On Fri, Oct 31, 2008 at 07:55:28 -0400, David Roundy wrote: > Except that we *know* that the new code without that indirection layer is > still slower than OldFastPackedString in some tests. Which means that > there's an unresolved performance problem with bytestrings. Or perhaps > it's a new regression, rather than a remaining regression.
As I understand it, you are still intending [*] to apply this patch bundle. Doing so would be really great so that we can clear up this logjam and get back to hacking :-) We should probably forget about the actual performance details for now, and just think of this as modernising the code to use the new API. The cleaner code can be used as a basis for future optimisation work that will let us catch up and surpass any regressions. Note that I have fossilised a copy of both pre and post salvo 8 darcses, so performance forensics are always possible after the fact. I will attach context files and some new numbers (with new bytestring, and taking the minimum across three runs, not the mean) later. Meanwhile, is there anything we can do to help? [*] http://lists.osuosl.org/pipermail/darcs-users/2008-October/015291.html -- Eric Kow <http://www.nltg.brighton.ac.uk/home/Eric.Kow> PGP Key ID: 08AC04F9
signature.asc
Description: Digital signature
_______________________________________________ darcs-users mailing list [email protected] http://lists.osuosl.org/mailman/listinfo/darcs-users
