On Wed, Feb 06, 2008 at 08:34:57PM -0700, zooko wrote: > > On Feb 6, 2008, at 11:58 AM, David Roundy wrote: > > > On Wed, Feb 06, 2008 at 11:29:42AM -0700, zooko wrote: > >> time tar xjvf ciphercycles-20070205.tar.bz2 > > > > Any chance you could write up a little script that generates something > > functionally equivalent to this directory for a benchmarking script? > > You can get that actual tarball with:
Thanks for the test case, Zooko, and for pointing out this regression! Could you rerun your timings when you have pulled this change: Thu Feb 7 15:00:51 EST 2008 David Roundy <[EMAIL PROTECTED]> * speed up pending handling. In fixing a bug, I thoughtlessly made pending handling O(N^2) more often than it needs to be... which was the "safe" choice, because it was previously the "fast" code that had the bug. I believe I've fixed the main regression. On my computer, your record took 45s, and I got it down to just under twice the time that darcs 1.0.9rc1 took. I gather your computer is faster. At this point, I think I've fixed the scaling issues (although I'm not certain, there could remain some O(N^2) issues with a smaller prefactor), and the remainder may just be micro-optimization. We changed our list type for holding sequences of patches (besides generalizing to support alternate patch semantics), and it could well be that there's a factor of two slowdown somewhere in there. Hopefully soon someone will volunteer to create a timings framework. I've now found two handy inverse sequences time darcs obliterate --last 100 -a && time darcs pull -a and time darcs unrecord --last 1 -a && time darcs record -m foo -a where this latter test came about because of your test case. The first was Simon M's test case. If we can generate more of these pairs of inverse operations that display slowdowns that'd be great. Once I've got something like this that I can reproduce (and also reproduce with darcs 1, so I know that we've got a regression), it's generally only a matter of a few hours to find the main causes. -- David Roundy Department of Physics Oregon State University _______________________________________________ darcs-devel mailing list darcs-devel@darcs.net http://lists.osuosl.org/mailman/listinfo/darcs-devel
