compilers: prioritise LLVM clang over GCC on macOS

[cho-m]

• 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 typecheck with your changes locally?
• Have you successfully run brew tests with your changes locally?

---

Prioritise LLVM clang over GCC on macOS. This has a higher chance of succeeding when there is no manual override of cc/cxx as many formulae are linked to libc++.

However, when the formula has a GCC dependency but no LLVM, then try GCC first. This usually means that we want GCC/libstdc++

---

Ideally, formulae should work by adding fails_with :clang and then depends_on "gcc" or depends_on "llvm" => :build but right now we need to manually force ENV.llvm_clang (often from gcc dep for gfortran) breaking compiler selection.

LLVM Clang is usually our 2nd preference on macOS due to libc++. And 3rd is GCC (since using GCC often causes linkage to libstdc++ or libgomp which can cause issues in dependency trees)

---

No changes done on Linux side for now.

[cho-m]

I haven't been able to figure out where this test failure is happening.

  1) Homebrew::Cmd::Migrate migrates a renamed Formula
     Failure/Error:
       expect { brew "migrate", "testball1" }
         .to output(/Migrating formula testball1 to testball2/).to_stdout
         .and not_to_output.to_stderr
         .and be_a_success

       expected block to output /Migrating formula testball1 to testball2/ to stdout, but output nothing
       Diff for (output /Migrating formula testball1 to testball2/ to stdout):
       @@ -1 +1 @@
       -/Migrating formula testball1 to testball2/
       +""
     # ./test/cmd/migrate_spec.rb:15:in 'block (2 levels) in <top (required)>'
     # ./test/support/helper/spec/shared_context/integration_test.rb:50:in 'block (2 levels) in <top (required)>'

Not able to reproduce locally but is hit on re-run for CI.

[MikeMcQuaid]

As long as we're confident the our LLVM clang compiler is (more) stable for formulae on macOS than GCC: seems good, thanks 👍🏻

[cho-m]

As long as we're confident the our LLVM clang compiler is (more) stable for formulae on macOS than GCC: seems good, thanks 👍🏻

On arm64 macOS, I would consider it more stable since GCC still needs a large patch to build.

On Intel macOS, harder to say. But we at least use LLVM Clang more than GCC in Homebrew/core (excluding Fortran parts) so it has at least been okay for the last few years.

There are still quirks to using LLVM Clang outside of superenv (due to the libc++ situation) but currently should be good for building formulae.

[MikeMcQuaid]

Ok, sounds good, thanks @cho-m!