From: Junio C Hamano <gitster@pobox.com>
Date: Wed, 30 Apr 2008 09:19:27 GMT
Subject: Re: [PATCH v2 04/13] Teach rebase interactive the mark command
Message-ID: <7vr6cn1yzk.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <alpine.DEB.1.00.0804300938190.17469@eeepc-johanness>

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

>> But instead you are thinking of letting me just say "X", and somehow 
>> make the machinery guess by noticing "Ah, original X is a merge between 
>> original A and B, and we have a merge between rewritten A and rewritten 
>> B, so we will treat that merge as rewritten "X"?
>> 
>> I actually was hoping we could avoid that, which feels messy.
> ...
> But this got me thinking, and I think that to leave out the first parent 
> was another mistake I made, so I really would like to have this syntax:
>
> 	merge <orig-commit> <parent1> <parent2>... <message>
>
> This would allow to change the parents in the interactive rebase, and if 
> <parent1> is not the current commit at that point, it would implicitly 
> call "reset".
>
> What appeals to me is the simplicity of this approach: you refer to the 
> commits by calling them by their (original) name.

Ok, that clears my confusion, but it raises another issue.

In the context of "rebase -i", this may not be a problem, but by forcing
us to name commits always with original commits, we cannot build (instead
of rebuild) a history that does not yet exist using the sequencer
machinery, can we?

If we want to transplant "^C ^N O" in this history elsewhere:

      --o---C---N
               / 
          B---M
         /   /
        O---A

while inserting a new fix-up commit F on top of B before we merge that
side branch to rewritten A:

              --o---C---N'
                       / 
              B'--X---M'
             /       /
        O---Q-------A'

Would/Should the machinery somehow figure out that the merge between the
rewritten A (which is A') and an inserted commit X (which is made on top
of the rewritten B) corresponds to M in the original history?

This is not a made up example, but something I have to do once (on my
non-git days) or many more times (on my git days) every day when
rebuilding 'pu' on top of updated 'next' using updated tips of topic
branches.  I was hoping that the sequencer mechanism can help me
automating the process a bit more.