usefixtures as a Class Decorator

This is part of a series of posts on Python unit testing.

In the fixtures post we saw that usefixtures is a way for a test to use a fixture without taking it as an argument:

@pytest.mark.usefixtures('routes')
def test_something():
    ...

The test doesn’t take the routes fixture as a method argument, and doesn’t need access to the routes fixture in the test body, but the test does need pytest to call the routes fixture function for it before calling the test. The @pytest.mark.usefixtures('routes') decorator on the test function tells pytest to do that.

You also can put a @pytest.mark.usefixtures(...) decorator on a test class and pytest will automatically call those fixtures for every test method in the class, even if the methods themselves don’t take the fixtures as arguments.

We often use this with patch-based fixtures when all tests for a given function should ensure that the function under test uses a mock version of an imported library and it would be a mistake for any one of this function’s tests to be missing the fixture and use the real library.

For example, h/views/api.py::create() is the view function that’s called when someone POSTs a new annotation to the https://hypothes.is/api/annotations URL. It calls the storage module to save the new annotation to the database. storage has its own tests and we never want any of the tests for create() to be accessing the real database, so we always want to ensure that storage is replaced with a mock object for all of views.py::create()’s tests. This is done with a usefixtures() on the test class:

@pytest.mark.usefixtures('storage')
class TestCreate(object):

    def test_something(self):
        # The storage fixture will be used in this test, even though the test
        # method itself didn't declare it.

    # If some tests _do_ want to use the mock storage object in their test
    # method bodies, they can just take the fixture as an argument as usual.
    def test_it_creates_the_annotation_in_storage(self, storage):
        ... # (Call the create() view to create an annotation)

        # Test that it would have saved the annotation to storage.
        storage.create_annotation.assert_called_once_with(...)

    @pytest.fixture
    def storage(self, patch):
        return patch('h.views.api.storage')

The @pytest.mark.usefixtures('storage') class decorator means that pytest will automatically call the storage() fixture before every test method in the TestCreate class, whether the test method declares the fixture itself or not. If any particular test method wants to use the fixture value it can just have an argument named storage, as with any other fixture, but test methods that don’t need to use the fixture value don’t need to have the argument.

Using usefixtures as a test class decorator reduces the number of boilerplate and duplicate lines by not having to put a usefixtures decorator on every test method, reduces the chances of a mistake where one test method that should be using a mock is instead using the real dependency because it’s missing a usefixtures decorator, and is one more reason to group test methods into classes.

In the next and last post in the tutorial we’ll cover one final technique used in the Hypothesis Python tests - matcher objects.

Sean Hammond,