Re: [PATCH] last-modified: implement faster algorithm

From: Jeff King <peff@peff.net>
Date: Oct 21, 2025, 08:20 UTC
Message-ID: <20251021082021.GF259661@coredump.intra.peff.net>
In-Reply-To: <aPJYvYs8W6LrV+0Q@nand.local>

On Fri, Oct 17, 2025 at 10:54:53AM -0400, Taylor Blau wrote:

Show 16 quoted lines

> > Hmm, sounds like the mark_path() discussion from:
> >
> >   https://lore.kernel.org/git/aHmPHcNQYlhGo8JB@nand.local/
> >
> > coming home to roost. I'm sure you already knew that, but there's maybe
> > an interesting process observation here: in pulling a battle-tested
> > implementation apart into patches to be applied in chunks, we ended up
> > missing a critical part of that original implementation and getting a
> > bug.
> 
> Hmm. Is that what happened in this case, though?
> [...]
> I was wondering if perhaps I had made an error when pulling these
> patches out of GitHub's fork, but even in my b0ae8b3cc0 (blame-tree:
> introduce '--go-faster' mode, 2025-03-27) from my fork, you can see the
> same diff in mark_path() as above.

Yeah, I think the patches in your fork are correct, and it got lost in Toon's rewrite. It is probably naive to think we could diff the endpoint (your fork vs Toon's patches) to find such changes, though. There have been too many other cleanups and changes as it was upstreamed. So you can ignore most of my other email as philosophical musing.

> [...some more clever optimizations...]

All of that looked plausibly correct to me. ;) I'll leave it to Toon to experiment with it for correctness and performance improvements.

-Peff