Junio C Hamano <junkio@xxxxxxx> wrote: > Shawn Pearce <spearce@xxxxxxxxxxx> writes: > > > So in your test case where you thought you entered 20 pages, what you > > really had was just 2 pages. We rounded 20 bytes down to 0 pages, > > then forced that to a minimum of 2 pages. :-) > > Yes. > > And I found something else that is interesting. I still had 2 > pages windows but reduced packedgitlimit to 256kB and re-run the > same test to reproduce all merges in git.git. It still did not > fail, but it took even shorter wallclock time (again, it is only > 3 seconds or so out of 5 1/2 minuts, so it is probably > statistically insignificant). Its possible the packed_git.windows list is hurting us here. Its O(n) to locate a given window. The larger the ratio between packedGitWindowSize and packedGitLimit the longer that list can get, and the longer it can take to locate a window. Maybe that should be a simple red-black tree? More testing probably needs to be done though. -- Shawn. - To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html