threads / patch / 33101

patchFix revision walk for commits with the same dates

Subject: [PATCH] Fix revision walk for commits with the same dates

## tl;dr

5 messages between Mar 7, 2013 and Mar 24, 2013. Diffs are folded; open one to read it.

replies: 4people: 3as markdown or json

Kacper Kornet· Mar 7, 2013, 18:03 UTC · lore

git rev-list A^! --not B provides wrong answer if all commits in the range A..B had the same commit times and there are more then 8 of them. This commits fixes the logic in still_interesting function to prevent this error.

Signed-off-by: Kacper Kornet <draenog@pld-linux.org>
---
 revision.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Show changes to revision.c +1 −1

diff --git a/revision.c b/revision.c
index ef60205..cf620c6 100644
--- a/revision.c
+++ b/revision.c
@@ -709,7 +709,7 @@ static int still_interesting(struct commit_list *src, unsigned long date, int sl
 	 * Does the destination list contain entries with a date
 	 * before the source list? Definitely _not_ done.
 	 */
-	if (date < src->item->date)
+	if (date <= src->item->date)
 		return SLOP;
 
 	/*

-- 
1.8.2.rc2

Kacper Kornet· Mar 22, 2013, 18:38 UTC · re: Kacper Kornet · lore

[PATCH v2] Fix revision walk for commits with the same dates

Logic in still_interesting function allows to stop the commits traversing if the oldest processed commit is not older then the youngest commit on the list to process and the list contains only commits marked as not interesting ones. It can be premature when dealing with a set of coequal commits. For example git rev-list A^! --not B provides wrong answer if all commits in the range A..B had the same commit time and there are more then 7 of them.

To fix this problem the relevant part of the logic in still_interesting is changed to: the walk can be stopped if the oldest processed commit is younger then the youngest commit on the list to processed.

Signed-off-by: Kacper Kornet <draenog@pld-linux.org>
---

I don't know whether the first version was overlooked or deemed as not worthy. So just in case I resend it. Changes since the first version:

1. The test has been added
2. The commit log has been rewritten

 revision.c                 |  2 +-
 t/t6009-rev-list-parent.sh | 13 +++++++++++++
 2 files changed, 14 insertions(+), 1 deletion(-)

Show changes to 2 files +14 −1

revision.c, t/t6009-rev-list-parent.sh

diff --git a/revision.c b/revision.c
index ef60205..cf620c6 100644
--- a/revision.c
+++ b/revision.c
@@ -709,7 +709,7 @@ static int still_interesting(struct commit_list *src, unsigned long date, int sl
 	 * Does the destination list contain entries with a date
 	 * before the source list? Definitely _not_ done.
 	 */
-	if (date < src->item->date)
+	if (date <= src->item->date)
 		return SLOP;
 
 	/*
diff --git a/t/t6009-rev-list-parent.sh b/t/t6009-rev-list-parent.sh
index 3050740..66cda17 100755
--- a/t/t6009-rev-list-parent.sh
+++ b/t/t6009-rev-list-parent.sh
@@ -133,4 +133,17 @@ test_expect_success 'dodecapus' '
 	check_revlist "--min-parents=13" &&
 	check_revlist "--min-parents=4 --max-parents=11" tetrapus
 '
+
+test_expect_success 'ancestors with the same commit time' '
+
+	test_tick_keep=$test_tick &&
+	for i in 1 2 3 4 5 6 7 8; do
+		test_tick=$test_tick_keep
+		test_commit t$i
+	done &&
+	git rev-list t1^! --not t$i >result &&
+	>expect &&
+	test_cmp expect result
+'
+
 test_done

-- 
1.8.2

-- 
  Kacper Kornet

Junio C Hamano· Mar 22, 2013, 20:45 UTC · re: Kacper Kornet · lore

Re: [PATCH v2] Fix revision walk for commits with the same dates

Kacper Kornet <draenog@pld-linux.org> writes:

Show 11 quoted lines

> Logic in still_interesting function allows to stop the commits
> traversing if the oldest processed commit is not older then the
> youngest commit on the list to process and the list contains only
> commits marked as not interesting ones. It can be premature when dealing
> with a set of coequal commits. For example git rev-list A^! --not B
> provides wrong answer if all commits in the range A..B had the same
> commit time and there are more then 7 of them.
>
> To fix this problem the relevant part of the logic in still_interesting
> is changed to: the walk can be stopped if the oldest processed commit is
> younger then the youngest commit on the list to processed.

