refactor: split spnl monolith into 5 crates under crates/ layout

.github/scripts/verify-feature-consistency.sh

@@ -7,21 +7,25 @@ set -e
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REPO_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
 CORE_YML="$REPO_ROOT/.github/workflows/core.yml"
-SPNL_CARGO_TOML="$REPO_ROOT/spnl/Cargo.toml"
+SPNL_CARGO_TOML="$REPO_ROOT/crates/spnl/Cargo.toml"
 
 echo "Verifying feature consistency between core.yml and spnl/Cargo.toml..."
 
 # Extract features from spnl/Cargo.toml
-# Parse the [features] section, excluding only:
+# Parse the [features] section, excluding:
 # - default = [...]
 # - local (CPU inference - mistralrs unconditionally depends on CUDA via candle-core/cudarc)
 # - cuda* (CUDA-specific features)
+# - ffi, pypi, run_py (pyo3/extension-module can't link into test binary; tested via python.yml)
 CARGO_FEATURES=$(sed -n '/^\[features\]/,/^\[/p' "$SPNL_CARGO_TOML" | \
     grep -E '^[a-z]' | \
     sed 's/ *=.*//' | \
     grep -v '^default$' | \
     grep -v '^local$' | \
     grep -v '^cuda' | \
+    grep -v '^ffi$' | \
+    grep -v '^pypi$' | \
+    grep -v '^run_py$' | \
     sort)
 
 # Extract features from core.yml cargo test command
@@ -34,7 +38,7 @@ CORE_YML_FEATURES=$(grep 'cargo test -p spnl --features' "$CORE_YML" | \
     uniq)
 
 echo ""
-echo "Features in spnl/Cargo.toml (excluding default, local, cuda*):"
+echo "Features in spnl/Cargo.toml (excluding default, local, cuda*, ffi, pypi, run_py):"
 echo "$CARGO_FEATURES"
 echo ""
 echo "Features tested in core.yml (macOS cargo test -p spnl):"
@@ -67,6 +71,7 @@ echo "Excluded features (not tested in core.yml for spnl package):"
 echo "  - default (tested implicitly)"
 echo "  - local (CPU inference - mistralrs unconditionally depends on CUDA via candle-core/cudarc)"
 echo "  - cuda* (CUDA-specific features)"
+echo "  - ffi, pypi, run_py (pyo3/extension-module can't link into test binary; tested via python.yml)"
 echo ""
 echo "Note: 'local' feature IS tested in spnl-cli package tests, which uses it differently."
 

.github/workflows/README-release-cli.md

@@ -251,7 +251,7 @@ To add support for a new platform, add an entry to the `matrix.platform` array:
 Modify the `cargo build` command in the workflow:
 
 ```yaml
-cargo build --features your,features,here --release --package spnl-cli --target $TARGET
+cargo build --manifest-path crates/spnl-cli/Cargo.toml --features your,features,here --release --target $TARGET
 ```
 
 ### Adjusting Compression
@@ -374,5 +374,5 @@ sha256sum -c checksums.txt
 ## Related Files
 
 - Workflow: `.github/workflows/release-cli.yml`
-- CLI Cargo.toml: `cli/Cargo.toml`
+- CLI Cargo.toml: `crates/spnl-cli/Cargo.toml`
 - Main Cargo.toml: `Cargo.toml`

.github/workflows/RELEASE-GUIDE.md

@@ -12,7 +12,7 @@
 
 3. Commit and push changes:
    ```bash
-   git add cli/Cargo.toml Cargo.lock
+   git add crates/spnl-cli/Cargo.toml Cargo.lock
    git commit -m "Bump version to 0.14.0"
    git push origin main
    ```
@@ -118,7 +118,7 @@ Users on Apple Silicon can run either version (Rosetta 2 compatibility).
 
 Before publishing a release:
 
-- [ ] Version bumped in `cli/Cargo.toml`
+- [ ] Version bumped in `crates/spnl-cli/Cargo.toml`
 - [ ] `Cargo.lock` updated
 - [ ] Changes committed and pushed to main
 - [ ] Release notes prepared

