e2e-matrix.yml

name: E2E Coverage Matrix

on:
  push:
    branches: [main]
    paths:
      - "cppmega_v4/**"
      - "cppmega_mlx/**"
      - "vbgui/**"
      - "tests/fixtures/build_e2e_matrix.py"
      - "tests/fixtures/tokenizers/**"
      - "tests/fixtures/parquet/**"
      - ".github/workflows/e2e-matrix.yml"
  pull_request:
    branches: [main]
    paths:
      - "cppmega_v4/**"
      - "vbgui/**"
      - "tests/fixtures/build_e2e_matrix.py"
  schedule:
    # Nightly full run — includes the mini-train matrix on macOS.
    - cron: "0 4 * * *"
  workflow_dispatch:

concurrency:
  group: e2e-matrix-${{ github.ref }}
  cancel-in-progress: true

jobs:
  # On every push/PR: smoke matrix (912 cells). The backend imports MLX, so
  # this runs on the repository's persistent Apple Silicon runner.
  preset-matrix:
    name: Preset matrix (smoke, self-hosted macOS)
    runs-on: [self-hosted, macOS, ARM64, cppmega-mlx-macos]
    timeout-minutes: 25
    strategy:
      fail-fast: false
      matrix:
        shard: [1, 2, 3, 4]
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: "22"
          cache: "npm"
          cache-dependency-path: vbgui/package-lock.json
      - name: Bind repository MLX Python
        run: |
          python_bin=/Volumes/external/sources/cppmega.mlx/.venv/bin/python
          test -x "$python_bin"
          echo "VBGUI_E2E_PYTHON=$python_bin" >> "$GITHUB_ENV"
      - name: Python deps
        run: |
          "$VBGUI_E2E_PYTHON" -m pip install --upgrade pip
          "$VBGUI_E2E_PYTHON" -m pip install -e ".[gui,parquet,widget]"
          "$VBGUI_E2E_PYTHON" -m pip install jsonschema pyarrow tokenizers anywidget
      - name: Generate fixtures
        run: |
          "$VBGUI_E2E_PYTHON" tests/fixtures/build_e2e_matrix.py
      - name: Vbgui install + typecheck
        working-directory: vbgui
        env:
          NODE_ENV: development
        run: |
          npm ci
          npm run typecheck
      - name: Playwright browsers
        working-directory: vbgui
        run: npx playwright install chromium
      - name: Run preset matrix shard ${{ matrix.shard }}/4
        working-directory: vbgui
        env:
          NODE_ENV: development
          CI: "1"
        run: |
          npx playwright test --config=e2e/playwright.config.ts \
            e2e/scenarios/02_preset_matrix.spec.ts \
            --shard=${{ matrix.shard }}/4
      - name: Upload artefacts
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: preset-matrix-shard-${{ matrix.shard }}
          path: |
            vbgui/e2e/test-results/
            vbgui/e2e/screenshots/
            vbgui/e2e/logs/
          retention-days: 3

  # Specialised + canvas smoke scenarios — run on every push (lightweight).
  specialised:
    name: Specialised + canvas smoke
    runs-on: [self-hosted, macOS, ARM64, cppmega-mlx-macos]
    timeout-minutes: 15
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: "22", cache: "npm",
                cache-dependency-path: vbgui/package-lock.json }
      - name: Bind repository MLX Python
        run: |
          python_bin=/Volumes/external/sources/cppmega.mlx/.venv/bin/python
          test -x "$python_bin"
          echo "VBGUI_E2E_PYTHON=$python_bin" >> "$GITHUB_ENV"
      - run: |
          "$VBGUI_E2E_PYTHON" -m pip install --upgrade pip
          "$VBGUI_E2E_PYTHON" -m pip install -e ".[gui,parquet,widget]"
          "$VBGUI_E2E_PYTHON" -m pip install jsonschema pyarrow tokenizers anywidget
      - run: |
          "$VBGUI_E2E_PYTHON" tests/fixtures/build_e2e_matrix.py
      - working-directory: vbgui
        env: { NODE_ENV: development }
        run: |
          npm ci
          npm run typecheck
      - working-directory: vbgui
        run: npx playwright install chromium
      - working-directory: vbgui
        env: { NODE_ENV: development, CI: "1" }
        run: |
          npx playwright test --config=e2e/playwright.config.ts \
            e2e/scenarios/01_canvas_smoke.spec.ts \
            e2e/scenarios/04_tokenizer_playground.spec.ts \
            e2e/scenarios/05_data_inspector.spec.ts \
            e2e/scenarios/06_sharding_proposals.spec.ts \
            e2e/scenarios/07_gotchas.spec.ts
      - if: always()
        uses: actions/upload-artifact@v4
        with:
          name: specialised-results
          path: |
            vbgui/e2e/test-results/
            vbgui/e2e/screenshots/
            vbgui/e2e/logs/
          retention-days: 3

  # Mini-train matrix (192 cells, real mlx) — macOS only, nightly.
  mini-train-matrix:
    name: Mini-train matrix (macOS, nightly)
    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
    runs-on: [self-hosted, macOS, ARM64, cppmega-mlx-macos]
    timeout-minutes: 60
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: "22", cache: "npm",
                cache-dependency-path: vbgui/package-lock.json }
      - name: Bind repository MLX Python
        run: |
          python_bin=/Volumes/external/sources/cppmega.mlx/.venv/bin/python
          test -x "$python_bin"
          echo "VBGUI_E2E_PYTHON=$python_bin" >> "$GITHUB_ENV"
      - run: |
          "$VBGUI_E2E_PYTHON" -m pip install --upgrade pip
          "$VBGUI_E2E_PYTHON" -m pip install -e ".[gui,parquet,widget]"
          "$VBGUI_E2E_PYTHON" -m pip install jsonschema pyarrow tokenizers anywidget mlx mlx-lm
      - run: |
          "$VBGUI_E2E_PYTHON" tests/fixtures/build_e2e_matrix.py
      - working-directory: vbgui
        env: { NODE_ENV: development }
        run: |
          npm ci
          npm run typecheck
      - working-directory: vbgui
        run: npx playwright install chromium
      - working-directory: vbgui
        env: { NODE_ENV: development, CI: "1" }
        run: |
          npx playwright test --config=e2e/playwright.config.ts \
            e2e/scenarios/03_train_matrix.spec.ts
      - if: always()
        uses: actions/upload-artifact@v4
        with:
          name: mini-train-results
          path: |
            vbgui/e2e/test-results/
            vbgui/e2e/screenshots/
            vbgui/e2e/logs/
          retention-days: 7

  # Matrix report — aggregates shard artefacts into one Markdown
  # summary, posted as a build summary + uploaded as standalone artefact.
  matrix-report:
    name: Matrix report
    needs: [preset-matrix, specialised]
    runs-on: [self-hosted, Linux, X64, cppmega-mlx]
    if: always()
    steps:
      - uses: actions/checkout@v4
      - uses: actions/download-artifact@v4
        with: { path: artifacts }
      - uses: actions/setup-python@v5
        with: { python-version: "3.13" }
      - name: Build matrix report
        run: |
          python tools/build_e2e_matrix_report.py \
            --artifacts artifacts \
            --output e2e_matrix_report.md
          cat e2e_matrix_report.md >> "$GITHUB_STEP_SUMMARY"
      - uses: actions/upload-artifact@v4
        with:
          name: e2e_matrix_report
          path: e2e_matrix_report.md
          retention-days: 14