ci-test.yml

# ============================================================================
# 🧪 CI · Test
# ----------------------------------------------------------------------------
# Purpose : Run unit + Playwright e2e tests across Linux & Windows
# Trigger : Push to `main`, PRs targeting `main`, manual dispatch
# Jobs    : unit  — `bun turbo test` + config_assistant Go tests on linux
#                    only (windows dropped — see
#                    unit-tests matrix comment; free windows-latest runners
#                    can't fit the suite in a reasonable CI budget)
#           e2e   — Playwright chromium on linux + windows (matrix)
# Gate    : The full unit, HttpAPI, generated-code and Linux/Windows E2E
#           suite runs on every PR targeting `main` before merge, and on
#           every push to `main`. Required status checks retain their names.
# Notes   : `cancel-in-progress: false` — every main push gets a full run
#           No trigger on feat/* or fix/* (frequent changes).
# Runners : All events use GitHub-hosted Ubuntu and Windows runners.
# ============================================================================

name: 🧪 CI · Test

on:
  push:
    branches:
      - main
  pull_request:
    # Re-run checks when a draft becomes ready for review.
    types: [opened, synchronize, reopened, ready_for_review]
    branches:
      - main
  workflow_dispatch:

concurrency:
  group: ${{ format('{0}-{1}', github.workflow, github.run_id) }}
  cancel-in-progress: false

permissions:
  contents: read
  actions: read
  checks: write

env:
  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
  GOTOOLCHAIN: local

jobs:
  unit-tests:
    name: Unit Tests (${{ matrix.settings.name }})
    strategy:
      fail-fast: false
      matrix:
        settings:
          - name: linux
            host: ubuntu-latest
          # windows dropped: this fork runs on free windows-latest runners
          # (not the paid Blacksmith 4vCPU hosts upstream uses), and the
          # opencode:test suite (3048 tests, many spawning real CLI
          # subprocesses) doesn't fit the standard runner's slower
          # process-spawn/IO within a reasonable CI budget. E2E Tests
          # (windows) is unaffected and still covers the platform.
    runs-on: ${{ matrix.settings.host }}
    defaults:
      run:
        shell: bash
    steps:
      - name: Checkout repository
        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}

      - name: Setup Node
        uses: ./.github/actions/setup-node

      - name: Setup GitHub CLI
        uses: ./.github/actions/setup-gh

      - name: Check previously verified content
        id: evidence
        uses: ./.github/actions/verified-content
        with:
          check: unit
          job-name: Unit Tests (${{ matrix.settings.name }})
          runner-label: ${{ matrix.settings.host }}

      - name: Setup Go
        if: steps.evidence.outputs.reused != 'true' && (runner.os == 'Linux')
        uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7.0.0
        with:
          go-version-file: config_assistant/go.mod
          cache-dependency-path: config_assistant/go.sum

      - name: Setup Bun
        if: steps.evidence.outputs.reused != 'true'
        uses: ./.github/actions/setup-bun
        with:
          # ci-typecheck.yml's Linux job is the designated Linux cache
          # saver (runs on every push, fast, low collision risk) — this job
          # only restores, to avoid racing on the same {OS}-bun-{hash} key.
          save-cache: false

      - name: Verify Go toolchain
        if: steps.evidence.outputs.reused != 'true' && runner.os == 'Linux'
        run: node script/toolchain.mjs check --go

      - name: Configure Git Identity
        if: steps.evidence.outputs.reused != 'true'
        run: |
          git config --global user.email "bot@opencode.ai"
          git config --global user.name "opencode"

      - name: Ensure ripgrep prerequisite
        if: steps.evidence.outputs.reused != 'true'
        # tool.glob tests hit ripgrep; without the system binary, binary.ts
        # downloads rg from GitHub releases every run (temp XDG per preload),
        # gambling on network stability — ECONNRESET fails the test.
        run: |
          if command -v rg >/dev/null 2>&1; then
            exit 0
          fi
          sudo apt-get update && sudo apt-get install -y ripgrep

