[darcs-users] performance (was GHC and Darcs)

[zooko]

On Jul 30, 2008, at 8:21 AM, Eric Kow wrote:
>
> Some bugs filed by the GHC team
...
> [b] still too slow:  http://bugs.darcs.net/issue973
...
>
>> I have the idea that darcs-2 has not yet benefitted from newly  
>> developed
>> Haskell tools such as profilers.
>
> As I mentioned on Reddit, darcs essentially suffers from the day job
> problem.  If we could get a Haskeller whose job is to work on darcs,
> at least for a few months, I think we could make a lot of progress.
> (The best candidate would of course have some experience with writing
> fast Haskell, and of course, funding).

You're probably right that this would help, but on the other hand it  
probably isn't necessary.  Many successful open source projects are  
done for love, not for money.  For example, I'm pretty sure that hg  
(a candidate distributed revision control tool that the GHC  
developers are considering using) was produced solely as a hobby/ 
independent/pro bono project, and that bzr (one which the GHC  
developers have rejected) was developed by a team of people who were  
paid full-time to work on bzr and related technologies.

On the other hand, many successful open source projects are supported  
financially.

In any case, the next step for anyone who wants to improve darcs  
performance is probably to run a profiler on darcs on a large  
repository (e.g. ghc) in hashed format and in old-fashioned-inventory  
format.

I don't know anything about the new Haskell profilers, but I bet that  
oprofile would yield interesting insights, and I'll bet a simple  
strace -T -r would, too.

I've posted my notes about strace results before to the darcs-users  
list, but I didn't get any response.  Perhaps I'll do so again using  
darcs-2.0.2 and a ghc repository.

Regards,

Zooko