.github/workflows/core.yml

@@ -88,16 +88,21 @@ jobs:
             # IMPORTANT: Cannot use --all-features because mistralrs (used by local/metal/cuda features)
             # unconditionally depends on candle-core, which depends on cudarc (CUDA).
             #
-            # When adding new features to spnl/Cargo.toml, update this list!
-            # Current features tested (from spnl/Cargo.toml):
-            # - Included: cli_support, print, lisp, run, ollama, openai, gemini, pull, yaml, metal, rag, rag-deep-debug, spnl-api, vllm, k8s, gce, ffi, pypi, run_py, tok, openssl-vendored
+            # When adding new features to crates/spnl/Cargo.toml, update this list!
+            # Current features tested (from crates/spnl/Cargo.toml):
+            # - Included: cli_support, print, lisp, run, ollama, openai, gemini, pull, yaml, metal, rag, rag-deep-debug, spnl-api, vllm, k8s, gce, tok, openssl-vendored
+            # - Excluded: ffi, pypi, run_py (pyo3/extension-module can't link into test binary; tested via python.yml)
             # - Excluded: local (CPU inference - pulls in CUDA deps via mistralrs)
             # - Excluded: cuda, cuda-flash-attn, cuda-flash-attn-v3 (CUDA features)
-            cargo test -p spnl --features cli_support,print,lisp,run,ollama,openai,gemini,pull,yaml,metal,rag,rag-deep-debug,spnl-api,vllm,k8s,gce,ffi,pypi,run_py,tok,openssl-vendored -- --nocapture
-            cargo test -p spnl-cli --features bench,rag,spnl-api,vllm,k8s,gce,local,metal -- --nocapture
+            cargo test -p spnl-core --features lisp,yaml,tok -- --nocapture
+            cargo test -p spnl-run --features cli_support,print,lisp,run,ollama,openai,gemini,pull,yaml,metal,rag,rag-deep-debug,spnl-api,vllm,k8s,gce,tok,openssl-vendored -- --nocapture
+            cargo test -p spnl --features cli_support,print,lisp,run,ollama,openai,gemini,pull,yaml,metal,rag,rag-deep-debug,spnl-api,vllm,k8s,gce,tok,openssl-vendored -- --nocapture
+            cargo test --manifest-path crates/spnl-cli/Cargo.toml --features bench,rag,spnl-api,vllm,k8s,gce,local,metal -- --nocapture
           else
             # Test default features on Linux (no GPU features)
-            cargo test -p spnl -p spnl-cli -- --nocapture
+            # spnl-cli is excluded from workspace, test separately
+            cargo test -p spnl-core -p spnl-run -p spnl -- --nocapture
+            cargo test --manifest-path crates/spnl-cli/Cargo.toml -- --nocapture
           fi
 
       - name: cargo clippy
@@ -109,8 +114,12 @@ jobs:
         # When adding new features, update the feature list above in cargo test!
         if: runner.os == 'macOS'
         run: |
-          cargo clippy -p spnl --features cli_support,print,lisp,run,ollama,openai,gemini,pull,yaml,metal,rag,rag-deep-debug,spnl-api,vllm,k8s,gce,ffi,pypi,run_py,tok,openssl-vendored --tests --no-deps -- -D warnings
-          cargo clippy -p spnl-cli --features bench,rag,spnl-api,vllm,k8s,gce,local,metal --tests --no-deps -- -D warnings
+          cargo clippy -p spnl-core --features lisp,yaml,tok --tests --no-deps -- -D warnings
+          cargo clippy -p spnl-run --features cli_support,print,lisp,run,ollama,openai,gemini,pull,yaml,metal,rag,rag-deep-debug,spnl-api,vllm,k8s,gce,tok,openssl-vendored --tests --no-deps -- -D warnings
+          cargo clippy -p spnl --features cli_support,print,lisp,run,ollama,openai,gemini,pull,yaml,metal,rag,rag-deep-debug,spnl-api,vllm,k8s,gce,tok,openssl-vendored --tests --no-deps -- -D warnings
+          cargo clippy --manifest-path crates/spnl-cli/Cargo.toml --features bench,rag,spnl-api,vllm,k8s,gce,local,metal --tests --no-deps -- -D warnings
 
       - name: rustfmt
-        run: cargo fmt --all -- --check
+        run: |
+          cargo fmt --all -- --check
+          cargo fmt --manifest-path crates/spnl-cli/Cargo.toml -- --check

.github/workflows/publish_crate.yml

@@ -39,7 +39,16 @@ jobs:
       - name: Publish
         # Don't use --all-features to avoid CUDA compilation on macOS
         # Users can enable CUDA features explicitly when needed
-        run: cargo publish --manifest-path spnl/Cargo.toml $DRY_RUN
+        run: |
+          cargo publish --manifest-path crates/spnl-core/Cargo.toml $DRY_RUN
+          # spnl-run and spnl depend on spnl-core being on crates.io, so
+          # dry-run (PR) can only validate spnl-core; real publish runs all three.
+          if [ "$DRY_RUN" = " " ]; then
+            sleep 30  # wait for crates.io to index spnl-core
+            cargo publish --manifest-path crates/spnl-run/Cargo.toml
+            sleep 30  # wait for crates.io to index spnl-run
+            cargo publish --manifest-path crates/spnl/Cargo.toml
+          fi
         env:
           CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ github.event_name == 'pull_request' && '--dry-run' || ' ' }}

.github/workflows/publish_pypi.yml

@@ -1,7 +1,7 @@
 # This file is autogenerated by maturin v1.9.4
 # To update, run
 #
-#    maturin generate-ci -m spnl/Cargo.toml github
+#    maturin generate-ci -m crates/spnl-ffi/Cargo.toml github
 #
 name: publish_pypi
 
