Tue Apr 2 16:03:18 EDT 2013


Not necessarily.  The problem is that the upsampling/downsampling
points don't have much structure.

mixing state update and local nodes computation does make things
entagnled and harder to manipulate.

So I really see no other way: evaluate the code twice:

- mode 'setup' -> run the update method, and pull (force) only the
  updated state nodes and loop constants

- mode 'update' -> don't run the update method, but plug in the nodes
  from the 'setup phase.

the trouble here is associating each feedback/n node with the nodes
computed in the previous phase.