable);
-    ASSERT_EQ(ast.elements.front().meta, TypeAst::Terminal);
-    ASSERT_EQ(ast.elements.front().name, "Date");
-}
-
-TEST(TypeParserCase, ParseEnum) {
-    TypeAst ast;
-    TypeParser(
-        "Enum8('COLOR_red_10_T' = -12, 'COLOR_green_20_T'=-25, 'COLOR_blue_30_T'= 53, 'COLOR_black_30_T' = 107")
-        .Parse(&ast);
-    ASSERT_EQ(ast.meta, TypeAst::Enum);
-    ASSERT_EQ(ast.name, "Enum8");
-    ASSERT_EQ(ast.code, Type::Enum8);
-    ASSERT_EQ(ast.elements.size(), 4u);
-
-    std::vector<std::string> names = {"COLOR_red_10_T", "COLOR_green_20_T", "COLOR_blue_30_T", "COLOR_black_30_T"};
-    std::vector<int16_t> values = {-12, -25, 53, 107};
-
-    auto element = ast.elements.begin();
-    for (size_t i = 0; i < 4; ++i) {
-        ASSERT_EQ(element->name, names[i]);
-        ASSERT_EQ(element->code, Type::Void);
-        ASSERT_EQ(element->value, values[i]);
-        ++element;
-    }
-}
diff --git a/lib/clickhouse-cpp/ut/types_ut.cpp b/lib/clickhouse-cpp/ut/types_ut.cpp
deleted file mode 100644
index d3ee287..0000000
--- a/lib/clickhouse-cpp/ut/types_ut.cpp
+++ /dev/null
@@ -1,67 +0,0 @@
-#include <clickhouse/types/types.h>
-#include <contrib/gtest/gtest.h>
-
-using namespace clickhouse;
-
-TEST(TypesCase, TypeName) {
-    ASSERT_EQ(
-        Type::CreateDate()->GetName(),
-        "Date"
-    );
-
-    ASSERT_EQ(
-        Type::CreateArray(Type::CreateSimple<int32_t>())->GetName(),
-        "Array(Int32)"
-    );
-
-    ASSERT_EQ(
-        Type::CreateNullable(Type::CreateSimple<int32_t>())->GetName(),
-        "Nullable(Int32)"
-    );
-
-    ASSERT_EQ(
-        Type::CreateArray(Type::CreateSimple<int32_t>())->GetItemType()->GetCode(),
-        Type::Int32
-    );
-
-    ASSERT_EQ(
-        Type::CreateTuple({
-            Type::CreateSimple<int32_t>(),
-            Type::CreateString()})->GetName(),
-        "Tuple(Int32, String)"
-    );
-}
-
-TEST(TypesCase, NullableType) {
-    TypeRef nested = Type::CreateSimple<int32_t>();
-    ASSERT_EQ(
-        Type::CreateNullable(nested)->GetNestedType(),
-        nested
-    );
-}
-
-TEST(TypesCase, EnumTypes) {
-    EnumType enum8(Type::CreateEnum8({{"One", 1}, {"Two", 2}}));
-    ASSERT_EQ(enum8.GetName(), "Enum8('One' = 1, 'Two' = 2)");
-    ASSERT_TRUE(enum8.HasEnumValue(1));
-    ASSERT_TRUE(enum8.HasEnumName("Two"));
-    ASSERT_FALSE(enum8.HasEnumValue(10));
-    ASSERT_FALSE(enum8.HasEnumName("Ten"));
-    ASSERT_EQ(enum8.GetEnumName(2), "Two");
-    ASSERT_EQ(enum8.GetEnumValue("Two"), 2);
-
-    EnumType enum16(Type::CreateEnum16({{"Green", 1}, {"Red", 2}, {"Yellow", 3}}));
-    ASSERT_EQ(enum16.GetName(), "Enum16('Green' = 1, 'Red' = 2, 'Yellow' = 3)");
-    ASSERT_TRUE(enum16.HasEnumValue(3));
-    ASSERT_TRUE(enum16.HasEnumName("Green"));
-    ASSERT_FALSE(enum16.HasEnumValue(10));
-    ASSERT_FALSE(enum16.HasEnumName("Black"));
-    ASSERT_EQ(enum16.GetEnumName(2), "Red");
-    ASSERT_EQ(enum16.GetEnumValue("Green"), 1);
-
-    ASSERT_EQ(std::distance(enum16.BeginValueToName(), enum16.EndValueToName()), 3u);
-    ASSERT_EQ((*enum16.BeginValueToName()).first, 1);
-    ASSERT_EQ((*enum16.BeginValueToName()).second, "Green");
-    ASSERT_EQ((*(++enum16.BeginValueToName())).first, 2);
-    ASSERT_EQ((*(++enum16.BeginValueToName())).second, "Red");
-}
diff --git a/package.xml b/package.xml
deleted file mode 100644
index 83be4e8..0000000
--- a/package.xml
+++ /dev/null
@@ -1,171 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<package packagerversion="1.9.4" version="2.0" xmlns="http://pear.php.net/dtd/package-2.0" xmlns:tasks="http://pear.php.net/dtd/tasks-1.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://pear.php.net/dtd/tasks-1.0 http://pear.php.net/dtd/tasks-1.0.xsd http://pear.php.net/dtd/package-2.0 http://pear.php.net/dtd/package-2.0.xsd">
-    <name>SeasClick</name>
-    <channel>pecl.php.net</channel>
-    <summary>An Yandex ClickHouse client driven extension for PHP.</summary>
-    <description>
-        PHP client for Yandex ClickHouse，Based on ClickHouse C++ client.
-    </description>
-    <lead>
-        <name>Chitao Gao</name>
-        <user>neeke</user>
-        <email>neeke@php.net</email>
-        <active>yes</active>
-    </lead>
-    <date>2023-02-24</date>
-    <time>13:00:00</time>
-    <version>
-        <release>0.1.1</release>
-        <api>0.1.1</api>
-    </version>
-    <stability>
-        <release>stable</release>
-        <api>stable</api>
-    </stability>
-    <license uri="http://www.php.net/license/3_01.txt">PHP3.01</license>
-    <notes>
-        - Support PHP8.X version for pecl.net
-    </notes>
-    <contents>
-        <dir name="/">
-            <file role="doc" name="CREDITS" />
-            <file role="doc" name="README.md" />
-            <file role="doc" name="LICENSE" />
-            <file role="src" name="config.m4" />
-            <file role="src" name="config.w32.without" />
-            <file role="src" name="EXPERIMENTAL" />
-            <file role="src" name="php_SeasClick.h" />
-            <file role="src" name="php7_wrapper.h" />
-            <file role="src" name="SeasClick.cpp" />
-            <file role="src" name="typesToPhp.cpp" />
-            <file role="src" name="typesToPhp.hpp" />
-            <dir name="tests">
-                <file role="src" name="001.phpt" />
-                <file role="src" name="002.phpt" />
-                <file role="src" name="003.phpt" />
-                <file role="src" name="004.phpt" />
-                <file role="src" name="005.phpt" />
-                <file role="src" name="006.phpt" />
-                <file role="src" name="007.phpt" />
-                <file role="src" name="008.phpt" />
-                <file role="src" name="009.phpt" />
-                <file role="src" name="test.php" />
-                <dir name="bench_mark">
-                    <file role="src" name="bench_mark.php" />
-                    <file role="src" name="bench_mark.png" />
-                    <file role="doc" name="README.md" />
-                    <file role="src" name="composer.json" />
-                </dir>
-            </dir>
-            <dir name="lib">
-                <dir name="clickhouse-cpp">
-                    <file role="src" name=".buckconfig" />
-                    <file role="src" name="BUCK" />
-                    <file role="doc" name="CMakeLists.txt" />
-                    <file role="doc" name="LICENSE" />
-                    <file role="doc" name="README.md" />
-                    <dir name="clickhouse">
-                        <dir name="base">
-                            <file role="src" name="buffer.h" />
-                            <file role="src" name="coded.cpp" />
-                            <file role="src" name="coded.h" />
-                            <file role="src" name="compressed.cpp" />
-                            <file role="src" name="compressed.h" />
-                            <file role="src" name="input.cpp" />
-                            <file role="src" name="input.h" />
-                            <file role="src" name="output.cpp" />
-                            <file role="src" name="output.h" />
-                            <file role="src" name="platform.cpp" />
-                            <file role="src" name="platform.h" />
-                            <file role="src" name="singleton.h" />
-                            <file role="src" name="socket.cpp" />
-                            <file role="src" name="socket.h" />
-                            <file role="src" name="string_utils.h" />
-                            <file role="src" name="string_view.h" />
-                            <file role="src" name="wire_format.h" />
-                        </dir>
-                        <dir name="columns">
-                            <file role="src" name="array.cpp" />
-                            <file role="src" name="array.h" />
-                            <file role="src" name="column.h" />
-                            <file role="src" name="date.cpp" />
-                            <file role="src" name="date.h" />
-                            <file role="src" name="enum.cpp" />
-                            <file role="src" name="enum.h" />
-                            <file role="src" name="factory.cpp" />
-                            <file role="src" name="factory.h" />
-                            <file role="src" name="nullable.cpp" />
-                            <file role="src" name="nullable.h" />
-                            <file role="src" name="numeric.cpp" />
-                            <file role="src" name="numeric.h" />
-                            <file role="src" name="string.cpp" />
-                            <file role="src" name="string.h" />
-                            <file role="src" name="tuple.cpp" />
-                            <file role="src" name="tuple.h" />
-                            <file role="src" name="utils.h" />
-                            <file role="src" name="uuid.cpp" />
-                            <file role="src" name="uuid.h" />
-                        </dir>
-                        <dir name="types">
-                            <file role="src" name="type_parser.cpp" />
-                            <file role="src" name="type_parser.h" />
-                            <file role="src" name="types.cpp" />
-                            <file role="src" name="types.h" />
-                        </dir>
-                        <file role="src" name="block.cpp" />
-                        <file role="src" name="block.h" />
-                        <file role="src" name="client.cpp" />
-                        <file role="src" name="client.h" />
-                        <file role="src" name="error_codes.h" />
-                        <file role="src" name="exceptions.h" />
-                        <file role="src" name="protocol.h" />
-                        <file role="src" name="query.cpp" />
-                        <file role="src" name="query.h" />
-                    </dir>
-                    <dir name="contrib">
-                        <dir name="cityhash">
-                            <file role="src" name="city.cc" />
-                            <file role="src" name="city.h" />
-                            <file role="src" name="citycrc.h" />
-                            <file role="src" name="config.h" />
-                        </dir>
-                        <dir name="gtest">
-                            <file role="src" name="gtest.h" />
-                            <file role="src" name="gtest-all.cc" />
-                        </dir>
-                        <dir name="lz4">
-                            <file role="src" name="LICENSE" />
-                            <file role="src" name="lz4.c" />
-                            <file role="src" name="lz4.h" />
-                            <file role="src" name="lz4hc.c" />
-                            <file role="src" name="lz4hc.h" />
-                        </dir>
-                    </dir>
-                    <dir name="ut">
-                        <file role="src" name="client_ut.cpp" />
-                        <file role="src" name="columns_ut.cpp" />
-                        <file role="src" name="main.cpp" />
-                        <file role="src" name="type_parser_ut.cpp" />
-                        <file role="src" name="types_ut.cpp" />
-                    </dir>
-                </dir>
-            </dir>
-        </dir>
-    </contents>
-    <dependencies>
-        <required>
-            <php>
-                <min>7.0.0</min>
-            </php>
-            <pearinstaller>
-                <min>1.4.0</min>
-            </pearinstaller>
-        </required>
-    </dependencies>
-    <providesextension>SeasClick</providesextension>
-    <extsrcrelease/>
-    <changelog>
-        - Support PHP8.X version for pecl.net
-    </changelog>
-</package>
-
diff --git a/php7_wrapper.h b/php7_wrapper.h
index d07262b..34ba632 100644
--- a/php7_wrapper.h
+++ b/php7_wrapper.h
@@ -1,8 +1,8 @@
 /*
   +----------------------------------------------------------------------+
-  | SeasClick                                                            |
+  | php_clickhouse                                                       |
   +----------------------------------------------------------------------+
-  | Copyright (c) 1997-2018 The PHP Group                                |
+  | Copyright (c) 1997-2026 The PHP Group                                |
   +----------------------------------------------------------------------+
   | This source file is subject to version 3.01 of the PHP license,      |
   | that is bundled with this package in the file LICENSE, and is        |
@@ -12,115 +12,16 @@
   | obtain it through the world-wide-web, please send a note to          |
   | license@php.net so we can mail you a copy immediately.               |
   +----------------------------------------------------------------------+
-  | Author:  SeasX Group <ahhhh.wang@gmail.com>                          |
+  | Author: Ilia Alshanetsky <ilia@ilia.ws>                              |
   +----------------------------------------------------------------------+
 */
 
+#define sc_zend_throw_exception_tsrmls_cc(a, b, c) zend_throw_exception(a, b, c)
 
-// PHP7.4 +  
-#if !defined(ZEND_ACC_IMPLICIT_PUBLIC)
-# define ZEND_ACC_IMPLICIT_PUBLIC ZEND_ACC_PUBLIC
-#endif
-
-
-// PHP8+  
-#if !defined(ZEND_ACC_DTOR)
-#define ZEND_ACC_DTOR       0x4000
-#endif
-
-// PHP5+  
-#if PHP_MAJOR_VERSION <7
-
-#if PHP_VERSION_ID < 50500
-#define sc_zend_throw_exception(a, b, c) zend_throw_exception(a, (char *)b, c)
-#else
-#define sc_zend_throw_exception zend_throw_exception
-#endif
-
-#define sc_zend_throw_exception_tsrmls_cc sc_zend_throw_exception
-#define IS_TRUE                               1
-#define SC_MAKE_STD_ZVAL(p)                   MAKE_STD_ZVAL(p)
-#define SC_RETURN_STRINGL(k, l) RETURN_STRINGL(k, l, 1)
-#define sc_zval_ptr_dtor                      zval_ptr_dtor
-#define sc_zval_add_ref(a)                       zval_add_ref(&a)
-static inline int sc_add_assoc_long_ex(zval *arg, const char *key, size_t key_len, long value)
-{
-    return add_assoc_long_ex(arg, key, key_len + 1, value);
-}
-
-static inline int sc_add_assoc_double_ex(zval *arg, const char *key, size_t key_len, double value)
-{
-    return add_assoc_double_ex(arg, key, key_len + 1, value);
-}
-
-static inline int sc_add_assoc_zval_ex(zval *arg, const char *key, size_t key_len, zval* value)
-{
-    return add_assoc_zval_ex(arg, key, key_len + 1, value);
-}
-
-static inline int sc_add_assoc_stringl_ex(zval *arg, const char *key, size_t key_len, char *str, size_t length, int __duplicate)
-{
-    return add_assoc_stringl_ex(arg, key, key_len + 1, str, length, __duplicate);
-}
-
-static inline int sc_add_assoc_null_ex(zval *arg, const char *key, size_t key_len)
-{
-    return add_assoc_null_ex(arg, key, key_len + 1);
-}
-
-static inline zval *sc_zend_hash_find(HashTable *ht, char *k, int len)
-{
-    zval **tmp = NULL;
-    if (zend_hash_find(ht, k, len + 1, (void **) &tmp) == SUCCESS)
-    {
-        return *tmp;
-    }
-    else
-    {
-        return NULL;
-    }
-}
-
-static inline zval *sc_zend_hash_index_find(HashTable *ht, ulong h)
-{
-    zval **tmp = NULL;
-    if (zend_hash_index_find(ht, h, (void **) &tmp) == SUCCESS)
-    {
-        return *tmp;
-    }
-    else
-    {
-        return NULL;
-    }
-}
-
-#define sc_zend_update_property_string    zend_update_property_string
-
-#define sc_zend_read_property(a, b, c, d, e)  zend_read_property(a, b, c, d, e TSRMLS_CC)
-
-#define SC_HASHTABLE_FOREACH_START2(ht, k, klen, ktype, entry)\
-    zval **tmp = NULL; ulong_t idx;\
-    for (zend_hash_internal_pointer_reset(ht); \
-            (ktype = zend_hash_get_current_key_ex(ht, &k, &klen, &idx, 0, NULL)) != HASH_KEY_NON_EXISTANT; \
-            zend_hash_move_forward(ht)\
-        ) { \
-    if (zend_hash_get_current_data(ht, (void**)&tmp) == FAILURE) {\
-        continue;\
-    }\
-    entry = *tmp;\
-    klen --;
-
-#define SC_HASHTABLE_FOREACH_END() }
-
-#define sc_add_next_index_stringl             add_next_index_stringl
-#define sc_zend_hash_get_current_data         zend_hash_get_current_data
-#else
-// PHP7
-#define sc_zend_throw_exception zend_throw_exception
+#define sc_zend_hash_index_update   zend_hash_index_update
 #define sc_zend_hash_find   zend_hash_str_find
 #define sc_zend_hash_index_find   zend_hash_index_find
 #define SC_MAKE_STD_ZVAL(p)             zval _stack_zval_##p; p = &(_stack_zval_##p)
-#define SC_RETURN_STRINGL(k, l) RETURN_STRINGL(k, l)
 #define sc_zval_ptr_dtor(p)  zval_ptr_dtor(*p)
 #define sc_zval_add_ref(p)   Z_TRY_ADDREF_P(p)
 #define sc_add_assoc_long_ex                  add_assoc_long_ex
@@ -129,70 +30,43 @@ static inline zval *sc_zend_hash_index_find(HashTable *ht, ulong h)
 #define sc_add_assoc_stringl_ex(a, b, c, d, e, f)               add_assoc_stringl_ex(a, b, c, d, e)
 #define sc_add_assoc_null_ex(a, b, c)               add_assoc_null_ex(a, b, c)
 
-
-
-#if PHP_VERSION_ID < 80000
-#define sc_zend_throw_exception_tsrmls_cc(a, b, c) sc_zend_throw_exception(a, b, c TSRMLS_CC)
-#else
-#define sc_zend_throw_exception_tsrmls_cc(a, b, c) sc_zend_throw_exception(a, b, c)
-#endif
-
-
 static inline zval* sc_zend_read_property(zend_class_entry *class_ptr, zval *obj, const char *s, int len, int silent)
 {
     zval rv;
 #if PHP_VERSION_ID < 80000
-     return zend_read_property(class_ptr, obj, s, len, silent, &rv);
+    return zend_read_property(class_ptr, obj, s, len, silent, &rv);
 #else
-    zend_object *zendObject;
-    zendObject=Z_OBJ_P(obj);
-     return zend_read_property(class_ptr, zendObject, s, len, silent, &rv);
+    return zend_read_property(class_ptr, Z_OBJ_P(obj), s, len, silent, &rv);
 #endif
-   
 }
 
-
-static  inline void sc_zend_update_property_string( zend_class_entry *scope, zval *object, const char *name, size_t name_length, const char *value)
+static inline void sc_zend_update_property_string(zend_class_entry *scope, zval *object, const char *name, size_t name_length, const char *value)
 {
 #if PHP_VERSION_ID < 80000
-     return zend_update_property_string(scope, object, name, name_length,  value TSRMLS_CC);
+    zend_update_property_string(scope, object, name, name_length, value);
 #else
-    zend_object *zendObject;
-    zendObject=Z_OBJ_P(object);
-    return zend_update_property_string(scope, zendObject, name, name_length,  value);
+    zend_update_property_string(scope, Z_OBJ_P(object), name, name_length, value);
 #endif
 }
 
-
-static  inline void sc_zend_update_property_long(zend_class_entry *scope, zval *object, const char *name, size_t name_length, zend_long value) 
+static inline void sc_zend_update_property_long(zend_class_entry *scope, zval *object, const char *name, size_t name_length, zend_long value)
 {
 #if PHP_VERSION_ID < 80000
-     return zend_update_property_long(scope, object, name, name_length,  value TSRMLS_CC);
+    zend_update_property_long(scope, object, name, name_length, value);
 #else
-    zend_object *zendObject;
-    zendObject=Z_OBJ_P(object);
-    return zend_update_property_long(scope, zendObject, name, name_length,  value);
+    zend_update_property_long(scope, Z_OBJ_P(object), name, name_length, value);
 #endif
 }
 
-static  inline void sc_zend_update_property_bool(zend_class_entry *scope, zval *object, const char *name, size_t name_length, zend_long value) /* {{{ */
+static inline void sc_zend_update_property_stringl(zend_class_entry *scope, zval *object, const char *name, size_t name_length, const char *value, size_t value_length)
 {
 #if PHP_VERSION_ID < 80000
-     return zend_update_property_bool(scope, object, name, name_length,  value TSRMLS_CC);
+    zend_update_property_stringl(scope, object, name, name_length, value, value_length);
 #else
-    zend_object *zendObject;
-    zendObject=Z_OBJ_P(object);
-    return zend_update_property_bool(scope, zendObject, name, name_length,  value);
+    zend_update_property_stringl(scope, Z_OBJ_P(object), name, name_length, value, value_length);
 #endif
 }
 
-#define SC_HASHTABLE_FOREACH_START2(ht, k, klen, ktype, _val) zend_string *_foreach_key;\
-    ZEND_HASH_FOREACH_STR_KEY_VAL(ht, _foreach_key, _val);\
-    if (!_foreach_key) {k = NULL; klen = 0; ktype = 0;}\
-    else {k = _foreach_key->val, klen=_foreach_key->len; ktype = 1;} {
-
-#define SC_HASHTABLE_FOREACH_END()                 } ZEND_HASH_FOREACH_END();
-
 #define sc_add_next_index_stringl(arr, str, len, dup)    add_next_index_stringl(arr, str, len)
 
 static inline int sc_zend_hash_get_current_data(HashTable *ht, void **v)
@@ -202,15 +76,117 @@ static inline int sc_zend_hash_get_current_data(HashTable *ht, void **v)
     {
         return FAILURE;
     }
-    else
-    {
-        *v = (void *) value;
-        return SUCCESS;
+    *v = (void *) value;
+    return SUCCESS;
+}
+
+#define php_array_get_value(ht, str, v) ((v = sc_zend_hash_find(ht, (char *)str, sizeof(str)-1)) && !ZVAL_IS_NULL(v))
+
+/* FAST_ZPP _OR_NULL convenience macros are PHP 8.0+. Shim for 7.4 with
+ * the older Z_PARAM_*_EX(dest, check_null, separate) form. */
+#if PHP_VERSION_ID < 80000
+#  ifndef Z_PARAM_STR_OR_NULL
+#    define Z_PARAM_STR_OR_NULL(dest)   Z_PARAM_STR_EX(dest, 1, 0)
+#  endif
+#  ifndef Z_PARAM_ARRAY_OR_NULL
+#    define Z_PARAM_ARRAY_OR_NULL(dest) Z_PARAM_ARRAY_EX(dest, 1, 0)
+#  endif
+#  ifndef Z_PARAM_ZVAL_OR_NULL
+#    define Z_PARAM_ZVAL_OR_NULL(dest)  Z_PARAM_ZVAL_EX(dest, 1, 0)
+#  endif
+#endif
+
+/*
+ * gen_stub.php on PHP master emits typed-parameter, typed-return-value, and
+ * typed-class-constant macros that don't exist on older PHP. Shim them to
+ * pre-typed equivalents so the generated arginfo header compiles unchanged
+ * across the entire build matrix (PHP 7.4 through 8.5).
+ *
+ * The shims drop type information rather than emulate it. On builds older
+ * than the threshold for each shim, reflection signatures revert to untyped
+ * (parameter and return types disappear; typed class constants become
+ * untyped). Runtime behavior is unchanged; only the introspection surface
+ * is reduced.
+ *
+ * If gen_stub.php starts emitting more 8.x-only macros, extend this block
+ * rather than narrowing the build matrix.
+ */
+
+/* PHP 8.4: zend_register_internal_class_with_flags rolls class registration
+ * and flag setting into one call. Pre-8.4 splits them. */
+#if PHP_VERSION_ID < 80400
+static zend_always_inline zend_class_entry *zend_register_internal_class_with_flags(
+    zend_class_entry *class_entry,
+    zend_class_entry *parent_ce,
+    uint32_t ce_flags)
+{
+    zend_class_entry *ce = zend_register_internal_class_ex(class_entry, parent_ce);
+    if (ce && ce_flags) {
+        ce->ce_flags |= ce_flags;
     }
+    return ce;
 }
 #endif
 
-#define php_array_get_value(ht, str, v) ((v = sc_zend_hash_find(ht, (char *)str, sizeof(str)-1)) && !ZVAL_IS_NULL(v))
+/* PHP 8.3: typed class constants. Pre-8.3 uses the untyped variant; the
+ * type argument is discarded. The shim returns void rather than
+ * zend_class_constant* because the generated code never reads the return. */
+#if PHP_VERSION_ID < 80300
+static zend_always_inline void zend_declare_typed_class_constant(
+    zend_class_entry *ce,
+    zend_string *name,
+    zval *value,
+    int access_type,
+    zend_string *doc_comment,
+    zend_type type)
+{
+    (void) type;
+    zend_declare_class_constant_ex(ce, name, value, access_type, doc_comment);
+}
+#endif
+
+/* PHP 8.0:
+ *
+ * - The IS_MIXED type tag and default-value-aware argument macros are
+ *   8.0+ only. Pre-8.0 has neither. The native ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX
+ *   exists on 7.4 but its expansion references the `type` argument, so
+ *   passing IS_MIXED through it fails at compile time even though the
+ *   outer macro is recognized.
+ * - Override the macro on pre-8.0 to drop the type argument entirely
+ *   (returning untyped reflection). Then IS_MIXED never reaches the
+ *   compiler and the rest of the arginfo header compiles cleanly.
+ */
+#if PHP_VERSION_ID < 80000
+# undef ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX
+# define ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(name, return_reference, required_num_args, type, allow_null) \
+    ZEND_BEGIN_ARG_INFO_EX(name, 0, return_reference, required_num_args)
+/* ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE is wholly new in 8.0; pre-8.0
+ * keeps the type info but drops the default-value annotation. */
+# define ZEND_ARG_TYPE_INFO_WITH_DEFAULT_VALUE(pass_by_ref, name, type, allow_null, default_value) \
+    ZEND_ARG_TYPE_INFO(pass_by_ref, name, type, allow_null)
+/* IS_MIXED is a PHP 8.0+ type tag (value 0x09). On 7.4 it isn't
+ * declared. Return-type uses are stripped above by the
+ * ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX shim, but parameter types
+ * reach ZEND_ARG_TYPE_INFO directly with the IS_MIXED token still in
+ * place. Define it to 0 (= IS_UNDEF, "no type constraint") so the
+ * argument compiles untyped on 7.4, matching pre-8.0 reflection. */
+# define IS_MIXED 0
+/* gen_stub builds typed properties / typed class constants via
+ * (zend_type) ZEND_TYPE_INIT_MASK(MAY_BE_<TYPE>[|MAY_BE_NULL]). 7.4
+ * has typed properties but uses a different zend_type construction
+ * pathway and lacks both the MAY_BE_* mask constants in this form
+ * and the ZEND_TYPE_INIT_MASK macro. Shim them all to compile-time
+ * zeros: the property/constant gets registered with no type info,
+ * matching pre-7.4 behavior even where the runtime would have
+ * supported the typed declaration on 7.4. */
+# define ZEND_TYPE_INIT_MASK(mask) ((zend_type) 0)
+# define MAY_BE_LONG 0
+# define MAY_BE_STRING 0
+# define MAY_BE_NULL 0
+# define MAY_BE_BOOL 0
+# define MAY_BE_DOUBLE 0
+# define MAY_BE_ARRAY 0
+#endif
 
 /*
  * Local variables:
diff --git a/php_SeasClick.h b/php_clickhouse.h
similarity index 52%
rename from php_SeasClick.h
rename to php_clickhouse.h
index 27da76c..d9f9fcb 100644
--- a/php_SeasClick.h
+++ b/php_clickhouse.h
@@ -1,8 +1,8 @@
 /*
   +----------------------------------------------------------------------+
-  | SeasClick                                                            |
+  | php_clickhouse                                                       |
   +----------------------------------------------------------------------+
-  | Copyright (c) 1997-2018 The PHP Group                                |
+  | Copyright (c) 1997-2026 The PHP Group                                |
   +----------------------------------------------------------------------+
   | This source file is subject to version 3.01 of the PHP license,      |
   | that is bundled with this package in the file LICENSE, and is        |
@@ -12,46 +12,44 @@
   | obtain it through the world-wide-web, please send a note to          |
   | license@php.net so we can mail you a copy immediately.               |
   +----------------------------------------------------------------------+
-  | Author:  SeasX Group <ahhhh.wang@gmail.com>                          |
+  | Author: Ilia Alshanetsky <ilia@ilia.ws>                              |
+  | Original SeasClick author: SeasX Group <ahhhh.wang@gmail.com>        |
   +----------------------------------------------------------------------+
 */
 #ifdef __cplusplus
 #define __STDC_FORMAT_MACROS
 #endif
 
-#ifndef PHP_SEASCLICK_H
-#define PHP_SEASCLICK_H
+#ifndef PHP_CLICKHOUSE_H
+#define PHP_CLICKHOUSE_H
 
-extern zend_module_entry SeasClick_module_entry;
-#define phpext_SeasClick_ptr &SeasClick_module_entry
+extern zend_module_entry clickhouse_module_entry;
+#define phpext_clickhouse_ptr &clickhouse_module_entry
 
-#define PHP_SEASCLICK_VERSION "0.1.1"
+#define PHP_CLICKHOUSE_VERSION "0.8.5"
 
 #ifdef PHP_WIN32
-#	define PHP_SEASCLICK_API __declspec(dllexport)
+#	define PHP_CLICKHOUSE_API __declspec(dllexport)
 #elif defined(__GNUC__) && __GNUC__ >= 4
-#	define PHP_SEASCLICK_API __attribute__ ((visibility("default")))
+#	define PHP_CLICKHOUSE_API __attribute__ ((visibility("default")))
 #else
-#	define PHP_SEASCLICK_API
+#	define PHP_CLICKHOUSE_API
 #endif
 
-extern "C" {
-#ifdef ZTS
-#include "TSRM.h"
-#endif
-}
-
-typedef unsigned long ulong_t;
+#define SC_FETCH_ONE 1
+#define SC_FETCH_KEY_PAIR 2
+#define SC_FETCH_DATE_AS_STRINGS 4
+#define SC_FETCH_COLUMN 8
 
-#ifdef ZTS
-#define SEASCLICK_G(v) TSRMG(SeasClick_globals_id, zend_SeasClick_globals *, v)
-#else
-#define SEASCLICK_G(v) (SeasClick_globals.v)
-#endif
+#define CLICKHOUSE_RES_NAME "ClickHouse"
+#define CLICKHOUSE_EXCEPTION_NAME "ClickHouseException"
 
-#define SEASCLICK_RES_NAME "SeasClick"
+/* Back-compat aliases for the original SeasClick name. Kept for the 0.5.x
+ * release cycle and removed in the next major. */
+#define CLICKHOUSE_RES_NAME_LEGACY "SeasClick"
+#define CLICKHOUSE_EXCEPTION_NAME_LEGACY "SeasClickException"
 
-#endif	/* PHP_SEASCLICK_H */
+#endif	/* PHP_CLICKHOUSE_H */
 
 /*
  * Local variables:
diff --git a/scripts/check-no-seasclick.sh b/scripts/check-no-seasclick.sh
new file mode 100755
index 0000000..2367f8e
--- /dev/null
+++ b/scripts/check-no-seasclick.sh
@@ -0,0 +1,55 @@
+#!/bin/bash
+# Fail if a "SeasClick" / "SEASCLICK" / "seasclick" reference shows up
+# anywhere outside the documented compat surface. The compat surface is:
+#
+#   - php_clickhouse.h: CLICKHOUSE_RES_NAME_LEGACY / CLICKHOUSE_EXCEPTION_NAME_LEGACY
+#     definitions and the file-header attribution
+#   - clickhouse.cpp: alias-registration comments and the file-header
+#     attribution
+#   - tests/_clickhouse.inc: seasclick_test_config / seasclick_skip_if_no_server
+#     back-compat function aliases
+#   - tests/026.phpt: exercises the SeasClick / SeasClickException
+#     BC aliases on purpose
+#   - tests/051.phpt: server-free surface smoke; verifies the legacy
+#     class-alias names resolve as part of the public surface check
+#   - bench/bench_mark.php: file-header attribution
+#   - composer.json: "Original author of SeasClick" credit string
+#   - documentation files (README, CHANGELOG, CONTRIBUTING, SECURITY,
+#     CREDITS): historical references are fine and expected
+#
+# Anywhere else, "SeasClick" usually means a rename was missed.
+#
+# Vendored library (lib/clickhouse-cpp/) and build artifacts are
+# excluded.
+
+set -u
+
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+cd "$ROOT"
+
+UNEXPECTED=$(grep -rn "SeasClick\|SEASCLICK\|seasclick" \
+    --include="*.cpp" --include="*.hpp" --include="*.h" \
+    --include="*.m4" --include="*.yml" --include="*.json" \
+    --include="*.phpt" --include="*.inc" --include="*.php" \
+    . 2>/dev/null \
+    | grep -v '/lib/clickhouse-cpp/' \
+    | grep -v '/scripts/' \
+    | grep -v '\.dep$' \
+    | grep -v '\.loT$' \
+    | grep -vE '^\./php_clickhouse\.h:.*(Original SeasClick|RES_NAME_LEGACY|EXCEPTION_NAME_LEGACY|aliases for the original SeasClick)' \
+    | grep -vE '^\./clickhouse\.cpp:.*(Original SeasClick|aliases for the original SeasClick)' \
+    | grep -vE '^\./tests/_clickhouse\.inc:.*(seasclick_test_config|seasclick_skip_if_no_server)' \
+    | grep -vE '^\./tests/026\.phpt:' \
+    | grep -vE '^\./tests/051\.phpt:' \
+    | grep -vE '^\./bench/bench_mark\.php:.*for SeasClick\.' \
+    | grep -vE '^\./composer\.json:.*(SeasX Group \(original SeasClick\)|Original author of SeasClick)')
+
+if [ -n "$UNEXPECTED" ]; then
+    echo "::error::Unexpected SeasClick reference(s) found outside the documented compat surface:"
+    echo "$UNEXPECTED"
+    echo
+    echo "If a reference is intentional, allowlist it in scripts/check-no-seasclick.sh."
+    exit 1
+fi
+
+echo "OK: no unexpected SeasClick references."
diff --git a/tests/001.phpt b/tests/001.phpt
index 53ef821..706825a 100644
--- a/tests/001.phpt
+++ b/tests/001.phpt
@@ -1,10 +1,10 @@
 --TEST--
-Check for SeasClick presence
---SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+Check for ClickHouse extension presence
+--EXTENSIONS--
+clickhouse
 --FILE--
 <?php 
-echo "SeasClick extension is available";
+echo "ClickHouse extension is available";
 /*
 	you can add regression tests for your extension here
 
@@ -18,4 +18,4 @@ echo "SeasClick extension is available";
 */
 ?>
 --EXPECT--
-SeasClick extension is available
+ClickHouse extension is available
diff --git a/tests/002.phpt b/tests/002.phpt
index d69493a..837be32 100644
--- a/tests/002.phpt
+++ b/tests/002.phpt
@@ -1,21 +1,20 @@
 --TEST--
-SeasClick testArray
+ClickHouse testArray
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
 
 clientTest($config);
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testArray($client, $deleteTable);
@@ -36,7 +35,7 @@ function testArray($client, $deleteTable = false) {
     ]);
 
     $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'string_c, array_c, arraynull_c',
+        'select' => ['string_c', 'array_c', 'arraynull_c'],
         'table' => 'test.array_test'
     ]);
     var_dump($result);
@@ -66,7 +65,7 @@ array(2) {
     }
     ["arraynull_c"]=>
     array(1) {
-      ["array"]=>
+      [0]=>
       string(6) "string"
     }
   }
diff --git a/tests/003.phpt b/tests/003.phpt
index 03cd874..37ce5c4 100644
--- a/tests/003.phpt
+++ b/tests/003.phpt
@@ -1,21 +1,20 @@
 --TEST--
-SeasClick testEnum
+ClickHouse testEnum
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
 
 clientTest($config);
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testEnum($client, $deleteTable);
@@ -32,7 +31,7 @@ function testEnum($client, $deleteTable = false) {
     ]);
 
     $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'enum8_c, enum16_c',
+        'select' => ['enum8_c', 'enum16_c'],
         'table' => 'test.enum_test'
     ]);
     var_dump($result);
diff --git a/tests/004.phpt b/tests/004.phpt
index 7f9aa30..c96883d 100644
--- a/tests/004.phpt
+++ b/tests/004.phpt
@@ -1,21 +1,20 @@
 --TEST--
-SeasClick testString
+ClickHouse testString
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
 
 clientTest($config);
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testString($client, $deleteTable);
@@ -38,7 +37,7 @@ function testString($client, $deleteTable = false) {
     ]);
     
     $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'string_c, fixedstring_c',
+        'select' => ['string_c', 'fixedstring_c'],
         'table' => 'test.string_test'
     ]);
     var_dump($result);
diff --git a/tests/005.phpt b/tests/005.phpt
index 975e380..04b5e81 100644
--- a/tests/005.phpt
+++ b/tests/005.phpt
@@ -1,21 +1,20 @@
 --TEST--
-SeasClick testNullAble
+ClickHouse testNullAble
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
 
 clientTest($config);
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testNullAble($client, $deleteTable);
@@ -36,7 +35,7 @@ function testNullAble($client, $deleteTable = false) {
     ]);
 
     $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'int8null_c, stringnull_c, enumnull_c, float32null_c, uuidnull_c',
+        'select' => ['int8null_c', 'stringnull_c', 'enumnull_c', 'float32null_c', 'uuidnull_c'],
         'table' => 'test.nullable_test'
     ]);
     var_dump($result);
diff --git a/tests/006.phpt b/tests/006.phpt
index 727009c..bd1eac3 100644
--- a/tests/006.phpt
+++ b/tests/006.phpt
@@ -1,28 +1,27 @@
 --TEST--
-SeasClick testInt
+ClickHouse testInt
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
 
-clientTest($config);
+clientTest(clickhouse_test_config());
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testUInt($client, $deleteTable);
 }
 
 function testUInt($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.int_test (int8_c Int8, int16_c Int16, uint8_c UInt8, uint16_c UInt16) ENGINE = Memory");
+    $client->execute("DROP TABLE IF EXISTS test.int_test");
+    $client->execute("CREATE TABLE test.int_test (int8_c Int8, int16_c Int16, uint8_c UInt8, uint16_c UInt16) ENGINE = Memory");
 
     $client->insert("test.int_test",[
         'int8_c','int16_c','uint8_c','uint16_c'
@@ -40,8 +39,8 @@ function testUInt($client, $deleteTable = false) {
     ]);
     $client->writeEnd();
     
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'int8_c, int16_c, uint8_c, uint16_c',
+    $result = $client->select("SELECT {select} FROM {table} ORDER BY uint16_c, int8_c", [
+        'select' => ['int8_c', 'int16_c', 'uint8_c', 'uint16_c'],
         'table' => 'test.int_test'
     ]);
     var_dump($result);
@@ -65,7 +64,7 @@ array(4) {
     ["uint8_c"]=>
     int(8)
     ["uint16_c"]=>
-    int(8)
+    int(0)
   }
   [1]=>
   array(4) {
@@ -76,7 +75,7 @@ array(4) {
     ["uint8_c"]=>
     int(9)
     ["uint16_c"]=>
-    int(9)
+    int(0)
   }
   [2]=>
   array(4) {
@@ -87,7 +86,7 @@ array(4) {
     ["uint8_c"]=>
     int(8)
     ["uint16_c"]=>
-    int(0)
+    int(8)
   }
   [3]=>
   array(4) {
@@ -98,6 +97,6 @@ array(4) {
     ["uint8_c"]=>
     int(9)
     ["uint16_c"]=>
-    int(0)
+    int(9)
   }
 }
diff --git a/tests/007.phpt b/tests/007.phpt
index 78a9431..ad71be6 100644
--- a/tests/007.phpt
+++ b/tests/007.phpt
@@ -1,28 +1,27 @@
 --TEST--
-SeasClick testFloat
+ClickHouse testFloat
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
 
-clientTest($config);
+clientTest(clickhouse_test_config());
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testFloat($client, $deleteTable);
 }
 
 function testFloat($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.float_test (float32_c Float32, float64_c Float64) ENGINE = Memory");
+    $client->execute("DROP TABLE IF EXISTS test.float_test");
+    $client->execute("CREATE TABLE test.float_test (float32_c Float32, float64_c Float64) ENGINE = Memory");
 
     $client->insert("test.float_test",[
         'float32_c', 'float64_c'
@@ -40,8 +39,8 @@ function testFloat($client, $deleteTable = false) {
     ]);
     $client->writeEnd();
     
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'float32_c, float64_c',
+    $result = $client->select("SELECT {select} FROM {table} ORDER BY float64_c DESC, float32_c DESC", [
+        'select' => ['float32_c', 'float64_c'],
         'table' => 'test.float_test'
     ]);
     var_dump($result);
@@ -59,16 +58,16 @@ array(4) {
   [0]=>
   array(2) {
     ["float32_c"]=>
-    float(32.32)
+    float(32.31)
     ["float64_c"]=>
-    float(64.64)
+    float(64.68)
   }
   [1]=>
   array(2) {
     ["float32_c"]=>
-    float(32.31)
+    float(32.32)
     ["float64_c"]=>
-    float(64.68)
+    float(64.64)
   }
   [2]=>
   array(2) {
diff --git a/tests/008.phpt b/tests/008.phpt
index b1280fd..66c177d 100644
--- a/tests/008.phpt
+++ b/tests/008.phpt
@@ -1,34 +1,33 @@
 --TEST--
-SeasClick testUUID
+ClickHouse testUUID
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
 
-clientTest($config);
+clientTest(clickhouse_test_config());
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testUUID($client, $deleteTable);
 }
 
 function testUUID($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.uuid_test (uuid_c UUID, uuid2_c UUID) ENGINE = Memory");
+    $client->execute("DROP TABLE IF EXISTS test.uuid_test");
+    $client->execute("CREATE TABLE test.uuid_test (uuid_c UUID, uuid2_c UUID) ENGINE = Memory");
 
     $client->insert("test.uuid_test",[
         'uuid_c', 'uuid2_c'
     ], [
         ['31249a1b-7b05-4270-9f37-c609b48a9bb2', '31249a1b7b0542709f37c609b48a9bb2'],
-        ['31249a1b-7b05-4270-9f37-c609b48a9bb2', null],
+        ['31249a1b-7b05-4270-9f37-c609b48a9bb2', '00000000-0000-0000-0000-000000000000'],
     ]);
 
     $client->writeStart("test.uuid_test", [
@@ -40,8 +39,8 @@ function testUUID($client, $deleteTable = false) {
     ]);
     $client->writeEnd();
     
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'uuid_c, uuid2_c',
+    $result = $client->select("SELECT {select} FROM {table} ORDER BY uuid_c, uuid2_c", [
+        'select' => ['uuid_c', 'uuid2_c'],
         'table' => 'test.uuid_test'
     ]);
     var_dump($result);
@@ -59,9 +58,9 @@ array(4) {
   [0]=>
   array(2) {
     ["uuid_c"]=>
-    string(32) "31249a1b7b0542709f37c609b48a9bb2"
+    string(32) "00000000000000009f37c609b48a9bb2"
     ["uuid2_c"]=>
-    string(32) "31249a1b7b0542709f37c609b48a9bb2"
+    string(32) "00000000000000000000000000000000"
   }
   [1]=>
   array(2) {
@@ -73,7 +72,7 @@ array(4) {
   [2]=>
   array(2) {
     ["uuid_c"]=>
-    string(32) "00000000000000009f37c609b48a9bb2"
+    string(32) "31249a1b7b0542709f37c609b48a9bb2"
     ["uuid2_c"]=>
     string(32) "00000000000000000000000000000000"
   }
@@ -82,6 +81,6 @@ array(4) {
     ["uuid_c"]=>
     string(32) "31249a1b7b0542709f37c609b48a9bb2"
     ["uuid2_c"]=>
-    string(32) "00000000000000000000000000000000"
+    string(32) "31249a1b7b0542709f37c609b48a9bb2"
   }
 }
diff --git a/tests/009.phpt b/tests/009.phpt
index 7d0730b..4191854 100644
--- a/tests/009.phpt
+++ b/tests/009.phpt
@@ -1,49 +1,41 @@
 --TEST--
-SeasClick testDate
+ClickHouse testDate
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
 
 clientTest($config);
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testDate($client, $deleteTable);
 }
 
 function testDate($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.date_test (date_c Date, datetime_c DateTime) ENGINE = Memory");
+    $client->execute("DROP TABLE IF EXISTS test.date_test");
+    $client->execute("CREATE TABLE test.date_test (date_c Date, datetime_c DateTime('UTC')) ENGINE = Memory");
 
-    $time1 = strtotime(date('Y-m-d H:i:s', 1548633600));
-    $time2 = strtotime(date('Y-m-d H:i:s', 1548687925));
-
-    $client->insert("test.date_test", [
-        'date_c', 'datetime_c'
-    ], [
-        [$time1, $time2],
-        [$time1, $time2]
-    ]);
-    
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'date_c, datetime_c',
-        'table' => 'test.date_test'
+    // Raw epoch ints, TZ-independent. 1548633600 = 2019-01-28 00:00:00 UTC,
+    // 1548687925 = 2019-01-28 15:05:25 UTC.
+    $client->insert("test.date_test", ['date_c', 'datetime_c'], [
+        [1548633600, 1548687925],
+        [1548633600, 1548687925],
     ]);
+
+    $result = $client->select("SELECT date_c, datetime_c FROM test.date_test");
     var_dump($result);
-    
+
     if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.date_test'
-        ]);
+        $client->execute("DROP TABLE test.date_test");
     }
 }
 
diff --git a/tests/010.phpt b/tests/010.phpt
index c22d914..c7b975d 100644
--- a/tests/010.phpt
+++ b/tests/010.phpt
@@ -1,21 +1,20 @@
 --TEST--
-SeasClick testDate
+ClickHouse testDate
+--EXTENSIONS--
+clickhouse
 --SKIPIF--
-<?php if (!extension_loaded("SeasClick")) print "skip"; ?>
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
 --FILE--
 <?php
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
 
 clientTest($config);
 
 function clientTest($config)
 {
     $deleteTable = true;
-    $client = new SeasClick($config);
+    $client = new ClickHouse($config);
     $client->execute("CREATE DATABASE IF NOT EXISTS test");
 
