Yeah I should do that.

A user needed a wheel for 3.12 on linux today and it successfully compiled from your 522-wheels-for-python-312-are-missing branch with especially the linker fix being helpful. Is it the mac support which is blocking? If so I'll just drop that a partial release of the linux wheels would definitely be of interest to some users. Thanks for the great tool btw!

I was actually having issues linking locally, and then got distracted. I'll try to get back to it.

I think I figured out the linking issues.

Any updates?

Been sick, but, might try tomorrow.

This also applies to Python 3.13 now

Yeah... I have been distracted by paid work, being sick for a month, vacation, political organizing, etc.. The main bottleneck is that the output is wrong on Python 3.12 (and presumably 3.13) and I can't figure out why. If anyone who understands Python's C API wants to poke at the branch, that'd be helpful, but I will try again.

Some progress in understanding:

• Tests fail on 3.12 because mmap.mmap() used to not release GIL, and now does as of 3.12.
• Put another way, Fil fails to profile correctly in some cases when GIL is not held.
• I was relying on PyTrace_C_CALL op in Python's tracing/profiling API to get accurate line number info, but... turns out that doesn't get used in some cases! For example, for mmap.mmap(), perhaps because it's a type constructor rather than a standalone function. This seems like a bug in CPython to me (filed as sys.setprofile doesn't do c_call in all cases python/cpython#126056) but even if it'll get fixed that is not very helpful anytime soon.