[PATCH] line-log: fix crash when --first-parent is used

[PATCH] line-log: fix crash when --first-parent is used

[Tzvetan Mikov]

line-log tries to access all parents of a commit, but only the first
parent has been loaded if "--first-parent" is specified, resulting
in a crash.

Limit the number of parents to one if "--first-parent" is specified.

Reported-by: Eric N. Vander Weele <ericvw@gmail.com>
Signed-off-by: Tzvetan Mikov <tmikov@gmail.com>
---
 line-log.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/line-log.c b/line-log.c
index 1008e72..86e7274 100644
--- a/line-log.c
+++ b/line-log.c
@@ -1141,6 +1141,9 @@ static int process_ranges_merge_commit(struct rev_info *rev, struct commit *comm
 	int i;
 	int nparents = commit_list_count(commit->parents);
 
+	if (nparents > 1 && rev->first_parent_only)
+	    nparents = 1;
+
 	diffqueues = xmalloc(nparents * sizeof(*diffqueues));
 	cand = xmalloc(nparents * sizeof(*cand));
 	parents = xmalloc(nparents * sizeof(*parents));
-- 
1.9.1

Re: [PATCH] line-log: fix crash when --first-parent is used

[Junio C Hamano]

Tzvetan Mikov <tmikov@gmail.com> writes:

> line-log tries to access all parents of a commit, but only the first
> parent has been loaded if "--first-parent" is specified, resulting
> in a crash.
>
> Limit the number of parents to one if "--first-parent" is specified.
>
> Reported-by: Eric N. Vander Weele <ericvw@gmail.com>
> Signed-off-by: Tzvetan Mikov <tmikov@gmail.com>
> ---

Thomas, how does this one look?

Tzvetan, can we have a test for this one?

Thanks.

>  line-log.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/line-log.c b/line-log.c
> index 1008e72..86e7274 100644
> --- a/line-log.c
> +++ b/line-log.c
> @@ -1141,6 +1141,9 @@ static int process_ranges_merge_commit(struct rev_info *rev, struct commit *comm
>  	int i;
>  	int nparents = commit_list_count(commit->parents);
>  
> +	if (nparents > 1 && rev->first_parent_only)
> +	    nparents = 1;
> +
>  	diffqueues = xmalloc(nparents * sizeof(*diffqueues));
>  	cand = xmalloc(nparents * sizeof(*cand));
>  	parents = xmalloc(nparents * sizeof(*parents));

Re: [PATCH] line-log: fix crash when --first-parent is used

[Tzvetan Mikov]

On Mon, Nov 3, 2014 at 12:58 PM, Junio C Hamano <gitster@pobox.com> wrote:

>> line-log tries to access all parents of a commit, but only the first
>> parent has been loaded if "--first-parent" is specified, resulting
>> in a crash.
>>
>> Limit the number of parents to one if "--first-parent" is specified.
>>

> [...]
> Tzvetan, can we have a test for this one?

Here is a sequence of commands to reproduce the crash:

git init
echo "1" > a.txt && git add a.txt && git commit -m "a"
git checkout -b branch
echo "2" > b.txt && git add b.txt && git commit -m "b"
git checkout master
git merge branch --no-ff -m "merge"
git log --first-parent -L 1,1:a.txt

I am not sure whether you have a requirement for a formal test
included with the patch, but if there is one, I am happy to
rework it.

This is a very simple fix for the crash, though possibly
a "real" comprehensive fix may be needed. I am not sure.
The real cause is that with "--first-parent" some commit
objects are never loaded from disk but may still be accessed.

(sorry about my formatting - have to use GMail)

regards,
Tzvetan

Re: [PATCH] line-log: fix crash when --first-parent is used

[Michael J Gruber]

Tzvetan Mikov schrieb am 03.11.2014 um 23:09:
> On Mon, Nov 3, 2014 at 12:58 PM, Junio C Hamano <gitster@pobox.com> wrote:
> 
>>> line-log tries to access all parents of a commit, but only the first
>>> parent has been loaded if "--first-parent" is specified, resulting
>>> in a crash.
>>>
>>> Limit the number of parents to one if "--first-parent" is specified.
>>>
> 
>> [...]
>> Tzvetan, can we have a test for this one?
> 
> Here is a sequence of commands to reproduce the crash:
> 
> git init
> echo "1" > a.txt && git add a.txt && git commit -m "a"
> git checkout -b branch
> echo "2" > b.txt && git add b.txt && git commit -m "b"
> git checkout master
> git merge branch --no-ff -m "merge"
> git log --first-parent -L 1,1:a.txt
> 
> I am not sure whether you have a requirement for a formal test
> included with the patch, but if there is one, I am happy to
> rework it.
> 
> This is a very simple fix for the crash, though possibly
> a "real" comprehensive fix may be needed. I am not sure.
> The real cause is that with "--first-parent" some commit
> objects are never loaded from disk but may still be accessed.
> 
> (sorry about my formatting - have to use GMail)
> 
> regards,
> Tzvetan
> 

A test would be nice, yes. Look in t/t4211-line-log.sh - you should be
OK if you add a test at the end which does the following:

"git checkout parallel-change" so that you have commits with more than 1
parent in the history

test "git log -L1,1 b.c" (it core dumps with current git)

That is, you don't have to set up merges and history, you can use the
existing one.