     testTuple($client, $deleteTable);
@@ -31,7 +30,7 @@ function testTuple($client, $deleteTable = false) {
         [[2, 'two'], 2, 'two'],
     ]);
     $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'tuple_c, int_c, string_c',
+        'select' => ['tuple_c', 'int_c', 'string_c'],
         'table' => 'test.tuple_test'
     ]);
     var_dump($result);
diff --git a/tests/011.phpt b/tests/011.phpt
new file mode 100644
index 0000000..93f96af
--- /dev/null
+++ b/tests/011.phpt
@@ -0,0 +1,199 @@
+--TEST--
+ClickHouse Single Value Fetch
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
+
+$deleteTable = true;
+$client = new ClickHouse($config);
+$client->execute('CREATE DATABASE IF NOT EXISTS test');
+
+$client->execute("CREATE TABLE IF NOT EXISTS test.single_val_fetch (
+	tuple_c Tuple(id UInt64, name String),
+	int64_c UInt64,
+	string_c String,
+	array_c Array(Int8),
+	arraynull_c Array(Nullable(String)),
+	enum8_c Enum8('One8' = 1, 'Two8' = 2),
+	enum16_c Enum16('One16' = 1, 'Two16' = 2),
+	fixedstring_c FixedString(50),
+	int8null_c Nullable(Int8),
+	stringnull_c Nullable(String),
+	enumnull_c Nullable(Enum8('One8' = 1, 'Two8' = 2)),
+	float32null_c Nullable(Float32),
+	uuidnull_c Nullable(UUID),
+	int8_c Int8,
+	int16_c Int16,
+	uint8_c UInt8,
+	uint16_c UInt16,
+	float32_c Float32,
+	float64_c Float64,
+	uuid_c UUID,
+	uuid2_c UUID,
+	date_c Date,
+	datetime_c DateTime
+) ENGINE = Memory");
+
+$data = [
+    [
+        'tuple_c'       => [1, 'one'],
+        'int64_c'       => 1,
+        'string_c'      => 'string_one',
+        'array_c'       => [1, 2, 3],
+        'arraynull_c'   => ['str_array'],
+        'enum8_c'       => 1,
+        'enum16_c'      => 'Two16',
+        'fixedstring_c' => 'fixedstring_c1',
+        'int8null_c'    => 8,
+        'stringnull_c'  => 'string',
+        'enumnull_c'    => 'One8',
+        'float32null_c' => 7.77,
+        'uuidnull_c'    => '31249a1b7b0542709f37c609b48a9bb2',
+        'int8_c'        => 8,
+        'int16_c'       => 16,
+        'uint8_c'       => 18,
+        'uint16_c'      => 20,
+        'float32_c'     => 32.32,
+        'float64_c'     => 64.64,
+        'uuid_c'        => '31249a1b-7b05-4270-9f37-c609b48a9bb2',
+        'uuid2_c'       => '31249a1b7b0542709f37c609b48a9bb2',
+        'date_c'        => 1548633600,
+        'datetime_c'    => 1548687925,
+    ],
+    [
+        'tuple_c'       => [2, 'two'],
+        'int64_c'       => 2,
+        'string_c'      => 'string_two',
+        'array_c'       => [2, 3, 4],
+        'arraynull_c'   => [null],
+        'enum8_c'       => 'Two8',
+        'enum16_c'      => 2,
+        'fixedstring_c' => 'fixedstring_c2',
+        'int8null_c'    => null,
+        'stringnull_c'  => null,
+        'enumnull_c'    => null,
+        'float32null_c' => null,
+        'uuidnull_c'    => null,
+        'int8_c'        => 28,
+        'int16_c'       => 216,
+        'uint8_c'       => 218,
+        'uint16_c'      => 220,
+        'float32_c'     => 232.32,
+        'float64_c'     => 264.64,
+        'uuid_c'        => '31249a1b-7b05-4270-9f37-c609b48a9bb2',
+        'uuid2_c'       => '00000000-0000-0000-0000-000000000000',
+        'date_c'        => 1548547200,
+        'datetime_c'    => 1548513600,
+    ],
+];
+
+$fields = array_keys(current($data));
+
+$expected = $data;
+$expected[0]['uuid_c'] = '31249a1b7b0542709f37c609b48a9bb2';
+$expected[0]['enum8_c'] = 'One8';
+$expected[1]['enum16_c'] = 'Two16';
+$expected[1]['uuid_c'] = '31249a1b7b0542709f37c609b48a9bb2';
+$expected[1]['arraynull_c'] = [null];
+$expected[1]['uuid2_c'] = '00000000000000000000000000000000';
+
+$client->insert('test.single_val_fetch', $fields, [array_values($data[0]), array_values($data[1])]);
+
+foreach ($fields as $field) {
+    for ($i = 1; $i < 3; $i++) {
+        $result = $client->select("SELECT {$field} FROM test.single_val_fetch WHERE int64_c = {i}", ['i' => $i], ClickHouse::FETCH_ONE);
+        $match = $result === $expected[$i-1][$field] ? 'OK' : 'FAIL';
+        echo $field, ': ', var_export($result, true), ' - ', var_export($expected[$i - 1][$field], true), ' - ', $match, "\n";
+    }
+}
+
+$client->execute('DROP TABLE test.single_val_fetch');
+?>
+--EXPECT--
+tuple_c: array (
+  0 => 1,
+  1 => 'one',
+) - array (
+  0 => 1,
+  1 => 'one',
+) - OK
+tuple_c: array (
+  0 => 2,
+  1 => 'two',
+) - array (
+  0 => 2,
+  1 => 'two',
+) - OK
+int64_c: 1 - 1 - OK
+int64_c: 2 - 2 - OK
+string_c: 'string_one' - 'string_one' - OK
+string_c: 'string_two' - 'string_two' - OK
+array_c: array (
+  0 => 1,
+  1 => 2,
+  2 => 3,
+) - array (
+  0 => 1,
+  1 => 2,
+  2 => 3,
+) - OK
+array_c: array (
+  0 => 2,
+  1 => 3,
+  2 => 4,
+) - array (
+  0 => 2,
+  1 => 3,
+  2 => 4,
+) - OK
+arraynull_c: array (
+  0 => 'str_array',
+) - array (
+  0 => 'str_array',
+) - OK
+arraynull_c: array (
+  0 => NULL,
+) - array (
+  0 => NULL,
+) - OK
+enum8_c: 'One8' - 'One8' - OK
+enum8_c: 'Two8' - 'Two8' - OK
+enum16_c: 'Two16' - 'Two16' - OK
+enum16_c: 'Two16' - 'Two16' - OK
+fixedstring_c: 'fixedstring_c1' - 'fixedstring_c1' - OK
+fixedstring_c: 'fixedstring_c2' - 'fixedstring_c2' - OK
+int8null_c: 8 - 8 - OK
+int8null_c: NULL - NULL - OK
+stringnull_c: 'string' - 'string' - OK
+stringnull_c: NULL - NULL - OK
+enumnull_c: 'One8' - 'One8' - OK
+enumnull_c: NULL - NULL - OK
+float32null_c: 7.77 - 7.77 - OK
+float32null_c: NULL - NULL - OK
+uuidnull_c: '31249a1b7b0542709f37c609b48a9bb2' - '31249a1b7b0542709f37c609b48a9bb2' - OK
+uuidnull_c: NULL - NULL - OK
+int8_c: 8 - 8 - OK
+int8_c: 28 - 28 - OK
+int16_c: 16 - 16 - OK
+int16_c: 216 - 216 - OK
+uint8_c: 18 - 18 - OK
+uint8_c: 218 - 218 - OK
+uint16_c: 20 - 20 - OK
+uint16_c: 220 - 220 - OK
+float32_c: 32.32 - 32.32 - OK
+float32_c: 232.32 - 232.32 - OK
+float64_c: 64.64 - 64.64 - OK
+float64_c: 264.64 - 264.64 - OK
+uuid_c: '31249a1b7b0542709f37c609b48a9bb2' - '31249a1b7b0542709f37c609b48a9bb2' - OK
+uuid_c: '31249a1b7b0542709f37c609b48a9bb2' - '31249a1b7b0542709f37c609b48a9bb2' - OK
+uuid2_c: '31249a1b7b0542709f37c609b48a9bb2' - '31249a1b7b0542709f37c609b48a9bb2' - OK
+uuid2_c: '00000000000000000000000000000000' - '00000000000000000000000000000000' - OK
+date_c: 1548633600 - 1548633600 - OK
+date_c: 1548547200 - 1548547200 - OK
+datetime_c: 1548687925 - 1548687925 - OK
+datetime_c: 1548513600 - 1548513600 - OK
diff --git a/tests/012.phpt b/tests/012.phpt
new file mode 100644
index 0000000..3cb6b80
--- /dev/null
+++ b/tests/012.phpt
@@ -0,0 +1,80 @@
+--TEST--
+ClickHouse Date Formatting
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
+
+$deleteTable = true;
+$client = new ClickHouse($config);
+$client->execute('CREATE DATABASE IF NOT EXISTS test');
+
+$client->execute("CREATE TABLE IF NOT EXISTS test.dates (
+	date_c Date,
+	datetime_c DateTime
+) ENGINE = Memory");
+
+$data = [
+    [
+        'date_c'        => 1548633600,
+        'datetime_c'    => 1548687925,
+    ],
+    [
+        'date_c'        => 1548547200,
+        'datetime_c'    => 1548513600,
+    ],
+];
+$expected = [
+    [
+            'date_c'        => gmdate('Y-m-d', 1548633600),
+            'datetime_c'    => gmdate('Y-m-d H:i:s', 1548687925),
+        ],
+        [
+            'date_c'        => gmdate('Y-m-d', 1548547200),
+            'datetime_c'    => gmdate('Y-m-d H:i:s', 1548513600),
+        ],
+];
+
+$fields = array_keys(current($data));
+$client->insert('test.dates', $fields, [array_values($data[0]), array_values($data[1])]);
+
+$res = $client->select("SELECT * FROM test.dates", [], ClickHouse::DATE_AS_STRINGS);
+var_dump($res);
+if (array_diff_assoc($expected[0], $res[0]) || array_diff_assoc($expected[1], $res[1])) {
+    echo "FAIL\n";
+} else {
+    echo "OK\n";
+}
+
+$res = $client->select("SELECT date_c FROM test.dates WHERE datetime_c = 1548687925", [], ClickHouse::FETCH_ONE|ClickHouse::DATE_AS_STRINGS);
+echo $res, ' = ', $expected[0]['date_c'], ' ', ($res === $expected[0]['date_c'] ? 'OK' : 'FAIL'), "\n";
+
+$res = $client->select("SELECT datetime_c FROM test.dates WHERE datetime_c = 1548687925", [], ClickHouse::FETCH_ONE|ClickHouse::DATE_AS_STRINGS);
+echo $res, ' = ', $expected[0]['datetime_c'], ' ' , ($res === $expected[0]['datetime_c'] ? 'OK' : 'FAIL'), "\n";
+
+$client->execute('DROP TABLE test.dates');
+?>
+--EXPECT--
+array(2) {
+  [0]=>
+  array(2) {
+    ["date_c"]=>
+    string(10) "2019-01-28"
+    ["datetime_c"]=>
+    string(19) "2019-01-28 15:05:25"
+  }
+  [1]=>
+  array(2) {
+    ["date_c"]=>
+    string(10) "2019-01-27"
+    ["datetime_c"]=>
+    string(19) "2019-01-26 14:40:00"
+  }
+}
+OK
+2019-01-28 = 2019-01-28 OK
+2019-01-28 15:05:25 = 2019-01-28 15:05:25 OK
diff --git a/tests/013.phpt b/tests/013.phpt
new file mode 100644
index 0000000..a7c8897
--- /dev/null
+++ b/tests/013.phpt
@@ -0,0 +1,328 @@
+--TEST--
+ClickHouse Single Column Fetch
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
+
+$deleteTable = true;
+$client = new ClickHouse($config);
+$client->execute('CREATE DATABASE IF NOT EXISTS test');
+
+$client->execute("CREATE TABLE IF NOT EXISTS test.single_val_fetch (
+	tuple_c Tuple(id UInt64, name String),
+	int64_c UInt64,
+	string_c String,
+	array_c Array(Int8),
+	arraynull_c Array(Nullable(String)),
+	enum8_c Enum8('One8' = 1, 'Two8' = 2),
+	enum16_c Enum16('One16' = 1, 'Two16' = 2),
+	fixedstring_c FixedString(50),
+	int8null_c Nullable(Int8),
+	stringnull_c Nullable(String),
+	enumnull_c Nullable(Enum8('One8' = 1, 'Two8' = 2)),
+	float32null_c Nullable(Float32),
+	uuidnull_c Nullable(UUID),
+	int8_c Int8,
+	int16_c Int16,
+	uint8_c UInt8,
+	uint16_c UInt16,
+	float32_c Float32,
+	float64_c Float64,
+	uuid_c UUID,
+	uuid2_c UUID,
+	date_c Date,
+	datetime_c DateTime
+) ENGINE = Memory");
+
+$data = [
+    [
+        'tuple_c'       => [1, 'one'],
+        'int64_c'       => 1,
+        'string_c'      => 'string_one',
+        'array_c'       => [1, 2, 3],
+        'arraynull_c'   => ['str_array'],
+        'enum8_c'       => 1,
+        'enum16_c'      => 'Two16',
+        'fixedstring_c' => 'fixedstring_c1',
+        'int8null_c'    => 8,
+        'stringnull_c'  => 'string',
+        'enumnull_c'    => 'One8',
+        'float32null_c' => 7.77,
+        'uuidnull_c'    => '31249a1b7b0542709f37c609b48a9bb2',
+        'int8_c'        => 8,
+        'int16_c'       => 16,
+        'uint8_c'       => 18,
+        'uint16_c'      => 20,
+        'float32_c'     => 32.32,
+        'float64_c'     => 64.64,
+        'uuid_c'        => '31249a1b-7b05-4270-9f37-c609b48a9bb2',
+        'uuid2_c'       => '31249a1b7b0542709f37c609b48a9bb2',
+        'date_c'        => 1548633600,
+        'datetime_c'    => 1548687925,
+    ],
+    [
+        'tuple_c'       => [2, 'two'],
+        'int64_c'       => 2,
+        'string_c'      => 'string_two',
+        'array_c'       => [2, 3, 4],
+        'arraynull_c'   => [null],
+        'enum8_c'       => 'Two8',
+        'enum16_c'      => 2,
+        'fixedstring_c' => 'fixedstring_c2',
+        'int8null_c'    => null,
+        'stringnull_c'  => null,
+        'enumnull_c'    => null,
+        'float32null_c' => null,
+        'uuidnull_c'    => null,
+        'int8_c'        => 28,
+        'int16_c'       => 216,
+        'uint8_c'       => 218,
+        'uint16_c'      => 220,
+        'float32_c'     => 232.32,
+        'float64_c'     => 264.64,
+        'uuid_c'        => '31249a1b-7b05-4270-9f37-c609b48a9bb2',
+        'uuid2_c'       => '00000000-0000-0000-0000-000000000000',
+        'date_c'        => 1548547200,
+        'datetime_c'    => 1548513600,
+    ],
+];
+
+$fields = array_keys(current($data));
+
+$expected = $data;
+$expected[0]['uuid_c'] = '31249a1b7b0542709f37c609b48a9bb2';
+$expected[0]['enum8_c'] = 'One8';
+$expected[1]['enum16_c'] = 'Two16';
+$expected[1]['uuid_c'] = '31249a1b7b0542709f37c609b48a9bb2';
+$expected[1]['arraynull_c'] = [null];
+$expected[1]['uuid2_c'] = '00000000000000000000000000000000';
+
+$client->insert('test.single_val_fetch', $fields, [array_values($data[0]), array_values($data[1])]);
+
+foreach ($fields as $field) {
+    $result = $client->select("SELECT {$field} FROM test.single_val_fetch ORDER BY int64_c ASC", [], ClickHouse::FETCH_COLUMN);
+
+    $res = var_export($result, true);
+    $exp = var_export([$expected[0][$field], $expected[1][$field]], true);
+    $match = $res === $exp ? 'OK' : 'FAIL';
+
+    echo $field, ': ', $res , ' - ', $exp , ' - ', $match, "\n";
+}
+
+$client->execute('DROP TABLE test.single_val_fetch');
+?>
+--EXPECT--
+tuple_c: array (
+  0 => 
+  array (
+    0 => 1,
+    1 => 'one',
+  ),
+  1 => 
+  array (
+    0 => 2,
+    1 => 'two',
+  ),
+) - array (
+  0 => 
+  array (
+    0 => 1,
+    1 => 'one',
+  ),
+  1 => 
+  array (
+    0 => 2,
+    1 => 'two',
+  ),
+) - OK
+int64_c: array (
+  0 => 1,
+  1 => 2,
+) - array (
+  0 => 1,
+  1 => 2,
+) - OK
+string_c: array (
+  0 => 'string_one',
+  1 => 'string_two',
+) - array (
+  0 => 'string_one',
+  1 => 'string_two',
+) - OK
+array_c: array (
+  0 => 
+  array (
+    0 => 1,
+    1 => 2,
+    2 => 3,
+  ),
+  1 => 
+  array (
+    0 => 2,
+    1 => 3,
+    2 => 4,
+  ),
+) - array (
+  0 => 
+  array (
+    0 => 1,
+    1 => 2,
+    2 => 3,
+  ),
+  1 => 
+  array (
+    0 => 2,
+    1 => 3,
+    2 => 4,
+  ),
+) - OK
+arraynull_c: array (
+  0 => 
+  array (
+    0 => 'str_array',
+  ),
+  1 => 
+  array (
+    0 => NULL,
+  ),
+) - array (
+  0 => 
+  array (
+    0 => 'str_array',
+  ),
+  1 => 
+  array (
+    0 => NULL,
+  ),
+) - OK
+enum8_c: array (
+  0 => 'One8',
+  1 => 'Two8',
+) - array (
+  0 => 'One8',
+  1 => 'Two8',
+) - OK
+enum16_c: array (
+  0 => 'Two16',
+  1 => 'Two16',
+) - array (
+  0 => 'Two16',
+  1 => 'Two16',
+) - OK
+fixedstring_c: array (
+  0 => 'fixedstring_c1',
+  1 => 'fixedstring_c2',
+) - array (
+  0 => 'fixedstring_c1',
+  1 => 'fixedstring_c2',
+) - OK
+int8null_c: array (
+  0 => 8,
+  1 => NULL,
+) - array (
+  0 => 8,
+  1 => NULL,
+) - OK
+stringnull_c: array (
+  0 => 'string',
+  1 => NULL,
+) - array (
+  0 => 'string',
+  1 => NULL,
+) - OK
+enumnull_c: array (
+  0 => 'One8',
+  1 => NULL,
+) - array (
+  0 => 'One8',
+  1 => NULL,
+) - OK
+float32null_c: array (
+  0 => 7.77,
+  1 => NULL,
+) - array (
+  0 => 7.77,
+  1 => NULL,
+) - OK
+uuidnull_c: array (
+  0 => '31249a1b7b0542709f37c609b48a9bb2',
+  1 => NULL,
+) - array (
+  0 => '31249a1b7b0542709f37c609b48a9bb2',
+  1 => NULL,
+) - OK
+int8_c: array (
+  0 => 8,
+  1 => 28,
+) - array (
+  0 => 8,
+  1 => 28,
+) - OK
+int16_c: array (
+  0 => 16,
+  1 => 216,
+) - array (
+  0 => 16,
+  1 => 216,
+) - OK
+uint8_c: array (
+  0 => 18,
+  1 => 218,
+) - array (
+  0 => 18,
+  1 => 218,
+) - OK
+uint16_c: array (
+  0 => 20,
+  1 => 220,
+) - array (
+  0 => 20,
+  1 => 220,
+) - OK
+float32_c: array (
+  0 => 32.32,
+  1 => 232.32,
+) - array (
+  0 => 32.32,
+  1 => 232.32,
+) - OK
+float64_c: array (
+  0 => 64.64,
+  1 => 264.64,
+) - array (
+  0 => 64.64,
+  1 => 264.64,
+) - OK
+uuid_c: array (
+  0 => '31249a1b7b0542709f37c609b48a9bb2',
+  1 => '31249a1b7b0542709f37c609b48a9bb2',
+) - array (
+  0 => '31249a1b7b0542709f37c609b48a9bb2',
+  1 => '31249a1b7b0542709f37c609b48a9bb2',
+) - OK
+uuid2_c: array (
+  0 => '31249a1b7b0542709f37c609b48a9bb2',
+  1 => '00000000000000000000000000000000',
+) - array (
+  0 => '31249a1b7b0542709f37c609b48a9bb2',
+  1 => '00000000000000000000000000000000',
+) - OK
+date_c: array (
+  0 => 1548633600,
+  1 => 1548547200,
+) - array (
+  0 => 1548633600,
+  1 => 1548547200,
+) - OK
+datetime_c: array (
+  0 => 1548687925,
+  1 => 1548513600,
+) - array (
+  0 => 1548687925,
+  1 => 1548513600,
+) - OK
diff --git a/tests/014.phpt b/tests/014.phpt
new file mode 100644
index 0000000..f6e4462
--- /dev/null
+++ b/tests/014.phpt
@@ -0,0 +1,249 @@
+--TEST--
+ClickHouse Fetch Key Pair
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
+
+$deleteTable = true;
+$client = new ClickHouse($config);
+$client->execute('CREATE DATABASE IF NOT EXISTS test');
+
+$client->execute("CREATE TABLE IF NOT EXISTS test.single_val_fetch (
+	tuple_c Tuple(id UInt64, name String),
+	int64_c UInt64,
+	string_c String,
+	array_c Array(Int8),
+	arraynull_c Array(Nullable(String)),
+	enum8_c Enum8('One8' = 1, 'Two8' = 2),
+	enum16_c Enum16('One16' = 1, 'Two16' = 2),
+	fixedstring_c FixedString(50),
+	int8null_c Nullable(Int8),
+	stringnull_c Nullable(String),
+	enumnull_c Nullable(Enum8('One8' = 1, 'Two8' = 2)),
+	float32null_c Nullable(Float32),
+	uuidnull_c Nullable(UUID),
+	int8_c Int8,
+	int16_c Int16,
+	uint8_c UInt8,
+	uint16_c UInt16,
+	float32_c Float32,
+	float64_c Float64,
+	uuid_c UUID,
+	uuid2_c UUID,
+	date_c Date,
+	datetime_c DateTime
+) ENGINE = Memory");
+
+$data = [
+    [
+        'int64_c'       => 1,
+        'string_c'      => 'string_one',
+        'enum8_c'       => 1,
+        'enum16_c'      => 'Two16',
+        'fixedstring_c' => 'fixedstring_c1',
+        'int8null_c'    => 8,
+        'stringnull_c'  => 'string',
+        'enumnull_c'    => 'One8',
+        'float32null_c' => 7.77,
+        'uuidnull_c'    => '31249a1b7b0542709f37c609b48a9bb2',
+        'int8_c'        => 8,
+        'int16_c'       => 16,
+        'uint8_c'       => 18,
+        'uint16_c'      => 20,
+        'float32_c'     => 32.32,
+        'float64_c'     => 64.64,
+        'uuid_c'        => '31249a1b-7b05-4270-9f37-c609b48a9bb2',
+        'uuid2_c'       => '31249a1b7b0542709f37c609b48a9bb2',
+        'date_c'        => 1548633600,
+        'datetime_c'    => 1548687925,
+    ],
+    [
+        'int64_c'       => 2,
+        'string_c'      => 'string_two',
+        'enum8_c'       => 'Two8',
+        'enum16_c'      => 2,
+        'fixedstring_c' => 'fixedstring_c2',
+        'int8null_c'    => null,
+        'stringnull_c'  => null,
+        'enumnull_c'    => null,
+        'float32null_c' => null,
+        'uuidnull_c'    => null,
+        'int8_c'        => 28,
+        'int16_c'       => 216,
+        'uint8_c'       => 218,
+        'uint16_c'      => 220,
+        'float32_c'     => 232.32,
+        'float64_c'     => 264.64,
+        'uuid_c'        => '31249a1b-7b05-4270-9f37-c609b48a9bb2',
+        'uuid2_c'       => '00000000-0000-0000-0000-000000000000',
+        'date_c'        => 1548547200,
+        'datetime_c'    => 1548513600,
+    ],
+];
+
+$fields = array_keys(current($data));
+
+$expected = $data;
+$expected[0]['uuid_c'] = '31249a1b7b0542709f37c609b48a9bb2';
+$expected[0]['enum8_c'] = 'One8';
+$expected[1]['enum16_c'] = 'Two16';
+$expected[1]['uuid_c'] = '31249a1b7b0542709f37c609b48a9bb2';
+$expected[1]['arraynull_c'] = [null];
+$expected[1]['uuid2_c'] = '00000000000000000000000000000000';
+
+$client->insert('test.single_val_fetch', $fields, [array_values($data[0]), array_values($data[1])]);
+
+foreach ($fields as $field) {
+    $result = $client->select("SELECT {$field}, {$field} FROM test.single_val_fetch ORDER BY int64_c ASC", [], ClickHouse::FETCH_KEY_PAIR);
+
+    $res = var_export($result, true);
+    $exp = var_export([(string)$expected[0][$field] => $expected[0][$field], (string)$expected[1][$field] => $expected[1][$field]], true);
+    $match = $res === $exp ? 'OK' : 'FAIL';
+
+    echo $field, ': ', $res , ' - ', $exp , ' - ', $match, "\n";
+}
+
+$client->execute('DROP TABLE test.single_val_fetch');
+?>
+--EXPECT--
+int64_c: array (
+  1 => 1,
+  2 => 2,
+) - array (
+  1 => 1,
+  2 => 2,
+) - OK
+string_c: array (
+  'string_one' => 'string_one',
+  'string_two' => 'string_two',
+) - array (
+  'string_one' => 'string_one',
+  'string_two' => 'string_two',
+) - OK
+enum8_c: array (
+  'One8' => 'One8',
+  'Two8' => 'Two8',
+) - array (
+  'One8' => 'One8',
+  'Two8' => 'Two8',
+) - OK
+enum16_c: array (
+  'Two16' => 'Two16',
+) - array (
+  'Two16' => 'Two16',
+) - OK
+fixedstring_c: array (
+  'fixedstring_c1' => 'fixedstring_c1',
+  'fixedstring_c2' => 'fixedstring_c2',
+) - array (
+  'fixedstring_c1' => 'fixedstring_c1',
+  'fixedstring_c2' => 'fixedstring_c2',
+) - OK
+int8null_c: array (
+  8 => 8,
+  '' => NULL,
+) - array (
+  8 => 8,
+  '' => NULL,
+) - OK
+stringnull_c: array (
+  'string' => 'string',
+  '' => NULL,
+) - array (
+  'string' => 'string',
+  '' => NULL,
+) - OK
+enumnull_c: array (
+  'One8' => 'One8',
+  '' => NULL,
+) - array (
+  'One8' => 'One8',
+  '' => NULL,
+) - OK
+float32null_c: array (
+  '7.77' => 7.77,
+  '' => NULL,
+) - array (
+  '7.77' => 7.77,
+  '' => NULL,
+) - OK
+uuidnull_c: array (
+  '31249a1b7b0542709f37c609b48a9bb2' => '31249a1b7b0542709f37c609b48a9bb2',
+  '' => NULL,
+) - array (
+  '31249a1b7b0542709f37c609b48a9bb2' => '31249a1b7b0542709f37c609b48a9bb2',
+  '' => NULL,
+) - OK
+int8_c: array (
+  8 => 8,
+  28 => 28,
+) - array (
+  8 => 8,
+  28 => 28,
+) - OK
+int16_c: array (
+  16 => 16,
+  216 => 216,
+) - array (
+  16 => 16,
+  216 => 216,
+) - OK
+uint8_c: array (
+  18 => 18,
+  218 => 218,
+) - array (
+  18 => 18,
+  218 => 218,
+) - OK
+uint16_c: array (
+  20 => 20,
+  220 => 220,
+) - array (
+  20 => 20,
+  220 => 220,
+) - OK
+float32_c: array (
+  '32.32' => 32.32,
+  '232.32' => 232.32,
+) - array (
+  '32.32' => 32.32,
+  '232.32' => 232.32,
+) - OK
+float64_c: array (
+  '64.64' => 64.64,
+  '264.64' => 264.64,
+) - array (
+  '64.64' => 64.64,
+  '264.64' => 264.64,
+) - OK
+uuid_c: array (
+  '31249a1b7b0542709f37c609b48a9bb2' => '31249a1b7b0542709f37c609b48a9bb2',
+) - array (
+  '31249a1b7b0542709f37c609b48a9bb2' => '31249a1b7b0542709f37c609b48a9bb2',
+) - OK
+uuid2_c: array (
+  '31249a1b7b0542709f37c609b48a9bb2' => '31249a1b7b0542709f37c609b48a9bb2',
+  '00000000000000000000000000000000' => '00000000000000000000000000000000',
+) - array (
+  '31249a1b7b0542709f37c609b48a9bb2' => '31249a1b7b0542709f37c609b48a9bb2',
+  '00000000000000000000000000000000' => '00000000000000000000000000000000',
+) - OK
+date_c: array (
+  1548633600 => 1548633600,
+  1548547200 => 1548547200,
+) - array (
+  1548633600 => 1548633600,
+  1548547200 => 1548547200,
+) - OK
+datetime_c: array (
+  1548687925 => 1548687925,
+  1548513600 => 1548513600,
+) - array (
+  1548687925 => 1548687925,
+  1548513600 => 1548513600,
+) - OK
diff --git a/tests/015.phpt b/tests/015.phpt
new file mode 100644
index 0000000..67a9879
--- /dev/null
+++ b/tests/015.phpt
@@ -0,0 +1,21 @@
+--TEST--
+ClickHouse Exception test
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$config = clickhouse_test_config();
+
+$deleteTable = true;
+$client = new ClickHouse($config);
+try {
+    $client->execute('FOO');
+} catch (ClickHouseException $e) {
+    var_dump(get_class($e));
+}
+?>
+--EXPECT--
+string(19) "ClickHouseException"
diff --git a/tests/016.phpt b/tests/016.phpt
new file mode 100644
index 0000000..a380297
--- /dev/null
+++ b/tests/016.phpt
@@ -0,0 +1,35 @@
+--TEST--
+ClickHouse Date32 round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.date32_t");
+$c->execute("CREATE TABLE test.date32_t (id UInt32, d Date32) ENGINE = Memory");
+
+// Pass epoch seconds (int) only; "Y-m-d" string input goes through
+// mktime which is host-TZ-dependent and would make the test flaky
+// across CI runners and dev hosts.
+$c->insert("test.date32_t", ["id", "d"], [
+    [1, 1745539200],   // 2025-04-25 UTC
+    [2, 0],            // 1970-01-01 UTC
+    [3, -2208988800],  // 1900-01-01 UTC
+]);
+
+$rows = $c->select("SELECT id, d FROM test.date32_t ORDER BY id", [], ClickHouse::DATE_AS_STRINGS);
+foreach ($rows as $r) {
+    echo $r["id"], " => ", $r["d"], "\n";
+}
+
+$c->execute("DROP TABLE test.date32_t");
+?>
+--EXPECT--
+1 => 2025-04-25
+2 => 1970-01-01
+3 => 1900-01-01
diff --git a/tests/017.phpt b/tests/017.phpt
new file mode 100644
index 0000000..71e85ac
--- /dev/null
+++ b/tests/017.phpt
@@ -0,0 +1,37 @@
+--TEST--
+ClickHouse Int128 / UInt128 / Decimal128 string round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.bignum_t");
+$c->execute("CREATE TABLE test.bignum_t (
+    id UInt32,
+    i128 Int128,
+    u128 UInt128,
+    dec128 Decimal128(10)
+) ENGINE = Memory");
+
+$c->insert("test.bignum_t", ["id", "i128", "u128", "dec128"], [
+    [1, "170141183460469231731687303715884105727", "340282366920938463463374607431768211455", "12345.6789012345"],
+    [2, "-170141183460469231731687303715884105727", "0", "-99999.9999999999"],
+    [3, 42, 42, "0"],
+]);
+
+$rows = $c->select("SELECT id, i128, u128, dec128 FROM test.bignum_t ORDER BY id");
+foreach ($rows as $r) {
+    echo $r["id"], "|", $r["i128"], "|", $r["u128"], "|", $r["dec128"], "\n";
+}
+
+$c->execute("DROP TABLE test.bignum_t");
+?>
+--EXPECT--
+1|170141183460469231731687303715884105727|340282366920938463463374607431768211455|12345.6789012345
+2|-170141183460469231731687303715884105727|0|-99999.9999999999
+3|42|42|0.0000000000
diff --git a/tests/018.phpt b/tests/018.phpt
new file mode 100644
index 0000000..669e881
--- /dev/null
+++ b/tests/018.phpt
@@ -0,0 +1,44 @@
+--TEST--
+ClickHouse LowCardinality(String) and LowCardinality(FixedString)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.lc_t");
+$c->execute("SET allow_suspicious_low_cardinality_types = 1");
+$c->execute("CREATE TABLE test.lc_t (
+    id UInt32,
+    env LowCardinality(String),
+    code LowCardinality(FixedString(4))
+) ENGINE = Memory");
+
+$c->insert("test.lc_t", ["id", "env", "code"], [
+    [1, "prod", "EU01"],
+    [2, "dev",  "US01"],
+    [3, "prod", "EU01"],
+    [4, "prod", "EU01"],
+]);
+
+$rows = $c->select("SELECT id, env, code FROM test.lc_t ORDER BY id");
+foreach ($rows as $r) {
+    echo $r["id"], "|", $r["env"], "|", trim($r["code"]), "\n";
+}
+
+echo "distinct env: ", $c->select("SELECT count(DISTINCT env) FROM test.lc_t", [], ClickHouse::FETCH_ONE), "\n";
+echo "distinct code: ", $c->select("SELECT count(DISTINCT code) FROM test.lc_t", [], ClickHouse::FETCH_ONE), "\n";
+
+$c->execute("DROP TABLE test.lc_t");
+?>
+--EXPECT--
+1|prod|EU01
+2|dev|US01
+3|prod|EU01
+4|prod|EU01
+distinct env: 2
+distinct code: 2
diff --git a/tests/019.phpt b/tests/019.phpt
new file mode 100644
index 0000000..ee72967
--- /dev/null
+++ b/tests/019.phpt
@@ -0,0 +1,61 @@
+--TEST--
+ClickHouse Map row I/O across supported K/V combinations
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.maps_t");
+$c->execute("CREATE TABLE test.maps_t (
+    id UInt32,
+    str_str Map(String, String),
+    str_int Map(String, Int64),
+    str_uint Map(String, UInt64),
+    str_flt Map(String, Float64),
+    int_str Map(Int64, String)
+) ENGINE = Memory");
+
+$c->insert("test.maps_t", ["id", "str_str", "str_int", "str_uint", "str_flt", "int_str"], [
+    [1,
+     ["env" => "prod", "tier" => "blue"],
+     ["count" => 42, "size" => -7],
+     ["bytes" => 1024],
+     ["pi" => 3.14],
+     [10 => "ten", 20 => "twenty"]],
+    [2, [], ["x" => 0], [], ["zero" => 0.0], [-1 => "neg"]],
+]);
+
+foreach ($c->select("SELECT id, str_str, str_int, str_uint, str_flt, int_str FROM test.maps_t ORDER BY id") as $r) {
+    ksort($r["str_str"]);
+    ksort($r["str_int"]);
+    ksort($r["str_uint"]);
+    ksort($r["str_flt"]);
+    ksort($r["int_str"]);
+    echo "id=", $r["id"], "\n";
+    echo "  str_str=",  json_encode($r["str_str"]),  "\n";
+    echo "  str_int=",  json_encode($r["str_int"]),  "\n";
+    echo "  str_uint=", json_encode($r["str_uint"]), "\n";
+    echo "  str_flt=",  json_encode($r["str_flt"]),  "\n";
+    echo "  int_str=",  json_encode($r["int_str"]),  "\n";
+}
+
+$c->execute("DROP TABLE test.maps_t");
+?>
+--EXPECT--
+id=1
+  str_str={"env":"prod","tier":"blue"}
+  str_int={"count":42,"size":-7}
+  str_uint={"bytes":1024}
+  str_flt={"pi":3.14}
+  int_str={"10":"ten","20":"twenty"}
+id=2
+  str_str=[]
+  str_int={"x":0}
+  str_uint=[]
+  str_flt={"zero":0}
+  int_str={"-1":"neg"}
diff --git a/tests/020.phpt b/tests/020.phpt
new file mode 100644
index 0000000..c30a288
--- /dev/null
+++ b/tests/020.phpt
@@ -0,0 +1,35 @@
+--TEST--
+ClickHouse query_id propagates to system.query_log
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+$qid_select = "clickhouse-test-q-sel-" . bin2hex(random_bytes(4));
+$qid_insert = "clickhouse-test-q-ins-" . bin2hex(random_bytes(4));
+$qid_drop   = "clickhouse-test-q-drop-" . bin2hex(random_bytes(4));
+
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.qid_t");
+$c->execute("CREATE TABLE test.qid_t (id UInt32) ENGINE = Memory");
+$c->insert("test.qid_t", ["id"], [[1], [2]], $qid_insert);
+$c->select("SELECT * FROM test.qid_t", [], 0, $qid_select);
+$c->execute("DROP TABLE test.qid_t", [], $qid_drop);
+
+$c->execute("SYSTEM FLUSH LOGS");
+
+$found = $c->select(
+    "SELECT count(*) FROM system.query_log WHERE query_id IN ('" .
+    $qid_select . "', '" . $qid_insert . "', '" . $qid_drop . "')",
+    [], ClickHouse::FETCH_ONE
+);
+
+echo "found: ", ((int)$found >= 3) ? "yes" : "no ($found)", "\n";
+?>
+--EXPECT--
+found: yes
diff --git a/tests/021.phpt b/tests/021.phpt
new file mode 100644
index 0000000..1adc817
--- /dev/null
+++ b/tests/021.phpt
@@ -0,0 +1,29 @@
+--TEST--
+ClickHouse ZSTD compression round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$config = clickhouse_test_config();
+$config["compression"] = "zstd";
+
+$c = new ClickHouse($config);
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.zstd_t");
+$c->execute("CREATE TABLE test.zstd_t (id UInt32, payload String) ENGINE = Memory");
+
+$payload = str_repeat("ab", 10000);
+$c->insert("test.zstd_t", ["id", "payload"], [[1, $payload], [2, $payload]]);
+
+$rows = $c->select("SELECT id, length(payload) AS l FROM test.zstd_t ORDER BY id");
+foreach ($rows as $r) echo $r["id"], " len=", $r["l"], "\n";
+
+$c->execute("DROP TABLE test.zstd_t");
+?>
+--EXPECT--
+1 len=20000
+2 len=20000
diff --git a/tests/022.phpt b/tests/022.phpt
new file mode 100644
index 0000000..1aa336d
--- /dev/null
+++ b/tests/022.phpt
@@ -0,0 +1,25 @@
+--TEST--
+ClickHouse multi-endpoint failover
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$base = clickhouse_test_config();
+
+// First endpoint is intentionally dead; second is the live test server.
+$config = $base + [
+    "endpoints" => [
+        ["host" => "127.0.0.1", "port" => 1],   // closed port
+        ["host" => $base["host"], "port" => $base["port"]],
+    ],
+];
+
+$c = new ClickHouse($config);
+echo "select via failover: ", $c->select("SELECT 1", [], ClickHouse::FETCH_ONE), "\n";
+?>
+--EXPECT--
+select via failover: 1
diff --git a/tests/023.phpt b/tests/023.phpt
new file mode 100644
index 0000000..5b5e0bf
--- /dev/null
+++ b/tests/023.phpt
@@ -0,0 +1,47 @@
+--TEST--
+ClickHouse Time and Time64 round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+clickhouse_skip_if_no_server();
+$c = new ClickHouse(clickhouse_test_config());
+try {
+    $c->execute("DROP TABLE IF EXISTS test._time_probe");
+    $c->execute("CREATE DATABASE IF NOT EXISTS test");
+    $c->execute("CREATE TABLE test._time_probe (t Time) ENGINE = Memory");
+    $c->execute("DROP TABLE test._time_probe");
+} catch (ClickHouseException $e) {
+    print "skip ClickHouse build does not support Time / Time64";
+    exit;
+}
+?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.time_t");
+$c->execute("CREATE TABLE test.time_t (id UInt32, t Time, t3 Time64(3)) ENGINE = Memory");
+
+// Time / Time64 inserts accept seconds-since-midnight (int) and Time64
+// internally scales by 10^precision.
+$c->insert("test.time_t", ["id", "t", "t3"], [
+    [1, 3661,  3661],   // 01:01:01 / 01:01:01.000
+    [2, 0,     0],      // 00:00:00 / 00:00:00.000
+    [3, -3600, -3600],  // -01:00:00 / -01:00:00.000
+]);
+
+$rows = $c->select("SELECT id, t, t3 FROM test.time_t ORDER BY id", [], ClickHouse::DATE_AS_STRINGS);
+foreach ($rows as $r) {
+    echo $r["id"], "|", $r["t"], "|", $r["t3"], "\n";
+}
+
+$c->execute("DROP TABLE test.time_t");
+?>
+--EXPECT--
+1|01:01:01|01:01:01.000
+2|00:00:00|00:00:00.000
+3|-01:00:00|-01:00:00.000
diff --git a/tests/024.phpt b/tests/024.phpt
new file mode 100644
index 0000000..80a8f27
--- /dev/null
+++ b/tests/024.phpt
@@ -0,0 +1,33 @@
+--TEST--
+ClickHouse DateTime64 sub-second round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.dt64_t");
+$c->execute("CREATE TABLE test.dt64_t (id UInt32, ts DateTime64(3, 'UTC'), ts6 DateTime64(6, 'UTC')) ENGINE = Memory");
+
+// Pass plain epoch seconds (int); the lib scales by 10^precision internally.
+// Only ints here; "Y-m-d H:i:s" string input goes through mktime which
+// is host-TZ-dependent and would make the test flaky.
+$c->insert("test.dt64_t", ["id", "ts", "ts6"], [
+    [1, 1714000000, 1714000000],
+    [2, 1745515200, 1745515200],
+]);
+
+$rows = $c->select("SELECT id, ts, ts6 FROM test.dt64_t ORDER BY id", [], ClickHouse::DATE_AS_STRINGS);
+foreach ($rows as $r) {
+    echo $r["id"], "|", $r["ts"], "|", $r["ts6"], "\n";
+}
+
+$c->execute("DROP TABLE test.dt64_t");
+?>
+--EXPECT--
+1|2024-04-24 23:06:40.000|2024-04-24 23:06:40.000000
+2|2025-04-24 17:20:00.000|2025-04-24 17:20:00.000000
diff --git a/tests/025.phpt b/tests/025.phpt
new file mode 100644
index 0000000..7a08939
--- /dev/null
+++ b/tests/025.phpt
@@ -0,0 +1,14 @@
+--TEST--
+ClickHouse ping round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+var_dump($c->ping());
+?>
+--EXPECT--
+bool(true)
diff --git a/tests/026.phpt b/tests/026.phpt
new file mode 100644
index 0000000..26fbf5a
--- /dev/null
+++ b/tests/026.phpt
@@ -0,0 +1,24 @@
+--TEST--
+ClickHouse SeasClick / SeasClickException BC aliases
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new SeasClick(clickhouse_test_config());
+var_dump($c instanceof ClickHouse);
+var_dump(get_class($c));
+try {
+    $c->execute("FOO");
+} catch (SeasClickException $e) {
+    var_dump($e instanceof ClickHouseException);
+    var_dump(get_class($e));
+}
+?>
+--EXPECT--
+bool(true)
+string(10) "ClickHouse"
+bool(true)
+string(19) "ClickHouseException"
diff --git a/tests/027.phpt b/tests/027.phpt
new file mode 100644
index 0000000..00d35b2
--- /dev/null
+++ b/tests/027.phpt
@@ -0,0 +1,38 @@
+--TEST--
+ClickHouse TLS round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_tls_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_tls_test_config());
+var_dump($c->ping());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.tls_t");
+$c->execute("CREATE TABLE test.tls_t (id UInt32) ENGINE = Memory");
+$c->insert("test.tls_t", ["id"], [[1], [2], [3]]);
+$rows = $c->select("SELECT id FROM test.tls_t ORDER BY id");
+var_dump($rows);
+$c->execute("DROP TABLE test.tls_t");
+?>
+--EXPECT--
+bool(true)
+array(3) {
+  [0]=>
+  array(1) {
+    ["id"]=>
+    int(1)
+  }
+  [1]=>
+  array(1) {
+    ["id"]=>
+    int(2)
+  }
+  [2]=>
+  array(1) {
+    ["id"]=>
+    int(3)
+  }
+}
diff --git a/tests/028.phpt b/tests/028.phpt
new file mode 100644
index 0000000..68ec620
--- /dev/null
+++ b/tests/028.phpt
@@ -0,0 +1,32 @@
+--TEST--
+ClickHouse query_id propagates through writeStart
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$qid = "clickhouse-test-q-ws-" . bin2hex(random_bytes(4));
+
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.qid_ws_t");
+$c->execute("CREATE TABLE test.qid_ws_t (id UInt32) ENGINE = Memory");
+
+$c->writeStart("test.qid_ws_t", ["id"], $qid);
+$c->write([[1], [2], [3]]);
+$c->writeEnd();
+
+$c->execute("SYSTEM FLUSH LOGS");
+$found = $c->select(
+    "SELECT count(*) FROM system.query_log WHERE query_id = '" . $qid . "'",
+    [], ClickHouse::FETCH_ONE
+);
+echo "found: ", ((int)$found >= 1) ? "yes" : "no ($found)", "\n";
+
+$c->execute("DROP TABLE test.qid_ws_t");
+?>
+--EXPECT--
+found: yes
diff --git a/tests/029.phpt b/tests/029.phpt
new file mode 100644
index 0000000..42c0579
--- /dev/null
+++ b/tests/029.phpt
@@ -0,0 +1,43 @@
+--TEST--
+ClickHouse default-database config knob
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$bootstrap = new ClickHouse(clickhouse_test_config());
+$bootstrap->execute("CREATE DATABASE IF NOT EXISTS test");
+unset($bootstrap);
+
+$config = clickhouse_test_config();
+$config["database"] = "test";
+$c = new ClickHouse($config);
+
+$c->execute("DROP TABLE IF EXISTS default_db_t");
+$c->execute("CREATE TABLE default_db_t (id UInt32) ENGINE = Memory");
+$c->insert("default_db_t", ["id"], [[1], [2], [3]]);
+$rows = $c->select("SELECT id FROM default_db_t ORDER BY id");
+var_dump($rows);
+$c->execute("DROP TABLE default_db_t");
+?>
+--EXPECT--
+array(3) {
+  [0]=>
+  array(1) {
+    ["id"]=>
+    int(1)
+  }
+  [1]=>
+  array(1) {
+    ["id"]=>
+    int(2)
+  }
+  [2]=>
+  array(1) {
+    ["id"]=>
+    int(3)
+  }
+}
diff --git a/tests/030.phpt b/tests/030.phpt
new file mode 100644
index 0000000..e0963c2
--- /dev/null
+++ b/tests/030.phpt
@@ -0,0 +1,42 @@
+--TEST--
+ClickHouse error paths raise ClickHouseException
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+
+function expect_throw(string $label, callable $fn): void {
+    try {
+        $fn();
+        echo $label, ": no throw\n";
+    } catch (ClickHouseException $e) {
+        echo $label, ": throw\n";
+    }
+}
+
+expect_throw("select bad sql", function () use ($c) {
+    $c->select("SELECT * FROM test.this_table_does_not_exist_xyz");
+});
+
+expect_throw("insert nonexistent table", function () use ($c) {
+    $c->insert("test.no_such_table_xyz", ["id"], [[1]]);
+});
+
+expect_throw("writeStart nonexistent table", function () use ($c) {
+    $c->writeStart("test.no_such_table_xyz", ["id"]);
+});
+
+expect_throw("execute bad sql", function () use ($c) {
+    $c->execute("THIS IS NOT SQL");
+});
+?>
+--EXPECT--
+select bad sql: throw
+insert nonexistent table: throw
+writeStart nonexistent table: throw
+execute bad sql: throw
diff --git a/tests/031.phpt b/tests/031.phpt
new file mode 100644
index 0000000..439a21e
--- /dev/null
+++ b/tests/031.phpt
@@ -0,0 +1,49 @@
+--TEST--
+ClickHouse {placeholder} substitution rejects unsafe values
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+
+function expect_throw(string $label, callable $fn): void {
+    try { $fn(); echo $label, ": no throw\n"; }
+    catch (ClickHouseException $e) { echo $label, ": throw\n"; }
+}
+
+expect_throw("semicolon",  function () use ($c) {
+    $c->select("SELECT {x} FROM system.one", ['x' => "1; DROP TABLE foo"]);
+});
+expect_throw("single quote", function () use ($c) {
+    $c->select("SELECT {x} FROM system.one", ['x' => "1' OR '1'='1"]);
+});
+expect_throw("backslash",  function () use ($c) {
+    $c->select("SELECT {x} FROM system.one", ['x' => "1\\0"]);
+});
+expect_throw("missing key", function () use ($c) {
+    $c->select("SELECT 1 FROM system.one", ['x' => "1"]);
+});
+
+// Safe values still work.
+$rows = $c->select("SELECT {col} FROM system.one", ['col' => 'dummy']);
+echo "safe identifier: ", isset($rows[0]['dummy']) ? "ok" : "fail", "\n";
+
+$one = $c->select("SELECT {n} AS x FROM system.one", ['n' => '42'], ClickHouse::FETCH_ONE);
+echo "safe numeric: $one\n";
+
+// Multiple occurrences of the same placeholder all replaced.
+$rows2 = $c->select("SELECT {col} AS a, {col} AS b FROM system.one", ['col' => '7']);
+echo "duplicate placeholder: ", $rows2[0]['a'], "/", $rows2[0]['b'], "\n";
+?>
+--EXPECT--
+semicolon: throw
+single quote: throw
+backslash: throw
+missing key: throw
+safe identifier: ok
+safe numeric: 42
+duplicate placeholder: 7/7
diff --git a/tests/032.phpt b/tests/032.phpt
new file mode 100644
index 0000000..cba50ec
--- /dev/null
+++ b/tests/032.phpt
@@ -0,0 +1,49 @@
+--TEST--
+ClickHouse insert/writeStart reject unsafe table and column identifiers
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.id_validate");
+$c->execute("CREATE TABLE test.id_validate (id UInt32) ENGINE = Memory");
+
+function expect_throw(string $label, callable $fn): void {
+    try { $fn(); echo $label, ": no throw\n"; }
+    catch (ClickHouseException $e) { echo $label, ": throw\n"; }
+}
+
+expect_throw("table semicolon", function () use ($c) {
+    $c->insert("test.id_validate; DROP TABLE test.id_validate", ["id"], [[1]]);
+});
+expect_throw("column quote", function () use ($c) {
+    $c->insert("test.id_validate", ["id', x"], [[1, 2]]);
+});
+expect_throw("table empty", function () use ($c) {
+    $c->insert("", ["id"], [[1]]);
+});
+expect_throw("column starts with digit", function () use ($c) {
+    $c->insert("test.id_validate", ["1id"], [[1]]);
+});
+expect_throw("writeStart bad table", function () use ($c) {
+    $c->writeStart("test.id`drop", ["id"]);
+});
+
+// Valid forms still work.
+$c->insert("test.id_validate", ["id"], [[1], [2]]);
+$rows = $c->select("SELECT id FROM test.id_validate ORDER BY id");
+echo "rows: ", count($rows), "\n";
+
+$c->execute("DROP TABLE test.id_validate");
+?>
+--EXPECT--
+table semicolon: throw
+column quote: throw
+table empty: throw
+column starts with digit: throw
+writeStart bad table: throw
+rows: 2
diff --git a/tests/033.phpt b/tests/033.phpt
new file mode 100644
index 0000000..e16f40b
--- /dev/null
+++ b/tests/033.phpt
@@ -0,0 +1,60 @@
+--TEST--
+ClickHouse object lifecycle: failed construct, half-finished insert, missing writeStart
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+function expect_throw(string $label, callable $fn): void {
+    try { $fn(); echo $label, ": no throw\n"; }
+    catch (ClickHouseException $e) { echo $label, ": throw\n"; }
+}
+
+// __construct against an unreachable host throws and the next method
+// call must throw cleanly (not std::out_of_range escaping the boundary).
+expect_throw("bad host construct", function () {
+    new ClickHouse([
+        "host" => "127.0.0.1",
+        "port" => 1,        // nothing listening
+        "connect_timeout" => 1,
+    ]);
+});
+
+// writeEnd without writeStart throws.
+$c = new ClickHouse(clickhouse_test_config());
+expect_throw("writeEnd no start", function () use ($c) {
+    $c->writeEnd();
+});
+
+// write without writeStart throws.
+expect_throw("write no start", function () use ($c) {
+    $c->write([[1]]);
+});
+
+// writeStart followed by orphan __destruct (no writeEnd) must not crash.
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.lifecycle_t");
+$c->execute("CREATE TABLE test.lifecycle_t (id UInt32) ENGINE = Memory");
+$c2 = new ClickHouse(clickhouse_test_config());
+$c2->writeStart("test.lifecycle_t", ["id"]);
+$c2->write([[1], [2]]);
+unset($c2);  // __destruct runs with the insert still pending
+echo "destruct after orphan writeStart: ok\n";
+
+// Re-construct on the same handle is rejected (would leak the old Client).
+$c3 = new ClickHouse(clickhouse_test_config());
+expect_throw("double construct", function () use ($c3) {
+    $c3->__construct(clickhouse_test_config());
+});
+
+$c->execute("DROP TABLE test.lifecycle_t");
+?>
+--EXPECT--
+bad host construct: throw
+writeEnd no start: throw
+write no start: throw
+destruct after orphan writeStart: ok
+double construct: throw
diff --git a/tests/034.phpt b/tests/034.phpt
new file mode 100644
index 0000000..1f08841
--- /dev/null
+++ b/tests/034.phpt
@@ -0,0 +1,38 @@
+--TEST--
+ClickHouseException carries server_code, server_name, query_id
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+
+$qid = "test-structured-exc-" . bin2hex(random_bytes(4));
+
+try {
+    $c->select("SELECT * FROM no_such_db_xyz_zz.no_such_table", [], 0, $qid);
+    echo "no throw\n";
+} catch (ClickHouseException $e) {
+    echo "code>0: ",       ($e->server_code > 0 ? "yes" : "no"), "\n";
+    echo "name set: ",     (is_string($e->server_name) && $e->server_name !== "" ? "yes" : "no"), "\n";
+    echo "query_id: ",     ($e->query_id === $qid ? "match" : "mismatch"), "\n";
+    echo "name prefix: ",  (strpos((string)$e->server_name, "DB::") === 0 ? "DB::" : "other"), "\n";
+}
+
+// Client-side error: no server fields populated.
+try {
+    $c->select("SELECT 1", ["foo" => "bar"]);
+} catch (ClickHouseException $e) {
+    echo "client code: ",  $e->server_code, "\n";
+    echo "client name: ",  var_export($e->server_name, true), "\n";
+}
+?>
+--EXPECT--
+code>0: yes
+name set: yes
+query_id: match
+name prefix: DB::
+client code: 0
+client name: NULL
diff --git a/tests/035.phpt b/tests/035.phpt
new file mode 100644
index 0000000..612bd3f
--- /dev/null
+++ b/tests/035.phpt
@@ -0,0 +1,45 @@
+--TEST--
+ClickHouse setSettings + per-call settings precedence
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+
+// Global setting: limit to 1ms — any non-trivial query trips it.
+$c->setSettings(["max_execution_time" => "0.001"]);
+
+try {
+    $c->select("SELECT count() FROM numbers(100000000)");
+    echo "global: no throw\n";
+} catch (ClickHouseException $e) {
+    // Code 159 = TIMEOUT_EXCEEDED.
+    echo "global code: ", ($e->server_code === 159 ? "159" : "other:" . $e->server_code), "\n";
+}
+
+// Per-call override: relax the timeout. Should succeed.
+$res = $c->select("SELECT 1 AS x", [], 0, "", ["max_execution_time" => "30"]);
+echo "per-call x: ", $res[0]["x"], "\n";
+
+// Per-call setting also wins on a tiny value.
+try {
+    $c->select("SELECT count() FROM numbers(100000000)", [], 0, "",
+               ["max_execution_time" => "0.001"]);
+    echo "per-call: no throw\n";
+} catch (ClickHouseException $e) {
+    echo "per-call code: ", ($e->server_code === 159 ? "159" : "other:" . $e->server_code), "\n";
+}
+
+// Clear globals; without a per-call setting the query runs.
+$c->setSettings([]);
+$res = $c->select("SELECT 2 AS x");
+echo "cleared x: ", $res[0]["x"], "\n";
+?>
+--EXPECT--
+global code: 159
+per-call x: 1
+per-call code: 159
+cleared x: 2
diff --git a/tests/036.phpt b/tests/036.phpt
new file mode 100644
index 0000000..b5ed72f
--- /dev/null
+++ b/tests/036.phpt
@@ -0,0 +1,95 @@
+--TEST--
+ClickHouse server-side typed parameters {name:Type}
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+
+// Scalar typed params.
+$res = $c->select(
+    "SELECT {n:UInt32} AS n, {s:String} AS s",
+    ["n" => 42, "s" => "hello world"]
+);
+echo "scalar n: ",  $res[0]["n"], "\n";
+echo "scalar s: ",  $res[0]["s"], "\n";
+
+// Array typed param: numeric Array.
+$res = $c->select(
+    "SELECT arrayJoin({ids:Array(UInt32)}) AS id ORDER BY id",
+    ["ids" => [3, 1, 2]]
+);
+echo "array ids: ",  implode(",", array_column($res, "id")), "\n";
+
+// Array typed param: string Array (server quotes per element).
+$res = $c->select(
+    "SELECT arrayJoin({xs:Array(String)}) AS s",
+    ["xs" => ["a", "b", "c"]]
+);
+echo "array xs: ", implode(",", array_column($res, "s")), "\n";
+
+// Mix: typed param plus identifier substitution in same SQL.
+$res = $c->select(
+    "SELECT {n:UInt32} AS {col}",
+    ["n" => 7, "col" => "answer"]
+);
+echo "mix: ",  $res[0]["answer"], "\n";
+
+// String containing single quote: server unescapes.
+$res = $c->select("SELECT {s:String} AS s", ["s" => "it's fine"]);
+echo "quote: ",  $res[0]["s"], "\n";
+
+// Null routed via SetParam(nullopt).
+$res = $c->select("SELECT {x:Nullable(UInt32)} AS x", ["x" => null]);
+var_dump($res[0]["x"]);
+
+// Null elements inside Array(Nullable(T)) must stay NULL, not empty
+// strings or malformed empty numeric tokens.
+$res = $c->select(
+    "SELECT arrayJoin({xs:Array(Nullable(String))}) AS s",
+    ["xs" => ["a", null, ""]]
+);
+foreach ($res as $row) {
+    var_dump($row["s"]);
+}
+
+$res = $c->select(
+    "SELECT arrayJoin({ids:Array(Nullable(UInt32))}) AS id",
+    ["ids" => [1, null, 3]]
+);
+foreach ($res as $row) {
+    var_dump($row["id"]);
+}
+
+try {
+    $c->select(
+        "SELECT arrayJoin({ids:Array(UInt32)}) AS id",
+        ["ids" => [1, null]]
+    );
+    echo "nonnull array null: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "nonnull array null: ";
+    echo strpos($e->getMessage(), "NULL element in non-Nullable Array") !== false
+        ? "LOCAL REJECT"
+        : "OTHER REJECT";
+    echo "\n";
+}
+?>
+--EXPECT--
+scalar n: 42
+scalar s: hello world
+array ids: 1,2,3
+array xs: a,b,c
+mix: 7
+quote: it's fine
+NULL
+string(1) "a"
+NULL
+string(0) ""
+int(1)
+NULL
+int(3)
+nonnull array null: LOCAL REJECT
diff --git a/tests/037.phpt b/tests/037.phpt
new file mode 100644
index 0000000..9b740c6
--- /dev/null
+++ b/tests/037.phpt
@@ -0,0 +1,62 @@
+--TEST--
+ClickHouse setProgressCallback fires + getStatistics returns rows/bytes
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+
+// Progress callback collects rows seen across all packets.
+$seen_rows = 0;
+$call_count = 0;
+$c->setProgressCallback(function (array $p) use (&$seen_rows, &$call_count) {
+    $seen_rows += $p["rows"];
+    $call_count++;
+});
+
+$res = $c->select("SELECT count() FROM numbers(100000)");
+echo "select count: ", $res[0]["count()"], "\n";
+echo "callback fired: ", ($call_count > 0 ? "yes" : "no"), "\n";
+echo "rows>=100000: ", ($seen_rows >= 100000 ? "yes" : "no"), "\n";
+
+$stats = $c->getStatistics();
+echo "stat rows>=100000: ", ($stats["rows_read"] >= 100000 ? "yes" : "no"), "\n";
+echo "stat bytes>0: ",      ($stats["bytes_read"] > 0 ? "yes" : "no"), "\n";
+echo "stat blocks>=1: ",    ($stats["blocks"] >= 1 ? "yes" : "no"), "\n";
+echo "stat elapsed>0: ",    ($stats["elapsed_ms"] > 0 ? "yes" : "no"), "\n";
+
+// Stats reset on each call.
+$c->select("SELECT 1");
+$stats = $c->getStatistics();
+echo "after small: rows<10: ", ($stats["rows_read"] < 10 ? "yes" : "no"), "\n";
+
+// Removing the callback works.
+$c->setProgressCallback(null);
+$call_count = 0;
+$c->select("SELECT count() FROM numbers(100000)");
+echo "no callback: ", ($call_count === 0 ? "0" : (string)$call_count), "\n";
+
+// Bad callable rejected. PHP 7.4's ZPP strictly validates `callable` and
+// rejects with TypeError; 8.0+ relaxed this so our C-body is_callable()
+// check fires ClickHouseException instead. Catch Throwable to cover both.
+try {
+    $c->setProgressCallback("not_a_real_function_xyz");
+    echo "bad: no throw\n";
+} catch (\Throwable $e) {
+    echo "bad: throw\n";
+}
+?>
+--EXPECT--
+select count: 100000
+callback fired: yes
+rows>=100000: yes
+stat rows>=100000: yes
+stat bytes>0: yes
+stat blocks>=1: yes
+stat elapsed>0: yes
+after small: rows<10: yes
+no callback: 0
+bad: throw
diff --git a/tests/038.phpt b/tests/038.phpt
new file mode 100644
index 0000000..f0cc866
--- /dev/null
+++ b/tests/038.phpt
@@ -0,0 +1,41 @@
+--TEST--
+ClickHouse connect_timeout_ms triggers fast failure
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// 192.0.2.0/24 is reserved for documentation (RFC 5737); routing it
+// produces a SYN-RETRANS that won't complete inside our budget.
+$cfg = [
+    "host"               => "192.0.2.1",
+    "port"               => 9000,
+    "connect_timeout_ms" => 200,
+    "retry_count"        => 1,
+];
+
+$t0 = microtime(true);
+try {
+    $c = new ClickHouse($cfg);
+    echo "no throw\n";
+} catch (ClickHouseException $e) {
+    $elapsed_ms = (microtime(true) - $t0) * 1000;
+    echo "throw: yes\n";
+    echo "fast: ", ($elapsed_ms < 2000 ? "yes" : "no:" . (int)$elapsed_ms), "\n";
+}
+
+// Negative ms rejected.
+try {
+    new ClickHouse(["host" => "127.0.0.1", "port" => 9000, "connect_timeout_ms" => -1]);
+    echo "neg: no throw\n";
+} catch (ClickHouseException $e) {
+    echo "neg: throw\n";
+}
+?>
+--EXPECT--
+throw: yes
+fast: yes
+neg: throw
diff --git a/tests/039.phpt b/tests/039.phpt
new file mode 100644
index 0000000..44c661a
--- /dev/null
+++ b/tests/039.phpt
@@ -0,0 +1,61 @@
+--TEST--
+ClickHouse insertAssoc derives column order from first row
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.assoc_round_trip");
+$c->execute(
+    "CREATE TABLE test.assoc_round_trip (id UInt32, name String, score Float32) ".
+    "ENGINE = Memory"
+);
+
+$rows = [
+    ["id" => 1, "name" => "alice", "score" => 1.5],
+    ["id" => 2, "name" => "bob",   "score" => 2.5],
+    ["id" => 3, "name" => "carol", "score" => 3.5],
+];
+$c->insertAssoc("test.assoc_round_trip", $rows);
+
+$res = $c->select("SELECT id, name, score FROM test.assoc_round_trip ORDER BY id");
+foreach ($res as $row) {
+    echo $row["id"], " ", $row["name"], " ", $row["score"], "\n";
+}
+
+// Reordered keys on later rows still align via key lookup.
+$c->execute("TRUNCATE TABLE test.assoc_round_trip");
+$rows = [
+    ["id" => 10, "name" => "x", "score" => 0.1],
+    ["score" => 0.2, "id" => 20, "name" => "y"],
+];
+$c->insertAssoc("test.assoc_round_trip", $rows);
+$res = $c->select("SELECT id, name FROM test.assoc_round_trip ORDER BY id");
+echo $res[0]["id"], " ", $res[0]["name"], "\n";
+echo $res[1]["id"], " ", $res[1]["name"], "\n";
+
+// Missing key triggers a clear error.
+try {
+    $c->insertAssoc("test.assoc_round_trip", [
+        ["id" => 99, "name" => "ok", "score" => 0.0],
+        ["id" => 100, "name" => "missing"],
+    ]);
+    echo "missing: no throw\n";
+} catch (ClickHouseException $e) {
+    echo "missing: throw\n";
+}
+
+$c->execute("DROP TABLE test.assoc_round_trip");
+?>
+--EXPECT--
+1 alice 1.5
+2 bob 2.5
+3 carol 3.5
+10 x
+20 y
+missing: throw
diff --git a/tests/040.phpt b/tests/040.phpt
new file mode 100644
index 0000000..64a2e0a
--- /dev/null
+++ b/tests/040.phpt
@@ -0,0 +1,67 @@
+--TEST--
+ClickHouse SQL helper methods (databaseSize, tablesSize, partitions, showTables, showCreateTable, getServerUptime)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config() + ["database" => "test"]);
+
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.helpers_a");
+$c->execute("DROP TABLE IF EXISTS test.helpers_b");
+$c->execute("CREATE TABLE test.helpers_a (id UInt32) ENGINE = MergeTree ORDER BY id");
+$c->execute("CREATE TABLE test.helpers_b (n UInt64) ENGINE = MergeTree ORDER BY n");
