more wording clarification
Linda Stanberry (linda@k2.llnl.gov)
Thu, 12 Jun 1997 17:54:09 -0700 (PDT)
I believe that the wording for compatibility of etypes and filetypes in
the 5/8 edition of MPI_FILE_SET_VIEW (someone may already have reported
this!?) is erroneous. The sentence that is troublesome for me is "In
addition, the typemap displacements for a filetype must be multiples of
the extent of the etype (...)." But typemap displacements are for a
sequence of primitive datatypes, and etype is not required to be a
primitive datatype. I understand the intent (i.e., that any holes are
to be etype-extent multiples), but I don't think you can state this as a
requirement on the typemap displacements. If the intent is clear to
the users, this can just be stated as "In addition, any holes in the
filetype must be multiples of the etype's extent."
It is unclear to me if the onus is on the user to guarantee the etype and
filetype meet these compatibility and other type requirements, or if it
is the burden of MPI_SET_VIEW to decode etypes and filetypes to verify
these conditions. That is, when the MPI2 spec says "it is erroneous"
does that mean that the user can expect an error is returned or that they
will simply get erroneous file I/O? I realize that this is mostly a QOI
issue, but I do wish the spec gave more consistent guidance in this area,
such as the error classes to be returned when error returns are expected.
(This observation applies to other parts of the MPI2 spec, not just MPI-IO!)
Linda Stanberry
lstanberry@llnl.gov