> The concepts often apply to less'pure' languages and certainly stretch you to think in new ways.
Purity is great help, but it isn't essential. The essential part is having a type system that actually captures what's going on in your program, without burdening you with manual annotations, and ML-family languages in general do very well in this regard.
I've gotten much more value from purity and immutability than I ever have from advanced types although I'm speaking from a F#/Clojure background rather than a Haskell one.
I'm not talking about fancy types either. Standard ML's type system is dead simple (no higher-kinded types, no GADTs, no type families, not even type classes), and I got a lot more value from it than from Haskell's type system, mostly thanks to:
(0) Abstract types, generated by means of opaque signature ascription, which Haskell doesn't have. This ensures that different modules can't “accidentally” each other's internal invariants.
(1) The fact variables stand for values (due to ML being strict), rather than potentially diverging computations (due to Haskell being lazy), which makes useful algebraic laws sound even in the presence of nontermination or whatever effects.
Purity is great help, but it isn't essential. The essential part is having a type system that actually captures what's going on in your program, without burdening you with manual annotations, and ML-family languages in general do very well in this regard.