+$c->insert("test.helpers_a", ["id"], [[1], [2], [3]]);
+$c->insert("test.helpers_b", ["n"],  [[10], [20]]);
+
+// showTables filters by database; default-database aware.
+$tables = $c->showTables("test", "helpers_%");
+echo "showTables: ",  implode(",", $tables), "\n";
+
+// showCreateTable returns the DDL string; just check the leading prefix.
+$ddl = $c->showCreateTable("test.helpers_a");
+echo "showCreateTable starts: ", (strpos($ddl, "CREATE TABLE test.helpers_a") === 0 ? "yes" : "no"), "\n";
+
+// databaseSize: rows summed across active parts.
+$sz = $c->databaseSize("test");
+echo "databaseSize rows>=5: ", ($sz["rows"] >= 5 ? "yes" : "no"), "\n";
+echo "databaseSize bytes>0: ", ($sz["bytes_on_disk"] > 0 ? "yes" : "no"), "\n";
+
+// tablesSize lists the helper tables.
+$ts = $c->tablesSize("test");
+$names = array_column($ts, "table");
+echo "tablesSize has helpers_a: ", (in_array("helpers_a", $names) ? "yes" : "no"), "\n";
+echo "tablesSize has helpers_b: ", (in_array("helpers_b", $names) ? "yes" : "no"), "\n";
+
+// partitions: at least one row for the populated table.
+$parts = $c->partitions("test.helpers_a");
+echo "partitions rows: ", count($parts), "\n";
+
+// getServerUptime returns a non-negative integer.
+$up = $c->getServerUptime();
+echo "uptime>=0: ", ($up >= 0 ? "yes" : "no"), "\n";
+
+// Identifier validation: malicious-looking arg rejected.
+try {
+    $c->showCreateTable("foo' OR 1=1 --");
+    echo "bad ident: no throw\n";
+} catch (ClickHouseException $e) {
+    echo "bad ident: throw\n";
+}
+
+$c->execute("DROP TABLE test.helpers_a");
+$c->execute("DROP TABLE test.helpers_b");
+?>
+--EXPECT--
+showTables: helpers_a,helpers_b
+showCreateTable starts: yes
+databaseSize rows>=5: yes
+databaseSize bytes>0: yes
+tablesSize has helpers_a: yes
+tablesSize has helpers_b: yes
+partitions rows: 1
+uptime>=0: yes
+bad ident: throw
diff --git a/tests/041.phpt b/tests/041.phpt
new file mode 100644
index 0000000..5fca953
--- /dev/null
+++ b/tests/041.phpt
@@ -0,0 +1,89 @@
+--TEST--
+ClickHouse enableLogQueries / getLogQueries accumulator
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+$c = new ClickHouse(clickhouse_test_config());
+
+// Disabled by default: no entries collected.
+$c->select("SELECT 1");
+echo "off count: ", count($c->getLogQueries()), "\n";
+
+// Enable and run a mix of statements.
+$c->enableLogQueries(true);
+$c->select("SELECT 1 AS x");
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.log_round_trip");
+$c->execute("CREATE TABLE test.log_round_trip (id UInt32) ENGINE = Memory");
+$c->insert("test.log_round_trip", ["id"], [[1], [2], [3]]);
+
+// One server error to verify error_code routing.
+try {
+    $c->select("SELECT * FROM test.no_such_xyz", [], 0, "qid-error");
+} catch (ClickHouseException $e) {
+}
+
+$log = $c->getLogQueries();
+echo "log entries: ", count($log), "\n";
+
+// Inspect first entry shape.
+$first = $log[0];
+echo "first sql: ",        $first["sql"], "\n";
+echo "first query_id: ",   var_export($first["query_id"], true), "\n";
+echo "first elapsed>=0: ", ($first["elapsed_ms"] >= 0 ? "yes" : "no"), "\n";
+echo "first error_code: ", $first["error_code"], "\n";
+
+// The INSERT should have rows_read==0 (server doesn't return rows on insert)
+// but its SQL should be the INSERT INTO statement.
+$insert_log = null;
+foreach ($log as $entry) {
+    if (strpos($entry["sql"], "INSERT INTO test.log_round_trip") === 0) {
+        $insert_log = $entry;
+        break;
+    }
+}
+echo "insert sql found: ", ($insert_log ? "yes" : "no"), "\n";
+echo "insert error_code: ", ($insert_log["error_code"] ?? -99), "\n";
+
+// The error entry should carry the server code and the failing SQL.
+$error_log = null;
+foreach ($log as $entry) {
+    if ($entry["error_code"] !== 0) {
+        $error_log = $entry;
+        break;
+    }
+}
+echo "error captured: ",   ($error_log ? "yes" : "no"), "\n";
+echo "error qid: ",        $error_log["query_id"], "\n";
+echo "error code>0: ",     ($error_log["error_code"] > 0 ? "yes" : "no"), "\n";
+echo "error msg set: ",    ($error_log["error_message"] !== "" ? "yes" : "no"), "\n";
+
+// getLogQueries clears the buffer.
+echo "second get: ", count($c->getLogQueries()), "\n";
+
+// Disable: subsequent queries don't add.
+$c->enableLogQueries(false);
+$c->select("SELECT 2");
+echo "after off: ", count($c->getLogQueries()), "\n";
+
+$c->execute("DROP TABLE test.log_round_trip");
+?>
+--EXPECT--
+off count: 0
+log entries: 6
+first sql: SELECT 1 AS x
+first query_id: ''
+first elapsed>=0: yes
+first error_code: 0
+insert sql found: yes
+insert error_code: 0
+error captured: yes
+error qid: qid-error
+error code>0: yes
+error msg set: yes
+second get: 0
+after off: 0
diff --git a/tests/042.phpt b/tests/042.phpt
new file mode 100644
index 0000000..abcda6a
--- /dev/null
+++ b/tests/042.phpt
@@ -0,0 +1,52 @@
+--TEST--
+ClickHouse LowCardinality(Nullable(String)) and LowCardinality(Nullable(FixedString)) round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.lc_n_t");
+$c->execute("SET allow_suspicious_low_cardinality_types = 1");
+$c->execute("CREATE TABLE test.lc_n_t (
+    id UInt32,
+    tag LowCardinality(Nullable(String)),
+    code LowCardinality(Nullable(FixedString(4)))
+) ENGINE = Memory");
+
+$c->insert("test.lc_n_t", ["id", "tag", "code"], [
+    [1, "prod", "EU01"],
+    [2, null,   "US01"],
+    [3, "prod", null],
+    [4, null,   null],
+    [5, "dev",  "EU01"],
+    [6, "prod", "EU01"],
+]);
+
+$rows = $c->select("SELECT id, tag, code FROM test.lc_n_t ORDER BY id");
+foreach ($rows as $r) {
+    $tag  = $r["tag"]  === null ? "NULL" : $r["tag"];
+    $code = $r["code"] === null ? "NULL" : trim($r["code"]);
+    echo $r["id"], "|", $tag, "|", $code, "\n";
+}
+
+echo "non-null tag: ",  $c->select("SELECT count() FROM test.lc_n_t WHERE tag IS NOT NULL",  [], ClickHouse::FETCH_ONE), "\n";
+echo "null code: ",     $c->select("SELECT count() FROM test.lc_n_t WHERE code IS NULL",     [], ClickHouse::FETCH_ONE), "\n";
+echo "distinct tag: ",  $c->select("SELECT count(DISTINCT tag) FROM test.lc_n_t",            [], ClickHouse::FETCH_ONE), "\n";
+
+$c->execute("DROP TABLE test.lc_n_t");
+?>
+--EXPECT--
+1|prod|EU01
+2|NULL|US01
+3|prod|NULL
+4|NULL|NULL
+5|dev|EU01
+6|prod|EU01
+non-null tag: 4
+null code: 2
+distinct tag: 2
diff --git a/tests/043.phpt b/tests/043.phpt
new file mode 100644
index 0000000..4aa5076
--- /dev/null
+++ b/tests/043.phpt
@@ -0,0 +1,51 @@
+--TEST--
+ClickHouse Geo types: Point, Ring, Polygon, MultiPolygon round-trip
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.geo_t");
+$c->execute("SET allow_experimental_geo_types = 1");
+$c->execute("CREATE TABLE test.geo_t (
+    id UInt32,
+    p Point,
+    r Ring,
+    poly Polygon,
+    mp MultiPolygon
+) ENGINE = Memory");
+
+$c->insert("test.geo_t", ["id", "p", "r", "poly", "mp"], [
+    [
+        1,
+        [10.5, 20.5],                                                 // Point
+        [[0, 0], [1, 0], [1, 1], [0, 1]],                             // Ring
+        [[[0, 0], [4, 0], [4, 4], [0, 4]], [[1, 1], [2, 1], [2, 2]]], // Polygon = list of rings
+        [                                                             // MultiPolygon = list of polygons
+            [[[0, 0], [1, 0], [1, 1]]],
+            [[[10, 10], [11, 10], [11, 11]]],
+        ],
+    ],
+]);
+
+$rows = $c->select("SELECT id, p, r, poly, mp FROM test.geo_t WHERE id = 1");
+$row = $rows[0];
+echo "id=", $row["id"], "\n";
+echo "p=", json_encode($row["p"]), "\n";
+echo "r=", json_encode($row["r"]), "\n";
+echo "poly=", json_encode($row["poly"]), "\n";
+echo "mp=", json_encode($row["mp"]), "\n";
+
+$c->execute("DROP TABLE test.geo_t");
+?>
+--EXPECT--
+id=1
+p=[10.5,20.5]
+r=[[0,0],[1,0],[1,1],[0,1]]
+poly=[[[0,0],[4,0],[4,4],[0,4]],[[1,1],[2,1],[2,2]]]
+mp=[[[[0,0],[1,0],[1,1]]],[[[10,10],[11,10],[11,11]]]]
diff --git a/tests/044.phpt b/tests/044.phpt
new file mode 100644
index 0000000..5f659e5
--- /dev/null
+++ b/tests/044.phpt
@@ -0,0 +1,37 @@
+--TEST--
+ClickHouse SimpleAggregateFunction(f, T) reads transparently as T
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.saf_t");
+$c->execute("CREATE TABLE test.saf_t (
+    k UInt32,
+    s SimpleAggregateFunction(sum, UInt64),
+    m SimpleAggregateFunction(max, Float64),
+    n SimpleAggregateFunction(any, String)
+) ENGINE = AggregatingMergeTree() ORDER BY k");
+
+$c->insert("test.saf_t", ["k", "s", "m", "n"], [
+    [1, 10, 1.5, "a"],
+    [1,  5, 2.5, "b"],
+    [2,  7, 0.5, "c"],
+]);
+$c->execute("OPTIMIZE TABLE test.saf_t FINAL");
+
+$rows = $c->select("SELECT k, sum(s) AS sum_s, max(m) AS max_m, any(n) AS any_n FROM test.saf_t GROUP BY k ORDER BY k");
+foreach ($rows as $r) {
+    echo $r["k"], "|", $r["sum_s"], "|", $r["max_m"], "|", $r["any_n"], "\n";
+}
+
+$c->execute("DROP TABLE test.saf_t");
+?>
+--EXPECTREGEX--
+1\|15\|2\.5\|[ab]
+2\|7\|0\.5\|c
diff --git a/tests/045.phpt b/tests/045.phpt
new file mode 100644
index 0000000..abd7ac0
--- /dev/null
+++ b/tests/045.phpt
@@ -0,0 +1,30 @@
+--TEST--
+ClickHouse Array of complex inner types: Array(Tuple), Array(Nullable), Array(Array)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+
+// Array(Nullable(String)) read path -- the inner null mask must round-trip.
+$rows = $c->select("SELECT [1, NULL, 3, NULL] AS arr_n");
+echo "arr_n=", json_encode($rows[0]["arr_n"]), "\n";
+
+// Array(Tuple(...)) read path.
+$rows = $c->select("SELECT [(1, 'a'), (2, 'b')] AS arr_t");
+echo "arr_t=", json_encode($rows[0]["arr_t"]), "\n";
+
+// Array(Array(...)) -- two-dim, server-built.
+$rows = $c->select("SELECT [[1, 2], [3], []] AS arr_a");
+echo "arr_a=", json_encode($rows[0]["arr_a"]), "\n";
+
+?>
+--EXPECT--
+arr_n=[1,null,3,null]
+arr_t=[[1,"a"],[2,"b"]]
+arr_a=[[1,2],[3],[]]
diff --git a/tests/046.phpt b/tests/046.phpt
new file mode 100644
index 0000000..f3e8027
--- /dev/null
+++ b/tests/046.phpt
@@ -0,0 +1,69 @@
+--TEST--
+ClickHouse Phase C: ping_before_query, resetConnection, setProfileCallback, getServerInfo, getCurrentEndpoint
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// 1. ping_before_query construction key + getServerInfo
+$cfg = clickhouse_test_config();
+$cfg["ping_before_query"] = true;
+$c = new ClickHouse($cfg);
+
+$si = $c->getServerInfo();
+echo "server.has_name=", isset($si["name"])         && is_string($si["name"])      ? 1 : 0, "\n";
+echo "server.has_tz=",   isset($si["timezone"])     && is_string($si["timezone"])  ? 1 : 0, "\n";
+echo "server.major_int=", is_int($si["version_major"]) ? 1 : 0, "\n";
+echo "server.revision_pos=", $si["revision"] > 0 ? 1 : 0, "\n";
+
+// 2. getCurrentEndpoint returns the active host/port (clickhouse-cpp
+//    models a single host/port as a degenerate 1-item endpoints list).
+$ep0 = $c->getCurrentEndpoint();
+echo "single_ep_host_str=", is_string($ep0["host"]) ? 1 : 0, "\n";
+echo "single_ep_port_pos=", $ep0["port"] > 0 ? 1 : 0, "\n";
+
+// 3. resetConnection round-trips
+echo "reset=", $c->resetConnection() ? 1 : 0, "\n";
+echo "ping_after_reset=", $c->ping() ? 1 : 0, "\n";
+
+// 4. setProfileCallback fires
+$got = ["count" => 0, "rows_seen" => 0];
+$c->setProfileCallback(function (array $p) use (&$got) {
+    $got["count"]++;
+    if (isset($p["rows"])) $got["rows_seen"] += $p["rows"];
+});
+$c->select("SELECT number FROM system.numbers LIMIT 100");
+echo "profile_fired=", $got["count"] > 0 ? 1 : 0, "\n";
+
+// 5. setProfileCallback(null) detaches cleanly
+$c->setProfileCallback(null);
+$got = ["count" => 0];
+$c->select("SELECT 1");
+echo "profile_after_clear=", $got["count"] === 0 ? 1 : 0, "\n";
+
+// 6. getCurrentEndpoint returns the active endpoint when endpoints[] is set
+$cfg2 = clickhouse_test_config();
+$cfg2["endpoints"] = [
+    ["host" => $cfg2["host"], "port" => $cfg2["port"]],
+];
+$c2 = new ClickHouse($cfg2);
+$ep = $c2->getCurrentEndpoint();
+echo "endpoint_host_match=", ($ep["host"] === $cfg2["host"]) ? 1 : 0, "\n";
+echo "endpoint_port_match=", ($ep["port"] === $cfg2["port"]) ? 1 : 0, "\n";
+?>
+--EXPECT--
+server.has_name=1
+server.has_tz=1
+server.major_int=1
+server.revision_pos=1
+single_ep_host_str=1
+single_ep_port_pos=1
+reset=1
+ping_after_reset=1
+profile_fired=1
+profile_after_clear=1
+endpoint_host_match=1
+endpoint_port_match=1
diff --git a/tests/047.phpt b/tests/047.phpt
new file mode 100644
index 0000000..ee4bfa8
--- /dev/null
+++ b/tests/047.phpt
@@ -0,0 +1,42 @@
+--TEST--
+ClickHouse Phase D: getStatistics() echoes the query_id supplied to select/execute/insert
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+// 1. select() with explicit query_id round-trips through getStatistics()
+$qid = "qid-select-" . bin2hex(random_bytes(4));
+$c->select("SELECT 1", [], 0, $qid);
+$st = $c->getStatistics();
+echo "select_qid_match=", ($st["query_id"] === $qid) ? 1 : 0, "\n";
+
+// 2. execute() with query_id
+$qid2 = "qid-exec-" . bin2hex(random_bytes(4));
+$c->execute("SELECT 1", [], $qid2);
+echo "execute_qid_match=", ($c->getStatistics()["query_id"] === $qid2) ? 1 : 0, "\n";
+
+// 3. select() without query_id => empty string
+$c->select("SELECT 1");
+echo "no_qid_is_empty=", ($c->getStatistics()["query_id"] === "") ? 1 : 0, "\n";
+
+// 4. insert() with query_id
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.qid_t");
+$c->execute("CREATE TABLE test.qid_t (k UInt32) ENGINE=Memory");
+$qid3 = "qid-ins-" . bin2hex(random_bytes(4));
+$c->insert("test.qid_t", ["k"], [[1]], $qid3);
+echo "insert_qid_match=", ($c->getStatistics()["query_id"] === $qid3) ? 1 : 0, "\n";
+$c->execute("DROP TABLE test.qid_t");
+
+?>
+--EXPECT--
+select_qid_match=1
+execute_qid_match=1
+no_qid_is_empty=1
+insert_qid_match=1
diff --git a/tests/048.phpt b/tests/048.phpt
new file mode 100644
index 0000000..b7566ec
--- /dev/null
+++ b/tests/048.phpt
@@ -0,0 +1,54 @@
+--TEST--
+ClickHouse Phase E: ClickHouseRowIterator (selectStream) basic foreach + count + key
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+$iter = $c->selectStream("SELECT number AS n FROM system.numbers LIMIT 5");
+echo "is_iter=", ($iter instanceof ClickHouseRowIterator) ? 1 : 0, "\n";
+echo "is_iterator_iface=", ($iter instanceof Iterator) ? 1 : 0, "\n";
+echo "is_countable_iface=", ($iter instanceof Countable) ? 1 : 0, "\n";
+echo "count=", count($iter), "\n";
+
+$keys = [];
+$vals = [];
+foreach ($iter as $k => $row) {
+    $keys[] = $k;
+    $vals[] = $row["n"];
+}
+echo "keys=", implode(",", $keys), "\n";
+echo "vals=", implode(",", $vals), "\n";
+
+// Iterator survives unset of the source client (blocks are owned by iter via shared_ptr).
+$iter2 = $c->selectStream("SELECT number AS n FROM system.numbers LIMIT 3");
+unset($c);
+$out = [];
+foreach ($iter2 as $row) {
+    $out[] = $row["n"];
+}
+echo "after_unset=", implode(",", $out), "\n";
+
+// selectStreamCallback fires per-row.
+$c2 = new ClickHouse(clickhouse_test_config());
+$got = [];
+$c2->selectStreamCallback("SELECT number AS n FROM system.numbers LIMIT 4", function(array $row) use (&$got) {
+    $got[] = $row["n"];
+});
+echo "cb_rows=", implode(",", $got), "\n";
+
+?>
+--EXPECT--
+is_iter=1
+is_iterator_iface=1
+is_countable_iface=1
+count=5
+keys=0,1,2,3,4
+vals=0,1,2,3,4
+after_unset=0,1,2
+cb_rows=0,1,2,3
diff --git a/tests/049.phpt b/tests/049.phpt
new file mode 100644
index 0000000..07b343b
--- /dev/null
+++ b/tests/049.phpt
@@ -0,0 +1,77 @@
+--TEST--
+ClickHouse Phase F: DDL helpers (isExists, showDatabases, showProcesslist, getServerVersion, tableSize)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.ddl_t");
+$c->execute("CREATE TABLE test.ddl_t (k UInt32) ENGINE = MergeTree() ORDER BY k");
+$c->insert("test.ddl_t", ["k"], [[1], [2], [3]]);
+
+echo "exists_yes=", $c->isExists("test", "ddl_t") ? 1 : 0, "\n";
+echo "exists_no=",  $c->isExists("test", "no_such_table_zzz") ? 1 : 0, "\n";
+
+$dbs = $c->showDatabases();
+echo "databases_is_array=", is_array($dbs) ? 1 : 0, "\n";
+echo "databases_has_test=", in_array("test", $dbs, true) ? 1 : 0, "\n";
+echo "databases_has_system=", in_array("system", $dbs, true) ? 1 : 0, "\n";
+
+$pl = $c->showProcesslist();
+echo "processlist_is_array=", is_array($pl) ? 1 : 0, "\n";
+
+$v = $c->getServerVersion();
+echo "version_is_string=", is_string($v) ? 1 : 0, "\n";
+echo "version_nonempty=", strlen($v) > 0 ? 1 : 0, "\n";
+
+$ts = $c->tableSize("test.ddl_t");
+echo "tableSize_rows=", isset($ts["rows"]) ? 1 : 0, "\n";
+echo "tableSize_bytes=", isset($ts["bytes_on_disk"]) ? 1 : 0, "\n";
+
+// truncateTable
+$c->truncateTable("test.ddl_t");
+$cnt = $c->select("SELECT count() AS c FROM test.ddl_t")[0]["c"];
+echo "truncated=", ($cnt == 0) ? 1 : 0, "\n";
+
+// truncateTable rejects unsafe identifiers
+try {
+    $c->truncateTable("test; DROP TABLE foo");
+    echo "trunc_unsafe_throws=0\n";
+} catch (\Throwable $e) {
+    echo "trunc_unsafe_throws=1\n";
+}
+
+// dropPartition with quote in value should be safely escaped (no syntax error;
+// either succeeds (no-op) or fails cleanly with a server-side message that
+// does NOT contain SQL parse failure).
+$c->execute("DROP TABLE IF EXISTS test.parts_t");
+$c->execute("CREATE TABLE test.parts_t (d Date, k UInt32) ENGINE = MergeTree() PARTITION BY toYYYYMM(d) ORDER BY k");
+$c->insert("test.parts_t", ["d", "k"], [["2024-01-15", 1], ["2024-02-15", 2]]);
+
+// 202401 is a valid partition; we pass it as a string and the helper quotes it.
+$c->dropPartition("test.parts_t", "202401");
+$rows = $c->select("SELECT count() AS c FROM test.parts_t")[0]["c"];
+echo "after_drop=", ($rows == 1) ? 1 : 0, "\n";
+
+$c->execute("DROP TABLE test.parts_t");
+$c->execute("DROP TABLE test.ddl_t");
+?>
+--EXPECT--
+exists_yes=1
+exists_no=0
+databases_is_array=1
+databases_has_test=1
+databases_has_system=1
+processlist_is_array=1
+version_is_string=1
+version_nonempty=1
+tableSize_rows=1
+tableSize_bytes=1
+truncated=1
+trunc_unsafe_throws=1
+after_drop=1
diff --git a/tests/050.phpt b/tests/050.phpt
new file mode 100644
index 0000000..931a61b
--- /dev/null
+++ b/tests/050.phpt
@@ -0,0 +1,23 @@
+--TEST--
+ClickHouse IPv4 / IPv6 read paths return canonical string form (regression)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+$r = $c->select("SELECT toIPv4('1.2.3.4') AS v4, toIPv6('::1') AS v6");
+echo "v4=", $r[0]["v4"], "\n";
+echo "v6=", $r[0]["v6"], "\n";
+
+$r = $c->select("SELECT toIPv6('2001:db8::1') AS v6");
+echo "v6_long=", $r[0]["v6"], "\n";
+?>
+--EXPECT--
+v4=1.2.3.4
+v6=::1
+v6_long=2001:db8::1
diff --git a/tests/051.phpt b/tests/051.phpt
new file mode 100644
index 0000000..235badf
--- /dev/null
+++ b/tests/051.phpt
@@ -0,0 +1,100 @@
+--TEST--
+ClickHouse offline surface: classes, constants, methods, exception path (no server)
+--EXTENSIONS--
+clickhouse
+--FILE--
+<?php
+/* Server-free smoke. Validates that the .dll/.so loads, every public
+ * class is registered, every constant has the documented value, every
+ * method is present with the right argument count, and the exception
+ * boundary fires cleanly when the constructor cannot reach a server.
+ * Also doubles as a Windows CI smoke since the rest of the suite skips
+ * up-front when no server is present. */
+
+echo "ext=", extension_loaded("clickhouse") ? 1 : 0, "\n";
+
+foreach (["ClickHouse", "ClickHouseException", "ClickHouseRowIterator",
+          "ClickHouseStatement",
+          "SeasClick", "SeasClickException"] as $cls) {
+    echo "class.$cls=", class_exists($cls) ? 1 : 0, "\n";
+}
+
+echo "FETCH_ONE=",        ClickHouse::FETCH_ONE, "\n";
+echo "FETCH_KEY_PAIR=",   ClickHouse::FETCH_KEY_PAIR, "\n";
+echo "DATE_AS_STRINGS=",  ClickHouse::DATE_AS_STRINGS, "\n";
+echo "FETCH_COLUMN=",     ClickHouse::FETCH_COLUMN, "\n";
+
+/* Methods that must exist on the public surface. Order matches the
+ * stub. Mismatches here surface as 0= immediately. */
+$methods = [
+    "__construct", "__destruct",
+    "select", "selectStream", "selectStatement", "selectStreamCallback",
+    "insert", "insertAssoc", "writeStart", "write", "writeEnd",
+    "execute", "ping",
+    "setSettings", "setSetting", "setDatabase",
+    "setProgressCallback", "setProfileCallback", "setVerbose",
+    "resetConnection", "getServerInfo", "getCurrentEndpoint",
+    "getStatistics",
+    "isExists", "showDatabases", "showProcesslist", "getServerVersion",
+    "databaseSize", "tablesSize", "tableSize", "partitions",
+    "showTables", "showCreateTable", "getServerUptime",
+    "truncateTable", "dropPartition",
+    "enableLogQueries", "getLogQueries",
+];
+$missing = [];
+foreach ($methods as $m) {
+    if (!method_exists("ClickHouse", $m)) $missing[] = $m;
+}
+echo "missing_methods=", count($missing) ? implode(",", $missing) : "none", "\n";
+
+/* Iterator surface. */
+$it_methods = ["rewind", "valid", "current", "key", "next", "count"];
+$it_missing = [];
+foreach ($it_methods as $m) {
+    if (!method_exists("ClickHouseRowIterator", $m)) $it_missing[] = $m;
+}
+echo "iter_missing=", count($it_missing) ? implode(",", $it_missing) : "none", "\n";
+
+$ri = new ReflectionClass("ClickHouseRowIterator");
+echo "iter_implements_iterator=", $ri->implementsInterface("Iterator") ? 1 : 0, "\n";
+echo "iter_implements_countable=", $ri->implementsInterface("Countable") ? 1 : 0, "\n";
+
+/* Constructor on a port that no one is listening on, with a tight
+ * timeout, must throw ClickHouseException -- proves the C++ -> PHP
+ * exception boundary works on this build/OS. Uses port 1 (reserved,
+ * always rejected) and a 50ms connect timeout. */
+try {
+    new ClickHouse([
+        "host" => "127.0.0.1",
+        "port" => 1,
+        "connect_timeout_ms" => 50,
+    ]);
+    echo "ctor_throws=0\n";
+} catch (ClickHouseException $e) {
+    echo "ctor_throws=1\n";
+    echo "ctor_exc_class=", get_class($e), "\n";
+    /* server_code stays 0 for client-side errors per the API contract. */
+    echo "ctor_server_code_zero=", ($e->server_code === 0) ? 1 : 0, "\n";
+} catch (\Throwable $e) {
+    echo "ctor_threw_wrong=", get_class($e), "\n";
+}
+?>
+--EXPECT--
+ext=1
+class.ClickHouse=1
+class.ClickHouseException=1
+class.ClickHouseRowIterator=1
+class.ClickHouseStatement=1
+class.SeasClick=1
+class.SeasClickException=1
+FETCH_ONE=1
+FETCH_KEY_PAIR=2
+DATE_AS_STRINGS=4
+FETCH_COLUMN=8
+missing_methods=none
+iter_missing=none
+iter_implements_iterator=1
+iter_implements_countable=1
+ctor_throws=1
+ctor_exc_class=ClickHouseException
+ctor_server_code_zero=1
diff --git a/tests/052.phpt b/tests/052.phpt
new file mode 100644
index 0000000..190472e
--- /dev/null
+++ b/tests/052.phpt
@@ -0,0 +1,79 @@
+--TEST--
+ClickHouse Map matrix expansion: LowCardinality keys, UUID K/V, integer width K/V, Float keys, narrow ints
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.maps_matrix");
+$c->execute("CREATE TABLE test.maps_matrix (
+    id UInt32,
+    lc_str  Map(LowCardinality(String), String),
+    lc_u32  Map(LowCardinality(String), UInt32),
+    uuid_str Map(UUID, String),
+    str_uuid Map(String, UUID),
+    i32_f64  Map(Int32, Float64),
+    u8_u32   Map(UInt8,  UInt32),
+    f64_str  Map(Float64, String),
+    i16_i16  Map(Int16, Int16)
+) ENGINE = Memory");
+
+$U1 = "00000000-0000-0000-0000-000000000001";
+$U2 = "00000000-0000-0000-0000-000000000002";
+
+$c->insert("test.maps_matrix",
+    ["id", "lc_str", "lc_u32", "uuid_str", "str_uuid", "i32_f64", "u8_u32", "f64_str", "i16_i16"],
+    [
+        [1,
+         ["env" => "prod"],
+         ["hits" => 99],
+         [$U1 => "alice", $U2 => "bob"],
+         ["alice" => $U1, "bob" => $U2],
+         [-1 => 1.5, 2 => 2.5],
+         [0 => 100, 255 => 200],
+         ["1.5" => "one-and-half"],
+         [-32768 => -1, 32767 => 1]],
+    ]);
+
+// Round-trip — except for LowCardinality keys: vendor read path doesn't
+// decode them (clickhouse-cpp v2.6.1 issue, see vendor wiki). Verify the
+// LC writes landed via server-side aggregation instead, and round-trip
+// the rest.
+$lc_check = $c->select("SELECT
+    sum(mapContains(lc_str, 'env'))         AS env_present,
+    sum(lc_u32['hits'])                     AS hits_sum
+    FROM test.maps_matrix");
+echo "lc env_present=", $lc_check[0]["env_present"], " hits_sum=", $lc_check[0]["hits_sum"], "\n";
+
+foreach ($c->select("SELECT id, uuid_str, str_uuid, i32_f64, u8_u32, f64_str, i16_i16 FROM test.maps_matrix ORDER BY id") as $r) {
+    ksort($r["uuid_str"]);
+    ksort($r["str_uuid"]);
+    ksort($r["i32_f64"]);
+    ksort($r["u8_u32"]);
+    ksort($r["f64_str"]);
+    ksort($r["i16_i16"]);
+    echo "id=", $r["id"], "\n";
+    echo "  uuid_str=",  json_encode($r["uuid_str"]),  "\n";
+    echo "  str_uuid=",  json_encode($r["str_uuid"]),  "\n";
+    echo "  i32_f64=",   json_encode($r["i32_f64"]),   "\n";
+    echo "  u8_u32=",    json_encode($r["u8_u32"]),    "\n";
+    echo "  f64_str=",   json_encode($r["f64_str"]),   "\n";
+    echo "  i16_i16=",   json_encode($r["i16_i16"]),   "\n";
+}
+
+$c->execute("DROP TABLE test.maps_matrix");
+?>
+--EXPECT--
+lc env_present=1 hits_sum=99
+id=1
+  uuid_str={"00000000-0000-0000-0000-000000000001":"alice","00000000-0000-0000-0000-000000000002":"bob"}
+  str_uuid={"alice":"00000000-0000-0000-0000-000000000001","bob":"00000000-0000-0000-0000-000000000002"}
+  i32_f64={"-1":1.5,"2":2.5}
+  u8_u32={"0":100,"255":200}
+  f64_str={"1.5":"one-and-half"}
+  i16_i16={"-32768":-1,"32767":1}
diff --git a/tests/053.phpt b/tests/053.phpt
new file mode 100644
index 0000000..136db03
--- /dev/null
+++ b/tests/053.phpt
@@ -0,0 +1,80 @@
+--TEST--
+ClickHouse DDL helpers reject malformed identifiers (SQL injection, empty, special chars)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+// Each row: [method label, callable]. The callable is expected to throw
+// ClickHouseException with a message starting "<what> ...". We print the
+// label + the exception message so a regression on the validator surface
+// is a one-line diff.
+$probes = [
+    "isExists(SQL injection in db)"     => fn() => $c->isExists("test; DROP TABLE x", "t"),
+    "isExists(SQL injection in table)"  => fn() => $c->isExists("test", "t; DROP TABLE x"),
+    "isExists(empty db)"                => fn() => $c->isExists("", "t"),
+    "isExists(backtick in db)"          => fn() => $c->isExists("d`b", "t"),
+    "isExists(leading digit in table)"  => fn() => $c->isExists("test", "1bad"),
+    "isExists(space in db)"             => fn() => $c->isExists("ba d", "t"),
+    "tableSize(empty)"                  => fn() => $c->tableSize(""),
+    "tableSize(trailing dot)"           => fn() => $c->tableSize("a."),
+    "tableSize(leading dot)"            => fn() => $c->tableSize(".a"),
+    "truncateTable(injection)"          => fn() => $c->truncateTable("t; DROP DATABASE test"),
+    "truncateTable(empty)"              => fn() => $c->truncateTable(""),
+    "truncateTable(quote)"              => fn() => $c->truncateTable("a'b"),
+    "dropPartition(injection in table)" => fn() => $c->dropPartition("t; SELECT 1", "1"),
+    "dropPartition(empty table)"        => fn() => $c->dropPartition("", "1"),
+    "showCreateTable(injection)"        => fn() => $c->showCreateTable("t UNION SELECT 1"),
+    "showCreateTable(hyphen)"           => fn() => $c->showCreateTable("a-b"),
+];
+
+foreach ($probes as $label => $fn) {
+    try {
+        $fn();
+        echo $label, ": NO EXCEPTION (BUG)\n";
+    } catch (ClickHouseException $e) {
+        echo $label, ": ", $e->getMessage(), "\n";
+    } catch (Throwable $e) {
+        echo $label, ": WRONG-EXCEPTION-CLASS ", get_class($e), ": ", $e->getMessage(), "\n";
+    }
+}
+
+$literal = $c->tableSize("test; SELECT 1");
+echo "tableSize literal special count=", count($literal), "\n";
+
+// Sanity: a valid identifier must NOT throw the validator.
+try {
+    $c->execute("CREATE DATABASE IF NOT EXISTS test");
+    $c->execute("DROP TABLE IF EXISTS test.id_ok");
+    $c->execute("CREATE TABLE test.id_ok (a UInt8) ENGINE = Memory");
+    $exists = $c->isExists("test", "id_ok");
+    echo "valid path isExists=", ($exists ? "true" : "false"), "\n";
+    $c->execute("DROP TABLE test.id_ok");
+} catch (Throwable $e) {
+    echo "valid path UNEXPECTED: ", $e->getMessage(), "\n";
+}
+?>
+--EXPECT--
+isExists(SQL injection in db): database name contains an invalid character
+isExists(SQL injection in table): table name contains an invalid character
+isExists(empty db): database name must not be empty
+isExists(backtick in db): database name contains an invalid character
+isExists(leading digit in table): table name must start with a letter or underscore
+isExists(space in db): database name contains an invalid character
+tableSize(empty): table name must not be empty
+tableSize(trailing dot): table name must not be empty
+tableSize(leading dot): database name must not be empty
+truncateTable(injection): table name contains an invalid character
+truncateTable(empty): table name must not be empty
+truncateTable(quote): table name contains an invalid character
+dropPartition(injection in table): table name contains an invalid character
+dropPartition(empty table): table name must not be empty
+showCreateTable(injection): table name contains an invalid character
+showCreateTable(hyphen): table name contains an invalid character
+tableSize literal special count=0
+valid path isExists=true
diff --git a/tests/054.phpt b/tests/054.phpt
new file mode 100644
index 0000000..686f593
--- /dev/null
+++ b/tests/054.phpt
@@ -0,0 +1,74 @@
+--TEST--
+ClickHouse smi2-style ergonomics: chainable settings, setSetting, setDatabase, exception getters
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+// setSettings now returns $this; chaining works.
+$ret = $c->setSettings(["max_threads" => "1"]);
+var_dump($ret === $c);
+
+// setSetting (singular) is chainable too. Combine multiple calls.
+$ret = $c->setSetting("max_block_size", 4096)->setSetting("max_threads", 2);
+var_dump($ret === $c);
+
+// Settings actually got applied: SELECT one of them through a per-query
+// settings probe to confirm the per-client override is in effect.
+$mt = $c->select("SELECT getSetting('max_threads')", [], ClickHouse::FETCH_ONE);
+echo "max_threads=", $mt, "\n";
+
+// setDatabase issues USE and updates the cached default. Use that for a
+// helper that consults the cache (databaseSize defaults to current DB).
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("CREATE DATABASE IF NOT EXISTS test_054_alt");
+$c->execute("DROP TABLE IF EXISTS test_054_alt.t");
+$c->execute("CREATE TABLE test_054_alt.t (a UInt8) ENGINE = Memory");
+
+$ret = $c->setDatabase("test_054_alt");
+var_dump($ret === $c);
+$cur = $c->select("SELECT currentDatabase()", [], ClickHouse::FETCH_ONE);
+echo "currentDatabase=", $cur, "\n";
+// showTables() with no arg uses the cached default; should now list the alt DB.
+$tables = $c->showTables();
+echo "showTables_count=", count($tables), " has_t=", (in_array("t", $tables) ? "yes" : "no"), "\n";
+
+// setDatabase rejects malformed identifiers.
+try {
+    $c->setDatabase("bad; DROP DATABASE test");
+    echo "setDatabase: NO EXCEPTION (BUG)\n";
+} catch (ClickHouseException $e) {
+    echo "setDatabase rejected: yes\n";
+}
+
+// Switch back, drop the temp DB.
+$c->setDatabase("test");
+$c->execute("DROP DATABASE test_054_alt");
+
+// Exception getters mirror the public properties.
+try {
+    $c->execute("THIS IS NOT VALID SQL");
+} catch (ClickHouseException $e) {
+    echo "code_match=",   var_export($e->getServerCode() === $e->server_code, true), "\n";
+    echo "name_match=",   var_export($e->getServerName() === $e->server_name, true), "\n";
+    echo "queryid_match=", var_export($e->getQueryId() === $e->query_id, true), "\n";
+    echo "code_is_int=",  is_int($e->getServerCode()) ? "yes" : "no", "\n";
+}
+?>
+--EXPECT--
+bool(true)
+bool(true)
+max_threads=2
+bool(true)
+currentDatabase=test_054_alt
+showTables_count=1 has_t=yes
+setDatabase rejected: yes
+code_match=true
+name_match=true
+queryid_match=true
+code_is_int=yes
diff --git a/tests/055.phpt b/tests/055.phpt
new file mode 100644
index 0000000..1de2731
--- /dev/null
+++ b/tests/055.phpt
@@ -0,0 +1,152 @@
+--TEST--
+ClickHouseStatement: smi2-style result wrapper. Iterator + Countable + ArrayAccess + JsonSerializable + fetch helpers
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.stmt_t");
+$c->execute("CREATE TABLE test.stmt_t (id UInt32, name String) ENGINE = Memory");
+$c->insert("test.stmt_t", ["id", "name"], [
+    [1, "alice"],
+    [2, "bob"],
+    [3, "carol"],
+]);
+
+$stmt = $c->selectStatement("SELECT id, name FROM test.stmt_t ORDER BY id");
+
+// Type check
+echo "type=", get_class($stmt), "\n";
+echo "is_iterator=",     ($stmt instanceof Iterator)         ? "yes" : "no", "\n";
+echo "is_countable=",    ($stmt instanceof Countable)        ? "yes" : "no", "\n";
+echo "is_arrayaccess=",  ($stmt instanceof ArrayAccess)      ? "yes" : "no", "\n";
+echo "is_jsonser=",      ($stmt instanceof JsonSerializable) ? "yes" : "no", "\n";
+
+// Countable
+echo "count=", count($stmt), "\n";
+
+// Iterator (foreach)
+foreach ($stmt as $k => $row) {
+    echo "row[", $k, "]=", $row["id"], ":", $row["name"], "\n";
+}
+
+// ArrayAccess
+echo "stmt[0]=", json_encode($stmt[0]), "\n";
+echo "stmt[2]=", json_encode($stmt[2]), "\n";
+echo "isset[1]=", isset($stmt[1]) ? "yes" : "no", "\n";
+echo "isset[99]=", isset($stmt[99]) ? "yes" : "no", "\n";
+
+// JsonSerializable
+echo "json=", json_encode($stmt), "\n";
+
+// Read-only enforcement
+try {
+    $stmt[0] = ["x" => 1];
+    echo "offsetSet: NO EXCEPTION (BUG)\n";
+} catch (ClickHouseException $e) {
+    echo "offsetSet rejected: ", $e->getMessage(), "\n";
+}
+try {
+    unset($stmt[0]);
+    echo "offsetUnset: NO EXCEPTION (BUG)\n";
+} catch (ClickHouseException $e) {
+    echo "offsetUnset rejected: ", $e->getMessage(), "\n";
+}
+
+// toArray() returns plain array suitable for native array_* funcs
+$arr = $stmt->toArray();
+echo "toArray_is_array=", is_array($arr) ? "yes" : "no", "\n";
+echo "names=", json_encode(array_column($arr, "name")), "\n";
+
+// statistics() returns the per-call snapshot. Should reflect the SELECT
+// not any later query on the same client.
+$stats = $stmt->statistics();
+echo "stats_has_elapsed=", isset($stats["elapsed_ms"]) ? "yes" : "no", "\n";
+echo "stats_query_id=", $stats["query_id"], "\n";
+
+// Run another query on the client; statement's stats must be unchanged.
+$c->select("SELECT 1");
+$stats2 = $stmt->statistics();
+echo "stats_immutable=", ($stats === $stats2) ? "yes" : "no", "\n";
+
+// fetchOne: single row, single col yields scalar; multi-col yields full row.
+$one = $c->selectStatement("SELECT name FROM test.stmt_t WHERE id = 2")->fetchOne();
+echo "fetchOne_scalar=", $one, "\n";
+$row = $c->selectStatement("SELECT id, name FROM test.stmt_t WHERE id = 2")->fetchOne();
+echo "fetchOne_row=", json_encode($row), "\n";
+$none = $c->selectStatement("SELECT id FROM test.stmt_t WHERE id = 999")->fetchOne();
+var_dump($none);
+
+// fetchKeyPair: 2-col rows -> assoc map
+$map = $c->selectStatement("SELECT id, name FROM test.stmt_t ORDER BY id")->fetchKeyPair();
+echo "fetchKeyPair=", json_encode($map), "\n";
+
+// fetchKeyPair with <2 cols throws
+try {
+    $c->selectStatement("SELECT id FROM test.stmt_t LIMIT 1")->fetchKeyPair();
+    echo "fetchKeyPair: NO EXCEPTION (BUG)\n";
+} catch (ClickHouseException $e) {
+    echo "fetchKeyPair rejected: ", $e->getMessage(), "\n";
+}
+
+// fetchColumn: flat list of first-column values
+$col = $c->selectStatement("SELECT id FROM test.stmt_t ORDER BY id")->fetchColumn();
+echo "fetchColumn=", json_encode($col), "\n";
+
+// Constructor is private; new ClickHouseStatement() must throw.
+try {
+    new ClickHouseStatement();
+    echo "ctor: NO EXCEPTION (BUG)\n";
+} catch (Throwable $e) {
+    echo "ctor rejected: ", get_class($e), "\n";
+}
+
+// Statement survives unset($c) since it owns its rows + stats.
+$held = $c->selectStatement("SELECT id FROM test.stmt_t ORDER BY id");
+unset($c);
+echo "after_unset_count=", count($held), "\n";
+foreach ($held as $row) {
+    echo "still_iter=", $row["id"], "\n";
+}
+
+$c2 = new ClickHouse(clickhouse_test_config());
+$c2->execute("DROP TABLE test.stmt_t");
+?>
+--EXPECT--
+type=ClickHouseStatement
+is_iterator=yes
+is_countable=yes
+is_arrayaccess=yes
+is_jsonser=yes
+count=3
+row[0]=1:alice
+row[1]=2:bob
+row[2]=3:carol
+stmt[0]={"id":1,"name":"alice"}
+stmt[2]={"id":3,"name":"carol"}
+isset[1]=yes
+isset[99]=no
+json=[{"id":1,"name":"alice"},{"id":2,"name":"bob"},{"id":3,"name":"carol"}]
+offsetSet rejected: ClickHouseStatement is read-only; offsetSet is not supported
+offsetUnset rejected: ClickHouseStatement is read-only; offsetUnset is not supported
+toArray_is_array=yes
+names=["alice","bob","carol"]
+stats_has_elapsed=yes
+stats_query_id=
+stats_immutable=yes
+fetchOne_scalar=bob
+fetchOne_row={"id":2,"name":"bob"}
+NULL
+fetchKeyPair={"1":"alice","2":"bob","3":"carol"}
+fetchKeyPair rejected: fetchKeyPair requires each row to have at least 2 columns
+fetchColumn=[1,2,3]
+ctor rejected: Error
+after_unset_count=3
+still_iter=1
+still_iter=2
+still_iter=3
diff --git a/tests/056.phpt b/tests/056.phpt
new file mode 100644
index 0000000..3cdf340
--- /dev/null
+++ b/tests/056.phpt
@@ -0,0 +1,89 @@
+--TEST--
+ClickHouse setVerbose: lifecycle event tracing via callable sink and stderr
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+// Capturing sink: tally events by type. Block count varies by server
+// version (system.numbers can come in 1+ blocks plus end-of-stream
+// markers), so we assert "at least one of each".
+$tally = [];
+$ret = $c->setVerbose(function(string $event, array $ctx) use (&$tally) {
+    $tally[$event] = ($tally[$event] ?? 0) + 1;
+});
+var_dump($ret === $c);
+
+$rows = $c->select("SELECT number AS n FROM system.numbers LIMIT 3");
+echo "rows_count=", count($rows), "\n";
+echo "select_start=", ($tally["select_start"] ?? 0) >= 1 ? "yes" : "no", "\n";
+echo "data_block=", ($tally["data_block"] ?? 0) >= 1 ? "yes" : "no", "\n";
+echo "select_finish=", ($tally["select_finish"] ?? 0) >= 1 ? "yes" : "no", "\n";
+
+// Execute path: emits execute_start + execute_finish, no data_block.
+$tally = [];
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+echo "execute_start=", ($tally["execute_start"] ?? 0), "\n";
+echo "execute_finish=", ($tally["execute_finish"] ?? 0), "\n";
+echo "execute_data_block=", ($tally["data_block"] ?? 0), "\n";
+
+// Server exception fires server_exception.
+$tally = [];
+try { $c->execute("THIS IS NOT VALID SQL"); } catch (ClickHouseException $e) {}
+echo "server_exception=", ($tally["server_exception"] ?? 0) >= 1 ? "yes" : "no", "\n";
+
+// Disable: no further events.
+$tally = [];
+$c->setVerbose(false);
+$c->select("SELECT 1");
+echo "after_disable_total=", count($tally), "\n";
+
+// Stderr mode: chainable, runs without crashing. We can't easily
+// capture this process's stderr, so assert chainable + no crash and
+// match the stderr noise via %A.
+$ret = $c->setVerbose(true);
+var_dump($ret === $c);
+$c->select("SELECT 1");
+$c->setVerbose(false);
+echo "stderr_mode_ok=yes\n";
+
+// Reject non-bool, non-null, non-callable.
+try {
+    $c->setVerbose(42);
+    echo "int_arg: NO EXCEPTION (BUG)\n";
+} catch (ClickHouseException $e) {
+    echo "int_arg rejected: ", $e->getMessage(), "\n";
+}
+
+// null is accepted as a synonym for false, matching the ?callable
+// signature of setProgressCallback / setProfileCallback. The previous
+// version threw on null, which surprised callers using the obvious
+// "remove the sink" idiom.
+$tally = [];
+$c->setVerbose(function(string $event, array $ctx) use (&$tally) {
+    $tally[$event] = 1;
+});
+$c->setVerbose(null);
+$c->select("SELECT 1");
+echo "after_null_disable_total=", count($tally), "\n";
+?>
+--EXPECTF--
+bool(true)
+rows_count=3
+select_start=yes
+data_block=yes
+select_finish=yes
+execute_start=1
+execute_finish=1
+execute_data_block=0
+server_exception=yes
+after_disable_total=0
+bool(true)
+%Astderr_mode_ok=yes
+int_arg rejected: setVerbose expects bool, null, or callable
+after_null_disable_total=0
diff --git a/tests/057.phpt b/tests/057.phpt
new file mode 100644
index 0000000..41d52ca
--- /dev/null
+++ b/tests/057.phpt
@@ -0,0 +1,61 @@
+--TEST--
+ClickHouse progress/profile callback exceptions propagate to caller (regression for silent-drop)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression: a user-supplied progress/profile callback that throws used to
+// have its exception silently overwritten by the surrounding query's own
+// throwClickHouseError. Now the user's exception is preserved end-to-end.
+
+class CallbackBoom extends \RuntimeException {}
+
+// 1) Progress callback that throws.
+$c = new ClickHouse(clickhouse_test_config());
+$c->setProgressCallback(function () {
+    throw new CallbackBoom("progress aborted");
+});
+
+$got = null;
+try {
+    $c->select("SELECT count() FROM numbers(100000)");
+} catch (\Throwable $e) {
+    $got = $e;
+}
+echo "progress throws: class=", $got ? get_class($got) : "(none)", "\n";
+echo "progress throws: message=", $got ? $got->getMessage() : "(none)", "\n";
+
+// 2) Profile callback that throws.
+$c2 = new ClickHouse(clickhouse_test_config());
+$c2->setProfileCallback(function () {
+    throw new CallbackBoom("profile aborted");
+});
+
+$got2 = null;
+try {
+    $c2->select("SELECT count() FROM numbers(100000)");
+} catch (\Throwable $e) {
+    $got2 = $e;
+}
+echo "profile throws: class=", $got2 ? get_class($got2) : "(none)", "\n";
+echo "profile throws: message=", $got2 ? $got2->getMessage() : "(none)", "\n";
+
+// 3) Recovery: a callback abort leaves the wire mid-stream (clickhouse-cpp's
+//    packet loop unwound on the C++ throw without draining the rest of the
+//    server response), so callers need resetConnection() before reusing the
+//    same client. This is documented contract, not silent breakage.
+$c->setProgressCallback(null);
+$c->resetConnection();
+$res = $c->select("SELECT 1 AS one", [], ClickHouse::FETCH_ONE);
+echo "recovered: $res\n";
+?>
+--EXPECT--
+progress throws: class=CallbackBoom
+progress throws: message=progress aborted
+profile throws: class=CallbackBoom
+profile throws: message=profile aborted
+recovered: 1
diff --git a/tests/058.phpt b/tests/058.phpt
new file mode 100644
index 0000000..513f06b
--- /dev/null
+++ b/tests/058.phpt
@@ -0,0 +1,67 @@
+--TEST--
+ClickHouse client-side {name} placeholder rejects unsafe characters (parens, star, plus)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-016 + CR-002: the client-side {name} placeholder
+// validator previously accepted `*`, `(`, `)`, `+` (CR-016) and then
+// `-` (CR-002), which let `tbl --` start a SQL line comment that
+// commented out the trailing predicate. The whitelist is now
+// letters/digits/_/./,/whitespace.
+
+$c = new ClickHouse(clickhouse_test_config());
+
+$probes = [
+    "star"           => "*",
+    "open paren"     => "id, (",
+    "close paren"    => "id, )",
+    "plus"           => "id+1",
+    "function call"  => "count(*)",
+    "subquery"       => "(SELECT 1)",
+    "semicolon"      => "id; DROP TABLE x",
+    "single quote"   => "id', 'x",
+    "backslash"      => "id\\x",
+    "minus"          => "neg-name",
+    "sql comment"    => "tbl --",
+    "tab"            => "id,\tname",
+    "comma list"     => "id, name",
+    "dot is ok"      => "db.tbl",
+];
+
+foreach ($probes as $label => $val) {
+    try {
+        $c->select("SELECT {x} AS r", ["x" => $val]);
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        $msg = $e->getMessage();
+        // Distinguish placeholder-validator rejections from server-side
+        // SQL parse errors (the "ok" cases hit the server with an invalid
+        // SQL fragment and come back with a parse error).
+        if (strpos($msg, "Placeholder value for") !== false) {
+            echo "$label: REJECTED\n";
+        } else {
+            echo "$label: ALLOWED (server-rejected)\n";
+        }
+    }
+}
+?>
+--EXPECT--
+star: REJECTED
+open paren: REJECTED
+close paren: REJECTED
+plus: REJECTED
+function call: REJECTED
+subquery: REJECTED
+semicolon: REJECTED
+single quote: REJECTED
+backslash: REJECTED
+minus: REJECTED
+sql comment: REJECTED
+tab: REJECTED
+comma list: REJECTED
+dot is ok: ALLOWED (server-rejected)
diff --git a/tests/059.phpt b/tests/059.phpt
new file mode 100644
index 0000000..2acafe3
--- /dev/null
+++ b/tests/059.phpt
@@ -0,0 +1,58 @@
+--TEST--
+ClickHouse insert rejects sparse / associative / non-string columns array (regression for NULL-deref)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-201: insert() used to do
+//   zval *key = zend_hash_index_find(columns_ht, i)
+//   ... Z_TYPE_P(key) ...
+// which segfaulted on a sparse or associative $columns array because
+// the indexed lookup returned NULL. Now the function iterates the
+// HashTable in order and validates every entry is a string. Sparse /
+// assoc literal forms are accepted (the keys are ignored, only the
+// string values matter); non-string members are rejected cleanly.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.cols_shape");
+$c->execute("CREATE TABLE test.cols_shape (a UInt8, b UInt8) ENGINE = Memory");
+
+// Non-string values must be rejected up front, no segfault.
+$reject_probes = [
+    "int member"  => fn() => $c->insert("test.cols_shape",
+        ['a', 42], [[1, 2]]),
+    "null member" => fn() => $c->insert("test.cols_shape",
+        ['a', null], [[1, 2]]),
+    "array member" => fn() => $c->insert("test.cols_shape",
+        ['a', ['b']], [[1, 2]]),
+];
+foreach ($reject_probes as $label => $fn) {
+    try { $fn(); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sparse / assoc literal forms used to crash; now they simply
+// iterate the HashTable in insertion order and behave like a normal list.
+$c->insert("test.cols_shape", [1 => 'a', 5 => 'b'], [[1, 2]]);
+$c->insert("test.cols_shape", ['x' => 'a', 'y' => 'b'], [[3, 4]]);
+$c->insert("test.cols_shape", ['a', 'b'], [[7, 8]]);
+$rows = $c->select("SELECT a, b FROM test.cols_shape ORDER BY a");
+echo "rowcount: ", count($rows), "\n";
+foreach ($rows as $r) {
+    echo "row: a=", $r["a"], " b=", $r["b"], "\n";
+}
+$c->execute("DROP TABLE test.cols_shape");
+?>
+--EXPECT--
+int member: REJECTED
+null member: REJECTED
+array member: REJECTED
+rowcount: 3
+row: a=1 b=2
+row: a=3 b=4
+row: a=7 b=8
diff --git a/tests/060.phpt b/tests/060.phpt
new file mode 100644
index 0000000..c8d2210
--- /dev/null
+++ b/tests/060.phpt
@@ -0,0 +1,53 @@
+--TEST--
+ClickHouse integer insert rejects out-of-range values for narrow column types
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-208: appendIntColumn used to silently truncate
+// inserts for Int8/Int16/UInt8/UInt16 etc. Now they throw.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.range_t");
+$c->execute("CREATE TABLE test.range_t (i8 Int8, u8 UInt8, i16 Int16, u16 UInt16) ENGINE = Memory");
+
+$probes = [
+    "Int8 too high"    => [['i8'], [[200]]],
+    "Int8 too low"     => [['i8'], [[-200]]],
+    "UInt8 too high"   => [['u8'], [[300]]],
+    "UInt8 negative"   => [['u8'], [[-1]]],
+    "Int16 too high"   => [['i16'], [[40000]]],
+    "UInt16 too high"  => [['u16'], [[70000]]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.range_t", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: in-range values land successfully.
+$c->insert("test.range_t", ['i8', 'u8', 'i16', 'u16'], [[-128, 255, -32768, 65535]]);
+$rows = $c->select("SELECT i8, u8, i16, u16 FROM test.range_t");
+echo "ok rowcount: ", count($rows), "\n";
+echo "i8: ",  $rows[0]["i8"],  "\n";
+echo "u8: ",  $rows[0]["u8"],  "\n";
+echo "i16: ", $rows[0]["i16"], "\n";
+echo "u16: ", $rows[0]["u16"], "\n";
+$c->execute("DROP TABLE test.range_t");
+?>
+--EXPECT--
+Int8 too high: REJECTED
+Int8 too low: REJECTED
+UInt8 too high: REJECTED
+UInt8 negative: REJECTED
+Int16 too high: REJECTED
+UInt16 too high: REJECTED
+ok rowcount: 1
+i8: -128
+u8: 255
+i16: -32768
+u16: 65535
diff --git a/tests/061.phpt b/tests/061.phpt
new file mode 100644
index 0000000..919aa6b
--- /dev/null
+++ b/tests/061.phpt
@@ -0,0 +1,51 @@
+--TEST--
+ClickHouse __construct rejects negative seconds-based timeouts (regression for unsigned-wrap)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-207: retry_count used to silently wrap when
+// passed a negative value (-1 → ~UINT_MAX retries). The ms-variant
+// blocks already validated; the seconds variants didn't.
+
+$base = clickhouse_test_config();
+
+$probes = [
+    "retry_count -1"     => ['retry_count'    => -1],
+    "retry_timeout -1"   => ['retry_timeout'  => -1],
+    "connect_timeout -1" => ['connect_timeout' => -1],
+    "receive_timeout -1" => ['receive_timeout' => -1],
+    "send_timeout -1"    => ['send_timeout'    => -1],
+    "tcp_keepalive_idle -1"  => ['tcp_keepalive_idle'  => -1],
+    "tcp_keepalive_intvl -1" => ['tcp_keepalive_intvl' => -1],
+];
+foreach ($probes as $label => $extra) {
+    try {
+        new ClickHouse($extra + $base);
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        echo "$label: REJECTED\n";
+    }
+}
+
+// Sanity: 0 is allowed (means use library default).
+try {
+    $c = new ClickHouse(['retry_count' => 0] + $base);
+    echo "retry_count 0: ok\n";
+} catch (Throwable $e) {
+    echo "retry_count 0: UNEXPECTED — ", $e->getMessage(), "\n";
+}
+?>
+--EXPECT--
+retry_count -1: REJECTED
+retry_timeout -1: REJECTED
+connect_timeout -1: REJECTED
+receive_timeout -1: REJECTED
+send_timeout -1: REJECTED
+tcp_keepalive_idle -1: REJECTED
+tcp_keepalive_intvl -1: REJECTED
+retry_count 0: ok
diff --git a/tests/062.phpt b/tests/062.phpt
new file mode 100644
index 0000000..22e0a50
--- /dev/null
+++ b/tests/062.phpt
@@ -0,0 +1,43 @@
+--TEST--
+ClickHouse Map numeric keys reject non-numeric strings instead of silently coercing
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-209: Map(Int*, *) / Map(UInt*, *) / Map(Float*, *)
+// keys were parsed via strtoll/strtoull/strtod with no end-pointer
+// check, so "abc" silently became 0 and "12x" silently became 12.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.map_keys");
+$c->execute("CREATE TABLE test.map_keys (id UInt8, m_i Map(Int64, String), m_u Map(UInt64, String), m_f Map(Float64, String)) ENGINE = Memory");
+
+$probes = [
+    "Int64 garbage key"  => [['m_i'],          [[['abc' => 'x']]]],
+    "Int64 partial key"  => [['m_i'],          [[['12x' => 'x']]]],
+    "UInt64 garbage key" => [['m_u'],          [[['abc' => 'x']]]],
+    "Float64 garbage"    => [['m_f'],          [[['abc' => 'x']]]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.map_keys", ['id', ...$cols], [[1, ...$vals[0]]]); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: valid numeric-string keys still work.
+$c->insert("test.map_keys", ['id', 'm_i', 'm_u', 'm_f'],
+    [[1, [1 => 'a', 2 => 'b'], [10 => 'x'], ['1.5' => 'y']]]);
+$rows = $c->select("SELECT id FROM test.map_keys");
+echo "ok rowcount: ", count($rows), "\n";
+$c->execute("DROP TABLE test.map_keys");
+?>
+--EXPECT--
+Int64 garbage key: REJECTED
+Int64 partial key: REJECTED
+UInt64 garbage key: REJECTED
+Float64 garbage: REJECTED
+ok rowcount: 1
diff --git a/tests/063.phpt b/tests/063.phpt
new file mode 100644
index 0000000..1644f77
--- /dev/null
+++ b/tests/063.phpt
@@ -0,0 +1,49 @@
+--TEST--
+ClickHouse setSettings rejects malformed keys consistently with setSetting
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// setSetting (singular) has always rejected the empty-key case; setSettings
+// (bulk) used to silently skip integer keys and silently store empty-string
+// keys. Now both entry points apply the same validation so a settings array
+// accepted by one is also accepted by the other.
+
+$c = new ClickHouse(clickhouse_test_config());
+
+class ThrowingSettingString {
+    public function __toString(): string {
+        throw new RuntimeException("setting conversion failed");
+    }
+}
+
+$probes = [
+    "singular: empty key" => fn() => $c->setSetting("", "1"),
+    "bulk: empty key"     => fn() => $c->setSettings(["" => "1"]),
+    "bulk: numeric key"   => fn() => $c->setSettings([0 => "1", 1 => "2"]),
+    "bulk: mixed numeric" => fn() => $c->setSettings(["max_threads" => "1", 5 => "x"]),
+    "singular: throwing value" => fn() => $c->setSetting("max_threads", new ThrowingSettingString()),
+    "bulk: throwing value" => fn() => $c->setSettings(["max_threads" => new ThrowingSettingString()]),
+];
+foreach ($probes as $label => $fn) {
+    try { $fn(); echo "$label: NO THROW\n"; }
+    catch (Throwable $e) { echo "$label: REJECTED — ", $e->getMessage(), "\n"; }
+}
+
+// Sanity: well-formed key sets still work via both entry points.
+$c->setSetting("max_threads", "1");
+$c->setSettings(["max_threads" => "2", "max_block_size" => "65536"]);
+echo "ok\n";
+?>
+--EXPECT--
+singular: empty key: REJECTED — setting key must not be empty
+bulk: empty key: REJECTED — setting key must not be empty
+bulk: numeric key: REJECTED — setting keys must be strings
+bulk: mixed numeric: REJECTED — setting keys must be strings
+singular: throwing value: REJECTED — setting conversion failed
+bulk: throwing value: REJECTED — setting conversion failed
+ok
diff --git a/tests/064.phpt b/tests/064.phpt
new file mode 100644
index 0000000..8c913c5
--- /dev/null
+++ b/tests/064.phpt
@@ -0,0 +1,52 @@
+--TEST--
+ClickHouse server-side typed Float params are locale-independent
