Functional encapsulation and type reconstruction in a strongly-typed, polymorphic language

Static type systems are traditionally used to prevent run-time type-errors in user programs and to assign appropriate storage representations to objects during compilation. In this thesis, we explore some new ways of using static type information in the design, compilation, and execution of programs written in a strongly-typed, polymorphic language. Programmers often nd it useful to know whether or not a particular data-structure may be updated outside a given control block. Information about an object's non-mutability helps compiler optimizations, improves aliasing and dependence analyses, and permits unrestricted caching of functional data at run-time. In the rst part of this thesis, we present a safe, static mechanism for functional encapsulation of imperative data-structures using a powerful type system based on closure types and regions. We introduce a new language construct called close which delimits the scope of side-e ects on imperative objects and converts them into functional objects outside that scope. This mechanism may be used to build e cient, highlevel, functional data-abstractions within a language using its low-level, imperative constructs. Type-safety and non-mutability of closed objects is guaranteed by a semantic soundness theorem that ensures consistency between the static and the dynamic semantics. The type system is presented in the context of Id, which is a strongly-typed, polymorphic, higher-order language, and it easily simpli es to a rst-order, monomorphic language such as C or Fortran. In the second part of the thesis, we develop a general, compiler-directed methodology for complete type reconstruction of run-time objects in a polymorphic language without using any run-time type-tags. Run-time type reconstruction is carried out by instantiating static type information for each function activation frame present within the dynamic call tree. Additional type-hints are inserted automatically at compile-time and are decoded at run-time to ensure complete type reconstruction. We present the necessary compiler analysis and the type reconstruction algorithm and prove their correctness. This technique has been used successfully for displaying run-time objects within the Id source debugger for Monsoon and to perform tagless garbage collection in the *T architecture. We describe the latter application in detail, comparing its performance with other schemes for automatic storage reclamation.