[FFmpeg-devel] [PATCH] fix bug in af_pan channel coefficient parser

[Tobias Rapp]

On 19.11.2017 17:39, Nicolas George wrote:
> L'octidi 28 brumaire, an CCXXVI, Michael Roitzsch a écrit :
>> Hi FFmpeg team,
>>
>> I was using af_pan to subtract one audio file from another. I first used an amerge filter and then configured the pan filter with a channel formula like c0=c0-c2|c1=c1-c3.
>>
>> However, this does not work as expected. In debug output, I can see that channel 0 correctly uses gain coefficients 1 and -1 for c0 and c2. But channel 1 incorrectly uses -1 and -1 for c1 and c3.
>>
>> I looked into the code and found that the sign handling is currently wrong. When the last contribution of a formula is subtracted, the sign variable will be correctly set to -1 in line 191. But when then leaving the gains loop in line 189, because we reached the end of the formula for an output channel, sign is not reset. We therefore re-enter the gains loop for the next output channel with sign still -1. The first gain in this new formula will therefore be negated.
>>
>> The attached patch fixes the problem for me.
> 
> Thanks, applied.

Would it be OK to backport the fix into release/3.4? I can do the 
cherry-picking but am unsure about the policies and what other older 
releases should possibly be adapted, too.

Regards,
Tobias