Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] 1.3 branch
From: Terry Dontje (Terry.Dontje_at_[hidden])
Date: 2008-07-24 14:00:09


Jeff Squyres wrote:
> I think that this is exactly the problem -- when a developer puts
> something back to the trunk, they (including me!) almost always commit
> what they think is the fix to the problem. But hindsight is 20/20.
> Case in point: it took Ralph and me and others over a week to fully
> fix the SM/paffinity issue, even though we thought at each commit,
> "yep, that's it. This commit will fix the problem." Looking back, we
> obviously missed some things during that process, but we didn't
> realize that at the time, even though we were being as careful as we
> could.
>
> If I could be so bold -- I think that's what Terry was asking: how are
> we supposed to know?
>
> My $0.02: how to know "it really solves the problem without
> introducing new ones" is a really, really hard determination. Even
> for very small code changes. :-)
>
To beat this horse into the ground. A good example is the latest
performance regression due to the paffinity changes. If you were
testing on RH 5.1 you would not have found the problem. And I think
that is true with a lot of our changes in that we test on a limited set
of platforms locally so there is definitely a risk here.

So you can stand by the "Do the right thing" mantra but at the same time
we need to realize problems will happen and the only way to reduce them
is by shrinking the window of ambiguity.

--td
>
>
> On Jul 24, 2008, at 10:44 AM, George Bosilca wrote:
>
>> Terry,
>>
>> I did not and I will not enforce any policy at this point. I'm
>> confident developers in this community can take such decisions by
>> themselves, without restrictions from the RM. As a hint, the most
>> basic common sense (make sure it compile and it really solve the
>> problem it is supposed to solve without introducing new ones) is a
>> good decision metric.
>>
>> george.
>>
>> On Jul 24, 2008, at 3:55 PM, Terry Dontje wrote:
>>
>>> It might be worthwhile to spell out the conditions of when someone
>>> should let changes soak or not. Considering your changeset 19011
>>> was putback without much soak time. I am not saying 19011 needed
>>> more soak time just that I think it adds potential confusion as to
>>> what one really needs to do versus amount of code a change.
>>>
>>> --td
>>>
>>> George Bosilca wrote:
>>>> Unfortunately over the last couple of days I realize that the
>>>> patches from the trunk are moved to the 1.3 too rapidly and usually
>>>> without much testing. I would like to remember to everybody that
>>>> the 1.3, while opened for community commits, is supposed to become
>>>> stable at one point and that we should do the best efforts to keep
>>>> it that way as long as possible.
>>>>
>>>> Please allows few days of testing time before moving your patches
>>>> from the trunk to the 1.3.
>>>>
>>>> george.
>>>>
>>>> ------------------------------------------------------------------------
>>>>
>>>>
>>>> _______________________________________________
>>>> devel mailing list
>>>> devel_at_[hidden]
>>>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>>>>
>>>
>>> _______________________________________________
>>> devel mailing list
>>> devel_at_[hidden]
>>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>>
>> _______________________________________________
>> devel mailing list
>> devel_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>
>