On Thu, Nov 13, 2014 at 11:23 AM, Andras Slemmer <0slemi0@gmail.com> wrote:
Actually this is a great way to shed light on the difference between monadic and applicative: In the original function the context chaining itself depends on a computed value (short circuiting), meaning it "properly" relies on (>>=). liftM2 (<|>) - or rather liftA2 (<|>) - does not, it doesn't unbox anything, so it cannot possibly be correct.
That's what I meant by "I think 'm' has to be a monad", sorry if I wasn't clear! I admit I'm weak on Alternative, and haven't really found many uses for it, other than the parser combinator thing, so the conversation was interesting anyway. Come to think of it I've used (<|>) for choice and mzero for failure in parsers, so that's really confusing, why one of each? Perhaps because parsec predates Applicative and Alternative. I'm also not too clear on the uses of MonadPlus, aside from being the pure version of my 'try' function, or the relation between Alternative and MonadPlus. There was some discussion recently with the whole AMP thing that implied that it's a historical relic of Applicative coming after Monad, but then some implication that maybe it's not. I guess you'd need a MonadPlus if choice relied on the value inside, but presumably if there's an Alternative, then MonadPlus should have the same implementation.