A GitHub Action to enable running pytest --last-failed in your CI

This composite GitHub Action makes use of GitHub's Cache action to enable pytest's --last-failed functionality for your CI workflow to speed up testing in PRs, without compromising on having the full test suite running and passing before a merge.

This action requires that Python is already setup and that pytest is already installed in the environment.

Then, if your workflow currently contains a pytest step as follows:

  - name: pytest
    run: pytest --my --pytest --arguments

you can replace it with:

  - name: pytest
    uses: sjvrijn/pytest-last-failed@v3
    with:
      # Additional arguments to be passed to pytest
      pytest-args: '--my --pytest --arguments'

      # Whether to perform a full run of the test-suite once any previously failed
      # tests pass again. This may catch any unintended side effects of fixing the
      # previously failed tests, at the cost of running the entire suite again.
      # Default: false
      run-all-on-success: ''

That's all! Note that this means you don't have to add the --last-failed argument yourself, that's taken care of by using this action.

For any first run, it will always run the entire test suite. On subsequent runs, if the pytest cache indicates any last-failed tests, it will run pytest --last-failed instead.

If that run succeded, i.e., all previously failed tests now pass again, the run-all-on-success flag can be set to perform another run of the whole test suite to check for any unintended side effects of the implemented fixes.

In pseudo-code:

if not cache.has_last_failed_entries():
    pytest --cache-clear $PYTEST-ARGS
else:
    result = $( pytest --last-failed $PYTEST-ARGS )

    if result == 'success' and $RUN-ALL-ON-SUCCESS:
        pytest --cache-clear $PYTEST-ARGS