name: tests-macos

# Separate from tests.yml on purpose. This runs on a laptop, and a laptop is
# sometimes asleep. Keeping it in its own workflow means a queued macOS job
# leaves *this* run pending instead of holding the main test result hostage.

on:
  push:
    branches: [main]
  workflow_dispatch:

concurrency:
  group: tests-macos-${{ github.ref }}
  cancel-in-progress: true

jobs:
  macos:
    # Homebrew act_runner executes jobs directly on the host — there is no
    # container. Change this label if the runner was registered with a
    # different one; it has to match what the runner declared.
    runs-on: macos-latest
    steps:
      - uses: actions/checkout@v4

      # No setup-python / setup-node here, deliberately. Those actions install
      # a toolchain into a container that does not exist in host mode. Using
      # whatever Homebrew provides is also the better test: it is the
      # environment an actual macOS user would have, rather than a pinned one
      # nobody runs.
      - name: Show toolchain
        run: |
          python3 --version
          node --version
          npm --version
          uname -m

      # A venv rather than --user: Homebrew's Python is externally managed
      # (PEP 668), so a plain pip install into it is refused outright.
      - name: Install Python dependencies
        run: |
          python3 -m venv .ci-venv
          .ci-venv/bin/python -m pip install --quiet --upgrade pip
          .ci-venv/bin/python -m pip install -r requirements-ci.txt

      # requirements-ci.txt omits llama-cpp-python, so this covers suite
      # loading, grading, reporting and the HTTP surface — not inference.
      # Exercising .apple_silicon needs model weights CI does not have, and
      # remains a manual check.
      - name: Run Python tests
        run: .ci-venv/bin/python tests_py/run_all.py

      - name: Install web dependencies
        working-directory: web
        run: npm ci

      - name: Type-check and build
        working-directory: web
        run: npm run build

      - name: Run UI tests
        working-directory: web
        run: npm test

      # Proves the shipped artifact actually works on macOS, which is a
      # different claim from "it built". Catches launcher and layout problems
      # that never show up in a source checkout.
      - name: Smoke-test the packaged archive
        run: |
          .ci-venv/bin/python packaging/build.py --platform macos --skip-web
          cd dist
          tar xzf LM-Gambit-*-macos.tar.gz
          cd LM-Gambit-*-macos
          test -x LM-Gambit.sh || { echo "launcher is not executable"; exit 1; }
          head -1 LM-Gambit.sh | grep -q '^#!/usr/bin/env bash$' || { echo "bad shebang"; exit 1; }
          ../../.ci-venv/bin/python auto-test.py --suites