audit: check Accelerate linkage for core formulae

[jonchang]

• Have you followed the guidelines in our Contributing document?
• Have you checked to ensure there aren't other open Pull Requests for the same change?
• Have you added an explanation of what your changes do and why you'd like us to include them?
• Have you written new tests for your changes? Here's an example.
• Have you successfully run brew style with your changes locally?
• Have you successfully run brew tests with your changes locally?

---

In concert with Homebrew/homebrew-core#39809, Homebrew/homebrew-core#39226, and Homebrew/homebrew-core#37898 (comment); this new audit will check core tap formulae for undesired linkage against system Accelerate.

No new tests written since none exist for this file.

[MikeMcQuaid]

Makes sense, thanks @jonchang!

[zbeekman]

Nice! N.B., however, that veclibfort.rb, by design, will always link to Accelerate and should be an exception to this rule.

Should I change this to make an exception for veclibfort? Or should we end up removing veclibfort? (I do not favor removal, personally.)

[MikeMcQuaid]

Should I change this to make an exception for veclibfort?

Yes, please

[jonchang]

Good point. I think adding an exception is probably the best idea. Want to file a follow up? I can do it later this week as well.

[zbeekman]

@MikeMcQuaid I think this is actually a bit too strict. There are cases in which users may want to link accelerate, as it provides more than just linear algebra.

https://developer.apple.com/documentation/accelerate

I think instead we should check for linkage against veclibfort and lapack. (Veclibfort is specifically for BLAS/LAPACK provided through Accelerate, and lapack reference implementation is old and minimally maintained and outdated.)

I can open a new PR with this. I hit an instance in core where a package links to Accelerate, seemingly for reasons other than linear algebra: Homebrew/homebrew-core#39418

[zbeekman]

I'm tempted to add -DBLA_VENDOR=OpenBLAS to the std_cmake_flags which will prevent CMake formula using the included standard FindBLAS and FindLAPACK intrinsic modules from picking up Accelerate. The only issue I foresee arising may be if they want, e.g., scalapack over OpenBLAS.

[zbeekman]

In fact, both tests could apply on linux too. 1) veclibfort should not be ported to linux (hopefully this is already true) and 2) lapack is outdated/crusty on linux also.