[FFmpeg-devel] [PATCH 1/2] avcodec/mlp_parser: don't try to combine frames when full frames are provided

[James Almer]

On 1/19/2018 5:06 PM, wm4 wrote:
> On Fri, 19 Jan 2018 16:51:45 -0300
> James Almer <jamrial at gmail.com> wrote:
> 
>> Attempting full frame reconstruction is unnecessary for containers
>> like Matroska, so just skip it altogether.
>>
>> Signed-off-by: James Almer <jamrial at gmail.com>
>> ---
>>  libavcodec/mlp_parser.c | 4 ++++
>>  1 file changed, 4 insertions(+)
>>
>> diff --git a/libavcodec/mlp_parser.c b/libavcodec/mlp_parser.c
>> index 3c0330f777..4827354f18 100644
>> --- a/libavcodec/mlp_parser.c
>> +++ b/libavcodec/mlp_parser.c
>> @@ -256,6 +256,9 @@ static int mlp_parse(AVCodecParserContext *s,
>>      if (buf_size == 0)
>>          return 0;
>>  
>> +    if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
>> +        next = buf_size;
>> +    } else {
>>      if (!mp->in_sync) {
>>          // Not in sync - find a major sync header
>>  
>> @@ -315,6 +318,7 @@ static int mlp_parse(AVCodecParserContext *s,
>>      }
>>  
>>      mp->bytes_left = 0;
>> +    }
>>  
>>      sync_present = (AV_RB32(buf + 4) & 0xfffffffe) == 0xf8726fba;
>>  
> 
> Not so sure about this. I think some mkv TrueHD files contain packets
> that are not properly starting on frame boundaries. But I don't really
> remember.

As in badly muxed files? Couldn't the same thing happen with any other
codec? We're not bothering to consider such cases in other parses as far
as i could see.