[mpiwg-tools] Binding of predefined handles

[Marc-André Hermanns]

Dear all,

Howard, Dan, and I discussed the binding problem on Monday during the
sessions WG time slot (thanks again for the time and input) and we
came to quite a change in direction this errata is going.

We were discussing what was available when for predefined handles
(contemplating what *link-time constant* means) and live-testing a bit
with Open-MPI.

Result was that a test with

MPI_Comm comm = MPI_COMM_WORLD;
MPI_Init(...);
MPI_Send(..., comm);

works as we would expect from a link-time constant, in that the value
(of the handle, not necessarily the object behind it) is known to MPI
before Init and does not change.

That gave rise to the question, "If that is possible, what prevents
the implementation of 'lazy binding' for an MPI libary?"

So this would be a question to all implementors on the list.

I updated the Errata with a new suggested text (now actually allowing
binding to predefined handles at any time between MPI_T initialization
and finalization, as a start for discussing this thoroughly before
September.

https://github.com/mpi-forum/mpi-standard/pull/161/files#diff-f6fa82f9df45c241a7dedbdce0bde29cR144-R148

The core idea here is that, as a library should always be able to tell
apart a predefined handle from a user-defined one, it could catch any
binding and postpone it until the handle is initialized.


Cheers,
Marc-Andre

-- 
Dr. rer. nat. Marc-André Hermanns

IT Center
Group: High Performance Computing
Division: Computational Science and Engineering
RWTH Aachen University
Seffenter Weg 23
52074 Aachen
Phone: +49 241 80-24381
hermanns at itc.rwth-aachen.de
www.itc.rwth-aachen.de

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5336 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-tools/attachments/20200708/943f20bc/attachment.p7s>