The text for MPI_WAITALL (Section 3.7.5, page 86) is the only place where MPI_ERR_PENDING is mentioned in the text itself (and serves as main index entry).
The text says on requests in general:
Active persistent requests are marked inactive. Requests of any other type are deallocated
and the corresponding handles in the array are set to MPI_REQUEST_NULL.
This paragraph introduces MPI_ERR_PENDING:
When one or more of the communication operations completed by a call to
MPI_WAITALL fail, it is desirable to return specific information on each communication. The
function MPI_WAITALL will return in such case the error code MPI_ERR_IN_STATUS and will
set the error field of each status to a specific error code. This code will be MPI_SUCCESS, if
the specific communication completed; it will be another specific error code, if it failed; or it
can be MPI_ERR_PENDING if it has neither failed nor completed.
The text makes no mention of what happens to these pending requests and there is no other place (I can find) throughout the document that explains it. Intuitively, pending requests should be returned to the caller who then decides what to do with it but that needs to be stated explicitly.
Proposal
The text should be amended to describe the semantics of MPI_ERR_PENDING and what applications should expect happens if they see the error.
Changes to the Text
Add language like the following:
When one or more of the communication operations completed by a call to
MPI_WAITALL fail, it is desirable to return specific information on each communication. The
function MPI_WAITALL will return in such case the error code MPI_ERR_IN_STATUS and will
set the error field of each status to a specific error code. This code will be MPI_SUCCESS, if
the specific communication completed; it will be another specific error code, if it failed; or it
can be MPI_ERR_PENDING if it has neither failed nor completed. In the latter case, the corresponding
request handle will not be reset to MPI_REQUEST_NULL.
Impact on Implementations
Check handling of MPI_ERR_PENDING.
Impact on Users
Clarity on how pending requests are to be handled if an error occurs.
Problem
The text for
MPI_WAITALL(Section 3.7.5, page 86) is the only place whereMPI_ERR_PENDINGis mentioned in the text itself (and serves as main index entry).The text says on requests in general:
This paragraph introduces
MPI_ERR_PENDING:The text makes no mention of what happens to these pending requests and there is no other place (I can find) throughout the document that explains it. Intuitively, pending requests should be returned to the caller who then decides what to do with it but that needs to be stated explicitly.
Proposal
The text should be amended to describe the semantics of
MPI_ERR_PENDINGand what applications should expect happens if they see the error.Changes to the Text
Add language like the following:
Impact on Implementations
Check handling of
MPI_ERR_PENDING.Impact on Users
Clarity on how pending requests are to be handled if an error occurs.
References and Pull Requests
TBD