python-package.yml

name: Validate Install

on:
  push:
    branches: [ main ]
  pull_request:
    branches: [ main ]

jobs:
  test:
    runs-on: ${{ matrix.os }}
    timeout-minutes: 45

    strategy:
      matrix:
        os: [ubuntu-latest, windows-latest, macos-latest]
        python-version: ["3.10.11", 3.11, 3.12, 3.13, 3.14]
      fail-fast: false

    steps:
    - name: Check out the repository
      uses: actions/checkout@v4

    - name: Set up Python ${{ matrix.python-version }}
      uses: actions/setup-python@v5
      with:
        python-version: ${{ matrix.python-version }}

    - name: Cache pip
      uses: actions/cache@v4
      with:
        path: ~/.cache/pip
        key: ${{ runner.os }}-pip-${{ matrix.python-version }}-${{ hashFiles('chainforge/requirements.txt', 'chainforge/constraints.txt', 'setup.py') }}
        restore-keys: |
          ${{ runner.os }}-pip-${{ matrix.python-version }}-
          ${{ runner.os }}-pip-

    - name: Upgrade pip
      run: python -m pip install --upgrade pip

    - name: Install package
      run: |
        pip install -e .
    
    - name: Install with RAG extras
      run: |
        pip install -e .[rag]

    - name: Install pytest
      run: |
        pip install pytest

    # The RAG tests download sentence-transformers / model2vec models from the
    # HuggingFace Hub. Cache them so that every matrix entry doesn't re-fetch
    # a few hundred MB on each run.
    - name: Cache HuggingFace models
      uses: actions/cache@v4
      with:
        path: ~/.cache/huggingface
        key: ${{ runner.os }}-hf-${{ hashFiles('tests/*.py') }}
        restore-keys: |
          ${{ runner.os }}-hf-

    # TODO: Check that the server runs

    # NOTE: faiss-cpu is deliberately NOT installed here. It bundles its own
    # OpenMP runtime, and a second libomp in the same interpreter as torch's
    # segfaults (verified on macOS: merely having faiss-cpu installed, without
    # importing it, turns the embedding tests into SIGSEGV). The FaissVectorStore
    # tests therefore run in the separate step below.
    - name: Run tests
      env:
        HF_HUB_DISABLE_TELEMETRY: "1"
        # Tests needing a provider API key skip themselves when it is unset,
        # so this job stays green (and free) without secrets.
      run: pytest -v -m "not needs_faiss"

    # A second, isolated invocation for the optional FAISS backend. ChainForge
    # no longer offers a faiss extra -- the pip wheels collide with any other
    # OpenMP library, so we do not install it for anyone -- but the
    # faiss_vector_store retrieval method still works when a user installs it
    # themselves, and this keeps that path from rotting. faiss-cpu is named
    # directly here for the same reason.
    #
    # Installing it only now means it cannot destabilise the run above, and
    # excluding `slow` keeps this process from loading a real ML model, which
    # is the combination that crashes.
    - name: Install faiss and run the FAISS backend tests
      env:
        HF_HUB_DISABLE_TELEMETRY: "1"
      run: |
        pip install faiss-cpu
        pytest -v -m "needs_faiss and not slow"