Hmm, the Address Sanitizer build failure seems related in some way. I think we would need to figure that out before merging. I reran it and it failed, printing out AddressSanitizer:DEADLYSIGNAL repeatedly. The same build on main works.

Feels like it, if only because the sanitiser job installs setuptools before running the build. I think that might be because older versions of this script were doing python setup.py build to install (where setup.py imports from setuptools). It shouldn't have that same bootstrapping problem with a pyproject.toml-based config.

I'm...going to try removing this just on a hunch. I'll drop the commit if this hunch fails, squash if it works.

Nope, that wasn't it. Actually, it succeeded, but hung forever on exactly the same thing. That exhausts my thoughts for why this might be the case.

Well that's interesting. It succeeded on the most recent run.

Huh, that is weird. Maybe it is flaky. I don't know that it has a history of being flaky, but it is possible something recently changed in the image to make it so.

Given the above, I am going to go ahead and merge this. Thank you for taking a closer look.

Thank you for the merge. I was reluctant to suggest merging given that it only started happening with me - but it does rather seem like we just had some bad luck with one particular RAM stick in the world or something.

Yeah, I am not sure. I don't really see how the issue could be related to your change; I would have expected an explicit build failure if it was that. It did pass again in main.

I released 2.6.0 with this change.

That was fast. Thank you very much!