API to access and update MPI_Status public fields

[jeffhammond]

Problem

MPI_Status is the only non-opaque handle and users are expected to access and update the three public fields directly.

This is not good for users of languages that are not C/C++ or Fortran. Currently, users are working around this by writing their own get/set functions (e.g. mpi4py's pympistatus.h) or linking libclang to parse header files and generate these functions (e.g. rsmpi).

Proposal

Add MPI_Status_{get,set}_{source,tag,error} functions that allow non-first-party languages to support the field access.

I don't know if the set routines make sense, but we have them for elements and cancelled, which are the "hidden" status fields.

Changes to the Text

Add the following in LIS way and define them in the obvious way.

int MPI_Status_get_source(MPI_Status * status, int * source);
int MPI_Status_set_source(MPI_Status * status, int source);
int MPI_Status_get_tag(MPI_Status * status, int * tag);
int MPI_Status_set_tag(MPI_Status * status, int tag);
int MPI_Status_get_error(MPI_Status * status, int * error);
int MPI_Status_set_error(MPI_Status * status, int error);

Impact on Implementations

These are trivial to implement.

Impact on Users

Users of C/C++ and Fortran are not required to use these.

Implementers of MPI support for Python, Julia, Rust, etc. will find it easier to use MPI.

References and Pull Requests

[simonbyrne]

This would be very helpful for those of us maintaining 3rd-party language bindings, and it makes it consistent with other accessors like MPI_Get_count.

[Wee-Free-Scot]

Latest PDF, see section 13.3 page 608-610:

mpi41-report_Issue645_API_access_for_status_public_fields.pdf

[Wee-Free-Scot]

Should these new procedures be referenced within the text of section 3.2.5?
That is the first (lexically) mention of status and of how obtain the values of the fields.

[wesbland]

This passed a no-no vote on 2023-02-01.

Yes	No	Abstain
31	0	0