[Theora-dev] Experimental decoder

[Timothy B. Terriberry]

The problem with motion vectors in the experimental decoder should now
be resolved. The issue was that in low-bitrate/low-motion conditions, it
would sometimes blow away the contents of the golden reference frame
with the contents of the following inter frame (and then possibly
several more inter frames after that). This caused big errors in blocks
coded using the golden frame as a reference frame (and of course never
occurred in my test files). What I get for trying to be cute and
minimize buffer copies, I guess.