Type Specialisation of a Subset of Haskell Per Sjörs

John Hughes presents a new method for performing partial evaluation in [Hug96b]. The method is called type specialisation and functions much like a type checker. It infers its results. The static content of every sub-expression of the source program is derived. By letting this static content be the type of the expression, new specialised types are produced. The types are then propagated through the specialisation process independently of how code is residualised. This enables strong specialisations. This paper describes an implementation with a subset of Haskell as both the source and the residual (target) language. It is capable of handling Haskell’s data types, including specialising constructors. One problem with Hughes ’ specialiser was that it could not handle static tuples and projections on them properly. Here, a solution is presented; a post-phase called projection unfolding. The method is capable of removing all static tuples provided neither the type of the program nor a sub-type of the type o