Re: Rebasing evil merges with --rebase-merges

From: Sergey Organov <sorganov@gmail.com>
Date: Jan 16, 2020, 07:42 UTC
Message-ID: <87h80vg849.fsf@osv.gnss.ru>
In-Reply-To: <9355c545-08a5-ef63-f7bf-65201d50acc8@gmail.com>

Igor Djordjevic <igor.d.djordjevic@gmail.com> writes:

Show 24 quoted lines

> On 15/01/2020 19:14, Junio C Hamano wrote:
>> 
>> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:
>> >
>> > Having said that, if you ever find yourself wanting Just One Feature 
>> > in `--rebase-merges` that would make it worthwhile for you to think 
>> > about switching your patch-based workflow to a `rebase -ir`-based 
>> > one, please let me know, and I will try my best to accommodate.
>> 
>> Another thing I noticed was that we may want to attempt to recreate
>> an evil merge and then stop to ask confirmation.  The "rebase -ri" I
>> did to sanity-check my revert for example failed to bring in the
>> change made in the existing evil merge when trying to recreate the
>> merge of the dl/merge-autostash topic into master..pu chain and
>> silently created a fails-to-build-from-the-source tree instead.
>
> FYI (and anyone interested), it`s something we actually brought up 
> some two years ago, at the time of introducing `--rebase-merges` 
> (known as `--recreate-merges` back at the time), see[1].
>
> It ended being a lengthy and heated discussion (inside a few 
> different topics as well, like original RFC[2] and it`s v2 update[3]), 
> myself being guilty for dropping out eventually and not following it 
> through, though, life taking me in another direction at the moment...

For reference, there is a nice summary in "Git Rev News Edition 38":

https://git.github.io/rev_news/2018/04/18/edition-38

Show 5 quoted lines

> but I still find this functionality to be very useful, not to say 
> essential, even, for reliable complex merge _rebasing_ (meaning 
> keeping "evil merge" changes, too), and not just merge _recreating_ 
> (loosing "evil merge" changes, and worse - doing it silently, as you 
> experienced yourself now as well).

Yeah, dropping user content silently and by default is still the most weird thing for git to do, be it a merge or not a merge.

As an additional note, I came to conclusion that there is actually no such thing as "evil merge" that is somehow different from "evil commit" in general (a commit containing unrelated changes).

Then, as "evil commit" belongs to user domain, we need to finally realize that "evil merge" belongs entirely to user domain as well, and thus, as it's out of git domain, we should stop using the term "evil merge" to excuse any kinds of weird git behaviors.

Regards, Sergey