[Bug bzip2/28904] libbzip2: ‘cost[3]’ may be used uninitialized in this function [-Wmaybe-uninitialized]

["demerphq at gmail dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28904

--- Comment #3 from demerphq <demerphq at gmail dot com> ---
On Wed, 20 Apr 2022 at 00:23, mark at klomp dot org
<sourceware-bugzilla@sourceware.org> wrote:
>
> https://sourceware.org/bugzilla/show_bug.cgi?id=28904
>
> --- Comment #2 from Mark Wielaard <mark at klomp dot org> ---
> A different solution might be to use the constant BZ_N_GROUPS instead of the
> variable nGroups when resetting the arrays:
>
> diff --git a/compress.c b/compress.c
> index 5dfa002..2dc5dc1 100644
> --- a/compress.c
> +++ b/compress.c
> @@ -321,7 +321,7 @@ void sendMTFValues ( EState* s )
>     ---*/
>     for (iter = 0; iter < BZ_N_ITERS; iter++) {
>
> -      for (t = 0; t < nGroups; t++) fave[t] = 0;
> +      for (t = 0; t < BZ_N_GROUPS; t++) fave[t] = 0;
>
>        for (t = 0; t < nGroups; t++)
>           for (v = 0; v < alphaSize; v++)
> @@ -353,7 +353,7 @@ void sendMTFValues ( EState* s )
>              Calculate the cost of this group as coded
>              by each of the coding tables.
>           --*/
> -         for (t = 0; t < nGroups; t++) cost[t] = 0;
> +         for (t = 0; t < BZ_N_GROUPS; t++) cost[t] = 0;
>
>           if (nGroups == 6 && 50 == ge-gs+1) {
>              /*--- fast track the common case ---*/
>
> Which might also help the compiler to simply zero the whole (6 element) arrays
> in one go (the common case), instead of having to check whether nGroups is
> smaller in this particular case.

I am fine with both of your proposals. I was wondering why you don't
use memset() instead of the explicit loop? From what I understand it
is faster.

Yves

-- 
You are receiving this mail because:
You are on the CC list for the bug.