[PATCH v6 5/5] pack-redundant: consistent sort method
To
Junio C Hamano
Git List
SZEDER Gábor
Cc
Jiang Xin
Sun Chao
Jiang Xin
From
Jiang Xin
See Also
Prev
Date
2019-01-12 09:17:54 UTC
From: Jiang Xin <zhiyou.jx@alibaba-inc.com>

SZEDER reported that test case t5323 has different test result on MacOS.
This is because `cmp_pack_list_reverse` cannot give identical result
when two pack being sorted has the same size of remaining_objects.

Changes to the sorting function will make consistent test result for
t5323.

The new algorithm to find redundant packs is a trade-off to save memory
resources, and the result of it may be different with old one, and may
be not the best result sometimes.  Update t5323 for the new algorithm.

Reported-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Jiang Xin <zhiyou.jx@alibaba-inc.com>
---
 builtin/pack-redundant.c  | 24 ++++++++++++++++--------
 t/t5323-pack-redundant.sh |  2 +-
 2 files changed, 17 insertions(+), 9 deletions(-)

diff --git a/builtin/pack-redundant.c b/builtin/pack-redundant.c
index 64eec3e297..4448e58a10 100644
--- a/builtin/pack-redundant.c
+++ b/builtin/pack-redundant.c
@@ -33,6 +33,7 @@ static struct pack_list {
 	struct packed_git *pack;
 	struct llist *unique_objects;
 	struct llist *remaining_objects;
+	size_t all_objects_size;
 } *local_packs = NULL, *altodb_packs = NULL;
 
 static struct llist_item *free_nodes;
@@ -335,19 +336,25 @@ static inline off_t pack_set_bytecount(struct pack_list *pl)
 	return ret;
 }
 
-static int cmp_pack_list_reverse(const void *a, const void *b)
+static int cmp_remaining_objects(const void *a, const void *b)
 {
 	struct pack_list *pl_a = *((struct pack_list **)a);
 	struct pack_list *pl_b = *((struct pack_list **)b);
-	size_t sz_a = pl_a->remaining_objects->size;
-	size_t sz_b = pl_b->remaining_objects->size;
 
-	if (sz_a == sz_b)
-		return 0;
-	else if (sz_a < sz_b)
+	if (pl_a->remaining_objects->size == pl_b->remaining_objects->size) {
+		/* have the same remaining_objects, big pack first */
+		if (pl_a->all_objects_size == pl_b->all_objects_size)
+			return 0;
+		else if (pl_a->all_objects_size < pl_b->all_objects_size)
+			return 1;
+		else
+			return -1;
+	} else if (pl_a->remaining_objects->size < pl_b->remaining_objects->size) {
+		/* sort by remaining objects, more objects first */
 		return 1;
-	else
+	} else {
 		return -1;
+	}
 }
 
 /* Sort pack_list, greater size of remaining_objects first */
@@ -365,7 +372,7 @@ static void sort_pack_list(struct pack_list **pl)
 	for (n = 0, p = *pl; p; p = p->next)
 		ary[n++] = p;
 
-	QSORT(ary, n, cmp_pack_list_reverse);
+	QSORT(ary, n, cmp_remaining_objects);
 
 	/* link them back again */
 	for (i = 0; i < n - 1; i++)
@@ -507,6 +514,7 @@ static struct pack_list * add_pack(struct packed_git *p)
 		llist_insert_back(l.remaining_objects, (const struct object_id *)(base + off));
 		off += step;
 	}
+	l.all_objects_size = l.remaining_objects->size;
 	/* this list will be pruned in cmp_two_packs later */
 	l.unique_objects = llist_copy(l.remaining_objects);
 	if (p->pack_local)
diff --git a/t/t5323-pack-redundant.sh b/t/t5323-pack-redundant.sh
index 407838f0e8..663328ab30 100755
--- a/t/t5323-pack-redundant.sh
+++ b/t/t5323-pack-redundant.sh
@@ -133,7 +133,7 @@ test_expect_success 'create pack 4, 5' '
 '
 
 cat >expected <<EOF
-P2:$P2
+P3:$P3
 EOF
 
 test_expect_success 'one of pack-2/pack-3 is redundant' '
-- 
2.20.0.3.gc45e608566