Confused about diff-tree --cc output

Confused about diff-tree --cc output

[Jay Soffian]

I'm a bit confused about the output of "git show 8554af" (i.e. git
diff-tree --cc 8554af) on the git tree.

This is a merge of a topic branch of two patches I submitted. If I
understand what Junio did, he branched js/maint-remote-remove-mirror
from maint, applied the two patches I sent via email, then merged that
to next. (Given the topic name he chose, after cooking for a while, I
guess it will eventually merge to maint.)

My understanding of "git show <merge commit>" is that it should only
show changes if the merge resulted in a conflict that needed to be
touched up (ignoring the possibility of an evil merge). Yet git show
on this commit shows this diff:

diff --cc builtin-remote.c
index abc8dd8,07cfdac..db18bcf
--- a/builtin-remote.c
+++ b/builtin-remote.c
@@@ -539,11 -362,14 +549,14 @@@ static int rm(int argc, const char **ar
  		OPT_END()
  	};
  	struct remote *remote;
 -	struct strbuf buf;
 +	struct strbuf buf = STRBUF_INIT;
  	struct known_remotes known_remotes = { NULL, NULL };
  	struct string_list branches = { NULL, 0, 0, 1 };
- 	struct branches_for_remote cb_data = { NULL, &branches, &known_remotes };
- 	int i;
+ 	struct string_list skipped = { NULL, 0, 0, 1 };
+ 	struct branches_for_remote cb_data = {
+ 		NULL, &branches, &skipped, &known_remotes
+ 	};
+ 	int i, result;

  	if (argc != 2)
  		usage_with_options(builtin_remote_usage, options);

This is the same output shown here:

  http://repo.or.cz/w/git.git?a=commitdiff;h=8554af

However, kernel.org's git shows something different (and more what I
would expect, if any output is to be shown):

  http://git.kernel.org/?p=git/git.git;a=commitdiff;h=8554af

(I guess kernel.org is using something like diff-tree -p 3613d1 8554af
--, but I'm not sure why.)

I don't see any other changes to builtin-remote.c that would have
caused a merge conflict, so, color me confused.

j.

Re: Confused about diff-tree --cc output

[Junio C Hamano]

Jay Soffian <jaysoffian@gmail.com> writes:

> My understanding of "git show <merge commit>" is that it should only
> show changes if the merge resulted in a conflict that needed to be
> touched up (ignoring the possibility of an evil merge). Yet git show
> on this commit shows this diff:

I think this is "more than two versions", in

    http://thread.gmane.org/gmane.comp.version-control.git/15486/focus=15491

Nothing new to see here, I think.

Re: Confused about diff-tree --cc output

[Jay Soffian]

On Wed, Feb 4, 2009 at 4:56 PM, Junio C Hamano <gitster@pobox.com> wrote:
> I think this is "more than two versions", in
>
>    http://thread.gmane.org/gmane.comp.version-control.git/15486/focus=15491
>
> Nothing new to see here, I think.

Okay, let me just make sure I understand. The output of git show has:

diff --cc builtin-remote.c
index abc8dd8,07cfdac..db18bcf

So, I can do:

git diff abc8dd8 db18bcf > parent1.diff
git diff 07cfdac db18bcf > parent2.diff

Then look at each for overlapping hunks. If I do so, I find that this
hunk in parent1.diff (I've annotated w/line numbers):

@@ -542,8 +552,11 @@ static int rm(int argc, const char **argv)
552 	struct strbuf buf = STRBUF_INIT;
553 	struct known_remotes known_remotes = { NULL, NULL };
554 	struct string_list branches = { NULL, 0, 0, 1 };
   -	struct branches_for_remote cb_data = { NULL, &branches, &known_remotes };
   -	int i;
555+	struct string_list skipped = { NULL, 0, 0, 1 };
556+	struct branches_for_remote cb_data = {
557+		NULL, &branches, &skipped, &known_remotes
558+	};
559+	int i, result;
560
561 	if (argc != 2)
562 		usage_with_options(builtin_remote_usage, options);

and this hunk in parent2.diff:

@@ -362,7 +549,7 @@ static int rm(int argc, const char **argv)
549 		OPT_END()
550 	};
551 	struct remote *remote;
   -	struct strbuf buf;
552+	struct strbuf buf = STRBUF_INIT;
553 	struct known_remotes known_remotes = { NULL, NULL };
554 	struct string_list branches = { NULL, 0, 0, 1 };
555 	struct string_list skipped = { NULL, 0, 0, 1 };

overlap (lines 552-555). Since the hunks overlap, and each introduces
a unique change, both hunks are shown in the combined format. I guess
this makes sense now that I know how it works, but my expectation had
been that I'd only see output if there was a merge conflict.

I also note that "git diff-tree -U2 --cc 8554af" produces no output,
and this is because there is no longer overlap (well, the context
lines overlap, but are identical, which I guess is good enough).

Is that about right?

(Aside, to reconstruct the merge manually I did the following, but I
bet there is an easier way:

git show 3613d1d:builtin-remote.c > mine
git show 441adf0:builtin-remote.c > other
git show $(git merge-base 3613d1d 441adf0):builtin-remote.c > base
git merge-file -p mine base other
)

j.

Re: Confused about diff-tree --cc output

[Junio C Hamano]

Jay Soffian <jaysoffian@gmail.com> writes:

> Is that about right?

Yes, not just about right but it is how it works.

The patch in:

    http://article.gmane.org/gmane.comp.version-control.git/15500

gave the final implementation of the rule.

The rule is that if there are only one or two variations to choose from,
and the result matches one of them, it is not interesting.  Otherwise it
is interesting.

In an octopus where more than two parents disagree, no matter what the
result is, it is interesting.

In an extreme case, even if both parents in a two-parent merge had the
same contents (i.e. only one to choose from), if you amend the merge
result to make it different from it, it becomes interesting.

In your example, two pieces, each of which took everything from one parent
(hence is uninteresting by itself), happened to lie within the context
range.  It is not very interesting, and it automerges cleanly, but it is
rare that these merges happen close together, and that "close miss" is
what makes this case "interesting".

It is exactly the same as the example in the message I referred you to.
In the same thread:

    http://article.gmane.org/gmane.comp.version-control.git/15598

has a very clear explanation about this exact kind of borderline "is it
really interesting?" case.

Also the follow-up to it:

    http://article.gmane.org/gmane.comp.version-control.git/15600

would give you further insights, the most interesting of which is that
what is shown in --cc output does *NOT* have anything to do with "did the
merge result in a manual conflict resolution?"

By the way, the latter message talks about gitk not showing the right
thing, but that was an ancient story.  These days gitk just shows what
diff --cc feeds it and there is no discrepancy between the two.

Re: Confused about diff-tree --cc output

[Jay Soffian]

On Thu, Feb 5, 2009 at 3:38 AM, Junio C Hamano <gitster@pobox.com> wrote:
> Jay Soffian <jaysoffian@gmail.com> writes:
>
>> Is that about right?
>
> Yes, not just about right but it is how it works.

Got it. I did read the thread you pointed me to earlier, but I had to
work through it w/my example to make sure I understood completely. I
appreciate your patience.

j.