test.yml

name: tests

on:
  workflow_call:
  workflow_dispatch:
  merge_group:
  push:
    branches: [main]
  workflow_run:
    workflows:
      - "contributor check"
    types:
      - completed

# The path to the sha of the commit we want to test will depend
# on the trigger type. If a workflow_run trigger, then we want
# the sha from the head of the branch that triggered it. For
# pushes and manual triggers, we want the sha of the branch.
env:
  CI: true
  HEAD_SHA: |
    ${{
      github.event.workflow_run.head_sha ||
      github.event.pull_request.head.sha ||
      github.sha
    }}

jobs:
  test:
    runs-on: 32-core-ubuntu
    # Explicit `name:` overrides GitHub Actions' default check-run name.
    # Without it, GitHub auto-expands EVERY field of the object-typed
    # `shard` matrix entry — including `shard.paths` — into a
    # >100-character truncated string like
    # `test (3.14, core, tests/core/models tests/core/graph ...`.
    # That diverges from the short custom status emitted by
    # multi-trigger-setup, forcing branch protection to pick one event
    # type's names and break the others.
    #
    # We use a literal job key (`test`) rather than `${{ github.job }}`
    # because `github.job` is only populated inside the execution of
    # `steps:` per the GitHub docs, and resolves to null at job-level
    # `name:` evaluation time. The literal matches what `github.job`
    # would have produced and matches the multi-trigger-setup template
    # verbatim, so the auto-generated check-run name and the custom
    # commit status now share a single string set across all event
    # types (push, merge_group, workflow_run, workflow_dispatch).
    name: 'test (${{ matrix.python_version }}, ${{ matrix.shard.name }})'
    permissions:
      contents: read
      # Needed to manually set commit status
      statuses: write
    strategy:
      max-parallel: 10
      fail-fast: false
      matrix:
        python_version: ['3.11', '3.14']
        # tests/perf is run by perf-test.yml on a separate cadence.
        # tests/applications and tests/lammps are run by integration-test.yml
        # and test_lammps_gpu respectively. Both shards below intentionally
        # exclude those paths.
        # YAML anchor: shard list is reused by test_cpu_sweep below.
        # Do not reorder jobs (the alias must follow the anchor textually).
        shard: &cpu_shards
          - name: units
            paths: tests/core/units
          - name: core
            paths: tests/core/models tests/core/graph tests/core/calculate tests/core/components tests/core/common tests/core/datasets tests/core/evaluator tests/core/modules tests/core/scripts tests/core/launchers tests/core/test_cli.py tests/core/test_download_large_files.py tests/data tests/demo

    steps:
      - name: Checkout code
        uses: actions/checkout@v7
        with:
          ref: ${{ env.HEAD_SHA }}
          # Required since actions/checkout began refusing fork PR code in
          # workflow_run jobs. Safe here: fork PRs from all external
          # contributors require maintainer approval before this trusted,
          # secret-holding run can execute (repo Actions setting).
          # See https://gh.io/securely-using-pull_request_target
          allow-unsafe-pr-checkout: true
          # Optional: get full history if needed
          fetch-depth: 0

      - name: Setup
        uses: ./.github/actions/multi-trigger-setup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }})'

      - name: Set up Python ${{ matrix.python_version }}
        uses: actions/setup-python@v6
        with:
          python-version:  ${{ matrix.python_version }}

      - name: Cache pip
        uses: actions/cache@v6
        with:
          path: ~/.cache/pip
          key: ${{ runner.os }}-pip-${{ hashFiles('requirements.txt') }}
          restore-keys: |
            ${{ runner.os }}-pip-
            ${{ runner.os }}-

      - name: Install core dependencies and package
        run: |
          python -m pip install --upgrade pip
          pip install \
            packages/fairchem-core[dev,extras] \
            packages/fairchem-data-oc[dev] \
            packages/fairchem-data-omat \
            packages/fairchem-data-omol[dev] \
            packages/fairchem-demo-ocpapi[dev] \
            -r tests/requirements.txt # pin test packages

