TST: Add test with large shape to check_below_min_count

[FactorizeD]

The function in question didn't have any tests in the first place so I thought that I will add them and simply use the problematic tuple for shape argument.

• closes BUG: ValueError: Cannot convert non-finite values (NA or inf) to integer only when DF exceed certain size #35227
• All code checks passed.

[jreback]

pls rebase, this is failing on windows

[jreback]

on 32-bit (some other builds as well i think)

> =================================== FAILURES ===================================
_________ test_check_below_min_count__positive_min_count[1-False-None] _________

mask = None, min_count = 1, expected_result = False

    @pytest.mark.parametrize(
        "mask", [None, np.array([False, False, True]), np.array([True] + 2244366 * [False])]
    )
    @pytest.mark.parametrize("min_count, expected_result", [(1, False), (2812191852, True)])
    def test_check_below_min_count__positive_min_count(mask, min_count, expected_result):
        # shape taken to show that the issue with GH35227 is fixed
        shape = (2244367, 1253)
        result = nanops.check_below_min_count(shape, mask, min_count)
>       assert result == expected_result
E       assert True == False

[FactorizeD]

Right, didn't test for 32-bit, my bad. I will work on it later this week. BTW, I am still new to the pandas testing process, is there any way to check 32-bit builds apart from configuring continuous integration tests as specified in https://pandas.pydata.org/docs/development/contributing_codebase.html#testing-with-continuous-integration?

[FactorizeD]

All should be good now - it took me a while to figure out that np.prod will evaluate the dtype to 32 bits on all Windows machines (even 64 bit ones). In the latest version we have three - 2 running on all platforms and 1 running on 64 bit non-windows ones (it is the one with large shape that was initially causing the issue)

[jreback]

All should be good now - it took me a while to figure out that np.prod will evaluate the dtype to 32 bits on all Windows machines (even 64 bit ones). In the latest version we have three - 2 running on all platforms and 1 running on 64 bit non-windows ones (it is the one with large shape that was initially causing the issue)

hmm yeah you can specify np.prod(..., dtype='float64') but directly yeah on windows this would easily overflow.

thanks

[jreback]

thanks @FactorizeD