Re: [PATCH] Rename detection: Avoid repeated filespec population

From: Jeff King <peff@peff.net>
Date: Jan 20, 2009, 22:12 UTC
Message-ID: <20090120221214.GA13668@coredump.intra.peff.net>
In-Reply-To: <20090120212723.GA10967@coredump.intra.peff.net>

On Tue, Jan 20, 2009 at 04:27:23PM -0500, Jeff King wrote:

Show 6 quoted lines

> I have a pathological real-world rename test case from a long time ago.
> It's so awful on rename because almost the whole repo was reorganized,
> but almost every path got its content adjusted, too. I was hoping it
> would be better with your patch, but it isn't: as it turns out, it is
> _too_ uniform. The cheap size checks don't work because all of the files
> are almost the same size (they're all jpgs from a camera).

Nevermind, I'm stupid. Your patch _does_ yield a significant improvement for my test case. I was accidentally comparing your patch to some of my previous optimization attempts, not to vanilla git.

Without your patch, I get:

  $ time git show -l0 06d28867 >/dev/null
  841.88user 146.95system 17:18.99elapsed 95%CPU (0avgtext+0avgdata 0maxresident)k
  2476880inputs+0outputs (19372major+27092797minor)pagefaults 0swaps

With it, I get:

  $ time git show -l0 06d28867 >/dev/null
  100.51user 5.76system 2:31.07elapsed 70%CPU (0avgtext+0avgdata 0maxresident)k
  2020176inputs+0outputs (15771major+836127minor)pagefaults 0swaps

So much much better.

-Peff