[Mpi-forum] Errata handling
rabenseifner at hlrs.de
Wed Dec 5 11:23:56 CST 2012
I detected that there seems to be a discussion.
>From my viewpoint based on experience since MPI-2.0,
the processing should be very simple and clear:
- only important errata are published, not the typos.
- important is that the interface was definitely inconsistent
or obviously wrong.
- enough review by committee members, i.e., it is clear that
the correction is correct.
- real reading in the forum, because it is about the interface.
- single reading + vote for corrections of inconsistencies.
Errata should pass always in the next meeting after they
Errata are voted and "3.0 compiant" always means 3.0+errata compliant.
Errata are for obvious bugs and inconsistencies.
- Being compliant to a bug is stupied and must be corrected
as soon as detected.
(Example: It was never intented that you are MPI-2.2 compliant
when having BOOL with 4 bytes in the external32 datarep.)
- Being compliant to an inconsistency is impossible,
because MPI-3.0 says A and not-A;
you can never be compliant to both A and not-A.
There should be only ONE errata document, which is
updated after each meeting if new errata are added.
The errata should be sorted by page and linenumbers in MPI-3.0.
By this, there are older versions (visible at the date)
of the errata document, but only the newest is binding.
If we want to have the history in, we can add for each
erratum the following parantheses:
"(detected in ticket #nnn on May 1, 2099, voted in on Jun 16, 2099)".
Because we always should try to circulate errata in a ticket,
all implementors have early access to the information.
It should be voted in in the same meeting as it was read.
That's enough. This should need 30-60 minutes in a forum meeting.
That's fast compared to a lot of emailing or email-voting.
That's the way we did it in the past for all corrections
in the voted Fortran interface - and it worked.
My strong recommendation:
- Do it simple and effective!
- And start it now!
- Use the meeting time for content of the MPI standard
and not for complicating the errata process.
Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner at hlrs.de
High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530
University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832
Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner
Nobelstr. 19, D-70550 Stuttgart, Germany . (Office: Room 1.307)
More information about the mpi-forum