That cleanup patch is good, but I've found a bug in it. In the item removal
code
> + /* p->path not in q->queue[]; drop it */
> + struct combine_diff_path *next = p->next;
> +
> + if ((*tail = next) != NULL)
> + tail = &next->next;
> + free(p);
> continue;
*tail = next
is right, but
tail = &next->next
is wrong, because it will lead to skipping the element after removed
one. Proof:
tail p
| |
| |
v v
+-+ +------+-+ +------+-+ +------+-+
| | | | | | | | | | |
|o------->| |o------->| |o------->| |o------->
|0| | 1| | | 2| | | 3| |
+-+ +------+-+ +------+-+ +------+-+
suppose, we are removing element 1. p->next points to 2, after
*tail = next
0 points to 2, which != NULL. After
tail = &next->next
tail points to &2->next.
On the next cycle iteration, after
p = *tail
we'll have
p = *2->next = 3
so 2 was skipped, oops.
I've actually hit the bug - when generating combine-diff, for merges with
should-be empty `log -c` output, every second path was present in the diff.
That, by the way, means we need to extend combine-diff tests somewhat.
Signed-off-by: Kirill Smelkov <kirr@xxxxxxxxxx>
---
combine-diff.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/combine-diff.c b/combine-diff.c
index 0809e79..a03147c 100644
--- a/combine-diff.c
+++ b/combine-diff.c
@@ -58,8 +58,7 @@ static struct combine_diff_path *intersect_paths(struct combine_diff_path *curr,
/* p->path not in q->queue[]; drop it */
struct combine_diff_path *next = p->next;
- if ((*tail = next) != NULL)
- tail = &next->next;
+ *tail = next;
free(p);
continue;
}
--
1.9.rc1.181.g641f458
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html