Add warning about generators for parametrizing class or module

[andrewwillowen]

I stumbled across this issue when trying to write some tests and took a while to figure out the issue.
Based on #13409, it sounds like this is expected behavior.

This PR adds a warning to https://docs.pytest.org/en/stable/how-to/parametrize.html so that hopefully other people don't waste cycles debugging this behavior.

[bluetech]

Hi @andrewwillowen,

Your case seems different from @steadyfirmness's case which was running pytest.main() twice (a rarer case than yours, but still should be perfectly supported). Can you give an example demonstrating your case, just so we're clear about it?

[andrewwillowen]

Taking the example from the documentation.
Modify the list to use a generator instead.
The second test will be skipped because the generator has been exhausted.

test.py

import pytest

@pytest.mark.parametrize("n,expected", (i for i in [(1, 2), (3, 4)]))
class TestClass:
    def test_simple_case(self, n, expected):
        assert n + 1 == expected

    def test_weird_simple_case(self, n, expected):
        assert (n * 1) + 1 == expected

Then run

pytest -rs ./test.py

======================= test session starts ========================
platform linux -- Python 3.13.1, pytest-8.4.2, pluggy-1.6.0
rootdir: xxxxx
configfile: pyproject.toml
plugins: cov-7.0.0, hypothesis-6.138.15
collected 3 items

test.py ..s                                                  [100%]

===================== short test summary info ======================
SKIPPED [1] test.py: got empty parameter set for (n, expected)
=================== 2 passed, 1 skipped in 0.24s ===================

[bluetech]

Thanks. In order to keep the discussion in one place, I'll comment further in the issue #13409.

[bluetech]

We ended up deprecating this case, see #13877.