MPI_Psend_init_c & MPI_Precv_init_c

[cniethammer]

Problem

The bindings of MPI_Psend_init and MPI_Precv_init do not follow the naming scheme of procedures taking MPI_Count arguments / misses a regular int version.

Note: check examples for correct usage and argument declarations!

Proposal

As discussed in the Forum meeting 14/09/2023: Make it consistent and have an integer and a embiggened (_c) version.

Note: Check other places in the standard for more issues of this kind.

Changes to the Text

Rename the current MPI_P(send|recv)_init to MPI_P(send|recv)_c and add another binding for MPI_P(send|recv)_init that uses an int count|INTEGER count

Impact on Implementations

References and Pull Requests

https://github.com/mpi-forum/mpi-standard/pull/888

[cniethammer]

Just realized that the Fortran binding used INTEGER(KIND=MPI_COUNT_KIND). Should this be changed back to pure INTEGER and reflect the bindings of regular send operations?

@martinruefenacht Is there a way to generate regular+'_c' bingings for C and F08 but MPI_COUNT_KIND ones for Fortran?

[jprotze]

The function has this signature since MPI-4.0. The proposed change breaks the API.
Why didn't 4.0 replace all int count by MPI_Count count in the first place?

[jeffhammond]

This breaks API for approximately zero users. The reason we didn't change ALL the APIs for large count is that would have broken every MPI code in existence.

[jeffhammond]

Breaking backwards compatibility for non-existent users is a very bad reason for not fixing an obvious mistake in the standard. This issue describes an inconsistency relative to all other features that has multiple negative effects on users. Fixing this will ensure once this feature has users, they are satisfied.

[cniethammer]

Note: this PR will change the old Fortran module interface from INTEGER(KIND=MPI_COUNT_KIND) to INTEGER without a replacement - however there are no other old Fortran module interfaces and users of MPI_Count should anyway use mpi_f08, already.

[raffenet]

What are the negative effects on users with the current interface?