Actually, looking a bit closer, I see warnings fromexamples/scales/log_test.py, which we definitely don't want to run.

Locally, I seepytest --collect-only | rg examples shows warnings from the examples, but they aren't collected. That's because they have top-level code, but no test functions/classes. So I wonder if we should add apytest.ini by default that tells pytest to ignore those directories.

And then also turn warnings into errors in the nightly build (by mostly reverting this PR), so we catch things from upstream?

After some further investigation, I think I see the reason for the inconsistency. For stuff inlib, we have aconftest.py that configures warnings into errors and ignores a few:

We don't have that forexamples, so that's why the warnings depend on the config inpytest.ini. Thedistutils warning is not ignored above though, and I assume from that that the upstream code has been fixed to not trigger those.

What that means is that I'm going to keep the commit above, dropping thepytest.ini in the nightly test. But I amalso going to add apytest.ini that tells pytest to only look inlib, as we don't want it trying to importexamples,tutorials, etc.