Re: [PATCH 4/4] Invalidate cache-tree entries for touched paths in git-apply.

Junio C Hamano [off-list ref] writes:

 * ... then the big rock falls.  With this, I tried to apply and
   then write-tree "diff-tree -p $commit^1 $commit" on top of
   "$commit^1" for the last 20 or so commits in the kernel tree.
   The "master" version takes 0.15 second per patch on my Duron
   750 with 700MB, while this one does that in 0.06 second.
   This also helps the memory pressure because we do not have to
   regenerate unchanged trees.  810 minor faults with the patch
   vs 2150 minor faults without.

Sorry, but not really.  The patch is wrong and the measurement
was flawed.

It was doing unnecessarily more work in git-apply, which made
git-write-tree a no-op, and I was measuring only git-write-tree.

The following goes on top of the series to remove that
unnecessary work from git-apply.  Unfortunately this makes the
overall combination a bit slower than before X-<.

But I have not optimized cache-tree.c for speed; for example,
its subtree lists are not even sorted.  Once that is done we may
get decent speedups from the combo.

---

diff --git a/apply.c b/apply.c
index 5fa2c1e..e283df3 100644
--- a/apply.c
+++ b/apply.c

@@ -1935,8 +1935,6 @@ static int apply_patch(int fd, const cha
 				  active_cache_sha1) ||
 		    commit_index_file(&cache_file))
 			die("Unable to write new cachefile");
-		cache_tree_update(active_cache_tree,
-				  active_cache, active_nr, 1);
 		write_cache_tree(active_cache_sha1, active_cache_tree);
 	}