Revisit the use of return code vs. error class in MPI_T

[schulzm]

Problem

When we introduced the MPI_T interface in MPI 3.0, we decided to use return codes for all MPI_T functions instead of error classes, to allow continued operation on error and to enable the usage before Init. With recent changes this may be obsolete.

Proposal

Change occurrences of error code to return code to make clear that these return codes behave differently from the "usual" MPI return codes.

Changes to the Text

See PR#646

Impact on Implementations

Should be none.

Impact on Users

Should be none, except for consistency - possible better usage of error handling.

References

[mahermanns]

@schulzm The introduction of this issue suggests this to be very close to this one: #385 and could be a duplicate.

So the 4.1 part of handling return codes should probably discussed there, and this issue is more about fixing the language/typo? Is that already fixed by some other PR?

[mahermanns]

This might be a duplicate of #385 or should at least be handled in the same way.

[mahermanns]

Tools WG, Dec 2, 2021: We favor to keep the current semantics but maybe make it clearer in the text that MPI_T behaves different than the rest of MPI. The text should rather talk about "MPI_T error codes" to make the difference clear to the user.

[mahermanns]

Let's close this as a duplicate of #542 and continue discussion there.