X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.2 
Date: Wed, 31 Jan 2001 16:56:55 -0600
To: Karl Feind <kaf@XXXXXXX>
From: William Gropp <gropp@XXXXXXXXXXX>
Subject: Re: MPI_Abort nit
Cc: mpi-core@XXXXXXXXXXX
In-Reply-To: <200101311806.MAA02617@XXXXXXXXXXXXXXXXXXXXXXXXX>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-mpi-core@XXXXXXXXXXXXX
Precedence: bulk

At 12:06 PM 1/31/2001 -0600, Karl Feind wrote:

In the MPI-1 standard, we have this description of MPI_Abort():

   MPI_ABORT(COMM, ERRORCODE, IERROR)
   INTEGER COMM, ERRORCODE, IERROR

   This routine makes a ``best attempt'' to abort all tasks in the group
   of comm. This function does not require that the invoking environment
   take any action with the error code. However, a Unix or POSIX
   environment should handle this as a return errorcode from the main
   program or an abort(errorcode).

This is a minor nit, but POSIX and ANSI C specify that the abort()
function does not take an argument.  Hence the meaning of
"abort(errorcode)" above is unclear.  I think we should queue up a
clarification/fix to the standard unless there's a meaning here that I'm
misunderstanding.

Thanks; I've added it to the clarification list.

My interpretation followed the "handle this as a return errorcode from the main program" branch; but either interpretation is ok.  However, note that the "should" here was not intended to be mandatory; it merely suggests what a user-friendly implementation might do.   As a *user*, I'd like the specified error code returned by mpirun to the invoking environment.  However, any clarification should make clear that there is no required return value (this is a quality of implementation issue).

I can see environments where the return status indicates not only the code but the processes that caused the abort ...

Bill