HI All,

It looks like we need to discuss an issue open by Rolf last week:

https://github.com/mpi-forum/mpi-issues/issues/435

He does have a point that as the Sessions Model related text currently stands, there is no explicit prohibition of applications using Sessions asymmetrically.

By asymmetric I mean that different MPI processes derive MPI communicators, files, or windows from sessions  handles differently.   The current text implies that this is not allowed, or at least leads to logical inconsistencies, owing to the description of the collective behavior of Session finalize in the event that there are communicators, windows, or files still associated with a session handle when Session finalize is invoked.

I propose that for 4.1, we add an additional paragraph to 11.3 that states something to the effect that this asymmetry is not allowed, or at the minimum that if an application does so, the behavior of Session finalize may be undefined.

Other ideas on how to address 435 would be welcome!

Thanks,

Howard

Howard Pritchard Research Scientist HPC-ENV Los Alamos National Laboratory lanl.gov