From: "Omar Polo" <op@omarpolo.com>
Subject: Re: fix off-by-one causing invalid deltas
To: Stefan Sperling <stsp@stsp.name>
Cc: gameoftrees@openbsd.org
Date: Mon, 21 Jul 2025 23:39:10 +0200

Stefan Sperling <stsp@stsp.name> wrote:
> I found a seperate bug while trying to write a test for the problem
> reported by Aaron.
> 
> When deltification attempts to stretch a common block of data to the
> maximum possible size, we have an off-by-one which can cause the block
> to be larger than the maximum block size which can be represented in
> an encoded delta. The result is an invalid delta which does not copy
> enough data and gets flagged by both gotadmin indexpack and git index-pack.
> 
> The new regression test included with this patch triggers the problem:
> 
> got-index-pack: delta application result size mismatch: \
>         actual: 65536 expected: 16777216: bad delta
> 
> 16777216 in the error above equals (1 << 24)
> 
> The problem is on this line, where we effectively need - 2 instead of - 1:
> 
> 	while (buf_equal && *blocklen < (1 << 24) - 1) {
> 
> Also, the inner loop which runs (*blocklen)++ without a bounds check
> could end up moving *blocklen way past the limit.
> 
> I am including the diff I already sent earlier with Subject
>   "delta.c overrides errors in some cases"
> because diagnostic changes I made depend on this other diff.
> I would commit them separately.
> 
> ok?

another great catch! :)

diffs looks good to me, ok op@

(and sorry that i missed the "delta.c overrids  errors in some cases")