TST: Handle PY_SSIZE_T_CLEAN warnings - #11019

Remote data failure is unrelated.

Looks straightforward and the warning seems correctly handled in CI now (mpl302 is still running in Python 3.7, so cannot directly test the numpy_warnings case, but I noted there are further PY_SSIZE_T_CLEAN checks in test_slicing_warnings).

One thing I have been wondering about before, why not just adding
"PY_SSIZE_T_CLEAN will be required for '#' formats" in w_msg
to the anticipated warnings – is this because the test is supposed to probe only for immediate warnings from numpy or mpl?

Re: #11019 (comment)

Not sure. I didn't write the test. Will have to ask @astrofrog or @Cadair .

Re: #11019 (comment)

Not sure. I didn't write the test. Will have to ask @astrofrog or @Cadair .

Looks like originally this was intended to make sure no warnings from numpy were issued, but as of #10504 some others (matplotlib-related?) needed to be intercepted, + the DeprecationWarning as of numpy 1.16. Then it might be reasonable to handle the Python 3.8 warning there as well.

So, you are saying we should not skip at all and use pytest.warns to catch them all?

Yes, since we clearly anticipate that warning, I'd say it's better to run the test and catch it (it is not skipped for Numpy >= 1.16 either...). Only possible drawback I can see is too many potential warnings accumulating.

RTD reached concurrency limit but we don't really need it (but we cannot cancel). You can ignore the "allowed failure" job. If this looks fine to you, please approve and merge. Thanks!