[Haskell-cafe] Printing and Referential transparency excuse

[Derek Elkins]

On Mon, 2007-12-24 at 11:15 +0200, Cristian Baboi wrote:
> While reading the Haskell language report I noticed that function type is  
> not an instance of class Read.
> 
> I was told that one cannot define them as an instance of class Show  
> without breaking "referential transparency" or printing a constant.
> 
>    f :: (a->b)->String
>    f x = "bla bla bla"
> 
> How can I define a function to do the inverse operation ?
>    g :: String -> ( a -> b )
> 
> This time I cannot see how referential transparency will deny it.
> What's the excuse now ?

Referential transparency has nothing to with this.  The problem here is
parametricity.  However, if you give a non-parametric type (or add
appropriate class constraints on a and b, or add (phantom) type
information to the input), e.g. String -> (Int -> Int) then there is no
trouble at all.

The compiler (or something non-portably using extremely unsafe features)
could provide a primitive String -> a that reads in a string and
interprets it somehow as a value of type a.  If the value represented by
the string doesn't match the type that a is instantiated to, something
bad is going to happen.  Adding such a primitive would utterly destroy
parametricity.