      - name: Cache Turbo
        if: steps.evidence.outputs.reused != 'true'
        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
          path: node_modules/.cache/turbo
          key: turbo-${{ runner.os }}-${{ hashFiles('turbo.json', '**/package.json') }}-${{ github.sha }}
          restore-keys: |
            turbo-${{ runner.os }}-${{ hashFiles('turbo.json', '**/package.json') }}-
            turbo-${{ runner.os }}-

      - name: Run context-folding lifecycle preflight
        if: steps.evidence.outputs.reused != 'true' && (runner.os == 'Linux')
        working-directory: packages/opencode
        timeout-minutes: 3
        run: |
          bun test --timeout 30000 --only-failures test/session/prompt.test.ts \
            --test-name-pattern '^(runs builtin read/grep/glob settlement through stored history into the next (AI SDK|Native) request|automatically compacts (AI SDK|Native) folded history at the token threshold|recovers (AI SDK|Native) overflow through an unprojected summary and a continued turn|terminates (AI SDK|Native) overflow when the full compaction request is also rejected)$'

      - name: Run unit tests
        if: steps.evidence.outputs.reused != 'true'
        # opencode:test alone (244 files / 3048 tests, many spawning real CLI
        # subprocesses via cliIt) measured at 803s / 13m24s locally, and this
        # job has been seen at 16-17m in CI. turbo buffers a concurrent
        # task's output until it completes, so the step is silent the whole
        # time it runs — that silence can look like a hang but isn't one.
        # 20m was never sized against real data; give it margin above the
        # measured baseline instead of racing it.
        # Hosted runners execute workspace tasks one at a time. This avoids
        # CPU contention between builds and tests timing out 250ms Node workers.
        timeout-minutes: 35
        run: GITHUB_ACTIONS=false bun turbo test --concurrency=1
        env:
          OPENCODE_EXPERIMENTAL_DISABLE_FILEWATCHER: ${{ runner.os == 'Windows' && 'true' || 'false' }}

      - name: Run config assistant tests
        if: steps.evidence.outputs.reused != 'true' && (runner.os == 'Linux')
        working-directory: config_assistant
        run: go test ./...

      - name: Run VS Code connection tests
        if: steps.evidence.outputs.reused != 'true' && runner.os == 'Linux'
        working-directory: sdks/vscode
        timeout-minutes: 5
        run: |
          bun install --frozen-lockfile
          bun run test:unit

      - name: Check generated client
        if: steps.evidence.outputs.reused != 'true' && (runner.os == 'Linux')
        working-directory: packages/client
        run: bun run check:generated

      - name: Check generated SDK
        if: steps.evidence.outputs.reused != 'true' && (runner.os == 'Linux')
        working-directory: packages/sdk/js
        run: bun run check:generated

      - name: Run HttpAPI Exerciser Gates
        if: steps.evidence.outputs.reused != 'true' && (runner.os == 'Linux')
        working-directory: packages/opencode
        # The exerciser aggregates its report and prints it in one flush, so
        # this step is silent while it runs. A latent uninterruptible hang
        # (bare Effect.promise in scenario call/cleanup paths that the 30s
        # scenario timeout cannot interrupt) froze the effect mode twice on
        # 2026-07-27 — with no step timeout that occupies the runner for the
        # default 6h. The full gate baseline is ~6m in CI, so 15m is generous;
        # test:httpapi:ci adds --progress/--trace to effect mode so the log
        # names the exact scenario and phase if it ever hangs again.
        # 2026-08-05: a native-level freeze survived every in-process guard
        # (timers die with the event loop) and burned the full 15m silently.
        # --progress now also arms an out-of-process watchdog that SIGKILLs
        # the runner after 120s without progress, naming the last scenario/
        # phase — the step timeout stays as the final backstop.
        timeout-minutes: 15
        run: bun run test:httpapi:ci