@@ -54,7 +54,7 @@ jobs:
         uses: ./maturin-action #PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path spnl/Cargo.toml --no-default-features -F tok,run_py
+          args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path crates/spnl-ffi/Cargo.toml --no-default-features -F tok,run_py --strip
           sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
           manylinux: 2_28
       - name: Upload wheels
@@ -89,7 +89,7 @@ jobs:
   #       uses: ./maturin-action #PyO3/maturin-action@v1
   #       with:
   #         target: ${{ matrix.platform.target }}
-  #         args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path spnl/Cargo.toml
+  #         args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path crates/spnl-ffi/Cargo.toml
   #         sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
   #         manylinux: auto #musllinux_1_2
   #     - name: Upload wheels
@@ -126,7 +126,7 @@ jobs:
         uses: ./maturin-action #PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path spnl/Cargo.toml --no-default-features -F tok,run_py
+          args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path crates/spnl-ffi/Cargo.toml --no-default-features -F tok,run_py --strip
           sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
       - name: Upload wheels
         if: ${{ github.event_name != 'pull_request' }}
@@ -160,7 +160,7 @@ jobs:
         uses: ./maturin-action #PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path spnl/Cargo.toml --no-default-features -F tok,run_py
+          args: ${{ env.BUILD_RELEASE_ARG }} --out dist --find-interpreter --manifest-path crates/spnl-ffi/Cargo.toml --no-default-features -F tok,run_py --strip
           sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
       - name: Upload wheels
         if: ${{ github.event_name != 'pull_request' }}
@@ -181,7 +181,7 @@ jobs:
         uses: ./maturin-action #PyO3/maturin-action@v1
         with:
           command: sdist
-          args: --out dist --manifest-path spnl/Cargo.toml
+          args: --out dist --manifest-path crates/spnl-ffi/Cargo.toml
       - name: Upload sdist
         uses: actions/upload-artifact@v6
         with:

.github/workflows/python.yml

@@ -51,7 +51,7 @@ jobs:
       - name: install maturin
         run: pip install maturin
       - name: maturin develop
-        run: maturin develop -m spnl/Cargo.toml --no-default-features -F pypi
+        run: maturin develop -m crates/spnl-ffi/Cargo.toml --no-default-features -F pypi,tok
 
       - name: invoke the python tokenize api
         run: |

.github/workflows/release-cli.yml

@@ -171,7 +171,7 @@ jobs:
         run: |
           # Only use openssl-vendored for musl builds (static linking)
           # Other platforms use system OpenSSL
-          cargo build --features $LOCAL_FEATURE,bench,rag,k8s,gce,spnl-api $BUILD_RELEASE_ARG --package spnl-cli --target $TARGET
+          cargo build --manifest-path crates/spnl-cli/Cargo.toml --target-dir target --features $LOCAL_FEATURE,bench,rag,k8s,gce,spnl-api $BUILD_RELEASE_ARG --target $TARGET
 
       - name: Check for Apple signing secrets
         if: runner.os == 'macOS'
@@ -329,7 +329,7 @@ jobs:
       - name: Build CLI
         shell: sh
         run: |
-          cargo build --features $LOCAL_FEATURE,bench,rag,k8s,gce,spnl-api,openssl-vendored $BUILD_RELEASE_ARG --package spnl-cli --target $TARGET
+          cargo build --manifest-path crates/spnl-cli/Cargo.toml --target-dir target --features $LOCAL_FEATURE,bench,rag,k8s,gce,spnl-api,openssl-vendored $BUILD_RELEASE_ARG --target $TARGET
 
       - name: Prepare binary
         shell: sh
@@ -516,8 +516,8 @@ jobs:
           echo "Building with features: $FEATURES,rag,k8s,gce,spnl-api"
           echo "CUDA_COMPUTE_CAP: $CUDA_COMPUTE_CAP"
           echo "CUDA_NVCC_FLAGS: $CUDA_NVCC_FLAGS"
-          cargo build --features $FEATURES,bench,rag,k8s,gce,spnl-api $BUILD_RELEASE_ARG \
-            --package spnl-cli --target $TARGET
+          cargo build --manifest-path crates/spnl-cli/Cargo.toml --target-dir target --features $FEATURES,bench,rag,k8s,gce,spnl-api $BUILD_RELEASE_ARG \
+            --target $TARGET
 
       - name: Prepare binary
         # Skip flash-attn variants on PRs (only run on release)

.github/workflows/vllm-gce.yml

@@ -65,7 +65,7 @@ jobs:
       - name: Install spnl CLI
         if: "!contains(github.event.pull_request.title, 'chore(deps)') && !contains(github.event.pull_request.title, 'fix(deps)') && env.GCP_CREDENTIALS && env.HF_TOKEN"
         run: |
-          cargo build --features gce
+          cargo build --manifest-path crates/spnl-cli/Cargo.toml --target-dir target --features gce
           sudo cp target/debug/spnl /usr/local/bin/
 
       - name: Launch GCE VM

.gitignore

@@ -1,5 +1,5 @@
 *~
-/target
+target/
 /target.cross
 history.txt