ci.yml

name: CI

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

on:
  pull_request:
    branches:
    - main
  push:
    branches:
    - main
  release:
    types: [published]
  schedule:
  - cron: '37 11 * * 1,2,3,4,5'  # early morning (11:37 UTC / 4:37 AM PDT) Monday - Friday

env:
  # NOTE: Need to update `TORCH_VERSION`, and `TORCH_*_INSTALL` for new torch releases.
  TORCH_VERSION: 1.13.1
  # TORCH_CPU_INSTALL: conda install pytorch torchvision torchaudio cpuonly -c pytorch
  TORCH_CPU_INSTALL: pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu -c constraints.txt
  # Change this to invalidate existing cache.
  CACHE_PREFIX: v11
  # Disable tokenizers parallelism because this doesn't help, and can cause issues in distributed tests.
  TOKENIZERS_PARALLELISM: 'false'
  # Disable multithreading with OMP because this can lead to dead-locks in distributed tests.
  OMP_NUM_THREADS: '1'
  # See https://github.com/pytorch/pytorch/issues/37377#issuecomment-677851112.
  MKL_THREADING_LAYER: 'GNU'
  DEFAULT_PYTHON_VERSION: 3.8
  # For debugging GPU tests.
  CUDA_LAUNCH_BLOCKING: '1'

defaults:
  run:
    shell: bash -l {0}

jobs:
  changelog:
    name: CHANGELOG
    runs-on: ubuntu-latest
    if: github.event_name == 'pull_request'

    steps:
    - uses: actions/checkout@v3

    - name: Check if source files have changed
      run: |
        git diff --name-only $(git merge-base origin/main HEAD) | grep '^allennlp/.*\.py$' && echo "source_files_changed=true" >> $GITHUB_ENV || echo "source_files_changed=false" >> $GITHUB_ENV

    - name: Check that CHANGELOG has been updated
      if: env.source_files_changed == 'true'
      run: |
        # If this step fails, this means you haven't updated the CHANGELOG.md
        # file with notes on your contribution.
        git diff --name-only $(git merge-base origin/main HEAD) | grep '^CHANGELOG.md$' && echo "Thanks for helping keep our CHANGELOG up-to-date!"

  style:
    name: Style
    runs-on: ubuntu-latest

    steps:
    - uses: actions/checkout@v3

    - name: Setup Python
      uses: actions/setup-python@v4
      with:
        python-version: ${{ env.DEFAULT_PYTHON_VERSION }}

    - name: Install requirements
      run: |
        grep -E '^black' dev-requirements.txt | xargs pip install

    - name: Debug info
      run: |
        pip freeze

    - name: Run black
      run: |
        black --check .

  checks:
    name: ${{ matrix.task.name }}
    runs-on: ${{ matrix.task.runs_on }}
    timeout-minutes: 30
    strategy:
      fail-fast: false
      matrix:
        task:
        - name: Lint
          runs_on: ubuntu-latest
          coverage_report: false
          torch_platform: cpu
          run: |
            make flake8
            make typecheck

        - name: CPU Tests
          runs_on: ubuntu-latest
          coverage_report: true
          torch_platform: cpu
          run: make test

        - name: Model Tests
          runs_on: ubuntu-latest
          coverage_report: true
          torch_platform: cpu
          run: |
            cd allennlp-models
            make test-with-cov COV=allennlp
            mv coverage.xml ../

    steps:
    - uses: actions/checkout@v3

    - uses: conda-incubator/setup-miniconda@v2
      with:
        miniconda-version: "latest"
        python-version: ${{ env.DEFAULT_PYTHON_VERSION }}

    - name: Set build variables
      run: |
        # Get the exact Python version to use in the cache key.
        echo "PYTHON_VERSION=$(python --version)" >> $GITHUB_ENV
        echo "RUNNER_ARCH=$(uname -m)" >> $GITHUB_ENV
        # Use week number in cache key so we can refresh the cache weekly.
        echo "WEEK_NUMBER=$(date +%V)" >> $GITHUB_ENV

    - name: Set build variables (CPU only)
      if: matrix.task.torch_platform == 'cpu'
      run: |
        echo "TORCH_INSTALL=$TORCH_CPU_INSTALL" >> $GITHUB_ENV

    - uses: actions/cache@v3
      id: virtualenv-cache
      with:
        path: .venv
        key: ${{ env.CACHE_PREFIX }}-${{ env.WEEK_NUMBER }}-${{ runner.os }}-${{ env.RUNNER_ARCH }}-${{ env.PYTHON_VERSION }}-${{ matrix.task.torch_platform }}-${{ hashFiles('setup.py') }}-${{ hashFiles('*requirements.txt', 'constraints.txt') }}

    - name: Setup virtual environment (no cache hit)
      if: steps.virtualenv-cache.outputs.cache-hit != 'true'
      run: |
        python${{ env.DEFAULT_PYTHON_VERSION }} -m venv .venv
        source .venv/bin/activate
        make install TORCH_INSTALL="$TORCH_INSTALL"

    - name: Setup virtual environment (cache hit)
      if: steps.virtualenv-cache.outputs.cache-hit == 'true'
      run: |
        source .venv/bin/activate
        pip install --no-deps -e .[all]
        make download-extras

    - name: Pull and install models repo
      if: matrix.task.name == 'Model Tests'
      env:
        ALLENNLP_VERSION_OVERRIDE: ""  # Don't replace the core library.
      run: |
        source .venv/bin/activate
        git clone https://github.com/allenai/allennlp-models.git
        cd allennlp-models
        # git checkout dependabot/pip/torch-gte-1.7.0-and-lt-1.13.0
        pip install -e .[dev,all]

    - name: Debug info
      run: |
        source .venv/bin/activate
        pip freeze

    - name: Ensure torch up-to-date
      run: |
        source .venv/bin/activate
        python scripts/check_torch_version.py

    - name: ${{ matrix.task.name }}
      run: |
        source .venv/bin/activate
        ${{ matrix.task.run }}

    - name: Prepare coverage report
      if: matrix.task.coverage_report
      run: |
        mkdir coverage
        mv coverage.xml coverage/

    - name: Save coverage report
      if: matrix.task.coverage_report
      uses: actions/upload-artifact@v3
      with:
        name: ${{ matrix.task.name }}-coverage
        path: ./coverage

    - name: Clean up
      if: always()
      run: |
        # Could run into issues with the cache if we don't uninstall the editable.
        # See https://github.com/pypa/pip/issues/4537.
        source .venv/bin/activate
        pip uninstall --yes allennlp allennlp-models

  upload_coverage:
    name: Upload Coverage Report
    timeout-minutes: 5
    if: github.repository == 'EducationalTestingService/allennlp' && (github.event_name == 'push' || github.event_name == 'pull_request')
    runs-on: ubuntu-latest
    needs: [checks]

    steps:
      # Need to checkout code to get the coverage config.
    - uses: actions/checkout@v3

    - name: Download coverage report from CPU tests
      uses: actions/download-artifact@v3
      with:
        name: CPU Tests-coverage
        path: coverage/cpu_tests

    - name: Download coverage report from model tests
      uses: actions/download-artifact@v3
      with:
        name: Model Tests-coverage
        path: coverage/model_tests

    - name: Upload coverage to Codecov
      uses: codecov/codecov-action@v2
      with:
        files: coverage/cpu_tests/coverage.xml,coverage/model_tests/coverage.xml
        # Ignore codecov failures as the codecov server is not
        # very reliable but we don't want to report a failure
        # in the github UI just because the coverage report failed to
        # be published.
        fail_ci_if_error: false