Date: Mon, 3 Aug 1998 09:44:36 -0600 (MDT) From: Leslie Hart To: Steve Huss-Lederman cc: mpi-comments@XXXXXXXXXXXXX,mpi-core@XXXXXXXXXXX Subject: Re: MPI_ALLOC_MEM & no mem In-Reply-To: <199808031231.HAA22325@XXXXXXXXXXXXXXXXXX> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-mpi-core@XXXXXXXXXXX Precedence: bulk X-UIDL: ae1b54f1bea7dde82d9bbf75f52e7521 Steve, I did make the original proposal for "continuable" errors. There were a number of arguments along the lines of Eric's that perhaps any "error" that MPIF wants to label as continuable is perhaps not an error. Returning a "NULL" value from MPI_ALLOC_MEM can be labeled to not be an error. On the otherhand, there may be a desire to abort here. I have the TEX of the original proposal. It had a list of error classes that were defined to be continuable and a list that bore no guarantees. There was a function int MPI_Err_is_continuable (int Error, int Flag) where Flag came back true if the error was continuable, false otherwise. A later proposal (I can't remember from whom) suggested an error handler to return on continuable errors and to abort on non-continuable. Perhaps a proposal consistent with Raja's comments would be to revamp the list of continuable errors, have MPI_Err_is_continuable and a new error handler. Have we established a mechanism to udate MPI 2? Do we want to? As an applications developer, I have always been uncomfortable with the notion of no guarantees regarding the state of MPI after an error was encountered. Regards, Leslie Leslie Hart (hart@XXXXXXXXXXXXX (303) 497 7253 http://www-ad.fsl.noaa.gov/ac FAX 6301 NOAA/Forecast Systems Laboratories Aviation Division Advanced Computing Branch R/E/FS5 325 Broadway Boulder, CO 80303 On Mon, 3 Aug 1998, Steve Huss-Lederman wrote: > This whole discussion of continuing after an error is making brain > cells tingle in my head. I tried to find the text this morning but > failed. I am pretty certain the JOD (and maybe external interfaces) > had a discussion of continuable errors. I don't think we ever came to > a satisfactory conclusion but recall the text and discussion. I want > to say the Leslie Hart was one of the leads but am not certain. Can > someone help out here to see if we can recall the history to avoid > recreating the wheel. > > Steve >