Wed May 22 12:13:31 EDT 2013
This works well for real I/O as the pack/unpack operations can be done
explicitly. However for the behind-the-scenes state feedback it might
It would be good to allow for some kind of "lazy lifting" on the state
nodes. I.e. whenever a map operation is attempted on a single node,
splice it open.
Trouble there is to put it back together again. I.e. how to bind a
vector node to a list of scalars.
This can actually be done in the ai-array.rkt implementation. When a
return value of any function is not a single node, insert a parallel
( While there is definitely some worth in just doing it, running into
walls, and climbing over them, I do miss a bit of a global idea. )
Let's see where this goes incrementally.
As long as the size of the array is known, it's fairly
straightforward. However, there is one point where it isn't, which is
state input. How to split that into a proper list?
Problem: type inference needs to complete before the operation can be