Re: vote on first batch of corrections for MPI 2.1

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

At 04:39 PM 6/8/2001 +0200, Hans-Christian.Hoppe@pallas.com wrote:

>Dear group,
>
>while most of the items in the first batch are a clear "yes", I'd like to
>make a few exceptions:
>
>3., 9., 12., 13.: these create compatibility problems and will require
>recompilation of codes. Alternative: introduce the new, correct names, and
>keep the old ones as duplicates.

While we are voting on these now as part of the whole set, all four of 
these are part of the original Errata that has been posted on the MPI Forum 
web site since just after the standard was originally released (May 20, 
1998).  The most popular C++ implementation already reflects these errata 
items for the MPI-1 functions.


>8.: that'll break implementations that do in-place conversions; Eric Salo
>at one point was fighting to keep that door open in the MPI-2
>deliberations. Just to mention it - he probably never went ahead and
>implemented this.

This was also in the original list (though I agree with your comment, users 
don't like the part of the standard that prohibits access to the send 
buffer while a send is taking place).

Bill