Abstract

We present an algorithm which finds typing derivations for ML typeable expressions, such that polymorphic abstraction is minimized where possible. Consequently, unnecessary boxing or code duplication can be avoided, allowing more efficient ML implementations. 1 Introduction This paper has been inspired by the use of typing derivations in the analysis of ML programs. In several applications, polymorphism imposes a strong restriction on the kinds of optimizations one can perform. Unboxing analysis [5] is based on typing derivations. It inserts suitable coercions such that boxed value passing is mostly avoided. Further boxing can be reduced by avoiding unnecessary polymorphism. In the specialization of ad-hoc polymorphic primitives (such as equality), more efficient code can be generated when the types of the primitives are known at compile-time [1]. Tag-free garbage collection [9] also benefits from a typing derivation that avoids polymorphism. Finally, extensions of the results presen...

Citations