Thu Feb 28 17:20:14 CET 2013

Lifting semantics : adding type information

Adding type information to a network seems straightforward: just
prefix the values of the parent semantics with node types.  Trouble
here is that types are also outputs.  How to handle that extra node?
Ha! convert to CPS!

So at least the principle seems straightforward.

The implementation seems a bit problematic.  Mostly the delay
involved: something needs to be suspended until the whole type node
system is resolved.  I.e. primtives cannot be called directly.  How to
do that?

Each computation node can be represented as a function of all the
program's input nodes.  Maybe CPS is actually essential to make this