Structural Polymorphism

This paper describes structural polymorphism, a new form of type polymorphism appropriate to functional languages featuring user-defined algebraic data types (e.g., Standard ML, Haskell and Miranda1). The approach extends the familiar notion of parametric polymorphism by allowing the definition of functions which are generic with respect to data structures as well as to individual types. For example, structural polymorphism accommodates generalizations of the usual length and map functions which may be applied not only to lists, but also to trees, binary trees or similar algebraic structures. Under traditional polymorphic type systems, these functions may be defined for arbitrary component types, but must be (laboriously) re-defined for every distinct data structure. In this sense, our approach also extends the spirit of parametric polymorphism, in that it provides the programmer relief from the burden of unnecessary repetitive effort. The mechanism we will use to realize this form of polymorphism is inspired by a feature familiar to functional programmers, namely the pattern abstraction. Pattern abstractions generalize the usual lambda abstraction (¬x.e) in that they are comprised of multiple pattern/expression clauses, rather than just a single bound-variable/expression pair. By analogy with pattern abstractions, we generalize polymorphic type abstractions (Òå.e) to type-pattern abstractions, which are comprised of multiple type-pattern/expression pairs. The types given to type-pattern abstractions are universally quantified, just as for traditional type abstractions, but the universal quantifiers are now justified by a recursive analysis of the forms of all possible type instantiations, rather than by parametric independence with respect to a type variable.