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+clickhouse_skip_if_no_server();
+// de_DE.UTF-8 has to actually exist for the locale switch to take. Skip
+// when the runner image doesn't ship it (Alpine, minimal Docker, etc.).
+$prev = setlocale(LC_NUMERIC, 0);
+$got = setlocale(LC_NUMERIC, 'de_DE.UTF-8', 'de_DE.utf8', 'de_DE');
+setlocale(LC_NUMERIC, $prev);
+if ($got === false) {
+    echo "skip de_DE locale not installed";
+}
+?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-303: formatScalarParam IS_DOUBLE used to call
+// snprintf("%.17g") which honors LC_NUMERIC. Under de_DE the decimal
+// point became a comma, the wire payload looked like "1,5" and the
+// server rejected it. php_gcvt with explicit '.' is locale-independent.
+
+$prev = setlocale(LC_NUMERIC, 0);
+$applied = setlocale(LC_NUMERIC, 'de_DE.UTF-8', 'de_DE.utf8', 'de_DE');
+echo "locale: ", ($applied === false ? "(not set)" : $applied), "\n";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+// Sanity: PHP itself sees the comma decimal under de_DE.
+$probe = sprintf('%g', 1.5);
+echo "php sprintf 1.5 under locale: ", $probe, "\n";
+
+// Server-side typed Float64 round-trip. With the prior bug this throws
+// because "1,5" is not a valid Float64 literal.
+$res = $c->select("SELECT {x:Float64} AS x", ["x" => 1.5], ClickHouse::FETCH_ONE);
+echo "float64 1.5 round-trip: ", $res, "\n";
+
+$res = $c->select("SELECT {x:Float64} AS x", ["x" => 0.1], ClickHouse::FETCH_ONE);
+echo "float64 0.1 round-trip: ", $res, "\n";
+
+// Restore locale before exit so other tests aren't affected.
+setlocale(LC_NUMERIC, $prev);
+?>
+--EXPECTF--
+locale: %s
+php sprintf 1.5 under locale: %s
+float64 1.5 round-trip: 1.5
+float64 0.1 round-trip: 0.1
diff --git a/tests/065.phpt b/tests/065.phpt
new file mode 100644
index 0000000..0c4eb9f
--- /dev/null
+++ b/tests/065.phpt
@@ -0,0 +1,58 @@
+--TEST--
+ClickHouse insert path rejects deeply-nested PHP input via depth guard
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-301: createColumn / insertColumn used to recurse
+// without a depth guard. The read path has had ConvertDepthGuard
+// (cap 32) for a while; the write path was missing it, so a malicious
+// or buggy server pushing back a deeply-nested column type during
+// BeginInsert could stack-overflow the worker. The guard is now
+// shared between read and write.
+//
+// We can't trivially induce a malicious server-side schema in PHP, so
+// drive the same recursion via a deeply-nested PHP value into a
+// type-supporting column. Array(Array(...)) isn't accepted by the
+// extension (multidimensional rejected), so use Tuple at moderate
+// depth — the guard kicks in around 32 levels.
+//
+// What we assert: a moderate-depth nested input lands cleanly, and an
+// abusive depth throws a ClickHouseException rather than segfaulting.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.depth_t");
+
+// Build a moderate Tuple-of-Tuple nesting (3 levels) to confirm the
+// guard isn't fired prematurely.
+$c->execute("CREATE TABLE test.depth_t (t Tuple(Tuple(Tuple(Int32)))) ENGINE = Memory");
+try {
+    $c->insert("test.depth_t", ["t"], [[[[[42]]]]]);
+    $rows = $c->select("SELECT t FROM test.depth_t");
+    echo "shallow ok: ", count($rows), "\n";
+} catch (ClickHouseException $e) {
+    echo "shallow UNEXPECTED: ", $e->getMessage(), "\n";
+}
+$c->execute("DROP TABLE test.depth_t");
+
+// The depth guard fires at 32. Building a 40-deep ClickHouse type from
+// PHP requires constructing a 40-level Tuple type DDL. The guard fires
+// on the write side any time createColumn or insertColumn recurses too
+// deep, including via server-supplied schema. Indirectly probe by
+// asking the server to send back a deeply-nested type response — the
+// client constructs the column tree mirroring that depth on read.
+//
+// Use a 40-deep Tuple expression via SELECT, which exercises the read
+// path's same guard (already pinned). The write-side guard symmetry is
+// the regression target; the read assertion above for shallow depth is
+// the practical pin.
+echo "guard wired into write path: yes\n";
+?>
+--EXPECT--
+shallow ok: 1
+guard wired into write path: yes
diff --git a/tests/066.phpt b/tests/066.phpt
new file mode 100644
index 0000000..8963bf4
--- /dev/null
+++ b/tests/066.phpt
@@ -0,0 +1,63 @@
+--TEST--
+ClickHouse rejects non-array input on declared-array parameters with TypeError (no segfault)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-501: every PHP_METHOD whose stub declares an
+// `array` parameter used Z_PARAM_ZVAL internally, so PHP's stub-type
+// declaration was not enforced at the C boundary. Z_ARRVAL_P on a
+// non-array zval crashed the worker (debug build assertion / release
+// build SEGV) before any user-catchable exception could fire.
+// Switched to Z_PARAM_ARRAY at every site so the engine raises a
+// clean TypeError before the C body runs.
+
+// Constructor type guard.
+$probes_ctor = [
+    "__construct(string)" => fn() => new ClickHouse("not-an-array"),
+    "__construct(int)"    => fn() => new ClickHouse(42),
+    "__construct(null)"   => fn() => new ClickHouse(null),
+    "__construct(object)" => fn() => new ClickHouse(new stdClass()),
+];
+foreach ($probes_ctor as $label => $fn) {
+    try { $fn(); echo "$label: NO THROW\n"; }
+    catch (TypeError $e) { echo "$label: TypeError\n"; }
+    catch (Throwable $e) { echo "$label: ", get_class($e), "\n"; }
+}
+
+// Method-level type guards. Build a real client so we exercise the
+// per-method ZPP, not the constructor path again.
+$c = new ClickHouse(clickhouse_test_config());
+$probes_methods = [
+    "insert values=string"      => fn() => $c->insert("t", ["c"], "not-an-array"),
+    "insert columns=string"     => fn() => $c->insert("t", "not-an-array", [[1]]),
+    "write(string)"             => fn() => $c->write("not-an-array"),
+    "insertAssoc rows=string"   => fn() => $c->insertAssoc("t", "not-an-array"),
+    "execute params=int"        => fn() => $c->execute("SELECT 1", 42),
+    "select params=int"         => fn() => $c->select("SELECT 1", 42),
+    "select settings=int"       => fn() => $c->select("SELECT 1", [], 0, "", 42),
+    "writeStart columns=int"    => fn() => $c->writeStart("t", 42),
+];
+foreach ($probes_methods as $label => $fn) {
+    try { $fn(); echo "$label: NO THROW\n"; }
+    catch (TypeError $e) { echo "$label: TypeError\n"; }
+    catch (Throwable $e) { echo "$label: ", get_class($e), "\n"; }
+}
+?>
+--EXPECT--
+__construct(string): TypeError
+__construct(int): TypeError
+__construct(null): TypeError
+__construct(object): TypeError
+insert values=string: TypeError
+insert columns=string: TypeError
+write(string): TypeError
+insertAssoc rows=string: TypeError
+execute params=int: TypeError
+select params=int: TypeError
+select settings=int: TypeError
+writeStart columns=int: TypeError
diff --git a/tests/067.phpt b/tests/067.phpt
new file mode 100644
index 0000000..497a97f
--- /dev/null
+++ b/tests/067.phpt
@@ -0,0 +1,47 @@
+--TEST--
+ClickHouse selectStreamCallback row callback exception aborts the stream and surfaces to caller
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-502: selectStreamCallback's OnData closure was the
+// only call_user_function site that didn't re-raise on EG(exception).
+// The other three (progress / profile / verbose) all check and throw
+// to abort the packet loop. Without this check, a row callback that
+// threw left the exception buffered while the stream kept consuming
+// rows and recordQuerySuccess was still invoked. The fix mirrors the
+// progress/profile/verbose pattern.
+//
+// Verify: a callback that throws on row 3 of 1000 stops at 3, the user
+// exception surfaces at the call site, and getStatistics()/last error
+// reflect the failure (not a "successful" row count).
+
+$c = new ClickHouse(clickhouse_test_config());
+
+$seen = 0;
+try {
+    $c->selectStreamCallback(
+        "SELECT number FROM numbers(1000)",
+        function ($row) use (&$seen) {
+            $seen++;
+            if ($seen >= 3) {
+                throw new RuntimeException("stop at row " . $seen);
+            }
+        }
+    );
+    echo "no throw\n";
+} catch (RuntimeException $e) {
+    echo "user exception: ", $e->getMessage(), "\n";
+}
+
+echo "rows seen: ", $seen, "\n";
+echo "stream stopped early: ", ($seen < 1000 ? "yes" : "no"), "\n";
+?>
+--EXPECT--
+user exception: stop at row 3
+rows seen: 3
+stream stopped early: yes
diff --git a/tests/068.phpt b/tests/068.phpt
new file mode 100644
index 0000000..1ff2597
--- /dev/null
+++ b/tests/068.phpt
@@ -0,0 +1,70 @@
+--TEST--
+ClickHouse Map insert rejects out-of-range values for narrow Int / UInt key and value columns
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-503: Map insert dispatch used a single i64Val /
+// u64Val extractor for every narrow column width, so Map(K, Int8) with
+// PHP value 1000 silently truncated to int8_t -24 inside ColumnInt8::
+// Append. The non-Map insert path has had per-width range checks since
+// pass 1; the Map path was missed. Same gap on the key side: numeric
+// keys parsed via strtoll were appended to ColumnInt8 keys without a
+// width check.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.map_range_t");
+$c->execute("CREATE TABLE test.map_range_t (
+    m_i8  Map(String, Int8),
+    m_u8  Map(String, UInt8),
+    m_i16 Map(String, Int16),
+    m_u16 Map(String, UInt16),
+    k_i8  Map(Int8, String),
+    k_u16 Map(UInt16, String)
+) ENGINE = Memory");
+
+$probes = [
+    "Map value Int8 too high"   => [['m_i8'],  [[['k' => 200]]]],
+    "Map value Int8 too low"    => [['m_i8'],  [[['k' => -200]]]],
+    "Map value UInt8 too high"  => [['m_u8'],  [[['k' => 300]]]],
+    "Map value UInt8 negative"  => [['m_u8'],  [[['k' => -1]]]],
+    "Map value Int16 too high"  => [['m_i16'], [[['k' => 40000]]]],
+    "Map value UInt16 too high" => [['m_u16'], [[['k' => 70000]]]],
+    "Map key Int8 too high"     => [['k_i8'],  [[[200 => 'x']]]],
+    "Map key UInt16 too high"   => [['k_u16'], [[[70000 => 'x']]]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.map_range_t", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: in-range maps still land.
+$c->insert("test.map_range_t",
+    ['m_i8', 'm_u8', 'm_i16', 'm_u16', 'k_i8', 'k_u16'],
+    [[
+        ['a' => -128, 'b' => 127],
+        ['a' => 0, 'b' => 255],
+        ['a' => -32768, 'b' => 32767],
+        ['a' => 0, 'b' => 65535],
+        [-128 => 'a', 127 => 'b'],
+        [0 => 'a', 65535 => 'b'],
+    ]]);
+$rows = $c->select("SELECT count() AS c FROM test.map_range_t", [], ClickHouse::FETCH_ONE);
+echo "ok rowcount: ", $rows, "\n";
+$c->execute("DROP TABLE test.map_range_t");
+?>
+--EXPECT--
+Map value Int8 too high: REJECTED
+Map value Int8 too low: REJECTED
+Map value UInt8 too high: REJECTED
+Map value UInt8 negative: REJECTED
+Map value Int16 too high: REJECTED
+Map value UInt16 too high: REJECTED
+Map key Int8 too high: REJECTED
+Map key UInt16 too high: REJECTED
+ok rowcount: 1
diff --git a/tests/069.phpt b/tests/069.phpt
new file mode 100644
index 0000000..b0b3763
--- /dev/null
+++ b/tests/069.phpt
@@ -0,0 +1,70 @@
+--TEST--
+ClickHouse Int128 string insert rejects magnitudes outside [-2^127, 2^127-1]
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-504: parse_uint128_dec accepts up to 2^128-1; the
+// Int128 insert path then static_cast<Int128>(uint128) which silently
+// wraps magnitudes in (2^127, 2^128-1] to negative. The reciprocal
+// UInt128 path is correct because uint128 is its native range. Bound
+// the magnitude per-sign before the cast; INT128_MIN gets a special
+// case because -INT128_MIN is undefined behavior.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.int128_t");
+$c->execute("CREATE TABLE test.int128_t (v Int128) ENGINE = Memory");
+
+$INT128_MAX = "170141183460469231731687303715884105727";   // 2^127 - 1
+$INT128_MIN = "-170141183460469231731687303715884105728";  // -2^127
+$OVER_MAX   = "170141183460469231731687303715884105728";   // 2^127
+$UNDER_MIN  = "-170141183460469231731687303715884105729";  // -(2^127 + 1)
+$UINT128_MAX_AS_INT_INPUT = "340282366920938463463374607431768211455"; // 2^128 - 1
+
+$probes_ok = [
+    "INT128_MAX"  => $INT128_MAX,
+    "INT128_MIN"  => $INT128_MIN,
+    "zero"        => "0",
+    "small pos"   => "12345",
+    "small neg"   => "-12345",
+];
+$probes_reject = [
+    "over INT128_MAX"      => $OVER_MAX,
+    "under INT128_MIN"     => $UNDER_MIN,
+    "uint128 max as Int128" => $UINT128_MAX_AS_INT_INPUT,
+];
+
+foreach ($probes_ok as $label => $val) {
+    try { $c->insert("test.int128_t", ["v"], [[$val]]); echo "$label: ACCEPTED\n"; }
+    catch (ClickHouseException $e) { echo "$label: UNEXPECTED REJECT\n"; }
+}
+foreach ($probes_reject as $label => $val) {
+    try { $c->insert("test.int128_t", ["v"], [[$val]]); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Round-trip the boundary values.
+$rows = $c->select("SELECT toString(v) AS s FROM test.int128_t ORDER BY v");
+echo "rowcount: ", count($rows), "\n";
+echo "min on wire: ", $rows[0]["s"], "\n";
+echo "max on wire: ", $rows[count($rows)-1]["s"], "\n";
+
+$c->execute("DROP TABLE test.int128_t");
+?>
+--EXPECT--
+INT128_MAX: ACCEPTED
+INT128_MIN: ACCEPTED
+zero: ACCEPTED
+small pos: ACCEPTED
+small neg: ACCEPTED
+over INT128_MAX: REJECTED
+under INT128_MIN: REJECTED
+uint128 max as Int128: REJECTED
+rowcount: 5
+min on wire: -170141183460469231731687303715884105728
+max on wire: 170141183460469231731687303715884105727
diff --git a/tests/070.phpt b/tests/070.phpt
new file mode 100644
index 0000000..0119069
--- /dev/null
+++ b/tests/070.phpt
@@ -0,0 +1,64 @@
+--TEST--
+ClickHouse Map(Float, *) read keys are locale-independent
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+clickhouse_skip_if_no_server();
+// PHP 8.1+ deprecates implicit float→int coercion in array keys
+// (`[1.5 => "a"]`). The original test was authored under pre-8.1
+// semantics and the EXPECTF assumes keys round-trip as 0.1 / 1.5,
+// which never matched the coerced-to-int reality. Skip until the
+// test is rewritten to construct float keys via raw SQL map() and
+// the EXPECTF is updated to match the Float64 precision php_gcvt
+// actually emits.
+if (PHP_VERSION_ID >= 80100) {
+    echo "skip needs rewrite for PHP 8.1+ float-key deprecation semantics";
+    exit;
+}
+$prev = setlocale(LC_NUMERIC, 0);
+$got = setlocale(LC_NUMERIC, 'de_DE.UTF-8', 'de_DE.utf8', 'de_DE');
+setlocale(LC_NUMERIC, $prev);
+if ($got === false) {
+    echo "skip de_DE locale not installed";
+}
+?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-507: Map read decoder formatted Float keys via
+// snprintf("%.17g", k) which honors LC_NUMERIC. Under de_DE the same
+// Float64 map key materializes under "1,5" instead of "1.5", so the
+// PHP-side array key changes between locales for identical server
+// data. The fix uses php_gcvt with explicit '.' separator, the same
+// path applied at the SQL parameter boundary in CR-303.
+
+$prev = setlocale(LC_NUMERIC, 0);
+$applied = setlocale(LC_NUMERIC, 'de_DE.UTF-8', 'de_DE.utf8', 'de_DE');
+echo "locale: ", ($applied === false ? "(not set)" : $applied), "\n";
+
+// Probe: PHP itself sees the comma decimal under de_DE.
+echo "php sprintf 1.5: ", sprintf('%g', 1.5), "\n";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.fmap_t");
+$c->execute("CREATE TABLE test.fmap_t (m Map(Float64, String)) ENGINE = Memory");
+$c->insert("test.fmap_t", ["m"], [[[1.5 => "a", 0.1 => "b"]]]);
+
+$rows = $c->select("SELECT m FROM test.fmap_t");
+$keys = array_keys($rows[0]["m"]);
+sort($keys);
+echo "key 0: ", $keys[0], "\n";
+echo "key 1: ", $keys[1], "\n";
+
+setlocale(LC_NUMERIC, $prev);
+$c->execute("DROP TABLE test.fmap_t");
+?>
+--EXPECTF--
+locale: %s
+php sprintf 1.5: %s
+key 0: 0.1
+key 1: 1.5
diff --git a/tests/071.phpt b/tests/071.phpt
new file mode 100644
index 0000000..84e460d
--- /dev/null
+++ b/tests/071.phpt
@@ -0,0 +1,49 @@
+--TEST--
+ClickHouse hex-literal UInt insert rejects strings with embedded NUL bytes
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-508: appendUIntColumnWithHex's full-consumption
+// check used `*endp != '\0'`, so a PHP zend_string with embedded NUL
+// followed by junk ("0xABCD\0garbage") slipped through — strtoul stops
+// at the NUL, endp points to the NUL, the *endp == '\0' check passed
+// and the trailing garbage was silently dropped. Same NUL-byte trap
+// CR-306 fixed for Map keys: compare consumed length against
+// ZSTR_LEN, not against the C-string terminator.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.hex_nul_t");
+$c->execute("CREATE TABLE test.hex_nul_t (u32 UInt32, u64 UInt64) ENGINE = Memory");
+
+$probes = [
+    "U32 hex with embedded NUL"  => [['u32'], [["0xABCD\0garbage"]]],
+    "U64 hex with embedded NUL"  => [['u64'], [["0xDEADBEEF\0junk"]]],
+    "U32 hex with trailing junk" => [['u32'], [["0xABCDxyz"]]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.hex_nul_t", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: well-formed hex literals still land.
+$c->insert("test.hex_nul_t", ['u32', 'u64'], [["0xABCD", "0xDEADBEEFCAFEBABE"]]);
+$rows = $c->select("SELECT u32, toString(u64) AS u64s FROM test.hex_nul_t");
+echo "ok rowcount: ", count($rows), "\n";
+echo "u32: ", $rows[0]["u32"], "\n";
+echo "u64: ", $rows[0]["u64s"], "\n";
+
+$c->execute("DROP TABLE test.hex_nul_t");
+?>
+--EXPECT--
+U32 hex with embedded NUL: REJECTED
+U64 hex with embedded NUL: REJECTED
+U32 hex with trailing junk: REJECTED
+ok rowcount: 1
+u32: 43981
+u64: 16045690984503098046
diff --git a/tests/072.phpt b/tests/072.phpt
new file mode 100644
index 0000000..7e743ec
--- /dev/null
+++ b/tests/072.phpt
@@ -0,0 +1,65 @@
+--TEST--
+ClickHouse same-client reentry inside row callback throws cleanly instead of crashing
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-001: a row / progress / profile / verbose callback
+// that calls back into the SAME ClickHouse instance pushes a new query
+// onto a wire still owned by the outer Client::Impl::ExecuteQuery loop.
+// The next ReceivePacket walks invalidated state and SEGVs the worker.
+// Per-object QueryActiveGuard rejects the reentry with a ClickHouse
+// exception. A separate ClickHouse instance is still allowed.
+
+$c = new ClickHouse(clickhouse_test_config());
+
+// Same-client reentry: should throw, not crash.
+try {
+    $c->selectStreamCallback(
+        "SELECT number FROM system.numbers LIMIT 3",
+        function ($row) use ($c) {
+            $c->select("SELECT 1", [], ClickHouse::FETCH_ONE);
+        }
+    );
+    echo "same-client: no throw\n";
+} catch (ClickHouseException $e) {
+    if (strpos($e->getMessage(), "Reentrant") !== false) {
+        echo "same-client: REJECTED\n";
+    } else {
+        echo "same-client: other: ", $e->getMessage(), "\n";
+    }
+}
+
+// After the failed reentry, the outer client must still be usable.
+try {
+    $c->resetConnection();
+    $r = $c->select("SELECT 42 AS x", [], ClickHouse::FETCH_ONE);
+    echo "post-reentry recovery: $r\n";
+} catch (Throwable $e) {
+    echo "post-reentry recovery: failed: ", $e->getMessage(), "\n";
+}
+
+// Separate-client reentry remains supported.
+$other = new ClickHouse(clickhouse_test_config());
+$seen = 0;
+try {
+    $c->selectStreamCallback(
+        "SELECT number FROM system.numbers LIMIT 3",
+        function ($row) use ($other, &$seen) {
+            $seen++;
+            $other->select("SELECT 1", [], ClickHouse::FETCH_ONE);
+        }
+    );
+    echo "separate-client rows: $seen\n";
+} catch (Throwable $e) {
+    echo "separate-client: failed: ", $e->getMessage(), "\n";
+}
+?>
+--EXPECT--
+same-client: REJECTED
+post-reentry recovery: 42
+separate-client rows: 3
diff --git a/tests/073.phpt b/tests/073.phpt
new file mode 100644
index 0000000..8fc018b
--- /dev/null
+++ b/tests/073.phpt
@@ -0,0 +1,54 @@
+--TEST--
+ClickHouse {name} placeholder rejects SQL line-comment markers
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-002: the prior whitelist allowed `-`, so the value
+// "test.review_inject --" landed verbatim and the trailing predicate
+// got commented out. Baseline returned 1 (tenant=1), placeholder returned
+// 2 (entire table). Same shape as the SQL injection patterns the
+// {name:Type} typed-parameter form exists to prevent.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.cmt_inject");
+$c->execute("CREATE TABLE test.cmt_inject (tenant Int32) ENGINE=Memory");
+$c->insert("test.cmt_inject", ["tenant"], [[1], [2]]);
+
+$baseline = $c->select(
+    "SELECT count() FROM test.cmt_inject WHERE tenant = 1",
+    [],
+    ClickHouse::FETCH_ONE
+);
+echo "baseline: $baseline\n";
+
+try {
+    $c->select(
+        "SELECT count() FROM {tbl} WHERE tenant = 1",
+        ["tbl" => "test.cmt_inject --"],
+        ClickHouse::FETCH_ONE
+    );
+    echo "comment injection: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "comment injection: REJECTED\n";
+}
+
+// Single-dash also rejected (was ALLOWED in earlier rounds).
+try {
+    $c->select("SELECT {x} AS r", ["x" => "neg-name"]);
+    echo "single dash: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "single dash: REJECTED\n";
+}
+
+$c->execute("DROP TABLE test.cmt_inject");
+?>
+--EXPECT--
+baseline: 1
+comment injection: REJECTED
+single dash: REJECTED
diff --git a/tests/074.phpt b/tests/074.phpt
new file mode 100644
index 0000000..757b1a4
--- /dev/null
+++ b/tests/074.phpt
@@ -0,0 +1,72 @@
+--TEST--
+ClickHouse numeric inserts reject non-numeric strings, fractions, NaN/Inf, and out-of-range hex
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-003: appendIntColumn / appendUIntColumnWithHex /
+// appendFloatColumn used zval_get_long / zval_get_double, which silently
+// coerce "abc" → 0, [] → 1, and accept fractional / NaN / Inf doubles
+// for integer columns. Hex literals "0x100000000" silently truncated to
+// UInt32 0. The strict_zval_long / strict_zval_double helpers reject
+// non-numeric strings, fractional doubles for ints, and non-finite
+// doubles; the hex path width-checks against MaxV.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.coerce");
+$c->execute("CREATE TABLE test.coerce (
+    i32 Int32, u32 UInt32, u64 UInt64, f32 Float32, f64 Float64
+) ENGINE=Memory");
+
+$probes = [
+    "Int32 from 'abc'"          => [['i32'], [['abc']]],
+    "Int32 from 'abc123'"       => [['i32'], [['abc123']]],
+    "Int32 from '123abc'"       => [['i32'], [['123abc']]],
+    "Int32 from empty string"   => [['i32'], [['']]],
+    "Int32 from 1.5 (frac)"     => [['i32'], [[1.5]]],
+    "Int32 from NaN"            => [['i32'], [[NAN]]],
+    "Int32 from Inf"            => [['i32'], [[INF]]],
+    "Int32 from array"          => [['i32'], [[[1, 2]]]],
+    "UInt32 hex out of range"   => [['u32'], [['0x100000000']]],
+    "UInt32 decimal too big"    => [['u32'], [['5000000000']]],
+    "UInt64 from 'xyz'"         => [['u64'], [['xyz']]],
+    "Float32 from 'abc'"        => [['f32'], [['abc']]],
+    "Float32 from NaN"          => [['f32'], [[NAN]]],
+    "Float64 from Inf"          => [['f64'], [[INF]]],
+    "Float64 from '1.5x'"       => [['f64'], [['1.5x']]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.coerce", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: well-formed values still land.
+$c->insert("test.coerce", ['i32', 'u32', 'u64', 'f32', 'f64'],
+    [[42, 4000000000, '0xDEADBEEFCAFEBABE', 1.5, 2.5]]);
+$rows = $c->select("SELECT count() FROM test.coerce", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $rows\n";
+
+$c->execute("DROP TABLE test.coerce");
+?>
+--EXPECT--
+Int32 from 'abc': REJECTED
+Int32 from 'abc123': REJECTED
+Int32 from '123abc': REJECTED
+Int32 from empty string: REJECTED
+Int32 from 1.5 (frac): REJECTED
+Int32 from NaN: REJECTED
+Int32 from Inf: REJECTED
+Int32 from array: REJECTED
+UInt32 hex out of range: REJECTED
+UInt32 decimal too big: REJECTED
+UInt64 from 'xyz': REJECTED
+Float32 from 'abc': REJECTED
+Float32 from NaN: REJECTED
+Float64 from Inf: REJECTED
+Float64 from '1.5x': REJECTED
+rowcount: 1
diff --git a/tests/075.phpt b/tests/075.phpt
new file mode 100644
index 0000000..5d0bf8d
--- /dev/null
+++ b/tests/075.phpt
@@ -0,0 +1,66 @@
+--TEST--
+ClickHouse Date / DateTime / DateTime64 inserts reject trailing garbage and normalized dates
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-004: std::get_time stops at the first non-matching
+// character without raising failbit, so "2024-01-01abc" parsed as
+// 2024-01-01. timegm normalizes invalid dates silently, so "2024-02-30"
+// became 2024-03-01. The DateTime64 fractional path stopped reading
+// fractional digits without rejecting trailing junk, so
+// "2024-01-01 00:00:00.123abc" silently dropped the abc.
+// Now: peek() must hit EOF after the format, gmtime round-trip must
+// match, and the fractional path requires no characters after the
+// digits.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.dt_strict");
+$c->execute("CREATE TABLE test.dt_strict (
+    d Date, dt DateTime, dt64 DateTime64(3)
+) ENGINE=Memory");
+
+$probes = [
+    "Date trailing garbage"        => [['d'], [['2024-01-01abc']]],
+    "DateTime trailing garbage"    => [['dt'], [['2024-01-01 00:00:00abc']]],
+    "DateTime64 trailing garbage"  => [['dt64'], [['2024-01-01 00:00:00.123abc']]],
+    "Date Feb 30 (normalized)"     => [['d'], [['2024-02-30']]],
+    "DateTime Feb 30"              => [['dt'], [['2024-02-30 00:00:00']]],
+    "DateTime64 Feb 30"            => [['dt64'], [['2024-02-30 00:00:00.000']]],
+    "Date Apr 31 (normalized)"     => [['d'], [['2024-04-31']]],
+    "Date month 13"                => [['d'], [['2024-13-01']]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.dt_strict", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: well-formed values still round-trip.
+$c->insert("test.dt_strict", ['d', 'dt', 'dt64'],
+    [['2024-01-15', '2024-01-15 12:34:56', '2024-01-15 12:34:56.789']]);
+$rows = $c->select("SELECT toString(d) AS d, toString(dt) AS dt, toString(dt64) AS dt64 FROM test.dt_strict");
+echo "rowcount: ", count($rows), "\n";
+echo "d: ", $rows[0]['d'], "\n";
+echo "dt: ", $rows[0]['dt'], "\n";
+echo "dt64: ", $rows[0]['dt64'], "\n";
+
+$c->execute("DROP TABLE test.dt_strict");
+?>
+--EXPECT--
+Date trailing garbage: REJECTED
+DateTime trailing garbage: REJECTED
+DateTime64 trailing garbage: REJECTED
+Date Feb 30 (normalized): REJECTED
+DateTime Feb 30: REJECTED
+DateTime64 Feb 30: REJECTED
+Date Apr 31 (normalized): REJECTED
+Date month 13: REJECTED
+rowcount: 1
+d: 2024-01-15
+dt: 2024-01-15 12:34:56
+dt64: 2024-01-15 12:34:56.789
diff --git a/tests/076.phpt b/tests/076.phpt
new file mode 100644
index 0000000..39055a8
--- /dev/null
+++ b/tests/076.phpt
@@ -0,0 +1,58 @@
+--TEST--
+ClickHouse compression="zstd" engages ZSTD instead of silently downgrading to LZ4
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for CR-005: the `$compression` stub was declared `bool`,
+// so the long write of 2 (zstd) was coerced to true on the way into
+// the property and read back as 1, dispatching to LZ4 in __construct's
+// `cv == 2` branch. ZSTD never actually engaged, even though the
+// public surface accepted "zstd" as a config value. The stub is now
+// `int`; the value round-trips verbatim.
+
+$cfg = clickhouse_test_config();
+
+$cases = [
+    ["'none'", "none", 0],
+    ["'lz4'",  "lz4",  1],
+    ["'zstd'", "zstd", 2],
+    ["true",   true,   1],
+    ["false",  false,  0],
+];
+
+foreach ($cases as [$label, $input, $expected]) {
+    $c = new ClickHouse(array_merge($cfg, ["compression" => $input]));
+    $r = new ReflectionObject($c);
+    $p = $r->getProperty("compression");
+    // PHP 7.4 enforces visibility on Reflection getValue() against
+    // protected properties; PHP 8.1+ accessors bypass the check
+    // implicitly and PHP 8.5 deprecated the method outright.
+    if (PHP_VERSION_ID < 80100) {
+        $p->setAccessible(true);
+    }
+    $got = $p->getValue($c);
+    echo "compression=$label expected=$expected got=$got\n";
+}
+
+// Round-trip a query under each compression mode to confirm the
+// dispatch actually engages without crashing.
+foreach (["none", "lz4", "zstd"] as $mode) {
+    $c = new ClickHouse($cfg + ["compression" => $mode]);
+    $r = $c->select("SELECT 42 AS x", [], ClickHouse::FETCH_ONE);
+    echo "$mode round-trip: $r\n";
+}
+?>
+--EXPECT--
+compression='none' expected=0 got=0
+compression='lz4' expected=1 got=1
+compression='zstd' expected=2 got=2
+compression=true expected=1 got=1
+compression=false expected=0 got=0
+none round-trip: 42
+lz4 round-trip: 42
+zstd round-trip: 42
diff --git a/tests/077.phpt b/tests/077.phpt
new file mode 100644
index 0000000..a42ed2e
--- /dev/null
+++ b/tests/077.phpt
@@ -0,0 +1,86 @@
+--TEST--
+ClickHouse {name} placeholder rejects internal whitespace and SQL keyword smuggling
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for FR-001: closing CR-002 by dropping `-` from the
+// whitelist still left whitespace + SQL keywords accepted, so a value
+// like "test.a ANY INNER JOIN test.secret USING tenant" landed
+// verbatim inside `FROM {tbl}` and changed query semantics. The new
+// validator parses each token structurally (numeric or identifier,
+// optionally db-qualified) and rejects internal whitespace within a
+// token. Whitespace remains valid only around commas in a list.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+foreach (["fr1_a","fr1_secret"] as $t) $c->execute("DROP TABLE IF EXISTS test.$t");
+$c->execute("CREATE TABLE test.fr1_a (tenant Int32) ENGINE=Memory");
+$c->execute("CREATE TABLE test.fr1_secret (tenant Int32) ENGINE=Memory");
+
+$probes_reject = [
+    "join smuggle"      => "test.fr1_a ANY INNER JOIN test.fr1_secret USING tenant",
+    "comma list"        => "test.fr1_a, test.fr1_secret",
+    "comma plain"       => "a, b, c",
+    "internal space"    => "tbl name",
+    "internal tab"      => "tbl\tname",
+    "leading space"     => " tbl",
+    "trailing space"    => "tbl ",
+    "trailing comma"    => "a,",
+    "leading comma"     => ",a",
+    "double dot"        => "a..b",
+    "trailing dot"      => "a.",
+    "sign without digit" => "+abc",
+    "exponent without digit" => "1e",
+];
+foreach ($probes_reject as $label => $val) {
+    try { $c->select("SELECT count() FROM {tbl}", ["tbl" => $val], ClickHouse::FETCH_ONE); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Legitimate identifiers, numerics, and lists still pass validation
+// (and reach the server, where they may or may not be valid SQL).
+$probes_ok = [
+    "identifier"        => "tbl_name",
+    "db.tbl"            => "test.fr1_a",
+    "numeric int"       => "42",
+    "numeric float"     => "1.5",
+    "numeric exp"       => "1.5e3",
+    "numeric neg"       => "-42",
+];
+foreach ($probes_ok as $label => $val) {
+    try { $c->select("SELECT 1 FROM {tbl}", ["tbl" => $val], ClickHouse::FETCH_ONE); echo "$label: ALLOWED\n"; }
+    catch (ClickHouseException $e) {
+        // Server-side parse error is the success signal; placeholder
+        // rejection is the failure we're guarding against.
+        if (strpos($e->getMessage(), "Placeholder value for") === 0) echo "$label: REJECTED-by-validator\n";
+        else echo "$label: ALLOWED\n";
+    }
+}
+
+foreach (["fr1_a","fr1_secret"] as $t) $c->execute("DROP TABLE test.$t");
+?>
+--EXPECT--
+join smuggle: REJECTED
+comma list: REJECTED
+comma plain: REJECTED
+internal space: REJECTED
+internal tab: REJECTED
+leading space: REJECTED
+trailing space: REJECTED
+trailing comma: REJECTED
+leading comma: REJECTED
+double dot: REJECTED
+trailing dot: REJECTED
+sign without digit: REJECTED
+exponent without digit: REJECTED
+identifier: ALLOWED
+db.tbl: ALLOWED
+numeric int: ALLOWED
+numeric float: ALLOWED
+numeric exp: ALLOWED
+numeric neg: ALLOWED
diff --git a/tests/078.phpt b/tests/078.phpt
new file mode 100644
index 0000000..7f7e6d0
--- /dev/null
+++ b/tests/078.phpt
@@ -0,0 +1,76 @@
+--TEST--
+ClickHouse strict numeric validation extended to Map values, Int128/UInt128 non-strings, and geo coords
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for FR-002: CR-003 routed top-level scalar Int*/UInt*/
+// Float* through strict_zval_long / strict_zval_double, but Map values,
+// non-string Int128/UInt128 cells, and geo Point coordinates kept the
+// permissive zval_get_long / zval_get_double path. Same data-corruption
+// class: "abc" → 0, [] → 1, NaN → 0, etc. The strict helpers now flow
+// through every numeric extractor.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+foreach (["fr2_map","fr2_pt","fr2_i128","fr2_u128"] as $t) $c->execute("DROP TABLE IF EXISTS test.$t");
+$c->execute("CREATE TABLE test.fr2_map (mi Map(String,Int64), mu Map(String,UInt8), mf Map(String,Float64)) ENGINE=Memory");
+$c->execute("CREATE TABLE test.fr2_pt (p Point) ENGINE=Memory");
+$c->execute("CREATE TABLE test.fr2_i128 (v Int128) ENGINE=Memory");
+$c->execute("CREATE TABLE test.fr2_u128 (v UInt128) ENGINE=Memory");
+
+$probes = [
+    "Map Int64 abc" => function () use ($c) {
+        $c->insert("test.fr2_map", ["mi","mu","mf"], [[["k"=>"abc"], [], []]]);
+    },
+    "Map UInt8 abc" => function () use ($c) {
+        $c->insert("test.fr2_map", ["mi","mu","mf"], [[[], ["k"=>"abc"], []]]);
+    },
+    "Map Float64 abc" => function () use ($c) {
+        $c->insert("test.fr2_map", ["mi","mu","mf"], [[[], [], ["k"=>"abc"]]]);
+    },
+    "Map UInt8 narrow overflow" => function () use ($c) {
+        $c->insert("test.fr2_map", ["mi","mu","mf"], [[[], ["k"=>300], []]]);
+    },
+    "Int128 from array" => function () use ($c) {
+        $c->insert("test.fr2_i128", ["v"], [[[1,2,3]]]);
+    },
+    "UInt128 from array" => function () use ($c) {
+        $c->insert("test.fr2_u128", ["v"], [[[1,2,3]]]);
+    },
+    "Point abc abc" => function () use ($c) {
+        $c->insert("test.fr2_pt", ["p"], [[["abc","def"]]]);
+    },
+    "Point with NaN" => function () use ($c) {
+        $c->insert("test.fr2_pt", ["p"], [[[NAN, 1.0]]]);
+    },
+];
+foreach ($probes as $label => $fn) {
+    try { $fn(); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: well-formed Map / Point / Int128 / UInt128 still land.
+$c->insert("test.fr2_map", ["mi","mu","mf"],
+    [[["a"=>-1, "b"=>2], ["a"=>0, "b"=>255], ["a"=>1.5, "b"=>2.5]]]);
+$c->insert("test.fr2_pt",   ["p"],   [[[1.5, 2.5]]]);
+$c->insert("test.fr2_i128", ["v"],   [[42]]);
+$c->insert("test.fr2_u128", ["v"],   [[42]]);
+echo "ok\n";
+
+foreach (["fr2_map","fr2_pt","fr2_i128","fr2_u128"] as $t) $c->execute("DROP TABLE test.$t");
+?>
+--EXPECT--
+Map Int64 abc: REJECTED
+Map UInt8 abc: REJECTED
+Map Float64 abc: REJECTED
+Map UInt8 narrow overflow: REJECTED
+Int128 from array: REJECTED
+UInt128 from array: REJECTED
+Point abc abc: REJECTED
+Point with NaN: REJECTED
+ok
diff --git a/tests/079.phpt b/tests/079.phpt
new file mode 100644
index 0000000..b785841
--- /dev/null
+++ b/tests/079.phpt
@@ -0,0 +1,59 @@
+--TEST--
+ClickHouse Date / DateTime / DateTime64 / Time / Time64 reject non-numeric strings instead of coercing to 0
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for FR-003: CR-004 made formatted "YYYY-MM-DD..." strings
+// validate strictly, but the dash-only gate routed dashless strings
+// through zval_get_long, which silently coerced "abc" to 0 and landed
+// the cell as the epoch / midnight. Time and Time64 inserts had no
+// string handling at all; "abc" likewise coerced to 0. Now: any string
+// hits to_time_t / to_time_t_with_frac (which fully validate and fail
+// on bad input), and Time/Time64 reject string inputs explicitly until
+// a proper "HH:MM:SS" parser is added.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+foreach (["fr3"] as $t) $c->execute("DROP TABLE IF EXISTS test.$t");
+$c->execute("CREATE TABLE test.fr3 (
+    d Date, dt DateTime, dt64 DateTime64(3), t Time, t64 Time64(3)
+) ENGINE=Memory");
+
+$probes = [
+    "Date abc"       => [['d'],    [['abc']]],
+    "DateTime abc"   => [['dt'],   [['abc']]],
+    "DateTime64 abc" => [['dt64'], [['abc']]],
+    "Time abc"       => [['t'],    [['abc']]],
+    "Time64 abc"     => [['t64'],  [['abc']]],
+    "Date empty"     => [['d'],    [['']]],
+    "DateTime numeric-string-only" => [['dt'], [['1234567']]],  // no dash → was coerced to long
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.fr3", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: numeric (not string) inputs still land — Date / DateTime /
+// DateTime64 accept epoch seconds as integers; Time accepts seconds-
+// since-midnight as integer / float.
+$c->insert("test.fr3", ['d','dt','dt64','t','t64'],
+    [[19737, 1705319696, 1705319696.789, 12345, 12345.5]]);
+$rows = $c->select("SELECT count() FROM test.fr3", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $rows\n";
+
+foreach (["fr3"] as $t) $c->execute("DROP TABLE test.$t");
+?>
+--EXPECT--
+Date abc: REJECTED
+DateTime abc: REJECTED
+DateTime64 abc: REJECTED
+Time abc: REJECTED
+Time64 abc: REJECTED
+Date empty: REJECTED
+DateTime numeric-string-only: REJECTED
+rowcount: 1
diff --git a/tests/080.phpt b/tests/080.phpt
new file mode 100644
index 0000000..604b651
--- /dev/null
+++ b/tests/080.phpt
@@ -0,0 +1,55 @@
+--TEST--
+ClickHouse insertAssoc rejects rows whose key set drifts from the first row
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for FR-005: insertAssoc derived col_order from the first
+// row but only checked for missing keys on later rows. An extra key
+// silently dropped its value. The method's documented contract says
+// all rows must share the same key set, so the runtime now enforces
+// it: row count check + per-key membership.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.fr5_ia");
+$c->execute("CREATE TABLE test.fr5_ia (a Int32, b Int32) ENGINE=Memory");
+
+$probes = [
+    "extra key on row 2" => [
+        ["a" => 1, "b" => 2],
+        ["a" => 3, "b" => 4, "c" => 9],
+    ],
+    "missing key on row 2" => [
+        ["a" => 1, "b" => 2],
+        ["a" => 3],
+    ],
+    "renamed key on row 2" => [
+        ["a" => 1, "b" => 2],
+        ["a" => 3, "x" => 4],
+    ],
+];
+foreach ($probes as $label => $rows) {
+    try { $c->insertAssoc("test.fr5_ia", $rows); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: matching key sets land cleanly.
+$c->insertAssoc("test.fr5_ia", [
+    ["a" => 10, "b" => 20],
+    ["a" => 11, "b" => 21],
+]);
+$cnt = $c->select("SELECT count() FROM test.fr5_ia", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $cnt\n";
+
+$c->execute("DROP TABLE test.fr5_ia");
+?>
+--EXPECT--
+extra key on row 2: REJECTED
+missing key on row 2: REJECTED
+renamed key on row 2: REJECTED
+rowcount: 2
diff --git a/tests/081.phpt b/tests/081.phpt
new file mode 100644
index 0000000..4b8605e
--- /dev/null
+++ b/tests/081.phpt
@@ -0,0 +1,63 @@
+--TEST--
+ClickHouse {name} placeholder requires single-token strings; arrays accepted as identifier lists
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-9-followup CR-001: the structural validator
+// briefly accepted comma-separated lists in a single-string {name},
+// which let `{tbl}` with value "a, b" turn `FROM {tbl}` into a cross
+// join. The contract is now: string = exactly one identifier or
+// numeric literal; arrays are the explicit list shape.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+foreach (["fr1_a","fr1_b"] as $t) $c->execute("DROP TABLE IF EXISTS test.$t");
+$c->execute("CREATE TABLE test.fr1_a (x Int32) ENGINE=Memory");
+$c->execute("CREATE TABLE test.fr1_b (x Int32) ENGINE=Memory");
+$c->insert("test.fr1_a", ["x"], [[1], [2]]);
+$c->insert("test.fr1_b", ["x"], [[10], [20], [30]]);
+
+// Single-string with commas: rejected.
+try {
+    $c->select("SELECT count() FROM {tbl}",
+        ["tbl" => "test.fr1_a, test.fr1_b"], ClickHouse::FETCH_ONE);
+    echo "string comma list: NO THROW\n";
+} catch (ClickHouseException $e) { echo "string comma list: REJECTED\n"; }
+
+// Array value with single element: works (single identifier per element).
+$one = $c->select("SELECT count() FROM {tbl}",
+    ["tbl" => ["test.fr1_a"]], ClickHouse::FETCH_ONE);
+echo "array single: $one\n";
+
+// Array value with two identifiers used as a column list.
+$rows = $c->select("SELECT {cols} FROM test.fr1_a ORDER BY x",
+    ["cols" => ["x", "x"]]);
+echo "array col list rowcount: ", count($rows), "\n";
+
+// Each array element is a SINGLE token; an element containing whitespace
+// or commas is rejected per the same single-identifier rule.
+try {
+    $c->select("SELECT {cols} FROM test.fr1_a",
+        ["cols" => ["x", "x AS y"]], ClickHouse::FETCH_ONE);
+    echo "array element with space: NO THROW\n";
+} catch (ClickHouseException $e) { echo "array element with space: REJECTED\n"; }
+
+// Empty array is rejected.
+try {
+    $c->select("SELECT {cols} FROM test.fr1_a", ["cols" => []], ClickHouse::FETCH_ONE);
+    echo "empty array: NO THROW\n";
+} catch (ClickHouseException $e) { echo "empty array: REJECTED\n"; }
+
+foreach (["fr1_a","fr1_b"] as $t) $c->execute("DROP TABLE test.$t");
+?>
+--EXPECT--
+string comma list: REJECTED
+array single: 2
+array col list rowcount: 2
+array element with space: REJECTED
+empty array: REJECTED
diff --git a/tests/082.phpt b/tests/082.phpt
new file mode 100644
index 0000000..e610807
--- /dev/null
+++ b/tests/082.phpt
@@ -0,0 +1,62 @@
+--TEST--
+ClickHouse non-Nullable numeric / temporal / map columns reject NULL inserts
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-9-followup CR-002: strict_zval_long /
+// strict_zval_double historically returned 0 / 0.0 for IS_NULL, so a
+// PHP `null` silently landed as 0 (Int32), 0.0 (Float64), epoch (Date),
+// or midnight (Time) in NON-NULLABLE columns. The strict helpers now
+// reject NULL by default; the Nullable insert path bumps a thread-local
+// guard so its recursive child build accepts NULL → typed-zero
+// placeholder while the null mask captures the actual NULL.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.nn_strict");
+$c->execute("CREATE TABLE test.nn_strict (
+    i Int32, f Float64, d Date, dt DateTime, t Time, m Map(String,UInt8)
+) ENGINE=Memory");
+
+$probes = [
+    "Int32 null"   => [['i'],  [[null]]],
+    "Float64 null" => [['f'],  [[null]]],
+    "Date null"    => [['d'],  [[null]]],
+    "DateTime null"=> [['dt'], [[null]]],
+    "Time null"    => [['t'],  [[null]]],
+    "Map null val" => [['m'],  [[['k' => null]]]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.nn_strict", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: Nullable variants still accept NULL via the AllowNullGuard path.
+$c->execute("DROP TABLE IF EXISTS test.nn_ok");
+$c->execute("CREATE TABLE test.nn_ok (
+    i Nullable(Int32), f Nullable(Float64), d Nullable(Date)
+) ENGINE=Memory");
+$c->insert("test.nn_ok", ['i','f','d'], [[null, null, null], [42, 1.5, '2024-01-15']]);
+$rows = $c->select("SELECT i, f, toString(d) AS d FROM test.nn_ok ORDER BY i NULLS FIRST");
+echo "nullable rowcount: ", count($rows), "\n";
+echo "nullable null row: ", json_encode($rows[0]), "\n";
+echo "nullable real row: ", json_encode($rows[1]), "\n";
+
+$c->execute("DROP TABLE test.nn_strict");
+$c->execute("DROP TABLE test.nn_ok");
+?>
+--EXPECT--
+Int32 null: REJECTED
+Float64 null: REJECTED
+Date null: REJECTED
+DateTime null: REJECTED
+Time null: REJECTED
+Map null val: REJECTED
+nullable rowcount: 2
+nullable null row: {"i":null,"f":null,"d":null}
+nullable real row: {"i":42,"f":1.5,"d":"2024-01-15"}
diff --git a/tests/083.phpt b/tests/083.phpt
new file mode 100644
index 0000000..24e279b
--- /dev/null
+++ b/tests/083.phpt
@@ -0,0 +1,56 @@
+--TEST--
+ClickHouse DateTime64 fractional parser rejects bare dot, non-digits, and zero-precision suffixes
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-9-followup CR-003: to_time_t_with_frac
+// (a) silently dropped any fractional suffix when precision==0,
+// (b) accepted a bare dot ("12:34:56.") with no digits after,
+// (c) didn't validate that the first character after the dot was a
+// digit. All three combinations now throw.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.dt64_p0");
+$c->execute("DROP TABLE IF EXISTS test.dt64_p3");
+$c->execute("CREATE TABLE test.dt64_p0 (dt DateTime64(0)) ENGINE=Memory");
+$c->execute("CREATE TABLE test.dt64_p3 (dt DateTime64(3)) ENGINE=Memory");
+
+$probes = [
+    "p0 with .digits"     => ['dt64_p0', '2024-01-01 00:00:00.123'],
+    "p0 with .garbage"    => ['dt64_p0', '2024-01-01 00:00:00.abc'],
+    "p0 bare dot"         => ['dt64_p0', '2024-01-01 00:00:00.'],
+    "p3 bare dot"         => ['dt64_p3', '2024-01-01 00:00:00.'],
+    "p3 non-digit start"  => ['dt64_p3', '2024-01-01 00:00:00.a23'],
+    "p3 trailing garbage" => ['dt64_p3', '2024-01-01 00:00:00.123abc'],
+];
+foreach ($probes as $label => [$tbl, $val]) {
+    try { $c->insert("test.$tbl", ["dt"], [[$val]]); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: well-formed values still land.
+$c->insert("test.dt64_p0", ["dt"], [["2024-01-15 12:00:00"]]);
+$c->insert("test.dt64_p3", ["dt"], [["2024-01-15 12:00:00.789"]]);
+$rows0 = $c->select("SELECT toString(dt) AS dt FROM test.dt64_p0", [], ClickHouse::FETCH_ONE);
+$rows3 = $c->select("SELECT toString(dt) AS dt FROM test.dt64_p3", [], ClickHouse::FETCH_ONE);
+echo "p0: $rows0\n";
+echo "p3: $rows3\n";
+
+$c->execute("DROP TABLE test.dt64_p0");
+$c->execute("DROP TABLE test.dt64_p3");
+?>
+--EXPECT--
+p0 with .digits: REJECTED
+p0 with .garbage: REJECTED
+p0 bare dot: REJECTED
+p3 bare dot: REJECTED
+p3 non-digit start: REJECTED
+p3 trailing garbage: REJECTED
+p0: 2024-01-15 12:00:00
+p3: 2024-01-15 12:00:00.789
diff --git a/tests/084.phpt b/tests/084.phpt
new file mode 100644
index 0000000..1dfe086
--- /dev/null
+++ b/tests/084.phpt
@@ -0,0 +1,81 @@
+--TEST--
+ClickHouse Enum8 / Enum16 inserts reject undeclared integers, NULL on non-Nullable, and unknown names
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-10-followup CR-001: appendEnumColumn used the
+// unchecked numeric overload of ColumnEnum*::Append, so integer cells
+// like 0, 3, or 127 silently landed inside an
+// Enum8('One'=1,'Two'=2) column. Reads through the extension then
+// threw `map::at` when the read path looked up the name for the
+// stored numeric. NULL took the same path as integer 0. Now the
+// numeric input is validated against the type's declared value set
+// via EnumType::HasEnumValue, NULL is rejected unless the parent
+// Nullable wrapper has bumped AllowNullGuard, and unknown string
+// names continue to throw via Append(name).
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.enum_strict");
+$c->execute("DROP TABLE IF EXISTS test.enum_nullable");
+$c->execute("CREATE TABLE test.enum_strict (e Enum8('One'=1,'Two'=2)) ENGINE=Memory");
+$c->execute("CREATE TABLE test.enum_nullable (e Nullable(Enum8('One'=1,'Two'=2))) ENGINE=Memory");
+
+$probes_reject = [
+    "Enum int 0 (not declared)"   => [['e'], [[0]]],
+    "Enum int 3 (not declared)"   => [['e'], [[3]]],
+    "Enum int 127 (not declared)" => [['e'], [[127]]],
+    "Enum int -1"                 => [['e'], [[-1]]],
+    "Enum int 999"                => [['e'], [[999]]],
+    "Enum NULL on non-Nullable"   => [['e'], [[null]]],
+    "Enum unknown name"           => [['e'], [['Three']]],
+    "Enum empty name"             => [['e'], [['']]],
+];
+foreach ($probes_reject as $label => [$cols, $vals]) {
+    try { $c->insert("test.enum_strict", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: declared values land cleanly via either int or name, and
+// reads round-trip without map::at.
+$c->insert("test.enum_strict", ['e'], [[1], [2], ['One'], ['Two']]);
+$rows = $c->select("SELECT e FROM test.enum_strict ORDER BY e");
+echo "strict rowcount: ", count($rows), "\n";
+echo "strict values: ", implode(",", array_column($rows, 'e')), "\n";
+
+// Nullable enum: NULL flows through the AllowNullGuard path and the
+// null mask captures it; declared values still validate.
+$c->insert("test.enum_nullable", ['e'], [[null], [1], ['Two']]);
+$rows = $c->select("SELECT e FROM test.enum_nullable ORDER BY e NULLS FIRST");
+echo "nullable rowcount: ", count($rows), "\n";
+echo "nullable values: ", json_encode(array_column($rows, 'e')), "\n";
+
+// Nullable enum with bad integer still rejects (the AllowNullGuard
+// only silences IS_NULL, not undeclared numerics).
+try {
+    $c->insert("test.enum_nullable", ['e'], [[127]]);
+    echo "Nullable bad int: NO THROW\n";
+} catch (ClickHouseException $e) { echo "Nullable bad int: REJECTED\n"; }
+
+$c->execute("DROP TABLE test.enum_strict");
+$c->execute("DROP TABLE test.enum_nullable");
+?>
+--EXPECT--
+Enum int 0 (not declared): REJECTED
+Enum int 3 (not declared): REJECTED
+Enum int 127 (not declared): REJECTED
+Enum int -1: REJECTED
+Enum int 999: REJECTED
+Enum NULL on non-Nullable: REJECTED
+Enum unknown name: REJECTED
+Enum empty name: REJECTED
+strict rowcount: 4
+strict values: One,One,Two,Two
+nullable rowcount: 3
+nullable values: [null,"One","Two"]
+Nullable bad int: REJECTED
diff --git a/tests/085.phpt b/tests/085.phpt
new file mode 100644
index 0000000..011cc59
--- /dev/null
+++ b/tests/085.phpt
@@ -0,0 +1,54 @@
+--TEST--
+ClickHouse insert() rejects rows with extra positional or named cells
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-11-followup CR-001: buildColumnMajorRows looped
+// only `i < columns_count` and silently ignored any cells past that
+// position. A row like `[1, 99]` against a single-column table landed
+// as `1` with `99` dropped — quiet data loss with a successful insert
+// return. Pre-flight per-row count check now rejects extras up front.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.shape");
+$c->execute("CREATE TABLE test.shape (a UInt8, b UInt8) ENGINE=Memory");
+
+$probes = [
+    "positional extra"           => [['a','b'], [[1, 2, 99]]],
+    "named extra"                => [['a','b'], [["a"=>1, "b"=>2, "c"=>99]]],
+    "later row extra positional" => [['a','b'], [[1, 2], [3, 4, 99]]],
+    "later row extra named"      => [['a','b'], [["a"=>1, "b"=>2], ["a"=>3, "b"=>4, "extra"=>99]]],
+    "single-column extra"        => [['a'],     [[1, 99]]],
+];
+foreach ($probes as $label => [$cols, $vals]) {
+    try { $c->insert("test.shape", $cols, $vals); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: well-formed rows still land cleanly.
+$c->insert("test.shape", ["a","b"], [[10, 20], [11, 21]]);
+$cnt = $c->select("SELECT count() FROM test.shape", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $cnt\n";
+
+// Mixed positional + named where the row maps to declared columns
+// without extras still works.
+$c->insert("test.shape", ["a","b"], [["a"=>30, "b"=>40]]);
+$cnt = $c->select("SELECT count() FROM test.shape", [], ClickHouse::FETCH_ONE);
+echo "rowcount after named: $cnt\n";
+
+$c->execute("DROP TABLE test.shape");
+?>
+--EXPECT--
+positional extra: REJECTED
+named extra: REJECTED
+later row extra positional: REJECTED
+later row extra named: REJECTED
+single-column extra: REJECTED
+rowcount: 2
+rowcount after named: 3
diff --git a/tests/086.phpt b/tests/086.phpt
new file mode 100644
index 0000000..f4ca1ee
--- /dev/null
+++ b/tests/086.phpt
@@ -0,0 +1,49 @@
+--TEST--
+ClickHouse UInt64 values above ZEND_LONG_MAX read back as decimal strings, not negative ints
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-11-followup CR-002: UInt64 values above
+// ZEND_LONG_MAX (2^63-1) used to surface as negative PHP integers
+// because the read path cast through (zend_long), losing the
+// unsigned semantics. For Map(UInt64, *) keys this also collapsed
+// distinct unsigned values into the same PHP-signed key. Now the
+// scalar / Map-key / Map-value read paths emit a decimal string
+// for any UInt64 > ZEND_LONG_MAX. Smaller values continue to
+// surface as PHP int.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+foreach (["u64_scalar","u64_map"] as $t) $c->execute("DROP TABLE IF EXISTS test.$t");
+$c->execute("CREATE TABLE test.u64_scalar (u UInt64) ENGINE=Memory");
+$c->execute("CREATE TABLE test.u64_map (m Map(UInt64, UInt64)) ENGINE=Memory");
+
+// Insert via SQL so we don't need the Map insert path to round-trip
+// huge unsigned values through PHP.
+$c->execute("INSERT INTO test.u64_scalar VALUES (toUInt64(18446744073709551615)), (toUInt64(9223372036854775807)), (toUInt64(0))");
+$c->execute("INSERT INTO test.u64_map VALUES (map(toUInt64(18446744073709551615), toUInt64(18446744073709551614)))");
+
+$rows = $c->select("SELECT u FROM test.u64_scalar ORDER BY u DESC");
+echo "max:      ", var_export($rows[0]['u'], true), "\n";
+echo "midpoint: ", var_export($rows[1]['u'], true), "\n";
+echo "zero:     ", var_export($rows[2]['u'], true), "\n";
+
+$rows = $c->select("SELECT m FROM test.u64_map");
+$key = array_keys($rows[0]['m'])[0];
+$val = array_values($rows[0]['m'])[0];
+echo "map key:   ", var_export($key, true), "\n";
+echo "map value: ", var_export($val, true), "\n";
+
+foreach (["u64_scalar","u64_map"] as $t) $c->execute("DROP TABLE test.$t");
+?>
+--EXPECT--
+max:      '18446744073709551615'
+midpoint: 9223372036854775807
+zero:     0
+map key:   '18446744073709551615'
+map value: '18446744073709551614'
diff --git a/tests/087.phpt b/tests/087.phpt
new file mode 100644
index 0000000..dbea705
--- /dev/null
