Howard Pritchard

hmm..looks like the python test is tripping up on the comm compare. investigating that.

something else seems to be going on with this self with multiple ranks test. If each rank uses a unique tag argument in the call to MPI_Comm_create_from_group then I don't...

okay it turns out that this python test is hitting an edge case concerning the use of the tag argument to MPI_Comm_create_from_group and the only difference being the list of...

@dalcinl could you open a new issue to track this specific problem - supplying a non-unique tag to concurrent calls to MPI_Comm_create_from_group with non-overlapping group arguments?

@gpaulsen please review this Pr. there are now numerous threads going on here that don't have anything to do with what this Pr is suppose to address.

I doubt it has to do with ordering of the proc array. The way dalcini's test case works - using the group from MPI_COMM_SELF to create a communicator congruent to...

I looked in to this more and I think the current error code being returned is more accurate owing to this verbiage about ```MPI_Group_from_session_pset``` in the MPI 4 standard: ```...

closed via #10744 and #10784

Is there a mechanism for alerting spack users about this default change in behavior? I would expect users of spack-based installs of newer Open MPI versions and using UCX PML...

You may want to take a look at the workarounds discussed at https://github.com/open-mpi/ompi/issues/8531