Okay, I thought about this some. I walked around the house, talked to myself, played with some fresh fruit (moving them here and there, describing their properties), and I think I've come up with a very vague idea of this fourth control paradigm.
First, here's how I started. I figured, I'm trying to classify the three existing control paradigms (imperative, functional, logical) into two orthogonal properties. Well, it has to be one of these:
A B
X | Imperative | Functional |
Y | Logical | ??? |
A B
X | Functional | Imperative |
Y | Logical | ??? |
A B
X | Logical | Functional |
Y | Imperative | ??? |
Re:FWIW
luqui on 2005-11-19T22:51:17
> What we learn in these classes is, respectively,
> Functional, Imperative, Logic and OO programming.
Yep, those are the four that I identified in my last post. However, I argue that OO is an "abstraction style", and not a "control style", which is what I'm looking for.Re:FWIW
chromatic on 2005-11-22T01:04:49
If you want a control style, where does event-driven programming fit? What if it's asynchronous?
Re:FWIW
luqui on 2005-11-23T09:00:22
> If you want a control style, where does
> event-driven programming fit? What if it's
> asynchronous?
I define the control paradigm by the input (code) and the output (what happens when you run it). In the event driven case, you give it actions to execute when certain things happen. But the first part is the important part, you're giving it actions, and it tells you what happens when those actions are executed, even though those actions are triggered rather than sequenced. Therefore, I think event-driven programming is a form of imperative, which isn't surprising, since most programming these days is imperative (it is a, er, logical thing to program in, because it maps most closely to what the computer actually does).
However, the point of this little exercise was to identify the binary dimensions on which the different languages lie. So maybe event-driven programming is one of those dimensions:
Event-driven functional: Compute what properties of the world change as objects are introduced and removed (this is what I called forward-chaining logic, and I want a system that does it badly!)
Event-driven logical: Compute the set of objects satisfying a property continually as objects are introduced and removed
Event-driven imperative: Do stuff to the world as you detect it change
Event-driven logiparitive (or whatever this fourth control style is called): Compute the changes the world is undergoing as it changes
Some of those are a little bit stretchy, but it kinda seems to work.
As far as asynchronous programming, I think that's definitely a binary dimension, as it is referring to the *order* in which things are done (which serial or parallel), and the control style doesn't refer to that at all. I have also seen asynchronous functional languages and logic engines, so it seems like it can be put into any paradigm (though async Haskell is all within the IO monad, so it's arguable whether it really fits in the functional world).
Re:Hmm...
Ruud on 2005-11-21T21:51:25
I was thinking in the direction of Genetic Algorithms: the programming is in the training.Re:Hmm...
n1vux on 2005-11-21T22:07:10
Genetic algorithms, and also perhaps other relaxation / constraint techniques like the Simplex method of Linear programming?Re:Hmm...
luqui on 2005-11-23T09:10:26
> Genetic algorithms, and also perhaps other
> relaxation / constraint techniques like the
> Simplex method of Linear programming?
The only thing I know about linear programming is that my grad-student friend is taking a course in it and he's working his ass off...:-)
Genetic algorithms are an interesting idea for this control style. My thinking (and my toy implementation) was that you would specify a class of functions, and it would find which function in that class that mapped a given input to a given output. But it could be argued that that's just logic programming (find me function objects that satisfy the constraint that they map this input to this output). Genetic algorithms take a bunch of inputs and outputs and try to find the most general function that covers all of them, which seems closer to the mark, in that it is farther from logic programming. Genetic algorithms are a specific "solution method" to this style of problem. I wonder what the others are...