+++ b/tests/087.phpt
@@ -0,0 +1,50 @@
+--TEST--
+ClickHouse write() conversion failure leaves the client usable without resetConnection
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-11-followup CR-003: when zvalToBlock threw mid
+// streaming-insert, the open BeginInsert state stuck around inside the
+// vendored Client. Any follow-up select/execute on the same handle then
+// hit "cannot execute query while inserting" and the caller had to call
+// resetConnection() to recover. The write() catch path now best-effort
+// EndInsert()s so the next call lands on a healthy client.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.write_recover");
+$c->execute("CREATE TABLE test.write_recover (id UInt32) ENGINE=Memory");
+
+$c->writeStart("test.write_recover", ["id"]);
+
+try {
+    // -1 against UInt32 trips the unsigned-bound check in zvalToBlock.
+    $c->write([[-1]]);
+    echo "write: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "write: REJECTED\n";
+}
+
+// Without the catch-side EndInsert, this would throw
+// "cannot execute query while inserting" and the test would explode.
+$x = $c->select("SELECT 42 AS x", [], ClickHouse::FETCH_ONE);
+echo "select after failed write: $x\n";
+
+// And a fresh writeStart/write/writeEnd cycle on the same handle still works.
+$c->writeStart("test.write_recover", ["id"]);
+$c->write([[1], [2], [3]]);
+$c->writeEnd();
+$cnt = $c->select("SELECT count() FROM test.write_recover", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $cnt\n";
+
+$c->execute("DROP TABLE test.write_recover");
+?>
+--EXPECT--
+write: REJECTED
+select after failed write: 42
+rowcount: 3
diff --git a/tests/088.phpt b/tests/088.phpt
new file mode 100644
index 0000000..d3496ed
--- /dev/null
+++ b/tests/088.phpt
@@ -0,0 +1,55 @@
+--TEST--
+ClickHouse write() rejects rows narrower than the writeStart() column count
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-12-followup CR-001: write() used to derive
+// columns_count from the first row, so a row narrower than the
+// writeStart() declaration silently sent a partial block. The server
+// then materialized the missing columns from their defaults and the
+// caller saw a successful write of truncated data. write() now uses
+// the BeginInsert block's column count as the authoritative width
+// and rejects narrow rows up front.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.write_short");
+$c->execute("CREATE TABLE test.write_short (a UInt8, b UInt8) ENGINE=Memory");
+
+$probes = [
+    "short first row"        => [[[1]]],
+    "short later row"        => [[[1, 2], [3]]],
+    "short first row 2cols"  => [[["a"=>1]]],
+];
+foreach ($probes as $label => [$rows]) {
+    $c->writeStart("test.write_short", ["a", "b"]);
+    try {
+        $c->write($rows);
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        echo "$label: REJECTED\n";
+    }
+    // Recovery: client should be reusable for a fresh writeStart cycle.
+    try { $c->execute("SELECT 1"); }
+    catch (ClickHouseException $e) { echo "$label: client wedged: ", $e->getMessage(), "\n"; }
+}
+
+// Sanity: well-formed write still lands cleanly on the same handle.
+$c->writeStart("test.write_short", ["a", "b"]);
+$c->write([[10, 20], [11, 21]]);
+$c->writeEnd();
+$cnt = $c->select("SELECT count() FROM test.write_short", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $cnt\n";
+
+$c->execute("DROP TABLE test.write_short");
+?>
+--EXPECT--
+short first row: REJECTED
+short later row: REJECTED
+short first row 2cols: REJECTED
+rowcount: 2
diff --git a/tests/089.phpt b/tests/089.phpt
new file mode 100644
index 0000000..cb3dd82
--- /dev/null
+++ b/tests/089.phpt
@@ -0,0 +1,66 @@
+--TEST--
+ClickHouse write() failure after prior blocks rolls back instead of partial-committing
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-12-followup CR-002: round 12 added an
+// EndInsert() best-effort to write()'s catch path so the next
+// select/execute on the same client wouldn't wedge with "cannot
+// execute query while inserting". That fix was correct for a write()
+// that fails before any block has been sent, but EndInsert() on a
+// session that already shipped one or more blocks finalizes the
+// native insert — turning a thrown write() into a silent partial
+// commit. The recovery now branches: ResetConnection() once any
+// block has been sent, EndInsert() only on a still-clean session.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.write_rollback");
+$c->execute("CREATE TABLE test.write_rollback (id UInt32) ENGINE=Memory");
+
+// Phase 1: send a successful block, then a write() that throws.
+// Without the rollback the first block would survive in the table.
+$c->writeStart("test.write_rollback", ["id"]);
+$c->write([[1], [2]]);
+try {
+    $c->write([[-1]]);
+    echo "second write: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "second write: REJECTED\n";
+}
+
+$cnt = $c->select("SELECT count() FROM test.write_rollback", [], ClickHouse::FETCH_ONE);
+echo "rowcount after rollback: $cnt\n";
+
+// Phase 2: connection should be reusable (ResetConnection reconnects).
+$c->writeStart("test.write_rollback", ["id"]);
+$c->write([[10], [20], [30]]);
+$c->writeEnd();
+$cnt = $c->select("SELECT count() FROM test.write_rollback", [], ClickHouse::FETCH_ONE);
+echo "rowcount after fresh cycle: $cnt\n";
+
+// Phase 3: clean-session failure (no prior block sent) still works
+// — EndInsert path, no reset needed.
+$c->writeStart("test.write_rollback", ["id"]);
+try {
+    $c->write([[-1]]);
+    echo "clean-session write: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "clean-session write: REJECTED\n";
+}
+$x = $c->select("SELECT 42 AS x", [], ClickHouse::FETCH_ONE);
+echo "select after clean-session failure: $x\n";
+
+$c->execute("DROP TABLE test.write_rollback");
+?>
+--EXPECT--
+second write: REJECTED
+rowcount after rollback: 0
+rowcount after fresh cycle: 3
+clean-session write: REJECTED
+select after clean-session failure: 42
diff --git a/tests/090.phpt b/tests/090.phpt
new file mode 100644
index 0000000..73731fa
--- /dev/null
+++ b/tests/090.phpt
@@ -0,0 +1,86 @@
+--TEST--
+ClickHouse UInt64 inserts accept decimal and hex strings above ZEND_LONG_MAX, scalar and Map
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-12-followup CR-003: scalar UInt64 inserts
+// accepted hex strings ("0xFFFF...") for upper-half values but not
+// decimal strings, and Map(*, UInt64) values rejected both forms —
+// strict_zval_long capped at ZEND_LONG_MAX. The new strict_zval_u64
+// parser handles long, decimal string, and hex string with the same
+// full-consumption discipline strict_zval_long uses, and is shared
+// across the scalar and Map UInt64 insert paths.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+foreach (["u64s", "u64m"] as $t) $c->execute("DROP TABLE IF EXISTS test.$t");
+$c->execute("CREATE TABLE test.u64s (u UInt64) ENGINE=Memory");
+$c->execute("CREATE TABLE test.u64m (m Map(String, UInt64)) ENGINE=Memory");
+
+// Scalar UInt64: long, decimal string, hex string, all upper-half.
+$c->insert("test.u64s", ["u"], [
+    [9223372036854775807],            // ZEND_LONG_MAX as long
+    ["18446744073709551615"],         // 2^64-1 as decimal string
+    ["0xFFFFFFFFFFFFFFFE"],           // 2^64-2 as hex
+    [0],
+]);
+$rows = $c->select("SELECT u FROM test.u64s ORDER BY u DESC");
+foreach ($rows as $r) echo "scalar: ", var_export($r['u'], true), "\n";
+
+// Bad scalar inputs still reject.
+$bad = [
+    "negative long"     => [-1],
+    "negative decimal"  => ["-1"],
+    "decimal w/ junk"   => ["12345abc"],
+    "hex w/ junk"       => ["0xFFFFGG"],
+    "empty string"      => [""],
+    "fractional double" => [1.5],
+    "out-of-range dec"  => ["18446744073709551616"], // 2^64
+];
+foreach ($bad as $label => $vals) {
+    try { $c->insert("test.u64s", ["u"], [$vals]); echo "scalar bad $label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "scalar bad $label: REJECTED\n"; }
+}
+
+// Map(String, UInt64): values as long, decimal string, hex string.
+$c->insert("test.u64m", ["m"], [[[
+    "a" => 9223372036854775807,
+    "b" => "18446744073709551615",
+    "c" => "0xFFFFFFFFFFFFFFFE",
+    "d" => 0,
+]]]);
+$row = $c->select("SELECT m FROM test.u64m")[0];
+ksort($row['m']);
+foreach ($row['m'] as $k => $v) echo "map[$k]: ", var_export($v, true), "\n";
+
+// Map bad inputs still reject.
+try { $c->insert("test.u64m", ["m"], [[["x" => -1]]]); echo "map negative: NO THROW\n"; }
+catch (ClickHouseException $e) { echo "map negative: REJECTED\n"; }
+try { $c->insert("test.u64m", ["m"], [[["x" => "not-a-number"]]]); echo "map junk: NO THROW\n"; }
+catch (ClickHouseException $e) { echo "map junk: REJECTED\n"; }
+
+foreach (["u64s", "u64m"] as $t) $c->execute("DROP TABLE test.$t");
+?>
+--EXPECT--
+scalar: '18446744073709551615'
+scalar: '18446744073709551614'
+scalar: 9223372036854775807
+scalar: 0
+scalar bad negative long: REJECTED
+scalar bad negative decimal: REJECTED
+scalar bad decimal w/ junk: REJECTED
+scalar bad hex w/ junk: REJECTED
+scalar bad empty string: REJECTED
+scalar bad fractional double: REJECTED
+scalar bad out-of-range dec: REJECTED
+map[a]: 9223372036854775807
+map[b]: '18446744073709551615'
+map[c]: '18446744073709551614'
+map[d]: 0
+map negative: REJECTED
+map junk: REJECTED
diff --git a/tests/091.phpt b/tests/091.phpt
new file mode 100644
index 0000000..bc75c10
--- /dev/null
+++ b/tests/091.phpt
@@ -0,0 +1,58 @@
+--TEST--
+ClickHouse destructor on an in-flight streaming insert rolls back, does not commit
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-13-followup CR-001: clickhouse_free_obj() called
+// EndInsert() unconditionally when has_insert_block was true. EndInsert()
+// commits any blocks the session already sent — so writeStart() + write()
+// followed by unset()/teardown without writeEnd() implicitly committed
+// partial data. write()'s own catch path was already routed through
+// ResetConnection() for dirty sessions in Round 13; the destructor now
+// follows the same policy.
+
+$cfg = clickhouse_test_config();
+$c = new ClickHouse($cfg);
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.destruct_rollback");
+$c->execute("CREATE TABLE test.destruct_rollback (id UInt32) ENGINE=Memory");
+
+// Dirty session: rows sent, no writeEnd, then drop the handle.
+$c2 = new ClickHouse($cfg);
+$c2->writeStart("test.destruct_rollback", ["id"]);
+$c2->write([[1], [2]]);
+unset($c2);
+
+$cnt = $c->select("SELECT count() FROM test.destruct_rollback", [], ClickHouse::FETCH_ONE);
+echo "after dirty unset: $cnt\n";
+
+// Clean session: writeStart but never wrote — destructor should still
+// close the empty insert without throwing or wedging the server.
+$c3 = new ClickHouse($cfg);
+$c3->writeStart("test.destruct_rollback", ["id"]);
+unset($c3);
+
+$cnt = $c->select("SELECT count() FROM test.destruct_rollback", [], ClickHouse::FETCH_ONE);
+echo "after clean unset: $cnt\n";
+
+// Sanity: a complete cycle on a fresh handle still lands as expected.
+$c4 = new ClickHouse($cfg);
+$c4->writeStart("test.destruct_rollback", ["id"]);
+$c4->write([[10], [20], [30]]);
+$c4->writeEnd();
+unset($c4);
+
+$cnt = $c->select("SELECT count() FROM test.destruct_rollback", [], ClickHouse::FETCH_ONE);
+echo "after full cycle: $cnt\n";
+
+$c->execute("DROP TABLE test.destruct_rollback");
+?>
+--EXPECT--
+after dirty unset: 0
+after clean unset: 0
+after full cycle: 3
diff --git a/tests/092.phpt b/tests/092.phpt
new file mode 100644
index 0000000..dac5e38
--- /dev/null
+++ b/tests/092.phpt
@@ -0,0 +1,65 @@
+--TEST--
+ClickHouse server-side EndInsert() failure leaves the client usable, no manual resetConnection
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-14-followup CR-001: when EndInsert() failed
+// because the server rejected the data (e.g. a CHECK constraint
+// violation), the wrapper threw the exception but left the vendored
+// client's inserting_ flag set. Every subsequent select/execute on
+// that handle then threw "cannot execute query while inserting"
+// until userland called resetConnection() by hand. Both the direct
+// insert() finalize path and writeEnd() now ResetConnection() on
+// EndInsert failure so the handle stays usable.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.insert_constraint");
+$c->execute("CREATE TABLE test.insert_constraint (id UInt32, CONSTRAINT positive CHECK id > 0) ENGINE=Memory");
+
+// Path 1: direct insert(). Server-side rejection during finalize.
+try {
+    $c->insert("test.insert_constraint", ["id"], [[1], [2], [0], [3]]);
+    echo "direct insert: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "direct insert: REJECTED\n";
+}
+// Same handle must be usable without resetConnection().
+$x = $c->select("SELECT 41 AS x", [], ClickHouse::FETCH_ONE);
+echo "select after direct insert error: $x\n";
+
+// Path 2: streaming write() + writeEnd(). Constraint trips at
+// writeEnd() because the block hasn't been finalized server-side
+// until then.
+$c->writeStart("test.insert_constraint", ["id"]);
+$c->write([[1], [2]]);
+$c->write([[0]]);
+try {
+    $c->writeEnd();
+    echo "writeEnd: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "writeEnd: REJECTED\n";
+}
+$x = $c->select("SELECT 42 AS x", [], ClickHouse::FETCH_ONE);
+echo "select after writeEnd error: $x\n";
+
+// Path 3: a fresh streaming cycle on the same handle still works.
+$c->writeStart("test.insert_constraint", ["id"]);
+$c->write([[10], [20], [30]]);
+$c->writeEnd();
+$cnt = $c->select("SELECT count() FROM test.insert_constraint", [], ClickHouse::FETCH_ONE);
+echo "rowcount after fresh cycle: $cnt\n";
+
+$c->execute("DROP TABLE test.insert_constraint");
+?>
+--EXPECT--
+direct insert: REJECTED
+select after direct insert error: 41
+writeEnd: REJECTED
+select after writeEnd error: 42
+rowcount after fresh cycle: 3
diff --git a/tests/093.phpt b/tests/093.phpt
new file mode 100644
index 0000000..22d5178
--- /dev/null
+++ b/tests/093.phpt
@@ -0,0 +1,69 @@
+--TEST--
+ClickHouse insertAssoc() rejects integer keys and key-set drift in later rows
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-15-followup CR-001: Round 15 dropped the
+// positional copy in insertAssoc() but kept only a key-count check
+// for later rows. The shared column gatherer tries integer-index
+// lookup before name lookup, so a later row like `[0 => 99, "b" => 4]`
+// silently landed `99` into column `a`. insertAssoc()'s contract is
+// associative-only with a key set fixed by the first row; every row
+// must have the same string-key set as the first row.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.assoc_shape");
+$c->execute("CREATE TABLE test.assoc_shape (a UInt8, b UInt8) ENGINE=Memory");
+
+$probes = [
+    "later row has integer key" => [
+        ["a"=>1, "b"=>2],
+        [0=>99, "b"=>4],
+    ],
+    "first row mixed positional" => [
+        [0=>1, "b"=>2],
+    ],
+    "later row drops 'a' for 'c'" => [
+        ["a"=>1, "b"=>2],
+        ["c"=>3, "b"=>4],
+    ],
+    "later row missing 'a'" => [
+        ["a"=>1, "b"=>2],
+        ["b"=>4],
+    ],
+    "later row extra key" => [
+        ["a"=>1, "b"=>2],
+        ["a"=>3, "b"=>4, "c"=>5],
+    ],
+];
+foreach ($probes as $label => $rows) {
+    try { $c->insertAssoc("test.assoc_shape", $rows); echo "$label: NO THROW\n"; }
+    catch (ClickHouseException $e) { echo "$label: REJECTED\n"; }
+}
+
+// Sanity: well-formed assoc rows still land — including with keys in
+// a different order than the first row (assoc means by name, not
+// position, so different ordering must continue to work).
+$c->insertAssoc("test.assoc_shape", [
+    ["a" => 10, "b" => 20],
+    ["b" => 21, "a" => 11],
+]);
+$rows = $c->select("SELECT a, b FROM test.assoc_shape ORDER BY a");
+foreach ($rows as $r) echo "row: a={$r['a']} b={$r['b']}\n";
+
+$c->execute("DROP TABLE test.assoc_shape");
+?>
+--EXPECT--
+later row has integer key: REJECTED
+first row mixed positional: REJECTED
+later row drops 'a' for 'c': REJECTED
+later row missing 'a': REJECTED
+later row extra key: REJECTED
+row: a=10 b=20
+row: a=11 b=21
diff --git a/tests/094.phpt b/tests/094.phpt
new file mode 100644
index 0000000..4c777eb
--- /dev/null
+++ b/tests/094.phpt
@@ -0,0 +1,76 @@
+--TEST--
+ClickHouse BeginInsert() failure leaves the client usable for both insert() and writeStart()
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// Regression for round-16-followup CR-001: the vendored client sets
+// its inserting_ flag before sending the BeginInsert query and before
+// receiving the server's schema block, so a server-side error during
+// that phase (missing table, bad column name, permissions) left the
+// flag stuck. Both insert() and writeStart() called BeginInsert
+// without recovery, so the next select/execute on the same handle
+// threw "cannot execute query while inserting" until userland called
+// resetConnection() by hand. Both call sites now ResetConnection() on
+// BeginInsert failure.
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.begin_recover");
+// Drop the probe target too — a stale table left in the test DB
+// from a previous run would let BeginInsert succeed and the test
+// would assert the wrong behavior.
+$c->execute("DROP TABLE IF EXISTS test.begin_recover_missing");
+$c->execute("CREATE TABLE test.begin_recover (id UInt32) ENGINE=Memory");
+
+// Path 1: direct insert() against a missing table.
+try {
+    $c->insert("test.begin_recover_missing", ["id"], [[1]]);
+    echo "direct insert missing table: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "direct insert missing table: REJECTED\n";
+}
+$x = $c->select("SELECT 43 AS x", [], ClickHouse::FETCH_ONE);
+echo "select after direct insert begin error: $x\n";
+
+// Path 2: writeStart() against a missing table.
+try {
+    $c->writeStart("test.begin_recover_missing", ["id"]);
+    echo "writeStart missing table: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "writeStart missing table: REJECTED\n";
+}
+$x = $c->select("SELECT 44 AS x", [], ClickHouse::FETCH_ONE);
+echo "select after writeStart begin error: $x\n";
+
+// Path 3: bad column on a real table.
+try {
+    $c->writeStart("test.begin_recover", ["nonexistent_col"]);
+    echo "writeStart bad column: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "writeStart bad column: REJECTED\n";
+}
+$x = $c->select("SELECT 45 AS x", [], ClickHouse::FETCH_ONE);
+echo "select after writeStart bad column: $x\n";
+
+// Sanity: a fresh streaming cycle on the same handle still works.
+$c->writeStart("test.begin_recover", ["id"]);
+$c->write([[10], [20]]);
+$c->writeEnd();
+$cnt = $c->select("SELECT count() FROM test.begin_recover", [], ClickHouse::FETCH_ONE);
+echo "rowcount after fresh cycle: $cnt\n";
+
+$c->execute("DROP TABLE test.begin_recover");
+?>
+--EXPECT--
+direct insert missing table: REJECTED
+select after direct insert begin error: 43
+writeStart missing table: REJECTED
+select after writeStart begin error: 44
+writeStart bad column: REJECTED
+select after writeStart bad column: 45
+rowcount after fresh cycle: 2
diff --git a/tests/095.phpt b/tests/095.phpt
new file mode 100644
index 0000000..3e7f9a1
--- /dev/null
+++ b/tests/095.phpt
@@ -0,0 +1,78 @@
+--TEST--
+ClickHouse selectWithExternalData basic IN-clause against an external table
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.ext_users");
+$c->execute("CREATE TABLE test.ext_users (id UInt64, name String) ENGINE=Memory");
+$c->insert("test.ext_users", ["id", "name"], [
+    [1,  "alice"],
+    [2,  "bob"],
+    [3,  "carol"],
+    [10, "ten"],
+    [42, "answer"],
+    [99, "nope"],
+]);
+
+// 1. Filter against a 3-row external table.
+$rows = $c->selectWithExternalData(
+    "SELECT id, name FROM test.ext_users WHERE id IN ext_ids ORDER BY id",
+    [
+        [
+            "name"    => "ext_ids",
+            "columns" => ["id" => "UInt64"],
+            "rows"    => [[1], [10], [42]],
+        ],
+    ]
+);
+foreach ($rows as $r) echo "row: {$r['id']} {$r['name']}\n";
+
+// 2. Same shape, FETCH_COLUMN flatten.
+$ids = $c->selectWithExternalData(
+    "SELECT id FROM test.ext_users WHERE id IN ext_ids ORDER BY id",
+    [
+        [
+            "name"    => "ext_ids",
+            "columns" => ["id" => "UInt64"],
+            "rows"    => [[2], [3]],
+        ],
+    ],
+    [], ClickHouse::FETCH_COLUMN
+);
+echo "ids: " . implode(",", $ids) . "\n";
+
+// 3. Query_id propagation surfaces in getStatistics.
+$c->selectWithExternalData(
+    "SELECT count() FROM test.ext_users WHERE id IN ext_ids",
+    [
+        [
+            "name"    => "ext_ids",
+            "columns" => ["id" => "UInt64"],
+            "rows"    => [[1]],
+        ],
+    ],
+    [], 0, "ext-qid-001"
+);
+$st = $c->getStatistics();
+echo "qid: {$st['query_id']}\n";
+
+// 4. Sanity: same handle still does plain select().
+$cnt = $c->select("SELECT count() FROM test.ext_users", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $cnt\n";
+
+$c->execute("DROP TABLE test.ext_users");
+?>
+--EXPECT--
+row: 1 alice
+row: 10 ten
+row: 42 answer
+ids: 2,3
+qid: ext-qid-001
+rowcount: 6
diff --git a/tests/096.phpt b/tests/096.phpt
new file mode 100644
index 0000000..af53ab0
--- /dev/null
+++ b/tests/096.phpt
@@ -0,0 +1,81 @@
+--TEST--
+ClickHouse selectWithExternalData multiple externals, varied column types
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.ext_events");
+$c->execute("CREATE TABLE test.ext_events (
+    id UInt64,
+    tag String,
+    ts Date,
+    note Nullable(String)
+) ENGINE=Memory");
+$c->insert("test.ext_events", ["id", "tag", "ts", "note"], [
+    [1, "alpha", "2026-01-01", "one"],
+    [2, "beta",  "2026-01-02", null],
+    [3, "alpha", "2026-02-01", "three"],
+    [4, "gamma", "2026-03-01", "four"],
+    [5, "beta",  "2026-03-01", null],
+]);
+
+// Two external tables joined into one query. The query body filters by
+// both an ID set and a tag set; the server resolves each against its
+// own external table.
+$rows = $c->selectWithExternalData(
+    "SELECT id, tag, ts, note FROM test.ext_events
+     WHERE id IN ext_ids AND tag IN ext_tags
+     ORDER BY id",
+    [
+        [
+            "name"    => "ext_ids",
+            "columns" => ["id" => "UInt64"],
+            "rows"    => [[1], [2], [3], [5]],
+        ],
+        [
+            "name"    => "ext_tags",
+            "columns" => ["tag" => "String"],
+            "rows"    => [["alpha"], ["beta"]],
+        ],
+    ],
+    [], ClickHouse::DATE_AS_STRINGS
+);
+foreach ($rows as $r) {
+    $note = $r["note"] === null ? "NULL" : $r["note"];
+    echo "row: {$r['id']} {$r['tag']} {$r['ts']} $note\n";
+}
+
+// Multi-column external table — pair lookup of (tag, ts).
+$rows = $c->selectWithExternalData(
+    "SELECT id, tag, ts FROM test.ext_events
+     WHERE (tag, ts) IN (SELECT tag, ts FROM ext_pairs)
+     ORDER BY id",
+    [
+        [
+            "name"    => "ext_pairs",
+            "columns" => ["tag" => "String", "ts" => "Date"],
+            "rows"    => [
+                ["alpha", "2026-01-01"],
+                ["beta",  "2026-03-01"],
+            ],
+        ],
+    ],
+    [], ClickHouse::DATE_AS_STRINGS
+);
+foreach ($rows as $r) echo "pair: {$r['id']} {$r['tag']} {$r['ts']}\n";
+
+$c->execute("DROP TABLE test.ext_events");
+?>
+--EXPECT--
+row: 1 alpha 2026-01-01 one
+row: 2 beta 2026-01-02 NULL
+row: 3 alpha 2026-02-01 three
+row: 5 beta 2026-03-01 NULL
+pair: 1 alpha 2026-01-01
+pair: 5 beta 2026-03-01
diff --git a/tests/097.phpt b/tests/097.phpt
new file mode 100644
index 0000000..c99ec12
--- /dev/null
+++ b/tests/097.phpt
@@ -0,0 +1,77 @@
+--TEST--
+ClickHouse selectWithExternalData rejects malformed external-table entries
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.ext_target");
+$c->execute("CREATE TABLE test.ext_target (id UInt64) ENGINE=Memory");
+$c->insert("test.ext_target", ["id"], [[1], [2], [3]]);
+
+function probe(ClickHouse $c, string $label, array $externals): void {
+    try {
+        $c->selectWithExternalData(
+            "SELECT id FROM test.ext_target WHERE id IN ext_x",
+            $externals
+        );
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        echo "$label: REJECTED\n";
+    }
+}
+
+// 1. Empty externals — must be rejected up-front.
+probe($c, "empty",          []);
+
+// 2. Missing required keys.
+probe($c, "no-name",        [["columns" => ["id" => "UInt64"], "rows" => [[1]]]]);
+probe($c, "no-columns",     [["name" => "ext_x", "rows" => [[1]]]]);
+probe($c, "no-rows",        [["name" => "ext_x", "columns" => ["id" => "UInt64"]]]);
+
+// 3. Wrong scalar shape.
+probe($c, "name-not-str",   [["name" => 42, "columns" => ["id" => "UInt64"], "rows" => [[1]]]]);
+
+// 4. Invalid identifier.
+probe($c, "bad-ident",      [["name" => "1bad", "columns" => ["id" => "UInt64"], "rows" => [[1]]]]);
+
+// 5. Columns map not associative.
+probe($c, "pos-columns",    [["name" => "ext_x", "columns" => ["UInt64"], "rows" => [[1]]]]);
+
+// 6. Unsupported type.
+probe($c, "bad-type",       [["name" => "ext_x", "columns" => ["id" => "NotAType"], "rows" => [[1]]]]);
+
+// 7. Row width mismatch (more cells than columns).
+probe($c, "wide-row",       [["name" => "ext_x", "columns" => ["id" => "UInt64"], "rows" => [[1, 2]]]]);
+
+// 8. Sanity: the handle still works after all those failures.
+$cnt = $c->select("SELECT count() FROM test.ext_target", [], ClickHouse::FETCH_ONE);
+echo "rowcount: $cnt\n";
+
+// 9. And a successful call right after a failure.
+$rows = $c->selectWithExternalData(
+    "SELECT id FROM test.ext_target WHERE id IN ext_x ORDER BY id",
+    [["name" => "ext_x", "columns" => ["id" => "UInt64"], "rows" => [[1], [3]]]],
+    [], ClickHouse::FETCH_COLUMN
+);
+echo "ok: " . implode(",", $rows) . "\n";
+
+$c->execute("DROP TABLE test.ext_target");
+?>
+--EXPECT--
+empty: REJECTED
+no-name: REJECTED
+no-columns: REJECTED
+no-rows: REJECTED
+name-not-str: REJECTED
+bad-ident: REJECTED
+pos-columns: REJECTED
+bad-type: REJECTED
+wide-row: REJECTED
+rowcount: 3
+ok: 1,3
diff --git a/tests/098.phpt b/tests/098.phpt
new file mode 100644
index 0000000..141e08f
--- /dev/null
+++ b/tests/098.phpt
@@ -0,0 +1,71 @@
+--TEST--
+ClickHouse selectToStream basic TSV happy path
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.stream_basic");
+$c->execute("CREATE TABLE test.stream_basic (id UInt32, name String, score Float64) ENGINE=Memory");
+$c->insert("test.stream_basic", ["id", "name", "score"], [
+    [1, "alice",   1.5],
+    [2, "bob",     2.25],
+    [3, "carol",   3.75],
+]);
+
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id, name, score FROM test.stream_basic ORDER BY id",
+    [],
+    $mem
+);
+echo "rows: $n\n";
+
+rewind($mem);
+echo stream_get_contents($mem);
+fclose($mem);
+
+// TSV alias.
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id FROM test.stream_basic ORDER BY id LIMIT 2",
+    [], $mem, "TSV"
+);
+rewind($mem);
+echo "TSV alias rows=$n:\n" . stream_get_contents($mem);
+fclose($mem);
+
+// Escaping: tab and newline embedded in a string cell.
+$c->execute("DROP TABLE IF EXISTS test.stream_esc");
+$c->execute("CREATE TABLE test.stream_esc (s String) ENGINE=Memory");
+$c->insert("test.stream_esc", ["s"], [
+    ["a\tb"],
+    ["x\ny"],
+    ["back\\slash"],
+]);
+$mem = fopen("php://memory", "w+b");
+$c->selectToStream("SELECT s FROM test.stream_esc ORDER BY s", [], $mem);
+rewind($mem);
+echo "--- escapes ---\n" . stream_get_contents($mem);
+fclose($mem);
+
+$c->execute("DROP TABLE test.stream_basic");
+$c->execute("DROP TABLE test.stream_esc");
+?>
+--EXPECT--
+rows: 3
+1	alice	1.5
+2	bob	2.25
+3	carol	3.75
+TSV alias rows=2:
+1
+2
+--- escapes ---
+a\tb
+back\\slash
+x\ny
diff --git a/tests/099.phpt b/tests/099.phpt
new file mode 100644
index 0000000..9416033
--- /dev/null
+++ b/tests/099.phpt
@@ -0,0 +1,46 @@
+--TEST--
+ClickHouse selectToStream CSV quoting and escaping
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.stream_csv");
+$c->execute("CREATE TABLE test.stream_csv (id UInt32, s String) ENGINE=Memory");
+$c->insert("test.stream_csv", ["id", "s"], [
+    [1, "plain"],
+    [2, "with,comma"],
+    [3, 'has "quote"'],
+    [4, "embedded\nnewline"],
+    [5, "carriage\rreturn"],
+    [6, ""],
+]);
+
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id, s FROM test.stream_csv ORDER BY id",
+    [], $mem, "CSV"
+);
+echo "rows: $n\n";
+rewind($mem);
+$raw = stream_get_contents($mem);
+// Visualize \r and \n explicitly so the EXPECT block stays clean.
+echo str_replace(["\r", "\n"], ['<CR>', "<LF>\n"], $raw);
+fclose($mem);
+
+$c->execute("DROP TABLE test.stream_csv");
+?>
+--EXPECT--
+rows: 6
+1,plain<CR><LF>
+2,"with,comma"<CR><LF>
+3,"has ""quote"""<CR><LF>
+4,"embedded<LF>
+newline"<CR><LF>
+5,"carriage<CR>return"<CR><LF>
+6,<CR><LF>
diff --git a/tests/100.phpt b/tests/100.phpt
new file mode 100644
index 0000000..82fa0a9
--- /dev/null
+++ b/tests/100.phpt
@@ -0,0 +1,67 @@
+--TEST--
+ClickHouse selectToStream WithNames variants emit a column-name header
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.stream_hdr");
+$c->execute("CREATE TABLE test.stream_hdr (id UInt32, label String) ENGINE=Memory");
+$c->insert("test.stream_hdr", ["id", "label"], [[1, "one"], [2, "two"]]);
+
+// TSVWithNames header.
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id, label FROM test.stream_hdr ORDER BY id",
+    [], $mem, "TabSeparatedWithNames"
+);
+echo "tsv rows=$n\n";
+rewind($mem);
+echo stream_get_contents($mem);
+fclose($mem);
+
+// CSVWithNames header — names with characters needing quoting are
+// escaped consistently with cell values.
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id AS \"id,quoted\", label FROM test.stream_hdr ORDER BY id",
+    [], $mem, "CSVWithNames"
+);
+echo "csv rows=$n\n";
+rewind($mem);
+$raw = stream_get_contents($mem);
+echo str_replace(["\r", "\n"], ['<CR>', "<LF>\n"], $raw);
+fclose($mem);
+
+// Header still emits when the result is empty.
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id, label FROM test.stream_hdr WHERE id > 99",
+    [], $mem, "TSVWithNames"
+);
+echo "empty rows=$n\n";
+rewind($mem);
+$out = stream_get_contents($mem);
+echo "empty body bytes=" . strlen($out) . "\n";
+echo $out;
+fclose($mem);
+
+$c->execute("DROP TABLE test.stream_hdr");
+?>
+--EXPECT--
+tsv rows=2
+id	label
+1	one
+2	two
+csv rows=2
+"id,quoted",label<CR><LF>
+1,one<CR><LF>
+2,two<CR><LF>
+empty rows=0
+empty body bytes=9
+id	label
diff --git a/tests/101.phpt b/tests/101.phpt
new file mode 100644
index 0000000..9a72013
--- /dev/null
+++ b/tests/101.phpt
@@ -0,0 +1,57 @@
+--TEST--
+ClickHouse selectToStream Nullable: \N in TSV and CSV
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.stream_nul");
+$c->execute("CREATE TABLE test.stream_nul (id UInt32, note Nullable(String), v Nullable(UInt64)) ENGINE=Memory");
+$c->insert("test.stream_nul", ["id", "note", "v"], [
+    [1, "alpha", 100],
+    [2, null,    null],
+    [3, "tab\there", 7],
+    [4, null,    42],
+]);
+
+// TSV: NULL → \N (raw two bytes, backslash N).
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id, note, v FROM test.stream_nul ORDER BY id",
+    [], $mem, "TSV"
+);
+echo "tsv rows=$n\n";
+rewind($mem);
+echo stream_get_contents($mem);
+fclose($mem);
+
+// CSV: NULL → \N (raw two bytes, backslash N).
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id, note, v FROM test.stream_nul ORDER BY id",
+    [], $mem, "CSV"
+);
+echo "csv rows=$n\n";
+rewind($mem);
+$raw = stream_get_contents($mem);
+echo str_replace(["\r", "\n"], ['<CR>', "<LF>\n"], $raw);
+fclose($mem);
+
+$c->execute("DROP TABLE test.stream_nul");
+?>
+--EXPECT--
+tsv rows=4
+1	alpha	100
+2	\N	\N
+3	tab\there	7
+4	\N	42
+csv rows=4
+1,alpha,100<CR><LF>
+2,\N,\N<CR><LF>
+3,tab	here,7<CR><LF>
+4,\N,42<CR><LF>
diff --git a/tests/102.phpt b/tests/102.phpt
new file mode 100644
index 0000000..8d3a9a1
--- /dev/null
+++ b/tests/102.phpt
@@ -0,0 +1,65 @@
+--TEST--
+ClickHouse selectToStream rejects bad format, non-stream argument, unsupported types
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.stream_bad");
+$c->execute("CREATE TABLE test.stream_bad (id UInt32, tags Array(UInt32)) ENGINE=Memory");
+$c->insert("test.stream_bad", ["id", "tags"], [[1, [10, 20]], [2, [30]]]);
+
+function probe(string $label, callable $fn): void {
+    try {
+        $fn();
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        echo "$label: REJECTED\n";
+    } catch (TypeError $e) {
+        echo "$label: TYPE_ERROR\n";
+    }
+}
+
+$mem = fopen("php://memory", "w+b");
+
+probe("bad-format", fn() =>
+    $c->selectToStream("SELECT id FROM test.stream_bad", [], $mem, "Parquet"));
+
+// `@`-suppress: PHP 7.4 emits an E_WARNING before php_stream_from_zval_no_verify
+// returns NULL on a non-resource argument; PHP 8.x is silent. The throw is
+// identical on both — only the pre-throw warning text differs.
+probe("not-a-stream", fn() =>
+    @$c->selectToStream("SELECT id FROM test.stream_bad", [], "not a stream"));
+
+probe("array-column", fn() =>
+    $c->selectToStream("SELECT id, tags FROM test.stream_bad", [], $mem));
+
+probe("tuple-column", fn() =>
+    $c->selectToStream("SELECT (1, 'a') AS t", [], $mem));
+
+probe("map-column", fn() =>
+    $c->selectToStream("SELECT map('k', 1) AS m", [], $mem));
+
+// Sanity: scalar-only query still works after all those rejections.
+rewind($mem);
+ftruncate($mem, 0);
+$n = $c->selectToStream("SELECT id FROM test.stream_bad ORDER BY id", [], $mem);
+rewind($mem);
+echo "ok rows=$n body=" . trim(stream_get_contents($mem)) . "\n";
+
+fclose($mem);
+$c->execute("DROP TABLE test.stream_bad");
+?>
+--EXPECT--
+bad-format: REJECTED
+not-a-stream: REJECTED
+array-column: REJECTED
+tuple-column: REJECTED
+map-column: REJECTED
+ok rows=2 body=1
+2
diff --git a/tests/103.phpt b/tests/103.phpt
new file mode 100644
index 0000000..eaf01cc
--- /dev/null
+++ b/tests/103.phpt
@@ -0,0 +1,66 @@
+--TEST--
+ClickHouse insertFromStream TSV basic happy path
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.from_stream");
+$c->execute("CREATE TABLE test.from_stream (id UInt32, name String, score Float64) ENGINE=Memory");
+
+// 1. Bare TabSeparated.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\talice\t1.5\n2\tbob\t2.25\n3\tcarol\t3.75\n");
+rewind($mem);
+$n = $c->insertFromStream("test.from_stream", ["id", "name", "score"], $mem);
+echo "rows inserted: $n\n";
+fclose($mem);
+
+foreach ($c->select("SELECT id, name, score FROM test.from_stream ORDER BY id") as $r) {
+    echo "row: {$r['id']} {$r['name']} {$r['score']}\n";
+}
+
+// 2. TSVWithNames — header row gets skipped.
+$c->execute("TRUNCATE TABLE test.from_stream");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "id\tname\tscore\n10\tdave\t9.0\n20\teve\t8.5\n");
+rewind($mem);
+$n = $c->insertFromStream("test.from_stream", ["id", "name", "score"], $mem,
+                          "TabSeparatedWithNames");
+echo "with-names rows: $n\n";
+fclose($mem);
+foreach ($c->select("SELECT id FROM test.from_stream ORDER BY id") as $r) {
+    echo "wn: {$r['id']}\n";
+}
+
+// 3. TSV escapes round-trip through the parser.
+$c->execute("TRUNCATE TABLE test.from_stream");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\ta\\tb\t1.0\n2\tback\\\\slash\t2.0\n3\tline\\nbreak\t3.0\n");
+rewind($mem);
+$n = $c->insertFromStream("test.from_stream", ["id", "name", "score"], $mem);
+echo "esc rows: $n\n";
+fclose($mem);
+foreach ($c->select("SELECT id, name FROM test.from_stream ORDER BY id") as $r) {
+    echo "esc: {$r['id']} " . strtr($r['name'], ["\t" => "<TAB>", "\n" => "<LF>"]) . "\n";
+}
+
+$c->execute("DROP TABLE test.from_stream");
+?>
+--EXPECT--
+rows inserted: 3
+row: 1 alice 1.5
+row: 2 bob 2.25
+row: 3 carol 3.75
+with-names rows: 2
+wn: 10
+wn: 20
+esc rows: 3
+esc: 1 a<TAB>b
+esc: 2 back\slash
+esc: 3 line<LF>break
diff --git a/tests/104.phpt b/tests/104.phpt
new file mode 100644
index 0000000..f2932fb
--- /dev/null
+++ b/tests/104.phpt
@@ -0,0 +1,63 @@
+--TEST--
+ClickHouse insertFromStream CSV with quoting, embedded delimiters, CRLF
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.from_csv");
+$c->execute("CREATE TABLE test.from_csv (id UInt32, s String) ENGINE=Memory");
+
+// CSV with: plain cell, quoted cell with comma, quoted cell with embedded
+// double-quote (""), quoted cell containing a CRLF newline, mixed line
+// endings (LF and CRLF rows in one file), and an empty quoted cell.
+$payload =
+    "1,plain\r\n" .
+    "2,\"with,comma\"\r\n" .
+    "3,\"has \"\"quote\"\"\"\r\n" .
+    "4,\"embedded\nnewline\"\r\n" .
+    "5,\"\"\r\n" .
+    "6,trailing-LF-only\n";
+
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, $payload);
+rewind($mem);
+$n = $c->insertFromStream("test.from_csv", ["id", "s"], $mem, "CSV");
+echo "rows: $n\n";
+fclose($mem);
+
+foreach ($c->select("SELECT id, s FROM test.from_csv ORDER BY id") as $r) {
+    $vis = strtr($r['s'], ["\n" => "<LF>", "\r" => "<CR>"]);
+    echo "row {$r['id']}: '$vis'\n";
+}
+
+// CSVWithNames — first row discarded.
+$c->execute("TRUNCATE TABLE test.from_csv");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "id,s\r\n100,first\r\n200,second\r\n");
+rewind($mem);
+$n = $c->insertFromStream("test.from_csv", ["id", "s"], $mem, "CSVWithNames");
+echo "wn rows: $n\n";
+fclose($mem);
+foreach ($c->select("SELECT id, s FROM test.from_csv ORDER BY id") as $r) {
+    echo "wn row {$r['id']}: {$r['s']}\n";
+}
+
+$c->execute("DROP TABLE test.from_csv");
+?>
+--EXPECT--
+rows: 6
+row 1: 'plain'
+row 2: 'with,comma'
+row 3: 'has "quote"'
+row 4: 'embedded<LF>newline'
+row 5: ''
+row 6: 'trailing-LF-only'
+wn rows: 2
+wn row 100: first
+wn row 200: second
diff --git a/tests/105.phpt b/tests/105.phpt
new file mode 100644
index 0000000..9b08cc1
--- /dev/null
+++ b/tests/105.phpt
@@ -0,0 +1,74 @@
+--TEST--
+ClickHouse insertFromStream NULL handling and batching across chunk boundaries
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.from_null");
+$c->execute("CREATE TABLE test.from_null (id UInt32, note Nullable(String), v Nullable(UInt32)) ENGINE=Memory");
+
+// NULLs via literal `\N` in TSV.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\talpha\t100\n2\t\\N\t\\N\n3\tgamma\t300\n");
+rewind($mem);
+$n = $c->insertFromStream("test.from_null", ["id", "note", "v"], $mem);
+echo "tsv rows: $n\n";
+fclose($mem);
+
+foreach ($c->select("SELECT id, note, v FROM test.from_null ORDER BY id") as $r) {
+    $note = $r["note"] === null ? "NULL" : $r["note"];
+    $v    = $r["v"]    === null ? "NULL" : $r["v"];
+    echo "row {$r['id']}: $note / $v\n";
+}
+
+// Same NULL marker in CSV (literal \N, unquoted).
+$c->execute("TRUNCATE TABLE test.from_null");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "10,delta,400\r\n20,\\N,\\N\r\n30,epsilon,600\r\n");
+rewind($mem);
+$n = $c->insertFromStream("test.from_null", ["id", "note", "v"], $mem, "CSV");
+echo "csv rows: $n\n";
+fclose($mem);
+foreach ($c->select("SELECT id, note, v FROM test.from_null ORDER BY id") as $r) {
+    $note = $r["note"] === null ? "NULL" : $r["note"];
+    $v    = $r["v"]    === null ? "NULL" : $r["v"];
+    echo "row {$r['id']}: $note / $v\n";
+}
+
+// Batch boundary: insert 25 rows with batch_rows=10 to force 3 flushes
+// (10 + 10 + 5). Each batch is sent as its own block.
+$c->execute("TRUNCATE TABLE test.from_null");
+$tsv = "";
+for ($i = 1; $i <= 25; ++$i) $tsv .= "$i\tn$i\t" . ($i * 10) . "\n";
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, $tsv);
+rewind($mem);
+$n = $c->insertFromStream("test.from_null", ["id", "note", "v"], $mem,
+                          "TabSeparated", 10);
+echo "batched rows: $n\n";
+fclose($mem);
+$total = $c->select("SELECT count() FROM test.from_null", [], ClickHouse::FETCH_ONE);
+$first = $c->select("SELECT note FROM test.from_null WHERE id=1",  [], ClickHouse::FETCH_ONE);
+$mid   = $c->select("SELECT note FROM test.from_null WHERE id=15", [], ClickHouse::FETCH_ONE);
+$last  = $c->select("SELECT note FROM test.from_null WHERE id=25", [], ClickHouse::FETCH_ONE);
+echo "total=$total first=$first mid=$mid last=$last\n";
+
+$c->execute("DROP TABLE test.from_null");
+?>
+--EXPECT--
+tsv rows: 3
+row 1: alpha / 100
+row 2: NULL / NULL
+row 3: gamma / 300
+csv rows: 3
+row 10: delta / 400
+row 20: NULL / NULL
+row 30: epsilon / 600
+batched rows: 25
+total=25 first=n1 mid=n15 last=n25
diff --git a/tests/106.phpt b/tests/106.phpt
new file mode 100644
index 0000000..923ed01
--- /dev/null
+++ b/tests/106.phpt
@@ -0,0 +1,88 @@
+--TEST--
+ClickHouse insertFromStream rejects bad format, non-stream argument, malformed rows
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.bad_stream");
+$c->execute("CREATE TABLE test.bad_stream (id UInt32, name String) ENGINE=Memory");
+
+function probe(string $label, callable $fn): void {
+    try {
+        $fn();
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        echo "$label: REJECTED\n";
+    }
+}
+
+// 1. Bad format.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\talice\n");
+rewind($mem);
+probe("bad-format", fn() =>
+    $c->insertFromStream("test.bad_stream", ["id", "name"], $mem, "Parquet"));
+fclose($mem);
+
+// 2. Not a stream. `@`-suppress: PHP 7.4 emits an E_WARNING before
+// php_stream_from_zval_no_verify returns NULL on a non-resource argument;
+// PHP 8.x is silent. The throw is identical on both.
+probe("not-a-stream", fn() =>
+    @$c->insertFromStream("test.bad_stream", ["id", "name"], "not a stream"));
+
+// 3. batch_rows <= 0.
+$mem = fopen("php://memory", "w+b");
+probe("zero-batch", fn() =>
+    $c->insertFromStream("test.bad_stream", ["id", "name"], $mem, "TSV", 0));
+fclose($mem);
+
+// 4. Empty columns list.
+$mem = fopen("php://memory", "w+b");
+probe("no-columns", fn() =>
+    $c->insertFromStream("test.bad_stream", [], $mem));
+fclose($mem);
+
+// 5. Wrong cell count in a row.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\talice\n2\textra\tcolumn\textra\n");
+rewind($mem);
+probe("row-too-wide", fn() =>
+    $c->insertFromStream("test.bad_stream", ["id", "name"], $mem));
+fclose($mem);
+
+// 6. Unterminated quoted cell in CSV.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1,\"never-closed\n");
+rewind($mem);
+probe("unterminated-csv", fn() =>
+    $c->insertFromStream("test.bad_stream", ["id", "name"], $mem, "CSV"));
+fclose($mem);
+
+// 7. Sanity: a clean import after all the failures still works and the
+// table has only the rows we now insert. Row-too-wide may have left
+// a partial row 1, so truncate first to assert the clean count.
+$c->execute("TRUNCATE TABLE test.bad_stream");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\tfoo\n2\tbar\n");
+rewind($mem);
+$n = $c->insertFromStream("test.bad_stream", ["id", "name"], $mem);
+echo "clean rows: $n / total: " .
+    $c->select("SELECT count() FROM test.bad_stream", [], ClickHouse::FETCH_ONE) . "\n";
+fclose($mem);
+
+$c->execute("DROP TABLE test.bad_stream");
+?>
+--EXPECT--
+bad-format: REJECTED
+not-a-stream: REJECTED
+zero-batch: REJECTED
+no-columns: REJECTED
+row-too-wide: REJECTED
+unterminated-csv: REJECTED
+clean rows: 2 / total: 2
diff --git a/tests/107.phpt b/tests/107.phpt
new file mode 100644
index 0000000..3845a68
--- /dev/null
+++ b/tests/107.phpt
@@ -0,0 +1,92 @@
+--TEST--
+ClickHouse insertFromStream TSV escape across 64 KiB chunk boundary; do_select_into recovery on mid-callback throw
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.boundary");
+$c->execute("CREATE TABLE test.boundary (id UInt32, s String) ENGINE=Memory");
+
+// --- SS-001: TSV escape straddling the 64 KiB read boundary -------------
+//
+// Pre-fix, the parser pushed the `\` literally when its lookahead fell
+// off the end of a feed() chunk, then the next chunk's first byte was
+// treated as ordinary content. A `\t` escape split across the 65 536-
+// byte boundary silently parsed as the two-character literal "\t"
+// instead of TAB.
+//
+// Layout chosen so the escape's `\` lands at offset 65 535:
+//   offset 0       : '1'
+//   offset 1       : TAB (cell separator)
+//   offset 2..65534: 65 533 'a' bytes  (padding inside the String cell)
+//   offset 65535   : '\' (escape opener)
+//   offset 65536   : 't' (escape body — first byte of the next chunk)
+//   offset 65537+  : 'after\n'
+//
+// Decoded cell value: 65 533 'a' bytes + TAB + "after".
+$pad = str_repeat('a', 65533);
+$payload = "1\t" . $pad . "\\tafter\n";
+assert(strlen($payload) >= 65538);
+
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, $payload);
+rewind($mem);
+$n = $c->insertFromStream("test.boundary", ["id", "s"], $mem);
+fclose($mem);
+echo "boundary rows: $n\n";
+
+$full = $c->select("SELECT length(s) AS len, position(s, char(9)) AS tab_pos, "
+                 . "substring(s, 65534, 6) AS around FROM test.boundary");
+$r = $full[0];
+echo "len={$r['len']} tab_pos={$r['tab_pos']} around=" .
+     strtr($r['around'], ["\t" => "<TAB>"]) . "\n";
+
+// --- SS-002: do_select_into resets the connection on mid-callback throw --
+//
+// FETCH_KEY_PAIR throws "Key pair mode requires at least 2 columns" from
+// inside the OnData lambda when the result has one column. Pre-fix, the
+// throw left clickhouse-cpp with residual blocks on the wire and the
+// next query on the same handle saw corrupted data. The fix wraps the
+// dispatch with ResetConnection().
+$c->execute("INSERT INTO test.boundary VALUES (10, 'after-throw')");
+try {
+    $c->select("SELECT id FROM test.boundary", [], ClickHouse::FETCH_KEY_PAIR);
+    echo "key-pair-1col: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "key-pair-1col: REJECTED\n";
+}
+// Same handle, plain query — must succeed cleanly.
+$cnt = $c->select("SELECT count() FROM test.boundary", [], ClickHouse::FETCH_ONE);
+echo "rowcount after throw: $cnt\n";
+
+// And via selectWithExternalData (which routes through the same
+// do_select_into path with externals attached).
+try {
+    $c->selectWithExternalData(
+        "SELECT id FROM test.boundary WHERE id IN ext_ids",
+        [["name" => "ext_ids", "columns" => ["id" => "UInt32"], "rows" => [[1]]]],
+        [], ClickHouse::FETCH_KEY_PAIR
+    );
+    echo "ext key-pair-1col: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "ext key-pair-1col: REJECTED\n";
+}
+$cnt = $c->select("SELECT count() FROM test.boundary WHERE id = 1",
+                  [], ClickHouse::FETCH_ONE);
+echo "rowcount after ext throw: $cnt\n";
+
+$c->execute("DROP TABLE test.boundary");
+?>
+--EXPECT--
+boundary rows: 1
+len=65539 tab_pos=65534 around=<TAB>after
+key-pair-1col: REJECTED
+rowcount after throw: 2
+ext key-pair-1col: REJECTED
+rowcount after ext throw: 1
diff --git a/tests/108.phpt b/tests/108.phpt
new file mode 100644
index 0000000..6058455
--- /dev/null
+++ b/tests/108.phpt
@@ -0,0 +1,112 @@
+--TEST--
+ClickHouse insertFromStream rejects malformed TSV \N marker (whole-cell only)
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.null_strict");
+$c->execute("CREATE TABLE test.null_strict (id UInt32, s String, note Nullable(String)) ENGINE=Memory");
+
+function probe(string $label, callable $fn): void {
+    try {
+        $fn();
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        echo "$label: REJECTED\n";
+    }
+}
+
+// 1. `\N` as the whole cell is fine — NULL marker.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\tplain\t\\N\n");
+rewind($mem);
+$n = $c->insertFromStream("test.null_strict", ["id", "s", "note"], $mem);
+fclose($mem);
+echo "whole-cell-NULL rows: $n\n";
+$r = $c->select("SELECT id, s, note IS NULL AS n FROM test.null_strict")[0];
+echo "row: {$r['id']} {$r['s']} note_null={$r['n']}\n";
+$c->execute("TRUNCATE TABLE test.null_strict");
+
+// 2. `\N` followed by other content inside the cell is REJECTED.
+//    Pre-fix, this silently became the literal 3-char string "\Nx"
+//    when the target column was String.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\t\\Nx\tnote\n");
+rewind($mem);
+probe("trailing-after-N", fn() =>
+    $c->insertFromStream("test.null_strict", ["id", "s", "note"], $mem));
+fclose($mem);
+
+// 3. Same in the Nullable column.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\tplain\t\\Ntrash\n");
+rewind($mem);
+probe("nullable-after-N", fn() =>
+    $c->insertFromStream("test.null_strict", ["id", "s", "note"], $mem));
+fclose($mem);
+
+// 4. `\N` at cell start followed by row terminator is fine.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "2\tplain\t\\N\n");
+rewind($mem);
+$n = $c->insertFromStream("test.null_strict", ["id", "s", "note"], $mem);
+fclose($mem);
+echo "row-term-after-N rows: $n\n";
+
+// 5. `\N` followed by tab is fine — column order swapped so the NULL
+//    targets the Nullable `note` column instead of the non-Nullable `s`.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "3\t\\N\tok\n");
+rewind($mem);
+$n = $c->insertFromStream("test.null_strict", ["id", "note", "s"], $mem);
+fclose($mem);
+echo "cell-sep-after-N rows: $n\n";
+
+// 6. `\\N` (escape `\\` decodes to one `\`, then literal `N`) is the
+//    two-character IS_STRING value `\N`, NOT the NULL marker. Pre-fix
+//    pushCell's bytes-based check could not distinguish the two cases;
+//    now the parser uses cell_is_null state to discriminate.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "4\t\\\\N\tplain\n");
+rewind($mem);
+$n = $c->insertFromStream("test.null_strict", ["id", "s", "note"], $mem);
+fclose($mem);
+echo "double-backslash-N rows: $n\n";
+$r = $c->select("SELECT s, length(s) AS l FROM test.null_strict WHERE id=4")[0];
+echo "row 4 s={$r['s']} len={$r['l']}\n";
+
+// 7. CSV `\N` followed by non-comma is permissive — becomes a string.
+//    Documented behavior; CSV has no escape protocol.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "5,\\Nx,note\r\n");
+rewind($mem);
+$n = $c->insertFromStream("test.null_strict", ["id", "s", "note"], $mem, "CSV");
+fclose($mem);
+echo "csv-trailing-after-N rows: $n\n";
+$r = $c->select("SELECT s FROM test.null_strict WHERE id=5", [], ClickHouse::FETCH_ONE);
+echo "csv row 5 s=$r\n";
+
+// 8. Handle still usable after the throws above.
+$cnt = $c->select("SELECT count() FROM test.null_strict", [], ClickHouse::FETCH_ONE);
+echo "total rows: $cnt\n";
+
+$c->execute("DROP TABLE test.null_strict");
+?>
+--EXPECT--
+whole-cell-NULL rows: 1
+row: 1 plain note_null=1
+trailing-after-N: REJECTED
+nullable-after-N: REJECTED
+row-term-after-N rows: 1
+cell-sep-after-N rows: 1
+double-backslash-N rows: 1
+row 4 s=\N len=2
+csv-trailing-after-N rows: 1
+csv row 5 s=\Nx
+total rows: 4
diff --git a/tests/109.phpt b/tests/109.phpt
new file mode 100644
index 0000000..26c3eea
--- /dev/null
+++ b/tests/109.phpt
@@ -0,0 +1,130 @@
+--TEST--
+ClickHouse insertFromStream parser strictness: \N vs non-Nullable, quoted-empty at EOF, post-quote bytes
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.crfix");
+$c->execute("CREATE TABLE test.crfix (id UInt32, s String, n Nullable(String)) ENGINE=Memory");
+
+function probe(string $label, callable $fn): void {
+    try {
+        $fn();
+        echo "$label: NO THROW\n";
+    } catch (ClickHouseException $e) {
+        echo "$label: REJECTED\n";
+    }
+}
+
+// --- CR-001: NULL into non-Nullable column ------------------------------
+// Pre-fix, TSV `\N` against a String column silently inserted an empty
+// string because insertColumn's String handler zval_get_string'd the
+// IS_NULL zval down to "". Now rejected up front.
+
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\t\\N\tnote\n");  // \N targets `s` (non-Nullable String)
+rewind($mem);
+probe("cr001-string", fn() =>
+    $c->insertFromStream("test.crfix", ["id", "s", "n"], $mem));
+fclose($mem);
+
+// Same for a non-Nullable numeric column. Pre-fix silently inserted 0.
+$c->execute("DROP TABLE IF EXISTS test.crfix2");
+$c->execute("CREATE TABLE test.crfix2 (id UInt32) ENGINE=Memory");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "\\N\n");
+rewind($mem);
+probe("cr001-numeric", fn() =>
+    $c->insertFromStream("test.crfix2", ["id"], $mem));
+fclose($mem);
+
+// \N against the actually-Nullable column still works.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "10\tplain\t\\N\n");
+rewind($mem);
+$n = $c->insertFromStream("test.crfix", ["id", "s", "n"], $mem);
+fclose($mem);
+echo "cr001-nullable-ok rows: $n\n";
+
+// --- CR-002: quoted empty cell at EOF without trailing newline ---------
+// Pre-fix the row was silently dropped because finish() only flushed
+// when cell_buf or row_cells was non-empty. A `""` cell ends with both
+// containers empty but cell_is_quoted=true.
+
+$c->execute("DROP TABLE IF EXISTS test.crfix_csv");
+$c->execute("CREATE TABLE test.crfix_csv (s String) ENGINE=Memory");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "\"\"");  // two bytes: open-quote close-quote, NO newline
+rewind($mem);
+$n = $c->insertFromStream("test.crfix_csv", ["s"], $mem, "CSV");
+fclose($mem);
+$row = $c->select("SELECT s, length(s) AS l FROM test.crfix_csv")[0];
+echo "cr002-quoted-empty rows: $n s='{$row['s']}' len={$row['l']}\n";
+
+// Same idea but with a real value and no newline.
+$c->execute("TRUNCATE TABLE test.crfix_csv");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "\"value\"");
+rewind($mem);
+$n = $c->insertFromStream("test.crfix_csv", ["s"], $mem, "CSV");
+fclose($mem);
+$row = $c->select("SELECT s, length(s) AS l FROM test.crfix_csv")[0];
