CI test with system MPI and local t8code installation

[benegee]

Motivated by #139 and by P4est.jl's tests, this PR adds tests using system's MPI libraries and locally compiled t8code libraries.

[JoshuaLampert]

Can you explain why there are problems here with the system MPI tests in CI that we do not have in P4est.jl? Your comment above about the version mismatch would also apply to P4est_jll.jl and P4est.jl, wouldn't it (see, e.g. here)? Or what am I missing?

[benegee]

Can you explain why there are problems here with the system MPI tests in CI that we do not have in P4est.jl?

I think this is the reason:
https://github.com/JuliaPackaging/Yggdrasil/blob/c47cf70b997286ebdf74548e31284085514d1865/P/P4est/build_tarballs.jl#L84-L93

Although OpenMPI_jll is listed as a dependency, it actually isn't.

[JoshuaLampert]

Ah yes, maybe. These issues are too long ago for my brain to remember the details, but yes, there was something. A similar workaround in the build script for t8code_jll.jl in Yggdrasil as the one you linked above for P4est_jll.jl does not work?

[benegee]

A similar workaround in the build script for t8code_jll.jl in Yggdrasil as the one you linked above for P4est_jll.jl does not work?

It would probably. It's just that when I started revising the Yggdrasil receipt I was eager to enable as much as possible.

[benegee]

I'm happy with the current state. Maybe you want to have a look, @JoshuaLampert @spenke91 .

[JoshuaLampert]

Thanks a lot! The whole setup seems a bit fragile, but if it the only way we see at the moment to run tests with a system MPI, I would be fine with it. Did you already let an LLM review this?

[benegee]

The whole setup seems a bit fragile

True!
Thanks for your approval, anyway!

Did you already let an LLM review this?

I just did, but it just raised two false positives.
In fact, the t8code part of the workflow was copied from libtrixi. And Claude helped with pinpointing the dlopen issue.