BLD: make macOS version check for Accelerate NEWLAPACK more robust - #25040

Thanks Ralf.

This doesn't want to backport. Do we need #25008 as well?

Ah fun with git submodules. We don't strictly need gh-25008 and it's a large PR, that's why I didn't mark it for a backport. However, we either need to backport that or we need to create a new branch specific to 1.26.x in our fork of meson. Either option works here, but I'm thinking that it's helpful to minimize differences between main and 1.26.x given that we may need to backport more changes (there's at least one more BLAS-related fix coming) and given that 1.26.x will likely be a fairly long-lived branch.

So let's backport both? If you want me to deal with it, just say the word.

The problems with #25008 are likely to be because of the because of the core -> _core rename, which only affects three files, so probably not too bad. If there were a git checkout ... as <new_name> it would be easy :) Let me give it a shot tomorrow.

@rgommers The way to deal with the renames is to download the PR as a patch and edit the files before applying. Unfortunately, you have been happily removing distutils related stuff :) The easy way to fix that is to give up on setup for 1.26, which seems reasonable to me. Or we could just go with a 1.27 ...

I think the moral of that story is "Don't try to support two build systems". We have been there before.

Yes, I think we can drop setup.py based builds in 1.26.x, I don't see a downside to that. Everyone who still needs them can use 1.25.2 or 1.26.1

For completeness: this fix did get backported to the maintenance/1.26.x branch, but after 1.26.4 was released. If you need this fix, build from the unreleased v1.26.5 tag or from the head of the maintenance/1.26.x branch.