CI update

[boriskaus]

This is an attempt to get CI to pass with the current version

[boriskaus]

I'm not sure why this is failing; it works fine on my local machine (macOS aarch64)

[boriskaus]

@ViralBShah All MPI tests fail in the CI, but work fine locally. A simple "hello world" MPI test does work, so it doesn't seem to be because of MPI itself.
As far as I can tell, the correct MPICH_jll is being employed.
Any idea?

[ViralBShah]

I have been noticing the same thing (passes locally on my mac but fails in CI). One thing is that the build complains it is built with OpenMPI but we are using MPICH. The other thing is that we are unable to use the latest MPICH.

Perhaps @eschnett can help us with MPI here?

[boriskaus]

The problem with MPICH was solved last week thanks to @vchuravy and @giordano and only occured if the PATH environmental variable was empty. As far as I can tell, we here use the working version of MPICH.

Where do you see that PETSc was build with OpenMPI? That would indeed be a problem - it should be using the MPICH build instead.

[boriskaus]

Problem solved. Turns out to be that one should add

using PETSc

at the beginning of runtests.jl...

[boriskaus]

I am deactivating CI on windows because the PETSc_jll binaries for windows were compiled without MPI support (it did compile with MicrosoftMPI, but we always get runtime errors). Perhaps someone motivated can solve this again at some stage (it did work at some point in the past), and we can allow windows again.

Until that stage, windows users can simply install WSL on their machine and run PETSc.jl though windows. This will also give access to additional external packages.

[ViralBShah]

Where do you see that PETSc was build with OpenMPI? That would indeed be a problem - it should be using the MPICH build instead.

I suspect I saw that in some CI logs, which would be very hard to find. If I see it again, I'll open a new issue.

[boriskaus]

I suspect I saw that in some CI logs, which would be very hard to find. If I see it again, I'll open a new issue.

ok. It appears to be solved meanwhile, so I'll merge this PR.