Hi all, Comments inline below. Martin ________________________________________________________________________ Martin Schulz, [email protected], http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA On 11/13/14, 9:06 AM, "Jeff Squyres (jsquyres)" <[email protected]> wrote:

Here's what I understand from Martin+me at the end of the call:

- name and desc are OUT strings, for both control and perf vars. So we need to have this whole mess apply to all 4 cases.

There are actually 6 cases - two more for the categories.

- we keep the idea of separating the two concepts: 1. the (conceptual) internal value for the string, which will be identical across connected processes 2. copying it to an output buffer, where it may be truncated

Yes, agreed.

- so up in 14.3.3, have a definition for "<foo> strings" (someone come up with a better name than "<foo>"):

MPI_T strings?

- an MPI implementation behaves if the string is identical across all connected processes

Not sure what you mean with behaves - I think you mean we define what identical means in 14.3.3 (which is trivial) and then say what this means when copied into output buffer. Something like: If two MPI_T strings are identical, the corresponding routine returning the string returns, for any length parameter passed into the routine, an identical content in the string buffer up to the null terminating character.

- when this string is copied to an output buffer, it may be truncated (e.g., if the user specifies an output buffer length that is too small).

- Then in the four cases (name/desc for control/perf vars), say that they are <foo> strings. For example, "desc is a <foo> string (see section 14.3.3)".

Yes, this should go into the actual parameter/argument specification of the prototype and then we just say if the name is identical, the following OUT parameters, Š, and the description has to be identical as well

Something along these lines...

-- Jeff Squyres [email protected] For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

_______________________________________________ mpiwg-tools mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-tools