Admitting that Functional Programming Can Be Awkward

My initial interest in functional programming was because it seemed so perverse.

At the time, I was the classic self-taught programmer, having learned BASIC and then 6502 assembly language so I could implement my own game designs. I picked up the August 1985 issue of Byte magazine to read about the then-new Amiga. It also happened to be the issue on declarative languages, featuring a reprint of Backus's famous Turing Award Lecture and a tutorial on Hope, among other articles.

This was all pretty crazy stuff for an Atari 800 game coder to be reading about. I understood some parts, completely missed vast swaths of others, but one key point caught my imagination: programming without modifiable variables. How could that possibly work? I couldn't write even the smallest game without storing values to memory. It appealed to me for its impossibility, much in the way that I had heard machine language was too difficult for most people to approach. But while I had pored over assembly language listings of games in magazines, and learned to write my own as a result, there wasn't such direct applicability for functional programming. It made me wonder, but I didn't use it.

Many years later when I first worked through tutorials for Haskell, Standard ML, and eventually Erlang, it was to figure out how programming without modifying variables could work. In the small, it's pretty easy. Much of what seemed weird back in 1985 had become commonplace: garbage collection, using complex data structures without worrying about memory layout, languages with much less bookkeeping than C or Pascal. But that "no destructive updates" thing was--and still is--tricky.

I suppose it's completely obvious to point out that there have been tens of thousands of video games written using an imperative programming style, and maybe a handful--maybe even just a couple of fingers worth--of games written in a purely functional manner. Sure, there have been games written in Lisp and some games written by language dilettantes fond of Objective Caml, but they never turn out to be programmed in a functional style. You can write imperative code in those languages easily enough. And the reason for going down that road is simple: it's not at all clear how to write many types of complex applications in functional languages.

Usually I can work through the data dependencies, and often I find that there's an underlying simplicity to the functional approach. But for other applications...well, they can turn into puzzles. Where I can typically slog through a messy solution in C, the purely functional solution either eludes me or takes some puzzling to figure out. In those cases I feel like I'm fighting the system, and I realize why it's the road less traveled. Don't believe it? Think that functional purity is always the road to righteousness? Here's an easy example.

I wrote a semi-successful Mac game a while back called Bumbler. At its heart it was your standard sprite-based game: lots of independent objects running some behavioral code and interacting with each other. That kind of code looks easy to write in a purely functional way. An ant, represented as a coordinate, marches across the screen in a straight line and is deleted when it hits the opposite screen edge. That's easy to see as a function. One small clod of data goes in, another comes out.

But the behaviors and interactions can be a lot more tangled than this. You could have an insect that chases other insects, so you've got to pass in a list of existing entities to it. You can have an insect that affects spawn rates of other other insects, but of course you can't modify those rates directly so you've got to return that data somehow. You can have an insect that latches onto eggs and turns them into something else, so now there's a behavioral function that needs to reach into the list of entities and make modifications, but of you're not allowed to do that. You can have an insect that modifies the physical environment (that is, the background of the game) and spawns other insects. And each of these is messier than it sounds, because there are so many counters and thresholds and limiters being managed and sounds being played in all kinds of situations, that the data flow isn't clean by any means.

What's interesting is that it would be trivial to write this in C. Some incrementing, some conditions, direct calls to sound playing routines and insect spawning functions, reading and writing from a pool of global counters and state variables. For a purely functional approach, I'm sure the data flow could be puzzled out...assuming that everything was all perfectly planned and all the behaviors were defined ahead of time. It's much more difficult to take a pure movement function and say "okay, what I'd like is for this object to gravitationally influence other objects once it has bounced off of the screen edges three times." Doable, yes. As directly implementable as the C equivalent? No way.

That's one option: to admit that functional programming is the wrong paradigm for some types of problems. Fair enough. I'd put money on that. But it also may be that almost no one has been thinking about problems like this, that functional programming attracts purists and enamored students. In the game example above, some of the issues are solvable, they just need different approaches. Other issues I don't know how to solve, or at least I don't have solutions that are as straightforward as writing sequential C. And there you go...I'm admitting that functional programming is awkward in some cases. It's also extremely useful in others.

(Also see the follow-up.)