diff: cache textconv output

Running a textconv filter can take a long time. It's
particularly bad for a large file which needs to be spooled
to disk, but even for small files, the fork+exec overhead
can add up for something like "git log -p".

This patch uses the notes-cache mechanism to keep a fast
cache of textconv output. Caches are stored in
refs/notes/textconv/$x, where $x is the userdiff driver
defined in gitattributes.

Caching is enabled only if diff.$x.cachetextconv is true.

In my test repo, on a commit with 45 jpg and avi files
changed and a textconv to show their exif tags:

  [before]
  $ time git show >/dev/null
  real    0m13.724s
  user    0m12.057s
  sys     0m1.624s

  [after, first run]
  $ git config diff.mfo.cachetextconv true
  $ time git show >/dev/null
  real    0m14.252s
  user    0m12.197s
  sys     0m1.800s

  [after, subsequent runs]
  $ time git show >/dev/null
  real    0m0.352s
  user    0m0.148s
  sys     0m0.200s

So for a slight (3.8%) cost on the first run, we achieve an
almost 40x speed up on subsequent runs.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

1 files changed, 4 insertions, 0 deletions

diff --git a/userdiff.h b/userdiff.h
index c315159..942d594 100644
--- a/userdiff.h
+++ b/userdiff.h
@@ -1,6 +1,8 @@
 #ifndef USERDIFF_H
 #define USERDIFF_H
 
+#include "notes-cache.h"
+
 struct userdiff_funcname {
 	const char *pattern;
 	int cflags;
@@ -13,6 +15,8 @@ struct userdiff_driver {
 	struct userdiff_funcname funcname;
 	const char *word_regex;
 	const char *textconv;
+	struct notes_cache *textconv_cache;
+	int textconv_want_cache;
 };
 
 int userdiff_config(const char *k, const char *v);