Re: [Haskell-cafe] Simple GADTs, type families and type classes combination with type error.

2011/7/22 Dan Doel :

On Fri, Jul 22, 2011 at 11:12 AM, Serguey Zefirov wrote: GHC cannot decide what instance of FuncVars to use. The signature of funcVars is:    funcVars :: FuncVars cpu => CPUFunc cpu -> [String]

This does not take any arguments that allow cpu to be determined. For instance, if there were instances (rolling them into one declaration for simplicity):

   instance FuncVars Int where      type CPUFunc Int = Int      ...

   instance FuncVars Char where      type CPUFunc Char = Int

Then GHC would see that CPUFunc cpu = Int, but from this, it cannot determine whether cpu = Int or cpu = Char. CPUFunc is not (necessarily) injective.

But "cpu" variable is the same in all places. If we don't dive into CPUFunc reduction (to Int or whatever) we can safely match funcVars argument and unify cpu. This is the case when we write generic functions over type family application.

Also, if you have a class whose only content is an associated type, there's really no need for the class at all. It desugars into:

   type family CPUFunc a :: *

   class CPU a

It would be somewhat inconvenient. I omitted some constraints in CPU class for the sake of presentation.