Re: MPI_Set_request_{tag,source,error,count}

Steve Huss-Lederman (lederman@cs.wisc.edu)
Tue, 22 Oct 1996 13:45:48 -0500

I believe in this case I did what the forum wanted. I rewrote this
and it became 7.3 (p. 119). The section numbers changed in an attempt
to make the text appear more logically. The intent of the change was
that if you use a generalized request then you can still set the
status as described in 7.3. This is done by the fact that a callback
that occurs at the MPI_WAIT/TEST gets to status passed to the
wait/test. Thus, just setting the status works and you don't need to
be able to set the request. This is along the lines of the mail from
Rajeev before the last meeting. The other changes in generalized
requests made this change even simplier so we did that. If you still
think that this does not meet the needs of layering IO then I would
very much like to hear that. If it is true then it needs to be
changed.

Steve

P.S. - David, with competition from Rajeev, is winning the free beer for
the most/best comments on the latest text :-) I very much appreciate
the feedback. I know there are mistakes!

> Date: Tue, 22 Oct 96 14:24:42 EDT
> From: David Taylor <taylor@think.com>
> Sender: owner-mpi-core@mcs.anl.gov
> Precedence: bulk
>
> In the previous version of the specification there was a section titled:
>
> Associating Information with Status
>
> which described the four functions:
>
> MPI_Set_request_tag
> MPI_Set_request_source
> MPI_Set_request_error
> MPI_Set_request_count
>
> [Section 7.12 in the September 25th version of the external interfaces
> chapter.]
>
> I don't believe it's possible to do a portable layered implemtnation of
> the `I' functions in the i/o chapter without having these functions.
>
> Where these deleted during the October meeting?
>