Instance instantiation is *not* search, and *not* similar to subclassing in OO languages. Both your instance declarations simply add constraints to functions that use it. Here's a more concrete example: class A a where doA :: a -> String class R a where doR :: a -> String instance R Int where doR = show instance R a => A a where doA = doR question x = doA x The question is, what is the type of "question"? Here are two valid choices with this program: question :: (A a) => a -> String question :: (R a) => a -> String The instance R a => A a says that every type "a" is an instance of "A"; if an instance for A is needed, the compiler says "OK, I know how to make one of those. But I now add a new constraint, R a." Adding another instance S a => A a makes the choice of what constraint to add ambiguous. In particular the following code does *not* work: class S a where doS :: a -> String instance S String where doS = id instance S a => A a where doA = doS question2 = question (2::Int) question3 = question "3" In my experience, if you are turning undecidable instances on and you don't know exactly why it's safe to do so, there is probably a mistake in your design. -- ryan 2008/10/24 Alberto G. Corona :

with:

...

{-# OPTIONS -fglasgow-exts -fallow-undecidable-instances #-}

...

class A a class R a

...

class S a

...

instance R a => A a

...

instance S a => A a

---

GHC gives

Duplicate instance declarations instance R a => A a instance S a => A a

Why? _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

On Fri, Oct 24, 2008 at 10:20:39AM +0200, Alberto G. Corona wrote:

with:

...

{-# OPTIONS -fglasgow-exts -fallow-undecidable-instances #-}

...

class A a class R a

...

class S a

...

instance R a => A a instance S a => A a

---

This point commonly trips people up (it used to trip me up). In addition to what others have said, I hope I can provide a bit more intuition. If you have

instance R a => A a

it *seems* like it should mean, "Any a which is an instance of R is also an instance of A". However, that's *not* what it means! It actually means "Any type which matches 'a' (that is, any type at all) can be an instance of A; and if some type is used as an instance of A, then it must also be an instance of R as well." GHC picks which instance to use by looking *only* at stuff to the right of => in instance declarations. Only after an instance has been chosen is the stuff to the left of the => considered. Hopefully now it is clear why the code above is a duplicate instance declaration; there's no way to distinguish the instances by looking only at stuff to the right of => . -Brent