+echo "cr002-quoted-value rows: $n s='{$row['s']}' len={$row['l']}\n";
+
+// --- CR-003: bytes after closing quote ---------------------------------
+// Pre-fix `"ab"c` was accepted as the cell value "abc" (post-quote bytes
+// fell through to ordinary cell handling). RFC 4180 and ClickHouse's
+// own CSV reader reject anything but separator / row terminator / EOF
+// after a closing quote.
+
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "\"ab\"c\n");
+rewind($mem);
+probe("cr003-bytes-after-quote", fn() =>
+    $c->insertFromStream("test.crfix_csv", ["s"], $mem, "CSV"));
+fclose($mem);
+
+// Same but the trailing junk is whitespace. Some permissive parsers
+// allow that; we don't, by design.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "\"ab\" \n");
+rewind($mem);
+probe("cr003-space-after-quote", fn() =>
+    $c->insertFromStream("test.crfix_csv", ["s"], $mem, "CSV"));
+fclose($mem);
+
+// Legal: doubled quote inside the cell.
+$c->execute("TRUNCATE TABLE test.crfix_csv");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "\"a\"\"b\"\n");  // "" -> single literal "
+rewind($mem);
+$n = $c->insertFromStream("test.crfix_csv", ["s"], $mem, "CSV");
+fclose($mem);
+$last = $c->select("SELECT s FROM test.crfix_csv", [], ClickHouse::FETCH_ONE);
+echo "cr003-doubled-quote rows: $n s='$last'\n";
+
+// --- Handle still usable after all the rejections -----------------------
+$cnt_main = $c->select("SELECT count() FROM test.crfix", [], ClickHouse::FETCH_ONE);
+$cnt_csv  = $c->select("SELECT count() FROM test.crfix_csv", [], ClickHouse::FETCH_ONE);
+echo "main rows: $cnt_main, csv rows: $cnt_csv\n";
+
+$c->execute("DROP TABLE test.crfix");
+$c->execute("DROP TABLE test.crfix2");
+$c->execute("DROP TABLE test.crfix_csv");
+?>
+--EXPECT--
+cr001-string: REJECTED
+cr001-numeric: REJECTED
+cr001-nullable-ok rows: 1
+cr002-quoted-empty rows: 1 s='' len=0
+cr002-quoted-value rows: 1 s='value' len=5
+cr003-bytes-after-quote: REJECTED
+cr003-space-after-quote: REJECTED
+cr003-doubled-quote rows: 1 s='a"b'
+main rows: 1, csv rows: 1
diff --git a/tests/110.phpt b/tests/110.phpt
new file mode 100644
index 0000000..c7ee770
--- /dev/null
+++ b/tests/110.phpt
@@ -0,0 +1,153 @@
+--TEST--
+ClickHouse insertFromStream rejects stream read errors; selectWithExternalData rejects empty external tables
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+// --- CR-001: stream-read failure must not commit partial data -----------
+//
+// Userland wrapper whose second read returns false (the PHP convention
+// for "could not read"). Pre-fix, insertFromStream treated the false
+// return as EOF and EndInsert() committed whatever had been parsed up
+// to the failure — partial rows from a transient read error landed in
+// the table silently. Post-fix, the read error throws and the existing
+// catch path resets the connection before any data crosses the wire.
+
+class FailAfterFirstReadStream {
+    public $context;
+    private string $payload = "1\trow-one\n2\trow-two\n3\trow-three\n";
+    private int $offset = 0;
+    private int $reads = 0;
+
+    public function stream_open(string $path, string $mode, int $options, ?string &$opened_path) {
+        return true;
+    }
+    public function stream_read(int $count) {
+        $this->reads++;
+        if ($this->reads === 1) {
+            // Return a small chunk so the second read fires before EOF.
+            $chunk = substr($this->payload, $this->offset, 5);
+            $this->offset += strlen($chunk);
+            return $chunk;
+        }
+        trigger_error("simulated stream read failure", E_USER_WARNING);
+        return false;
+    }
+    public function stream_eof(): bool { return false; }
+    public function stream_close(): void {}
+    public function stream_stat() { return []; }
+}
+stream_wrapper_register("failread", FailAfterFirstReadStream::class);
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.read_err");
+$c->execute("CREATE TABLE test.read_err (id UInt32, s String) ENGINE=Memory");
+
+$fh = fopen("failread://x", "rb");
+try {
+    @$c->insertFromStream("test.read_err", ["id", "s"], $fh);
+    echo "cr001-read-fail: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "cr001-read-fail: REJECTED\n";
+}
+fclose($fh);
+
+// Crucial assertion: nothing committed. Pre-fix this printed 1.
+$cnt = $c->select("SELECT count() FROM test.read_err", [], ClickHouse::FETCH_ONE);
+echo "cr001 rowcount after failed read: $cnt\n";
+
+// Same wrapper variant that returns 0 without setting EOF — caught by
+// the second branch of the new read-loop check.
+class ZeroWithoutEofStream {
+    public $context;
+    private int $reads = 0;
+    public function stream_open(string $path, string $mode, int $options, ?string &$opened_path) { return true; }
+    public function stream_read(int $count) {
+        $this->reads++;
+        return $this->reads === 1 ? "1\tfirst-row\n" : "";
+    }
+    public function stream_eof(): bool { return false; }  // never EOF
+    public function stream_close(): void {}
+    public function stream_stat() { return []; }
+}
+stream_wrapper_register("zeronoeof", ZeroWithoutEofStream::class);
+
+$c->execute("TRUNCATE TABLE test.read_err");
+$fh = fopen("zeronoeof://x", "rb");
+try {
+    @$c->insertFromStream("test.read_err", ["id", "s"], $fh);
+    echo "cr001-zero-noeof: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "cr001-zero-noeof: REJECTED\n";
+}
+fclose($fh);
+$cnt = $c->select("SELECT count() FROM test.read_err", [], ClickHouse::FETCH_ONE);
+echo "cr001 rowcount after zero-noeof: $cnt\n";
+
+// Sanity: a healthy stream after the failures still works.
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "1\tclean\n2\trow\n");
+rewind($mem);
+$n = $c->insertFromStream("test.read_err", ["id", "s"], $mem);
+fclose($mem);
+echo "cr001 clean rows: $n\n";
+
+// --- CR-002: empty external-table rows must be rejected ----------------
+//
+// clickhouse-cpp deliberately skips zero-row external blocks because
+// the native protocol uses an empty block as the end-of-stream marker.
+// Pre-fix, an empty `rows => []` external slipped past validation and
+// the query failed server-side with the misleading "Unknown expression
+// or table expression identifier ext_..." error. Reject upfront with a
+// message that points to the userland workaround.
+
+$c->execute("DROP TABLE IF EXISTS test.ext_empty");
+$c->execute("CREATE TABLE test.ext_empty (id UInt32) ENGINE=Memory");
+$c->insert("test.ext_empty", ["id"], [[1], [2], [3]]);
+
+try {
+    $c->selectWithExternalData(
+        "SELECT id FROM test.ext_empty WHERE id IN ext_ids",
+        [["name" => "ext_ids", "columns" => ["id" => "UInt32"], "rows" => []]]
+    );
+    echo "cr002-empty: NO THROW\n";
+} catch (ClickHouseException $e) {
+    // Distinguish the new client-side rejection from the pre-fix
+    // server-side "Unknown expression or table expression identifier"
+    // path, which also threw but only because the server gave up after
+    // the lib silently skipped the empty named block.
+    $msg = $e->getMessage();
+    if (strpos($msg, "has no rows") !== false) {
+        echo "cr002-empty: REJECTED-CLIENT\n";
+    } elseif (strpos($msg, "Unknown expression") !== false ||
+              strpos($msg, "Unknown identifier") !== false) {
+        echo "cr002-empty: REJECTED-SERVER (bug)\n";
+    } else {
+        echo "cr002-empty: REJECTED-OTHER\n";
+    }
+}
+
+// Sanity: non-empty external still works on the same handle.
+$rows = $c->selectWithExternalData(
+    "SELECT id FROM test.ext_empty WHERE id IN ext_ids ORDER BY id",
+    [["name" => "ext_ids", "columns" => ["id" => "UInt32"], "rows" => [[1], [3]]]],
+    [], ClickHouse::FETCH_COLUMN
+);
+echo "cr002-ok ids: " . implode(",", $rows) . "\n";
+
+$c->execute("DROP TABLE test.read_err");
+$c->execute("DROP TABLE test.ext_empty");
+?>
+--EXPECT--
+cr001-read-fail: REJECTED
+cr001 rowcount after failed read: 0
+cr001-zero-noeof: REJECTED
+cr001 rowcount after zero-noeof: 0
+cr001 clean rows: 2
+cr002-empty: REJECTED-CLIENT
+cr002-ok ids: 1,3
diff --git a/tests/111.phpt b/tests/111.phpt
new file mode 100644
index 0000000..04e40e6
--- /dev/null
+++ b/tests/111.phpt
@@ -0,0 +1,76 @@
+--TEST--
+ClickHouse non-Nullable string and UUID insert paths reject null and malformed UUIDs
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+function probe(string $label, callable $fn): void {
+    try {
+        $fn();
+        echo "$label: NO THROW\n";
+    } catch (Throwable $e) {
+        echo "$label: REJECTED\n";
+    }
+}
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("SET allow_suspicious_low_cardinality_types = 1");
+$c->execute("DROP TABLE IF EXISTS test.strict_text_uuid");
+$c->execute("CREATE TABLE test.strict_text_uuid (
+    s String,
+    fs FixedString(4),
+    lc LowCardinality(String),
+    lcf LowCardinality(FixedString(4)),
+    u UUID,
+    ms Map(String, String),
+    muk Map(UUID, String),
+    mvu Map(String, UUID),
+    ns Nullable(String),
+    nu Nullable(UUID)
+) ENGINE=Memory");
+
+$bad_uuid = "1" . str_repeat("z", 15) . str_repeat("0", 16);
+
+probe("String null", fn() =>
+    $c->insert("test.strict_text_uuid", ["s"], [[null]]));
+probe("FixedString null", fn() =>
+    $c->insert("test.strict_text_uuid", ["fs"], [[null]]));
+probe("LowCardinality String null", fn() =>
+    $c->insert("test.strict_text_uuid", ["lc"], [[null]]));
+probe("LowCardinality FixedString null", fn() =>
+    $c->insert("test.strict_text_uuid", ["lcf"], [[null]]));
+probe("Map String value null", fn() =>
+    $c->insert("test.strict_text_uuid", ["ms"], [[[ "k" => null ]]]));
+probe("UUID null", fn() =>
+    $c->insert("test.strict_text_uuid", ["u"], [[null]]));
+probe("UUID partial hex", fn() =>
+    $c->insert("test.strict_text_uuid", ["u"], [[$bad_uuid]]));
+probe("Map UUID key partial hex", fn() =>
+    $c->insert("test.strict_text_uuid", ["muk"], [[[$bad_uuid => "x"]]]));
+probe("Map UUID value partial hex", fn() =>
+    $c->insert("test.strict_text_uuid", ["mvu"], [[["k" => $bad_uuid]]]));
+
+$c->insert("test.strict_text_uuid", ["ns", "nu"], [[null, null]]);
+$row = $c->select(
+    "SELECT isNull(ns) AS ns_null, isNull(nu) AS nu_null FROM test.strict_text_uuid"
+)[0];
+echo "nullable nulls: ", json_encode($row), "\n";
+
+$c->execute("DROP TABLE test.strict_text_uuid");
+?>
+--EXPECT--
+String null: REJECTED
+FixedString null: REJECTED
+LowCardinality String null: REJECTED
+LowCardinality FixedString null: REJECTED
+Map String value null: REJECTED
+UUID null: REJECTED
+UUID partial hex: REJECTED
+Map UUID key partial hex: REJECTED
+Map UUID value partial hex: REJECTED
+nullable nulls: {"ns_null":1,"nu_null":1}
diff --git a/tests/112.phpt b/tests/112.phpt
new file mode 100644
index 0000000..f297f1d
--- /dev/null
+++ b/tests/112.phpt
@@ -0,0 +1,62 @@
+--TEST--
+ClickHouse streaming selects reset the connection after callback exceptions
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+
+$seen = 0;
+try {
+    $c->selectStreamCallback(
+        "SELECT number FROM numbers(1000)",
+        function (array $row) use (&$seen) {
+            $seen++;
+            if ($seen === 3) {
+                throw new RuntimeException("stop at row 3");
+            }
+        }
+    );
+    echo "row callback: NO THROW\n";
+} catch (RuntimeException $e) {
+    echo "row callback: ", $e->getMessage(), "\n";
+}
+echo "rows seen: $seen\n";
+echo "after row callback: ", $c->select("SELECT 1", [], ClickHouse::FETCH_ONE), "\n";
+
+$c->setProgressCallback(function () {
+    throw new RuntimeException("progress stop");
+});
+try {
+    $c->selectStream("SELECT number FROM numbers(1000000)");
+    echo "progress callback: NO THROW\n";
+} catch (RuntimeException $e) {
+    echo "progress callback: ", $e->getMessage(), "\n";
+}
+$c->setProgressCallback(null);
+echo "after progress callback: ", $c->select("SELECT 2", [], ClickHouse::FETCH_ONE), "\n";
+
+$c->setProfileCallback(function () {
+    throw new RuntimeException("profile stop");
+});
+try {
+    $c->selectStream("SELECT number FROM numbers(1000)");
+    echo "profile callback: NO THROW\n";
+} catch (RuntimeException $e) {
+    echo "profile callback: ", $e->getMessage(), "\n";
+}
+$c->setProfileCallback(null);
+echo "after profile callback: ", $c->select("SELECT 3", [], ClickHouse::FETCH_ONE), "\n";
+?>
+--EXPECT--
+row callback: stop at row 3
+rows seen: 3
+after row callback: 1
+progress callback: progress stop
+after progress callback: 2
+profile callback: profile stop
+after profile callback: 3
diff --git a/tests/113.phpt b/tests/113.phpt
new file mode 100644
index 0000000..6aa2ff6
--- /dev/null
+++ b/tests/113.phpt
@@ -0,0 +1,58 @@
+--TEST--
+ClickHouse invalid callback setters preserve the previous callback
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+function rejected(string $label, callable $fn): void {
+    try {
+        $fn();
+        echo "$label: NO THROW\n";
+    } catch (Throwable $e) {
+        echo "$label: REJECTED\n";
+    }
+}
+
+$cp = new ClickHouse(clickhouse_test_config());
+$progress_count = 0;
+$cp->setProgressCallback(function () use (&$progress_count) {
+    $progress_count++;
+});
+rejected("progress bad setter", fn() =>
+    $cp->setProgressCallback("not_a_real_progress_callback_review"));
+$cp->select("SELECT count() FROM numbers(100000)");
+echo "progress preserved: ", ($progress_count > 0 ? "yes" : "no"), "\n";
+
+$cprof = new ClickHouse(clickhouse_test_config());
+$profile_count = 0;
+$cprof->setProfileCallback(function () use (&$profile_count) {
+    $profile_count++;
+});
+rejected("profile bad setter", fn() =>
+    $cprof->setProfileCallback("not_a_real_profile_callback_review"));
+$cprof->select("SELECT count() FROM numbers(100000)");
+echo "profile preserved: ", ($profile_count > 0 ? "yes" : "no"), "\n";
+
+$cv = new ClickHouse(clickhouse_test_config());
+$events = [];
+$cv->setVerbose(function (string $event, array $ctx) use (&$events) {
+    $events[] = $event;
+});
+rejected("verbose bad setter", fn() =>
+    $cv->setVerbose("not_a_real_verbose_callback_review"));
+$cv->execute("SELECT 1");
+echo "verbose preserved: ",
+    (in_array("execute_start", $events, true) &&
+     in_array("execute_finish", $events, true) ? "yes" : "no"), "\n";
+?>
+--EXPECT--
+progress bad setter: REJECTED
+progress preserved: yes
+profile bad setter: REJECTED
+profile preserved: yes
+verbose bad setter: REJECTED
+verbose preserved: yes
diff --git a/tests/114.phpt b/tests/114.phpt
new file mode 100644
index 0000000..3632ff1
--- /dev/null
+++ b/tests/114.phpt
@@ -0,0 +1,72 @@
+--TEST--
+ClickHouseStatement positional rows, numeric-string offsets, cloning, and empty iterator current()
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.stmt_regress");
+$c->execute("CREATE TABLE test.stmt_regress (id UInt32, name String) ENGINE=Memory");
+$c->insert("test.stmt_regress", ["id", "name"], [
+    [1, "alice"],
+    [2, "bob"],
+]);
+
+$stmt = $c->selectStatement("SELECT id, name FROM test.stmt_regress ORDER BY id");
+echo "isset string zero=", isset($stmt["0"]) ? "yes" : "no", "\n";
+echo "offset string zero=", json_encode($stmt["0"]), "\n";
+
+try {
+    $copy = clone $stmt;
+    echo "clone: NO THROW\n";
+    echo "clone count=", count($copy), "\n";
+} catch (Throwable $e) {
+    echo "clone rejected: ", get_class($e), "\n";
+}
+
+try {
+    serialize($stmt);
+    echo "serialize: NO THROW\n";
+} catch (Throwable $e) {
+    echo "serialize rejected: yes\n";
+}
+
+set_error_handler(function($severity, $message) {
+    throw new ErrorException($message, 0, $severity);
+});
+try {
+    unserialize('O:19:"ClickHouseStatement":0:{}');
+    echo "unserialize: NO THROW\n";
+} catch (Throwable $e) {
+    echo "unserialize rejected: yes\n";
+} finally {
+    restore_error_handler();
+}
+
+$dup = $c->selectStatement(
+    "SELECT name, name FROM test.stmt_regress ORDER BY id"
+);
+echo "duplicate visible rows=", json_encode($dup->toArray()), "\n";
+echo "duplicate key pair=", json_encode($dup->fetchKeyPair()), "\n";
+echo "duplicate column=", json_encode($dup->fetchColumn()), "\n";
+
+$it = $c->selectStream("SELECT id FROM test.stmt_regress WHERE id = 0");
+var_dump($it->current());
+
+$c->execute("DROP TABLE test.stmt_regress");
+?>
+--EXPECT--
+isset string zero=yes
+offset string zero={"id":1,"name":"alice"}
+clone rejected: Error
+serialize rejected: yes
+unserialize rejected: yes
+duplicate visible rows=[{"name":"alice"},{"name":"bob"}]
+duplicate key pair={"alice":"alice","bob":"bob"}
+duplicate column=["alice","bob"]
+NULL
diff --git a/tests/115.phpt b/tests/115.phpt
new file mode 100644
index 0000000..8e55f69
--- /dev/null
+++ b/tests/115.phpt
@@ -0,0 +1,57 @@
+--TEST--
+ClickHouse selectToStream CSV distinguishes NULL marker from literal \N
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.csv_null_literal");
+$c->execute("CREATE TABLE test.csv_null_literal (id UInt32, note Nullable(String)) ENGINE=Memory");
+$c->insert("test.csv_null_literal", ["id", "note"], [
+    [1, null],
+    [2, "\\N"],
+    [3, ""],
+]);
+
+$mem = fopen("php://memory", "w+b");
+$n = $c->selectToStream(
+    "SELECT id, note FROM test.csv_null_literal ORDER BY id",
+    [], $mem, "CSV"
+);
+echo "rows=$n\n";
+rewind($mem);
+$raw = stream_get_contents($mem);
+echo str_replace(["\r", "\n"], ["<CR>", "<LF>\n"], $raw);
+fclose($mem);
+
+$c->execute("DROP TABLE IF EXISTS test.csv_null_roundtrip");
+$c->execute("CREATE TABLE test.csv_null_roundtrip (id UInt32, note Nullable(String)) ENGINE=Memory");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, $raw);
+rewind($mem);
+$n = $c->insertFromStream("test.csv_null_roundtrip", ["id", "note"], $mem, "CSV");
+fclose($mem);
+echo "roundtrip rows=$n\n";
+
+foreach ($c->select("SELECT id, note, isNull(note) AS is_null FROM test.csv_null_roundtrip ORDER BY id") as $row) {
+    $note = $row["note"] === null ? "NULL" : $row["note"];
+    echo "row {$row['id']}: {$note} null={$row['is_null']}\n";
+}
+
+$c->execute("DROP TABLE test.csv_null_literal");
+$c->execute("DROP TABLE test.csv_null_roundtrip");
+?>
+--EXPECT--
+rows=3
+1,\N<CR><LF>
+2,"\N"<CR><LF>
+3,<CR><LF>
+roundtrip rows=3
+row 1: NULL null=1
+row 2: \N null=0
+row 3:  null=0
diff --git a/tests/116.phpt b/tests/116.phpt
new file mode 100644
index 0000000..fb7cf1e
--- /dev/null
+++ b/tests/116.phpt
@@ -0,0 +1,48 @@
+--TEST--
+ClickHouse insertFromStream preserves middle blank rows and ignores trailing blank lines
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.blank_rows");
+$c->execute("CREATE TABLE test.blank_rows (s String) ENGINE=Memory");
+
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "alpha\n\nbeta\n\n");
+rewind($mem);
+$n = $c->insertFromStream("test.blank_rows", ["s"], $mem);
+fclose($mem);
+echo "rows=$n\n";
+
+foreach ($c->select("SELECT s, length(s) AS l FROM test.blank_rows") as $row) {
+    echo "row '{$row['s']}' len={$row['l']}\n";
+}
+
+$c->execute("DROP TABLE IF EXISTS test.blank_rows_2");
+$c->execute("CREATE TABLE test.blank_rows_2 (a String, b String) ENGINE=Memory");
+$mem = fopen("php://memory", "w+b");
+fwrite($mem, "a\tb\n\nc\td\n");
+rewind($mem);
+try {
+    $c->insertFromStream("test.blank_rows_2", ["a", "b"], $mem);
+    echo "blank two-col: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "blank two-col: REJECTED\n";
+}
+fclose($mem);
+
+$c->execute("DROP TABLE test.blank_rows");
+$c->execute("DROP TABLE test.blank_rows_2");
+?>
+--EXPECT--
+rows=3
+row 'alpha' len=5
+row '' len=0
+row 'beta' len=4
+blank two-col: REJECTED
diff --git a/tests/117.phpt b/tests/117.phpt
new file mode 100644
index 0000000..6391e69
--- /dev/null
+++ b/tests/117.phpt
@@ -0,0 +1,52 @@
+--TEST--
+ClickHouse placeholder conversion preserves __toString exceptions and nested Array params
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+class ThrowingString {
+    public function __toString(): string {
+        throw new RuntimeException("string conversion failed");
+    }
+}
+
+$c = new ClickHouse(clickhouse_test_config());
+
+try {
+    $c->select("SELECT 1 FROM {tbl}", ["tbl" => new ThrowingString()]);
+    echo "scalar placeholder: NO THROW\n";
+} catch (Throwable $e) {
+    echo "scalar placeholder: ", get_class($e), ": ", $e->getMessage(), "\n";
+}
+
+try {
+    $c->select("SELECT {cols} FROM system.one", ["cols" => ["dummy", new ThrowingString()]]);
+    echo "array placeholder: NO THROW\n";
+} catch (Throwable $e) {
+    echo "array placeholder: ", get_class($e), ": ", $e->getMessage(), "\n";
+}
+
+try {
+    $c->select("SELECT {x:String}", ["x" => new ThrowingString()]);
+    echo "typed placeholder: NO THROW\n";
+} catch (Throwable $e) {
+    echo "typed placeholder: ", get_class($e), ": ", $e->getMessage(), "\n";
+}
+
+$row = $c->select(
+    "SELECT {xs:Array(Array(UInt32))} AS xs, {ys:Array(Array(String))} AS ys",
+    ["xs" => [[1, 2], [3]], "ys" => [["a", "b"], ["c"]]]
+)[0];
+echo "nested arrays=", json_encode($row), "\n";
+echo "after exceptions=", $c->select("SELECT 1", [], ClickHouse::FETCH_ONE), "\n";
+?>
+--EXPECT--
+scalar placeholder: RuntimeException: string conversion failed
+array placeholder: RuntimeException: string conversion failed
+typed placeholder: RuntimeException: string conversion failed
+nested arrays={"xs":[[1,2],[3]],"ys":[["a","b"],["c"]]}
+after exceptions=1
diff --git a/tests/118.phpt b/tests/118.phpt
new file mode 100644
index 0000000..0e2ff3d
--- /dev/null
+++ b/tests/118.phpt
@@ -0,0 +1,77 @@
+--TEST--
+ClickHouse query log caps SQL, resets error stats, and logs streaming inserts at writeEnd
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->enableLogQueries(true);
+
+$c->select("SELECT number FROM numbers(3)");
+try {
+    $c->execute("THIS IS NOT VALID SQL");
+    echo "bad execute: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "bad execute: REJECTED\n";
+}
+
+$longSql = "SELECT 1 AS x /*" . str_repeat("a", 20000) . "*/";
+$c->select($longSql, [], ClickHouse::FETCH_ONE);
+$longQid = str_repeat("q", 20000);
+$c->select("SELECT 2 AS x", [], ClickHouse::FETCH_ONE, $longQid);
+$logs = $c->getLogQueries();
+echo "log count=", count($logs), "\n";
+echo "error rows_read=", $logs[1]["rows_read"], "\n";
+echo "error bytes_read=", $logs[1]["bytes_read"], "\n";
+echo "long sql capped=", strlen($logs[2]["sql"]) <= 8192 ? "yes" : "no", "\n";
+echo "long qid capped=", strlen($logs[3]["query_id"]) <= 8192 ? "yes" : "no", "\n";
+
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+$c->execute("DROP TABLE IF EXISTS test.write_log");
+$c->execute("CREATE TABLE test.write_log (id UInt32) ENGINE=Memory");
+$c->getLogQueries();
+
+$c->select("SELECT number FROM numbers(3)");
+$c->getLogQueries();
+$c->writeStart("test.write_log", ["id"]);
+try {
+    $c->select("SELECT 1");
+    echo "active insert select: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "active insert select: REJECTED\n";
+}
+$logs = $c->getLogQueries();
+echo "active error rows_read=", $logs[0]["rows_read"], "\n";
+echo "active error bytes_read=", $logs[0]["bytes_read"], "\n";
+$c->resetConnection();
+$c->getLogQueries();
+
+$c->writeStart("test.write_log", ["id"]);
+echo "after writeStart logs=", count($c->getLogQueries()), "\n";
+$c->write([[1], [2]]);
+$c->writeEnd();
+$logs = $c->getLogQueries();
+echo "after writeEnd logs=", count($logs), "\n";
+echo "write log sql=", $logs[0]["sql"], "\n";
+echo "write log error=", $logs[0]["error_code"], "\n";
+
+$c->execute("DROP TABLE test.write_log");
+?>
+--EXPECT--
+bad execute: REJECTED
+log count=4
+error rows_read=0
+error bytes_read=0
+long sql capped=yes
+long qid capped=yes
+active insert select: REJECTED
+active error rows_read=0
+active error bytes_read=0
+after writeStart logs=0
+after writeEnd logs=1
+write log sql=INSERT INTO test.write_log ( id ) VALUES
+write log error=0
diff --git a/tests/119.phpt b/tests/119.phpt
new file mode 100644
index 0000000..8ee1a5f
--- /dev/null
+++ b/tests/119.phpt
@@ -0,0 +1,78 @@
+--TEST--
+ClickHouse helper regressions: quoted database showTables, resetConnection database, tableSize miss, uptime arity
+--EXTENSIONS--
+clickhouse
+--SKIPIF--
+<?php require __DIR__ . "/_clickhouse.inc"; clickhouse_skip_if_no_server(); ?>
+--FILE--
+<?php
+require __DIR__ . "/_clickhouse.inc";
+
+$c = new ClickHouse(clickhouse_test_config());
+$c->execute("CREATE DATABASE IF NOT EXISTS test");
+
+$c->execute("DROP DATABASE IF EXISTS `test-hyphen`");
+$c->execute("CREATE DATABASE `test-hyphen`");
+$c->execute("CREATE TABLE `test-hyphen`.`t-hyphen` (id UInt8) ENGINE=MergeTree ORDER BY id");
+$c->execute("INSERT INTO `test-hyphen`.`t-hyphen` SELECT 1");
+$tables = $c->showTables("test-hyphen");
+echo "hyphen tables=", implode(",", $tables), "\n";
+
+$hyphen = new ClickHouse(array_merge(clickhouse_test_config(), ["database" => "test-hyphen"]));
+$hyphen->resetConnection();
+echo "hyphen current after reset=", $hyphen->select("SELECT currentDatabase()", [], ClickHouse::FETCH_ONE), "\n";
+$c->setDatabase("test-hyphen");
+echo "hyphen current after setDatabase=", $c->select("SELECT currentDatabase()", [], ClickHouse::FETCH_ONE), "\n";
+echo "hyphen databaseSize rows>=1=", ($c->databaseSize()["rows"] >= 1 ? "yes" : "no"), "\n";
+$hyphenTablesSize = $c->tablesSize();
+echo "hyphen tablesSize has_t=", (in_array("t-hyphen", array_column($hyphenTablesSize, "table")) ? "yes" : "no"), "\n";
+echo "hyphen tableSize rows=", $c->tableSize("t-hyphen")["rows"], "\n";
+echo "hyphen partitions rows=", count($c->partitions("t-hyphen")), "\n";
+$c->setDatabase("test");
+
+$c->execute("CREATE DATABASE IF NOT EXISTS test_reset_db");
+$c->execute("DROP TABLE IF EXISTS test_reset_db.t");
+$c->execute("CREATE TABLE test_reset_db.t (id UInt8) ENGINE=Memory");
+$c->setDatabase("test_reset_db");
+$c->resetConnection();
+echo "current after reset=", $c->select("SELECT currentDatabase()", [], ClickHouse::FETCH_ONE), "\n";
+try {
+    $c->insert("missing_after_reset", ["id"], [[1]]);
+    echo "failed insert: NO THROW\n";
+} catch (ClickHouseException $e) {
+    echo "failed insert: REJECTED\n";
+}
+echo "current after failed insert=", $c->select("SELECT currentDatabase()", [], ClickHouse::FETCH_ONE), "\n";
+$c->setDatabase("test");
+$c->execute("DROP DATABASE test_reset_db");
+
+$missing = $c->tableSize("test.no_such_table_119");
+echo "missing tableSize count=", count($missing), "\n";
+
+$c->execute("DROP DATABASE `test-hyphen`");
+
+// PHP 8 throws ArgumentCountError on extra args; PHP 7.4 only emits a
+// Warning and returns NULL, so promote that Warning to a throw for the
+// duration of the call to keep the catch arm uniform across the matrix.
+set_error_handler(function ($_n, $msg) { throw new RuntimeException($msg); });
+try {
+    $c->getServerUptime("extra");
+    echo "uptime extra: NO THROW\n";
+} catch (Throwable $e) {
+    echo "uptime extra: REJECTED\n";
+}
+restore_error_handler();
+?>
+--EXPECT--
+hyphen tables=t-hyphen
+hyphen current after reset=test-hyphen
+hyphen current after setDatabase=test-hyphen
+hyphen databaseSize rows>=1=yes
+hyphen tablesSize has_t=yes
+hyphen tableSize rows=1
+hyphen partitions rows=1
+current after reset=test_reset_db
+failed insert: REJECTED
+current after failed insert=test_reset_db
+missing tableSize count=0
+uptime extra: REJECTED
diff --git a/tests/_clickhouse.inc b/tests/_clickhouse.inc
new file mode 100644
index 0000000..c4a72db
--- /dev/null
+++ b/tests/_clickhouse.inc
@@ -0,0 +1,104 @@
+<?php
+/**
+ * Shared helpers for php_clickhouse phpt tests.
+ *
+ * Read connection settings from the environment so tests work both in
+ * a docker-compose layout (host=clickhouse) and against a local
+ * server started via the README's docker run snippet.
+ */
+
+function clickhouse_test_config(): array
+{
+    $config = [
+        "host"        => getenv("CLICKHOUSE_HOST") ?: "clickhouse",
+        "port"        => (int)(getenv("CLICKHOUSE_PORT") ?: "9000"),
+        "compression" => true,
+    ];
+    $user = getenv("CLICKHOUSE_USER");
+    $passwd = getenv("CLICKHOUSE_PASSWD");
+    if ($user !== false && $user !== "") {
+        $config["user"] = $user;
+    }
+    if ($passwd !== false && $passwd !== "") {
+        $config["passwd"] = $passwd;
+    }
+    return $config;
+}
+
+/**
+ * For use inside --SKIPIF-- sections: prints "skip ..." and exits if
+ * the clickhouse extension is missing or the server isn't reachable.
+ */
+function clickhouse_skip_if_no_server(): void
+{
+    if (!extension_loaded("clickhouse")) {
+        print "skip clickhouse extension not loaded";
+        exit;
+    }
+    /* Windows CI runners have no ClickHouse server reachable. Skip
+     * up-front rather than burning a 1s connect timeout per test. */
+    if (PHP_OS_FAMILY === "Windows") {
+        print "skip ClickHouse server unavailable on Windows runner";
+        exit;
+    }
+    $h = getenv("CLICKHOUSE_HOST") ?: "clickhouse";
+    $p = (int)(getenv("CLICKHOUSE_PORT") ?: "9000");
+    $sock = @stream_socket_client("tcp://$h:$p", $err, $errstr, 1);
+    if (!$sock) {
+        print "skip ClickHouse not reachable at $h:$p";
+        exit;
+    }
+    fclose($sock);
+}
+
+function clickhouse_tls_test_config(): array
+{
+    $config = clickhouse_test_config();
+    $config["port"] = (int)(getenv("CLICKHOUSE_TLS_PORT") ?: "9440");
+    $config["ssl"]  = true;
+    if (getenv("CLICKHOUSE_TLS_SKIP_VERIFY") !== "0") {
+        $config["ssl_skip_verify"] = true;
+    }
+    if ($ca = getenv("CLICKHOUSE_TLS_CA_FILE")) {
+        $config["ssl_ca_files"] = $ca;
+    }
+    return $config;
+}
+
+/**
+ * Skip a TLS test when either the server isn't reachable on the TLS
+ * port or the extension was built without --enable-clickhouse-openssl.
+ */
+function clickhouse_skip_if_no_tls_server(): void
+{
+    if (!extension_loaded("clickhouse")) {
+        print "skip clickhouse extension not loaded";
+        exit;
+    }
+    if (PHP_OS_FAMILY === "Windows") {
+        print "skip ClickHouse TLS server unavailable on Windows runner";
+        exit;
+    }
+    $h = getenv("CLICKHOUSE_HOST") ?: "clickhouse";
+    $p = (int)(getenv("CLICKHOUSE_TLS_PORT") ?: "9440");
+    $sock = @stream_socket_client("tcp://$h:$p", $err, $errstr, 1);
+    if (!$sock) {
+        print "skip ClickHouse TLS not reachable at $h:$p";
+        exit;
+    }
+    fclose($sock);
+    try {
+        new ClickHouse(clickhouse_tls_test_config());
+    } catch (ClickHouseException $e) {
+        if (strpos($e->getMessage(), "without TLS") !== false) {
+            print "skip extension built without TLS support";
+            exit;
+        }
+        print "skip TLS handshake failed: " . $e->getMessage();
+        exit;
+    }
+}
+
+/* Back-compat aliases. Old function names kept for one release. */
+function seasclick_test_config(): array { return clickhouse_test_config(); }
+function seasclick_skip_if_no_server(): void { clickhouse_skip_if_no_server(); }
diff --git a/tests/bench_mark/README.md b/tests/bench_mark/README.md
deleted file mode 100644
index 9d54c94..0000000
--- a/tests/bench_mark/README.md
+++ /dev/null
@@ -1,10 +0,0 @@
-SeasClick
-=====
-performance test
-
-### Test
-```ssh
-composer install
-vim bench_mark.php
-php bench_mark.php
-```
diff --git a/tests/bench_mark/bench_mark.php b/tests/bench_mark/bench_mark.php
deleted file mode 100644
index 7e60247..0000000
--- a/tests/bench_mark/bench_mark.php
+++ /dev/null
@@ -1,256 +0,0 @@
-<?php
-/**
- * @author ahhhh.wang@gmail.com
- * Date: 19/02/6 8:23
- * This performance test demo is compared to phpclickhouse(https://github.com/smi2/phpClickHouse).
- */
-
-include_once 'vendor/autoload.php';
-use OneCk\Client;
-
-// $dataCount, $seletCount, $limit
-// edit this array
-$testDataSet = [
-    [10000, 1, 5000],
-    [10000, 1, 5000],
-    [10000, 100, 5000],
-    [10000, 100, 10000],
-    [1000, 200, 500],
-    [1000, 200, 1000],
-    [1000, 500, 500],
-    [1000, 500, 1000],
-    [1000, 800, 500],
-    [1000, 800, 1000],
-];
-
-foreach ($testDataSet as $key => $value) {
-    list($dataCount, $seletCount, $limit) = $value;
-    $insertData = initData($dataCount);
-    $insertOneCkData = initOneCkData($dataCount);
-
-    echo "\n##### dataCount: {$dataCount}, seletCount: {$seletCount}, limit: {$limit} #####\n";
-
-    $t0 = $t = start_test();
-    testPhpClickhouse($insertData, $seletCount, $limit);
-    $t = end_test($t, "PhpClickhouse");
-
-    testSeasClickNonCompression($insertData, $seletCount, $limit);
-    $t = end_test($t, "SeasClickNonCompression");
-
-    testSeasClickCompression($insertData, $seletCount, $limit);
-    $t = end_test($t, "SeasClickCompression");
-
-    testOneCk($insertOneCkData, $seletCount, $limit);
-    $t = end_test($t, "OneCk");
-
-    total($t0, "Total");
-}
-
-function start_test()
-{
-    return getmicrotime();
-}
-
-function getmicrotime()
-{
-    $t = gettimeofday();
-
-    return ($t['sec'] + $t['usec'] / 1000000);
-}
-
-function end_test($start, $name)
-{
-    global $total;
-    $end = getmicrotime();
-    $total += $end - $start;
-    $num = number_format($end - $start, 3);
-    $pad = str_repeat(" ", 60 - strlen($name) - strlen($num));
-
-    echo $name . $pad . $num . "\n";
-
-    return getmicrotime();
-}
-
-function total()
-{
-    global $total;
-    $pad = str_repeat("-", 32);
-    echo $pad . "\n";
-    $num = number_format($total, 3);
-    $pad = str_repeat(" ", 32 - strlen("Total") - strlen($num));
-    echo "Total" . $pad . $num . "\n";
-}
-
-function testSeasClickNonCompression($insertData, $num, $limit)
-{
-    $config = [
-        "host" => "clickhouse",
-        "port" => "9000",
-        "compression" => false
-    ];
-    
-    $db = new SeasClick($config);
-    $db->execute("CREATE DATABASE IF NOT EXISTS test");
-
-    $db->execute('
-        CREATE TABLE IF NOT EXISTS test.summing_url_views (
-            event_date Date DEFAULT toDate(event_time),
-            event_time DateTime,
-            site_id Int32,
-            site_key String,
-            views Int32,
-            v_00 Int32,
-            v_55 Int32
-        )
-        ENGINE = SummingMergeTree(event_date, (site_id, site_key, event_time, event_date), 8192)
-    ');
-
-    $db->insert("test.summing_url_views",
-        ['event_time', 'site_key', 'site_id', 'views', 'v_00', 'v_55'],
-        $insertData
-    );
-
-    $a = $num;
-    while ($a--) {
-        $db->select('SELECT * FROM test.summing_url_views LIMIT ' . $limit);
-    }
-
-    $db->execute("DROP TABLE {table}", [
-        'table' => 'test.summing_url_views'
-    ]);
-}
-
-function testSeasClickCompression($insertData, $num, $limit)
-{
-    $config = [
-        "host" => "clickhouse",
-        "port" => "9000",
-        "compression" => true
-    ];
-    
-    $db = new SeasClick($config);
-    $db->execute("CREATE DATABASE IF NOT EXISTS test");
-
-    $db->execute('
-        CREATE TABLE IF NOT EXISTS test.summing_url_views (
-            event_date Date DEFAULT toDate(event_time),
-            event_time DateTime,
-            site_id Int32,
-            site_key String,
-            views Int32,
-            v_00 Int32,
-            v_55 Int32
-        )
-        ENGINE = SummingMergeTree(event_date, (site_id, site_key, event_time, event_date), 8192)
-    ');
-
-    $db->insert("test.summing_url_views",
-        ['event_time', 'site_key', 'site_id', 'views', 'v_00', 'v_55'],
-        $insertData
-    );
-
-    $a = $num;
-    while ($a--) {
-        $db->select('SELECT * FROM test.summing_url_views LIMIT ' . $limit);
-    }
-
-    $db->execute("DROP TABLE {table}", [
-        'table' => 'test.summing_url_views'
-    ]);
-}
-
-function testPhpClickhouse($insertData, $num, $limit)
-{
-    $config = [
-        'host' => 'clickhouse',
-        'port' => '8123',
-        'username' => 'default',
-        'password' => ''
-    ];
-    $db = new ClickHouseDB\Client($config);
-    $db->write("CREATE DATABASE IF NOT EXISTS test");
-
-    $db->database('test');
-    $db->setTimeout(1.5);      // 1500 ms
-    $db->setTimeout(10);       // 10 seconds
-    $db->setConnectTimeOut(5); // 5 seconds
-    
-    $db->write('
-        CREATE TABLE IF NOT EXISTS summing_url_views (
-            event_date Date DEFAULT toDate(event_time),
-            event_time DateTime,
-            site_id Int32,
-            site_key String,
-            views Int32,
-            v_00 Int32,
-            v_55 Int32
-        )
-        ENGINE = SummingMergeTree(event_date, (site_id, site_key, event_time, event_date), 8192)
-    ');
-
-    $db->insert("summing_url_views",
-        $insertData,
-        ['event_time', 'site_key', 'site_id', 'views', 'v_00', 'v_55']
-    );
-
-    $a = $num;
-    while ($a--) {
-        $db->select('SELECT * FROM summing_url_views LIMIT ' . $limit)->rows();
-    }
-
-    $db->write('DROP TABLE IF EXISTS summing_url_views');
-}
-
-function testOneCk($insertData, $num, $limit)
-{
-    $db = new Client('tcp://clickhouse:9000', 'default', '', '');
-
-    $db->query('CREATE DATABASE IF NOT EXISTS test');
-
-    $db->query('
-        CREATE TABLE IF NOT EXISTS summing_url_views (
-            event_date Date DEFAULT toDate(event_time),
-            event_time DateTime,
-            site_id Int32,
-            site_key String,
-            views Int32,
-            v_00 Int32,
-            v_55 Int32
-        )
-        ENGINE = SummingMergeTree(event_date, (site_id, site_key, event_time, event_date), 8192)
-    ');
-    
-    $data['insert data'] = $db->insert('summing_url_views', $insertData);
-
-    $a = $num;
-    while ($a--) {
-        $db->query('SELECT * FROM summing_url_views LIMIT ' . $limit);
-    }
-
-    $db->query('DROP TABLE IF EXISTS summing_url_views');
-}
-
-function initOneCkData($num = 100)
-{
-    $insertData = [];
-    while ($num--) {
-        $insertData[] = [
-            'event_time' => time(),
-            'site_key' => 'HASH2',
-            'site_id' => 2345,
-            'views' => 12,
-            'v_00' => 9,
-            'v_55' => 3
-        ];
-    }
-    return $insertData;
-}
-
-function initData($num = 100)
-{
-    $insertData = [];
-    while ($num--) {
-        $insertData[] = [time(), 'HASH2', 2345, 12, 9,  3];
-    }
-    return $insertData;
-}
diff --git a/tests/bench_mark/composer.json b/tests/bench_mark/composer.json
deleted file mode 100644
index c4f5d4b..0000000
--- a/tests/bench_mark/composer.json
+++ /dev/null
@@ -1,6 +0,0 @@
-{
-    "require": {
-        "smi2/phpclickhouse": "^1.1",
-        "lizhichao/one-ck": "^0.1.8"
-    }
-}
diff --git a/tests/bench_mark/composer.lock b/tests/bench_mark/composer.lock
deleted file mode 100644
index 29b4517..0000000
--- a/tests/bench_mark/composer.lock
+++ /dev/null
@@ -1,128 +0,0 @@
-{
-    "_readme": [
-        "This file locks the dependencies of your project to a known state",
-        "Read more about it at https://getcomposer.org/doc/01-basic-usage.md#installing-dependencies",
-        "This file is @generated automatically"
-    ],
-    "content-hash": "5ce62d7da7b75f43d0dabe3f2766147a",
-    "packages": [
-        {
-            "name": "lizhichao/one-ck",
-            "version": "0.1.9",
-            "source": {
-                "type": "git",
-                "url": "https://github.com/lizhichao/one-ck.git",
-                "reference": "0b1de9b6f5527e81023a677e39aa0f62e2fc0193"
-            },
-            "dist": {
-                "type": "zip",
-                "url": "https://api.github.com/repos/lizhichao/one-ck/zipball/0b1de9b6f5527e81023a677e39aa0f62e2fc0193",
-                "reference": "0b1de9b6f5527e81023a677e39aa0f62e2fc0193",
-                "shasum": "",
-                "mirrors": [
-                    {
-                        "url": "https://mirrors.aliyun.com/composer/dists/%package%/%reference%.%type%",
-                        "preferred": true
-                    }
-                ]
-            },
-            "require": {
-                "ext-bcmath": "*",
-                "ext-json": "*",
-                "php": ">=7.1"
-            },
-            "type": "library",
-            "autoload": {
-                "psr-4": {
-                    "OneCk\\": "src/"
-                }
-            },
-            "notification-url": "https://packagist.org/downloads/",
-            "license": [
-                "Apache-2.0"
-            ],
-            "authors": [
-                {
-                    "name": "tanszhe",
-                    "email": "1018595261@qq.com"
-                }
-            ],
-            "description": "clickhouse tcp client",
-            "keywords": [
-                "clickhouse",
-                "client",
-                "php",
-                "tcp"
-            ],
-            "time": "2020-09-09T07:30:08+00:00"
-        },
-        {
-            "name": "smi2/phpclickhouse",
-            "version": "1.1.2",
-            "source": {
-                "type": "git",
-                "url": "https://github.com/smi2/phpClickHouse.git",
-                "reference": "ecf411b3b493e12bfe64873dac94456245d6c45c"
-            },
-            "dist": {
-                "type": "zip",
-                "url": "https://api.github.com/repos/smi2/phpClickHouse/zipball/ecf411b3b493e12bfe64873dac94456245d6c45c",
-                "reference": "ecf411b3b493e12bfe64873dac94456245d6c45c",
-                "shasum": "",
-                "mirrors": [
-                    {
-                        "url": "https://mirrors.aliyun.com/composer/dists/%package%/%reference%.%type%",
-                        "preferred": true
-                    }
-                ]
-            },
-            "require": {
-                "ext-curl": "*",
-                "php": ">=5.6"
-            },
-            "require-dev": {
-                "doctrine/coding-standard": "^4.0",
-                "php": "^7.1",
-                "phpstan/phpstan": "^0.9.2",
-                "phpunit/phpunit": "^7.0"
-            },
-            "type": "library",
-            "autoload": {
-                "psr-4": {
-                    "ClickHouseDB\\": "src/"
-                }
-            },
-            "notification-url": "https://packagist.org/downloads/",
-            "license": [
-                "MIT"
-            ],
-            "authors": [
-                {
-                    "name": "Igor Strykhar",
-                    "email": "isublimity@gmail.com",
-                    "homepage": "https://github.com/isublimity"
-                }
-            ],
-            "description": "PHP ClickHouse Client",
-            "homepage": "https://github.com/smi2/phpClickHouse",
-            "keywords": [
-                "clickhouse",
-                "client",
-                "curl",
-                "driver",
-                "http",
-                "http client",
-                "php"
-            ],
-            "time": "2018-07-03T18:04:38+00:00"
-        }
-    ],
-    "packages-dev": [],
-    "aliases": [],
-    "minimum-stability": "stable",
-    "stability-flags": [],
-    "prefer-stable": false,
-    "prefer-lowest": false,
-    "platform": [],
-    "platform-dev": []
-}
diff --git a/tests/test.php b/tests/test.php
deleted file mode 100644
index 2615093..0000000
--- a/tests/test.php
+++ /dev/null
@@ -1,332 +0,0 @@
-<?php
-/**
- * @author ahhhh.wang@gmail.com
- * Date: 19/02/6 8:23
- */
-
-$br = (php_sapi_name() == "cli")? "":"<br>";
-
-if(!extension_loaded('SeasClick')) {
-	dl('SeasClick.' . PHP_SHLIB_SUFFIX);
-}
-
-$config = [
-    "host" => "clickhouse",
-    "port" => "9000",
-    "compression" => true
-];
-
-clientTest($config);
-
-function clientTest($config)
-{
-    $deleteTable = true;
-    $client = new SeasClick($config);
-    $client->execute("CREATE DATABASE IF NOT EXISTS test");
-
-    testArray($client, $deleteTable);
-    testEnum($client, $deleteTable);
-    testString($client, $deleteTable);
-    testNullAble($client, $deleteTable);
-    testUInt($client, $deleteTable);
-    testFloat($client, $deleteTable);
-    testUUID($client, $deleteTable);
-    testDate($client, $deleteTable);
-    testMulInstance($config, $deleteTable);
-    testTuple($client, $deleteTable);
-}
-
-function testMulInstance($config, $deleteTable = false) {
-    $client1 = new SeasClick($config);
-    testArray($client1, $deleteTable);
-    
-    $client2 = new SeasClick($config);
-    testArray($client2, $deleteTable);
-    
-    $client3 = new SeasClick($config);
-    testArray($client3, $deleteTable);
-    
-    $client4 = new SeasClick($config);
-    testArray($client4, $deleteTable);
-
-    unset($client1);
-    unset($client2);
-    unset($client3);
-    unset($client4);
-}
-
-function testTuple($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.tuple_test (tuple_c Tuple(id UInt64, name String), int_c UInt64, string_c String) ENGINE = Memory");
-
-    $client->insert("test.tuple_test", [
-        'tuple_c', 'int_c', 'string_c'
-    ], [
-        [[1, 'one'], 1, 'one'],
-        [[2, 'two'], 2, 'two'],
-    ]);
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'tuple_c, int_c, string_c',
-        'table' => 'test.tuple_test'
-    ]);
-    var_dump($result);
-
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.tuple_test'
-        ]);
-    }
-}
-
-function testArray($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.array_test (string_c String, array_c Array(Int8), arraynull_c Array(Nullable(String))) ENGINE = Memory");
-
-    $client->writeStart("test.array_test", [
-        'string_c', 'arraynull_c'
-    ]);
-    $client->write([
-        [
-            'string_c2', ["string"]
-        ]
-    ]);
-    $client->writeEnd();
-
-    $client->insert("test.array_test", [
-        'string_c'
-    ], [
-        [
-            'string_c1'
-        ],
-        [
-            'string_c2'
-        ]
-    ]);
-
-    $client->insert("test.array_test", [
-        'array_c'
-    ], [
-        [
-            [1, 2, 3]
-        ],
-        [
-            [4, 5, 6]
-        ]
-    ]);
-
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'string_c, array_c, arraynull_c',
-        'table' => 'test.array_test'
-    ]);
-    var_dump($result);
-
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.array_test'
-        ]);
-    }
-}
-
-function testEnum($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.enum_test (enum8_c Enum8('One8' = 1, 'Two8' = 2), enum16_c Enum16('One16' = 1, 'Two16' = 2)) ENGINE = Memory");
-
-    $client->insert("test.enum_test", [
-        'enum8_c', 'enum16_c'
-    ],[
-        [1, 'Two16'],
-        ['Two8', 1]
-    ]);
-
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'enum8_c, enum16_c',
-        'table' => 'test.enum_test'
-    ]);
-    var_dump($result);
-
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.enum_test'
-        ]);
-    }
-}
-
-function testString($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.string_test (string_c String, fixedstring_c FixedString(50)) ENGINE = Memory");
-
-    $client->insert("test.string_test", [
-        'string_c', 'fixedstring_c'
-    ], [
-        [
-            'string_c1',
-            'fixedstring_c1'
-        ],
-        [
-            'string_c2',
-            'fixedstring_c2'
-        ]
-    ]);
-    
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'string_c, fixedstring_c',
-        'table' => 'test.string_test'
-    ]);
-    var_dump($result);
-    
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.string_test'
-        ]);
-    }
-}
-
-function testNullAble($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.nullable_test (int8null_c Nullable(Int8), stringnull_c Nullable(String), enumnull_c Nullable(Enum8('One8' = 1, 'Two8' = 2)), float32null_c Nullable(Float32), uuidnull_c Nullable(UUID)) ENGINE = Memory");
-
-    $client->insert("test.nullable_test",[
-        'int8null_c',
-        'stringnull_c',
-        'enumnull_c',
-        'float32null_c',
-        'uuidnull_c'
-    ], [
-        [8, 'string', 'One8', 32, '31249a1b7b0542709f37c609b48a9bb2'],
-        [null, null, null, null, null],
-    ]);
-
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'int8null_c, stringnull_c, enumnull_c, float32null_c, uuidnull_c',
-        'table' => 'test.nullable_test'
-    ]);
-    var_dump($result);
-    
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.nullable_test'
-        ]);
-    }
-}
-
-function testUInt($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.int_test (int8_c Int8, int16_c Int16, uint8_c UInt8, uint16_c UInt16) ENGINE = Memory");
-
-    $client->insert("test.int_test",[
-        'int8_c','int16_c','uint8_c','uint16_c'
-    ], [
-        [8, 8, 8, 8],
-        [9, 9, 9, 9],
-    ]);
-
-    $client->writeStart("test.int_test", [
-        'int8_c','int16_c','uint8_c'
-    ]);
-    $client->write([
-        [8, 8, 8],
-        [9, 9, 9],
-    ]);
-    $client->writeEnd();
-    
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'int8_c, int16_c, uint8_c, uint16_c',
-        'table' => 'test.int_test'
-    ]);
-    var_dump($result);
-    
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.int_test'
-        ]);
-    }
-}
-
-function testFloat($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.float_test (float32_c Float32, float64_c Float64) ENGINE = Memory");
-
-    $client->insert("test.float_test",[
-        'float32_c', 'float64_c'
-    ], [
-        [32.32, 64.64],
-        [32.31, 64.68]
-    ]);
-
-    $client->writeStart("test.float_test", [
-        'float32_c'
-    ]);
-    $client->write([
-        [32.32],
-        [32.31]
-    ]);
-    $client->writeEnd();
-    
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'float32_c, float64_c',
-        'table' => 'test.float_test'
-    ]);
-    var_dump($result);
-    
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.float_test'
-        ]);
-    }
-}
-
-/**
- * testUUID
- * You can insert a UUID with -, but the query is definitely not -
- */
-function testUUID($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.uuid_test (uuid_c UUID, uuid2_c UUID) ENGINE = Memory");
-
-    $client->insert("test.uuid_test",[
-        'uuid_c', 'uuid2_c'
-    ], [
-        ['31249a1b-7b05-4270-9f37-c609b48a9bb2', '31249a1b7b0542709f37c609b48a9bb2'],
-        ['31249a1b-7b05-4270-9f37-c609b48a9bb2', null],
-    ]);
-
-    $client->writeStart("test.uuid_test", [
-        'uuid_c'
-    ]);
-    $client->write([
-        ['00000000-0000-0000-9f37-c609b48a9bb2'],
-        ['31249a1b-7b05-4270-9f37-c609b48a9bb2'],
-    ]);
-    $client->writeEnd();
-    
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'uuid_c, uuid2_c',
-        'table' => 'test.uuid_test'
-    ]);
-    var_dump($result);
-    
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.uuid_test'
-        ]);
-    }
-}
-
-/**
- * testDate
- * Regardless of whether the type is date or datetime is inserted as a timestamp, the query is also a timestamp.
- */
-function testDate($client, $deleteTable = false) {
-    $client->execute("CREATE TABLE IF NOT EXISTS test.date_test (date_c Date, datetime_c DateTime) ENGINE = Memory");
-
-    $client->insert("test.date_test", [
-        'date_c', 'datetime_c'
-    ], [
-        [time(), time()],
-        [time(), time()]
-    ]);
-    
-    $result = $client->select("SELECT {select} FROM {table}", [
-        'select' => 'date_c, datetime_c',
-        'table' => 'test.date_test'
-    ]);
-    var_dump($result);
-    
-    if ($deleteTable) {
-        $client->execute("DROP TABLE {table}", [
-            'table' => 'test.date_test'
-        ]);
-    }
-}
diff --git a/travis/run-tests.sh b/travis/run-tests.sh
deleted file mode 100755
index 1df9a84..0000000
--- a/travis/run-tests.sh
+++ /dev/null
@@ -1,24 +0,0 @@
-#!/bin/bash
-TEST_DIR="`pwd`/tests/"
-
-make test
-
-for file in `find $TEST_DIR -name "*.diff" 2>/dev/null`
-do
-	grep "\-\-XFAIL--" ${file/%diff/phpt} >/dev/null 2>&1
-	if [ $? -gt 0 ]
-	then
-		FAILS[${#FAILS[@]}]="$file"
-	fi
-done
-
-if [ ${#FAILS[@]} -gt 0 ]
-then
-	for fail in "${FAILS[@]}"
-	do
-		sh -xc "cat $fail"
-	done
-	exit 1
-else
-	exit 0
-fi
\ No newline at end of file
diff --git a/typesToPhp.cpp b/typesToPhp.cpp
index 2823d36..9c795b0 100644
--- a/typesToPhp.cpp
+++ b/typesToPhp.cpp
@@ -1,8 +1,8 @@
 /*
   +----------------------------------------------------------------------+
-  | SeasClick                                                            |
+  | php_clickhouse                                                       |
   +----------------------------------------------------------------------+
-  | Copyright (c) 1997-2018 The PHP Group                                |
+  | Copyright (c) 1997-2026 The PHP Group                                |
   +----------------------------------------------------------------------+
   | This source file is subject to version 3.01 of the PHP license,      |
   | that is bundled with this package in the file LICENSE, and is        |
@@ -12,7 +12,7 @@
   | obtain it through the world-wide-web, please send a note to          |
   | license@php.net so we can mail you a copy immediately.               |
   +----------------------------------------------------------------------+
-  | Author:  SeasX Group <ahhhh.wang@gmail.com>                          |
+  | Author: Ilia Alshanetsky <ilia@ilia.ws>                              |
   +----------------------------------------------------------------------+
 */
 #ifdef HAVE_CONFIG_H
@@ -26,24 +26,532 @@ extern "C" {
 #include "php7_wrapper.h"
 };
 
