Lambda the Ultimate

There exist various approaches to overloading, including:

* type classes of various kinds
* traits
* implicits [2]
* multimethods
* other systems [1,3]

What seems to offer the best value for investment effort, expressiveness, and type safety?

For instance, multimethods are pretty flexible and uniformally allow a form of type-based and case-based overloading, but it seems quite difficult to capture accurate types with all the expressiveness we expect of ML functions, eg. first-class functions require intersection types and subtyping.

Type classes and traits seem more or less equivalent, but have problems with overlapping and named instances, a problem shared by [1] I believe. Implicits seem promising, and properly handle named and overlapping instances.

[1] A Second Look at Overloading
[2] The Implicit Calculus: A New Foundation for Generic Programming
[3] A Theory of Overloading