I am a bit confused by GHC behaviour for the following variants of a program: data A = A (Int -> A) f = A g g _ = f h = g 'c' ghc:

Error: Couldn't match `Int' against `Char'

The problem is that a small change ommits the error: data A = A (Int -> A) f = A g where g _ = f h = g 'c' ghc:

Ok, modules loaded

But I really need the functions to be top-level, thus I added type signatures: data A = A (Int -> A) f :: A f = A g g :: a -> A g _ = f h :: A h = g 'c' ghc:

Ok, modules loaded

Fine, but what I really need, is this: data A = A (Int -> A) f :: A f = A g g :: Ord a => a -> A g _ = f h :: A h = g 'c'

Contexts differ in length The signature contexts in a mutually recursive group should all be identical ... Couldn't match `Int' against `Char'

Again the problem does not occur for local functions: data A = A (Int -> A) f :: A f = A g where g :: Ord a => a -> A g _ = f h :: A h = g 'c' Why is ghc treating local and global functions differently? Is there really a difference in decidability of the Polymorphic Type Inference? And regarding the "contexts differ in length", I read a comment in "Typing Haskell in Haskell":

a throw-away comment specifying that all explicit type signatures in a binding group must have the same context up to renaming of variables [10,Section 4.5.2]. This is a syntactic restriction that can easily be checked prior to type checking. Our comments here, however, suggest that it is unnecessarily restrictive.

Why does ghc still use that unnecessary restriction? Cheers Bernd