[theora] Liboggplay seeking artifacts

[Ralph Giles]

On Fri, Jul 24, 2009 at 4:56 PM, Chris Pearce<chris at pearce.org.nz> wrote:

> I think my main beef against the canonical approach is that it requires
> reading backwards, which is bad for apps that read over networks, like
> Firefox, because they require new networks requests.

The way I generally state this is that (for each stream you care about
playing) you need to bisection search for the page with the highest
whose time-equivalent is the lastest point *prior* to the target. That
way you can be sure, going forward to get your keyframe (for theora)
or preroll (for vorbis).

And no, it's not fun.

 -r