Tests not collected if parametrized with `itertools` objects and imported in multiple files

I'm observing a very odd behavior of tests that are imported from other files getting collected only for one test file when they have parametrized arguments that are generated through itertools.

Example with 3 files:

• common.py:

import pytest
import itertools

class CommonTestSuite:
    @pytest.mark.parametrize(
        "m,n",
        itertools.product(
            [100, 1_000],
            [100, 1_000],
        )
    )
    def test_no_negative_numbers(self, generator, m, n):
        assert generator(m,n).min() >= 0.

• test_poisson.py:

from common import CommonTestSuite
import pytest
import numpy as np

class PoissonGenerator:
    def __init__(self):
        self.rng = np.random.default_rng(seed=123)
    def generate(self, m, n):
        return self.rng.poisson(1, size=(m,n))

class TestPoisson(CommonTestSuite):
    obj = PoissonGenerator()

    @pytest.fixture
    def generator(self):
        return self.obj.generate

• test_gamma.py:

from common import CommonTestSuite
import pytest
import numpy as np

class GammaGenerator:
    def __init__(self):
        self.rng = np.random.default_rng(seed=123)
    def generate(self, m, n):
        return self.rng.standard_gamma(1, size=(m,n))

class TestGamma(CommonTestSuite):
    obj = GammaGenerator()

    @pytest.fixture
    def generator(self):
        return self.obj.generate

If I run pytest . --co, this is what I get:

============================================ test session starts ============================================
platform linux -- Python 3.11.0, pytest-7.4.0, pluggy-1.0.0
rootdir: /home/david/del/pytest_issue
plugins: anyio-3.5.0, mypy-0.10.3
collected 5 items                                                                                           

<Module test_gamma.py>
  <Class TestGamma>
    <Function test_no_negative_numbers[100-100]>
    <Function test_no_negative_numbers[100-1000]>
    <Function test_no_negative_numbers[1000-100]>
    <Function test_no_negative_numbers[1000-1000]>
<Module test_poisson.py>
  <Class TestPoisson>
    <Function test_no_negative_numbers[m0-n0]>

That is: the tests from the common class were only collected for the first file that is executed when running pytest .. Same behavior happens if I add more files importing from common.py.

If I modify the file common.py as follows (no itertools-generated parametrization):

import pytest

class CommonTestSuite:
    @pytest.mark.parametrize(
        "m,n",
        [
            (100, 10),
            (10, 100)
        ]
    )
    def test_no_negative_numbers(self, generator, m, n):
        assert generator(m,n).min() >= 0.

Then the tests get collected as expected:

platform linux -- Python 3.11.0, pytest-7.4.0, pluggy-1.0.0
rootdir: /home/david/del/pytest_issue
plugins: anyio-3.5.0, mypy-0.10.3
collected 4 items                                                                                           

<Module test_gamma.py>
  <Class TestGamma>
    <Function test_no_negative_numbers[100-10]>
    <Function test_no_negative_numbers[10-100]>
<Module test_poisson.py>
  <Class TestPoisson>
    <Function test_no_negative_numbers[100-10]>
    <Function test_no_negative_numbers[10-100]>

If instead of importing the common tests from the same file in both test_gamma.py and test_poisson.py I copy-paste the code with itertools or "import" the file by using file.read() + exec(), then the tests are collected correctly as expected.

Start by reproducing the report with common.py, test_gamma.py, and test_poisson.py, then run pytest . --co to compare collection with the list-based parametrization. Trace how pytest handles the shared itertools parametrization and confirm that both derived test classes collect all four cases without changing the reported test behavior.