Re: [PATCH 1/2] add -p: mark split hunks as undecided

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 27, 2025, 18:36 UTC
Message-ID: <xmqqjz9b6xr1.fsf@gitster.g>
In-Reply-To: <180271a6-eb0e-4c15-9916-b2ab5760f4ec@gmail.com>

phillip.wood123@gmail.com writes:

Show 14 quoted lines

>>> Currently after a selected hunk is split we always prompt the user to
>>> make a decision on the first mini-hunk even though it is marked as
>>> selected when it is split. This seems inconsistent and confused me
>>> when I first tried splitting a selected hunk which is why I wrote this
>>> patch.
>> Hmph, so there is an obvious alternative "fix" to the inconsistency,
>> i.e., after splitting, move to the first unselected hunk?
>
> We could do that but I think it would be even more confusing than the
> current behavior as it would make it harder to change the state of the
> mini-hunks. At least with the current behavior one can use 'J' to move
> through them immediately after splitting the original hunk. If we move
> to the next undecided hunk one has to know where the newly-created
> mini-hunks are relative to that.

True. After all, going back to an already selected hunk and then splitting the hunk is a clear indication that the user wants to visit some of them to change their state. Moving them back to "undecided" (not "deselected") instead of leaving them marked as "selected" (which is the current behaviour) looks like a better behaviour and I wish I knew about the possibility in late 2006 when I added the hunk splitting.

> I'm not sure either. I dislike the way it works at the moment and find
> it confusing but if there are a lot of people relying on it then I'd
> be reluctant to change it.

I share the sentiment, especially the latter.

> Unfortunately we don't have any way to know
> if anyone is relying on the current behavior without changing it and
> seeing if anyone complains. Given it is a bit of a corner case I'm not
> sure whether it is worth spending much more time on it.

Given our user base has grown quite a bit over the years, it almost is a given that any change to existing behaviour is a regression to somebody. Certainly a safe material for Git 3.0 but I do not know if it is safe enough for 2.50 for example. The strategy to leave it longer in 'next' did not work well to catch potential issues for another topic during this cycle, but we could try it out again.

> I can see the problem and asking for conformation before quitting
> would have been nice if we'd done it from the start. I'm not sure it
> is worth the disruption of changing it when one can re-run "reset/add
> -p" quite easily though.

Yup. That matches my assessment of it. I brought it up because I see this "selection should not stick across splitting" falls into the same "it would have been nice if it were that way from the beginning" bucket.

> I guess we could add an opt-in cofing that
> eventually becomes the default.

I'd prefer not to add configuration for tweaking such a small thing (this applies to "should selection stick across splitting?", too).

> While we're talking about tangential issues it would be nice if when a
> user revisited a hunk we told them its current state. At the moment
> there is no way to tell if a hunk has been selected or not.

The user came back with 'J' or 'K' probably because the hunk was skipped in their earlier navigation with 'k' or 'j', so users may be using it as a workaround, but I agree there should be an indicator for the (unselected, selected, undecided).

> Related to that the help for 'J' and 'K' talk about leaving the
> current hunk undecided when what they actually do is leave the
> current state unchanged.

Nice catch.

Thanks.