Validate PR title: I really think that tests is a better commit type than test, because to me the latter always feels like "this is just a test, not a for real". But here we are ...

Unit tests: looks like catch_warnings fails on older pythons, I guess the __init__ form changed? This makes me think, we should update CI to not have unit tests cancel each other -- it's possible for them to pass on some python versions and fail on others, and it's good to know which ones -- and they're cheap and fast to run so cancelling doesn't really save much anyway.

Validate PR title: I really think that tests is a better commit type than test, because to me the latter always feels like "this is just a test, not a for real". But here we are ...

I feel like I get this wrong about 50% of the time 😆.

Unit tests: looks like catch_warnings fails on older pythons, I guess the __init__ form changed?

Yeah, a bunch of parameters were added in 3.11, and I'd forgotten that. Maybe I ought to default to running tox with 3.8 locally - it's much more likely that it would catch things than using newer Python.

This makes me think, we should update CI to not have unit tests cancel each other -- it's possible for them to pass on some python versions and fail on others, and it's good to know which ones -- and they're cheap and fast to run so cancelling doesn't really save much anyway.

Interesting thought - feel free to add it to the team discussion topics. I do agree that at times (like this one) it's more informative to immediately see "ah, this is bad on [3.8 or macOS or ..]". At other times, when something is just generally wrong (hopefully only drafts, if one is properly running tox locally) it does feel wasteful to do all the matrix when it's going to fail every time.