[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: MPI-2.1 corrections, Batch 2 // Item 3



At 12:17 PM 7/17/2001 +0200, Rolf Rabenseifner wrote:
> > 3:  Page 61, after line 36.  Add the following (paralleling the errata to
> >   MPI-1.1):
> >
> > MPI_{COMM,WIN,FILE}_GET_ERRHANDLER behave as if a
> > new error handler object is created.
> > This provides behavior similar to that of MPI_COMM_GROUP.
>
>Sorry, but only looking on MPI-2 page 61 and the new text above,
>I can not understand this item.
>
>Looking also at
>http://www-unix.mcs.anl.gov/~gropp/mpi/mpi-errata/discuss/errhandlers/
>and there on the last mail
>
>   |  I propose the following clarification be added to the errata:
>   |
>   |  MPI_GET_ERRORHANDLER, MPI_{COMM,WIN,FILE}_GET_ERRORHANDLER behave as 
> if a
>   |  new error handler object is created.
>   |  This provides behavior similar to that of MPI_COMM_GROUP.
>   |
>   |  Rationale:  This case was overlooked in the MPI-1 document.
>   |
>   |  Any objections?
>   |
>   |  Bill
>
>I see that someone has removed MPI_GET_ERRORHANDLER and the Rationale.
>Looking at the text MPI-1, page 139, MPI_COMM_GROUP ...,
>I can not understand the sentese
>"This provides behavior similar to that of MPI_COMM_GROUP".
>
>It may be, that the text of this errata can be improved by someone
>who is familiar with the content of this problem.

I've added some text to the web page as well as included MPI_GET_ERRHANDLER 
in the list.  In short, we did not explicitly say that after an 
MPI_ERRHANDLER_GET the user should call MPI_ERRHANDLER_FREE, in the same 
way that after MPI_COMM_GROUP, the user must call MPI_GROUP_FREE when the 
group is no longer needed.  There was a long email discussion here, but the 
consensus was the most of the implementations implemented what this 
clarification says and that this is what users should expect.

Bill