From: Nico Williams <nico@cryptonector.com>
Date: Fri, 02 Oct 2026 03:19:48 GMT
Subject: Re: git-rebase-walk
Message-ID: <ar8i1Pz3Rh5F8ngx@ubby>
In-Reply-To: <f5859438-f91d-46d2-b80c-25d63937ed7c@hogyros.de>

On Fri, Oct 02, 2026 at 11:49:26AM +0900, Simon Richter wrote:
> On 10/1/26 8:58 PM, Alejandro Colomar wrote:
> > I use this little command to apply iterative rebases, which are easier
> > to handle when there are large conflicts.  Are you interested in it?
> 
> I use something similar:
> 
> [alias]
>         slowrebase = "!bash -c 'for i in $(git rev-list --reverse $(git
> merge-base HEAD @{u})..@{u}); do git rebase $i || break; done'"
>         slowrebasemerges = "!bash -c 'for i in $(git rev-list --merges
> --reverse $(git merge-base HEAD @{u})..@{u}); do git rebase $i || break;
> done'"

Nice!

> Alas, this breaks down with merges, and it is slow, so I've been thinking
> [...]

A slow-rebase or bisect-rebase works best when a) you follow a rebase
workflow, and b) the upstream has linear history (i.e., they also do a
rebase workflow).  When the upstream has merges then... improving this
experience gets difficult, and the easiest thing to do is to treat the
merge as a single [large] commit and not try to bisect-rebase on the
merge author's branch side.  But if you're looking for a slow-rebase or
a bisect-rebase then chances are you're doing (a), and if the upstream
doesn't have linear history then you accept the trouble.

> I wonder if it would make sense to have a rebase-bisect (bisect-rebase?)

We had a whole sub-thread on this thread about just that! :)

> command that finds the first commit that the branch cannot be cleanly
> rebased onto, optionally with a test command to see if there are semantic
> conflicts.
> 
> So given
> 
>     A --- B --- C --- D --- E --- F (main)
>       \
>         a --- b --- c (feature)
> 
> I'd like to be able to use "git bisect rebase main -x 'make check'" to
> attempt rebasing onto D first, and continue on to B or E, depending on
> whether the merge goes cleanly and "make check" succeeds, maybe with an
> option to try F first if the resolution is trivial.

I linked to my version of this, then Alejandro re-wrote it, on this
thread.

I've used mine a few times to rebase across thousands of upstream
commits.  It works very well, IMO.

Nico
--