after there was no counter-argument the last 2 days I may ask you
to add my MPI_INFO proposal as an amendment to the dynamic
chapter that we can discuss and vote on it next meeting.
(e.g. after the subsection MPI\_APPNUM)
\subsection{\mpifunc{MPI\_INFO}}
There is a predefined info-handle attribute \mpifunc{MPI\_INFO}
of \mpifunc{MPI\_COMM\_WORLD}.
\mpifunc{MPI\_INFO} gives an info object that represents the info
argument in the starting \code{mpiexec} or \mpifunc{MPI\_SPAWN(\_...)}.
\begin{rationale}
In conjuction with \code{argv/argc}, \mpifunc{MPI\_COMM\_PARENT}
and \mpifunc{MPI\_APPNUM} this gives the application the full
information of how it was started.
The returned info handle can be modified or freed without
changing the outcome of an additional call to
\mpifunc{MPI\_ATTR\_GET(comm, MPI\_INFO, &info\_handle)}.
\end{rationale}
\discuss{
Most implementations would have this information
anywhere inside and it is usual practice in Unix that applications
can examine how they were started.
Additionally this proposal supports glue library writers that
want to give some information (e.g. a port name) to the MPI processes
started with a dedicated MPI implementation. (The hack of adding and
removing arguments from argv+argc can be avoided then.)
}
Rolf
Rolf Rabenseifner (Computer Center )
Rechenzentrum Universitaet Stuttgart (University of Stuttgart)
Allmandring 30 Phone: ++49 711 6855530
D-70550 Stuttgart 80 FAX: ++49 711 6787626
Germany rabenseifner@rus.uni-stuttgart.de