[darcs-users] [Haskell-cafe] Re: poll: how can we help you contribute to darcs?

[Alberto Bertogli]

On Wed, Aug 06, 2008 at 10:11:16PM -0700, Jason Dagit wrote:
> On Wed, Aug 6, 2008 at 7:35 PM, Alberto Bertogli
> <albertito at blitiri.com.ar>wrote:
> > I don't want this to turn into "I'm faster/easier/whatever than you",
> > and personally I don't think the speed is the issue with small
> > repositories; but since you asked, and I've _just_ got into a situation
> > like this...
> 
> Good point.  No one wants to read a pissing contest.  But, maybe we should
> have a "Version Control Shootout" ala http://shootout.alioth.debian.org/.
> We could probably even use their existing framework.
> 
> You wouldn't need to be an expert at any particular version control system
> to establish a site like that and it could garner someone a decent bit of
> recognition as well as giving version control implementers some concrete
> goals.  The more I think about it, the more I want that site now.  Maybe I
> should start putting it together myself...

Sounds nice. There are a lot of feature comparison sites, but not so
many with performance numbers. Having something that can be automated
like the language shootout would be quite interesting.

If you are seriously going to do this and need a hand with git's side,
let me know.


> > That's easy math: 16 times slower (since darcs doesn't color the diff
> > but only the word "hunk", I think it's fair to use the 0.010s number).
> >
> > Yes, 0.16s may not seem like a lot to you, but when you're used to it
> > taking 0.01s, you notice.
> 
> 
> Wow, thanks.  I prefer facts over anecdotes.  Thanks for taking the time to
> run that experiment.  That is quite a difference.  Especially considering

You're welcome. I like facts too, that's why I ran the numbers.


> that the linux kernel has many more patches and files.  I think I know part
> of the problem too.  Was the git patch comparable in size to the darcs patch
> that you picked?

I would say so, yes. The Darcs patch has a one-line commit message (not
counting headers), 4 +, 4 -. The Linux patch has an 11-line commit
message (idem), 5 +, 1 -. They were chosen completely by random (I just
scrolled down the logs a bit and picked what was there).


I looked for a heavier patch in darcs' repo and found
20080123013642-20bb4-23996deb7f6aa8dbbc300ac82e66568f4648071d.gz of
around 4 files changed, 1008 +, 646 - (sorry, no diffstat, I seem unable
to get darcs to output the patch in diff format [*]). It takes 0.183s to
show it using darcs changes -v > /dev/null.

A similar change for Linux is ad68076e07fa01bd0c98278a959d0fd2bb26f1ac:
12 files changed, 1009 insertions(+), 642 deletions(-).
It takes 0.017s to show it using git show > /dev/null.


Thanks,
		Alberto


[*]: http://bugs.darcs.net/issue966