From: Paolo Bonzini <bonzini@gnu.org>
Date: Fri, 16 Apr 2010 21:52:15 GMT
Subject: Re: [PATCH] patch-id: Add support for mbox format
Message-ID: <4BC8DC0F.5010007@gnu.org>
In-Reply-To: <7vochjay8a.fsf@alter.siamese.dyndns.org>

On 04/16/2010 11:23 PM, Junio C Hamano wrote:
>>>>   I have an alias that takes two arguments and compares their patch IDs.
>>>>   I would like to use to make sure I've tested exactly what I submit
>>>>   (patch by patch), like
>>>>
>>>>       git patch-cmp origin/master.. file-being-sent
>>>> ...
>>>
>>> I somehow have a feeling that this is solving a wrong problem.
>>
>> In what sense?
>
> Why does file-being-sent have anything from origin/master to begin with?
> Perhaps the --ignore-if-in-upstream mechanism needs to be updated so that
> you won't have duplicates that patch-id could easily find in the first
> place?

I think we're speaking about different things, --ignore-if-in-upstream 
doesn't have anything to do with this.  The git patch-cmp alias is just 
checking that

   git show origin/master.. | git patch-id | tac | awk '{print $1}'

and

   git patch-id < file-being-sent | awk '{print $1}'

produce the exact same output.

I use this when I had to edit the file-being-sent, e.g. to add cover 
letters or an introduction to a patch series.  Since some time passes 
between format-patch and send-email, I want to test that the file I'm 
sending is exactly what I have in the repository, and that I'm not 
submitting the wrong series.

Alternatively, I could apply file-being-sent to a detached HEAD and 
compare the trees, like

   topic=`git rev-parse HEAD`
   git checkout origin/master
   git am file-being-sent || echo bad
   mine=HEAD
   for i in `git rev-list origin/master..$topic`; do
     git diff-tree $i $mine || echo bad
     mine=${mine}^
   done

However, comparing the patch-id is faster and perfect for this job. 
What were your intended uses for the git-patch-id utility?

Paolo