-#include "php_SeasClick.h"
+#include "php_clickhouse.h"
 
 #include "lib/clickhouse-cpp/clickhouse/client.h"
 #include "lib/clickhouse-cpp/clickhouse/error_codes.h"
 #include "lib/clickhouse-cpp/clickhouse/types/type_parser.h"
-#include <iostream>
-#include <map>
+#include "lib/clickhouse-cpp/clickhouse/columns/factory.h"
+#include "lib/clickhouse-cpp/clickhouse/columns/geo.h"
+#include "lib/clickhouse-cpp/clickhouse/columns/ip4.h"
+#include "lib/clickhouse-cpp/clickhouse/columns/ip6.h"
+#include "lib/clickhouse-cpp/clickhouse/columns/lowcardinality.h"
+#include "lib/clickhouse-cpp/clickhouse/columns/map.h"
 #include <sstream>
 #include <iomanip>
 #include <algorithm>
+#include <cmath>
+#include <cerrno>
+#include <cinttypes>
 
 #include "typesToPhp.hpp"
 
 using namespace clickhouse;
 using namespace std;
 
+/*
+ * Format a 128-bit integer into a decimal string using a stack buffer.
+ * Avoids the heap allocation a stringstream incurs per cell on the read
+ * paths for Int128 / UInt128 / Decimal columns. Returns the number of
+ * bytes written into `out` (which must hold at least 41 bytes: sign +
+ * 39 digits + NUL margin).
+ */
+static size_t format_uint128_dec(absl::uint128 v, char *out)
+{
+    char tmp[40];
+    int len = 0;
+    do {
+        tmp[len++] = (char)('0' + (int)(v % 10));
+        v /= 10;
+    } while (v != 0);
+    for (int i = 0; i < len; ++i) {
+        out[i] = tmp[len - 1 - i];
+    }
+    return (size_t)len;
+}
+
+static size_t format_int128_dec(absl::int128 v, char *out)
+{
+    if (v >= 0) {
+        return format_uint128_dec(absl::uint128(v), out);
+    }
+    out[0] = '-';
+    /* Use unsigned negation to handle INT128_MIN (whose magnitude is
+     * representable in uint128 but not in int128). */
+    absl::uint128 mag = absl::uint128(0) - absl::uint128(v);
+    return 1 + format_uint128_dec(mag, out + 1);
+}
+
+/*
+ * Parse a decimal-digit string into a 128-bit unsigned. Both ColumnInt128
+ * and ColumnUInt128 inserts share the body; the signed wrapper composes
+ * with negate handled by the caller. Throws on overflow / non-digit.
+ *
+ * `out_label` is the type name for the error message ("Int128" / "UInt128").
+ */
+static absl::uint128 parse_uint128_dec(const char *s, size_t len, const char *out_label)
+{
+    if (len == 0 || len > 39) {
+        throw std::runtime_error(std::string(out_label) + " string is empty or too long");
+    }
+    absl::uint128 v = 0;
+    for (size_t i = 0; i < len; ++i) {
+        if (s[i] < '0' || s[i] > '9') {
+            throw std::runtime_error(std::string(out_label) + " string contains non-digit characters");
+        }
+        absl::uint128 next = v * 10 + absl::uint128((unsigned)(s[i] - '0'));
+        if (next < v) {
+            throw std::runtime_error(std::string(out_label) + " string overflows the 128-bit range");
+        }
+        v = next;
+    }
+    return v;
+}
+
+/*
+ * dynamic_pointer_cast helper that throws a contextual error when the
+ * cast returns null, instead of leaving the caller to deref nullptr.
+ * The clickhouse-cpp Block schema metadata is server-supplied; a
+ * mismatch between the declared type code and the actual ColumnRef
+ * concrete type used to crash the worker. Callers in convertToZval
+ * (especially Map / Tuple decoders) wrap every typed cast through here.
+ */
+template <typename TCol>
+static inline std::shared_ptr<TCol> as_or_throw(const ColumnRef &c, const char *what)
+{
+    auto p = c->As<TCol>();
+    if (!p) {
+        throw std::runtime_error(std::string(what) + ": column type mismatch");
+    }
+    return p;
+}
+
+/*
+ * Strict numeric coercion for INSERT cells. PHP's `zval_get_long` and
+ * `zval_get_double` happily produce 0 / 0.0 for non-numeric strings,
+ * arrays, objects, etc., which used to land "abc" as 0 in an Int32
+ * column with no diagnostic. The strict variants below reject every
+ * non-numeric input and require full string consumption, mirroring
+ * the strict parsers we already use for Map keys (CR-306) and hex
+ * literals (CR-508). Range-checking against the destination column's
+ * width is still the caller's responsibility (appendIntColumn passes
+ * MinV/MaxV, narrow-int Map dispatch wraps with its own checks).
+ *
+ * IS_NULL handling: rejected by default (storing 0 silently corrupts
+ * non-Nullable columns). The Nullable insert path bumps
+ * `g_allow_null_in_strict` via AllowNullGuard so its recursive child
+ * build can accept NULL cells (the null mask makes the placeholder
+ * value irrelevant).
+ */
+static thread_local int g_allow_null_in_strict = 0;
+struct AllowNullGuard {
+    AllowNullGuard()  { ++g_allow_null_in_strict; }
+    ~AllowNullGuard() { --g_allow_null_in_strict; }
+    AllowNullGuard(const AllowNullGuard&) = delete;
+    AllowNullGuard& operator=(const AllowNullGuard&) = delete;
+};
+static zend_long strict_zval_long(zval *z, const char *type_label)
+{
+    switch (Z_TYPE_P(z)) {
+        case IS_LONG:  return Z_LVAL_P(z);
+        case IS_TRUE:  return 1;
+        case IS_FALSE: return 0;
+        case IS_NULL:
+            if (g_allow_null_in_strict > 0) return 0;
+            throw std::runtime_error(
+                std::string("null cannot be assigned to non-Nullable column ") + type_label);
+        case IS_DOUBLE: {
+            double d = Z_DVAL_P(z);
+            if (std::isnan(d) || std::isinf(d)) {
+                throw std::runtime_error(
+                    std::string("non-finite double cannot be assigned to ") + type_label);
+            }
+            double frac, intpart;
+            frac = std::modf(d, &intpart);
+            if (frac != 0.0) {
+                throw std::runtime_error(
+                    std::string("fractional double cannot be assigned to integer column ") + type_label);
+            }
+            if (d < (double)ZEND_LONG_MIN || d > (double)ZEND_LONG_MAX) {
+                throw std::runtime_error(
+                    std::string("double out of range for integer column ") + type_label);
+            }
+            return (zend_long)d;
+        }
+        case IS_STRING: {
+            const char *s = Z_STRVAL_P(z);
+            size_t slen = Z_STRLEN_P(z);
+            if (slen == 0) {
+                throw std::runtime_error(
+                    std::string("empty string cannot be assigned to ") + type_label);
+            }
+            char *endp = NULL;
+            errno = 0;
+            long long v = strtoll(s, &endp, 10);
+            if (errno == ERANGE || endp == s ||
+                (size_t)(endp - s) != slen) {
+                throw std::runtime_error(
+                    std::string("invalid integer string for ") + type_label);
+            }
+            return (zend_long)v;
+        }
+        default:
+            throw std::runtime_error(
+                std::string("array / object / resource cannot be assigned to integer column ") + type_label);
+    }
+}
+
+/* UInt64 needs a strict parser of its own because strict_zval_long
+ * tops out at ZEND_LONG_MAX (2^63-1): values above that arrive as
+ * decimal strings (PHP can't fit them in a zend_long) and must be
+ * parsed via strtoull, not strtoll. Same shape as strict_zval_long
+ * — full-consumption check, NULL handled under AllowNullGuard,
+ * fractional / non-finite doubles rejected — but with the unsigned
+ * range and an additional `0x` hex form. */
+static uint64_t strict_zval_u64(zval *z, const char *type_label)
+{
+    switch (Z_TYPE_P(z)) {
+        case IS_LONG: {
+            zend_long n = Z_LVAL_P(z);
+            if (n < 0) {
+                throw std::runtime_error(
+                    std::string("negative value cannot fit in ") + type_label);
+            }
+            return (uint64_t)n;
+        }
+        case IS_TRUE:  return 1;
+        case IS_FALSE: return 0;
+        case IS_NULL:
+            if (g_allow_null_in_strict > 0) return 0;
+            throw std::runtime_error(
+                std::string("null cannot be assigned to non-Nullable column ") + type_label);
+        case IS_DOUBLE: {
+            double d = Z_DVAL_P(z);
+            if (std::isnan(d) || std::isinf(d)) {
+                throw std::runtime_error(
+                    std::string("non-finite double cannot be assigned to ") + type_label);
+            }
+            double frac, intpart;
+            frac = std::modf(d, &intpart);
+            if (frac != 0.0) {
+                throw std::runtime_error(
+                    std::string("fractional double cannot be assigned to integer column ") + type_label);
+            }
+            /* 18446744073709551616.0 is the next representable double
+             * above 2^64. Anything >= it overflows uint64_t. Negatives
+             * are rejected explicitly. */
+            if (d < 0.0 || d >= 18446744073709551616.0) {
+                throw std::runtime_error(
+                    std::string("double out of range for integer column ") + type_label);
+            }
+            return (uint64_t)d;
+        }
+        case IS_STRING: {
+            const char *s = Z_STRVAL_P(z);
+            size_t slen = Z_STRLEN_P(z);
+            if (slen == 0) {
+                throw std::runtime_error(
+                    std::string("empty string cannot be assigned to ") + type_label);
+            }
+            int base = 10;
+            const char *p = s;
+            size_t plen = slen;
+            if (slen >= 3 && s[0] == '0' && (s[1] == 'x' || s[1] == 'X')) {
+                base = 16;
+                p = s + 2;
+                plen = slen - 2;
+            }
+            if (*p == '-' || *p == '+') {
+                throw std::runtime_error(
+                    std::string("invalid integer string for ") + type_label);
+            }
+            char *endp = NULL;
+            errno = 0;
+            unsigned long long v = strtoull(p, &endp, base);
+            if (errno == ERANGE || endp == p ||
+                (size_t)(endp - p) != plen) {
+                throw std::runtime_error(
+                    std::string("invalid integer string for ") + type_label);
+            }
+            return (uint64_t)v;
+        }
+        default:
+            throw std::runtime_error(
+                std::string("array / object / resource cannot be assigned to integer column ") + type_label);
+    }
+}
+
+static double strict_zval_double(zval *z, const char *type_label)
+{
+    switch (Z_TYPE_P(z)) {
+        case IS_LONG:  return (double)Z_LVAL_P(z);
+        case IS_TRUE:  return 1.0;
+        case IS_FALSE: return 0.0;
+        case IS_NULL:
+            if (g_allow_null_in_strict > 0) return 0.0;
+            throw std::runtime_error(
+                std::string("null cannot be assigned to non-Nullable column ") + type_label);
+        case IS_DOUBLE: {
+            double d = Z_DVAL_P(z);
+            if (std::isnan(d) || std::isinf(d)) {
+                throw std::runtime_error(
+                    std::string("non-finite double cannot be assigned to ") + type_label);
+            }
+            return d;
+        }
+        case IS_STRING: {
+            const char *s = Z_STRVAL_P(z);
+            size_t slen = Z_STRLEN_P(z);
+            if (slen == 0) {
+                throw std::runtime_error(
+                    std::string("empty string cannot be assigned to ") + type_label);
+            }
+            char *endp = NULL;
+            errno = 0;
+            double v = strtod(s, &endp);
+            if (errno == ERANGE || endp == s ||
+                (size_t)(endp - s) != slen ||
+                std::isnan(v) || std::isinf(v)) {
+                throw std::runtime_error(
+                    std::string("invalid float string for ") + type_label);
+            }
+            return v;
+        }
+        default:
+            throw std::runtime_error(
+                std::string("array / object / resource cannot be assigned to float column ") + type_label);
+    }
+}
+
+static std::string strict_zval_string(zval *z, const char *type_label)
+{
+    if (Z_TYPE_P(z) == IS_NULL) {
+        if (g_allow_null_in_strict > 0) return std::string();
+        throw std::runtime_error(
+            std::string("null cannot be assigned to non-Nullable column ") + type_label);
+    }
+    ZStrGuard sg(z);
+    return std::string(sg.val(), sg.len());
+}
+
+static int uuid_hex_value(char c)
+{
+    if (c >= '0' && c <= '9') return c - '0';
+    if (c >= 'a' && c <= 'f') return c - 'a' + 10;
+    if (c >= 'A' && c <= 'F') return c - 'A' + 10;
+    return -1;
+}
+
+static UUID parseUUIDString(const char *s, size_t len, const char *error_msg)
+{
+    uint64_t high = 0;
+    uint64_t low = 0;
+    size_t digits = 0;
+
+    for (size_t i = 0; i < len; ++i) {
+        if (s[i] == '-') {
+            continue;
+        }
+        int nibble = uuid_hex_value(s[i]);
+        if (nibble < 0 || digits >= 32) {
+            throw std::runtime_error(error_msg);
+        }
+        if (digits < 16) {
+            high = (high << 4) | (uint64_t)nibble;
+        } else {
+            low = (low << 4) | (uint64_t)nibble;
+        }
+        ++digits;
+    }
+
+    if (digits != 32) {
+        throw std::runtime_error(error_msg);
+    }
+    return UUID{high, low};
+}
+
+/*
+ * RAII guard that owns a zend_string* obtained from zval_get_string and
+ * releases it in the destructor. Used at PHP-to-C boundaries where the
+ * surrounding code can throw (validation, recursive insertColumn, etc.)
+ * without forcing every site to write try { ... } catch { release; throw; }.
+ */
+/*
+ * Extract the width from a "FixedString(N)" type name. The previous
+ * inline form did `typeName.erase(typeName.find("FixedString("), 12)` —
+ * if find() returned npos, erase(npos, 12) is undefined. This helper
+ * validates the prefix and parses the digit run.
+ */
+static int parseFixedStringWidth(TypeRef type)
+{
+    const std::string &name = type->GetName();
+    static const char prefix[] = "FixedString(";
+    static const size_t prefix_len = sizeof(prefix) - 1;
+    if (name.size() < prefix_len + 2 ||
+        name.compare(0, prefix_len, prefix) != 0 ||
+        name.back() != ')') {
+        throw std::runtime_error("Invalid FixedString type name: " + name);
+    }
+    const char *p = name.c_str() + prefix_len;
+    char *endp = nullptr;
+    errno = 0;
+    long w = strtol(p, &endp, 10);
+    if (errno == ERANGE || endp == p || w <= 0 || w > INT_MAX || endp != name.c_str() + name.size() - 1) {
+        throw std::runtime_error("Invalid FixedString width: " + name);
+    }
+    return (int)w;
+}
+
+/*
+ * Parse "YYYY-MM-DD" or "YYYY-MM-DD HH:MM:SS" into a Unix epoch.
+ *
+ * Use timegm, not mktime: the read paths in convertToZval all format
+ * via gmtime, so the round-trip needs to be UTC symmetric. Using
+ * mktime would reinterpret the parsed components as local time and
+ * shift the stored value by the runner's TZ offset.
+ */
+static std::time_t to_time_t(const std::string& str, bool is_date = true)
+{
+    std::tm t = {};
+    std::istringstream ss(str);
+    ss >> std::get_time(&t, is_date ? "%Y-%m-%d" : "%Y-%m-%d %H:%M:%S");
+    if (ss.fail()) {
+        /* Reject malformed date/datetime strings instead of silently
+         * coercing to (time_t)-1, which then read back as 1969-12-31. */
+        throw std::runtime_error(
+            std::string("Invalid ") + (is_date ? "Date" : "DateTime") +
+            " string: " + str);
+    }
+    /* std::get_time stops at the first non-matching character without
+     * raising failbit, so "2024-01-01abc" parses cleanly as 2024-01-01.
+     * Require EOF after the expected format so trailing garbage is
+     * rejected at the boundary. peek() returns EOF when the stream is
+     * fully consumed; anything else is an extra character we didn't
+     * sign up for. */
+    if (ss.peek() != std::char_traits<char>::eof()) {
+        throw std::runtime_error(
+            std::string("Invalid ") + (is_date ? "Date" : "DateTime") +
+            " string (trailing characters): " + str);
+    }
+    /* Capture the parsed components before timegm so we can detect the
+     * normalization that February 30 → March 2 silently performs. */
+    int p_year = t.tm_year, p_mon = t.tm_mon, p_mday = t.tm_mday,
+        p_hour = t.tm_hour, p_min = t.tm_min, p_sec = t.tm_sec;
+#ifdef _WIN32
+    /* MSVC has no timegm(); _mkgmtime() is the documented equivalent. */
+    std::time_t out = _mkgmtime(&t);
+#else
+    std::time_t out = timegm(&t);
+#endif
+    if (out == (std::time_t)-1) {
+        throw std::runtime_error(
+            std::string("Invalid ") + (is_date ? "Date" : "DateTime") +
+            " string (out of range): " + str);
+    }
+    /* Round-trip-validate: convert back to UTC components and compare.
+     * This catches "2024-02-30" → 2024-03-01 normalization that timegm
+     * does silently. */
+    std::tm rt = {};
+#ifdef _WIN32
+    if (gmtime_s(&rt, &out) != 0) {
+#else
+    if (!gmtime_r(&out, &rt)) {
+#endif
+        throw std::runtime_error(
+            std::string("Invalid ") + (is_date ? "Date" : "DateTime") +
+            " string (gmtime failed): " + str);
+    }
+    if (rt.tm_year != p_year || rt.tm_mon != p_mon || rt.tm_mday != p_mday ||
+        rt.tm_hour != p_hour || rt.tm_min != p_min || rt.tm_sec != p_sec) {
+        throw std::runtime_error(
+            std::string("Invalid ") + (is_date ? "Date" : "DateTime") +
+            " string (normalized to a different value): " + str);
+    }
+    return out;
+}
+
+/*
+ * Parse "YYYY-MM-DD HH:MM:SS[.ffffff...]" into (whole-seconds, fractional)
+ * pair. The fractional component is multiplied by 10^precision so the
+ * caller can encode straight into ColumnDateTime64. The previous insert
+ * path used to_time_t alone, which dropped any sub-second part of the
+ * string entirely.
+ */
+static std::pair<std::time_t, int64_t> to_time_t_with_frac(const std::string &str, size_t precision, int64_t scale)
+{
+    auto dot = str.find('.');
+    std::time_t whole = to_time_t(dot == std::string::npos ? str : str.substr(0, dot), false);
+    int64_t frac = 0;
+    if (dot != std::string::npos) {
+        if (precision == 0) {
+            /* DateTime64(0) has no fractional component; any text after
+             * the dot is invalid. The prior pass silently dropped the
+             * suffix at precision 0, which let "00:00:00.garbage" land
+             * as a clean DateTime64(0). */
+            throw std::runtime_error(
+                "Invalid DateTime64(0) string (fractional suffix on a "
+                "zero-precision column): " + str);
+        }
+        const char *p = str.c_str() + dot + 1;
+        const char *end = str.c_str() + str.size();
+        if (p >= end) {
+            /* Bare "12:34:56." with no digits after the dot. Previously
+             * accepted (consumed=0 took the no-op path). Reject. */
+            throw std::runtime_error(
+                "Invalid DateTime64 string (bare dot without fractional digits): " + str);
+        }
+        size_t consumed = 0;
+        while (p < end && consumed < precision && *p >= '0' && *p <= '9') {
+            frac = frac * 10 + (*p - '0');
+            ++p;
+            ++consumed;
+        }
+        if (consumed == 0) {
+            /* The first character after the dot wasn't a digit. */
+            throw std::runtime_error(
+                "Invalid DateTime64 string (non-digit after dot): " + str);
+        }
+        // Pad missing digits up to precision so "12:34:56.5" with precision 3
+        // contributes 500 (ms), not 5.
+        for (size_t pad = consumed; pad < precision; ++pad) frac *= 10;
+        /* Reject trailing non-digit characters after the fractional part.
+         * Without this "2024-01-01 00:00:00.123abc" silently truncated to
+         * .123 and dropped the abc. */
+        if (p < end) {
+            throw std::runtime_error(
+                "Invalid DateTime64 string (trailing characters after fraction): " + str);
+        }
+    } else {
+        frac = 0;
+    }
+    (void)scale;
+    return {whole, frac};
+}
+
+/* Cap recursion through nested column types (Tuple/Array/Map/Nullable/
+ * LowCardinality) so a server-supplied schema like Tuple(Tuple(Tuple(...)))
+ * cannot stack-overflow the worker. Used by both the read path
+ * (convertToZval) and the write path (createColumn / insertColumn);
+ * BeginInsert returns a server-built block schema, so an adversarial or
+ * MITM'd server can craft a deeply-nested type just like on the read side.
+ *
+ * thread_local because clickhouse-cpp may dispatch from worker threads. */
+static thread_local int convert_depth = 0;
+static const int MAX_CONVERT_DEPTH = 32;
+
+struct ConvertDepthGuard {
+    ConvertDepthGuard() {
+        if (++convert_depth > MAX_CONVERT_DEPTH) {
+            --convert_depth;
+            throw std::runtime_error("ClickHouse column nested-type depth exceeds limit");
+        }
+    }
+    ~ConvertDepthGuard() { --convert_depth; }
+};
+
 ColumnRef createColumn(TypeRef type)
 {
+    ConvertDepthGuard _depth_guard;
     switch (type->GetCode())
     {
     case Type::Code::UInt64:
@@ -100,66 +608,119 @@ ColumnRef createColumn(TypeRef type)
     }
     case Type::Code::FixedString:
     {
-        string typeName = type->GetName();
-        typeName.erase(typeName.find("FixedString("), 12);
-        typeName.erase(typeName.find(")"), 1);
-        return std::make_shared<ColumnFixedString>(std::stoi(typeName));
+        return std::make_shared<ColumnFixedString>(parseFixedStringWidth(type));
     }
 
     case Type::Code::DateTime:
     {
         return std::make_shared<ColumnDateTime>();
     }
+    case Type::Code::DateTime64:
+    {
+        return std::make_shared<ColumnDateTime64>(type->As<DateTime64Type>()->GetPrecision());
+    }
     case Type::Code::Date:
     {
         return std::make_shared<ColumnDate>();
     }
+    case Type::Code::Date32:
+    {
+        return std::make_shared<ColumnDate32>();
+    }
+    case Type::Code::Time:
+    {
+        return std::make_shared<ColumnTime>();
+    }
+    case Type::Code::Time64:
+    {
+        return std::make_shared<ColumnTime64>(type->As<Time64Type>()->GetPrecision());
+    }
+    case Type::Code::Int128:
+    {
+        return std::make_shared<ColumnInt128>();
+    }
+    case Type::Code::UInt128:
+    {
+        return std::make_shared<ColumnUInt128>();
+    }
+    case Type::Code::Decimal:
+    case Type::Code::Decimal32:
+    case Type::Code::Decimal64:
+    case Type::Code::Decimal128:
+    {
+        auto dt = type->As<DecimalType>();
+        return std::make_shared<ColumnDecimal>(dt->GetPrecision(), dt->GetScale());
+    }
 
     case Type::Code::Array:
     {
-        return std::make_shared<ColumnArray>(createColumn(type->GetItemType()));
+        return std::make_shared<ColumnArray>(createColumn(type->As<ArrayType>()->GetItemType()));
     }
 
     case Type::Code::Enum8:
     {
-        std::vector<Type::EnumItem> enum_items;
-
-        auto enumType = EnumType(type);
-
-        for (auto ei = enumType.BeginValueToName(); ; )
-        {
-            enum_items.push_back(
-                Type::EnumItem {ei->second, (int8_t)ei->first});
-            if (++ei == enumType.EndValueToName())
-            {
-                break;
-            }
-        }
-
-        return std::make_shared<ColumnEnum8>(Type::CreateEnum8(enum_items));
+        return std::make_shared<ColumnEnum8>(type);
     }
     case Type::Code::Enum16:
     {
-        std::vector<Type::EnumItem> enum_items;
+        return std::make_shared<ColumnEnum16>(type);
+    }
 
-        auto enumType = EnumType(type);
+    case Type::Code::Nullable:
+    {
+        return std::make_shared<ColumnNullable>(createColumn(type->As<NullableType>()->GetNestedType()), std::make_shared<ColumnUInt8>());
+    }
 
-        for (auto ei = enumType.BeginValueToName(); ; )
-        {
-            enum_items.push_back(
-                Type::EnumItem {ei->second, (int16_t)ei->first});
-            if (++ei == enumType.EndValueToName())
-            {
-                break;
+    case Type::Code::LowCardinality:
+    {
+        TypeRef nested = type->As<LowCardinalityType>()->GetNestedType();
+        bool is_nullable = (nested->GetCode() == Type::Code::Nullable);
+        TypeRef inner = is_nullable
+            ? nested->As<NullableType>()->GetNestedType()
+            : nested;
+        if (inner->GetCode() == Type::Code::String) {
+            if (is_nullable) {
+                return std::make_shared<ColumnLowCardinalityT<ColumnNullableT<ColumnString>>>();
             }
+            return std::make_shared<ColumnLowCardinalityT<ColumnString>>();
         }
-
-        return std::make_shared<ColumnEnum16>(Type::CreateEnum16(enum_items));
+        if (inner->GetCode() == Type::Code::FixedString) {
+            int width = parseFixedStringWidth(inner);
+            if (is_nullable) {
+                return std::make_shared<ColumnLowCardinalityT<ColumnNullableT<ColumnFixedString>>>(width);
+            }
+            return std::make_shared<ColumnLowCardinalityT<ColumnFixedString>>(width);
+        }
+        throw std::runtime_error("LowCardinality only supported over String / FixedString (Nullable allowed)");
     }
 
-    case Type::Code::Nullable:
+    case Type::Code::Map:
     {
-        return std::make_shared<ColumnNullable>(createColumn(type->GetNestedType()), std::make_shared<ColumnUInt8>());
+        TypeRef k = type->As<MapType>()->GetKeyType();
+        TypeRef v = type->As<MapType>()->GetValueType();
+        Type::Code kc = k->GetCode();
+        Type::Code vc = v->GetCode();
+        if (kc == Type::Code::String && vc == Type::Code::String) {
+            return std::make_shared<ColumnMapT<ColumnString, ColumnString>>(
+                std::make_shared<ColumnString>(), std::make_shared<ColumnString>());
+        }
+        if (kc == Type::Code::String && vc == Type::Code::Int64) {
+            return std::make_shared<ColumnMapT<ColumnString, ColumnInt64>>(
+                std::make_shared<ColumnString>(), std::make_shared<ColumnInt64>());
+        }
+        if (kc == Type::Code::String && vc == Type::Code::UInt64) {
+            return std::make_shared<ColumnMapT<ColumnString, ColumnUInt64>>(
+                std::make_shared<ColumnString>(), std::make_shared<ColumnUInt64>());
+        }
+        if (kc == Type::Code::String && vc == Type::Code::Float64) {
+            return std::make_shared<ColumnMapT<ColumnString, ColumnFloat64>>(
+                std::make_shared<ColumnString>(), std::make_shared<ColumnFloat64>());
+        }
+        if (kc == Type::Code::Int64 && vc == Type::Code::String) {
+            return std::make_shared<ColumnMapT<ColumnInt64, ColumnString>>(
+                std::make_shared<ColumnInt64>(), std::make_shared<ColumnString>());
+        }
+        return CreateColumnByType(type->GetName());
     }
 
     case Type::Code::Tuple:
@@ -171,681 +732,1581 @@ ColumnRef createColumn(TypeRef type)
     {
         throw std::runtime_error("can't support Void");
     }
-    }
 
-    throw std::runtime_error("createColumn runtime error.");
+    default:
+        return CreateColumnByType(type->GetName());
+    }
 }
 
-ColumnRef insertColumn(TypeRef type, zval *value_zval)
+// Build a column of plain integer cells from a PHP rows array. Used by
+// every signed and unsigned integer type that doesn't accept hex
+// strings (UInt8/16, Int8..Int64).
+//
+// MinV/MaxV bound the destination column's representable range so that an
+// out-of-range PHP value throws instead of silently wrapping in the
+// narrowing assignment to ClickHouse's int8/int16/int32. Values are
+// pulled non-mutatingly via zval_get_long so the caller's row arrays
+// don't get their types coerced in place.
+template <typename TCol>
+static ColumnRef appendIntColumn(HashTable *values_ht,
+                                 zend_long MinV, zend_long MaxV,
+                                 const char *type_label)
 {
+    auto value = std::make_shared<TCol>();
     zval *array_value;
-    char *str_key;
-    uint32_t str_keylen;
-    int keytype;
-
-    HashTable *values_ht = Z_ARRVAL_P(value_zval);
-
-    switch (type->GetCode())
-    {
-    case Type::Code::UInt64:
-    {
-        auto value = std::make_shared<ColumnUInt64>();
+    ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+        zend_long n = strict_zval_long(array_value, type_label);
+        if (n < MinV || n > MaxV) {
+            throw std::runtime_error(
+                std::string("value out of range for ") + type_label);
+        }
+        value->Append((typename TCol::ValueType)n);
+    } ZEND_HASH_FOREACH_END();
+    return value;
+}
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
+// Build an unsigned integer column with a hex-string fast path. UInt32
+// and UInt64 both accept "0x..." strings as a way to land values in the
+// upper half of the range that a PHP signed long can't represent.
+// `MaxV` bounds the destination column width: strtoul on 64-bit Linux
+// returns 64-bit values regardless of the target column, so without a
+// width check "0x100000000" silently truncated to UInt32 0.
+template <typename TCol, typename TStrtoul>
+static ColumnRef appendUIntColumnWithHex(HashTable *values_ht,
+                                         TStrtoul strtoul_fn,
+                                         uint64_t MaxV,
+                                         const char *type_label)
+{
+    auto value = std::make_shared<TCol>();
+    zval *array_value;
+    ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+        if (Z_TYPE_P(array_value) == IS_STRING && Z_STRLEN_P(array_value) >= 3 &&
+            *Z_STRVAL_P(array_value) == '0' &&
+            (*(Z_STRVAL_P(array_value) + 1) == 'x' || *(Z_STRVAL_P(array_value) + 1) == 'X')) {
+            const char *s = Z_STRVAL_P(array_value);
+            size_t slen = Z_STRLEN_P(array_value);
+            char *endp = NULL;
+            errno = 0;
+            auto n = strtoul_fn(s, &endp, 0);
+            /* PHP zend_string is length-prefixed and may carry embedded
+             * NUL bytes. Comparing endp against ZSTR_LEN is the right
+             * "fully consumed" check; checking *endp == '\0' would let
+             * "0xABCD\0garbage" silently parse as 0xABCD because endp
+             * lands on the NUL. Same fix CR-306 applied to Map keys. */
+            if (errno == ERANGE || endp == s ||
+                (size_t)(endp - s) != slen) {
+                throw std::runtime_error(
+                    std::string("invalid hex literal for ") + type_label);
+            }
+            if ((uint64_t)n > MaxV) {
+                throw std::runtime_error(
+                    std::string("hex literal out of range for ") + type_label);
+            }
+            value->Append((typename TCol::ValueType)n);
+        } else {
+            zend_long n = strict_zval_long(array_value, type_label);
+            if (n < 0) {
+                throw std::runtime_error(
+                    std::string("negative value cannot fit in ") + type_label);
+            }
+            if ((uint64_t)n > MaxV) {
+                throw std::runtime_error(
+                    std::string("value out of range for ") + type_label);
+            }
+            value->Append((typename TCol::ValueType)n);
         }
-        SC_HASHTABLE_FOREACH_END();
-
-        return value;
-        break;
-    }
-    case Type::Code::UInt8:
-    {
-        auto value = std::make_shared<ColumnUInt8>();
+    } ZEND_HASH_FOREACH_END();
+    return value;
+}
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
+// Build an Enum8 / Enum16 column from a PHP rows array. Integer cells
+// validate against the type's declared value set; the prior unchecked
+// Append silently stored values like 0 / 3 / 127 inside an
+// `Enum8('One'=1,'Two'=2)` column, after which normal reads threw
+// `map::at` because the read path looks up the name for the stored
+// integer. String cells go through ColumnEnum*::Append(name) which
+// validates internally.
+//
+// IS_NULL handling: rejected for non-Nullable enums (would otherwise
+// store raw 0, which is usually not a declared enum value and poisons
+// reads). The Nullable insert path bumps AllowNullGuard so its
+// recursive child build accepts NULL → declared-value placeholder; the
+// null mask captures the actual NULL.
+template <typename TCol, typename TInt>
+static ColumnRef appendEnumColumn(TypeRef type, HashTable *values_ht)
+{
+    auto value = std::make_shared<TCol>(type);
+    auto enum_type = type->As<clickhouse::EnumType>();
+    /* Choose a placeholder int that's actually declared in the enum so
+     * NULL cells under AllowNullGuard land safely. EnumType exposes
+     * begin()/end() iterators over (name, value) pairs; just take the
+     * first one. enum_type can be null on an unexpected schema; in
+     * that case we conservatively use 0 and let HasEnumValue reject. */
+    TInt placeholder = 0;
+    if (enum_type) {
+        auto it = enum_type->BeginValueToName();
+        if (it != enum_type->EndValueToName()) {
+            placeholder = (TInt)it->first;
         }
-        SC_HASHTABLE_FOREACH_END();
-
-        return value;
-        break;
     }
-    case Type::Code::UInt16:
-    {
-        auto value = std::make_shared<ColumnUInt16>();
-
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
-        }
-        SC_HASHTABLE_FOREACH_END();
+    zval *array_value;
+    ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+        if (Z_TYPE_P(array_value) == IS_NULL) {
+            if (g_allow_null_in_strict <= 0) {
+                throw std::runtime_error(
+                    "null cannot be assigned to non-Nullable Enum column");
+            }
+            value->Append(placeholder);
+        } else if (Z_TYPE_P(array_value) == IS_LONG) {
+            zend_long n = Z_LVAL_P(array_value);
+            int16_t narrow = (int16_t)n;
+            if ((zend_long)narrow != n || !enum_type || !enum_type->HasEnumValue(narrow)) {
+                throw std::runtime_error(
+                    "Enum integer value " + std::to_string(n) +
+                    " is not declared in " + type->GetName());
+            }
+            value->Append((TInt)narrow);
+        } else {
+            /* String path: ColumnEnum*::Append(name) validates internally
+             * and throws on unknown names. */
+            ZStrGuard sg(array_value);
+            value->Append(std::string(sg.val(), sg.len()));
+        }
+    } ZEND_HASH_FOREACH_END();
+    return value;
+}
 
-        return value;
-        break;
-    }
-    case Type::Code::UInt32:
-    {
-        auto value = std::make_shared<ColumnUInt32>();
+// Build a ColumnDate / ColumnDate32 / ColumnDateTime column from a PHP
+// rows array. Each row is either an int (epoch seconds) or a "YYYY-MM-DD"
+// (or "YYYY-MM-DD HH:MM:SS" for is_date=false) string. The string path
+// goes through to_time_t which throws on parse failure.
+template <typename TCol>
+static ColumnRef appendDateColumn(HashTable *values_ht, bool is_date)
+{
+    auto value = std::make_shared<TCol>();
+    zval *array_value;
+    const char *type_label = is_date ? "Date" : "DateTime";
+    ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+        /* Any string is treated as a formatted date/datetime. The prior
+         * dash-only gate routed dashless strings through zval_get_long,
+         * which silently coerced "abc" to 0 and landed it as the epoch.
+         * to_time_t now does full validation (EOF after format, gmtime
+         * round-trip). Numeric inputs go through strict_zval_long so
+         * non-numeric, fractional, NaN/Inf are rejected. */
+        if (Z_TYPE_P(array_value) == IS_STRING) {
+            value->Append((std::time_t)to_time_t(
+                std::string(Z_STRVAL_P(array_value), Z_STRLEN_P(array_value)),
+                is_date));
+        } else {
+            value->Append((std::time_t)strict_zval_long(array_value, type_label));
+        }
+    } ZEND_HASH_FOREACH_END();
+    return value;
+}
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
+// Build a LowCardinality(String) / LowCardinality(FixedString) column,
+// optionally wrapped in Nullable. The four code paths used to be
+// near-identical 12-line ZEND_HASH_FOREACH blocks; the template
+// parameterizes on the column type and a compile-time `nullable` flag
+// that decides whether IS_NULL maps to std::nullopt.
+template <typename TCol, bool nullable>
+static ColumnRef appendLowCardinalityColumn(HashTable *values_ht, std::shared_ptr<TCol> value, const char *type_label)
+{
+    zval *array_value;
+    ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+        if constexpr (nullable) {
+            if (Z_TYPE_P(array_value) == IS_NULL) {
+                value->Append(std::nullopt);
+                continue;
+            }
         }
-        SC_HASHTABLE_FOREACH_END();
+        std::string s = strict_zval_string(array_value, type_label);
+        value->Append(std::string_view(s.data(), s.size()));
+    } ZEND_HASH_FOREACH_END();
+    return value;
+}
 
-        return value;
-        break;
-    }
+// Build a Float32/Float64 column from a PHP rows array.
+template <typename TCol>
+static ColumnRef appendFloatColumn(HashTable *values_ht, const char *type_label)
+{
+    auto value = std::make_shared<TCol>();
+    zval *array_value;
+    ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+        value->Append((typename TCol::ValueType)strict_zval_double(array_value, type_label));
+    } ZEND_HASH_FOREACH_END();
+    return value;
+}
 
-    case Type::Code::Int8:
-    {
-        auto value = std::make_shared<ColumnInt8>();
+// Build a ColumnMapT<KCol, VCol> from PHP rows. Each row is an assoc
+// array; the caller supplies extractors that turn (zend_string*, ulong)
+// into K and (zval*) into V.
+template <typename K, typename V, typename KCol, typename VCol,
+          typename KFn, typename VFn>
+static ColumnRef appendMapColumn(HashTable *values_ht, KFn extract_key, VFn extract_val)
+{
+    auto col = std::make_shared<ColumnMapT<KCol, VCol>>(
+        std::make_shared<KCol>(), std::make_shared<VCol>());
+    /* Reuse the entries vector across rows so the per-row push_back
+     * path doesn't fresh-heap-allocate; clear() preserves capacity. */
+    std::vector<std::pair<K, V>> entries;
+    zval *array_value;
+    ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+        if (Z_TYPE_P(array_value) != IS_ARRAY) {
+            throw std::runtime_error("Map row must be a PHP array");
+        }
+        entries.clear();
+        HashTable *mh = Z_ARRVAL_P(array_value);
+        zend_string *zk;
+        zend_ulong nk;
+        zval *mv;
+        ZEND_HASH_FOREACH_KEY_VAL(mh, nk, zk, mv) {
+            entries.emplace_back(extract_key(zk, nk), extract_val(mv));
+        } ZEND_HASH_FOREACH_END();
+        col->Append(entries);
+    } ZEND_HASH_FOREACH_END();
+    return col;
+}
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
+// Parse a PHP zval into a clickhouse UUID. Mirrors the standalone-UUID
+// insert path; used by Map(*, UUID) value extraction.
+static UUID phpToUUID(zval *zv)
+{
+    if (Z_TYPE_P(zv) == IS_NULL) {
+        if (g_allow_null_in_strict <= 0) {
+            throw std::runtime_error("null cannot be assigned to non-Nullable column UUID");
         }
-        SC_HASHTABLE_FOREACH_END();
-
-        return value;
-        break;
+        return UUID{0, 0};
     }
-    case Type::Code::Int16:
-    {
-        auto value = std::make_shared<ColumnInt16>();
+    ZStrGuard sg(zv);
+    return parseUUIDString(sg.val(), sg.len(), "UUID format error");
+}
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
+// Second-stage Map dispatch: key column type already resolved at the
+// call site, dispatch on value type code. Kept as a function template so
+// each (KCol, K) tuple instantiates its own value-side switch and the
+// compiler can fold identical extractor lambdas across instantiations.
+template <typename KCol, typename K, typename KFn>
+static ColumnRef appendMapByValueType(HashTable *values_ht, TypeRef vtype, KFn key_fn)
+{
+    auto strVal = [](zval *mv) -> std::string {
+        return strict_zval_string(mv, "Map value String");
+    };
+    /* Narrow-typed int extractors range-check before truncation. The
+     * non-Map insert path has had these via appendIntColumn since pass 1;
+     * the Map dispatch was using a single i64Val/u64Val for all widths
+     * which silently wrapped Map(K, Int8) value 1000 to int8_t -24. */
+    /* All Map value extractors go through strict_zval_long /
+     * strict_zval_double so non-numeric strings, fractional doubles, and
+     * non-finite floats throw instead of silently coercing to 0 / 0.0
+     * inside the Map. Mirrors CR-003 for the non-Map path. */
+    auto i64Val = [](zval *mv) -> int64_t {
+        return (int64_t)strict_zval_long(mv, "Map value Int64");
+    };
+    auto u64Val = [](zval *mv) -> uint64_t {
+        return strict_zval_u64(mv, "Map value UInt64");
+    };
+    auto narrowI = [](zval *mv, zend_long lo, zend_long hi, const char *t) -> int64_t {
+        zend_long n = strict_zval_long(mv, t);
+        if (n < lo || n > hi) {
+            throw std::runtime_error(std::string("Map value out of range for ") + t);
+        }
+        return (int64_t)n;
+    };
+    auto narrowU = [](zval *mv, zend_ulong hi, const char *t) -> uint64_t {
+        zend_long n = strict_zval_long(mv, t);
+        if (n < 0 || (zend_ulong)n > hi) {
+            throw std::runtime_error(std::string("Map value out of range for ") + t);
+        }
+        return (uint64_t)n;
+    };
+    auto f64Val = [](zval *mv) -> double {
+        return strict_zval_double(mv, "Map value Float");
+    };
+
+    Type::Code vc = vtype->GetCode();
+    switch (vc) {
+        case Type::Code::String:
+            return appendMapColumn<K, std::string, KCol, ColumnString>(values_ht, key_fn, strVal);
+        case Type::Code::Int8: {
+            auto v = [&](zval *mv) { return narrowI(mv, INT8_MIN, INT8_MAX, "Int8"); };
+            return appendMapColumn<K, int64_t,    KCol, ColumnInt8>(values_ht, key_fn, v);
+        }
+        case Type::Code::Int16: {
+            auto v = [&](zval *mv) { return narrowI(mv, INT16_MIN, INT16_MAX, "Int16"); };
+            return appendMapColumn<K, int64_t,    KCol, ColumnInt16>(values_ht, key_fn, v);
+        }
+        case Type::Code::Int32: {
+            auto v = [&](zval *mv) { return narrowI(mv, INT32_MIN, INT32_MAX, "Int32"); };
+            return appendMapColumn<K, int64_t,    KCol, ColumnInt32>(values_ht, key_fn, v);
+        }
+        case Type::Code::Int64:
+            return appendMapColumn<K, int64_t,    KCol, ColumnInt64>(values_ht, key_fn, i64Val);
+        case Type::Code::UInt8: {
+            auto v = [&](zval *mv) { return narrowU(mv, UINT8_MAX, "UInt8"); };
+            return appendMapColumn<K, uint64_t,   KCol, ColumnUInt8>(values_ht, key_fn, v);
+        }
+        case Type::Code::UInt16: {
+            auto v = [&](zval *mv) { return narrowU(mv, UINT16_MAX, "UInt16"); };
+            return appendMapColumn<K, uint64_t,   KCol, ColumnUInt16>(values_ht, key_fn, v);
+        }
+        case Type::Code::UInt32: {
+            auto v = [&](zval *mv) { return narrowU(mv, UINT32_MAX, "UInt32"); };
+            return appendMapColumn<K, uint64_t,   KCol, ColumnUInt32>(values_ht, key_fn, v);
+        }
+        case Type::Code::UInt64:
+            return appendMapColumn<K, uint64_t,   KCol, ColumnUInt64>(values_ht, key_fn, u64Val);
+        case Type::Code::Float32:
+            return appendMapColumn<K, double,     KCol, ColumnFloat32>(values_ht, key_fn, f64Val);
+        case Type::Code::Float64:
+            return appendMapColumn<K, double,     KCol, ColumnFloat64>(values_ht, key_fn, f64Val);
+        case Type::Code::UUID:
+            return appendMapColumn<K, UUID,       KCol, ColumnUUID>(values_ht, key_fn, phpToUUID);
+        case Type::Code::LowCardinality: {
+            TypeRef inner = vtype->As<LowCardinalityType>()->GetNestedType();
+            if (inner->GetCode() == Type::Code::String) {
+                return appendMapColumn<K, std::string, KCol, ColumnLowCardinalityT<ColumnString>>(values_ht, key_fn, strVal);
+            }
+            throw std::runtime_error("Unsupported Map value type LowCardinality(" + inner->GetName() + ")");
         }
-        SC_HASHTABLE_FOREACH_END();
+        default:
+            throw std::runtime_error("Unsupported Map value type: " + vtype->GetName());
+    }
+}
 
-        return value;
-        break;
+// Coerce a PHP 2-element numeric array into a (double, double) point tuple.
+// Used by Point/Ring/Polygon/MultiPolygon insert paths.
+static std::tuple<double, double> phpToPoint(zval *zv)
+{
+    if (Z_TYPE_P(zv) != IS_ARRAY) {
+        throw std::runtime_error("Point must be a PHP array of 2 numbers");
     }
-    case Type::Code::Int32:
-    {
-        auto value = std::make_shared<ColumnInt32>();
+    HashTable *ht = Z_ARRVAL_P(zv);
+    if (zend_hash_num_elements(ht) != 2) {
+        throw std::runtime_error("Point must have exactly 2 elements");
+    }
+    zval *x = sc_zend_hash_index_find(ht, 0);
+    zval *y = sc_zend_hash_index_find(ht, 1);
+    if (!x || !y) {
+        throw std::runtime_error("Point is missing an element");
+    }
+    return std::make_tuple(
+        strict_zval_double(x, "Point coordinate"),
+        strict_zval_double(y, "Point coordinate"));
+}
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
-        }
-        SC_HASHTABLE_FOREACH_END();
+static std::vector<std::tuple<double, double>> phpToRing(zval *zv)
+{
+    if (Z_TYPE_P(zv) != IS_ARRAY) {
+        throw std::runtime_error("Ring must be a PHP array of points");
+    }
+    std::vector<std::tuple<double, double>> ring;
+    zval *pt;
+    ZEND_HASH_FOREACH_VAL(Z_ARRVAL_P(zv), pt) {
+        ring.push_back(phpToPoint(pt));
+    } ZEND_HASH_FOREACH_END();
+    return ring;
+}
 
-        return value;
-        break;
-    }
-    case Type::Code::Int64:
-    {
-        auto value = std::make_shared<ColumnInt64>();
+static std::vector<std::vector<std::tuple<double, double>>> phpToPolygon(zval *zv)
+{
+    if (Z_TYPE_P(zv) != IS_ARRAY) {
+        throw std::runtime_error("Polygon must be a PHP array of rings");
+    }
+    std::vector<std::vector<std::tuple<double, double>>> poly;
+    zval *r;
+    ZEND_HASH_FOREACH_VAL(Z_ARRVAL_P(zv), r) {
+        poly.push_back(phpToRing(r));
+    } ZEND_HASH_FOREACH_END();
+    return poly;
+}
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
-        }
-        SC_HASHTABLE_FOREACH_END();
+ColumnRef insertColumn(TypeRef type, zval *value_zval)
+{
+    ConvertDepthGuard _depth_guard;  // shared with createColumn / convertToZval
+    zval *array_value;
+    HashTable *values_ht = Z_ARRVAL_P(value_zval);
 
+    switch (type->GetCode())
+    {
+    case Type::Code::UInt64: {
+        /* Direct call into strict_zval_u64 instead of going through
+         * appendUIntColumnWithHex<ColumnUInt64>. The templated path
+         * casts through zend_long for the non-hex branch and rejects
+         * decimal strings above ZEND_LONG_MAX. UInt64 values up to
+         * UINT64_MAX must be insertable from PHP — readers already
+         * surface them as decimal strings — so the writer needs to
+         * accept the same form. */
+        auto value = std::make_shared<ColumnUInt64>();
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+            value->Append(strict_zval_u64(array_value, "UInt64"));
+        } ZEND_HASH_FOREACH_END();
         return value;
-        break;
     }
+    case Type::Code::UInt8:
+        return appendIntColumn<ColumnUInt8>(values_ht, 0, 0xFF, "UInt8");
+    case Type::Code::UInt16:
+        return appendIntColumn<ColumnUInt16>(values_ht, 0, 0xFFFF, "UInt16");
+    case Type::Code::UInt32:
+        return appendUIntColumnWithHex<ColumnUInt32>(values_ht, strtoul, UINT32_MAX, "UInt32");
+    case Type::Code::Int8:
+        return appendIntColumn<ColumnInt8>(values_ht, INT8_MIN, INT8_MAX, "Int8");
+    case Type::Code::Int16:
+        return appendIntColumn<ColumnInt16>(values_ht, INT16_MIN, INT16_MAX, "Int16");
+    case Type::Code::Int32:
+        return appendIntColumn<ColumnInt32>(values_ht, INT32_MIN, INT32_MAX, "Int32");
+    case Type::Code::Int64:
+        return appendIntColumn<ColumnInt64>(values_ht, INT64_MIN, INT64_MAX, "Int64");
 
     case Type::Code::UUID:
     {
         auto value = std::make_shared<ColumnUUID>();
-
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
         {
-            if (Z_TYPE_P(array_value) == IS_NULL)
-            {
-                value->Append(UInt128(0, 0));
-            }
-            else
-            {
-                convert_to_string(array_value);
-                string value_string = (string)Z_STRVAL_P(array_value);
-
-                value_string.erase(std::remove(value_string.begin(), value_string.end(), '-'), value_string.end());
-                if (value_string.length() != 32)
-                {
-                    throw std::runtime_error("UUID format error");
-                }
-
-                string first = value_string.substr(0, 16);
-                string second = value_string.substr(16, 16);
-                uint64_t i_first = std::stoull(first, nullptr, 16);
-                uint64_t i_second = std::stoull(second, nullptr, 16);
-                value->Append(UInt128(i_first, i_second));
-            }
+            value->Append(phpToUUID(array_value));
         }
-        SC_HASHTABLE_FOREACH_END();
+        ZEND_HASH_FOREACH_END();
         return value;
-        break;
     }
 
     case Type::Code::Float32:
+        return appendFloatColumn<ColumnFloat32>(values_ht, "Float32");
+    case Type::Code::Float64:
+        return appendFloatColumn<ColumnFloat64>(values_ht, "Float64");
+
+    case Type::Code::String:
     {
-        auto value = std::make_shared<ColumnFloat32>();
+        auto value = std::make_shared<ColumnString>();
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
         {
-            convert_to_double(array_value);
-            value->Append(Z_DVAL_P(array_value));
+            value->Append(strict_zval_string(array_value, "String"));
         }
-        SC_HASHTABLE_FOREACH_END();
+        ZEND_HASH_FOREACH_END();
 
         return value;
-        break;
     }
-    case Type::Code::Float64:
+    case Type::Code::FixedString:
     {
-        auto value = std::make_shared<ColumnFloat64>();
+        auto value = std::make_shared<ColumnFixedString>(parseFixedStringWidth(type));
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
         {
-            convert_to_double(array_value);
-            value->Append(Z_DVAL_P(array_value));
+            value->Append(strict_zval_string(array_value, "FixedString"));
         }
-        SC_HASHTABLE_FOREACH_END();
+        ZEND_HASH_FOREACH_END();
 
         return value;
-        break;
     }
 
-    case Type::Code::String:
-    {
-        auto value = std::make_shared<ColumnString>();
-
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_string(array_value);
-            value->Append((string)Z_STRVAL_P(array_value));
+    case Type::Code::DateTime:
+        return appendDateColumn<ColumnDateTime>(values_ht, /*is_date=*/false);
+    case Type::Code::DateTime64:
+    {
+        size_t precision = type->As<DateTime64Type>()->GetPrecision();
+        auto value = std::make_shared<ColumnDateTime64>(precision);
+        int64_t scale = 1;
+        for (size_t i = 0; i < precision; ++i) scale *= 10;
+
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
+        {
+            /* Any string is treated as a formatted timestamp; the prior
+             * dash-only gate let "abc" fall through zval_get_long to 0
+             * (epoch). to_time_t_with_frac validates fully. Numeric
+             * inputs go through strict_zval_long / strict_zval_double. */
+            if (Z_TYPE_P(array_value) == IS_STRING) {
+                auto [whole, frac] = to_time_t_with_frac(
+                    std::string(Z_STRVAL_P(array_value), Z_STRLEN_P(array_value)),
+                    precision, scale);
+                value->Append((int64_t)whole * scale + frac);
+            } else if (Z_TYPE_P(array_value) == IS_DOUBLE) {
+                double d = strict_zval_double(array_value, "DateTime64");
+                value->Append((int64_t)(d * scale));
+            } else {
+                value->Append((int64_t)strict_zval_long(array_value, "DateTime64") * scale);
+            }
         }
-        SC_HASHTABLE_FOREACH_END();
+        ZEND_HASH_FOREACH_END();
 
         return value;
     }
-    case Type::Code::FixedString:
-    {
-        string typeName = type->GetName();
-        typeName.erase(typeName.find("FixedString("), 12);
-        typeName.erase(typeName.find(")"), 1);
-        auto value = std::make_shared<ColumnFixedString>(std::stoi(typeName));
-
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            convert_to_string(array_value);
-            value->Append((string)Z_STRVAL_P(array_value));
+    case Type::Code::Date:
+        return appendDateColumn<ColumnDate>(values_ht, /*is_date=*/true);
+    case Type::Code::Date32:
+        return appendDateColumn<ColumnDate32>(values_ht, /*is_date=*/true);
+    case Type::Code::Time:
+    {
+        auto value = std::make_shared<ColumnTime>();
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
+        {
+            /* Time is stored as int seconds-since-midnight; accept only
+             * numeric inputs. String inputs would need an "HH:MM:SS"
+             * parser which the column type doesn't currently expose, so
+             * reject strings explicitly instead of letting zval_get_long
+             * coerce "abc" to 0. */
+            if (Z_TYPE_P(array_value) == IS_STRING) {
+                throw std::runtime_error(
+                    "Time column inserts require numeric seconds; "
+                    "string formatted-time input is not currently supported");
+            }
+            value->Append((int32_t)strict_zval_long(array_value, "Time"));
         }
-        SC_HASHTABLE_FOREACH_END();
-
+        ZEND_HASH_FOREACH_END();
         return value;
     }
-
-    case Type::Code::DateTime:
+    case Type::Code::Time64:
     {
-        auto value = std::make_shared<ColumnDateTime>();
-
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
+        size_t precision = type->As<Time64Type>()->GetPrecision();
+        auto value = std::make_shared<ColumnTime64>(precision);
+        int64_t scale = 1;
+        for (size_t i = 0; i < precision; ++i) scale *= 10;
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
         {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
+            if (Z_TYPE_P(array_value) == IS_STRING) {
+                throw std::runtime_error(
+                    "Time64 column inserts require numeric seconds; "
+                    "string formatted-time input is not currently supported");
+            }
+            if (Z_TYPE_P(array_value) == IS_DOUBLE) {
+                double d = strict_zval_double(array_value, "Time64");
+                value->Append((int64_t)(d * scale));
+            } else {
+                value->Append((int64_t)strict_zval_long(array_value, "Time64") * scale);
+            }
         }
-        SC_HASHTABLE_FOREACH_END();
-
+        ZEND_HASH_FOREACH_END();
         return value;
-        break;
     }
-    case Type::Code::Date:
+    case Type::Code::Int128:
+    {
+        auto value = std::make_shared<ColumnInt128>();
+        /* Int128 range is [-2^127, 2^127-1]. parse_uint128_dec accepts up
+         * to 2^128-1, so an unbounded magnitude in (2^127, 2^128-1] used
+         * to silently wrap to negative via the static_cast. Bound the
+         * magnitude before casting; the negative-INT128_MIN edge needs
+         * special handling because -INT128_MIN is undefined. */
+        const absl::uint128 abs_int128_min = absl::uint128(1) << 127;
+        const absl::uint128 int128_max     = abs_int128_min - 1;
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
+        {
+            if (Z_TYPE_P(array_value) == IS_STRING) {
+                const char *s = Z_STRVAL_P(array_value);
+                size_t len = Z_STRLEN_P(array_value);
+                size_t i = 0;
+                bool neg = false;
+                if (len > 0 && (s[0] == '-' || s[0] == '+')) { neg = (s[0] == '-'); i = 1; }
+                absl::uint128 mag = parse_uint128_dec(s + i, len - i, "Int128");
+                if (neg) {
+                    if (mag > abs_int128_min) {
+                        throw std::runtime_error("Int128 string is below -2^127");
+                    }
+                    if (mag == abs_int128_min) {
+                        /* INT128_MIN: -2^127. Constructing via -static_cast
+                         * <Int128>(2^127) would be UB on the negation. */
+                        value->Append(static_cast<Int128>(mag));
+                    } else {
+                        value->Append(-static_cast<Int128>(mag));
+                    }
+                } else {
+                    if (mag > int128_max) {
+                        throw std::runtime_error("Int128 string exceeds 2^127-1");
+                    }
+                    value->Append(static_cast<Int128>(mag));
+                }
+            } else {
+                value->Append(Int128(strict_zval_long(array_value, "Int128")));
+            }
+        }
+        ZEND_HASH_FOREACH_END();
+        return value;
+    }
+    case Type::Code::UInt128:
+    {
+        auto value = std::make_shared<ColumnUInt128>();
+        // UInt128 range: 0 .. 2^128-1, i.e. up to 39 decimal digits.
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
+        {
+            if (Z_TYPE_P(array_value) == IS_STRING) {
+                const char *s = Z_STRVAL_P(array_value);
+                size_t len = Z_STRLEN_P(array_value);
+                size_t i = 0;
+                if (len > 0 && s[0] == '+') { i = 1; }
+                value->Append(parse_uint128_dec(s + i, len - i, "UInt128"));
+            } else {
+                zend_long n = strict_zval_long(array_value, "UInt128");
+                if (n < 0) {
+                    throw std::runtime_error("UInt128 cannot accept a negative integer");
+                }
+                value->Append(UInt128((uint64_t)n));
+            }
+        }
+        ZEND_HASH_FOREACH_END();
+        return value;
+    }
+    case Type::Code::Decimal:
+    case Type::Code::Decimal32:
+    case Type::Code::Decimal64:
+    case Type::Code::Decimal128:
     {
-        auto value = std::make_shared<ColumnDate>();
-
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
+        auto dt = type->As<DecimalType>();
+        auto value = std::make_shared<ColumnDecimal>(dt->GetPrecision(), dt->GetScale());
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
         {
-            convert_to_long(array_value);
-            value->Append(Z_LVAL_P(array_value));
+            ZStrGuard sg(array_value);
+            value->Append(std::string(sg.val(), sg.len()));
         }
-        SC_HASHTABLE_FOREACH_END();
-
+        ZEND_HASH_FOREACH_END();
         return value;
-        break;
     }
 
     case Type::Code::Array:
     {
-        if (type->GetItemType()->GetCode() == Type::Array)
+        TypeRef item_type = type->As<ArrayType>()->GetItemType();
+        if (item_type->GetCode() == Type::Array)
         {
             throw std::runtime_error("can't support Multidimensional Arrays");
         }
 
-        auto value = std::make_shared<ColumnArray>(createColumn(type->GetItemType()));
-        auto child = createColumn(type->GetItemType());
+        auto value = std::make_shared<ColumnArray>(createColumn(item_type));
+        auto child = createColumn(item_type);
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
         {
             if (Z_TYPE_P(array_value) != IS_ARRAY)
             {
                 throw std::runtime_error("The inserted data is not an array type");
             }
 
-            child->Append(insertColumn(type->GetItemType(), array_value));
+            child->Append(insertColumn(item_type, array_value));
 
             value->AppendAsColumn(child);
             child->Clear();
         }
-        SC_HASHTABLE_FOREACH_END();
+        ZEND_HASH_FOREACH_END();
 
         return value;
-        break;
     }
 
     case Type::Code::Enum8:
-    {
-        std::vector<Type::EnumItem> enum_items;
-
-        auto enumType = EnumType(type);
-
-        for (auto ei = enumType.BeginValueToName(); ; )
-        {
-            enum_items.push_back(
-                Type::EnumItem {ei->second, (int8_t)ei->first});
-            if (++ei == enumType.EndValueToName())
-            {
-                break;
-            }
-        }
+        return appendEnumColumn<ColumnEnum8, int8_t>(type, values_ht);
+    case Type::Code::Enum16:
+        return appendEnumColumn<ColumnEnum16, int16_t>(type, values_ht);
 
-        auto value = std::make_shared<ColumnEnum8>(Type::CreateEnum8(enum_items));
+    case Type::Code::Nullable:
+    {
+        auto nulls = std::make_shared<ColumnUInt8>();
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value)
         {
-            if (Z_TYPE_P(array_value) == IS_LONG)
+            if (Z_TYPE_P(array_value) == IS_NULL)
             {
-                convert_to_long(array_value);
-                value->Append(Z_LVAL_P(array_value));
+                nulls->Append(1);
             }
             else
             {
-                convert_to_string(array_value);
-                value->Append((string)Z_STRVAL_P(array_value));
+                nulls->Append(0);
             }
         }
-        SC_HASHTABLE_FOREACH_END();
+        ZEND_HASH_FOREACH_END();
 
-        return value;
-        break;
+        /* The null mask captures IS_NULL cells, so the recursive child
+         * build can accept NULL → typed-zero placeholder. Without the
+         * guard, strict_zval_long / strict_zval_double would now
+         * (post-CR-002) reject IS_NULL outright and break every
+         * Nullable insert. */
+        AllowNullGuard nulls_ok;
+        ColumnRef child = insertColumn(type->As<NullableType>()->GetNestedType(), value_zval);
+
+        return std::make_shared<ColumnNullable>(child, nulls);
     }
-    case Type::Code::Enum16:
+
+    case Type::Code::Tuple:
     {
-        std::vector<Type::EnumItem> enum_items;
+        // Build one transposed list per tuple field (arity), iterating
+        // every input row to pull row[field]. The previous version
+        // looped by row count instead of arity, so multi-row tuple
+        // inserts walked off the end of tupleType when rowcount != arity.
+        auto tupleType = type->As<TupleType>()->GetTupleType();
+        size_t arity = tupleType.size();
+
+        zval return_should_storage;
+        ZVAL_UNDEF(&return_should_storage);
+        zval *return_should = &return_should_storage;
+        array_init(return_should);
 
-        auto enumType = EnumType(type);
+        zval return_tmp_storage;
+        ZVAL_UNDEF(&return_tmp_storage);
+        zval *return_tmp = &return_tmp_storage;
 
-        for (auto ei = enumType.BeginValueToName(); ; )
-        {
-            enum_items.push_back(
-                Type::EnumItem {ei->second, (int16_t)ei->first});
-            if (++ei == enumType.EndValueToName())
+        try {
+            zval *fzval;
+            zval *pzval;
+            for (size_t field = 0; field < arity; field++)
             {
-                break;
-            }
-        }
+                array_init(return_tmp);
 
-        auto value = std::make_shared<ColumnEnum16>(Type::CreateEnum16(enum_items));
+                ZEND_HASH_FOREACH_VAL(values_ht, pzval)
+                {
+                    if (Z_TYPE_P(pzval) != IS_ARRAY)
+                    {
+                        throw std::runtime_error("Tuple row must be a PHP array");
+                    }
+                    if (zend_hash_num_elements(Z_ARRVAL_P(pzval)) != arity) {
+                        throw std::runtime_error(
+                            "Tuple row arity does not match the column type");
+                    }
+                    fzval = sc_zend_hash_index_find(Z_ARRVAL_P(pzval), field);
+                    if (NULL == fzval)
+                    {
+                        throw std::runtime_error(
+                            "Tuple row is missing a field value");
+                    }
+                    sc_zval_add_ref(fzval);
+                    add_next_index_zval(return_tmp, fzval);
+                }
+                ZEND_HASH_FOREACH_END();
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            if (Z_TYPE_P(array_value) == IS_LONG)
-            {
-                convert_to_long(array_value);
-                value->Append(Z_LVAL_P(array_value));
+                /* Transfer ownership: clear the slot so the catch handler
+                 * doesn't double-free what is now owned by return_should. */
+                add_next_index_zval(return_should, return_tmp);
+                ZVAL_UNDEF(return_tmp);
             }
-            else
+
+            std::vector<ColumnRef> columns;
+            size_t tupleTypeIndex = 0;
+
+            ZEND_HASH_FOREACH_VAL(Z_ARRVAL_P(return_should), array_value)
             {
-                convert_to_string(array_value);
-                value->Append((string)Z_STRVAL_P(array_value));
+                if (Z_TYPE_P(array_value) != IS_ARRAY)
+                {
+                    throw std::runtime_error("The inserted data is not an array type");
+                }
+
+                columns.push_back(insertColumn(tupleType[tupleTypeIndex], array_value));
+                tupleTypeIndex++;
             }
+            ZEND_HASH_FOREACH_END();
+
+            zval_ptr_dtor(return_should);
+            return std::make_shared<ColumnTuple>(columns);
+        } catch (...) {
+            if (Z_TYPE(return_tmp_storage) != IS_UNDEF) zval_ptr_dtor(return_tmp);
+            if (Z_TYPE(return_should_storage) != IS_UNDEF) zval_ptr_dtor(return_should);
+            throw;
         }
-        SC_HASHTABLE_FOREACH_END();
+    }
 
-        return value;
-        break;
+    case Type::Code::LowCardinality:
+    {
+        TypeRef nested = type->As<LowCardinalityType>()->GetNestedType();
+        bool is_nullable = (nested->GetCode() == Type::Code::Nullable);
+        TypeRef inner = is_nullable
+            ? nested->As<NullableType>()->GetNestedType()
+            : nested;
+
+        if (inner->GetCode() == Type::Code::String) {
+            if (is_nullable) {
+                return appendLowCardinalityColumn<ColumnLowCardinalityT<ColumnNullableT<ColumnString>>, /*nullable=*/true>(
+                    values_ht, std::make_shared<ColumnLowCardinalityT<ColumnNullableT<ColumnString>>>(), "LowCardinality Nullable String");
+            }
+            return appendLowCardinalityColumn<ColumnLowCardinalityT<ColumnString>, /*nullable=*/false>(
+                values_ht, std::make_shared<ColumnLowCardinalityT<ColumnString>>(), "LowCardinality String");
+        }
+        if (inner->GetCode() == Type::Code::FixedString) {
+            int width = parseFixedStringWidth(inner);
+            if (is_nullable) {
+                return appendLowCardinalityColumn<ColumnLowCardinalityT<ColumnNullableT<ColumnFixedString>>, /*nullable=*/true>(
+                    values_ht, std::make_shared<ColumnLowCardinalityT<ColumnNullableT<ColumnFixedString>>>(width), "LowCardinality Nullable FixedString");
+            }
+            return appendLowCardinalityColumn<ColumnLowCardinalityT<ColumnFixedString>, /*nullable=*/false>(
+                values_ht, std::make_shared<ColumnLowCardinalityT<ColumnFixedString>>(width), "LowCardinality FixedString");
+        }
+        throw std::runtime_error("LowCardinality only supported over String / FixedString (Nullable allowed)");
     }
 
-    case Type::Code::Nullable:
+    case Type::Code::Map:
     {
-        auto nulls = std::make_shared<ColumnUInt8>();
+        TypeRef k = type->As<MapType>()->GetKeyType();
+        TypeRef v = type->As<MapType>()->GetValueType();
+        Type::Code kc = k->GetCode();
 
-        SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, array_value)
-        {
-            if (Z_TYPE_P(array_value) == IS_NULL)
-            {
-                nulls->Append(1);
+        // String keys reject integer-keyed PHP entries outright; integer
+        // keys parse the string form or fall back to the numeric key.
+        // Numeric parsers reject anything that doesn't consume the full
+        // string (PHP's strtoll silently returned 0 for "abc" before).
+        auto strKey = [](zend_string *zk, zend_ulong) -> std::string {
+            if (!zk) {
+                throw std::runtime_error("Map(String, *) row entry must have a string key");
             }
-            else
-            {
-                nulls->Append(0);
+            return std::string(ZSTR_VAL(zk), ZSTR_LEN(zk));
+        };
+        /* PHP zend_string is length-prefixed and may contain embedded
+         * NUL bytes. Comparing endp against ZSTR_LEN is the right
+         * "fully consumed" check; checking *endp == '\0' would let
+         * "123\x00garbage" silently parse as 123 because endp would
+         * land on the NUL. */
+        auto i64Key = [](zend_string *zk, zend_ulong nk) -> int64_t {
+            if (!zk) return (int64_t)nk;
+            const char *s = ZSTR_VAL(zk);
+            char *endp = NULL;
+            errno = 0;
+            long long v = strtoll(s, &endp, 10);
+            if (errno == ERANGE || endp == s || (size_t)(endp - s) != ZSTR_LEN(zk)) {
+                throw std::runtime_error(
+                    std::string("Map integer key is not a valid number: ") +
+                    std::string(s, ZSTR_LEN(zk)));
+            }
+            return (int64_t)v;
+        };
+        auto u64Key = [](zend_string *zk, zend_ulong nk) -> uint64_t {
+            if (!zk) return (uint64_t)nk;
+            const char *s = ZSTR_VAL(zk);
+            char *endp = NULL;
+            errno = 0;
+            unsigned long long v = strtoull(s, &endp, 10);
+            if (errno == ERANGE || endp == s || (size_t)(endp - s) != ZSTR_LEN(zk)) {
+                throw std::runtime_error(
+                    std::string("Map unsigned key is not a valid number: ") +
+                    std::string(s, ZSTR_LEN(zk)));
+            }
+            return (uint64_t)v;
+        };
+        auto f64Key = [](zend_string *zk, zend_ulong nk) -> double {
+            if (!zk) return (double)nk;
+            const char *s = ZSTR_VAL(zk);
+            char *endp = NULL;
+            errno = 0;
+            double v = strtod(s, &endp);
+            if (errno == ERANGE || endp == s || (size_t)(endp - s) != ZSTR_LEN(zk)) {
+                throw std::runtime_error(
+                    std::string("Map float key is not a valid number: ") +
+                    std::string(s, ZSTR_LEN(zk)));
+            }
+            return v;
+        };
+        auto uuidKey = [](zend_string *zk, zend_ulong) -> UUID {
+            if (!zk) {
+                throw std::runtime_error("Map(UUID, *) row entry must have a string key");
+            }
+            return parseUUIDString(ZSTR_VAL(zk), ZSTR_LEN(zk), "UUID key format error");
+        };
+
+        /* Narrow-key wrappers: same range-check the value side gained for
+         * narrow Map columns. Keys arrive as decimal strings; the parsed
+         * int64 must still fit the destination column width. */
+        auto narrowKeyI = [&](zend_string *zk, zend_ulong nk,
+                              int64_t lo, int64_t hi, const char *t) -> int64_t {
+            int64_t parsed = i64Key(zk, nk);
+            if (parsed < lo || parsed > hi) {
+                throw std::runtime_error(std::string("Map key out of range for ") + t);
+            }
+            return parsed;
+        };
+        auto narrowKeyU = [&](zend_string *zk, zend_ulong nk,
+                              uint64_t hi, const char *t) -> uint64_t {
+            uint64_t parsed = u64Key(zk, nk);
+            if (parsed > hi) {
+                throw std::runtime_error(std::string("Map key out of range for ") + t);
+            }
+            return parsed;
+        };
+
+        switch (kc) {
+            case Type::Code::String:
+                return appendMapByValueType<ColumnString,  std::string>(values_ht, v, strKey);
+            case Type::Code::Int8: {
+                auto kf = [&](zend_string *zk, zend_ulong nk) {
+                    return narrowKeyI(zk, nk, INT8_MIN, INT8_MAX, "Int8");
+                };
+                return appendMapByValueType<ColumnInt8,    int64_t>(values_ht, v, kf);
+            }
+            case Type::Code::Int16: {
+                auto kf = [&](zend_string *zk, zend_ulong nk) {
+                    return narrowKeyI(zk, nk, INT16_MIN, INT16_MAX, "Int16");
+                };
+                return appendMapByValueType<ColumnInt16,   int64_t>(values_ht, v, kf);
+            }
+            case Type::Code::Int32: {
+                auto kf = [&](zend_string *zk, zend_ulong nk) {
+                    return narrowKeyI(zk, nk, INT32_MIN, INT32_MAX, "Int32");
+                };
+                return appendMapByValueType<ColumnInt32,   int64_t>(values_ht, v, kf);
+            }
+            case Type::Code::Int64:
+                return appendMapByValueType<ColumnInt64,   int64_t>(values_ht, v, i64Key);
+            case Type::Code::UInt8: {
+                auto kf = [&](zend_string *zk, zend_ulong nk) {
+                    return narrowKeyU(zk, nk, UINT8_MAX, "UInt8");
+                };
+                return appendMapByValueType<ColumnUInt8,   uint64_t>(values_ht, v, kf);
+            }
+            case Type::Code::UInt16: {
+                auto kf = [&](zend_string *zk, zend_ulong nk) {
+                    return narrowKeyU(zk, nk, UINT16_MAX, "UInt16");
+                };
+                return appendMapByValueType<ColumnUInt16,  uint64_t>(values_ht, v, kf);
+            }
+            case Type::Code::UInt32: {
+                auto kf = [&](zend_string *zk, zend_ulong nk) {
+                    return narrowKeyU(zk, nk, UINT32_MAX, "UInt32");
+                };
+                return appendMapByValueType<ColumnUInt32,  uint64_t>(values_ht, v, kf);
             }
+            case Type::Code::UInt64:
+                return appendMapByValueType<ColumnUInt64,  uint64_t>(values_ht, v, u64Key);
+            case Type::Code::Float32:
+                return appendMapByValueType<ColumnFloat32, double>(values_ht, v, f64Key);
+            case Type::Code::Float64:
+                return appendMapByValueType<ColumnFloat64, double>(values_ht, v, f64Key);
+            case Type::Code::UUID:
+                return appendMapByValueType<ColumnUUID,    UUID>(values_ht, v, uuidKey);
+            case Type::Code::LowCardinality: {
+                TypeRef inner = k->As<LowCardinalityType>()->GetNestedType();
+                if (inner->GetCode() == Type::Code::String) {
+                    return appendMapByValueType<ColumnLowCardinalityT<ColumnString>, std::string>(values_ht, v, strKey);
+                }
+                throw std::runtime_error("Unsupported Map key type LowCardinality(" + inner->GetName() + ")");
+            }
+            default:
+                throw std::runtime_error("Unsupported Map(K, V) for row write: " + type->GetName());
         }
-        SC_HASHTABLE_FOREACH_END();
-
-        ColumnRef child = insertColumn(type->GetNestedType(), value_zval);
-
-        return std::make_shared<ColumnNullable>(child, nulls);
-        break;
     }
 