      - name: Install torchsim (Python 3.12+)
        if: ${{ matrix.python_version == '3.12' || matrix.python_version == '3.13' || matrix.python_version == '3.14' }}
        run: |
          pip install packages/fairchem-core[torchsim]

      - name: Install additional dependencies
        if: matrix.shard.name == 'core'
        run: |
          wget https://github.com/m3g/packmol/archive/refs/tags/v20.15.0.tar.gz
          tar -xzvf v20.15.0.tar.gz
          cd packmol-20.15.0
          ./configure
          make
          echo "$(readlink -f .)" >> $GITHUB_PATH

      - name: Download large test files
        if: matrix.shard.name == 'core'
        run: coverage run -m fairchem.core.scripts.download_large_files oc

      - name: Core tests (${{ matrix.shard.name }})
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        # --exclude-models deselects tests locked to sweep models and
        # removes sweep models from all-models parametrization. The
        # sweep jobs (test_cpu_sweep) handle those models.
        run: |
          NCCL_DEBUG=INFO pytest -n auto -m "not subprocess and not serial and not gpu" ${{ matrix.shard.paths }} --durations=0 -vv --ignore=tests/demo/ocpapi/tests/integration/ --exclude-models=uma-s-1p1,uma-s-1p2 --cov-report=xml --cov=fairchem --cov-append --junitxml=junit-${{ matrix.shard.name }}.xml -o junit_family=legacy -c ./packages/fairchem-core/pyproject.toml --suppress-no-test-exit-code

      - name: Serial CPU tests (${{ matrix.shard.name }})
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          pytest -m "serial and not gpu" ${{ matrix.shard.paths }} -vv --ignore=tests/demo/ocpapi/tests/integration/ --exclude-models=uma-s-1p1,uma-s-1p2 --cov-report=xml --cov=fairchem --cov-append --junitxml=junit-serial-${{ matrix.shard.name }}.xml -o junit_family=legacy -c ./packages/fairchem-core/pyproject.toml --suppress-no-test-exit-code

      - if: ${{ matrix.python_version == '3.14' }}
        name: codecov-coverage
        uses: codecov/codecov-action@v7
        with:
          fail_ci_if_error: false
          files: ./coverage.xml
          token: ${{ secrets.CODECOV_TOKEN }}
          verbose: true

      - if: ${{ matrix.python_version == '3.14' }}
        name: codecov-test-results
        uses: codecov/test-results-action@v1
        with:
          files: ./junit-${{ matrix.shard.name }}.xml,./junit-serial-${{ matrix.shard.name }}.xml
          token: ${{ secrets.CODECOV_TOKEN }}
          commit_parent: ${{ env.HEAD_SHA }}
          verbose: true

      - name: Cleanup
        if: always()
        uses: ./.github/actions/multi-trigger-cleanup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }})'

  test_cpu_sweep:
    runs-on: 32-core-ubuntu
    # See `test` job above for the `name:` rationale (literal job key,
    # not `${{ github.job }}`, to avoid the null-at-job-level pitfall).
    name: 'test_cpu_sweep (${{ matrix.python_version }}, ${{ matrix.shard.name }}, ${{ matrix.sweep_model }})'
    permissions:
      contents: read
      statuses: write
    strategy:
      fail-fast: false
      matrix:
        python_version: ['3.14']
        sweep_model: [uma-s-1p1, uma-s-1p2]
        shard: *cpu_shards

    steps:
      - name: Checkout code
        uses: actions/checkout@v7
        with:
          ref: ${{ env.HEAD_SHA }}
          # See the checkout step in the `test` job for why this is set
          # and why it is safe (external-contributor approval gate).
          allow-unsafe-pr-checkout: true
          fetch-depth: 0

      - name: Setup
        uses: ./.github/actions/multi-trigger-setup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }}, ${{ matrix.sweep_model }})'

      - name: Set up Python ${{ matrix.python_version }}
        uses: actions/setup-python@v6
        with:
          python-version: ${{ matrix.python_version }}

