Attempt at patching phist find mpi heuristic for updated phist version#32688
Conversation
for updated phist version
|
@jthies can you review this PR? This PR modifies the following package(s), for which you are listed as a maintainer:
|
| # MPI installation to be used than the one spack wants. | ||
| def patch(self): | ||
| if self.spec.satisfies("@1.9.6"): | ||
| if self.spec.satisfies("@1.9.6:"): |
There was a problem hiding this comment.
This patch undoes the switch from the unsafe 'mpi' to the better 'mpi_f08' module which is unfortunate. The actual fix should be to make it possible to query wether an mpi spec provides mpi_f08, and patch phist only if it doesn't.
Still I think this is OK for now.
There was a problem hiding this comment.
This PR is intended as a stopgap to allow the CI to continue functioning without error, but thank you for pointing this out. I'll leave a TODO to indicate that this should not languish in its current state.
adamjstewart
left a comment
adamjstewart
left a comment
There was a problem hiding this comment.
Hit this same problem in CI. Let's merge for now and think about a better long term solution at our leisure.
3 participants
This is a tentative fix for the current Gitlab CI issues plaguing develop where phist fails due to GCC complaining about a type mismatch in some calls to MPI.
This solve is based on a previous issue with phist, where its bundled findMPI.cmake is conflicting with Spack derived MPI location determination.