Newsgroups: comp.parallel.mpi
From: furnish@dino.ph.utexas.edu (Geoffrey Furnish)
Subject: Re: MPI_Finalize necessary ?
Organization: Institute for Fusion Studies, University of Texas at Austin
Date: 05 Sep 1995 18:29:59 GMT
Message-ID: <FURNISH.95Sep5132959@dino.ph.utexas.edu>

In article <42e07h$82e@murrow.corp.sgi.com> salo@mrjones.engr.sgi.com (Eric Salo) writes:

> From: salo@mrjones.engr.sgi.com (Eric Salo)

> > Is a MPI_Finalize essential ?
> 
> This is a very good question. To the best of my knowledge, the official
> answer from the MPI Forum is: We don't know.
> 
> Nowhere in the standard does it explicity say that MPI_Finalize *must* be
> called, although that may have been the original intent. (Help me out here,
> guys!) SGI's implementation of MPI does require it.

I am not an MPI implementor, but it is not hard for me to imagine that
an MPI implementation would want to use defered handling of some
things (where semanticaly acceptable), or even just "be a good citizen
and clean up after itself".

I personally don't have a lot of sympathy for C or Fortran users who
simply don't want to be bothered to call MPI_Finalize().  For object
oriented languages, such matters can always be handled through
instance destructors, or even through a global destructor, so that
clients would not have to know about it.  So there does not seem to me
to be any good excuse for dissallowing MPI implementors from relying
upon MPI_Finalize in order to provide a semantically correct
implementation. 

-- 
--
Geoffrey Furnish                  http://dino.ph.utexas.edu/~furnish
UT Institute for Fusion Studies,  furnish@dino.ph.utexas.edu   512-471-6147

	"Pushing back the boundary of inanity."

