Automatic conflict resolution to the state of working directory

[ilyapoz]

Let's say I have two independent branches applied.

I add two consecutive lines to a file:

context
+ line1
+ line2
context

Or I add one line, commit it to one branch and add another line just below it and commit to the other branch.

Now, this is not possible, because speaking just from the branches perspective, this is a conflict:

A:

context
+ line1
context

B:

context
+ line2
context

But (sic!) I did the edits in a consistent working copy. So there is a source of (possible) truth about how this conflict is/should/at least could be resolved. Can this information be used by gitbutler to "just work" in this case?

Otherwise this forces me to make branches dependent on each other. And this is problematic in case I already publeshed indepedent PRs for them.

[Byron]

Thanks for sharing, that's an interesting idea!

It's true that the resolution exists and that in theory could be used to make the merge work, when right now it won't unless these branches are dependent.
Technically, I think we are far from achieving this as it would mean that GB has fine-grained control over the merge to start with. Then of course, it would have to know what to do, much like git rerere somehow magically claims to be able to do that even though it never really worked for me (or I didn't notice it did).

[ilyapoz]

Yes, as random thoughts (I'm not familiar with inner workings of neither GB nor git rerere) I thought that GB could "show to git rerere" that such a resolution is available and then rerere would do the hard work. I for sure notice that rerere works. I often see that GB marks something as a conflict only for me to run but resolve and see that the conflict is auto resolved. I have rerere enabled always.

But really from what I feel the area of conflicts is not deeply worked through in but (yet?), because behavior in case of conflicts on the merge commit is often unobvious (unapplying branches, stopping absorb half way through without clear explanation about "why") etc. So probably this is long way off in the possible timeline.

[Byron]

Yes, first-class conflict support is definitely something we'd like to achieve, and doing auto-resolution seems like the cherry on top.