martinfleis/reverse-dependency-testing

GitHub Action for reverse dependency testing of Python packages

Reverse dependency testing

NOTE: This is an alpha version, proceed with caution.

This GitHub Action allows you to query conda-forge dependency tree and run tests included in packages that depend on your package against your development version.

The workflow is able to run tests only if the downstream packages do package them, including all the necessary files. Otherwise, it will skip the package with no collected tests or report a failure.

Usage

For the simple packages, the setup is simple, you just need to pass a package_name used for the dependency tree query:

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage

Ignore packages

In some cases, you may want to ignore certain packages (since they cannot be installed or for any other reason):

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    ignore: broken_package

Select only a subset of packages

Alternatively, you may run only a whitelisted subset of the dependents:

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    whitelist: >-
        package1
        package2

Include additional packages

Some packages you may want to test against may depend on your package only optionally and will not show up in the dependency tree. Those can be explicitly included in the workflow and tested with others using the include input.

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    include: >-
        package1
        package2

Specify environment

You can set up the environment with dependencies required to build your package by passing the conda environment file:

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    env: environment.yml

Alternatively, (or additionally) you can specify packages to be installed in the environment in the install input:

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    install: package1 package2

The default Python version installed in the environment is 3.11. You can override that either in your environment file or using python_version.

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    python_version: 3.9

If you need to run a specific command in the docker container prior the installation of the packages, you can specify it in the run input. The command will be run before setting up the conda environment. For example, to install GCC you can do:

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    run: apt-get update && apt install build-essential -y

Customize installation command

By default, the action attempts to install the package using pip install .. However, you can specify your own command:

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    installation_command: pip install --no-deps --no-build-isolation .

Fail or pass on test failure

The action returns an exit code 1 (failure) if there is at least one failed test. You can avoid it by setting fail_on_failure to false.

- uses: scientific-python/reverse-dependency-testing@main
with:
    package_name: mypackage
    fail_on_failure: false

How to ensure that tests are packaged?

The easiest way to ensure that the tests are packaged is to store them in the package folder (usually called package_name or src/package_name).

mypackage
├── pyproject.toml
|   README.md
|   LICENCE
└── mypackage
    ├── __init__.py
    ├── my_module.py
    └──tests
        └──test_my_module.py

You can verify that the tests can be found by running

pytest --pyargs mypackage

Packaging of tests is useful not only for this action but also to allow users to verify that your package is correctly installed in their environment.

How does the Action work?

The principle is simple:

1. Use micromamba repoquery whoneeds mypackage -c conda-forge to fetch the direct dependents of mypackage based on the conda-forge index.
2. De-duplicate the list to retrieve unique package names.
3. Create an environment with all the packages and install mypackage from main.
4. Loop over the package names and run pytest --pyargs packagename for each.