Re: first batch of corrections

[William Gropp <gropp@mcs.anl.gov>]

At 04:25 PM 5/15/2001 -0400, Richard Treumann wrote:

>I have reviewed the first batch of corrections and have the following
>comments:
>
>Item 1:
>
>I agree the MPI_Finalize is collective across all currently connected
>processes.  I do not see how any MPI program that would have a problem with
>this could be considered otherwise valid MPI.
>
>I endorse Marc's suggestion that we point to the discussion in 5.5.4 rather
>than paraphrase.

I agree.  I've added a reference to 5.5.4 to the draft errata.  I left in 
the paraphrase as an aid to reading the document:

MPI_FINALIZE is collective over all connected processes.
If no processes were spawned, accepted or connected then this means
over MPI_COMM_WORLD; otherwise it is collective over the
union of all processes that have been and continue to be connected,
as explained in Section 5.5.4.



>Item 6:
>
>This involves a set of obvious corrections to MPI_Alltoallw and I agree in
>principle.  However, there is a more difficult issue also to be decided and
>that is whether sdispls and rdispls should be MPI_Aint rather than integer.
>Using C int is clearly an unintended restrictiveness on the usability of
>MPI_Alltoallw for 64 bit MPI.  I am confident that if the Forum had noticed
>and discussed before MPI-2 was published, the type would have been
>MPI_Aint.
>
>I would not like to see any vote for making these array_of_int considered
>to be a resolution of the whole issue.  Since the vote is defined as
>approving new text I suggest we leave this one out of the first vote and
>come back to it when we deal with the MPI_Aint displacements question.

This is a change only to the language-independent description of the 
routine.  I've checked, and we consistently use the term "integer" in these 
descriptions for "whatever integer type the language binding chooses".  For 
example, see the new datatype creation routines whose C bindings use 
MPI_Aint for displacements - they still describe the values as "array of 
integer".

Thus, this change is independent of the choice of the type of integer used 
for the displacements.

Bill