Testing

coola.testing

Provide pytest marks and fixtures to test code that has optional dependencies (e.g. NumPy, pandas, PyTorch).

The marks are defined in coola.testing.marks. Import one from there and use it as a test decorator, e.g. from coola.testing.marks import numpy_available then @numpy_available skips the test unless NumPy is installed. The fixtures are defined in coola.testing.fixtures: load them with pytest_plugins = ["coola.testing.fixtures"] in the top-level conftest.py.

pytest is required: install coola with the testing extra (pip install coola[testing]).

coola.testing.marks

Define some pytest fixtures for testing.

pytest is required to use these fixtures.

Each optional dependency has a pair of marks: <name>_available skips the test unless the dependency is available, and <name>_not_available skips it if the dependency is available. They are created by requires from one row of _DEPENDENCIES, so supporting a new dependency takes one line.

Most marks are named after one package. The torch_cuda_*, torch_mps_* and torch_numpy_* marks are named after a feature of PyTorch or a pair of packages: they check that PyTorch is installed and that the CUDA device, the MPS device or NumPy (in a compatible version) is usable.

coola.testing.marks.requires

requires(
    is_available: bool, name: str
) -> tuple[MarkDecorator, MarkDecorator]

Create the pair of marks for a dependency.

Use it to support a new optional dependency: the marks are meant to be assigned to module-level names, so they can be imported and used as test decorators.

Parameters:

Name	Type	Description	Default
is_available	bool	Whether the dependency is available.	required
name	str	The name of the dependency in the skip reasons, e.g. "PyTorch with CUDA".	required

Returns:

Type	Description
tuple[MarkDecorator, MarkDecorator]	The (available, not_available) marks. The first one skips the test if the dependency is not available, the second one skips it if the dependency is available.

Example

>>> from coola.testing.marks import requires
>>> numpy_available, numpy_not_available = requires(True, "NumPy")
>>> numpy_available.kwargs["reason"]
'Requires NumPy'
>>> numpy_not_available.kwargs["reason"]
'Skip if NumPy is available'

coola.testing.fixtures

Define a pytest fixture to isolate the default registries.

pytest is required to use this fixture.

coola.testing.fixtures.isolated_default_registries

isolated_default_registries() -> Generator[None]

Restore the default registries after the test.

A test that registers a type in a default registry (for example with register_equality_testers) changes it for the rest of the process. With this fixture, the test works on copies of the default registries and the original ones are restored when it ends. Load the module as a plugin in the top-level conftest.py (pytest_plugins = ["coola.testing.fixtures"]) and use the fixture with pytest.mark.usefixtures("isolated_default_registries").

Yields:

Type	Description
Generator[None]	None. The default registries are restored when the test ends.

Example

>>> from coola.testing.fixtures import isolated_default_registries
>>> isolated_default_registries.__name__
'isolated_default_registries'