[mpiwg-rma] RMA chapter review: please review fixes; no meeting today

[Joseph Schuchart]

I have opened https://github.com/mpi-forum/mpi-standard/pull/912 to 
address the remaining issues.

I want to point out two pieces where I would welcome some eyes:

1) "As all \MPI/ \RMA/ communication functions are incomplete, they must 
be completed": 
https://github.com/mpi-forum/mpi-standard/pull/912/files#diff-abcde0810b89c0faf160df24683574bc0f017a2e677e3b8dd0d2d73116877a9eR1449
--> I rephrased this to differentiate between procedure and operation. I 
also added "eventually" to indicate that the completion call can happen 
at any time after the operation was started. I'm not sure that's the way 
to phrase it. Maybe it's not necessary.

2) Rule U1 referenced "the following exception". I assume that this was 
a reference to U2 because only U2 (and no other Rule) would apply as an 
exception but I want to make sure this is a correct change: 
https://github.com/mpi-forum/mpi-standard/pull/912/files#diff-abcde0810b89c0faf160df24683574bc0f017a2e677e3b8dd0d2d73116877a9eR3536

Please take a look at the changes and comment on them or approve them as 
you see fit.

With this (and the previous CC changes) we're in good shape for the 4.1 
release. I will update the Excel sheet referenced earlier once I get 
write access.

We'll hopefully return to more productive discussions soon :)

Many thanks
Joseph

On 9/21/23 09:18, Joseph Schuchart wrote:
> All,
>
> We've had the MPI Forum meeting last week and have the chapter reviews 
> from other Forum members in. There were a bunch of minor changes that 
> should be easy to take on [1] (starting at Line 187). I will try to 
> fix them this week. Some of them were already addressed in the CC 
> review PR [2] so this should be easy.
>
> I don't think we need a meeting over this so I suggest to skip today's 
> meeting.
>
> There are three PRs I've opened so far to address what I think are the 
> most important issues. I would appreciate some reviews by the WG/CC: 
> (ordered by importance)
>
> 1) MPI_REQUEST_FREE is not allowed to clean up RMA requests: when 
> changed the chapter to adapt to the MPI terms we added that 
> MPI_REQUEST_FREE could be used to release a request after the 
> operation is known to be completed (i.e., after a flush). However, 
> that collides with a statement in the above paragraph that 
> MPI_REQUEST_FREE is not permitted on RMA requests. This is somewhat 
> annoying but I am hesitant to change the restriction that late in the 
> 4.1 cycle. The proposed fix is to remove the use of MPI_REQUEST_FREE 
> from the list of procedures used to release the request: 
> https://github.com/mpi-forum/mpi-standard/pull/904
>
> 2) Remove some explicit mention of MPI-2: there are still explicit 
> mentions of MPI-2 in the text chapter, which is clearly outdated. The 
> proposed fix is to rephrase the text to refer to MPI-2 window creation 
> procedures as the "previously described window creation procedures" 
> since they appear before the dynamic window creation procedure in the 
> text: https://github.com/mpi-forum/mpi-standard/pull/905
>
> 3) MPI_REPLACE and MPI_NO_OP are not new but additional operators: 
> this is a rather minor issue but it warrants a fix. The suggested fix 
> is to replace "new" with "additional": 
> https://github.com/mpi-forum/mpi-standard/pull/903
>
> Thanks for looking at them. The timeline is tight because all changes 
> for 4.1 must be finalized by September 26. Most of the remaining 
> issues can probably be wrapped up into a single PR and I will send out 
> the link once I have that up.
>
> Many thanks
> Joseph
>
>
> [1] 
> https://docs.google.com/spreadsheets/d/1ZpsyVTnfNHv85K9Mzi_3niNXdmQ67yV2iwSaoYBTGpI/edit#gid=1606317136
> [2] https://github.com/mpi-forum/mpi-standard/pull/857