fixtures: fix incorrect "duplicate parametrization" when using indirect=[...]

[bluetech]

The code in _get_direct_parametrize_args only handled indirect=True/False, it didn't account for indirect=[...]. This caused the added test to start failing after d56b1af.

Fix by using the common _resolve_args_directness function. It's a bit heavy for the task but correctness first...

Fix #13974.

I am not marking for backport because the bug was laying dormant for a while and only uncovered by d56b1af which is only in main, not yet released.

[braingram]

I tested this with the downstream test suite which lead to #13974 and with this PR the tests are back to passing. Thanks for the quick fix!
https://github.com/asdf-format/asdf/actions/runs/19445301685/job/55638606133?pr=1990#step:10:113

[Zac-HD]

lgtm!

[dmakhno]

Hi @bluetech ,

Cannot find anything in changelog to 9.1.0 like compatibility change.
Just to acknowledge this is expect now behaviour

import pytest

DEFAULT = ["a", "b", "c"]

@pytest.fixture(params=DEFAULT)
def target(request):
    return request.param

@pytest.mark.parametrize(
    "target",
    ["a", "b"],  # override default
    indirect=True,
)
def test_foo(target):
    assert True

which used to work on 9.0.3 (the motivation fixture in conftest has default), is causing duplicate parametrization of 'target' on 9.1.0

And must be fixed using pytest_generate_tests, making "target" parametless.

@pytest.fixture
def target(request):
    return request.param

def pytest_generate_tests(self, metafunc: Metafunc) -> None:
    if not_missing_default(metafunc):  # function that iter_markers("parametrize") and its args
        return
    metafunc.parametrize('target', DEFAULT, indirect=True)

Like https://github.com/datachain-ai/datachain/pull/1818/changes

[bluetech]

@dmakhno Please see #14591, will be fixed in 9.1.1 which we'll hopefully release this weekend.