git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[RFC PATCH] rerere: fix overeager gc

From
SZEDER Gábor <szeder@ira.uka.de>
Date
Jun 29, 2010, 11:38 UTC
Message-ID
<1277811498-17288-1-git-send-email-szeder@ira.uka.de>

'rerere gc' prunes resolutions of conflicted merges that occurred long time ago, and when doing so it takes the creation time of the merge conflict resolution into account. This can cause the loss of frequently used merge resolutions (e.g. long-living topic branches are merged into a regularly rebuilt integration branch (think of git's pu)) when they become old enough to exceed 'rerere gc's threshold.

Prevent the loss of valuable merge resolutions by changing 'rerere gc' to take the time of last usage of the merge resolution into account when determining whether a merge resolution should be pruned.

Signed-off-by: SZEDER Gábor <szeder@ira.uka.de>
---

I was wondering that every once in a while when I got a merge conflict during rebuilding my integration branch then it was usually followed with a bunch of other conflicts as well, even though nothing really changed around the conflicting areas. Until today at last I noticed that it happens right after doing a 'git gc'...

RFC, because I would not say that I put in too much effort to fully understand how rerere works internally... As far as I observed rerere's behaviour and understood its code, thisimage is always rewritten each time a merge resolution is used. But I'm not sure I can rely on that when gc'ing.

 builtin/rerere.c  |   15 +++++++++++++--
 t/t4200-rerere.sh |   18 +++++++++++++++++-
 2 files changed, 30 insertions(+), 3 deletions(-)
diff --git a/builtin/rerere.c b/builtin/rerere.c
index 0048f9e..e095852 100644
--- a/builtin/rerere.c
+++ b/builtin/rerere.c
@@ -19,6 +19,12 @@ static time_t rerere_created_at(const char *name)
 	return stat(rerere_path(name, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
 }
 
+static time_t rerere_last_used_at(const char *name)
+{
+	struct stat st;
+	return stat(rerere_path(name, "thisimage"), &st) ? (time_t) 0 : st.st_mtime;
+}
+
 static void unlink_rr_item(const char *name)
 {
 	unlink(rerere_path(name, "thisimage"));
@@ -56,8 +62,13 @@ static void garbage_collect(struct string_list *rr)
 		then = rerere_created_at(e->d_name);
 		if (!then)
 			continue;
-		cutoff = (has_rerere_resolution(e->d_name)
-			  ? cutoff_resolve : cutoff_noresolve);
+		if (has_rerere_resolution(e->d_name)) {
+			then = rerere_last_used_at(e->d_name);
+			if (!then)
+				continue;
+			cutoff = cutoff_resolve;
+		} else
+			cutoff = cutoff_noresolve;
 		if (then < now - cutoff * 86400)
 			string_list_append(e->d_name, &to_remove);
 	}
diff --git a/t/t4200-rerere.sh b/t/t4200-rerere.sh
index 70856d0..45c9df8 100755
--- a/t/t4200-rerere.sh
+++ b/t/t4200-rerere.sh
@@ -154,33 +154,49 @@ test_expect_success 'clear removed the directory' "test ! -d $rr"
 mkdir $rr
 echo Hello > $rr/preimage
 echo World > $rr/postimage
+echo "!" > $rr/thisimage
 
 sha2=4000000000000000000000000000000000000000
 rr2=.git/rr-cache/$sha2
 mkdir $rr2
 echo Hello > $rr2/preimage
 
+sha3=ffffffffffffffffffffffffffffffffffffffff
+rr3=.git/rr-cache/$sha3
+mkdir $rr3
+echo Hello > $rr3/preimage
+echo World > $rr3/postimage
+echo "!" > $rr3/thisimage
+
 almost_15_days_ago=$((60-15*86400))
 just_over_15_days_ago=$((-1-15*86400))
 almost_60_days_ago=$((60-60*86400))
 just_over_60_days_ago=$((-1-60*86400))
 
 test-chmtime =$almost_60_days_ago $rr/preimage
+test-chmtime =$almost_60_days_ago $rr/thisimage
 test-chmtime =$almost_15_days_ago $rr2/preimage
+test-chmtime =$almost_60_days_ago $rr3/preimage
+test-chmtime =$almost_60_days_ago $rr3/thisimage
 
 test_expect_success 'garbage collection (part1)' 'git rerere gc'
 
 test_expect_success 'young records still live' \
-	"test -f $rr/preimage && test -f $rr2/preimage"
+	"test -f $rr/preimage && test -f $rr2/preimage && test -f $rr3/preimage"
 
 test-chmtime =$just_over_60_days_ago $rr/preimage
+test-chmtime =$just_over_60_days_ago $rr/thisimage
 test-chmtime =$just_over_15_days_ago $rr2/preimage
+test-chmtime =$just_over_60_days_ago $rr3/preimage
 
 test_expect_success 'garbage collection (part2)' 'git rerere gc'
 
 test_expect_success 'old records rest in peace' \
 	"test ! -f $rr/preimage && test ! -f $rr2/preimage"
 
+test_expect_success 'recently used records are still there' \
+	"test -f $rr3/preimage"
+
 test_expect_success 'file2 added differently in two branches' '
 	git reset --hard &&
 	git checkout -b fourth &&
-- 
1.7.2.rc0.42.g400d
Next: Junio C Hamano
Message 1 of 24 in “rerere: fix overeager gc”
  1. rerere: fix overeager gcSZEDER Gábor, Jun 29, 2010
  2. Junio C HamanoJun 29, 2010
  3. Johannes SixtJun 30, 2010
  4. SZEDER GáborJun 30, 2010
  5. Junio C HamanoJun 30, 2010
  6. Johannes SixtJun 30, 2010
  7. rerere: fix overeager gcSZEDER Gábor, Jul 1, 2010
  8. Johannes SixtJul 1, 2010
  9. 1/2 mingw: utime() handles NULL times parameterSZEDER Gábor, Jul 1, 2010
  10. 1/2 mingw_utime(): handle NULL times parameterJohannes Sixt, Jul 1, 2010
  11. 2/2 rerere: fix overeager gcSZEDER Gábor, Jul 1, 2010
  12. Junio C HamanoJul 1, 2010
  13. Johannes SixtJul 2, 2010
  14. Junio C HamanoJul 2, 2010
  15. Johannes SixtJul 5, 2010
  16. rerere: fix overeager gcSZEDER Gábor, Jul 8, 2010
  17. Junio C HamanoJul 9, 2010
  18. 1/2 mingw_utime(): handle NULL times parameterSZEDER Gábor, Jul 12, 2010
  19. 2/2 rerere: fix overeager gcSZEDER Gábor, Jul 12, 2010
  20. Junio C HamanoJul 13, 2010
  21. SZEDER GáborJul 14, 2010
  22. Junio C HamanoJul 14, 2010
  23. SZEDER GáborJul 14, 2010
  24. Johannes SixtJul 8, 2010

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.