tests.yml

name: Tests

# PR-time gate: runs Rust + plugin unit tests, Linux Docker e2e, and Windows e2e
# before merge. Tag pushes go through `release.yml` which independently runs the
# build matrix and publishes — both workflows call the SAME reusable e2e suite
# (`_e2e-suite.yml`) so PR-time and release-time e2e never drift.
#
# IMPORTANT: any change to bridge transport, bash spawning, ONNX install,
# locking, or platform-conditional code paths SHOULD touch the matching
# integration test or e2e scenario in `_e2e-suite.yml`. The Linux harness has
# caught real regressions before; the Windows e2e is here to extend that
# coverage to issue-#26-class Windows-specific bugs (bash timeouts, lock
# recovery, path separators).

on:
  pull_request:
    paths:
      - "crates/**"
      - "packages/**"
      - "tests/**"
      - "Cargo.toml"
      - "Cargo.lock"
      - "package.json"
      - "bun.lock"
      - ".github/opencode-version.txt"
      - ".github/workflows/tests.yml"
      - ".github/workflows/_e2e-suite.yml"
  push:
    branches:
      - main
    paths:
      - "crates/**"
      - "packages/**"
      - "tests/**"
      - "Cargo.toml"
      - "Cargo.lock"
      - "package.json"
      - "bun.lock"
      - ".github/opencode-version.txt"
      - ".github/workflows/tests.yml"
      - ".github/workflows/_e2e-suite.yml"
  workflow_dispatch:
    inputs:
      ref:
        description: "Git ref to test (branch name, tag, or commit SHA)"
        required: false
        type: string

# Cancel in-flight runs for the same PR/branch when a new commit arrives.
# Saves CI minutes and surfaces the latest result faster.
concurrency:
  group: tests-${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

env:
  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true

jobs:
  # ---------------------------------------------------------------------------
  # Rust + plugin unit tests on Linux (fast inner loop)
  # ---------------------------------------------------------------------------
  unit:
    name: Unit tests (Linux)
    runs-on: ubuntu-22.04
    timeout-minutes: 25
    steps:
      - uses: actions/checkout@v4

      - uses: dtolnay/rust-toolchain@stable

      - uses: Swatinem/rust-cache@v2

      - uses: oven-sh/setup-bun@v2
        with:
          bun-version: latest

      - name: Install workspace deps
        run: bun install --frozen-lockfile

      - name: Build aft-bridge dist (workspace consumers depend on it)
        run: bun run --cwd packages/aft-bridge build

      - name: Cargo build (debug — needed by plugin e2e tests)
        run: cargo build -p agent-file-tools

      - name: Cargo test
        run: cargo test --workspace

      - name: Bun typecheck
        run: bun run typecheck

      - name: Bun lint
        run: bun run lint

      - name: Bun test (all packages)
        run: bun run test
        env:
          # Tests don't need real cache locations — keep them out of the runner's
          # ~/.cache to avoid cross-test pollution.
          AFT_CACHE_DIR: ${{ runner.temp }}/aft-cache

  # ---------------------------------------------------------------------------
  # Rust + plugin tests on macOS
  # Catches macOS-specific code paths: FSEvents watcher behavior (different
  # coalescing latency from inotify), /var vs /private/var symlink
  # canonicalization, broken-symlink-chain fallback, bash_background SIGTERM
  # behavior, and Apple Silicon-specific Rust compilation. The build-darwin-*
  # jobs in release.yml only run `cargo build` — this is the only place we
  # actually execute tests on macOS in CI.
  # ---------------------------------------------------------------------------
  rust-macos:
    name: Unit tests (macOS)
    runs-on: macos-latest
    timeout-minutes: 25
    steps:
      - uses: actions/checkout@v4

      - uses: dtolnay/rust-toolchain@stable

      - uses: Swatinem/rust-cache@v2

      - uses: oven-sh/setup-bun@v2
        with:
          bun-version: latest

      - name: Install workspace deps
        run: bun install --frozen-lockfile

      - name: Build aft-bridge dist (workspace consumers depend on it)
        run: bun run --cwd packages/aft-bridge build

