I was re-reading "Out of the Tar Pit" earlier last week. It's a paper written by Ben Mosely and Peter Marks, discussing the complexity of software and approaches to software engineering that attempt to reduce software to its essential complexity. This is how I imagined their thought process:
- Functional programming feels really great and seems to cut down the complexity of our code!
- Why is this the case?
- Pure functions can't have side effects and are easy to reason about!
- But you can trivially compose any program of pure functions just by passing the state of the world into every function!
- Wait, that can't be the silver bullet, then... what else do well-designed (read: simple) functional programs have in common?
- OH! Good functional programs are explicit about their management of state!
This was something of a eureka moment for me, because it answered a lingering question I've had about functional programming for some time: if functional programming is the answer to the problem of software complexity, why have some of the best-written and simplest software systems I've seen (see plan 9) been written in simple imperative languages?
I'm going to posit that the effect of managing state well dominates the effect of programming with pure functions as programs get bigger.