Cheers,
Michael

[PATCH v2] line-log: fix crash when --first-parent is used

[Tzvetan Mikov]

line-log tries to access all parents of a commit, but only the first
parent has been loaded if "--first-parent" is specified, resulting
in a crash.

Limit the number of parents to one if "--first-parent" is specified.

Reported-by: Eric N. Vander Weele <ericvw@gmail.com>
Signed-off-by: Tzvetan Mikov <tmikov@gmail.com>
---
 PATCH v2: Add a test case (thanks Michael J Gruber)

 line-log.c          | 3 +++
 t/t4211-line-log.sh | 5 +++++
 2 files changed, 8 insertions(+)

diff --git a/line-log.c b/line-log.c
index 1008e72..86e7274 100644
--- a/line-log.c
+++ b/line-log.c
@@ -1141,6 +1141,9 @@ static int process_ranges_merge_commit(struct rev_info *rev, struct commit *comm
 	int i;
 	int nparents = commit_list_count(commit->parents);
 
+	if (nparents > 1 && rev->first_parent_only)
+	    nparents = 1;
+
 	diffqueues = xmalloc(nparents * sizeof(*diffqueues));
 	cand = xmalloc(nparents * sizeof(*cand));
 	parents = xmalloc(nparents * sizeof(*parents));
diff --git a/t/t4211-line-log.sh b/t/t4211-line-log.sh
index 7369d3c..0901b30 100755
--- a/t/t4211-line-log.sh
+++ b/t/t4211-line-log.sh
@@ -94,4 +94,9 @@ test_expect_success '-L ,Y (Y == nlines + 2)' '
 	test_must_fail git log -L ,$n:b.c
 '
 
+test_expect_success '-L with --first-parent and a merge' '
+	git checkout parallel-change &&
+	git log --first-parent -L 1,1:b.c
+'
+
 test_done
-- 
1.9.1

Re: [PATCH v2] line-log: fix crash when --first-parent is used

[Junio C Hamano]

Tzvetan Mikov <tmikov@gmail.com> writes:

> line-log tries to access all parents of a commit, but only the first
> parent has been loaded if "--first-parent" is specified, resulting
> in a crash.
>
> Limit the number of parents to one if "--first-parent" is specified.
>
> Reported-by: Eric N. Vander Weele <ericvw@gmail.com>
> Signed-off-by: Tzvetan Mikov <tmikov@gmail.com>
> ---
>  PATCH v2: Add a test case (thanks Michael J Gruber)

Thanks, both.  The patch looks good (modulo the indentation of
nparents assignment, which I'll locally fix up).

Will queue.

>  line-log.c          | 3 +++
>  t/t4211-line-log.sh | 5 +++++
>  2 files changed, 8 insertions(+)
>
> diff --git a/line-log.c b/line-log.c
> index 1008e72..86e7274 100644
> --- a/line-log.c
> +++ b/line-log.c
> @@ -1141,6 +1141,9 @@ static int process_ranges_merge_commit(struct rev_info *rev, struct commit *comm
>  	int i;
>  	int nparents = commit_list_count(commit->parents);
>  
> +	if (nparents > 1 && rev->first_parent_only)
> +	    nparents = 1;
> +
>  	diffqueues = xmalloc(nparents * sizeof(*diffqueues));
>  	cand = xmalloc(nparents * sizeof(*cand));
>  	parents = xmalloc(nparents * sizeof(*parents));
> diff --git a/t/t4211-line-log.sh b/t/t4211-line-log.sh
> index 7369d3c..0901b30 100755
> --- a/t/t4211-line-log.sh
> +++ b/t/t4211-line-log.sh
> @@ -94,4 +94,9 @@ test_expect_success '-L ,Y (Y == nlines + 2)' '
>  	test_must_fail git log -L ,$n:b.c
>  '
>  
> +test_expect_success '-L with --first-parent and a merge' '
> +	git checkout parallel-change &&
> +	git log --first-parent -L 1,1:b.c
> +'
> +
>  test_done

Re: [PATCH v2] line-log: fix crash when --first-parent is used

[Tzvetan Mikov]

On Tue, Nov 4, 2014 at 1:24 PM, Junio C Hamano <gitster@pobox.com> wrote:
> Thanks, both.  The patch looks good (modulo the indentation of
> nparents assignment, which I'll locally fix up).
>
> Will queue.

Awesome, thanks!
(I can't believe I missed that tab. Well, at least one of my two
lines was correct :-) )

regards,
Tzvetan