Re: FRP

James McCartney wrote:
> On 3/20/07, Shriram Krishnamurthi <[email protected]> wrote:
>> James,
>>
>> I hope we didn't give the impression that we're building the first
>> dataflow language.  If anything, our citations alone should put to
>> rest any such claim.
>>
>> So is SuperCollider also a dataflow language?  Sounds like it -- and
>> the more merrier.  Is SuperCollider the first dataflow language?  Not
>> by a long shot, either.
A lot of work in this area was done before 1980, but somehow the
declarative dataflow work has died out a bit.  FRP is very similar to
concurrent logic programming, but with extra scheduling constraints
to avoid nondeclarative "glitches".  The core language of such a
system is a functional language extended with threads, logic variable
synchronization, and a nondeterministic wait on multiple events.  At
that level it is not declarative, although at a higher level you can take
declarative "snapshots" (at any instant, it looks like a functional
calculation).  There is another form of concurrent reactive declarative
programming, pioneered as far as I can tell by Gilles Kahn in a 1974
paper.  This is completely deterministic: it cannot express nondeterminism
but it has a fully declarative functional semantics.  The nice thing is
that the default is deterministic - you can add nondeterminism but only
there where it is needed.  It also can work together well with lazy
evaluation, which I believe is harder for FRP (correct me if I'm wrong).
We have an exposition of this form of declarative dataflow in chapter 4
of the book "Concepts, Techniques, and Models of Computer
Programming".  It's implemented efficiently by the Mozart
Programming System (www.mozart-oz.org).

Peter Van Roy