Is the made-up test case to freeze the clock even interesting? The slop logic is merely a heuristic to compensate for effects caused by skewed or non-monototic clocks, so in a different repository you may even need to fuzz the timestamp comparison further

	if (date - 10 < src->item->date)

or something silly like that.

Show 51 quoted lines

> Signed-off-by: Kacper Kornet <draenog@pld-linux.org>
> ---
>
> I don't know whether the first version was overlooked or deemed as not
> worthy. So just in case I resend it. Changes since the first version:
>
> 1. The test has been added
> 2. The commit log has been rewritten
>
>
>  revision.c                 |  2 +-
>  t/t6009-rev-list-parent.sh | 13 +++++++++++++
>  2 files changed, 14 insertions(+), 1 deletion(-)
>
> diff --git a/revision.c b/revision.c
> index ef60205..cf620c6 100644
> --- a/revision.c
> +++ b/revision.c
> @@ -709,7 +709,7 @@ static int still_interesting(struct commit_list *src, unsigned long date, int sl
>  	 * Does the destination list contain entries with a date
>  	 * before the source list? Definitely _not_ done.
>  	 */
> -	if (date < src->item->date)
> +	if (date <= src->item->date)
>  		return SLOP;
>  
>  	/*
> diff --git a/t/t6009-rev-list-parent.sh b/t/t6009-rev-list-parent.sh
> index 3050740..66cda17 100755
> --- a/t/t6009-rev-list-parent.sh
> +++ b/t/t6009-rev-list-parent.sh
> @@ -133,4 +133,17 @@ test_expect_success 'dodecapus' '
>  	check_revlist "--min-parents=13" &&
>  	check_revlist "--min-parents=4 --max-parents=11" tetrapus
>  '
> +
> +test_expect_success 'ancestors with the same commit time' '
> +
> +	test_tick_keep=$test_tick &&
> +	for i in 1 2 3 4 5 6 7 8; do
> +		test_tick=$test_tick_keep
> +		test_commit t$i
> +	done &&
> +	git rev-list t1^! --not t$i >result &&
> +	>expect &&
> +	test_cmp expect result
> +'
> +
>  test_done
> -- 
> 1.8.2

Kacper Kornet· Mar 22, 2013, 21:07 UTC · re: Junio C Hamano · lore

Re: [PATCH v2] Fix revision walk for commits with the same dates

On Fri, Mar 22, 2013 at 01:45:47PM -0700, Junio C Hamano wrote:

> Kacper Kornet <draenog@pld-linux.org> writes:

Show 7 quoted lines

> > Logic in still_interesting function allows to stop the commits
> > traversing if the oldest processed commit is not older then the
> > youngest commit on the list to process and the list contains only
> > commits marked as not interesting ones. It can be premature when dealing
> > with a set of coequal commits. For example git rev-list A^! --not B
> > provides wrong answer if all commits in the range A..B had the same
> > commit time and there are more then 7 of them.

> > To fix this problem the relevant part of the logic in still_interesting
> > is changed to: the walk can be stopped if the oldest processed commit is
> > younger then the youngest commit on the list to processed.

> Is the made-up test case to freeze the clock even interesting?  The
> slop logic is merely a heuristic to compensate for effects caused by
> skewed or non-monototic clocks, so in a different repository you may
> even need to fuzz the timestamp comparison further

> 	if (date - 10 < src->item->date)

> or something silly like that.

I don't think it is a made-up test case. For example it is easy to get a number of coequal commits by using git rebase -i. So I argue that git should treat correctly ranges of such commits.

-- 
  Kacper Kornet

Eric Sunshine· Mar 24, 2013, 02:18 UTC · re: Kacper Kornet · lore

Re: [PATCH v2] Fix revision walk for commits with the same dates

On Fri, Mar 22, 2013 at 2:38 PM, Kacper Kornet <draenog@pld-linux.org> wrote:

> Logic in still_interesting function allows to stop the commits
> traversing if the oldest processed commit is not older then the

s/then/than/

Show 5 quoted lines

> youngest commit on the list to process and the list contains only
> commits marked as not interesting ones. It can be premature when dealing
> with a set of coequal commits. For example git rev-list A^! --not B
> provides wrong answer if all commits in the range A..B had the same
> commit time and there are more then 7 of them.