Fix pytest.parametrize when argnames are specified as kwarg

[leezu]

Previously only this worked:
@pytest.mark.parametrize("A", argvalues=[(A1, ), (A2, )])
Now argnames can be specified explicitly:
@pytest.mark.parametrize(argnames="A", argvalues=[(A1, ), (A2, )])

[RonnyPfannschmidt]

please also include a test for the behaviour

[RonnyPfannschmidt]

this line directly breaks testing pytest in various cases, see the broken ci results

the defensive coding in the surrounding lines may look ludicrous, but python doesn't have monadic handling of defensive coding so we have to bite the bullet or include null types

[coveralls]

Coverage decreased (-0.007%) to 92.224% when pulling e89abe6 on leezu:fix_kwargs_fixtures into 761d552 on pytest-dev:master.

[leezu]

Thanks @RonnyPfannschmidt . Please have a look at the updated version. I also added a test case. Turns out the bug is only triggered when parameterizing a test that relies on a parameterized fixture..

        import pytest

        @pytest.fixture(params=[1,2])
        def a(request):
            return request.param

        @pytest.mark.parametrize(argnames='b', argvalues=[1, 2])
        def test_func(a, b):
            pass

[coveralls]

Coverage increased (+0.003%) to 92.235% when pulling e86ba41 on leezu:fix_kwargs_fixtures into 761d552 on pytest-dev:master.

[coveralls]

Coverage increased (+0.003%) to 92.235% when pulling e86ba41 on leezu:fix_kwargs_fixtures into 761d552 on pytest-dev:master.