If I want to declare a newtype
such that type type of the value is constrained to have an instance for a type-class, it seems like I can do that with:
{-# LANGUAGE RankNTypes #-}
newtype ShowBox = ShowBox (forall a. Show a => a)
GHC compiles that just fine, but when I try and actually use ShowBox
with
ShowBox "hello"
I get a compiler error
<interactive>:1:18:
Could not deduce (a ~ [Char])
from the context (Show a)
bound by a type expected by the context: Show a => a
at <interactive>:1:10-24
`a' is a rigid type variable bound by
a type expected by the context: Show a => a at <interactive>:1:10
In the first argument of `ShowBox', namely `"hello"'
In the expression: ShowBox "hello"
In an equation for `a': a = ShowBox "hello"
Is there a way to make this work?
You're promising the compiler that the value you put inside a ShowBox
will have the type forall a. Show a => a
. There's only one possible value with that type, and it's _|_
. I think you probably want an existential type, which looks rather similar, but means something very different.
{-# LANGUAGE ExistentialQuantification #-}
data ShowBox = forall a. Show a => ShowBox a
This must be done with data
, rather than newtype
. Pattern-matching on the constructor is what brings the Show instance into scope, in this case. Since newtype
s have no run-time representation, they have no place to store the polymorphic witness the existential quantification implies.