learningcircuit--local-deep-research/.github/workflows/responsive-ui-tests-enhanced.yml

name: Responsive UI Tests

on:
  workflow_call:
  workflow_dispatch:

# No concurrency group — intentionally omitted.
# This workflow runs only via workflow_call (from release.yml's
# responsive-test-gate) and workflow_dispatch. The pull_request trigger
# was deliberately removed in #2248 to reduce PR CI load on what is a
# heavy ~20-minute matrix build (mobile + desktop). A shared concurrency
# key here previously caused workflow_call invocations to cancel each
# other mid-flight; see #3554 (reverted in #3599) for that history.

permissions:
  contents: read

jobs:
  ui-tests:
    runs-on: ubuntu-latest
    timeout-minutes: 20

    strategy:
      fail-fast: false
      matrix:
        viewport: [mobile, desktop]  # Test mobile and desktop on each PR

    services:
      postgres:
        image: postgres:14@sha256:ca25035f7e6f74552655a1c5e4a9eb21f85e9d316f1f70371f790ef70095dd58 # v14
        env:
          POSTGRES_USER: postgres
          POSTGRES_PASSWORD: postgres
          POSTGRES_DB: ldr_test
        ports:
          - 5432:5432
        options: >-
          --health-cmd pg_isready
          --health-interval 10s
          --health-timeout 5s
          --health-retries 5

    steps:
      - name: Harden the runner (Audit all outbound calls)
        uses: step-security/harden-runner@9af89fc71515a100421586dfdb3dc9c984fbf411 # v2.19.4
        with:
          egress-policy: audit

      - name: Checkout code
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0

        with:
          persist-credentials: false

      - name: Set up PDM
        uses: pdm-project/setup-pdm@973541a5febeafcfdadf8a51211435be6ecfd90f # v4.5
        with:
          python-version: '3.12'

      - name: Set up Node.js
        uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
        with:
          node-version: '24'
          cache: 'npm'
          cache-dependency-path: tests/ui_tests/package-lock.json

      - name: Free up disk space
        run: |
          # Remove unnecessary large packages to prevent disk space issues during cache save
          sudo rm -rf /usr/share/dotnet
          sudo rm -rf /usr/local/lib/android
          sudo rm -rf /opt/ghc
          sudo rm -rf /opt/hostedtoolcache/CodeQL
          df -h

      - name: Install system dependencies
        run: |
          sudo apt-get update
          sudo apt-get install -y \
            wget \
            gnupg \
            ca-certificates \
            fonts-liberation \
            libasound2t64 \
            libatk-bridge2.0-0 \
            libatk1.0-0 \
            libcups2 \
            libdbus-1-3 \
            libdrm2 \
            libgbm1 \
            libgtk-3-0 \
            libnspr4 \
            libnss3 \
            libx11-xcb1 \
            libxcomposite1 \
            libxdamage1 \
            libxrandr2 \
            xdg-utils \
            imagemagick

      - name: Install Python dependencies
        run: pdm install

      - name: Install root frontend dependencies
        run: npm ci

      - name: Build Vite frontend bundle
        # Generates src/local_deep_research/web/static/dist/, which the
        # Flask app loads via vite_helper. Without this the responsive UI
        # tests run against an unstyled page (no styles.css), which means
        # any CSS source changes between PRs are invisible to the test
        # baseline. See follow-up to PR #3985 for context.
        run: npm run build

      - name: Install Node test dependencies
        working-directory: tests/ui_tests
        run: |
          npm ci
          # Keep this version in sync with tests/ui_tests/package-lock.json's
          # puppeteer entry. A mismatch makes npx fetch a second puppeteer
          # that targets a different Chrome build, corrupting the browser
          # cache ("folder exists but executable is missing"). Matching the
          # locked version lets npx reuse the puppeteer just installed above.
          npx puppeteer@25.1.0 browsers install chrome

      - name: Set up test directories
        run: |
          mkdir -p ${{ github.workspace }}/data/encrypted_databases
          mkdir -p tests/ui_tests/screenshots
          echo "Created data and screenshots directories for tests"

