Hotfix: Run failing macos-latest tests on macos-13

[sloede]

Based on reports in #828

[sloede]

The two system/Homebrew tests on macos-latest fail as well:

• test-system-brew (mpich, 1, macos-latest)
• test-system-brew (openmpi, 1, macos-latest)

Switching them to macos-13 or macos-12 seems like a suitable workaround.

Running some more tests to see if test-mpitrampoline-jll (macos-latest, mpitrampoline, 1.6, x64) can be fixed by either going to an older macOS image or by providing the correct architecture.

[sloede]

@giordano @simonbyrne This PR provides a hotfix for the failing CI tests. Whether to merge this or not, I'll leave up to you. Unfortunately, I do not have the resources to tackle this issue at the root, so this is only a bandaid.

[giordano]

I think the problem is relatively simple: julia just can't find the MPI library because it isn't anywhere accessible. If I manually set DYLD_FALLBACK_LIBRARY_PATH then I can find libmpi:

$ DYLD_FALLBACK_LIBRARY_PATH=/opt/homebrew/Cellar/mpich/4.2.1/lib julia -q
julia> using Libdl

julia> find_library("libmpi")
"libmpi"

I have no clue of how this magically works in older versions of macOS

[giordano]

Ah, I think I understand now: up to the macos-13 image, the brew commands used to install packages in /usr/local (which is probably a default search path for the dynamic loader), but in macos-14 they're now installed under /opt/homebrew. I don't know whether this is a change in Homebrew, or macOS, or whatever, but the net result is that libmpi isn't anymore in a default search path for the dynamic loader, and so we have to do something #831 to reliably make loading libmpi. An alternative could be to tell Homebrew to install the packages again in /usr/local, but I think setting DYLD_FALLBACK_LIBRARY_PATH is slightly more general: we don't have to change the installation directory.

[sloede]

I don't know whether this is a change in Homebrew, or macOS, or whatever

https://apple.stackexchange.com/a/437622

[giordano]

Thanks for finding the reference! That confirms there isn't anything broken in MPI.jl, it was only an infrastructure change in the environment 🙂