[Haskell-cafe] Ambiguous type variable

[Jacek Generowicz]

On 2010 Oct 16, at 01:39, Daniel Fischer wrote:

> On Saturday 16 October 2010 01:18:55, Jacek Generowicz wrote:
>> On 2010 Oct 16, at 01:14, Jacek Generowicz wrote:
>>> On 2010 Oct 16, at 00:51, Ivan Lazar Miljenovic wrote:
>>>> "2" is a generic number.  If you don't specify a type, it usually
>>>> defaults to Integer.  All Num instances that come in the Prelude  
>>>> have
>>>> Show instances, so no matter which gets picked "show 2" works.
>>>> However, when you say "view 2" ghc/ghci doesn't know that you  
>>>> want 2
>>>> to be an Int (as that's the only type you have an instance for View
>>>> for).
>>>
>>> On 2010 Oct 16, at 00:51, Christopher Done wrote:
>>>> Don't integral literals default to Integer, of which there is a  
>>>> Show
>>>> instance but no View instance?
>>
>> An in both of the explanations above, it should then complain about
>> the lack of an instance of View, rather than about ambiguous type
>> variables, n'est-ce pas?
>
> Non.
>
> If you write
>
> x = view 2
>
> the type checker looks at the expression 2 and it sees it has the type
>
> Num a => a
>
> (because integer literals are actually shorthand for (fromInteger  
> literal)
> and fromInteger :: Num a => Integer -> a).
>
> Then (or before that) it looks at the context in which the 2 appears.
> That context is that the function view is applied to it.
>
> view :: View v => v -> String
>
> So, in order for the expression 2 to be well typed, the type checker  
> finds
> the two constraints
>
> 2 :: Num a => a
> 2 :: View v => v
>
> These constraints have to be unified (which is very easy here),  
> resulting
> in
>
> 2 :: (Num a, View a) => a
>
> But there's no way to find out what type a is/has to be.
> Hence a is an ambiguous type variable.

Exactly. Which is why I made the point that the two explanations  
offered by Christopher and Ivan (both of which suggested that the  
problem was related to a missing View instance) imply that the error  
report should mention missing View instances: As the error message  
does not mention them, I took this as further evidence that those  
suggestions were not correct.

> Now, under certain circumstances such ambiguous type variables are  
> resolved
> by defaulting. If you replace view with show, you get a Show  
> constraint
> instead of the View constraint and then defaulting may (and must)  
> happen.
> But since View is defined outside the standard libraries, the language
> report says defaulting mustn't take place, so it doesn't

Yup. That much was clear from you last message.

> (it may work if
> you specify
> {-# LANGUAGE ExtendedDefaultRules #-}
> at the top of your module).

Interesting. Thanks.