From: Patrick Steinhardt <ps@pks.im>
Date: Wed, 21 Jan 2026 08:51:36 GMT
Subject: Re: [PATCH v2 11/18] git-compat-util.h: introduce `u32_add()`
Message-ID: <aXCTkVpjJkTabx_0@pks.im>
In-Reply-To: <aWgwn2rk/qw+fRoA@nand.local>

On Wed, Jan 14, 2026 at 07:11:11PM -0500, Taylor Blau wrote:
> On Wed, Jan 14, 2026 at 05:03:56PM -0500, Taylor Blau wrote:
> > As for removing u64_add(), that should be straightforward as well since
> > there is also a single caller. Let me know if you think that makes sense
> > to take up as part of this series, or if you would prefer it done
> > separately. I tend to prefer the latter, since the state after applying
> > the above is that we avoid adding any new callers.
> 
> This appears to be easy enough. The following applies on top of 'master'
> if you want to pick it up separately:
> 
> --- 8< ---
> 
> Subject: [PATCH] git-compat-util.h: drop u64_add(), u64_mult() helpers
> 
> The u64_add() and u64_mult() helper functions were introduced in
> b103881d4f4 (midx repack: avoid integer overflow on 32 bit systems,
> 2025-05-22) to implement overflow checks during a fixed-point
> calculation when estimating pack sizes in the MIDX writing code.
> 
> However, those functions call die() when either the addition or
> multiplication of their operands (depending on which function is being
> called) would cause an overflow. This does not allow the caller to
> provide a more detailed message, presenting the user with an opaque
> message like:
> 
>     fatal: uint64_t overflow: M * N
> 
> Let's discourage these opaque error messages by dropping these functions
> entirely and instead having the caller use unsigned_mult_overflows() or
> unsigned_add_overflows() themselves, providing the caller the
> opportunity to come up with their own die() message.
> 
> Suggested-by: Junio C Hamano <gitster@pobox.com>
> Signed-off-by: Taylor Blau <me@ttaylorr.com>
> ---
>  git-compat-util.h | 16 ----------------
>  midx-write.c      | 15 +++++++++++++--
>  2 files changed, 13 insertions(+), 18 deletions(-)
> 
> diff --git a/git-compat-util.h b/git-compat-util.h
> index b0673d1a450..24edd68c671 100644
> --- a/git-compat-util.h
> +++ b/git-compat-util.h
> @@ -641,22 +641,6 @@ static inline int cast_size_t_to_int(size_t a)
>  	return (int)a;
>  }
> 
> -static inline uint64_t u64_mult(uint64_t a, uint64_t b)
> -{
> -	if (unsigned_mult_overflows(a, b))
> -		die("uint64_t overflow: %"PRIuMAX" * %"PRIuMAX,
> -		    (uintmax_t)a, (uintmax_t)b);
> -	return a * b;
> -}
> -
> -static inline uint64_t u64_add(uint64_t a, uint64_t b)
> -{
> -	if (unsigned_add_overflows(a, b))
> -		die("uint64_t overflow: %"PRIuMAX" + %"PRIuMAX,
> -		    (uintmax_t)a, (uintmax_t)b);
> -	return a + b;
> -}
> -
>  /*
>   * Limit size of IO chunks, because huge chunks only cause pain.  OS X
>   * 64-bit is buggy, returning EINVAL if len >= INT_MAX; and even in
> diff --git a/midx-write.c b/midx-write.c
> index 87b97c70872..6006b6569c8 100644
> --- a/midx-write.c
> +++ b/midx-write.c
> @@ -1738,8 +1738,19 @@ static void fill_included_packs_batch(struct repository *r,
>  		 */
>  		expected_size = (uint64_t)pack_info[i].referenced_objects << 14;
>  		expected_size /= p->num_objects;
> -		expected_size = u64_mult(expected_size, p->pack_size);
> -		expected_size = u64_add(expected_size, 1u << 13) >> 14;
> +
> +		if (unsigned_mult_overflows(expected_size,
> +					    (uint64_t)p->pack_size))
> +			die(_("overflow during fixed-point multiply (%"PRIu64" "
> +			      "* %"PRIu64")"),
> +			    expected_size, (uint64_t)p->pack_size);
> +		expected_size = expected_size * p->pack_size;
> +
> +		if (unsigned_add_overflows(expected_size, 1u << 13))
> +			die(_("overflow during fixed-point rounding (%"PRIu64" "
> +			      " + %"PRIu64")"),
> +			    expected_size, (uint64_t)(1ul << 13));
> +		expected_size = (expected_size + (1u << 13)) >> 14;

One downside this pattern has is that we repeat the computation, which
makes it easy to get it wrong or forget updating either the check or the
computation.

I think ideally, we would have interfaces that combine the two
approaches in `u64_mult()` and `unsigned_mult_overflows()`. Something
like this for example:

    static intline bool u64_mult(uint64_t a, uint64_t b, uint64_t *out)
    {
        if (unsigned_mult_overflows(a, b))
            return false;
        *out = a * b;
        return true;
    }

This would let the caller handle the failure and is thus quite flexible,
which results in the following code:

	if (!u64_mult(expected_size, (uint64_t)p->pack_size, &expected_size))
		die(_("overflow during fixed-point multiply (%"PRIu64" "
		      "* %"PRIu64")"), expected_size, (uint64_t)p->pack_size);

Patrick