MPICH-v4.2.3+0 libmpi.so not found

[andrew-saydjari]

The update to MPICH-v4.2.3+0 has triggered a new round of the usual libmpi.so not found problems for HDF5.jl. Is the behavior expected? It is solved when I roll back to MPICH-v4.2.2+0 and do an explicit add of MPICH_jll@v4.2.2.

[simonbyrne]

@giordano how do we fix this?

[simonbyrne]

@andrew-saydjari I can't reproduce it. Can you share a reproducer (along with a Project.toml and Manifest.toml)? What platform are you using?

[andrew-saydjari]

Hi, sorry, I have clearly not had time to follow up on this. I can reproduce it on one cluster but not another. It must have something to do with previous versions of julia/HDF5 installed in different projects on the system. There are no LMOD modules loaded or any environmental variables pointing to other MPIs, which is what makes this issue so frustrating, but I haven't had the time to run down creating multiple projects on a clean cluster to get this conflict.

Feel free to close since I can't provide enough information to reproduce, but it would be great if someone had time to run down problems like this.

The platform answer is easy:

  OS: Linux (x86_64-linux-gnu)
  CPU: 128 × AMD EPYC 7702P 64-Core Processor```