--durations is tricked by freezegun

--durations is tricked by freezegun, when freezegun is used in autouse fixture.

Example:

from freezegun import freeze_time

@pytest.yield_fixture(autouse=True)
def frozen_time():
    with freeze_time("2020-01-01T12:00:00Z"):
        yield
        
def test_my_django_view(admin_client):
    resp = admin_client.get(...)
    assert resp.status_code == 200

...

Command:

pytest --reuse-db -s -vv --durations=5

Result:

============================================= slowest 5 durations =============================================

1609479015.03s setup    test_my_django_view1
1609479014.85s setup    test_my_django_view2
1609479014.70s setup    test_my_django_view3
1609479014.60s setup    test_my_django_view4
1609479014.47s setup    test_my_django_view5

Versions (Python 3.10.2):

pytest                             7.1.2
pytest-django                      4.2.0
freezegun                          1.2.2

freezegun is latest version which has some work done regarding this: https://github.com/spulec/freezegun/pull/460

If every test function is decorated individually - then everything works as expected:

from freezegun import freeze_time

@freeze_time("2020-01-01T12:00:00Z")
def test_my_django_view(admin_client):
    resp = admin_client.get(...)
    assert resp.status_code == 200
    
...

Report:

3.75s setup    test_my_django_view1
0.49s call     test_my_django_view2
0.28s call     test_my_django_view3
0.27s call     test_my_django_view4
0.25s call     test_my_django_view5

Related issue: https://github.com/pytest-dev/pytest/issues/7764

Like in the linked issue - if you say that it's my problem of (mis)using freezegun (or that I need to speak with freezegun devs) - it's fine

Reproduce the report with an autouse freezegun fixture and pytest --durations=5, comparing it with a test-level decorator. Start by tracing pytest's duration reporting and setup timing; done means the command no longer reports frozen epoch timestamps as test durations, while real slow tests remain visible.