Strange behaviour of git diff branch1 ... branch2

Strange behaviour of git diff branch1 ... branch2

[Aaron Schrab]

I came across this odd question on stackoverflow:
	http://stackoverflow.com/q/13092854/1507392


If git diff is run with "..." as a separate argument between two 
commit-ish arguments causes it to produce strange output.  The 
differences seem to be the same as if "..." was left out, but change 
lines begin with 4 + or - characters rather than just 1.

Can anybody explain what is happening here?  I don't have any reason to 
want to use that form myself, but I'm very curious about why it produces 
this odd output.

Thanks.

Re: Strange behaviour of git diff branch1 ... branch2

[Nguyen Thai Ngoc Duy]

On Sat, Oct 27, 2012 at 4:26 AM, Aaron Schrab <aaron@schrab.com> wrote:
> I came across this odd question on stackoverflow:
>         http://stackoverflow.com/q/13092854/1507392
>
>
> If git diff is run with "..." as a separate argument between two commit-ish
> arguments causes it to produce strange output.  The differences seem to be
> the same as if "..." was left out, but change lines begin with 4 + or -
> characters rather than just 1.
>
> Can anybody explain what is happening here?  I don't have any reason to want
> to use that form myself, but I'm very curious about why it produces this odd
> output.

I'm curious too. It shows this to me

diff --cc .gitignore
index a188a82,a188a82,a188a82,a188a82..d4473d8
--- a/.gitignore
+++ b/.gitignore
@@@@@ -2,9 -2,9 -2,9 -2,9 +2,6 @@@@@
    /GIT-CFLAGS
    /GIT-LDFLAGS
    /GIT-GUI-VARS
----/GIT-PREFIX
----/GIT-SCRIPT-DEFINES
----/GIT-USER-AGENT
    /GIT-VERSION-FILE
    /bin-wrappers/
    /git

Notice the --cc in the first line, which is combined diff. Usually
combined-diff is between two points and one parent. Though somehow git
passes 4 parents down combined-diff.c:show_combined_header, as you can
see in the "index" line. I think we should fix rev parsing code as it
does not make sense to pass 4 identical parents this way.
-- 
Duy

Re: Strange behaviour of git diff branch1 ... branch2

[Junio C Hamano]

Nguyen Thai Ngoc Duy <pclouds@gmail.com> wrote:
>
>Notice the --cc in the first line, which is combined diff. Usually
>combined-diff is between two points and one parent. Though somehow git
>passes 4 parents down combined-diff.c:show_combined_header, as you can
>see in the "index" line. I think we should fix rev parsing code as it
>does not make sense to pass 4 identical parents this way.

The two heads home from HEAD...HEAD the user has on the command line.

The user is getting exactly what she asked; there is nothing to fix.

Re: Strange behaviour of git diff branch1 ... branch2

[Nguyen Thai Ngoc Duy]

On Sat, Oct 27, 2012 at 7:33 PM, Junio C Hamano <gitster@pobox.com> wrote:
>
>
> Nguyen Thai Ngoc Duy <pclouds@gmail.com> wrote:
>>
>>Notice the --cc in the first line, which is combined diff. Usually
>>combined-diff is between two points and one parent. Though somehow git
>>passes 4 parents down combined-diff.c:show_combined_header, as you can
>>see in the "index" line. I think we should fix rev parsing code as it
>>does not make sense to pass 4 identical parents this way.
>
> The two heads home from HEAD...HEAD the user has on the command line.
>
> The user is getting exactly what she asked; there is nothing to fix.

Is there any use case where HEAD...HEAD (or "..." alone) is actually useful?

I have re-read the git-diff man page and I don't think it explains
"git diff foo ... bar" syntax (from a user's point of view, not a git
guru's). We could improve the documentation if "git diff foo ... bar"
is useful, or reject it with an error to avoid confusion.
-- 
Duy

Re: Strange behaviour of git diff branch1 ... branch2

[Junio C Hamano]

Nguyen Thai Ngoc Duy <pclouds@gmail.com> writes:

> On Sat, Oct 27, 2012 at 7:33 PM, Junio C Hamano <gitster@pobox.com> wrote:
>>
>> Nguyen Thai Ngoc Duy <pclouds@gmail.com> wrote:
>>>
>>>Notice the --cc in the first line, which is combined diff. Usually
>>>combined-diff is between two points and one parent. Though somehow git
>>>passes 4 parents down combined-diff.c:show_combined_header, as you can
>>>see in the "index" line. I think we should fix rev parsing code as it
>>>does not make sense to pass 4 identical parents this way.
>>
>> The two heads home from HEAD...HEAD the user has on the command line.
>>
>> The user is getting exactly what she asked; there is nothing to fix.
>
> Is there any use case where HEAD...HEAD (or "..." alone) is actually useful?

That is an invalid question, implying a proposal that will introduce
more confusion to the users while hurting the code.

There is no use case where "cat A >>A" is actually useful, either,
but it does not mean "cat A >>A" should be special cased to do
something different from what it happens to do in line with the case
where "cat A >>B" does.  I think "diff ..." or "diff HEAD...HEAD"
are the same as that example.