Re: [RFC 0/5] LLMinus: LLM-Assisted Merge Conflict Resolution

[Sasha Levin]

On Mon, Dec 22, 2025 at 02:50:55PM +0000, Mark Brown wrote:
> On Sun, Dec 21, 2025 at 11:10:11AM -0500, Sasha Levin wrote:
>
> > I assigned categories to the various trees used by -next (see
> > https://gist.github.com/sashalevin/163df4ae1163e0e22a97edc40e14b7f5) and built
> > a simple wrapper script to generate per-category integration branches, letting
> > LLMinus resolve conflicts whenever we hit one.
>
> Those categories appear to be a bit randomly assigned FWIW.  I'm not

There should be some sense there :) but yes, we could fine tune it as we go.

> clear who would want the various intermediate merges either, I suppose
> that having some of the trees pulled into multiple places might help
> shake out some of the issues due to things getting sent to Linus in a
> different order but OTOH it will increase the total number of merges
> done and tested which is itself a cost.  We could also shake out
> ordering issues by doing something like randomise the ordering.  I think
> I'd want some demand or use case for doing more intermediate merges
> rather than just doing a bunch of them for the sake of it.

My thinking around it was to enable faster per-subsystem tests than what we
currently do. For example, we can quickly build mm-next and run mm focused
tests on it.

Since creating these per-subsystem trees is fairly cheap and can happen even
few times a day, we can help identify issues way earlier during the process.

> This seems like a very separate experiment to your LLM merge thing.

Right, just going off on a tangent based on the Maintainer's summit feedback of
how useful fs-next is.

> > The resulting branches were pushed to
> > https://git.kernel.org/pub/scm/linux/kernel/git/sashal/linux-next.git/refs/ .
> > Each category has a %s-next branch, and a larger all-next branch which merges
> > all of them together and is the equivalent of linux-next.
>
> > Please let me know what you think!
>
> It might be easier to tell what it's done if you ran this with the same
> inputs as the last -next (it's on Christmas break at the minute),
> there's quite large differences in the end result but most if not all of
> that is that the input trees you're using are fresher than the last
> -next.  Though I think even with the same base there'd be a bit of a
> needle in a haystack thing finding interesting cases, probably it'd be
> more useful to find and highlight specific cases where it does something
> interesting.

Yup, I figured I'd wait until break is over and compare the trees once the next
linux-next is released.

--
Thanks,
Sasha