Re: [PATCH] Fix rebase -p --onto

From: Johannes Sixt <j.sixt@viscovery.net>
Date: Jul 17, 2009, 08:40 UTC
Message-ID: <4A6038E8.1090402@viscovery.net>
In-Reply-To: <7vk52767el.fsf@alter.siamese.dyndns.org>

Junio C Hamano schrieb:

Show 57 quoted lines

> Johannes Sixt <j.sixt@viscovery.net> writes:
> 
>> Greg Price schrieb:
>>> In a rebase with --onto, the correct test for whether we can skip
>>> rewriting a commit is if it is already on top of $ONTO, not $UPSTREAM.
>>> Without --onto, this distinction does not exist and the behavior does
>>> not change.
>>>
>>>
>>> In the situation
>>>
>>>  X---o---o---o---M
>>>   \             /
>>>    x---x---x---x
>>>
>>>  Y
>>> ...
>>> The command `git rebase -p --onto Y X M` moves only the
>>> first-parent chain, like so:
>>>
>>>  X
>>>   \
>>>    x---x---x---x
>>>                 \
>>>  Y---o'--o'--o'--M'
>>>
>>> because it mistakenly drops the other branch(es) x---x---x---x from
>>> the TODO file.  This tests and fixes this behavior.
>> I think the current behavior is by design. There is nothing to fix.
>>
>> The purpose of rebase -p is to leave non-first children alone and rebase
>> only the first child parenthood chain. It is not the purpose to reseat an
>> entire history DAG.
> 
> Hmm, if the original history were
> 
>  .---X---o---o---o---M
>   \                 /
>    x---x---x---x---x
> 
>      Y
> 
> and the rebase is about moving history leading to M since X on top of Y,
> I would actually have agreed that this
> 
>  .---X---o---o---o---M
>   \                 /
>    x---x---x---x---x
>                     \
>      Y---o'--o'--o'--M'
> 
> would be the right thing to do (IOW, I would agree with you).
> 
> Can the current code distinguish the two cases?  More generally, can we
> always tell these two cases apart, or is it fundamentally not possible to
> differentiate the two cases and we should simplify the problem space by
> limiting ourselves by treating the first parent in a special way?

I have used rebase -i -p in the past to rewrite history that involves merges of topic branches like this:

  ---------Y--M--M--F     <-- master
             /  /
  ----a--a--a  /
              /
  --b--b--b--b

where F is a fixup that I want to insert between Y and M, and I thought rebase -i -p was intended for this use-case.

With this in mind, I do not see why should we distinguish the cases. I would even go so far to say that this is OK:

  X---a---o---o---M            X---a
       \         /    ===>          \
        x---x---x                    x---x---x
                                              \
  Y                            Y---a'--o'--o'--M'