      - name: Cache pip
        uses: actions/cache@v6
        with:
          path: ~/.cache/pip
          key: ${{ runner.os }}-pip-${{ hashFiles('requirements.txt') }}
          restore-keys: |
            ${{ runner.os }}-pip-
            ${{ runner.os }}-

      - name: Install core dependencies and package
        run: |
          python -m pip install --upgrade pip
          pip install \
            packages/fairchem-core[dev,extras] \
            packages/fairchem-data-oc[dev] \
            packages/fairchem-data-omat \
            packages/fairchem-data-omol[dev] \
            packages/fairchem-demo-ocpapi[dev] \
            -r tests/requirements.txt # pin test packages

      - name: Install torchsim (Python 3.12+)
        if: ${{ matrix.python_version == '3.12' || matrix.python_version == '3.13' || matrix.python_version == '3.14' }}
        run: |
          pip install packages/fairchem-core[torchsim]

      - name: Install additional dependencies
        if: matrix.shard.name == 'core'
        run: |
          wget https://github.com/m3g/packmol/archive/refs/tags/v20.15.0.tar.gz
          tar -xzvf v20.15.0.tar.gz
          cd packmol-20.15.0
          ./configure
          make
          echo "$(readlink -f .)" >> $GITHUB_PATH

      - name: Download large test files
        if: matrix.shard.name == 'core'
        run: coverage run -m fairchem.core.scripts.download_large_files oc

      - name: Sweep CPU tests (${{ matrix.shard.name }}, ${{ matrix.sweep_model }})
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          NCCL_DEBUG=INFO pytest -n auto -m "not subprocess and not serial and not gpu" ${{ matrix.shard.paths }} --durations=0 -vv --ignore=tests/demo/ocpapi/tests/integration/ --sweep-model=${{ matrix.sweep_model }} --junitxml=junit-sweep-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml -o junit_family=legacy -c ./packages/fairchem-core/pyproject.toml --suppress-no-test-exit-code

      - name: Sweep serial CPU tests (${{ matrix.shard.name }}, ${{ matrix.sweep_model }})
        # The core shard has no serial @pretrained tests, so the sweep
        # serial step would collect zero tests there (pytest exit 5).
        # Skip the step on core; only run on units.
        if: matrix.shard.name == 'units'
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          pytest -m "serial and not gpu" ${{ matrix.shard.paths }} -vv --ignore=tests/demo/ocpapi/tests/integration/ --sweep-model=${{ matrix.sweep_model }} --junitxml=junit-sweep-serial-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml -o junit_family=legacy -c ./packages/fairchem-core/pyproject.toml --suppress-no-test-exit-code

      - name: codecov-test-results
        if: always()
        uses: codecov/test-results-action@v1
        with:
          files: ./junit-sweep-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml,./junit-sweep-serial-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml
          token: ${{ secrets.CODECOV_TOKEN }}
          commit_parent: ${{ env.HEAD_SHA }}
          verbose: true

      - name: Cleanup
        if: always()
        uses: ./.github/actions/multi-trigger-cleanup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }}, ${{ matrix.sweep_model }})'

  test_gpu:
    runs-on: 4-core-ubuntu-gpu-t4
    # See `test` job above for the `name:` rationale.
    name: 'test_gpu (${{ matrix.python_version }}, ${{ matrix.shard.name }})'
    permissions:
      contents: read
      # Needed to manually set commit status
      statuses: write
    strategy:
      fail-fast: false
      matrix:
        python_version: ['3.14']
        # YAML anchor: shard list is reused by test_gpu_sweep below.
        # Do not reorder jobs (the alias must follow the anchor textually).
        shard: &gpu_shards
          - name: models
            paths: tests/core/models tests/core/graph tests/core/calculate tests/core/components tests/core/datasets tests/core/common tests/core/scripts
          - name: units
            paths: tests/core/units

    steps:
      - name: Checkout code
        uses: actions/checkout@v7
        with:
          ref: ${{ env.HEAD_SHA }}
          # Required since actions/checkout began refusing fork PR code in
          # workflow_run jobs. Safe here: fork PRs from all external
          # contributors require maintainer approval before this trusted,
          # secret-holding run can execute (repo Actions setting).
          # See https://gh.io/securely-using-pull_request_target
          allow-unsafe-pr-checkout: true
          # Optional: get full history if needed
          fetch-depth: 0

