Function type generalization

[girving]

If we apply a higher order function like "map" to an overloaded closure like negate and an int list, the type inference algorithm currently infers a type like

map :: negate -> [Int] -> [Int]

This means that if you apply map to a different function, like square, it infers a separate judgement

map :: square -> [Int] -> [Int]

This can be serious performance hit, causing degradation by factors of up to infinity. It would be much better to notice that in each case map only applies the closure to arguments of a single type, and infer the generalization

map :: (Int -> Int) -> [Int] -> [Int]

Note that this almost happens automatically if "map" has a standard type signature.