Performance regression on stable

[Durham Goode]

I noticed this in our perf charts today as well.  The code I added made
the performance a function of the number of arguments passed in, instead
of the size of the repo, but it turns out in some cases the arguments are
the entire repo, in which case my code was slower than before.

For now I won't resubmit because the correct fix is non obvious and the
gain wasn't massive.


On 11/25/12 10:20 PM, "Matt Mackall" <mpm at selenic.com> wrote:

>This changeset dropped perf on the hg repo itself rather noticeably. I
>just spent an entire evening rewriting util.chunkbuffer and wondering
>why it was consistently slower than 2.4 even though all of my
>microbenchmarks looked better. Going to back this out for now, the next
>attempt belongs on the default branch.
>
>changeset:   17952:54cedee86e51
>branch:      stable
>user:        Durham Goode <durham at fb.com>
>date:        Fri Nov 16 15:39:12 2012 -0800
>summary:     commit: increase perf by avoiding checks against entire
>repo subsets
>
>Before:
>
>$ time hgs clone --pull -U hg hg2
>requesting all changes
>adding changesets
>adding manifests  
>      
>adding file changes
>      
>added 17963 changesets with 35091 changes to 2171 files
>      
>
>real	0m12.488s
>user	0m11.783s
>sys	0m0.610s
>
>After:
>
>$ rm -rf hg2; time hgs clone --pull -U hg hg2
>requesting all changes
>adding changesets
>adding manifests
>adding file changes
>      
>added 17963 changesets with 35091 changes to 2171 files
>      
>
>real	0m14.748s
>user	0m13.936s
>sys	0m0.713s
>
>-- 
>Mathematics is the supreme nostalgia of our time.
>
>