From: Patrick Steinhardt <ps@pks.im>
Date: Mon, 23 Feb 2026 15:48:09 GMT
Subject: Re: [PATCH 4/4] pack-check: fix verification of large objects
Message-ID: <aZx2uRiLIaa21L-x@pks.im>
In-Reply-To: <20260223125843.GA215671@coredump.intra.peff.net>

On Mon, Feb 23, 2026 at 07:58:43AM -0500, Jeff King wrote:
> On Mon, Feb 23, 2026 at 12:30:35PM +0100, Patrick Steinhardt wrote:
> 
> > > A more concrete test would probably be something like:
> > > 
> > >    1. Two packs, $X and $Y, both contain the same object.
> > > 
> > >    2. The object is corrupt in $X but not in $Y.
> > > 
> > >    3. Running fsck detects that one copy is corrupt but the other is
> > >       not.
> > > 
> > > Right now it may or may not fail depending on the ordering of the packs
> > > in the MRU list (which we might be able to tweak via mtimes). But
> > > hopefully in the "after" state it should deterministically complain
> > > about $X.
> > 
> > Yeah. The problem I had here is that I'm not sure whether we have any
> > tools to reliably create a corrupted object, e.g. with a hash mismatch.
> > I'll have a look for v2.
> 
> You can see how do_corrupt_object() in t5303 does it. It's basically
> finding an offset via show-index and then writing a zero over it with
> dd.

Yeah, that's what I ended up doing indeed. I spotted such a test in t1450.

Patrick