Andrew Nelson

I'm intending to look into it, doing a similar thing for numpy at the moment. Debugging on a local Mac is actually one of the most powerful platforms that one...

For packages that make pyinstaller bundles it's a big win making scipy and numpy smaller. Two lots of openblas that dont have to be bundled, tests that don't have to...

To be honest I hadn't even thought about making `Accelerate` only wheels (`macosx_14_0_x86_64`) for Intel. I guess it makes sense. However: - there are going to be fewer and fewer...

I'm not sure why numpy uses that flag. I think I added it into pyproject.toml because I'd seen the option historically used for building wheels, e.g. https://github.com/numpy/numpy/blob/maintenance/1.22.x/.github/workflows/wheels.yml#L91. I'm not sure...

https://wiki.musl-libc.org/open-issues.html seems to indicate that there can be issues with musl-libc and complex math

Hmm it doesnt seem like it's all that easy to detect musl with a macro - https://wiki.musl-libc.org/faq.html. that would make it difficult to add to this Blocklist.

https://stackoverflow.com/questions/58177815/how-to-actually-detect-musl-libc/70211227#70211227 suggests a way of detecting MUSL

In the past `tol` has taken on various meanings for various optimisers. I'd like to work towards more uniformity. Here are some tols that one could use: - fatol, Absolute...

I'd prefer to use f_target for this PR and deprecate the rest

This is why conda environments are good, you can just remove them if you mess them up. You just have to remember to not install things into the base environment.