On 3/27/07, Jean-Marie Gaillourdet
I concur. The class declares T as being a ternary relation such that the following holds forall r p p' s s'. T(r,p,s) && T(r,p',s') -> s = s' Now, the instance `T root (Any root) sel' is satisfied when root=Int, sel = Bool and when root=Int, sel = Int. Does it not? That falsifies the above proposition. In other words, the instance T is not functional with respect to the first and the third arguments.
That is not surprising. What is surprising is why GHC 6.6 accepts such an instance?
GHC 6.6 does not accept such instances. Add the following code to the module TestCase
instance T Int Int Int instance T Int Int Bool
I'm not an expert as well, but I think oleg was actually talking about
instance T Int (Any Int) Int instance T Int (Any Int) Bool
which also fails (on GHC 6.6) with: Functional dependencies conflict between instance declarations: instance T Int (Any Int) Int -- Defined at ... instance T Int (Any Int) Bool -- Defined at ... although just
instance T root (Any root) sel
is accepted. -- Felipe.