      - name: Setup
        uses: ./.github/actions/multi-trigger-setup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }})'

      - name: Set up Python ${{ matrix.python_version }}
        uses: actions/setup-python@v6
        with:
          python-version:  ${{ matrix.python_version }}

      - name: Install core dependencies and package
        run: |
          python -m pip install --upgrade pip
          pip install packages/fairchem-core[dev,extras] \
          packages/fairchem-data-omol[dev] \
          packages/fairchem-data-omat \
          -r tests/requirements.txt # pin test packages

      - name: Install torchsim (Python 3.12+)
        if: ${{ matrix.python_version == '3.12' || matrix.python_version == '3.13' || matrix.python_version == '3.14' }}
        run: |
          pip install packages/fairchem-core[torchsim]

      - name: Core GPU tests (${{ matrix.shard.name }})
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        # Note: These tests seem to fail if "-n auto" is added
        # --exclude-models deselects tests locked to sweep models and
        # removes sweep models from all-models parametrization. The
        # sweep jobs (test_gpu_sweep) handle those models.
        run: |
          NCCL_DEBUG=INFO pytest ${{ matrix.shard.paths }} --durations=0 -vv -m "gpu and not compile_gpu" --exclude-models=uma-s-1p1,uma-s-1p2 -c ./packages/fairchem-core/pyproject.toml --junitxml=junit-gpu-${{ matrix.shard.name }}.xml --suppress-no-test-exit-code

      - name: Core GPU compile tests (${{ matrix.shard.name }})
        # compile_gpu tests exist in both models and units shards.
        # Sweep-eligible compile_gpu tests (e.g. @pretrained-locked to
        # UMA-S) run in test_gpu_sweep, not here.
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          NCCL_DEBUG=INFO pytest ${{ matrix.shard.paths }} --durations=0 -vv -m compile_gpu --exclude-models=uma-s-1p1,uma-s-1p2 -c ./packages/fairchem-core/pyproject.toml --junitxml=junit-gpu-compile-${{ matrix.shard.name }}.xml --suppress-no-test-exit-code

      - name: codecov-test-results
        if: always()
        uses: codecov/test-results-action@v1
        with:
          files: ./junit-gpu-${{ matrix.shard.name }}.xml,./junit-gpu-compile-${{ matrix.shard.name }}.xml
          token: ${{ secrets.CODECOV_TOKEN }}
          commit_parent: ${{ env.HEAD_SHA }}
          verbose: true

      - name: Cleanup
        if: always()
        uses: ./.github/actions/multi-trigger-cleanup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }})'

  test_gpu_sweep:
    runs-on: 4-core-ubuntu-gpu-t4
    # See `test` job above for the `name:` rationale.
    name: 'test_gpu_sweep (${{ matrix.python_version }}, ${{ matrix.shard.name }}, ${{ matrix.sweep_model }})'
    permissions:
      contents: read
      statuses: write
    strategy:
      fail-fast: false
      matrix:
        python_version: ['3.14']
        sweep_model: [uma-s-1p1, uma-s-1p2]
        shard: *gpu_shards

    steps:
      - name: Checkout code
        uses: actions/checkout@v7
        with:
          ref: ${{ env.HEAD_SHA }}
          # See the checkout step in the `test` job for why this is set
          # and why it is safe (external-contributor approval gate).
          allow-unsafe-pr-checkout: true
          fetch-depth: 0

      - name: Setup
        uses: ./.github/actions/multi-trigger-setup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }}, ${{ matrix.sweep_model }})'

      - name: Set up Python ${{ matrix.python_version }}
        uses: actions/setup-python@v6
        with:
          python-version: ${{ matrix.python_version }}

