-
Notifications
You must be signed in to change notification settings - Fork 10
Open
Description
When running numpy/_core/tests/test_finfo.py with pytest-run-parallel and 3.14t-dev, it results in test failures for the entire file. All errors seem to involve request.getfixturevalue, printing out recursive dependency involving fixture ... detected. The test file uses request.getfixturevalue to get fixtures through @pytest.mark.parametrize.
An error from the test file:
================================================================================================= ERRORS =================================================================================================
___________________________________________________________________ ERROR at call of test_finfo_properties[machep-float16-float16_ma] ____________________________________________________________________
file /home/brit1/git/numpy/build-install/usr/lib/python3.14t/site-packages/numpy/_core/tests/test_finfo.py, line 63
@pytest.mark.parametrize("dtype,ma_fixture", [
(np.half, "float16_ma"),
(np.float32, "float32_ma"),
(np.float64, "float64_ma"),
])
@pytest.mark.parametrize("prop", [
'machep', 'negep', 'minexp', 'maxexp', 'nmant', 'iexp',
'eps', 'epsneg', 'precision', 'resolution'
])
def test_finfo_properties(dtype, ma_fixture, prop, request):
E recursive dependency involving fixture 'float16_ma' detected
> available fixtures: add_np, cache, capfd, capfdbinary, caplog, capsys, capsysbinary, check_fpu_mode, cov, doctest_namespace, float16_ma, float32_ma, float64_ma, iteration_index, monkeypatch, no_cover, num_iterations, num_parallel_threads, pytestconfig, record_property, record_testsuite_property, record_xml_attribute, recwarn, testrun_uid, thread_comp, thread_index, tmp_path, tmp_path_factory, tmpdir, tmpdir_factory, worker_id
> use 'pytest --fixtures [testpath]' for help on them.Strangely enough, the test runs fine under pytest-run-parallel when using a GIL-enabled build of Python. So this request.getfixturevalue bug likely involves free-threaded Python in some way (in contrast, it runs fine under free-threaded Python when not ran in parallel).
Metadata
Metadata
Assignees
Labels
No labels