      - name: Record successful verification
        if: success() && steps.evidence.outputs.reused != 'true'
        uses: ./.github/actions/record-verification
        with:
          key: ${{ steps.evidence.outputs.key }}
          path: ${{ steps.evidence.outputs.path }}
          job-name: Unit Tests (${{ matrix.settings.name }})

  e2e-tests:
    name: E2E Tests (${{ matrix.settings.name }})
    strategy:
      fail-fast: false
      matrix:
        settings:
          - name: linux
            host: ubuntu-latest
          - name: windows
            host: windows-latest
    runs-on: ${{ matrix.settings.host }}
    env:
      PLAYWRIGHT_BROWSERS_PATH: ${{ github.workspace }}/.playwright-browsers
    defaults:
      run:
        shell: bash
    steps:
      - name: Checkout repository
        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}

      - name: Setup Node
        # Browser installation below checks extraction on the shared Node pin.
        uses: ./.github/actions/setup-node

      - name: Setup GitHub CLI
        uses: ./.github/actions/setup-gh

      - name: Check previously verified content
        id: evidence
        uses: ./.github/actions/verified-content
        with:
          check: e2e
          job-name: E2E Tests (${{ matrix.settings.name }})
          runner-label: ${{ matrix.settings.host }}

      - name: Setup Bun
        if: steps.evidence.outputs.reused != 'true'
        uses: ./.github/actions/setup-bun
        with:
          # Windows cache archiving can stall before DAG and Playwright tests.
          # Linux unit tests remain the single Linux cache saver.
          save-cache: false

      - name: Read Playwright version
        if: steps.evidence.outputs.reused != 'true'
        id: playwright-version
        run: |
          version=$(node -e 'console.log(require("./package.json").workspaces.catalog["@playwright/test"])')
          echo "version=$version" >> "$GITHUB_OUTPUT"

      - name: Cache Playwright browsers
        if: steps.evidence.outputs.reused != 'true'
        id: playwright-cache
        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
          path: ${{ github.workspace }}/.playwright-browsers
          key: ${{ runner.os }}-${{ runner.arch }}-playwright-${{ steps.playwright-version.outputs.version }}-chromium

      - name: Install Playwright system dependencies
        if: steps.evidence.outputs.reused != 'true' && (runner.os == 'Linux')
        working-directory: packages/app
        run: bunx playwright install-deps chromium

      - name: Install Playwright browsers
        if: steps.evidence.outputs.reused != 'true' && (steps.playwright-cache.outputs.cache-hit != 'true')
        working-directory: packages/app
        run: bunx playwright install chromium

      - name: Verify DAG artifact storage on Windows
        if: steps.evidence.outputs.reused != 'true' && runner.os == 'Windows'
        working-directory: packages/opencode
        run: bun test --timeout 30000 --only-failures test/dag/dag-output-ref.test.ts test/dag/dag-output-ref-result.test.ts test/dag/dag-input-mapping-runtime.test.ts test/dag/dag-artifact-permissions.test.ts
        timeout-minutes: 5

      - name: Verify shared Markdown in Chromium
        if: steps.evidence.outputs.reused != 'true'
        working-directory: packages/web
        run: bun test test/sanitize-markdown.test.ts --timeout 30000 --only-failures
        env:
          OPENCODE_TEST_BROWSER: "1"
        timeout-minutes: 3

      - name: Run app e2e tests
        if: steps.evidence.outputs.reused != 'true'
        run: bun --cwd packages/app test:e2e:local
        env:
          CI: true
        timeout-minutes: 30

      - name: Upload Playwright artifacts
        if: steps.evidence.outputs.reused != 'true' && (always())
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: playwright-${{ matrix.settings.name }}-${{ github.run_attempt }}
          if-no-files-found: ignore
          retention-days: 7
          path: |
            packages/app/e2e/test-results
            packages/app/e2e/playwright-report

      - name: Record successful verification
        if: success() && steps.evidence.outputs.reused != 'true'
        uses: ./.github/actions/record-verification
        with:
          key: ${{ steps.evidence.outputs.key }}
          path: ${{ steps.evidence.outputs.path }}
          job-name: E2E Tests (${{ matrix.settings.name }})