[FFmpeg-devel] [PATCH v9] lavf: palettized QuickTime video in Matroska

Mats Peterson matsp888 at yahoo.com
Mon Dec 28 03:59:47 CET 2015

On 12/28/2015 03:27 AM, Mats Peterson wrote:
> On 12/28/2015 03:21 AM, Mats Peterson wrote:
>> On 12/28/2015 03:16 AM, Michael Niedermayer wrote:
>>> On Mon, Dec 28, 2015 at 12:24:10AM +0000, Carl Eugen Hoyos wrote:
>>>> Mats Peterson <matsp888-at-yahoo.com <at> ffmpeg.org> writes:
>>>>> +        memcpy(st->codec->extradata, matroska->palette,
>>>>> +                AVPALETTE_SIZE);
>>>> As said, please remove this,
>>>> you must not fix MPlayer
>>>> issues in FFmpeg.
>>> this is true but iam not sure if this is really "just" a mplayer
>>> issue.
>>> If there is a global palette, then it is not unreasonable that it
>>> is placed in a stream global place. This is especially true if not
>>> all packets are copied and the palette from the first packet is
>>> lost as a result.
>>> either way, the commit message for this part just points at mplayer,
>>> which isnt really good for a description of a "bug fix" in libavformat
>>> Patch splited in move and matroska part
>>> i removed this memcpy() for now from what i commited as there is
>>> clearly no consenus on it.
>> Michael, the concensus is that MPlayer won't work without having the
>> palette in extradata, since it uses the extradata as the palette when it
>> adds ANOTHER palette side data packet AFTER the one added in
>> matroskadec.c. I'm not making this up. Carl is of course right in his
>> statement that we shouldn't fix MPlayer issues in FFmpeg, but perhaps it
>> could be a "quick fix" until the MPlayer developers wake up?
>> Mats
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> Especially since MPlayer is such an evident use case of FFmpeg. It even
> includes it inside its own source code.
> Mats

For the record, at line 423 in libmpdemux/demux_lavf.c in MPlayer, it 
uses whatever is in extradata to tack onto the end of its "fake" 
BITMAPINFOHEADER, and which is later used to add this "extraneous" 
palette side data packet:

   memcpy(sh_video->bih + 1, codec->extradata, codec->extradata_siz


Mats Peterson

More information about the ffmpeg-devel mailing list