Any comments on this? I'd love to get this merged.

Wow, quite a bit of work went into the tests refactor. I can see why you went there - wanting to extract the commonality of all the tests.

I'm pretty sure pytest_generate_tests has been superseded by parametrized fixtures, but that can be transformed later if needed.

The main concern I have about the refactoring is what does it do for the output? Is it still possible, for example, to select just one test to run? Is it still clear which test is being performed?

I'm also slightly concerned about the xfail syntax, which I presume was an unexpected introduction late. It's concerning in part because it's part of its own language, so it's not possible, for example to provide more detail (i.e. @pytest.mark.xfail(reason="..."))... without expanding the DSL.

I do like that the code seems to be more consolidated with this change. It does still feel a little uneasy that _finalize_requires gets called twice, but if that's what needs to happen, that's probably better than what was being done before.

If you're comfortable with the change being stable and addressed any concerns above to your satisfaction, would you add a CHANGELOG entry and then cut a release (release docs are in the repo)?

I'm pretty sure pytest_generate_tests has been superseded by parametrized fixtures, but that can be transformed later if needed.

This seemed to be the simplest way to parametrize those tests while still keeping the code in-class. But I agree it could become unwieldily if multiple methods are parametrized using the same mechanism. Would this be better?

The main concern I have about the refactoring is what does it do for the output? Is it still possible, for example, to select just one test to run? Is it still clear which test is being performed?

Yes, that where the ids parameter come into play:

> python -m pytest --collect-only -qk test_requires
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_marker]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_marker_in_setup_cfg]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra_in_setup_cfg]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra_and_marker]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra_and_marker_in_setup_cfg]

=========================== short test summary info ============================
SKIP [2] /home/bpierre/progs/src/setuptools/setuptools/tests/test_msvc.py:17: could not import 'distutils.msvc9compiler'
============================= 293 tests deselected =============================
2 skipped, 293 deselected in 0.65 seconds

> python -m pytest --collect-only -qk install_requires
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_marker]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_marker_in_setup_cfg]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra_in_setup_cfg]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra_and_marker]
setuptools/tests/test_egg_info.py::TestEggInfo::test_requires[install_requires_with_extra_and_marker_in_setup_cfg]

=========================== short test summary info ============================
SKIP [2] /home/bpierre/progs/src/setuptools/setuptools/tests/test_msvc.py:17: could not import 'distutils.msvc9compiler'
============================= 293 tests deselected =============================
2 skipped, 293 deselected in 0.64 seconds

I'm also slightly concerned about the xfail syntax, which I presume was an unexpected introduction late. It's concerning in part because it's part of its own language, so it's not possible, for example to provide more detail (i.e. @pytest.mark.xfail(reason="..."))... without expanding the DSL.

Yeah, I could not find a way to mark a test by name.

OK, I went ahead with the switch to pytest.mark.parametrize as it does make the code a little bit simpler.

Just tested this fix in setuptools 36.2.7 and it seems to work: I can now generate a wheel for Prequ with correct environment markers (with out-of-the-box setuptools and wheel packages). Thanks a lot! 😄