This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Feb's patch resolution rate


> Date: Tue, 23 Mar 2004 16:31:52 -0500
> From: Andrew Cagney <cagney@gnu.org>
> 
> >>> - typical (median) review time is 1 day
> > 
> > 
> >>> - average review time is ~2.5 days
> > 
> > 
> > I assume these only count the resolved patches?  Because:
> 
> Obviously, hence the ~.  The median, however, would not change.

Careful here: statistics are sometimes worse than a lie ;-)

Seriously, though: about my only comment to the figures that you
presented is that it is IMHO wrong in this case to use estimators,
such as the median, that are resistant to outliers.  That's because no
one, to the best of my knowledge, is claiming that GDB development is
dysfunctional.  As long as GDB maintenance as a whole works fairly
well, the average figures of any reasonable performance estimator will
be good.  IMHO, it is the (relatively rare) exceptions from the rule
that bothered and continue to bother those among us who came up with
suggestions to modify the existing practices.  And it is precisely
those exceptions that the median and its ilk give a zero weight.

In other words, to quantify the validity of complaints about the
current maintenance procedures, one needs to analyze and study the
outliers, those cases that are in the tails of the distribution, not
the average figures that are bound to be good by any measure, as GDB
is, by and large, a successful project.

These discussions grew from uneasy feelings, to put it mildly, shared
by several active maintainers.  As any good psychologist will tell
you, people are not using averages or medians when forming their
feelings, their reaction to outliers is acute and disproportional to
the actual percentiles.  If we want to solve these kinds of problems,
I think we need to summon techniques that closely follow these human
tendencies.

That said, I think it's a Good Thing that you posted those figures:
they allow us put the issue in perspective.  We just need to augment
the average figures with similarly quantitative analysis of the cases
where patch review took an exceptionally long time.  (Any takers? ;-)


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]