Underreact

Underreact is a framework for continuous reactive programming, using JavaScript as a runtime platform. David Barbour has been creating a programming model called RDP (Reactive Demand Programming), and Underreact is primarily my attempt to implement that vision--not just to prove it can be done, but to use it in my own projects. I may make some different choices that stray from what David would like to call RDP, partly because I still don't completely understand the reasons for the RDP design.

I've implemented a system for specifying and using a certain subset of RDP behaviors. This subset is an elegant system in itself, but its first-class behaviors are limited to a small, hardcoded library, and all its behaviors are limited to a single partition.

Most of the time, I've been doing tests using two ad hoc I/O behaviors behaviors dedicated to very specific tasks:

• Watch the mouse input.
• Dump a signal to a DOM element for visualization as text.

You can see these non-flashy tests here:

• test/tests.html

And now there's a slightly less ramshackle test here, which shows a box that follows the cursor:

• test/canvas.html

Last but not least, here's an upgrade of that test with a basic demonstration of animated reactive state transition:

• test/canvas-fancy.html

behClosure() and behCall()

Underreact does not yet have Sirea's beval for dynamic behaviors, but it does have first-class behaviors of a different sort. Invoking one of these behaviors, using behCall, will not perform implicit synchronization or communication at run time--but it may result in some initial setup at compile time or load time. The behClosure operator can create a first-class behavior with custom internal code and an encapsulated signal (suitable for implementing a syntax with lexical closures). Underreact will use these tools as an efficient basis for modeling capabilities, including resource spaces.

Programs built up using behClosure and behCall will essentially be fully inlined before execution begins. Even if it were possible to express a recursive program using these tools (and it isn't!), it would only grow divergently, and execution would never be able to start.

behYield()

As an experimental way to communicate with outside systems, Underreact models ambient access to a "membrane" that sends out demands to another system. This access is in the form of behYield, an operator which can pass a signal of serializable data to the environment and get another signal of serializable data in response, very much like a coroutine.

This direction would be expressive, but RDP is meant to be securable using object-capability model techniques, and I suspect this mechanism would encourage developers to lump all their privileges on the ambient environment rather than using fine-grained control. I therefore recommend trying to use behYield only in conjunction with behClosure, so that it's a first-class capability.

Possible future directions (in no particular order)

• Provide resource spaces.
• Provide persistent memory resources.
• Turn demand monitors into managed resources.
• Support dynamic evaluation of serialized behaviors (beval).
• Support remote partitions.
• Design a type inference layer, for convenient programming.
• Design a textual syntax, for convenient programming.

I intend for Era, my meaning-preserving module system, to coexist well with RDP, so Era and Underreact have some future directions in common:

• Let Era modules describe RDP programs which can be compiled to use Underreact.
• Provide an Underreact state resource that represents a set of installed Era modules.