[FFmpeg-devel] [PATCH] avcodec/g723_1_parser: keep avctx channels internally if value is > 0

[James Almer]

On 3/18/2022 10:07 PM, James Almer wrote:
> On 3/18/2022 10:03 PM, Paul B Mahol wrote:
>> Signed-off-by: Paul B Mahol <onemda at gmail.com>
>> ---
>>   libavcodec/g723_1_parser.c | 6 +++++-
>>   1 file changed, 5 insertions(+), 1 deletion(-)
>>
>> diff --git a/libavcodec/g723_1_parser.c b/libavcodec/g723_1_parser.c
>> index 2ed1a8ab19..03c165ce55 100644
>> --- a/libavcodec/g723_1_parser.c
>> +++ b/libavcodec/g723_1_parser.c
>> @@ -26,6 +26,7 @@
>>   typedef struct G723_1ParseContext {
>>       ParseContext pc;
>> +    int channels;
>>   } G723_1ParseContext;
>>   static int g723_1_parse(AVCodecParserContext *s1, AVCodecContext 
>> *avctx,
>> @@ -36,8 +37,11 @@ static int g723_1_parse(AVCodecParserContext *s1, 
>> AVCodecContext *avctx,
>>       ParseContext *pc = &s->pc;
>>       int next = END_NOT_FOUND;
>> +    if (avctx->ch_layout.nb_channels > 0)
>> +        s->channels = avctx->ch_layout.nb_channels;
>> +
>>       if (buf_size > 0)
>> -        next = frame_size[buf[0] & 3] * FFMAX(1, 
>> avctx->ch_layout.nb_channels);
>> +        next = frame_size[buf[0] & 3] * FFMAX(1, s->channels);
> 
> It would be nice to know why this is needed now but not before the 
> channel layout set.

327efa66331ebdc0087c6b656059a8df2f404019 introduced the regression. The 
demux code relies on sti->avctx to retain codec parameters after an 
avcodec_close() call, and it did for ch_layout before that commit 
(avcodec_close() calls av_opt_free(), which is now uninitializing it).

I can't revert that commit because it's obviously correct, so the two 
options are to either make the demux code keep a copy of ch_layout and 
put it back after avcodec_close(), or do that within avcodec_close() 
since it's not meant to clear codec parameters to begin with.

I'm leaning towards doing the latter.

> 
>>       if (ff_combine_frame(pc, next, &buf, &buf_size) < 0 || !buf_size) {
>>           *poutbuf      = NULL;