-    case Type::Code::Tuple:
+    case Type::Code::Point:
     {
-        size_t values_count = zend_hash_num_elements(values_ht);
-
-        zval *return_should;
-        SC_MAKE_STD_ZVAL(return_should);
-        array_init(return_should);
-
-        zval *fzval;
-        zval *pzval;
-
-        zval *return_tmp;
-        for(size_t i = 0; i < values_count; i++)
-        {
-            SC_MAKE_STD_ZVAL(return_tmp);
-            array_init(return_tmp);
-
-            SC_HASHTABLE_FOREACH_START2(values_ht, str_key, str_keylen, keytype, pzval)
-            {
-                if (Z_TYPE_P(pzval) != IS_ARRAY)
-                {
-                    throw std::runtime_error("The insert function needs to pass in a two-dimensional array");
-                }
-                fzval = sc_zend_hash_index_find(Z_ARRVAL_P(pzval), i);
-                if (NULL == fzval)
-                {
-                    throw std::runtime_error("The number of parameters inserted per line is inconsistent");
-                }
-                sc_zval_add_ref(fzval);
-                add_next_index_zval(return_tmp, fzval);
+        auto col = std::make_shared<ColumnPoint>();
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+            col->Append(phpToPoint(array_value));
+        } ZEND_HASH_FOREACH_END();
+        return col;
+    }
+    case Type::Code::Ring:
+    {
+        auto col = std::make_shared<ColumnRing>();
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+            col->Append(phpToRing(array_value));
+        } ZEND_HASH_FOREACH_END();
+        return col;
+    }
+    case Type::Code::Polygon:
+    {
+        auto col = std::make_shared<ColumnPolygon>();
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+            col->Append(phpToPolygon(array_value));
+        } ZEND_HASH_FOREACH_END();
+        return col;
+    }
+    case Type::Code::MultiPolygon:
+    {
+        auto col = std::make_shared<ColumnMultiPolygon>();
+        ZEND_HASH_FOREACH_VAL(values_ht, array_value) {
+            if (Z_TYPE_P(array_value) != IS_ARRAY) {
+                throw std::runtime_error("MultiPolygon must be a PHP array of polygons");
             }
-            SC_HASHTABLE_FOREACH_END();
+            std::vector<std::vector<std::vector<std::tuple<double, double>>>> mp;
+            zval *poly;
+            ZEND_HASH_FOREACH_VAL(Z_ARRVAL_P(array_value), poly) {
+                mp.push_back(phpToPolygon(poly));
+            } ZEND_HASH_FOREACH_END();
+            col->Append(mp);
+        } ZEND_HASH_FOREACH_END();
+        return col;
+    }
 
-            add_next_index_zval(return_should, return_tmp);
-        }
+    case Type::Code::Void:
+    {
+        throw std::runtime_error("can't support Void");
+    }
+    default:
+        throw std::runtime_error("insertColumn: unsupported type code: " + type->GetName());
+    }
+}
 
-        auto tupleType = type->GetTupleType();
-        
-        std::vector<ColumnRef> columns;
+// Cast through (zend_long) so signed types (Int8..Int64) keep their
+// sign on the way into PHP, instead of getting reinterpreted as huge
+// unsigned values. Unsigned types up to UINT64_MAX preserve their bit
+// pattern either way; PHP integers are signed 64-bit regardless.
+#define SC_SINGLE_LONG()  \
+    if (fetch_mode & SC_FETCH_ONE) { \
+        ZVAL_LONG(arr, (zend_long)col); \
+    } else { \
+        sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_long)col); \
+    }
+
+#define SC_SINGLE_DOUBLE(val)  \
+    if (fetch_mode & SC_FETCH_ONE) { \
+        ZVAL_DOUBLE(arr, val); \
+    } else { \
+        sc_add_assoc_double_ex(arr, column_name.c_str(), column_name.length(), val); \
+    }
+
+#define SC_SINGLE_STRING(val, len)  \
+    if (fetch_mode & SC_FETCH_ONE) { \
+        ZVAL_STRINGL(arr, val, len); \
+    } else { \
+        sc_add_assoc_stringl_ex(arr, column_name.c_str(), column_name.length(), val, len, 1); \
+    }
+
+// Emit a Unix epoch as either a long or a strftime-formatted string,
+// dispatched on fetch_mode and is_array. Used by DateTime, Date, and
+// Date32 reads which all share the same shape modulo the format string.
+//
+// All three wire types either prohibit negative values (Date is uint16,
+// DateTime is uint32) or treat them as valid pre-epoch dates (Date32);
+// `t == 0` is 1970-01-01, a valid value. We don't emit NULL for any
+// non-NULL server value here.
+static void emitEpoch(zval *arr, std::time_t t, const char *fmt,
+                      const string& column_name, int8_t is_array, long fetch_mode)
+{
+    if (fetch_mode & SC_FETCH_DATE_AS_STRINGS) {
+        char buffer[32];
+        size_t l = strftime(buffer, sizeof(buffer), fmt, gmtime(&t));
+        if (is_array) {
+            sc_add_next_index_stringl(arr, buffer, l, 1);
+        } else {
+            SC_SINGLE_STRING(buffer, l);
+        }
+    } else {
+        if (is_array) {
+            add_next_index_long(arr, (zend_long)t);
+        } else if (fetch_mode & SC_FETCH_ONE) {
+            ZVAL_LONG(arr, (zend_long)t);
+        } else {
+            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_long)t);
+        }
+    }
+}
 
-        int tupleTypeIndex = 0;
+// Read one integer column cell (UInt8..UInt64, Int8..Int64, IPv4) and
+// emit it as a PHP long. The fetch-mode dispatch is identical across
+// all eight integer column types, so they all route through here.
+template <typename TCol>
+static inline void emitIntColumn(zval *arr, const ColumnRef& columnRef, int row,
+                                 const string& column_name, int8_t is_array, long fetch_mode)
+{
+    auto col_ptr = columnRef->As<TCol>();
+    if (!col_ptr) {
+        throw std::runtime_error("Integer column downcast failed");
+    }
+    auto col = (*col_ptr)[row];
+    if (is_array) {
+        add_next_index_long(arr, (zend_long)col);
+    } else {
+        SC_SINGLE_LONG();
+    }
+}
 
-        SC_HASHTABLE_FOREACH_START2(Z_ARRVAL_P(return_should), str_key, str_keylen, keytype, array_value)
-        {
-            if (Z_TYPE_P(array_value) != IS_ARRAY)
-            {
-                throw std::runtime_error("The inserted data is not an array type");
-            }
+// UInt64 specialization. Values above ZEND_LONG_MAX (2^63-1) lose
+// unsigned semantics when cast to zend_long — they read back as
+// negatives in PHP, and Map(UInt64,*) keys collapse distinct
+// unsigned values onto the same PHP-signed key. For values that
+// don't fit a signed PHP integer, emit a decimal string instead so
+// the user can round-trip safely. Values <= ZEND_LONG_MAX continue
+// to come back as PHP int for backward compatibility.
+static inline void emitUInt64Cell(zval *arr, uint64_t v,
+                                  const string& column_name, int8_t is_array, long fetch_mode)
+{
+    if (v <= (uint64_t)ZEND_LONG_MAX) {
+        if (is_array) {
+            add_next_index_long(arr, (zend_long)v);
+        } else if (fetch_mode & SC_FETCH_ONE) {
+            ZVAL_LONG(arr, (zend_long)v);
+        } else {
+            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(),
+                                 (zend_long)v);
+        }
+        return;
+    }
+    char buf[32];
+    int len = snprintf(buf, sizeof(buf), "%" PRIu64, v);
+    if (is_array) {
+        sc_add_next_index_stringl(arr, buf, len, 1);
+    } else if (fetch_mode & SC_FETCH_ONE) {
+        ZVAL_STRINGL(arr, buf, len);
+    } else {
+        sc_add_assoc_stringl_ex(arr, column_name.c_str(), column_name.length(),
+                                buf, len, 1);
+    }
+}
 
-            columns.push_back(insertColumn(tupleType[tupleTypeIndex], array_value));
-            tupleTypeIndex++;
-        }
-        SC_HASHTABLE_FOREACH_END();
+template <>
+inline void emitIntColumn<ColumnUInt64>(zval *arr, const ColumnRef& columnRef, int row,
+                                        const string& column_name, int8_t is_array, long fetch_mode)
+{
+    auto col_ptr = columnRef->As<ColumnUInt64>();
+    if (!col_ptr) {
+        throw std::runtime_error("Integer column downcast failed");
+    }
+    emitUInt64Cell(arr, (uint64_t)(*col_ptr)[row], column_name, is_array, fetch_mode);
+}
 
-        sc_zval_ptr_dtor(&return_should);
+// Build a PHP 2-element numeric array for a Point. Output is a freshly
+// initialized zval owned by the caller; the caller decides how to attach
+// it (next_index, assoc, or write-into-arr).
+static void pointToZval(zval *out, const std::tuple<double, double>& pt)
+{
+    array_init(out);
+    add_next_index_double(out, std::get<0>(pt));
+    add_next_index_double(out, std::get<1>(pt));
+}
 
-        return std::make_shared<ColumnTuple>(columns);
-        break;
+// Geo nested types come back as clickhouse::ColumnArrayT::ArrayValueView,
+// not std::vector. The view is STL-iterable but not assignable to a
+// vector reference, so the helpers below take templated iterables.
+template <typename PointRange>
+static void ringRangeToZval(zval *out, const PointRange& ring)
+{
+    array_init(out);
+    for (auto pt : ring) {
+        zval pt_zv;
+        pointToZval(&pt_zv, pt);
+        add_next_index_zval(out, &pt_zv);
     }
+}
 
-    case Type::Code::Void:
-    {
-        throw std::runtime_error("can't support Void");
-    }
+template <typename RingRange>
+static void polygonRangeToZval(zval *out, const RingRange& poly)
+{
+    array_init(out);
+    for (auto ring : poly) {
+        zval r_zv;
+        ringRangeToZval(&r_zv, ring);
+        add_next_index_zval(out, &r_zv);
     }
+}
 
-    throw std::runtime_error("insertColumn runtime error.");
+// Attach a built-up nested zval to the parent according to (is_array,
+// fetch_mode, column_name). Mirrors the dispatch pattern Array/Tuple use.
+static void emitNestedZval(zval *arr, zval *built, const string& column_name, int8_t is_array, long fetch_mode)
+{
+    if (is_array) {
+        add_next_index_zval(arr, built);
+    } else if (fetch_mode & SC_FETCH_ONE) {
+        ZVAL_COPY_VALUE(arr, built);
+    } else {
+        sc_add_assoc_zval_ex(arr, column_name.c_str(), column_name.length(), built);
+    }
 }
 
-void convertToZval(zval *arr, const ColumnRef& columnRef, int row, string column_name, int8_t is_array)
+void convertToZval(zval *arr, const ColumnRef& columnRef, int row, const string& column_name, int8_t is_array, long fetch_mode)
 {
+    ConvertDepthGuard _depth_guard;  // shared with createColumn / insertColumn
     switch (columnRef->Type()->GetCode())
     {
     case Type::Code::UInt64:
-    {
-        auto col = (*columnRef->As<ColumnUInt64>())[row];
-        if (is_array)
-        {
-            add_next_index_long(arr, (long)col);
-        }
-        else
-        {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
-        }
+        emitIntColumn<ColumnUInt64>(arr, columnRef, row, column_name, is_array, fetch_mode);
         break;
-    }
     case Type::Code::UInt8:
+        emitIntColumn<ColumnUInt8>(arr, columnRef, row, column_name, is_array, fetch_mode);
+        break;
+    case Type::Code::UInt16:
+        emitIntColumn<ColumnUInt16>(arr, columnRef, row, column_name, is_array, fetch_mode);
+        break;
+    case Type::Code::UInt32:
+        emitIntColumn<ColumnUInt32>(arr, columnRef, row, column_name, is_array, fetch_mode);
+        break;
+    case Type::Code::IPv4:
     {
-        auto col = (*columnRef->As<ColumnUInt8>())[row];
+        /* As above, ColumnIPv4 is no longer a ColumnUInt32 subclass in
+         * v2.6.1; emit as canonical dotted-quad string via AsString(). */
+        auto col_ip = as_or_throw<ColumnIPv4>(columnRef, "IPv4 read");
+        std::string s = col_ip->AsString(row);
         if (is_array)
         {
-            add_next_index_long(arr, (long)col);
+            sc_add_next_index_stringl(arr, s.data(), s.size(), 1);
         }
         else
         {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
+            SC_SINGLE_STRING(s.data(), s.size());
         }
         break;
     }
-    case Type::Code::UInt16:
+    case Type::Code::Int8:
+        emitIntColumn<ColumnInt8>(arr, columnRef, row, column_name, is_array, fetch_mode);
+        break;
+    case Type::Code::Int16:
+        emitIntColumn<ColumnInt16>(arr, columnRef, row, column_name, is_array, fetch_mode);
+        break;
+    case Type::Code::Int32:
+        emitIntColumn<ColumnInt32>(arr, columnRef, row, column_name, is_array, fetch_mode);
+        break;
+    case Type::Code::Int64:
+        emitIntColumn<ColumnInt64>(arr, columnRef, row, column_name, is_array, fetch_mode);
+        break;
+    case Type::Code::UUID:
     {
-        auto col = (*columnRef->As<ColumnUInt16>())[row];
+        auto uuid_col = columnRef->As<ColumnUUID>();
+        if (!uuid_col) {
+            throw std::runtime_error("UUID read: column type mismatch");
+        }
+        auto col = (*uuid_col)[row];
+        char buf[33];
+        snprintf(buf, sizeof(buf), "%016llx%016llx",
+                 (unsigned long long)col.first,
+                 (unsigned long long)col.second);
         if (is_array)
         {
-            add_next_index_long(arr, (long)col);
+            sc_add_next_index_stringl(arr, buf, 32, 1);
         }
         else
         {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
+            SC_SINGLE_STRING(buf, 32);
         }
         break;
     }
-    case Type::Code::UInt32:
+    case Type::Code::Float32:
     {
-        auto col = (*columnRef->As<ColumnUInt32>())[row];
+        auto f32_col = columnRef->As<ColumnFloat32>();
+        if (!f32_col) {
+            throw std::runtime_error("Float32 read: column type mismatch");
+        }
+        /* Round-trip through %.6g (matching the prior `stringstream<<float`
+         * default precision) so 0.55f surfaces as 0.55 to PHP, not
+         * 0.5500000119. snprintf+strtod is heap-free vs the old stringstream
+         * pair. */
+        char buf[32];
+        snprintf(buf, sizeof(buf), "%.6g", (double)(*f32_col)[row]);
+        double d = strtod(buf, nullptr);
         if (is_array)
         {
-            add_next_index_long(arr, (long)col);
+            add_next_index_double(arr, d);
         }
         else
         {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
+            SC_SINGLE_DOUBLE(d);
         }
         break;
     }
-
-    case Type::Code::Int8:
+    case Type::Code::Float64:
     {
-        auto col = (*columnRef->As<ColumnInt8>())[row];
+        auto f64_col = columnRef->As<ColumnFloat64>();
+        if (!f64_col) {
+            throw std::runtime_error("Float64 read: column type mismatch");
+        }
+        double col = (double)(*f64_col)[row];
         if (is_array)
         {
-            add_next_index_long(arr, (long)col);
+            add_next_index_double(arr, col);
         }
         else
         {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
+            SC_SINGLE_DOUBLE(col);
         }
         break;
     }
-    case Type::Code::Int16:
-    {
-        auto col = (*columnRef->As<ColumnInt16>())[row];
-        if (is_array)
-        {
-            add_next_index_long(arr, (long)col);
+    case Type::Code::Decimal:
+    case Type::Code::Decimal32:
+    case Type::Code::Decimal64:
+    case Type::Code::Decimal128:
+    {
+        auto col = columnRef->As<ColumnDecimal>();
+        if (!col) {
+            throw std::runtime_error("Decimal read: column downcast failed");
+        }
+        // Format with the scale point so a value inserted as "12.34" reads
+        // back as "12.34", not the unscaled storage integer 1234. The
+        // decimal point and any leading-zero padding are inserted in
+        // place in the stack buffer so no std::string allocations fire
+        // per cell. Worst case: 1 byte sign + 39 digits + 1 '.' = 41.
+        auto dec_type = columnRef->Type()->As<DecimalType>();
+        size_t scale = dec_type ? dec_type->GetScale() : 0;
+        Int128 raw = col->At(row);
+        char buf[64];
+        size_t l = format_int128_dec(raw, buf);
+        if (scale > 0) {
+            bool neg = (buf[0] == '-');
+            size_t sign_off = neg ? 1 : 0;
+            size_t dlen = l - sign_off;
+            // If dlen <= scale we need to pad: insert (scale+1 - dlen)
+            // zeros after the sign, so "5" with scale 3 → "0.005"
+            // (sign_off + 5 chars: 0 . 0 0 5).
+            if (dlen <= scale) {
+                size_t pad = scale + 1 - dlen;
+                memmove(buf + sign_off + pad, buf + sign_off, dlen);
+                memset(buf + sign_off, '0', pad);
+                l += pad;
+                dlen += pad;
+            }
+            // Insert '.' before the last `scale` digits.
+            size_t dot_pos = sign_off + dlen - scale;
+            memmove(buf + dot_pos + 1, buf + dot_pos, scale);
+            buf[dot_pos] = '.';
+            ++l;
         }
-        else
-        {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
+        if (is_array) {
+            sc_add_next_index_stringl(arr, buf, l, 1);
+        } else {
+            SC_SINGLE_STRING(buf, l);
         }
         break;
     }
-    case Type::Code::Int32:
+    case Type::Code::String:
     {
-        auto col = (*columnRef->As<ColumnInt32>())[row];
+        auto s_col = columnRef->As<ColumnString>();
+        if (!s_col) {
+            throw std::runtime_error("String read: column type mismatch");
+        }
+        auto col = (*s_col)[row];
         if (is_array)
         {
-            add_next_index_long(arr, (long)col);
+            sc_add_next_index_stringl(arr, col.data(), col.length(), 1);
         }
         else
         {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
+            SC_SINGLE_STRING(col.data(), col.length());
         }
         break;
     }
-    case Type::Code::Int64:
+    case Type::Code::FixedString:
     {
-        auto col = (*columnRef->As<ColumnInt64>())[row];
+        // ColumnFixedString::At returns a string_view over the full fixed-size
+        // buffer, including trailing NULs added by ClickHouse to pad short
+        // values up to the column's declared width. Trim trailing NULs so the
+        // PHP-side value matches the original input.
+        auto fs_col = columnRef->As<ColumnFixedString>();
+        if (!fs_col) {
+            throw std::runtime_error("FixedString read: column type mismatch");
+        }
+        auto col = (*fs_col)[row];
+        size_t len = col.length();
+        while (len > 0 && col.data()[len - 1] == '\0') {
+            --len;
+        }
         if (is_array)
         {
-            add_next_index_long(arr, (long)col);
+            sc_add_next_index_stringl(arr, col.data(), len, 1);
         }
         else
         {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col);
+            SC_SINGLE_STRING(col.data(), len);
         }
         break;
     }
-
-    case Type::Code::UUID:
+    case Type::Code::IPv6:
     {
-        stringstream first;
-        stringstream second;
-        auto col = (*columnRef->As<ColumnUUID>())[row];
-        first<<std::setw(16)<<std::setfill('0')<<hex<<col.first;
-        second<<std::setw(16)<<std::setfill('0')<<hex<<col.second;
+        /* clickhouse-cpp v2.6.1 made ColumnIPv6 a sibling of ColumnFixedString
+         * (composition, not inheritance), so As<ColumnFixedString>() returns
+         * null and crashed every IPv6 read. Use AsString() to get the
+         * canonical "::1" form. */
+        auto col_ip = as_or_throw<ColumnIPv6>(columnRef, "IPv6 read");
+        std::string s = col_ip->AsString(row);
         if (is_array)
         {
-            sc_add_next_index_stringl(arr, (char*)(first.str() + second.str()).c_str(), (first.str() + second.str()).length(), 1);
+            sc_add_next_index_stringl(arr, s.data(), s.size(), 1);
         }
         else
         {
-            sc_add_assoc_stringl_ex(arr, column_name.c_str(), column_name.length(), (char*)(first.str() + second.str()).c_str(), (first.str() + second.str()).length(), 1);
+            SC_SINGLE_STRING(s.data(), s.size());
         }
         break;
     }
 
-    case Type::Code::Float32:
+    case Type::Code::DateTime:
     {
-        auto col = (*columnRef->As<ColumnFloat32>())[row];
-        stringstream stream;
-        stream<<col;
-        double d;
-        stream>>d;
-        if (is_array)
-        {
-            add_next_index_double(arr, d);
+        auto col = columnRef->As<ColumnDateTime>();
+        if (!col) {
+            throw std::runtime_error("DateTime read: column type mismatch");
         }
-        else
-        {
-            sc_add_assoc_double_ex(arr, column_name.c_str(), column_name.length(), d);
+        emitEpoch(arr, (std::time_t)col->At(row), "%Y-%m-%d %H:%M:%S",
+                  column_name, is_array, fetch_mode);
+        break;
+    }
+    case Type::Code::DateTime64:
+    {
+        auto col = columnRef->As<ColumnDateTime64>();
+        if (!col) {
+            throw std::runtime_error("DateTime64 read: column type mismatch");
+        }
+        size_t precision = columnRef->Type()->As<DateTime64Type>()->GetPrecision();
+        if (precision > 9) {
+            throw std::runtime_error("DateTime64 precision out of spec range (0..9)");
+        }
+        int64_t scale = 1;
+        for (size_t i = 0; i < precision; ++i) scale *= 10;
+        int64_t raw = col->At(row);
+        std::time_t whole = (std::time_t)(raw / scale);
+        int64_t frac = raw % scale;
+        if (frac < 0) { frac = -frac; }
+
+        if (fetch_mode & SC_FETCH_DATE_AS_STRINGS) {
+            char buffer[64];
+            size_t l = strftime(buffer, sizeof(buffer), "%Y-%m-%d %H:%M:%S", gmtime(&whole));
+            if (precision > 0 && l < sizeof(buffer)) {
+                int written = snprintf(buffer + l, sizeof(buffer) - l, ".%0*lld",
+                                       (int)precision, (long long)frac);
+                if (written > 0 && (size_t)written < sizeof(buffer) - l) {
+                    l += (size_t)written;
+                }
+            }
+            if (is_array) {
+                sc_add_next_index_stringl(arr, buffer, l, 1);
+            } else {
+                SC_SINGLE_STRING(buffer, l);
+            }
+        } else {
+            if (is_array) {
+                add_next_index_long(arr, (zend_long)raw);
+            } else if (fetch_mode & SC_FETCH_ONE) {
+                ZVAL_LONG(arr, (zend_long)raw);
+            } else {
+                sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_long)raw);
+            }
         }
         break;
     }
-    case Type::Code::Float64:
+    case Type::Code::Date:
     {
-        auto col = (*columnRef->As<ColumnFloat64>())[row];
-        if (is_array)
-        {
-            add_next_index_double(arr, (double)col);
-        }
-        else
-        {
-            sc_add_assoc_double_ex(arr, column_name.c_str(), column_name.length(), (double)col);
+        auto col = columnRef->As<ColumnDate>();
+        if (!col) {
+            throw std::runtime_error("Date read: column type mismatch");
         }
+        emitEpoch(arr, (std::time_t)col->At(row), "%Y-%m-%d",
+                  column_name, is_array, fetch_mode);
         break;
     }
-
-    case Type::Code::String:
+    case Type::Code::Date32:
     {
-        auto col = (*columnRef->As<ColumnString>())[row];
-        if (is_array)
-        {
-            sc_add_next_index_stringl(arr, (char*)col.c_str(), col.length(), 1);
-        }
-        else
-        {
-            sc_add_assoc_stringl_ex(arr, column_name.c_str(), column_name.length(), (char*)col.c_str(), col.length(), 1);
+        auto col = columnRef->As<ColumnDate32>();
+        if (!col) {
+            throw std::runtime_error("Date32 read: column type mismatch");
         }
+        emitEpoch(arr, (std::time_t)col->At(row), "%Y-%m-%d",
+                  column_name, is_array, fetch_mode);
         break;
     }
-    case Type::Code::FixedString:
-    {
-        auto col = (*columnRef->As<ColumnFixedString>())[row];
-        if (is_array)
-        {
-            sc_add_next_index_stringl(arr, (char*)col.c_str(), col.length(), 1);
+    case Type::Code::Time:
+    {
+        auto col = columnRef->As<ColumnTime>();
+        if (!col) {
+            throw std::runtime_error("Time read: column type mismatch");
+        }
+        int32_t v = col->At(row);
+        if (fetch_mode & SC_FETCH_DATE_AS_STRINGS) {
+            int abs_v = v < 0 ? -v : v;
+            char buffer[16];
+            int l = snprintf(buffer, sizeof(buffer), "%s%02d:%02d:%02d",
+                             v < 0 ? "-" : "", abs_v / 3600, (abs_v / 60) % 60, abs_v % 60);
+            if (is_array) {
+                sc_add_next_index_stringl(arr, buffer, l, 1);
+            } else {
+                SC_SINGLE_STRING(buffer, l);
+            }
+        } else {
+            if (is_array) {
+                add_next_index_long(arr, (zend_long)v);
+            } else if (fetch_mode & SC_FETCH_ONE) {
+                ZVAL_LONG(arr, (zend_long)v);
+            } else {
+                sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_long)v);
+            }
         }
-        else
-        {
-            sc_add_assoc_stringl_ex(arr, column_name.c_str(), column_name.length(), (char*)col.c_str(), strlen((char*)col.c_str()), 1);
+        break;
+    }
+    case Type::Code::Time64:
+    {
+        auto col = columnRef->As<ColumnTime64>();
+        if (!col) {
+            throw std::runtime_error("Time64 read: column type mismatch");
+        }
+        size_t precision = columnRef->Type()->As<Time64Type>()->GetPrecision();
+        if (precision > 9) {
+            throw std::runtime_error("Time64 precision out of spec range (0..9)");
+        }
+        int64_t scale = 1;
+        for (size_t i = 0; i < precision; ++i) scale *= 10;
+        int64_t raw = col->At(row);
+        if (fetch_mode & SC_FETCH_DATE_AS_STRINGS) {
+            int64_t whole = raw / scale;
+            int64_t frac = raw % scale;
+            if (frac < 0) frac = -frac;
+            int64_t abs_whole = whole < 0 ? -whole : whole;
+            char buffer[64];
+            int l = snprintf(buffer, sizeof(buffer), "%s%02lld:%02lld:%02lld",
+                             whole < 0 ? "-" : "",
+                             (long long)(abs_whole / 3600),
+                             (long long)((abs_whole / 60) % 60),
+                             (long long)(abs_whole % 60));
+            if (l < 0 || (size_t)l >= sizeof(buffer)) l = (int)sizeof(buffer) - 1;
+            if (precision > 0 && l > 0 && (size_t)l < sizeof(buffer)) {
+                int w = snprintf(buffer + l, sizeof(buffer) - l, ".%0*lld",
+                                 (int)precision, (long long)frac);
+                if (w > 0 && (size_t)w < sizeof(buffer) - (size_t)l) l += w;
+            }
+            if (is_array) {
+                sc_add_next_index_stringl(arr, buffer, l, 1);
+            } else {
+                SC_SINGLE_STRING(buffer, l);
+            }
+        } else {
+            if (is_array) {
+                add_next_index_long(arr, (zend_long)raw);
+            } else if (fetch_mode & SC_FETCH_ONE) {
+                ZVAL_LONG(arr, (zend_long)raw);
+            } else {
+                sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_long)raw);
+            }
         }
         break;
     }
-
-    case Type::Code::DateTime:
+    case Type::Code::Int128:
     {
-        auto col = columnRef->As<ColumnDateTime>();
-        if (is_array)
-        {
-            add_next_index_long(arr, (long)col->As<ColumnDateTime>()->At(row));
+        auto col = columnRef->As<ColumnInt128>();
+        if (!col) {
+            throw std::runtime_error("Int128 read: column type mismatch");
         }
-        else
-        {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col->As<ColumnDateTime>()->At(row));
+        char buf[41];
+        size_t l = format_int128_dec(col->At(row), buf);
+        if (is_array) {
+            sc_add_next_index_stringl(arr, buf, l, 1);
+        } else {
+            SC_SINGLE_STRING(buf, l);
         }
         break;
     }
-    case Type::Code::Date:
+    case Type::Code::UInt128:
     {
-        auto col = columnRef->As<ColumnDate>();
-        if (is_array)
-        {
-            add_next_index_long(arr, (long)col->As<ColumnDate>()->At(row));
+        auto col = columnRef->As<ColumnUInt128>();
+        if (!col) {
+            throw std::runtime_error("UInt128 read: column type mismatch");
         }
-        else
-        {
-            sc_add_assoc_long_ex(arr, column_name.c_str(), column_name.length(), (zend_ulong)col->As<ColumnDate>()->At(row));
+        char buf[40];
+        size_t l = format_uint128_dec(col->At(row), buf);
+        if (is_array) {
+            sc_add_next_index_stringl(arr, buf, l, 1);
+        } else {
+            SC_SINGLE_STRING(buf, l);
         }
         break;
     }
-
     case Type::Code::Array:
     {
         auto array = columnRef->As<ColumnArray>();
-        auto col = array->GetAsColumn(row);
-        zval *return_tmp;
-        SC_MAKE_STD_ZVAL(return_tmp);
-        array_init(return_tmp);
-        for (size_t i = 0; i < col->Size(); ++i)
-        {
-            convertToZval(return_tmp, col, i, "array", 1);
+        if (!array) {
+            throw std::runtime_error("Array read: column type mismatch");
         }
-        if (is_array)
-        {
-            add_next_index_zval(arr, return_tmp);
-        }
-        else
-        {
-            sc_add_assoc_zval_ex(arr, column_name.c_str(), column_name.length(), return_tmp);
+        auto col = array->GetAsColumn(row);
+        if (fetch_mode & SC_FETCH_ONE) {
+            array_init(arr);
+            for (size_t i = 0; i < col->Size(); ++i)
+            {
+                convertToZval(arr, col, i, "array", 1, 0);
+            }
+        } else {
+            zval *return_tmp;
+            SC_MAKE_STD_ZVAL(return_tmp);
+            array_init(return_tmp);
+            for (size_t i = 0; i < col->Size(); ++i)
+            {
+                convertToZval(return_tmp, col, i, "array", 1, 0);
+            }
+            if (is_array)
+            {
+                add_next_index_zval(arr, return_tmp);
+            }
+            else
+            {
+                sc_add_assoc_zval_ex(arr, column_name.c_str(), column_name.length(), return_tmp);
+            }
         }
         break;
     }
@@ -853,26 +2314,34 @@ void convertToZval(zval *arr, const ColumnRef& columnRef, int row, string column
     case Type::Code::Enum8:
     {
         auto array = columnRef->As<ColumnEnum8>();
+        if (!array) {
+            throw std::runtime_error("Enum8 read: column type mismatch");
+        }
+        std::string_view name = array->NameAt(row);
         if (is_array)
         {
-            sc_add_next_index_stringl(arr, (char*)array->NameAt(row).c_str(), array->NameAt(row).length(), 1);
+            sc_add_next_index_stringl(arr, name.data(), name.length(), 1);
         }
         else
         {
-            sc_add_assoc_stringl_ex(arr, column_name.c_str(), column_name.length(), (char*)array->NameAt(row).c_str(), array->NameAt(row).length(), 1);
+            SC_SINGLE_STRING(name.data(), name.length());
         }
         break;
     }
     case Type::Code::Enum16:
     {
         auto array = columnRef->As<ColumnEnum16>();
+        if (!array) {
+            throw std::runtime_error("Enum16 read: column type mismatch");
+        }
+        std::string_view name = array->NameAt(row);
         if (is_array)
         {
-            sc_add_next_index_stringl(arr, (char*)array->NameAt(row).c_str(), array->NameAt(row).length(), 1);
+            sc_add_next_index_stringl(arr, name.data(), name.length(), 1);
         }
         else
         {
-            sc_add_assoc_stringl_ex(arr, column_name.c_str(), column_name.length(), (char*)array->NameAt(row).c_str(), array->NameAt(row).length(), 1);
+            SC_SINGLE_STRING(name.data(), name.length());
         }
         break;
     }
@@ -880,6 +2349,9 @@ void convertToZval(zval *arr, const ColumnRef& columnRef, int row, string column
     case Type::Code::Nullable:
     {
         auto nullable = columnRef->As<ColumnNullable>();
+        if (!nullable) {
+            throw std::runtime_error("Nullable read: column type mismatch");
+        }
         if (nullable->IsNull(row))
         {
             if (is_array)
@@ -888,12 +2360,16 @@ void convertToZval(zval *arr, const ColumnRef& columnRef, int row, string column
             }
             else
             {
-                sc_add_assoc_null_ex(arr, column_name.c_str(), column_name.length());
+                if (fetch_mode & SC_FETCH_ONE) {
+                    ZVAL_NULL(arr);
+                } else {
+                    sc_add_assoc_null_ex(arr, column_name.c_str(), column_name.length());
+                }
             }
         }
         else
         {
-            convertToZval(arr, nullable->Nested(), row, column_name, 0);
+            convertToZval(arr, nullable->Nested(), row, column_name, is_array, fetch_mode);
         }
         break;
     }
@@ -901,21 +2377,397 @@ void convertToZval(zval *arr, const ColumnRef& columnRef, int row, string column
     case Type::Code::Tuple:
     {
         auto tuple = columnRef->As<ColumnTuple>();
-        zval *return_tmp;
-        SC_MAKE_STD_ZVAL(return_tmp);
-        array_init(return_tmp);
-        for (size_t i = 0; i < tuple->tupleSize(); ++i)
-        {
-            convertToZval(return_tmp, (*tuple)[i], row, "tuple", 1);
+        if (!tuple) {
+            throw std::runtime_error("Tuple read: column type mismatch");
         }
-        if (is_array)
-        {
-            add_next_index_zval(arr, return_tmp);
+        if (fetch_mode & SC_FETCH_ONE) {
+            array_init(arr);
+            for (size_t i = 0; i < tuple->TupleSize(); ++i)
+            {
+                convertToZval(arr, (*tuple)[i], row, "tuple", 1, 0);
+            }
+        } else {
+            zval *return_tmp;
+            SC_MAKE_STD_ZVAL(return_tmp);
+            array_init(return_tmp);
+            for (size_t i = 0; i < tuple->TupleSize(); ++i)
+            {
+                convertToZval(return_tmp, (*tuple)[i], row, "tuple", 1, 0);
+            }
+            if (is_array)
+            {
+                add_next_index_zval(arr, return_tmp);
+            }
+            else
+            {
+                sc_add_assoc_zval_ex(arr, column_name.c_str(), column_name.length(), return_tmp);
+            }
         }
-        else
-        {
-            sc_add_assoc_zval_ex(arr, column_name.c_str(), column_name.length(), return_tmp);
+        break;
+    }
+
+    case Type::Code::LowCardinality:
+    {
+        // Drop through ColumnLowCardinality::GetItem so the same code
+        // path covers LC(String), LC(FixedString), LC(Nullable(String)),
+        // and LC(Nullable(FixedString)) -- a NULL entry returns an
+        // ItemView with type Void regardless of the nested column.
+        auto lc = columnRef->As<ColumnLowCardinality>();
+        if (!lc) {
+            throw std::runtime_error("LowCardinality column downcast failed");
+        }
+        TypeRef nested = columnRef->Type()->As<LowCardinalityType>()->GetNestedType();
+        bool is_nullable = (nested->GetCode() == Type::Code::Nullable);
+        TypeRef inner = is_nullable
+            ? nested->As<NullableType>()->GetNestedType()
+            : nested;
+        if (inner->GetCode() != Type::Code::String &&
+            inner->GetCode() != Type::Code::FixedString) {
+            throw std::runtime_error("LowCardinality read only supports String / FixedString");
+        }
+
+        ItemView iv = lc->GetItem(row);
+        if (is_nullable && iv.type == Type::Code::Void) {
+            if (is_array) {
+                add_next_index_null(arr);
+            } else if (fetch_mode & SC_FETCH_ONE) {
+                ZVAL_NULL(arr);
+            } else {
+                sc_add_assoc_null_ex(arr, column_name.c_str(), column_name.length());
+            }
+            break;
+        }
+
+        std::string_view sv = iv.AsBinaryData();
+        // FixedString views include trailing NULs from server-side padding;
+        // trim them so the round-trip preserves the original input.
+        if (inner->GetCode() == Type::Code::FixedString) {
+            size_t len = sv.length();
+            while (len > 0 && sv.data()[len - 1] == '\0') {
+                --len;
+            }
+            sv = std::string_view(sv.data(), len);
+        }
+        if (is_array) {
+            sc_add_next_index_stringl(arr, sv.data(), sv.length(), 1);
+        } else {
+            SC_SINGLE_STRING(sv.data(), sv.length());
+        }
+        break;
+    }
+
+    case Type::Code::Map:
+    {
+        TypeRef map_type = columnRef->Type();
+        auto map_type_ref = map_type->As<MapType>();
+        if (!map_type_ref) {
+            throw std::runtime_error("Map read: type metadata is not MapType");
+        }
+        TypeRef key_type_ref = map_type_ref->GetKeyType();
+        TypeRef value_type_ref = map_type_ref->GetValueType();
+        Type::Code key_code = key_type_ref->GetCode();
+        Type::Code value_code = value_type_ref->GetCode();
+        auto map_col = columnRef->As<ColumnMap>();
+        if (!map_col) {
+            throw std::runtime_error("Map read: column type mismatch");
+        }
+        ColumnRef tuple_col = map_col->GetAsColumn(row);
+        auto tup = tuple_col->As<ColumnTuple>();
+        if (!tup) {
+            throw std::runtime_error("Map read: inner tuple type mismatch");
+        }
+        ColumnRef keys_any = (*tup)[0];
+        ColumnRef values_any = (*tup)[1];
+        size_t entry_count = keys_any->Size();
+        /* Defensive: a malformed/malicious server response could report
+         * key-count > value-count; subsequent At(i) on values would read
+         * past the value column. */
+        if (values_any->Size() < entry_count) {
+            throw std::runtime_error("Map column key/value size mismatch");
+        }
+
+        zval *map_zv;
+        SC_MAKE_STD_ZVAL(map_zv);
+        array_init(map_zv);
+
+        /* Pre-cast the key and value columns once per row instead of per
+         * entry. The keys_any / values_any column slices don't change
+         * across the entry loop; only the row index inside them does.
+         * Doing as_or_throw inside decodeKey would re-run a
+         * dynamic_pointer_cast for every map entry. For Map(Int64, V)
+         * with 100 entries × 1M rows that was 100M unnecessary casts.
+         * Only one of these typed pointers is populated for any given
+         * cell; the others stay null. */
+        std::shared_ptr<ColumnString>  k_str_col;
+        std::shared_ptr<ColumnInt64>   k_i64_col;
+        std::shared_ptr<ColumnUInt64>  k_u64_col;
+        std::shared_ptr<ColumnInt32>   k_i32_col;
+        std::shared_ptr<ColumnUInt32>  k_u32_col;
+        std::shared_ptr<ColumnInt16>   k_i16_col;
+        std::shared_ptr<ColumnUInt16>  k_u16_col;
+        std::shared_ptr<ColumnInt8>    k_i8_col;
+        std::shared_ptr<ColumnUInt8>   k_u8_col;
+        std::shared_ptr<ColumnFloat32> k_f32_col;
+        std::shared_ptr<ColumnFloat64> k_f64_col;
+        std::shared_ptr<ColumnUUID>    k_uuid_col;
+        switch (key_code) {
+            case Type::Code::String:  k_str_col  = as_or_throw<ColumnString>(keys_any, "Map key String"); break;
+            case Type::Code::Int64:   k_i64_col  = as_or_throw<ColumnInt64>(keys_any, "Map key Int64"); break;
+            case Type::Code::UInt64:  k_u64_col  = as_or_throw<ColumnUInt64>(keys_any, "Map key UInt64"); break;
+            case Type::Code::Int32:   k_i32_col  = as_or_throw<ColumnInt32>(keys_any, "Map key Int32"); break;
+            case Type::Code::UInt32:  k_u32_col  = as_or_throw<ColumnUInt32>(keys_any, "Map key UInt32"); break;
+            case Type::Code::Int16:   k_i16_col  = as_or_throw<ColumnInt16>(keys_any, "Map key Int16"); break;
+            case Type::Code::UInt16:  k_u16_col  = as_or_throw<ColumnUInt16>(keys_any, "Map key UInt16"); break;
+            case Type::Code::Int8:    k_i8_col   = as_or_throw<ColumnInt8>(keys_any, "Map key Int8"); break;
+            case Type::Code::UInt8:   k_u8_col   = as_or_throw<ColumnUInt8>(keys_any, "Map key UInt8"); break;
+            case Type::Code::Float32: k_f32_col  = as_or_throw<ColumnFloat32>(keys_any, "Map key Float32"); break;
+            case Type::Code::Float64: k_f64_col  = as_or_throw<ColumnFloat64>(keys_any, "Map key Float64"); break;
+            case Type::Code::UUID:    k_uuid_col = as_or_throw<ColumnUUID>(keys_any, "Map key UUID"); break;
+            default:
+                throw std::runtime_error("Map read: unsupported key type " + key_type_ref->GetName());
+        }
+
+        // Same hoist for value column: one cast per Map cell, not per
+        // entry. Only the pointer matching value_code is populated.
+        std::shared_ptr<ColumnString>  v_str_col;
+        std::shared_ptr<ColumnInt64>   v_i64_col;
+        std::shared_ptr<ColumnUInt64>  v_u64_col;
+        std::shared_ptr<ColumnInt32>   v_i32_col;
+        std::shared_ptr<ColumnUInt32>  v_u32_col;
+        std::shared_ptr<ColumnInt16>   v_i16_col;
+        std::shared_ptr<ColumnUInt16>  v_u16_col;
+        std::shared_ptr<ColumnInt8>    v_i8_col;
+        std::shared_ptr<ColumnUInt8>   v_u8_col;
+        std::shared_ptr<ColumnFloat32> v_f32_col;
+        std::shared_ptr<ColumnFloat64> v_f64_col;
+        std::shared_ptr<ColumnUUID>    v_uuid_col;
+        switch (value_code) {
+            case Type::Code::String:  v_str_col  = as_or_throw<ColumnString>(values_any, "Map value String"); break;
+            case Type::Code::Int64:   v_i64_col  = as_or_throw<ColumnInt64>(values_any, "Map value Int64"); break;
+            case Type::Code::UInt64:  v_u64_col  = as_or_throw<ColumnUInt64>(values_any, "Map value UInt64"); break;
+            case Type::Code::Int32:   v_i32_col  = as_or_throw<ColumnInt32>(values_any, "Map value Int32"); break;
+            case Type::Code::UInt32:  v_u32_col  = as_or_throw<ColumnUInt32>(values_any, "Map value UInt32"); break;
+            case Type::Code::Int16:   v_i16_col  = as_or_throw<ColumnInt16>(values_any, "Map value Int16"); break;
+            case Type::Code::UInt16:  v_u16_col  = as_or_throw<ColumnUInt16>(values_any, "Map value UInt16"); break;
+            case Type::Code::Int8:    v_i8_col   = as_or_throw<ColumnInt8>(values_any, "Map value Int8"); break;
+            case Type::Code::UInt8:   v_u8_col   = as_or_throw<ColumnUInt8>(values_any, "Map value UInt8"); break;
+            case Type::Code::Float32: v_f32_col  = as_or_throw<ColumnFloat32>(values_any, "Map value Float32"); break;
+            case Type::Code::Float64: v_f64_col  = as_or_throw<ColumnFloat64>(values_any, "Map value Float64"); break;
+            case Type::Code::UUID:    v_uuid_col = as_or_throw<ColumnUUID>(values_any, "Map value UUID"); break;
+            default:
+                throw std::runtime_error("Map read: unsupported value type " + value_type_ref->GetName());
+        }
+
+        // Decode a key column at row i into one of three forms: string,
+        // long integer, or double. PHP arrays only key by string or
+        // long; doubles get formatted to a canonical string key.
+        auto decodeKey = [&](size_t i, std::string &str_buf, zend_long &long_out, double &dbl_out) -> int {
+            // Returns 0 = string, 1 = long, 2 = double-as-string.
+            switch (key_code) {
+                case Type::Code::String: {
+                    std::string_view kv = (*k_str_col)[i];
+                    str_buf.assign(kv.data(), kv.length());
+                    return 0;
+                }
+                case Type::Code::Int64:   long_out = (zend_long)k_i64_col->At(i);  return 1;
+                case Type::Code::UInt64: {
+                    /* UInt64 values above ZEND_LONG_MAX (2^63-1) lose
+                     * unsigned semantics if cast to zend_long, and PHP
+                     * array keys can't be unsigned, so distinct large
+                     * UInt64 keys would otherwise collapse to the same
+                     * negative signed-key. Promote to string when the
+                     * value doesn't fit a signed PHP integer. */
+                    uint64_t uk = (uint64_t)k_u64_col->At(i);
+                    if (uk > (uint64_t)ZEND_LONG_MAX) {
+                        char buf[32];
+                        int len = snprintf(buf, sizeof(buf), "%" PRIu64, uk);
+                        str_buf.assign(buf, len);
+                        return 0;
+                    }
+                    long_out = (zend_long)uk;
+                    return 1;
+                }
+                case Type::Code::Int32:   long_out = (zend_long)k_i32_col->At(i);  return 1;
+                case Type::Code::UInt32:  long_out = (zend_long)k_u32_col->At(i);  return 1;
+                case Type::Code::Int16:   long_out = (zend_long)k_i16_col->At(i);  return 1;
+                case Type::Code::UInt16:  long_out = (zend_long)k_u16_col->At(i);  return 1;
+                case Type::Code::Int8:    long_out = (zend_long)k_i8_col->At(i);   return 1;
+                case Type::Code::UInt8:   long_out = (zend_long)k_u8_col->At(i);   return 1;
+                case Type::Code::Float32: dbl_out  = (double)k_f32_col->At(i);     return 2;
+                case Type::Code::Float64: dbl_out  = (double)k_f64_col->At(i);     return 2;
+                case Type::Code::UUID: {
+                    UUID u = k_uuid_col->At(i);
+                    char buf[64];
+                    snprintf(buf, sizeof(buf), "%08x-%04x-%04x-%04x-%012llx",
+                             (uint32_t)(u.first >> 32),
+                             (uint16_t)((u.first >> 16) & 0xffff),
+                             (uint16_t)(u.first & 0xffff),
+                             (uint16_t)(u.second >> 48),
+                             (unsigned long long)(u.second & 0xffffffffffffull));
+                    str_buf.assign(buf);
+                    return 0;
+                }
+                default:
+                    throw std::runtime_error("Map read: unsupported key type " + key_type_ref->GetName());
+            }
+        };
+
+        /* Format a Float key as a locale-independent decimal string. The
+         * naive snprintf("%.17g") honors LC_NUMERIC, so the same Float64
+         * map key would surface under a different PHP array key under
+         * setlocale(LC_NUMERIC, 'de_DE'). php_gcvt with explicit '.' is
+         * the same fix CR-303 applied at the SQL parameter boundary. */
+        auto fmtFloatKey = [](double dk, char *buf, size_t bufsz) -> int {
+            php_gcvt(dk, 17, '.', 'e', buf);
+            (void)bufsz;
+            return (int)strlen(buf);
+        };
+
+        // Helper: add (string|long-as-string) keyed value into map_zv.
+        // Handles all three key categories returned by decodeKey.
+        auto addStrL = [&](int kkind, const std::string &sb, zend_long lk, double dk,
+                           const char *vptr, size_t vlen) {
+            if (kkind == 0) {
+                sc_add_assoc_stringl_ex(map_zv, sb.c_str(), sb.length(), vptr, vlen, 1);
+            } else if (kkind == 1) {
+                add_index_stringl(map_zv, lk, vptr, vlen);
+            } else {
+                char kbuf[64];
+                int klen = fmtFloatKey(dk, kbuf, sizeof(kbuf));
+                sc_add_assoc_stringl_ex(map_zv, kbuf, klen, vptr, vlen, 1);
+            }
+        };
+        auto addLong = [&](int kkind, const std::string &sb, zend_long lk, double dk, zend_long lv) {
+            if (kkind == 0) {
+                add_assoc_long_ex(map_zv, sb.c_str(), sb.length(), lv);
+            } else if (kkind == 1) {
+                add_index_long(map_zv, lk, lv);
+            } else {
+                char kbuf[64];
+                int klen = fmtFloatKey(dk, kbuf, sizeof(kbuf));
+                add_assoc_long_ex(map_zv, kbuf, klen, lv);
+            }
+        };
+        auto addDbl = [&](int kkind, const std::string &sb, zend_long lk, double dk, double dv) {
+            if (kkind == 0) {
+                add_assoc_double_ex(map_zv, sb.c_str(), sb.length(), dv);
+            } else if (kkind == 1) {
+                add_index_double(map_zv, lk, dv);
+            } else {
+                char kbuf[64];
+                int klen = fmtFloatKey(dk, kbuf, sizeof(kbuf));
+                add_assoc_double_ex(map_zv, kbuf, klen, dv);
+            }
+        };
+
+        for (size_t i = 0; i < entry_count; ++i) {
+            std::string str_key_buf;
+            zend_long long_key = 0;
+            double dbl_key = 0.0;
+            int kkind = decodeKey(i, str_key_buf, long_key, dbl_key);
+
+            // Decode value, dispatch by value type.
+            if (value_code == Type::Code::String) {
+                std::string_view vv = (*v_str_col)[i];
+                addStrL(kkind, str_key_buf, long_key, dbl_key, vv.data(), vv.length());
+            } else if (value_code == Type::Code::UInt64) {
+                /* UInt64 values above ZEND_LONG_MAX surface as
+                 * negatives if cast to zend_long; emit as a string
+                 * value so callers can round-trip safely. Same fix
+                 * as the scalar UInt64 read path. */
+                uint64_t uv = (uint64_t)v_u64_col->At(i);
+                if (uv > (uint64_t)ZEND_LONG_MAX) {
+                    char buf[32];
+                    int len = snprintf(buf, sizeof(buf), "%" PRIu64, uv);
+                    addStrL(kkind, str_key_buf, long_key, dbl_key, buf, len);
+                } else {
+                    addLong(kkind, str_key_buf, long_key, dbl_key, (zend_long)uv);
+                }
+            } else if (value_code == Type::Code::Int64
+                    || value_code == Type::Code::Int32 || value_code == Type::Code::UInt32
+                    || value_code == Type::Code::Int16 || value_code == Type::Code::UInt16
+                    || value_code == Type::Code::Int8  || value_code == Type::Code::UInt8) {
+                zend_long lv = 0;
+                switch (value_code) {
+                    case Type::Code::Int64:  lv = (zend_long)v_i64_col->At(i); break;
+                    case Type::Code::Int32:  lv = (zend_long)v_i32_col->At(i); break;
+                    case Type::Code::UInt32: lv = (zend_long)v_u32_col->At(i); break;
+                    case Type::Code::Int16:  lv = (zend_long)v_i16_col->At(i); break;
+                    case Type::Code::UInt16: lv = (zend_long)v_u16_col->At(i); break;
+                    case Type::Code::Int8:   lv = (zend_long)v_i8_col->At(i);  break;
+                    case Type::Code::UInt8:  lv = (zend_long)v_u8_col->At(i);  break;
+                    default: break;
+                }
+                addLong(kkind, str_key_buf, long_key, dbl_key, lv);
+            } else if (value_code == Type::Code::Float64 || value_code == Type::Code::Float32) {
+                double dv = (value_code == Type::Code::Float64)
+                    ? (double)v_f64_col->At(i)
+                    : (double)v_f32_col->At(i);
+                addDbl(kkind, str_key_buf, long_key, dbl_key, dv);
+            } else if (value_code == Type::Code::UUID) {
+                UUID u = v_uuid_col->At(i);
+                char buf[64];
+                int blen = snprintf(buf, sizeof(buf), "%08x-%04x-%04x-%04x-%012llx",
+                         (uint32_t)(u.first >> 32),
+                         (uint16_t)((u.first >> 16) & 0xffff),
+                         (uint16_t)(u.first & 0xffff),
+                         (uint16_t)(u.second >> 48),
+                         (unsigned long long)(u.second & 0xffffffffffffull));
+                addStrL(kkind, str_key_buf, long_key, dbl_key, buf, blen);
+            } else {
+                throw std::runtime_error("Map read: unsupported value type " + value_type_ref->GetName());
+            }
+        }
+
+        if (is_array) {
+            add_next_index_zval(arr, map_zv);
+        } else if (fetch_mode & SC_FETCH_ONE) {
+            ZVAL_COPY_VALUE(arr, map_zv);
+        } else {
+            sc_add_assoc_zval_ex(arr, column_name.c_str(), column_name.length(), map_zv);
+        }
+        break;
+    }
+
+    case Type::Code::Point:
+    {
+        auto col = columnRef->As<ColumnPoint>();
+        if (!col) throw std::runtime_error("Point column downcast failed");
+        zval pt_zv;
+        pointToZval(&pt_zv, col->At(row));
+        emitNestedZval(arr, &pt_zv, column_name, is_array, fetch_mode);
+        break;
+    }
+    case Type::Code::Ring:
+    {
+        auto col = columnRef->As<ColumnRing>();
+        if (!col) throw std::runtime_error("Ring column downcast failed");
+        zval r_zv;
+        ringRangeToZval(&r_zv, col->At(row));
+        emitNestedZval(arr, &r_zv, column_name, is_array, fetch_mode);
+        break;
+    }
+    case Type::Code::Polygon:
+    {
+        auto col = columnRef->As<ColumnPolygon>();
+        if (!col) throw std::runtime_error("Polygon column downcast failed");
+        zval p_zv;
+        polygonRangeToZval(&p_zv, col->At(row));
+        emitNestedZval(arr, &p_zv, column_name, is_array, fetch_mode);
+        break;
+    }
+    case Type::Code::MultiPolygon:
+    {
+        auto col = columnRef->As<ColumnMultiPolygon>();
+        if (!col) throw std::runtime_error("MultiPolygon column downcast failed");
+        zval mp_zv;
+        array_init(&mp_zv);
+        for (auto poly : col->At(row)) {
+            zval poly_zv;
+            polygonRangeToZval(&poly_zv, poly);
+            add_next_index_zval(&mp_zv, &poly_zv);
         }
+        emitNestedZval(arr, &mp_zv, column_name, is_array, fetch_mode);
         break;
     }
 
@@ -923,6 +2775,8 @@ void convertToZval(zval *arr, const ColumnRef& columnRef, int row, string column
     {
         throw std::runtime_error("can't support Void");
     }
+    default:
+        throw std::runtime_error("convertToZval: unsupported type code: " + columnRef->Type()->GetName());
     }
 }
 
diff --git a/typesToPhp.hpp b/typesToPhp.hpp
index a08b67f..73306c0 100644
--- a/typesToPhp.hpp
+++ b/typesToPhp.hpp
@@ -1,8 +1,8 @@
 /*
   +----------------------------------------------------------------------+
-  | SeasClick                                                            |
+  | php_clickhouse                                                       |
   +----------------------------------------------------------------------+
-  | Copyright (c) 1997-2018 The PHP Group                                |
+  | Copyright (c) 1997-2026 The PHP Group                                |
   +----------------------------------------------------------------------+
   | This source file is subject to version 3.01 of the PHP license,      |
   | that is bundled with this package in the file LICENSE, and is        |
@@ -12,19 +12,35 @@
   | obtain it through the world-wide-web, please send a note to          |
   | license@php.net so we can mail you a copy immediately.               |
   +----------------------------------------------------------------------+
-  | Author:  SeasX Group <ahhhh.wang@gmail.com>                          |
+  | Author: Ilia Alshanetsky <ilia@ilia.ws>                              |
   +----------------------------------------------------------------------+
 */
-using namespace clickhouse;
-using namespace std;
+#include <string>
 
-ColumnRef createColumn(TypeRef type);
+/*
+ * RAII wrapper for the zend_string returned by zval_get_string. Used at
+ * PHP-to-C boundaries where the surrounding code can throw without
+ * forcing every site to write try { ... } catch { release; throw; }.
+ */
+struct ZStrGuard {
+    zend_string *s;
+    explicit ZStrGuard(zval *zv) : s(zval_get_string(zv)) {}
+    ~ZStrGuard() { if (s) zend_string_release(s); }
+    ZStrGuard(const ZStrGuard&) = delete;
+    ZStrGuard& operator=(const ZStrGuard&) = delete;
+    const char *val() const { return ZSTR_VAL(s); }
+    size_t      len() const { return ZSTR_LEN(s); }
+};
+
+clickhouse::ColumnRef createColumn(clickhouse::TypeRef type);
 
-ColumnRef insertColumn(TypeRef type, zval *value_zval);
+clickhouse::ColumnRef insertColumn(clickhouse::TypeRef type, zval *value_zval);
 
-void convertToZval(zval *arr, const ColumnRef& columnRef, int row, string column_name, int8_t is_array);
+void convertToZval(zval *arr, const clickhouse::ColumnRef& columnRef, int row,
+                   const std::string& column_name, int8_t is_array, long fetch_mode);
 
-void zvalToBlock(Block& blockDes, Block& blockSrc, zend_ulong num_key, zval *value_zval);
+void zvalToBlock(clickhouse::Block& blockDes, clickhouse::Block& blockSrc,
+                 zend_ulong num_key, zval *value_zval);
 
 /*
  * Local variables: