[<<][meta][>>][..]
Tue May 4 11:47:53 EDT 2010

Representing (single) assignment

See previous posts for prelude.  These ideas keep coming back:

  - Lexical scope is good  (stick to lambda/let for inputs)

  - In addition, explicit outputs are necessary.

It seems I'm looking for some kind of abstraction of assignment.
I.e. the "output" operator.

The problems that comes up then is: what is the return type of a
function that performs bindings to outputs?

  f (in1, in2) out = out <-- in1 + in2

Ha! It's the ``other side'' of the channel!

So the main abstraction I'm looking for is a channel with one input
and any number of outputs.  Funny, this relates directly to the
DFL / reactive work in Scheme[1].

The type of `f' above is ... -> [Channel], or even a tuple.  This
would then also bridge with the Arrow abstraction.

The problem is: how do you tie the two ends together?  How do you
implement "connect".  The flow of information is like:

     A "patcher" creates a node which has two ends, passes the input
     end as an output to a processor, and the output end as an input
     to any number of processors.

The question is then: what is a channel?  How to represent it?
Modeling a channel as a mutable object seems to make this simple.  Can
it be modeled as a function?  The important element is sharing.

Let's write `liftA2' for binary operators, or maybe Functor is enough
and simply use `mapf' for (x,x) -> x.

It seems that this needs a push or pull approach, i.e. we need to
"pick a side" of how to evaluate a network that's constructed by a
patcher.  Pull seems safest and closest to how we represent channels
(model a network as its eventual output), so a channel is something
that converts a "network context" into a number, triggering evaluation
whenever a certain value isn't present yet.

This is a State monad.  It's not a Reader monad since the context
(variable bindings) can and does change after an evaluation.

What we want to abstract is the memoization.

Full circle again: this is exactly the representation used in the
Staapl/Scheme approach to the dataflow problem.

So, let's restate: a channel is a State monad c->(v,c) where c is the
evaluation context.


[1] entry://../plt/20100327-160826



[Reply][About]
[<<][meta][>>][..]