[Mpi-forum] MPI_ERR_UNSUPPORTED_DATAREP error class definition scoped too tightly
thakur at mcs.anl.gov
Thu Mar 13 21:08:18 CDT 2014
Even if we remove the reference to File_set_view, people may not expect non-I/O functions to return any of those error classes. It is located in the I/O chapter, the section title is "I/O Error Classes", and the first line of the section is "The implementation dependent error codes returned by the I/O routines can be converted into the error classes defined in Table 13.3."
There is another error class there that could be used elsewhere:
MPI_ERR_NOT_SAME - Collective argument not identical on all processes, or collective routines called in a different order by different processes
On Mar 13, 2014, at 5:08 PM, Fab Tillier <ftillier at microsoft.com> wrote:
> Ok, how about removing the reference to MPI_FILE_SET_VIEW, then?
> I'd like to return MPI_ERR_UNSUPPORTED_DATAREP as the error class in MPI_PACK_* functions for that case, and the current text kind of prohibits that.
> We can then rationalize the location of error classes in MPI-4.
> Rajeev Thakur wrote on Thu, 13 Mar 2014 at 12:52:44
>> The table on pg 551 is about I/O error codes and classes. So
>> Pack_external/Unpack_external (from 9 chapters away) would look out of
>> place there. The real solution might be to move the I/O error codes
>> and classes from the I/O chapter to the chapter on error codes and
>> classes, but that would be more than a ticket 0 change.
>> On Mar 13, 2014, at 12:02 PM, Fab Tillier <ftillier at microsoft.com> wrote:
>>> Hi folks,
>>> The standard defines MPI_ERR_UNSUPPORTED_DATAREP as an I/O error
>> class, with the meaning "Unsupported datarep passed to
>> MPI_FILE_SET_VIEW". However, datareps are also passed as input to
>> MPI_PACK_EXTERNAL and MPI_UNPACK_EXTERNAL. Shouldn't these functions
>> use the same error class for unsupported datareps, rather than
>>> Can we update the description of MPI_ERR_UNSUPPORTED_DATAREP
>> (page 551, line 10) to also include MPI_PACK_EXTERNAL and
>> MPI_UNPACK_EXTERNAL as a ticket-0 item for MPI 3.1, or will this need
>> to happen in a later version?
>>> mpi-forum mailing list
>>> mpi-forum at lists.mpi-forum.org
>> mpi-forum mailing list
>> mpi-forum at lists.mpi-forum.org
> mpi-forum mailing list
> mpi-forum at lists.mpi-forum.org
More information about the mpi-forum