Re: new filter proposal

Bill Nitzberg (nitzberg@nas.nasa.gov)
Wed, 16 Apr 1997 11:45:50 -0700

In message <199704161821.NAA10285@abacus> Rajeev Thakur writes
>
> However, dtype_file_extent_fn can be passed either the etype or any of
> the basic datatypes used to construct etype (MPI_LB and MPI_UB not
> needed because their size is always 0).
>

Since dtype_file_extent_fn will also be called when the user calls
MPI_FILE_GET_TYPE_EXTENT (for user defined data representations),
I think we should either make no restrictions at all, or have
something like:

\MPI/ will only pass the following \mpiarg{datatype}s to
\func{dtype\_file\_extent\_fn}:
\begin{itemize}
\item a \mpiarg{datatype} passed to \func{MPI\_FILE\_GET\_TYPE\_EXTENT}
for a file with user defined data representation \mpiarg{datarep},
\item the \mpiarg{etype} used to set the view
for a file with user defined data representation \mpiarg{datarep}, or
\item an \MPI/ basic type used in the construction of either of the above.
\end{itemize}

- bill