sessions: oops, we made up a function #13379
Conversation
Turns out that in commit 6bd36a7 we had a function that is not part of the MPI standard. This showed while working on ABI support - which requires us to pay attention to the truth rather than make stuff up. This commit removes our made up MPI_Session_set_info method. Turns out who ever was doing the fortran bindings knew this wasn't a method in the standard so there's no need to change the fortran bindings. Same thing applies to the man pages. Related to open-mpi#13280 Signed-off-by: Howard Pritchard <[email protected]>
|
@devreal take a look when you have a chance |
|
And I was thinking that MPI finally got reasonable when adding sessions. We have support for setting infos for communicators, files and windows, and support for getting infos for sessions, communicators, files and windows. Why the discrepancy ? |
|
@bosilca I suspect that sessions are considered static once created. Unlike communicators, which can change behavior in between operations (well-defined, in-sync across processes). Do you see a benefit in dynamically setting info keys on sessions? |
|
Changing the info keys on an existing communicators was a bad idea especially taking in account how cheap is to create a new communicator with the updated info keys. But, if the Forum decided that it was worth allowing it for communicators, files and windows, they should have allowed it for sessions as well. Consistency does not seem to be a trend in the standardization committee. |
3 participants
Turns out that in commit 6bd36a7 we had a function that is not part of the MPI standard. This showed while working on ABI support - which requires us to pay attention to the truth rather than make stuff up.
This commit removes our made up MPI_Session_set_info method. Turns out who ever was doing the fortran bindings knew this wasn't a method in the standard so there's no need to change the fortran bindings. Same thing applies to the man pages.
Related to #13280