[darcs-users] is darcs 2.4 supposed to be faster? Re: darcs 2.4 beta 2 release

Eric Kow kowey at darcs.net
Sun Jan 24 09:05:06 UTC 2010

On Sun, Jan 24, 2010 at 00:26:31 -0700, Zooko Wilcox-O'Hearn wrote:
> This text makes it sound as though users who upgrade to darcs 2.4
> from earlier versions of darcs can expect performance improvements,
> Are we supposed to expect performance improvements from this
> release?  Perhaps the release announcement could be made more
> specific about what, if any, operations are expected to be faster.
> "Index-based diffing" means nothing to me, except possibly that the
> "darcs whatsnew" and "darcs diff" commands should be effected.

You're right, Zooko.  We are working to make sure our message on this is
as clear, accurate and precise a message as possible.

I think the heart of your answer lies in

Namely, I expect these operations to be faster than in darcs 2.3.1:
 - record
 - unrecord
 - amend-record
 - revert
 - unrevert

Furthermore, I think these things will have an impact:
 - darcs optimize --pristine [all commands?]
 - apply no longer sleeps for 1 second
 - darcs get now urges people to upgrade their servers to
   hashed [since darcs 2.3.1, getting from old-fashioned repos
   is slow because we convert to hashed for safety]

Things confusing the issue are:

 - Hashed repos vs old-fashioned?
   I think we have to ignore this question for now and just focus on
   comparing hashed repos with hashed repos.

 - Recent regressions in check/repair.
   In between the first two betas, Petr has gained a lot of ground on
   the regressions.  Not all of it, but it's much more reasonable

 - Many variables to account for: NFS? branches/symlinks? global
   cache size? size of repos in number of files? 

Some obstacles we still face (and which we need help on) are:

 - Insufficient grasp of statistics in the Darcs Team (Here's one way
   you can help out, darcs-users!)

 - Lack of benchmarks for the operations record/unrecord/amend-record/
   revert/unrevert <http://repos.mornfall.net/darcs/benchmark>

 - Darcs-benchmark buggy on Windows: we need a Windows hacker to swoop
   in and fix this.  Perhaps it's just another file handle leak?  So
   it doesn't have to be a Windows guy, necessarily, just somebody who
   can help track down this leak and work with Max to see if things are

 - No way to compare optimize --pristine repos against the unoptimised


Eric Kow <http://www.nltg.brighton.ac.uk/home/Eric.Kow>
PGP Key ID: 08AC04F9
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <http://lists.osuosl.org/pipermail/darcs-users/attachments/20100124/ed4d297a/attachment.pgp>

More information about the darcs-users mailing list