Hi Cafe, Here is another example of why 'let' should be sometimes generalised. I've been recently playing with indexed monads (for JSON processing) and found out that the following code fails to typecheck:
{-# LANGUAGE MonoLocalBinds #-}
data M t t' a = M
ipure :: a -> M t t a ipure a = M iseq :: M t t' a -> M t' t'' b -> M t t'' b iseq a b = M
np :: M () Bool () np = M
test = p `iseq` np `iseq` p where p = ipure ()
Test.hs:13:27: Couldn't match expected type `Bool' with actual type `()' Expected type: M Bool t''0 b0 Actual type: M () () () In the second argument of `iseq', namely `p' In the expression: p `iseq` np `iseq` p In practice that means that I need to provide a signature for almost every monadic local binding that is used more than once, which is unbearable, especially when monad transformers and complex indices are used.http://hackage.haskell.org/trac/ghc/blog/LetGeneralisationInGhc7