On Friday 10 July 2009 4:35:15 am Wolfgang Jeltsch wrote:
I fear that this instance doesn’t satisfy required laws. As far as I know, the following equalities should hold:
(*>) = (>>)
f *> empty = empty
IO already fails at this law, because (f *> empty) is not the same as empty, it is a failing computation with the side effects of f. empty is only a left- zero in the IO monad.
This implies the following:
(f >> empty) <|> g = g
But this wouldn’t hold with your instance. (f >> empty) <|> g would cause the side effects of f and of g, while g would (obviously) only cause the side effects of g.
Of course, this can be seen as a different symptom of the same underlying problem (although if the initial equation for empty held, it'd work). There's been talk before of splitting MonadPlus into multiple classes for choice, which pick and choose which of these sorts of laws apply to each particular class. MonadPlus vs. MonadOrElse is one splitting, for instance, but I can't recall if IO would satisfy all the laws for either.* -- Dan * For instance, you might expect MonadOrElse to satisfy: f <|> g = f OR f <|> g = g but IO will only satisfy: f <|> g = f OR f <|> g = g-with-side-effects-from-f