Re: [PATCH/RFC] Allow empty commits during rebase -i

From: Pete Harlan <pgit@pcharlan.com>
Date: Jan 18, 2010, 02:11 UTC
Message-ID: <4B53C355.1010109@pcharlan.com>
In-Reply-To: <7vljfww686.fsf@alter.siamese.dyndns.org>

On 01/17/2010 05:29 PM, Junio C Hamano wrote:

Show 26 quoted lines

> Pete Harlan <pgit@pcharlan.com> writes:
> 
>> I imagine an ideal version of this fix would make it so the use case I
>> presented here would work, but rebase -i would still prevent
>> introducing a new empty commit, or at least warn when it was
>> introducing one.  In the absence of that ideal fix, I think this
>> behavior is better than failing to handle this case.
> 
> Sorry, I actually tend to think that in the absense of that fix, your
> version introduces risky behaviour that only a corner-case use case
> benefits, and pros-and-cons doesn't look attractive enough.
> 
> Why not do something like:
> 
>     pick X a crap tree with a good message
>     pick Y revert X
>     pick Z a good tree with a crap message
> 
> -->
> 
>     # drop X
>     # drop Y
>     edit Z
> 
> and then run "git commit --amend -C X" when it is Z's turn to be
> processed?

That is another way to accomplish the same thing, but doesn't prevent the current behavior from being confusing.

Part of the problem is that with the current behavior the user is sent to the command line with:

  # Not currently on any branch.
  nothing to commit (working directory clean)

  Could not apply a0b17c5... Revert "Crap tree good message"

with HEAD pointed to X^. Unsure of how to proceed from here, I --aborted the rebase and copy/pasted the commit message I wanted and resolved to track this down and fix it when I got a chance.

As it happens, "git rebase --continue" does exactly what I would have wanted to happen, including putting me in an editor with all three commit messages and succeeding when I exit the editor. But without a better message from git I don't expect a user to discover that. And, when rebase can continue just by being told so it would be nice if it didn't require that user intervention.

If the introduction of empty commits that the user has asked for (perhaps inadvertently) is considered too undesirable, then perhaps my fix is too simple. I'll think about how to do something more sophisticated.

Thanks for your feedback,

--Pete