      - name: Cargo build (debug — needed by plugin e2e tests)
        run: cargo build -p agent-file-tools

      - name: Cargo test
        run: cargo test --workspace

      - name: Bun typecheck
        run: bun run typecheck

      - name: Bun test (all packages)
        run: bun run test
        env:
          AFT_CACHE_DIR: ${{ runner.temp }}/aft-cache

  # ---------------------------------------------------------------------------
  # Rust integration tests on Windows
  # Catches platform-conditional code paths (#[cfg(target_os = "windows")])
  # and Windows process/path/signal differences. Same suite as Linux but on a
  # real Windows runner — does NOT spin up OpenCode, that's the e2e job below.
  # ---------------------------------------------------------------------------
  rust-windows:
    name: Cargo test (Windows)
    runs-on: windows-2022
    timeout-minutes: 30
    # Non-blocking on Windows: many integration tests build NDJSON
    # requests by hand-formatting paths via `r#"...{path.display()}..."#`,
    # which produces invalid JSON on Windows because backslashes in
    # `C:\Users\...` get parsed as escape sequences. Migrating ~150
    # call sites to `serde_json::json!` is a separate cleanup; the
    # production code path is platform-clean (538 lib tests pass on
    # Windows runners, plus the Windows native E2E job below covers
    # the integration layer). Tracked for follow-up.
    continue-on-error: true
    steps:
      - uses: actions/checkout@v4

      - uses: dtolnay/rust-toolchain@stable
        with:
          targets: x86_64-pc-windows-msvc

      - uses: Swatinem/rust-cache@v2

      - name: Cargo test (lib only — integration tests need JSON path-escape cleanup)
        run: cargo test --workspace --lib
        shell: pwsh

  # ---------------------------------------------------------------------------
  # Bash permission flow on Windows
  # Scoped to bash.test.ts only: covers the runAsk + Effect runtime path on
  # Windows for OS-parity peace of mind. The runAsk fix itself is platform-
  # agnostic JavaScript semantics — Linux/macOS in the `unit` and `rust-macos`
  # jobs already cover that. Running the full plugin bun suite on Windows
  # blocks on broader integration-harness JSON-path escape issues (see
  # rust-windows comment), so we scope this to one test file.
  #
  # Why a real bun runner on Windows: this exercises the BinaryBridge
  # spawning aft.exe over NDJSON via stdio on the Windows process model. It
  # catches a class of bugs that Linux+macOS bun + Windows native E2E miss
  # individually — Effect-runtime semantics under bun's Windows process
  # spawn implementation. Non-blocking for now; promote to required once we
  # have N green runs.
  # ---------------------------------------------------------------------------
  bun-windows-bash:
    name: Bash permission e2e (Windows)
    runs-on: windows-2022
    timeout-minutes: 20
    continue-on-error: true
    steps:
      - uses: actions/checkout@v4

      - uses: dtolnay/rust-toolchain@stable

      - uses: Swatinem/rust-cache@v2

      - uses: oven-sh/setup-bun@v2
        with:
          bun-version: latest

      - name: Install workspace deps
        run: bun install --frozen-lockfile
        shell: pwsh

      - name: Build aft-bridge dist (workspace consumers depend on it)
        run: bun run --cwd packages/aft-bridge build
        shell: pwsh

      - name: Cargo build (debug — bash.test.ts spawns the real aft binary)
        run: cargo build -p agent-file-tools
        shell: pwsh

      - name: Bun test (bash permission flow — full bridge + plugin + Rust)
        run: bun test src/__tests__/e2e/bash.test.ts
        working-directory: packages/opencode-plugin
        shell: pwsh
        env:
          AFT_CACHE_DIR: ${{ runner.temp }}/aft-cache

  # ---------------------------------------------------------------------------
  # Linux Docker e2e + Windows native e2e — single source of truth shared
  # with release.yml. See `_e2e-suite.yml` for the actual job definitions.
  # ---------------------------------------------------------------------------
  e2e:
    name: E2E
    needs: unit
    uses: ./.github/workflows/_e2e-suite.yml