[Ffmpeg-devel] Compilation with --enable-theora fails

[Måns Rullgård]

Rich Felker <dalias at aerifal.cx> writes:

> On Wed, May 11, 2005 at 04:22:37PM +0200, M?ns Rullg?rd wrote:
>> Keenan Pepper <keenanpepper at gmail.com> writes:
>> 
>> > Viktor Kompaneyets wrote:
>> >> Is there missing files?  I'd used latest CVS version, but makefile
>> >> kan't find oggtheora.c in libavcodec
>> >
>> > I think oggtheora.c is missing simply because nobody's written it yet. I
>> > can write it when I have some spare time - it should just be some glue
>> > code to interface to libtheora, and also some changes to the ogg demuxer
>> > (currently it only recognizes one vorbis stream). For now you'll just
>> > have to omit --enable-theora (no idea why the option has been there so
>> > long with no actual code).
>> 
>> Since I'm sort of in charge of the Ogg demuxer, i guess this is my
>> call.  I haven't yet checked in the theora support for the Ogg demuxer
>> due to the incredibly insane way they've used to map frame number to
>> granule position.  Rich, will you help me out bashing Xiph a little?
>
> Are you talking about the old ogg demuxer or the new one lu ported?

The new one.

> IMO we should abandon the old libogg based one. Unfortunately I don't
> know how Xiph has done the granulepos for Theora, but if you say it's
> incredibly insane I won't be surprised..

The granulepos is the frame number of the last keyframe shifted left
by a variable amount (specified in the header) plus the number of
frames since the last keyframe.

> Just do whatever you can so that the ogg demuxer outputs valid
> timestamps (in units of the time base), which may mean doing ugly
> hacks/conversions from the granulepos system.. if you need to that's
> ok from my perspective.

I have a solution, just need to test it.

-- 
M?ns Rullg?rd
mru at inprovide.com