Consistency: replace "task" with "rank" or "MPI process" where appropriate

[devreal]

Problem

The standard document spuriously uses the term task instead of rank or MPI process. The term is never defined as such and quite overloaded these days.

Proposal

For the changes for MPI 4.0 please see the PR: https://github.com/mpi-forum/mpi-standard/pull/258

Impact on Implementations

None. (I am aware that some vendors may use the term task to refer to processes. I think it's a terrible choice and not the official MPI speak.)

Impact on Users

Being more consistent avoids some distraction.

[wesbland]

Is this replacing the job that @dholmes-epcc-ed-ac-uk started years ago?

[dholmes-epcc-ed-ac-uk]

This is a small piece of the effort by the Hybrid WG (of which I was/am a member). We clocked 2000+ instances of this type of change in the first 600 pages and then ground to a halt. Piecemeal seems like a better approach.

[cblaas]

I would like to see the 3 substitutions of "task" by "MPI process" in PR 258 already in MPI-4.0.

Piecemeal indeed seems to be a good approach here as these are the only three occurrences of "task" in a wrong context.

[wgropp]

A quick check shows 21 lines that contain the word "task". Historically, as noted above, some vendors used "task" where others were using "process". And the standard was a bit sloppy about saying MPI Process (the only kind of process that the standard discusses). And while multi-threaded processes were considered from the beginning (its why the global state approach in PVM was rejected), they weren't always at the top-of-mind. If we'd caught this in MPI 1.0, we would have replaced task with process everywhere.

[wesbland]

Closing this issue in favor of the chapter-specific issues created on 2021-03-04.