[dwhitty]

The methodology espoused in this article presents a good case for composing programs from discrete pieces of behavior with referential transparency (pure, or at least deterministic functions).

"Sometimes, the elegant implementation is just a function. Not a method. Not a class. Not a framework. Just a function."

- John Carmack (Twitter) http://twitter.com/#!/ID_AA_Carmack/status/53512300451201024

Not "sometimes", "mosttimes"!

[bad_user]

Carmack works on game engines and a game environment is by definition an environment in which objects have side effects and keep a lot of state. Referential transparency is nice and all, but developing games in this style (while doable with functional reactive programming) is a pain in the ass.

Here's a good series of articles about it: http://prog21.dadgum.com/23.html

[pjscott]

According to Tim Sweeney, the lead developer at Epic Games, there are three types of code in a modern first-person shooter: gameplay simulation, numeric computation, and shading. You've described gameplay simulation. Shading is very data-parallel. And the numeric code is essentially functional.

Even for gameplay code, it could be nice to have explicit tracking of what the mutable state is, to prevent bugs. He has some fascinating slides about this, if you google for "The Next Mainstream Programming Language". Extensive discussion on Lambda the Ultimate here:

http://lambda-the-ultimate.org/node/1277

Definitely one of the most interesting things I've seen in a while.

[dwhitty]

It is a perfectly defensible goal to have some or many pure functions within an imperative programming environment where state is persisted in variables outside of the library of pure functions. I don't think it is fair to equate pure functions with pure functional programming.

In a mixed model, you can have a mutable state but still have a large bank of trustable, testable, parallelizable behaviors defined in pure or deterministic functions with well defined contracts of input and output states.

[Periodic]

I think you could go a little further and say that it implies dependency injection. You want to be able to declaratively tell the program which implementation to use in difference places. This would be a little more familiar to the OO paradigms than pure/deterministic functions. However, I still argue that pure and deterministic functions have many desirable properties.