On Dec 30, 2007, at 5:23 PM, Brian Downing wrote:
On Sun, Dec 30, 2007 at 05:00:35PM +0100, Johannes Schindelin wrote:
I had the impression that Brian found a faster implementation,
which could
not be included for license reasons...
But for the time being, this is better than nothing, I guess.
It wasn't me. It was "Mike Ralphson" in the "Some git performance
measurements" thread on the vger list.
http://article.gmane.org/gmane.comp.version-control.git/67424
However, I came to the conclusion that he was actually comparing
against
glibc's quicksort implementation; his comment about being tuned for
"Sun 4/260" matches a comment in glibc's quicksort implementation,
not the mergesort that I used as a base. Later, in:
http://article.gmane.org/gmane.comp.version-control.git/67443
He said the mergesort in mingw was faster.
But apparently the discussion didn't result in a patch then. I
can't find a patch sent to the git list that would add a qsort
replacement.
The qsort replacement seems to be of general interest and I
therefore prefer if it went to official git ASAP. My main goal
is still to reduce differences between 4msysgit, mingw, and
official git.
I'm not sure if this is the right time to send a patch upstream,
because we're in the middle of the 1.5.4 freeze. So, I'll keep
it in 4msysgit only. But maybe later someone (Brian?) could
send the patch upstream.
The latest version of the patch is here:
http://repo.or.cz/w/git/mingw/4msysgit.git?a=commitdiff;h=bc8e4538
Steffen
