typically we just disable flake8 testing because its of no use to us.

Makes sense - I've re-tested the update on darwin and nixos. Thanks for the super-quick comments!

Upstream will be skipping some tests as well MagicStack/uvloop#318 (comment)

Thanks @jonringer - looks like the upstream HEAD now has a fix for the test_sock_cancel_add_reader_race test, in response to your feedback. It would definitely be nice to drop the workaround in nixpkgs.

The issue here, though, is with a different failing test (which also has an unreleased upstream fix). So the obvious options seem to be:

• drop both workarounds, pull in the latest dev commit, and trust that it's stable enough
• add a temporary workaround for this second issue (the current PR contents)

The first option doesn't seem totally outrageous - the dev branch was named 0.15.0.dev0 in October, and subsequent commits look like stability fixes. But I have no idea what behaviour might have changed since the 0.14.0 release that other packages think they're depending on.

Which option would you prefer? I'm happy to update the PR if it's the first one (or a totally different solution).

Any comment @FRidh or @jonringer about how you’d like to proceed with this? Currently this just disables flake8 testing, but happy to change if preferable.

Looks like this was fixed in #95591 now, thanks for the PR though :)