Abstract

Haskell’s type system provides mechanisms for type refinement within the scope of certain value expressions if GAD Ts or type classes are used. The type system propagates sufficient information to ensure that nothing can go wrong even if types are erased from the run-time representation of a program. This is not the case when we are using existential types, where we deliberately hide concrete types from the type system. Nevertheless, we may desire to eliminate existential types in a different part of a program in order to restore the original types.

For this purpose, we propose an extension to Haskell which allows programmers to restrict existential types within individual data constructors to finite, but open, domains of types. Each type in such a domain must be associated with a value tag that is then stored at run time to allow it to serve as witness in a case expression.