On 03/01/06, Donn Cave
On Tue, 3 Jan 2006, Chris Kuklewicz wrote: ...
I sometimes call a function with side-effects in IO a "command". But the terms are fungible. But calling putStr a "function" is correct. It is not a "pure function" however.
Is that the standard party line? I mean, we all know its type and semantics, whatever you want to call them, but if we want to put names to things, I had the impression that the IO monad is designed to work in a pure functional language - so that the functions are indeed actually pure, including putStr. It's the monad that actually incurs the side effects. Or something like that. So it isn't at all necessary to have another word for functions of type IO a.
I'd say it depends on the way that you're thinking about it. Strictly speaking, putStr is a pure function which returns an action. It's also referentially transparent, since it always returns the same action for the same string. However, if you think of functions a -> IO b as the arrows a -> b in a new category where composition is Kleisli composition: (@@) :: Monad m => (a -> m b) -> (t -> m a) -> (t -> m b) y @@ x = \u -> x u >>= y Then these arrows are "effectful" and would be considered impure. I think I prefer the first explanation, as it's a somewhat important property that the same action is computed for the same input. This doesn't hold of side-effectful arrows in general. Note that with Hughes' Arrows, this restriction can (often quite cleverly) be side-stepped in order to make some rather major optimisations. - Cale