Re: Proposed Data Access Conventions Section

Bill Nitzberg (nitzberg@nas.nasa.gov)
Tue, 15 Apr 1997 17:49:51 -0700

In message <19970415130547.CP30211@k2.llnl.gov> John May writes
> Bill Nitzberg writes:
> > I am trying to clean up the introductory material in the I/O
> > chapter, and it is clear that the chapter would be more readable
> > if the common conventions were all grouped into one place.
>
> Shouldn't there be a bit more discussion of the file pointer
> types here (shared vs. independent) and how they are updated,
> or at least a reference to that information elsewhere?

File pointer details show up in the introductions for those sections.

> You describe the buftypes pretty carefully but say nothing
> about filetypes. I assume that's because it's described in
> under Views. There should be a reference to that as well.

I added a reference to "File Views".

> > The number of \mpiarg{etype} or \mpiarg{datatype} elements accessed
> > can be extracted from \mpiarg{status}
> > using the routines \func{MPI\_GET\_COUNT} or \func{MPI\_GET\_ELEMENTS}.
>
> This sentence is still very misleading. You should say
> explicitly that you get the count of etypes and buftypes
> from MPI_Get_count (either directly or by doing the math yourself,
> whatever we decide), and you get the number of basic types from
> MPI_Get_elements.

Given the MPI-1 definitions for MPI_GET_COUNT and MPI_GET_ELEMENTS,
I have changed the sentence to:

The number of \mpiarg{datatype} entries and basic elements accessed
can be extracted from \mpiarg{status} using
\func{MPI\_GET\_COUNT} and \func{MPI\_GET\_ELEMENTS}, respectively.

I believe this is both legal and specific. My preference is to
leave off details like "the user can compute the number of etypes by ...",
as that is tutorial, and this section is supposed to be definitive.

- bill