Re: [RFC] git-rebase-rewind, nested rebases, remembering stgit
Show 48 quoted lines
> De: "Christian Couder" <christian.couder@gmail.com>
> À: "Yann Dirson" <ydirson@free.fr>
> Cc: "git" <git@vger.kernel.org>
> Envoyé: Lundi 22 Mars 2021 09:49:23
> Objet: Re: [RFC] git-rebase-rewind, nested rebases, remembering stgit
>
> Hi Yann,
>
> Nice to hear from you on the list!
>
> On Sat, Mar 13, 2021 at 5:45 PM <ydirson@free.fr> wrote:
> >
> > Hello there,
> >
> > I often find myself doing iterative refactorings, which can lead to
> > long branches, and while rebasing to edit HEAD~10 realize that I
> > first
> > need to edit HEAD~20 or to add more commits below that stack.
> >
> > If you've used stgit when it was a thing, you probably see how it
> > helped doing that. While git-rebase has grown to do much more than
> > stgit in most areas, this is still one area where with a pain point
> > for me.
> >
> > Here is a small git-rebase-rewind script I've been using for a few
> > weeks,
> > starting with my most common use-case: automate worklow "edit
> > git-rebase-todo
> > to prepend 'pick' commands for the N previous commits, then reset
> > --hard HEAD~N".
> >
> > As you will see from the new needs revealed by using this script
> > (see in the
> > script header), I believe it would be valuable to integrate such a
> > mechanism
> > directly into git-rebase. Notably, "git rebase -i" itself can be
> > seen as a
> > form of rewind, and this rewind feature would benefit from all the
> > interactive
> > rebase work.
> >
> > Does that sound like reasonable premises ?
>
> Sorry for the late answer. It sounds reasonable to me.
>
> It looks to me like a way to restart the whole interactive rebase
> process though, so I wonder if calling it "--restart" would be better
> than "--rewind".
Right, it is indeed like doing another interactive rebase while the current one is not finished. Maybe just running "rebase -i" without more option could be sufficient ? That could break the expectation when someone relies on rebase refusing to proceed while already rebasing, though it could be mitigated if a "rebase --abort" just aborts the most recent rebase.
There are quite a few UI issues to be thought through around here :)