[Haskell-cafe] A possibly stupid doubt about the GHC's overlapping instances flag's

[Daniel Fischer]

Am Mittwoch, 24. Oktober 2007 10:35 schrieb Bas van Dijk:

> Suppose you have:
>
>
> {-# OPTIONS_GHC -fglasgow-exts  -fallow-overlapping-instances #-}
>
> class    C a   b     where foo :: a -> b -> (a, b)
>
> instance C Int a     where foo n x  = (n+1, x)           -- (A)
> instance C a   Bool  where foo x b  = (x,   not b)       -- (B)
> instance C Int [a]   where foo n xs = (n+1, xs)          -- (C)
> instance C Int [Int] where foo n ns = (n+1, map (+1) ns) -- (D)
>
> f :: [b] -> [b]
> f xs = snd $ foo (1 :: Int) xs
>
>
> In the right hand sight of 'f', 'foo' is applied to an Int and a [b]
> so it seems that instance C should match. However GHC rejects this
> program because in a later call 'f' can be applied to  a list of Ints
> (like in: g = f ([1,2,3] :: [Int])) by which 'b' instantiates to an
> Int, by which instance D should really match.
>
> If you enable -fallow-incoherent-instances then 'f' will use instance
> C without complaining about the problem of subsequent instantiations.
>
> However if you then define 'g' you will get the error:
>
> Couldn't match expected type `Int' against inferred type `[a]'
>     In the first argument of `f', namely `([1, 2, 3] :: Int)'

This seems to be a typo. 
g = f ([1,2,3] :: [Int]) is accepted.

g = f ([1,2,3] :: Int) can never be, overlapping/incoherent instances or not

>
> regards,
>
> Bas.

Cheers,
Daniel