A treatment of higher-order features in logic programming

The logic programming paradigm provides the basis for a new intensional view of higherorder notions. This view is realized primarily by employing the terms of a typed lambda calculus as representational devices and by using a richer form of unification for probing their structures. These additions have important meta-programming applications but they also pose non-trivial implementation problems. One issue concerns the machine representation of lambda terms suitable to their intended use: an adequate encoding must facilitate comparison operations over terms in addition to supporting the usual reduction computation. Another aspect relates to the treatment of a unification operation that has a branching character and that sometimes calls for the delaying of the solution of unification problems. A final issue concerns the execution of goals whose structures become apparent only in the course of computation. These various problems are exposed in this paper and solutions to them are described. A satisfactory representation for lambda terms is developed by exploiting the nameless notation of de Bruijn as well as explicit encodings of substitutions. Special mechanisms are molded into the structure of traditional Prolog implementations to support branching in unification and carrying of unification problems over other computation steps; a premium is placed in this context on exploiting determinism and on emulating usual first-order behaviour. An extended compilation model is presented that treats higher-order unification and also handles dynamically emergent goals. The ideas described here have been employed in the Teyjus implementation of the λProlog language, a fact that is used to obtain a preliminary assessment of their efficacy.