      - name: Install core dependencies and package
        run: |
          python -m pip install --upgrade pip
          pip install packages/fairchem-core[dev,extras] \
          packages/fairchem-data-omol[dev] \
          packages/fairchem-data-omat \
          -r tests/requirements.txt # pin test packages

      - name: Install torchsim (Python 3.12+)
        if: ${{ matrix.python_version == '3.12' || matrix.python_version == '3.13' || matrix.python_version == '3.14' }}
        run: |
          pip install packages/fairchem-core[torchsim]

      - name: Sweep tests (${{ matrix.shard.name }}, ${{ matrix.sweep_model }})
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          NCCL_DEBUG=INFO pytest ${{ matrix.shard.paths }} --durations=0 -vv -m "gpu and not compile_gpu" --sweep-model=${{ matrix.sweep_model }} -c ./packages/fairchem-core/pyproject.toml --junitxml=junit-gpu-sweep-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml --suppress-no-test-exit-code

      - name: Sweep compile GPU tests (${{ matrix.shard.name }}, ${{ matrix.sweep_model }})
        # Covers @compile_gpu + @pretrained-locked-to-UMA-S tests that
        # the base test_gpu compile step excludes via --exclude-models.
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          NCCL_DEBUG=INFO pytest ${{ matrix.shard.paths }} --durations=0 -vv -m compile_gpu --sweep-model=${{ matrix.sweep_model }} -c ./packages/fairchem-core/pyproject.toml --junitxml=junit-gpu-sweep-compile-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml --suppress-no-test-exit-code

      - name: codecov-test-results
        if: always()
        uses: codecov/test-results-action@v1
        with:
          files: ./junit-gpu-sweep-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml,./junit-gpu-sweep-compile-${{ matrix.shard.name }}-${{ matrix.sweep_model }}.xml
          token: ${{ secrets.CODECOV_TOKEN }}
          commit_parent: ${{ env.HEAD_SHA }}
          verbose: true

      - name: Cleanup
        if: always()
        uses: ./.github/actions/multi-trigger-cleanup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }}, ${{ matrix.shard.name }}, ${{ matrix.sweep_model }})'

  test_lammps_gpu:
    runs-on: 4-core-ubuntu-gpu-t4
    permissions:
      contents: read
      # Needed to manually set commit status
      statuses: write
    strategy:
      max-parallel: 1
      matrix:
        python_version: ['3.14']

    steps:
      - uses: actions/checkout@v7
      - uses: mamba-org/setup-micromamba@v3
        with:
          generate-run-shell: true # This is the default behavior
          environment-name: test_env # Specifies a new environment to be created
          create-args: python=3.14 # Specifies packages to install into the new environment

      - name: Checkout code
        uses: actions/checkout@v7
        with:
          ref: ${{ env.HEAD_SHA }}
          # Required since actions/checkout began refusing fork PR code in
          # workflow_run jobs. Safe here: fork PRs from all external
          # contributors require maintainer approval before this trusted,
          # secret-holding run can execute (repo Actions setting).
          # See https://gh.io/securely-using-pull_request_target
          allow-unsafe-pr-checkout: true
          # Optional: get full history if needed
          fetch-depth: 0

      - name: Setup
        uses: ./.github/actions/multi-trigger-setup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }})'

      - name: Set up Python ${{ matrix.python_version }}
        uses: actions/setup-python@v6
        with:
          python-version:  ${{ matrix.python_version }}

      - name: Install core dependencies and package
        shell: micromamba-shell {0}
        run: |
          micromamba install lammps
          python -m pip install --upgrade pip
          pip install packages/fairchem-core[dev,extras] \
          packages/fairchem-lammps \
          -r tests/requirements.txt # pin test packages

      - name: Core GPU tests
        shell: micromamba-shell {0}
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          NCCL_DEBUG=INFO pytest tests/lammps --durations=0 -vv -m gpu

      - name: Cleanup
        if: always()
        uses: ./.github/actions/multi-trigger-cleanup
        with:
          status-context: '${{ github.job }} (${{ matrix.python_version }})'