How to add a fixture to tests under certain conditions at parsing time.

[cladmi]

How to add a fixture to tests in a hook using metafunc/request.

I saw a solution to add a fixture to tests during pytest_generate_tests with metafunc.fixturenames.append in pytest-repeat/#40

But this does not use the fixture implementation, it is just used during ordering/generation but is not executed at all during test execution. Or at least it is not the case anymore with pytest 7.0.

What would be the correct way of achieving this right now?

Thank you in advance.

Details

In a test library, we would like to be able to do pre-post checks for tests that
use a certain fixture.
Check that tests did not break the target fixture or allow doing recovery.

These checks could be defined in the test suite, so without
modifying the original target fixture, and would also need to be able to use fixtures.

---

Alternatives considered

Multi layers

This could be handled by having some intermediate fixture.
So def target_underlayer() and def target(target_underlayer).
So overwriting target in a project overwrites nothing, but then it may need to be done for each scopes if we want to allow checks at session/module/function scope.

Hooks

I thought about having instead hooks defined in the library, but then these
hooks callback cannot themselves rely on using extra fixtures.
(or I missed how to say "instanciate me this fixture dynamically")

Hacky autouse

We currently relied on the following pattern:

@pytest.fixture(autouse=True, scope="function")
def wrapper(request, tmp_path):
    if "target" not in request.fixturenames:
        yield
        return
    target = request.getfixturevalue("target")

    pre_checks(target, tmp_path)
    yield
    post_checks(target, tmp_path)

But that does not work reliably for anything else than function scope, as the
first test may not be using the fixture.

And the wrapper fixtures cannot use any fixture that rely on target as it
would be pulled in the same way.

Overwriting the fixture name

Some way could be the following syntax, but it does not work and it would force
overwriting the scope. (I could live with that)

@pytest.fixture(name="target", scope="function")
def target_wrapped(target):
    yield

Overwriting 'Fixturemanager'

I tried changing the "Fixturemanager" plugin, but I could not re-register a new
implementation as it was already loaded.

I could check more, but yeah not sure if it's a good way.

Idea

It could be phrased in the following way of writing:

@pytest.fixture(autouse=True, autouse_condition_callback=add_the_fixture_or_not)

Where it would get request or metafunc depending when its parsed.

I imagine the limitation of how fixture definition scopes, to which tests when
there are different conftest.py but yeah, did not look in these details yet.
I got blocked by not be able to overwrite fixturemanager.

Relying on "fixture(autouse=True)" has the issue that it trigger adding other
fixtures to the test even if not needed.
So we end-up having 'doctest' tests using a hardware target fixture so
triggering a flash or crashing as they do not have all the "expected" things a
test object normally has.

Last resort

The answer could also be "just copy-paste the implementation with all the private methods" but wanted to be sure before doing such things.

[cladmi]

Oh I just thought that I could rely on adding a pytest.mark.usefixtures to the function.

I will explore this and post update here.

[cladmi]

Did not manage to use mark.usefixtures.
The solution that worked for me is the following.

I was careful to not add duplicates but it may not be required.
It is working without the "if name in".

def pytest_generate_tests(metafunc):
    # Detect condition to do it
    usefixture(metafunc, "my_fixture")

def usefixture(metafunc, fixturename)
    if fixturename not in metafunc.fixturenames:
        # Prepend to try executing before normal ones.
        metafunc.fixturenames.insert(0, fixturename)

    fixtureinfo = metafunc.definition._fixtureinfo  # pylint: disable=protected-access
    # This is required to actually have the fixture instantiated
    if fixturename not in fixtureinfo.initialnames:
        fixtureinfo.initialnames = fixtureinfo.initialnames + (fixturename,)