[FFmpeg-devel] [PATCH] Fix for ogg timestamps when packets span multiple ogg pages

[David Conrad]

On Nov 11, 2008, at 5:07 AM, Michael Niedermayer wrote:

> On Tue, Nov 11, 2008 at 04:22:38AM -0500, David Conrad wrote:
>>
>> On Nov 9, 2008, at 3:58 AM, Michael Niedermayer wrote:
>>
>>> On Sun, Nov 09, 2008 at 09:49:00AM +0100, Michael Niedermayer wrote:
>>>> On Sun, Nov 09, 2008 at 02:48:01AM -0500, David Conrad wrote:
>>>>> On Nov 6, 2008, at 5:51 PM, David Conrad wrote:
>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> The problem with the current code comes from how an ogg page with
>>>>>> only a
>>>>>> partial packet uses the special granulepos of -1. The granulepos
>>>>>> needs to
>>>>>> be saved from the last page that completed a packet until the
>>>>>> next page
>>>>>> that completes a packet, where it is the timestamp of the first
>>>>>> complete
>>>>>> packet.
>>>>>
>>>>>
>>>>> After looking at ogg granule_position some more, there's another
>>>>> problem in
>>>>> that Dirac and ogm both assume that the timestamp is the beginning
>>>>> of the
>>>>> associated packet, wheras Xiph codecs define the granule as the
>>>>> time after
>>>>> the associated packet is fully decoded (I'm assuming, I've only
>>>>> found
>>>>> Vorbis and Theora to do so explicitly.)
>>>>
>>>> older versions of theora also use the ogm variant IIRC
>>>> also see the (non working AFAIK) hack in theora_gptopts() about  
>>>> that
>>
>> Are you sure r15785 isn't not working? Or were you thinking of
>> something else?
>> And you're right, Theora < 3.2.1 does use the granule as frame start.
>
> I was thinking of 15785 but i did not test it, so maybe iam wrong,  
> its a
> while since ive looked at theora in ogg. But i think the current  
> code is
> not guraneed to produce correct timestamps for some ways of  
> distributing
> theora packets in ogg pages.
> Also i suspect the keyframe flags will be wrong, this could be checked
> easily by printing demuxer and vp3_parser.c found keyframe flags.
> But again, this is just foggy memory and gut feeling ...


The only way I can think that the current code could be wrong (or not  
equivalent to interpreting the granule as frame start time) would be  
if the dts didn't increase by one each frame. But afaik that's  
required of theora in ogg.

As for theora keyframes, I don't think there's any difference between  
the current version and older versions (as in the current code is  
always wrong for both, and this patch should fix that.)  But perhaps  
I'm wrong...