opus - Dec 2013 - Possible pointer math issue in 1.1

While integrating 1.1, I ran into some errors with my compiler and, after
examining the code, saw what
might be some invalid pointer math in opus_encoder.c. In optimize_framesize(),
the first parameter 'x' is
typed as opus_val16 and pointer math is done on the variable (in the 'if
(buffering)' block) and then x is
passed to the downmix function. I see that the downmix function uses the proper
type, but it
appears to me that the pointer adjustment performed in that 'if
(buffering)' block is always using 16-bit.

Kevin O'Connor

On 12/10/2013 02:08 PM, O'Connor, Kevin wrote:
> While integrating 1.1, I ran into some errors with my compiler and,
> after examining the code, saw what might be some invalid pointer math
> in opus_encoder.c. In optimize_framesize(), the first parameter 'x'
> is typed as opus_val16 and pointer math is done on the variable (in
> the 'if (buffering)' block) and then x is passed to the downmix
> function. I see that the downmix function uses the proper type, but
> it appears to me that the pointer adjustment performed in that 'if
> (buffering)' block is always using 16-bit.
You're right, the first parameter optimize_framesize() should be a
"void
*x" and the pointer arithmetic on x is wrong. Fortunately, that code is
currently not enabled (and not exposed to the API) because it doesn't
work well (probably not only because of the bug you just found). Thanks
for spotting this!

	Jean-Marc

Oh, and the code is fixed now.

	Jean-Marc

On 12/10/2013 04:33 PM, Jean-Marc Valin wrote:
> On 12/10/2013 02:08 PM, O'Connor, Kevin wrote:
>> While integrating 1.1, I ran into some errors with my compiler and,
>> after examining the code, saw what might be some invalid pointer math
>> in opus_encoder.c. In optimize_framesize(), the first parameter
'x'
>> is typed as opus_val16 and pointer math is done on the variable (in
>> the 'if (buffering)' block) and then x is passed to the downmix
>> function. I see that the downmix function uses the proper type, but
>> it appears to me that the pointer adjustment performed in that 'if
>> (buffering)' block is always using 16-bit.
> 
> You're right, the first parameter optimize_framesize() should be a
"void
> *x" and the pointer arithmetic on x is wrong. Fortunately, that code
is
> currently not enabled (and not exposed to the API) because it doesn't
> work well (probably not only because of the bug you just found). Thanks
> for spotting this!
> 
> 	Jean-Marc
> _______________________________________________
> opus mailing list
> opus at xiph.org
> http://lists.xiph.org/mailman/listinfo/opus
>