Closed
Description
By adding
filterwarnings =
error
to our setup.cfg, we can easily make uncaught warnings turn into errors. What are people's thoughts on this? The only downside I see is that contributor's PR will fail due to unrelated changes to a downstream package, but we can quickly fix those.
It seems like uses of
with warnings.catch_warnings(record=True):
...does not work with however pytest is asserting that warnings haven't gotten through. We need to actually use tm.assert_produces_warning or pytest.warns, so it'll be a bit of work to transition.