Resolving overloading loops for circular constraint graph

[Stefan Holdermans]

Dear Martin,

> By "black-holing" you probably mean co-induction. That is,
> if the statement to proven appears again, we assume it must hold.
> However, type classes are by default inductive, so there's no
> easy fix to offer to your problem.

A propos: are there fundamental objections to coinductive resolving,  
i.e., constructing recursive dictionaries. I suppose termination is  
hard to guarantee then: is it enough to require constraints to be  
productive w.r.t. instance heads?

Cheers,

   Stefan