[Haskell-cafe] Specifying a function type in a where clause.

[Daniel Fischer]

Am Freitag, 19. September 2008 00:33 schrieb Rob deFriesse:
> I would like to know why I'm getting a particular compile time error
> message.
>
> In this program, I am specifying a function type on /combs'/ in the
> where clause:
>
> --------
> module Main where
>
> import List (delete)
>
> combs :: Eq a => [a] -> Int -> [[a]]
> combs l 1 = map (\x -> [x]) l
> combs l n = foldl combs' [] l
>     where combs' :: Eq a => [[a]] -> a -> [[a]]
>           combs' acc x = let sl = delete x l in (map (\i -> x:i) $ combs
> sl (n-1)) ++ acc
>
> main = do
>   print $ combs ["a","b","c","d"] 3
> --------
>
> I get this error message from GHC 6.8.3:
>
> cafe1.hs:9:43:
>     Couldn't match expected type `a1' against inferred type `a'
>       `a1' is a rigid type variable bound by
>            the type signature for `combs'' at cafe1.hs:8:23
>       `a' is a rigid type variable bound by
>           the type signature for `combs' at cafe1.hs:5:12
>       Expected type: [a1]
>       Inferred type: [a]
>     In the second argument of `delete', namely `l'
>     In the expression: delete x l
>
> I don't understand why I'm seeing this.  The type /a/ is the same type
> in /combs/ and /combs'/.

Not quite. As it is, the type signature of combs' promises that for all types 
b which belong to Eq, combs' has type [[b]] -> b -> [[b]], because the a from 
the top level type signature is not in scope. Therefore, ghc complains about 
the use of l in the definition of combs', since l has a specific type.
You can either remove the signature for combs' or bring the type variable a 
into scope by the pragma {-# LANGUAGE ScopedTypeVariables #-} at the top of 
your file or the flag -XScopedTypeVariables on the command line and changing 
the signature of combs to
combs :: forall a. (Eq a) => [a] -> Int -> [[a]],
I think then you must leave off the (Eq a) from the signature of combs'.

>  I realize I can remove the type specification
> from /combs'/, and the code will compile.  However, I'd like to get a
> better understanding of why GHC objects to this.
>
> Thank you,
> Rob.