[MEncoder-users] Adding harddup as an afterthought?

[Ilya Zakharevich]

[A complimentary Cc of this posting was NOT [per weedlist] sent to
Rich Felker 
<mencoder-users at mplayerhq.hu>], who wrote in article <20070929071846.GM246 at brightrain.aerifal.cx>:

   > > B-frames' mvs are predicted from the following P-frame's mvs.  And
   > > the prediction formula depends on the relative frame numbers, so
   > > inserting a new B-frame would break any existing adjacent B-frames
   > > even if you got the new frame to work.

   > Now I'm completely lost: how the player is suppose to number
   > B-frames in the presence of 0-frames?

   0-frames are not frames.  They are simply times at which there
   is no frame to present, for the purpose of fitting a
   fixed-time-increment container format. The decoder should never see
   them.

Let me restate my question: if decoder gets

  P 0 0 0 0 0 0 0 0 0 B P

how should it predict the movements of the B frame?  By 0.5 of
movement of (the second) P, or by 0.9 of it?

Puzzled,
Ilya