      - name: Start test server
        env:
          CI: true
          FLASK_ENV: testing
          TEST_ENV: true
          SECRET_KEY: test-secret-key-for-ci  # Security: CI test credential, not production secret
          LDR_DISABLE_RATE_LIMITING: true
          LDR_DATA_DIR: ${{ github.workspace }}/data
          DATABASE_URL: postgresql://postgres:postgres@localhost:5432/ldr_test  # Security: CI test database credentials, not production secrets
        run: |
          cd src
          # Start server and get its PID
          pdm run python -m local_deep_research.web.app > server.log 2>&1 &
          SERVER_PID=$!
          echo "Server PID: $SERVER_PID"
          # Persist the PID for the "Stop application server" step. That step
          # runs at the workspace root (no working-directory) while we are in
          # src/ here, so write an absolute path — otherwise the pidfile never
          # lines up and cleanup is a silent no-op.
          echo "$SERVER_PID" > "$GITHUB_WORKSPACE/server.pid"

          # Wait for server to start
          SERVER_READY=false
          for i in {1..60}; do
            # Probe the health endpoint, not bare `/`: `/` 302-redirects to
            # /auth/login and `curl -f` treats a 302 as success the instant
            # the socket binds — before DB/app init finishes — so it can pass
            # while the app is still not ready. /api/v1/health is a true
            # readiness signal and matches the docker-tests.yml gate.
            if curl -fsS --connect-timeout 2 --max-time 5 http://127.0.0.1:5000/api/v1/health 2>/dev/null; then
              echo "Server is ready after $i seconds"
              SERVER_READY=true
              break
            fi
            # Check if process is still running
            if ! kill -0 "$SERVER_PID" 2>/dev/null; then
              echo "Server process died!"
              echo "Server log:"
              cat server.log
              exit 1
            fi
            echo "Waiting for server... ($i/60)"
            sleep 1
          done

          # Fail fast if the server is alive but never became ready in time.
          # Without this guard the loop fell through silently on a slow-but-
          # not-crashed start, so "Register CI test user" and the UI tests ran
          # against a server that was not yet listening and failed with a
          # confusing `net::ERR_CONNECTION_REFUSED` in the *test* step instead
          # of a clear startup failure here. Window was also bumped 30s -> 60s
          # to match the proven puppeteer-e2e-tests.yml / docker-tests.yml
          # startup gates (Postgres-backed boot can exceed 30s on a loaded
          # runner).
          if [ "$SERVER_READY" != "true" ]; then
            echo "❌ Server did not become ready within 60 seconds"
            echo "Server log:"
            cat server.log
            exit 1
          fi

      - name: Register CI test user
        working-directory: tests/ui_tests
        run: node register_ci_user.js http://127.0.0.1:5000

      - name: Run responsive UI tests - ${{ matrix.viewport }}
        id: run-tests
        working-directory: tests/ui_tests
        env:
          VIEWPORT: ${{ matrix.viewport }}
        run: |
          set +e  # Don't exit on test failure

          # Run tests and capture output
          HEADLESS=true node test_responsive_ui_comprehensive.js "$VIEWPORT" 2>&1 | tee test-output.log
          TEST_EXIT_CODE="${PIPESTATUS[0]}"

          # Extract summary for PR comment
          echo "### 📱 $VIEWPORT Test Results" > test-summary.md
          echo "" >> test-summary.md

          # Extract pass/fail counts
          PASSED=$(grep -oP '\d+(?= passed)' test-output.log | tail -1 || echo "0")
          FAILED=$(grep -oP '\d+(?= failed)' test-output.log | tail -1 || echo "0")
          WARNINGS=$(grep -oP '\d+(?= warnings)' test-output.log | tail -1 || echo "0")

          if [ "$FAILED" -eq "0" ]; then
            echo "✅ **All tests passed!**" >> test-summary.md
          else
            echo "❌ **$FAILED critical issues found**" >> test-summary.md
          fi

          {
            echo ""
            echo "- ✅ Passed: $PASSED"
            echo "- ❌ Failed: $FAILED"
            echo "- ⚠️ Warnings: $WARNINGS"
          } >> test-summary.md

          {
            echo "test_exit_code=$TEST_EXIT_CODE"
            echo "test_failed=$FAILED"
          } >> "$GITHUB_OUTPUT"

          # Only fail if critical failures
          exit "$TEST_EXIT_CODE"

      - name: Upload viewport test results
        if: always()
        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: ui-test-results-${{ matrix.viewport }}
          path: |
            tests/ui_tests/test-output.log
            tests/ui_tests/test-summary.md
            tests/ui_tests/screenshots/
            tests/ui_tests/responsive/
            src/server.log
          retention-days: 7
          if-no-files-found: ignore

      # NOTE: Earlier screenshot generation/upload steps were removed to fix actionlint warnings
      # caused by "if: false" disablement. Screenshots are now captured into the artifact above
      # when they exist. Visual regression workflows can re-enable dedicated screenshot steps later.

