Re: [PATCH/RFC] git-post: the opposite of git-cherry-pick

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Oct 13, 2017, 10:51 UTC
Message-ID: <87wp3zs4la.fsf@evledraar.booking.com>
In-Reply-To: <3d362037-3eb6-83db-a17f-47a984135580@kdbg.org>

On Thu, Oct 05 2017, Johannes Sixt jotted:

Show 8 quoted lines

> Am 05.10.2017 um 21:33 schrieb Stefan Beller:
>> On Thu, Oct 5, 2017 at 12:13 PM, Johannes Sixt <j6t@kdbg.org> wrote:
>>> +git-post(1)
>>> +===========
>>> +
>>> +NAME
>>> +----
>>> +git-post - Apply a commit on top of a branch that is not checked out

It would be great to have this somehow, whatever the UI ends up being.

Show 13 quoted lines

>> Contrasted to:
>>    git-cherry-pick - Apply the changes introduced by some existing commits
>> Some questions on top of my head:
>> * Do we want to emphasize the cherry-pick to say checked out?
>
> Maybe. Maybe "cherry picking" is sufficiently clear that it is not needed.
>
>> * Is it a good design choice to have a different command, just because the
>>    target branch is [not] checked out?
>
> I would not want to make it a sub-mode of cherry-pick, if that is what
> you mean, because "cherry picking" is about getting something, not
> giving something away.

It occurs to me that a better long-term UI design might be to make git-{cherry-pick,pick) some sort of submodes for git-commit.

Right now git-commit picks the current index for committing, but "use a patch as the source instead" could be a submode.

Right now it commits that change on top of your checked out commit, but "other non-checked-out target commit" could be a mode instead, i.e. exposing more of the power of the underlying git-commit-tree.

Just food for thought.

Show 12 quoted lines

>> * Naming: I just searched for synonyms "opposite of cherry-picking" and
>>    came up with cherry-put, cherry-post, target-commit
>
> With command line completion, we have to type 'git cherr<TAB>-<TAB>'
> currently. If we add another command that begins with 'cherry-',
> another <TAB> is needed. Therefore, I do not want to use a name
> starting with 'cherry-'.
>
>> * What was the rationale for naming it "post" (it sounds very generic to me)?
>
> Yes, it is generic, but I did not find a better word that means "give
> away" a commit. Perhaps "tack"?

[Not entirely serious]. Well if cherry-picking is taking a thing and eating it here, maybe git-cherry-puke takes an already digested thing and "throws" it elsewhere ?:)

It's a silly name but it's somewhat symmetric :)

Show 49 quoted lines

>> * Does it play well with worktrees? (i.e. if a worktree has a branch checked
>>    out, we cannot post there, or can we?)
>
> Good point. It should be forbidden, but there are no safety checks,
> currently.
>
>>> +EXAMPLES
>>> +--------
>>> +
>>> +Assume, while working on a topic, you find and fix an unrelated bug.
>>> +Now:
>>> +
>>> +------------
>>> +$ git commit                                   <1>
>>> +$ git post master                              <2>
>>> +$ git show | git apply -R && git reset HEAD^   <3>
>>> +------------
>>> +
>>> +<1> create a commit with the fix on the current branch
>>> +<2> copy the fix onto the branch where it ought to be
>>> +<3> revert current topic branch to the unfixed state;
>>> +can also be done with `git reset --keep HEAD^` if there are no
>>> +unstaged changes in files that are modified by the fix
>>> +
>>> +Oftentimes, switching branches triggers a rebuild of a code base.
>>> +With the sequence above the branch switch can be avoided.
>>> +That said, it is good practice to test the bug fix on the
>>> +destination branch eventually.
>>
>> This is a cool and motivating example.
>
> Thanks. Another use case is when you receive a patch to be applied on
> a different branch while you are in the middle of some work. If it can
> be applied on the current branch, then you can post it to the
> destination, rewind, and continue with your work.
>
>>> +BUGS
>>> +----
>>> +
>>> +The change can be applied on `dest-branch` only if there is
>>> +no textual conflict.
>>
>> This is not a bug per se, it could be signaled via a return code that
>> the posting was unsuccessful.
>
> Oh, it does so return an error. But there are some cases that one
> expects that work, but where git-apply is not capable enough.
>
> -- Hannes