An intensional implementation technique for functional languages
The potential of functional programming languages has not been widely accepted yet. The reason lies in the difficulties associated with their implementation. In this dissertation we propose a new implementation technique for functional languages by compiling them into 'Intensional Logic' of R. Montague and R. Carnap. Our technique is not limited to a particular hardware or to a particular evaluation strategy; nevertheless it lends itself directly to demand-driven tagged dataflow architecture. Even though our technique can handle conventional languages as well, our main interest is exclusively with functional languages in general and with Lucid-like dataflow languages in particular. We give a brief general account of intensional logic and then introduce the concept of intensional algebras as structures (models) for intensional logic. We, formally, show the computability requirements for such algebras. The target language of our compilation is the family of languages DE (definitional equations over intensional expressions). A program in DE is a linear (not structured) set of non-ambiguous equations defining nullary variable symbols. One of these variable symbols should be the symbol result. We introduce the compilation of Iswim (a first order variant of Landin's ISWIM) as an example of compiling functions into intensional expressions. A compilation algorithm is given. Iswim(A), for any algebra of data types A, is compiled into DE(Flo(A)) where Flo(A) is a uniquely defined intensional algebra over the tree of function calls. The approach is extended to compiling Luswim and Lucid. We describe the demand-driven tagged dataflow (the eduction) approach to evaluating the intensional family of target languages DE. Furthermore, for each intensional algebra, we introduce a collection of rewrite rules. A justification of correctness is given. These rules are the basis for evaluating programs in the target DE by reduction. Finally, we discuss possible refinements and extensions to our approach.