Clarify return value for team split on SHMEM_TEAM_INVALID #461
Conversation
Specifically, team split operations should return nonzero if the parent_team argument is SHMEM_TEAM_INVALID. This is likely not a semantic change, just a clarification. Signed-off-by: David M. Ozog <[email protected]>
|
@davidozog Thanks for keeping track of this ticket. This feels more significant than a doc edit to me. Given that we have the time, I would prefer to read this and give the committee a chance to weigh in on whether the case that's being clarified is one where the function should return nonzero. |
|
@jdinan - sure. Maybe I can do an unofficial reading at the upcoming meeting to see if it generates any concerns. It's too late to announce an official reading this time... |
|
Note - this should go into errata (and/or pending errata?) Also need a changelog entry. |
|
@jdinan - I was about to cherry-pick 503317b from @kholland-intel's PR, which adds a v1.6 change-log that mentions |
This reverts commit ec66ff2.
|
@davidozog Please go ahead and add a changelog entry to the effect of "Clarified the return value of team split operations when the parent team is SHMEM_TEAM_INVALID". In the future, I'd prefer to include the changelog entry with the reading since it will help with consistency. Please don't cherry pick from another PR. I'll deal with any merge conflicts as they come up. |
|
@davidozog The changelog heading in question was merged into master. Please merge from master and add your changelog entry. |
Signed-off-by: James Dinan <[email protected]>
|
@davidozog Could you please review the changelog entry in 75559f8? |
|
@jdinan - I believe this PR is ready to merge. Is there anything pending on your end? (This should close Sandia-OpenSHMEM/SOS#969) |
This PR is meant to address #450 (as well as Sandia-OpenSHMEM/SOS#969 and Sandia-OpenSHMEM/SOS#972).
I don't believe 8b44a29 is a semantic change - just a clarification that a split operation on a SHMEM_TEAM_INVALID parent is indeed considered "unsuccessful" and the routine should return a nonzero value.
I also included a small change to the 2D team split example code (ec66ff2). This has no effect on the example, but leaving these variables uninitialized commonly causes a compiler warning. Please let me know if you prefer this in a separate PR...