From: Jeff King <peff@peff.net>
Date: Wed, 30 Sep 2026 20:55:35 GMT
Subject: Re: [PATCH 0/4] repack: various corner cases for cruft-less MIDXs
Message-ID: <20260930205535.GD747209@coredump.intra.peff.net>
In-Reply-To: <cover.1790731662.git.me@ttaylorr.com>

On Tue, Sep 29, 2026 at 08:28:34PM -0500, Taylor Blau wrote:

> This patch series fixes a few bugs I spotted while investigating the
> cruft-less MIDX feature.
> 
> The bugs addressed are found in various corner cases, and, when
> triggered, may result in a MIDX being written whose objects are not
> closed under reachability. When this happens while the caller is trying
> to write reachability bitmaps, bitmap generation may fail if one or more
> selected commits are descendants of the open portion of the MIDX.

I think all of these are making things strictly better, but I did find a
few spots where the fixes might be incomplete. I'm not sure if that
argues for a re-roll or for punting those to future work. ;)

I agree with Stolee that an oidset is perhaps a better data structure
for storing the extra roots (which are in a kind-of random order anyway,
since we're pulling them in pack order from various packs). But it also
probably doesn't make that big a difference in practice (we'll skip
duplicates during the traversal, and you probably don't have that many
duplicate objects in a repo in the first place).

-Peff