Chain programs for writing deterministic metainterpreters

摘要

Many metainterpreters found in the logic programming literature are nondeterministic in the sense that the selection of program clauses is not determined. Examples are the familiar 'demo' and 'vanilla' metainterpreters. For some applications this nondeterminism is convenient. In some cases, however, a deterministic metainterpreter, having an explicit selection of clauses, is needed. Such cases include (1) conversion of OR parallelism into AND parallelism for 'committed-choice' processors, (2) logic-based, imperative-language implementation of search strategies, and (3) simulation of bounded-resource reasoning. Deterministic metainterpreters are difficult to write because the programmer must be concerned about the set of unifiers of the children of a node in the derivation tree. We argue that it is both possible and advantageous to write these metainterpreters by reasoning in terms of object programs converted into a syntactically restricted form that we call 'chain' form, where we can forget about unification, except for unit clauses. We give two transformations converting logic programs into chain form, one for 'moded' programs (implicit in two existing exhaustive-traversal methods for committed-choice execution), and one for arbitrary definite programs. As illustrations of our approach we show examples of the three applications mentioned above.