      - name: Upload to GitHub Pages (if available)
        if: always() && github.event_name == 'pull_request'
        continue-on-error: true
        env:
          PR_NUMBER: ${{ github.event.pull_request.number }}
          VIEWPORT: ${{ matrix.viewport }}
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          GH_REPOSITORY: ${{ github.repository }}
          GH_REPO_OWNER: ${{ github.repository_owner }}
          GH_REPO_NAME: ${{ github.event.repository.name }}
        run: |
          # This requires GitHub Pages to be enabled for the repo
          # Create a branch for the screenshots
          BRANCH_NAME="pr-screenshots-$PR_NUMBER-$VIEWPORT"

          # Check if screenshots directory exists
          if [ ! -d "tests/ui_tests/responsive" ]; then
            echo "No screenshots directory found, skipping upload"
            exit 0
          fi

          cd tests/ui_tests/responsive
          git init
          git config user.name "GitHub Actions"
          git config user.email "actions@github.com"

          # Add all screenshots and gallery
          git add responsive-ui-tests screenshot-gallery.html
          git commit -m "Screenshots for PR #$PR_NUMBER"

          # Push to a dedicated branch (requires write permissions)
          git push --force "https://x-access-token:$GH_TOKEN@github.com/$GH_REPOSITORY.git" "HEAD:$BRANCH_NAME" 2>/dev/null || true

          # The URL would be: https://[owner].github.io/[repo]/pr-screenshots-[number]-[viewport]/screenshot-gallery.html
          echo "Screenshots available at: https://$GH_REPO_OWNER.github.io/$GH_REPO_NAME/$BRANCH_NAME/screenshot-gallery.html" || true

      - name: Stop application server
        if: always()
        run: |
          if [ -f server.pid ]; then
            kill "$(cat server.pid)" || true
            rm server.pid
          fi
          # `pdm run` stays alive as a wrapper, so the pidfile holds the
          # wrapper PID and killing it can orphan the actual python server.
          # Add the same pkill fallback the webkit workflow uses so the
          # python process (and port 5000) is actually released.
          pkill -f "python -m local_deep_research.web.app" || true

  post-results:
    needs: ui-tests
    runs-on: ubuntu-latest
    # Workflow has no pull_request trigger; gating on event_name == 'pull_request' kept this job
    # from ever running. Always run so the combined report is built for workflow_dispatch / release.
    if: always()

    steps:
      - name: Harden the runner (Audit all outbound calls)
        uses: step-security/harden-runner@9af89fc71515a100421586dfdb3dc9c984fbf411 # v2.19.4
        with:
          egress-policy: audit

      - name: Checkout code
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0

        with:
          persist-credentials: false

      - name: Download all artifacts
        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
        with:
          path: test-artifacts/

      - name: Generate combined report
        run: |
          # Create a combined markdown report with screenshot links
          {
            echo "# 📊 Responsive UI Test Report"
            echo ""
            echo "## Test Summary"
            echo ""
          } > combined-report.md

          # Process each viewport's results
          for dir in test-artifacts/ui-test-results-*/; do
            if [ -d "$dir" ]; then
              viewport=$(basename "$dir" | sed 's/ui-test-results-//')
              echo "### $viewport" >> combined-report.md

              if [ -f "$dir/test-summary.md" ]; then
                cat "$dir/test-summary.md" >> combined-report.md
              fi

              # Count screenshots
              screenshot_count=$(find "$dir" -name "*.png" 2>/dev/null | wc -l)
              if [ "$screenshot_count" -gt "0" ]; then
                {
                  echo ""
                  echo "📸 **$screenshot_count screenshots captured**"
                } >> combined-report.md
              fi

              echo "" >> combined-report.md
            fi
          done

          {
            echo "## 📥 Download Options"
            echo ""
            echo "- [Download all test artifacts](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})"
            echo "- View artifacts in the 'Artifacts' section below the workflow summary"
          } >> combined-report.md

      - name: Upload combined report
        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: combined-test-report
          path: combined-report.md
          retention-days: 30