bug#69220: [PATCH] smerge-mode: add a function to resolve all conflicts in a file

[Konstantin Kharlamov]

On Mon, 2024-02-19 at 20:07 +0300, Konstantin Kharlamov wrote:
> On Mon, 2024-02-19 at 15:17 +0300, Konstantin Kharlamov wrote:
> > On Mon, 2024-02-19 at 14:03 +0200, Eli Zaretskii wrote:
> > > > From: Konstantin Kharlamov <Hi-Angel@yandex.ru>
> > > > Date: Sat, 17 Feb 2024 13:16:14 +0300
> > > > 
> > > > This implements a feature request from here¹ about having a
> > > > function to
> > > > resolve all conflicts simultaneously. Although question author
> > > > didn't
> > > > reply, but either way I think it's a useful functional. I
> > > > needed
> > > > it
> > > > so
> > > > many times, but before stumbling upon this question I just
> > > > didn't
> > > > know
> > > > there are functions `smerge-keep-upper/base/lower`, and so ofc
> > > > I
> > > > never
> > > > though of writing a new one that would apply them to all
> > > > conflicts.
> > > 
> > > I use SMerge quite a lot, but never yet had a situation where the
> > > same
> > > resolution was applicable to all of the conflicts, let alone knew
> > > that
> > > in advance, before looking at each conflict.
> > 
> > Well, in Emacs it is allowed to create large commits with many
> > functional changes, which I think is why you never saw such
> > functional
> > to be necessary.
> > 
> > Offhand I can tell at least two situations where it is needed; both
> > imply you have more than one commit on the branch:
> > 
> > 1. You got a commit that does two different functional changes to a
> > hunk. So you want to split it. You do an interactive rebase to the
> > previous commit, then do one of the changes and create a commit
> > from
> > it. Then you do a `git rebase --continue` and you get conflicts;
> > but
> > you know beforehand exactly that you want it to be solved in
> > preference
> > of the newer commit.¹
> > 2. You noted, either yourself or as part of codereview, that one of
> > the
> > older commits on the branch has a bug; but you know the bug is non-
> > existent in newer commits. So you fix it in the older commit, then
> > upon
> > `git rebase --continue` you again know exactly that you want just
> > the
> > newer version.¹
> 
> Well, I understand these two points do not sound like something
> unsolvable with `git-checkout` theirs/ours options. It's just the
> general workflow that I remembered offhand.
> 
> I don't remember the distinction down to technical details, only that
> I
> stumbled upon that quite often (which I usually noted because I
> thought
> theirs/ours checkout is gonna work but then it wouldn't; and then I
> had
> to abort everything because I needed conflicts back lol).
> 
> I think this happens because git is often quite good in making
> conflict
> as small as possible. So I think if you have case like this: 1. you
> modify return value in older commit, 2. You do `git rebase --
> continue`,
> 3. you get conflicts because there're unrelated modifications in the
> same hunks as `return`s; then you might get conflicts that only
> contain
> lines you just modified and nothing else. So resolving every conflict
> becomes trivially choosing "ours" (IIRC, I confuse theirs/ours)
> everywhere; but you don't want to `checkout --ours`.
> 
> ----------------
> 
> Incidentally, for me it feels like having the case where you want to
> solve *all* conflicts in preference of either side happens more
> often,
> then the case where you want to solve only *only one* conflict in
> preference of either side. IOW, if I had to rate by frequency
> conflict
> types I meet during my everyday work, it would be (in order: most
> frequent to less frequent):
> 
> 1. Conflicts requiring manual intervention to take changes from both
> sides.
> 2. Conflicts, where all of them at once may be solved in preference
> of
> theirs or ours.
> 3. Conflicts where some require manual intervention and some may be
> solved in preference of either side.

Ok, did anyone order a case for "solve all to one side" that isn't
solvable with git's theirs/ours? Here, fresh from the bakery 😊

1. I edit a Makefile at hcl-mode¹ to try to introduce a separate option
for compiling tests and renamed the older `compile` one to `compile-
pkg`
2. While doing so I realize the Emacs call is wrong: it uses both `-Q`
and `-batch` options, whereas `-batch` implies "no init file". Strictly
speaking it implies `-q` not `-Q`, but it is very unlikely this
distinction is intentional. So I save the current changes and
interactively-rebase to the previous commit.
3. I remove `-Q` from all `-batch` calls and save it as a new commit
4. I do `git rebase --continue` and obviously I get conflicts

Now, git turns out to be very good in reducing conflicts, so it only
leaves me with the two lines that I change and nothing more from the
surrounding hunk. Now, since in the newer commit I don't have -Q
anymore, I know I want the newer version for all conflicts. But I can't
use `git --theirs/ours`, because that would return the `-Q`s that I
removed in the previous commit.

P.S.: with that said, in this case it was just one conflict, simply
because their Makefile is very small. But I hope you get the idea: if
there were more distinct lines where I'd removed `-Q` option both in
newer and older commits, they all would be solvable in preference of
the single side of the conflict.

1: https://github.com/hcl-emacs/hcl-mode