[Haskell-beginners] multi parameter classes and instance declarations

[Daniel Fischer]

Am Montag, 16. Februar 2009 15:40 schrieb Markus Barenhoff:
> On Mon 16.02 15:28, Markus Barenhoff wrote:
> > Good Afternoon,
> >
> > i'am trying to express something like this here:
> > > -- CI.hs http://pastebin.com/f5b17cf57
> > > {-# OPTIONS -XMultiParamTypeClasses #-}
> > >
> > > module CI(D,f) where
> > >
> > > class B b
> > >
> > > class (B b) => A a b where
> > >     f :: b
> > >
> > > data D = D
> > >
> > > instance B D
> > >
> > > instance A D D where
> > >     f = D
> > > -- CI.hs
> > >
> > > -- MultiClassTest.hs http://pastebin.com/f1c481e8d
> > > import CI
> > >
> > > a = f
> > > -- MultiClassTest.hs
>
> I forgot to mention that constrain a :: D does leed to the same
> ghc messages.
>
> > ghc output:
> > MultiClassTest.hs:4:4:
> >     No instance for (CI.A a D)
> >       arising from a use of `f' at MultiClassTest.hs:4:4
> >     Possible fix: add an instance declaration for (CI.A a D)
> >     In the expression: f
> >     In the definition of `a': a = f
> > Failed, modules loaded: CI.
>

GHC doesn't know that there's no other 

instance A a D

besides the given instance A D D, so it doesn't know which instance to select
(suppose there were an instance A Bool D where f = undefined, then it's clear 
that the instance to use can't be decided).
Since class A is not exported, there can't be without changing CI, so one 
might think it would be reasonable to use the only existing instance of A and 
be done with it. However, GHC treats type classes with an open world 
assumption (i.e. doesn't select instances according to which instances it 
currently knows about, but what instances there potentially might be).

To make f usable at all, you must make the type a somehow reachable from f 
(e.g. by adding a functional dependency:
{-# LANGUAGE FunctionalDependencies #-}

class (B b) => A a b | b -> a where
	f :: b

-- the | b -> a part means that type b uniquely determines type a
)

> Markus

Cheers,
Daniel