On Sun, Nov 02, 2008 at 06:35:53PM +0000, Ganesh Sittampalam wrote:
> On Fri, 31 Oct 2008, David Roundy wrote:
> >Performance patches are *always* low priority, and should therefore
> >should only be made when they are both certain to be correct
> 
> Please could you clarify what you mean by that? Much optimisation work 
> will never be "certain to be correct". For example, the darcs annotate 
> cache work will inevitably be too complicated to satisfy that criterion. 
> Even the recent series of patches I submitted, which were all quite 
> "local" in some sense, still needed some testing to shake out bugs, and 
> there's certainly no absolute guarantee that there are none left.

My first and absolute priority is to make darcs as bug-free as possible.  I
understand that changes are necessary, and that any change may introduce a
bug, and therefore every change needs to be carefully considered, what it's
benefits are versus its risks.

In the case of this series of patches, the best hope was for a moderate
constant-time improvement, which is very little performance gain.  For such
a small benefit, one should be particularly careful, rather than
particularly sloppy.

> In addition, I have to say that I find your attitude quite discouraging. 
> It might perhaps be helpful if you would elucidate your priorities and 
> attitudes as unstable maintainer, so that it's clear to everyone what 
> criteria they are working to when they contribute patches to you. For 
> example, I don't think I'd have bothered with any of my recent patches had 
> I known that you would consider them "low priority". What do you consider 
> important, beyond bug fixes? What about new features? If your vision for 
> darcs is simply to keep it in maintenance mode, then please be up front 
> about it. If not, then I think you need to recognise that blanket 
> statments like that are quite off-putting.

Scaling patches are qualitatively different from constant-factor
optimizations, and your patches were very well done.  You obviously tested
that these functions were a bottleneck, you made minimally invasive
changes, and explained why it was that your changes improved darcs'
behavior, and why it shouldn't hurt.  On top of which, you made a single
change at a time (well, in principle your foldr patch could have been done
as two separate changes), so it is very easy for me or anyone else to test
to see what effect these single changes had.

> >and have been tested to actually improve performace.
> 
> In case it's not clear, I do agree with this, and I apologise for some of 
> my recent sloppiness in this area.

I hope you didn't feel implicated, because your patches have been a breath
of fresh air!
-- 
David Roundy
http://www.darcs.net
_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users

Reply via email to