name: Test Count Monotonic

# Verify the PR doesn't reduce pytest collection count vs. base.
#
# Why: 5 days from v3.9.0 to v3.9.4.1 added 1463 -> 1561 tests
# (+98, 0 regression). Manual baseline tracking. This gate catches
# any PR that silently drops tests (deletion, accidental skip,
# fixture rename leaving orphans).
#
# Tolerance: net count must be >= base. Renames are fine; deletions
# require explicit justification.
#
# Override: include [skip-test-count] in PR body with justification
# (e.g., "removing dead suite X — N tests retired intentionally").

on:
  pull_request:
    branches: [main]

concurrency:
  group: test-count-${{ github.event.pull_request.number }}-${{ github.event_name }}
  cancel-in-progress: true

permissions:
  contents: read
  pull-requests: read

jobs:
  count-check:
    runs-on: ubuntu-latest
    steps:
      - name: Check out PR
        uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Set up Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'
          cache: pip
          cache-dependency-path: requirements-dev.txt

      - name: Install deps (PR head)
        run: |
          python -m pip install --upgrade pip
          pip install -r requirements-dev.txt pytest

      - name: Count tests on PR head
        id: head
        run: |
          set -euo pipefail
          # Capture pytest exit code separately from grep so a collection
          # error (broken import, missing fixture file, etc.) fails the gate
          # instead of being swallowed by the pipe (#155 / re-attempt of F4
          # from the v3.9.4.2 cycle, originally in PR #153 commit 8121dfa
          # before being reverted in 4abf9de while #154 was still open).
          set +e
          pytest --collect-only -q > pytest-collect.txt 2> pytest-collect.err
          PYTEST_EXIT=$?
          set -e
          # Exit 5 = pytest "no tests collected"; tolerable for degenerate
          # cases (empty repo, path filter that matches nothing). Any other
          # non-zero exit is a real collection failure and must fail the gate.
          if [ "$PYTEST_EXIT" -ne 0 ] && [ "$PYTEST_EXIT" -ne 5 ]; then
            echo "::error::pytest collection failed on PR head (exit $PYTEST_EXIT)"
            cat pytest-collect.err >&2 || true
            tail -20 pytest-collect.txt >&2 || true
            exit 1
          fi
          # grep -c exits 1 when there are zero matches — tolerate ONLY that;
          # any other non-zero exit from grep is a real error (unreadable
          # file, missing binary) and must propagate.
          set +e
          COUNT=$(grep -c '::' pytest-collect.txt)
          GREP_EXIT=$?
          set -e
          if [ "$GREP_EXIT" -ne 0 ] && [ "$GREP_EXIT" -ne 1 ]; then
            echo "::error::grep failed unexpectedly (exit $GREP_EXIT)"
            exit 1
          fi
          echo "Head test count: $COUNT"
          echo "count=$COUNT" >> "$GITHUB_OUTPUT"
          rm -f pytest-collect.txt pytest-collect.err

      - name: Add base via worktree
        env:
          BASE_SHA: ${{ github.event.pull_request.base.sha }}
        run: |
          set -euo pipefail
          git worktree add ../base "$BASE_SHA"

      - name: Install deps (base)
        working-directory: ../base
        run: |
          # Reuse pip cache; requirements may differ
          if [ -f requirements-dev.txt ]; then
            pip install -r requirements-dev.txt pytest --quiet
          else
            pip install pytest --quiet
          fi

      - name: Count tests on base
        id: base
        working-directory: ../base
        run: |
          set -euo pipefail
          # Symmetric strict collection — mirrors the PR-head step above.
          # The previous `2>/dev/null | grep -c '::' || true` pattern would
          # set BASE_COUNT to 0 on any collection error, making the
          # monotonic comparison (head >= base) vacuously pass and hiding
          # genuine drift. Fail the gate on real collection errors instead.
          set +e
          pytest --collect-only -q > pytest-collect.txt 2> pytest-collect.err
          PYTEST_EXIT=$?
          set -e
          if [ "$PYTEST_EXIT" -ne 0 ] && [ "$PYTEST_EXIT" -ne 5 ]; then
            echo "::error::pytest collection failed on base (exit $PYTEST_EXIT)"
            cat pytest-collect.err >&2 || true
            tail -20 pytest-collect.txt >&2 || true
            exit 1
          fi
          set +e
          COUNT=$(grep -c '::' pytest-collect.txt)
          GREP_EXIT=$?
          set -e
          if [ "$GREP_EXIT" -ne 0 ] && [ "$GREP_EXIT" -ne 1 ]; then
            echo "::error::grep failed unexpectedly (exit $GREP_EXIT)"
            exit 1
          fi
          echo "Base test count: $COUNT"
          echo "count=$COUNT" >> "$GITHUB_OUTPUT"
          rm -f pytest-collect.txt pytest-collect.err

      - name: Compare
        env:
          PR_BODY: ${{ github.event.pull_request.body }}
          HEAD_COUNT: ${{ steps.head.outputs.count }}
          BASE_COUNT: ${{ steps.base.outputs.count }}
        run: |
          set -euo pipefail
          DELTA=$((HEAD_COUNT - BASE_COUNT))
          echo "Base: $BASE_COUNT  Head: $HEAD_COUNT  Delta: $DELTA"

          if [ "$HEAD_COUNT" -ge "$BASE_COUNT" ]; then
            echo "OK — test count monotonic ($DELTA)."
            exit 0
          fi

          if echo "${PR_BODY:-}" | grep -qF '[skip-test-count]'; then
            echo "::warning::Test count dropped by $((BASE_COUNT - HEAD_COUNT)) but [skip-test-count] override present."
            exit 0
          fi

          echo "::error::Test count dropped: $BASE_COUNT -> $HEAD_COUNT (-$((BASE_COUNT - HEAD_COUNT)))."
          echo "::error::Add [skip-test-count] + justification to PR body if intentional."
          exit 1