@bsipocz I pushed a small fix for a merge conflict. But, could you remind me of the correct incantation (currently?), I already added:

 norecursedirs =
+    # Skip potential temporary environments:
+    build
+    benchmarks/env

to the pytest.ini (both can contain a conftest.py which fixed the below issue for it), but I keep getting import errors like:

_pytest.pathlib.ImportPathMismatchError: ('random', '/opt/homebrew/Caskroom/mambaforge/base/lib/python3.10/random.py', PosixPath('/Users/sebastianb/forks/numpy/numpy/typing/tests/data/pass/random.py'))

so I think I am probably just approaching running pytest wrong.

EDIT: I think the problem was only that I didn't add --continue-on-collection-errors

There are now quite a few plots popping up again and failures due to deprecations. Maybe we should just ignore deprecation warnings in the docs. Or if we don't want to do that, just remove the examples for clearly deprecated functionality?

remove the examples for clearly deprecated functionality

+1

Would a rebase help clear some of the collection issues?

I think the collection issues are maybe all expected, i.e. trying to find doctests in places we shouldn't look or just something to do with how I ran it.

Thanks Sebastian for the commits! I'm sorry for being too slow with this, I'm trying to come back to it in a more timely manner.

I'll note that the tool we currently use is tools/refguide_check.py. It is run in CircleCI only. It can also be run via runtests.py and should still be plugged into spin test.

If you want to iterate on CI, perhaps easiest to use that same job (or port it to a single GHA job).

We now have spin check-docs. @bsipocz, @seberg should we close this?

Yes, I think so. I think it's a pitty that we couldn't merge the approaches instead of having one tool doing the same another, less generically used was built up, but whatever.