.github/workflows/perf_regression.yml

# Sprint A5 (M-14) — anti-régression de performance OCR.
#
# Hebdomadaire (cron lundi 06:00 UTC) + manuel via workflow_dispatch.
# **Pas** déclenché à chaque PR (coût Tesseract + stabilité statistique
# CER nécessitent un corpus plus large que ce qu'on peut tolérer en PR
# bloquante). Le but : détecter une dérive franche introduite par un
# refactor de la normalisation, du runner, ou un upgrade de pytesseract.
#
# Sortie : un commentaire automatique sur l'issue #perf-baseline avec
# le CER mesuré pour chaque doc + agrégat. Échec dur si CER moyen
# > 15 % sur Tesseract (seuil large — détecte les régressions, pas
# les variations normales).

name: Perf regression (weekly)

on:
  schedule:
    - cron: '0 6 * * 1'  # Lundi 06:00 UTC
  workflow_dispatch:  # Déclenchement manuel
  pull_request:
    paths:
      # Une PR qui touche le runner de benchmark, la normalisation ou
      # les adapters OCR déclenche aussi le check (cas où on veut
      # prouver qu'un refactor ne dégrade rien).
      - 'picarones/app/services/run_orchestrator.py'
      - 'picarones/app/services/run_orchestrator_execution.py'
      - 'picarones/app/services/run_orchestrator_helpers/**'
      - 'picarones/app/services/benchmark_service.py'
      - 'picarones/pipeline/executor.py'
      - 'picarones/pipeline/runner.py'
      - 'picarones/formats/text/normalization.py'
      - 'picarones/adapters/ocr/**'
      - '.github/workflows/perf_regression.yml'

permissions:
  contents: read
  issues: write

jobs:
  perf:
    name: CER regression check
    runs-on: ubuntu-latest

    steps:
      - name: Checkout
        uses: actions/checkout@v4

      - name: Set up Python
        uses: actions/setup-python@v5
        with:
          python-version: "3.11"
          cache: pip

      - name: Install Tesseract (Ubuntu)
        run: |
          sudo apt-get update -qq
          sudo apt-get install -y tesseract-ocr tesseract-ocr-fra tesseract-ocr-eng

      - name: Install Picarones
        run: |
          python -m pip install --upgrade pip setuptools wheel
          pip install -e ".[dev,web]"

      - name: Regenerate reference corpus (idempotent)
        run: python tests/fixtures/reference_corpus/_generate.py

      - name: Run benchmark on reference corpus
        id: bench
        run: |
          mkdir -p /tmp/perf_artifacts
          picarones run \
            --corpus tests/fixtures/reference_corpus/ \
            --engines tesseract \
            --output /tmp/perf_artifacts/results.json \
            --fail-if-cer-above 0.15 \
            --no-progress

      - name: Generate report (lazy_images mode)
        if: always()
        run: |
          picarones report \
            --results /tmp/perf_artifacts/results.json \
            --output /tmp/perf_artifacts/perf_report.html \
            --lazy-images || true

      - name: Upload artifacts
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: perf-${{ github.run_id }}
          path: /tmp/perf_artifacts/
          retention-days: 30

      - name: Comment on tracking issue (success)
        if: success() && github.event_name == 'schedule'
        uses: actions/github-script@v7
        with:
          script: |
            const fs = require('fs');
            const data = JSON.parse(
              fs.readFileSync('/tmp/perf_artifacts/results.json', 'utf-8')
            );
            const eng = data.engine_reports?.[0] || {};
            const meanCer = eng.aggregated_metrics?.cer?.mean ?? 'n/a';
            const body = `Hebdomadaire — Tesseract CER moyen: **${meanCer}** ` +
                          `(commit \`${context.sha.slice(0,7)}\`, ` +
                          `${data.engine_reports?.[0]?.document_results?.length ?? 0} docs).`;
            // Cherche l'issue de tracking, en crée une si absente.
            const issues = await github.rest.issues.listForRepo({
              owner: context.repo.owner,
              repo: context.repo.repo,
              labels: 'perf-baseline',
              state: 'open',
            });
            let issueNumber;
            if (issues.data.length > 0) {
              issueNumber = issues.data[0].number;
            } else {
              const created = await github.rest.issues.create({
                owner: context.repo.owner,
                repo: context.repo.repo,
                title: '📈 Perf baseline (auto-tracking)',
                body: 'Issue de suivi du job hebdomadaire ' +
                      '`perf_regression.yml`. Chaque exécution y commente ' +
                      'le CER moyen pour Tesseract.',
                labels: ['perf-baseline'],
              });
              issueNumber = created.data.number;
            }
            await github.rest.issues.createComment({
              owner: context.repo.owner,
              repo: context.repo.repo,
              issue_number: issueNumber,
              body: body,
            });

      - name: Comment on tracking issue (failure)
        if: failure() && github.event_name == 'schedule'
        uses: actions/github-script@v7
        with:
          script: |
            const issues = await github.rest.issues.listForRepo({
              owner: context.repo.owner,
              repo: context.repo.repo,
              labels: 'perf-baseline',
              state: 'open',
            });
            if (issues.data.length > 0) {
              await github.rest.issues.createComment({
                owner: context.repo.owner,
                repo: context.repo.repo,
                issue_number: issues.data[0].number,
                body: '❌ Échec hebdomadaire — CER > 15 % ou crash. ' +
                      `Voir le run [${context.runId}]` +
                      `(${context.serverUrl}/${context.repo.owner}/` +
                      `${context.repo.repo}/actions/runs/${context.runId}).`,
              });
            }