SYNAPSE- A System for Logic Program Synthesis from Incomplete Specifications

A logic program development methodology has been proposed by the second author [3]. It aims at programming-in-the-small, and is (mainly) meant for “algorithmic” problems. It starts from a complete, yet informal, specification of the target problem. A crucial step is the design of the logic of the program, based solely on the declarative semantics of logic, and with exclusive concern about correctness issues. Procedural semantics and efficiency are taken care of in a later optimization and implementation step. The methodology gives rise to many computer-assistance, if not automation, opportunities. The FOLON environment [9] aims at this, and is described elsewhere in this volume. But, as specifications are non-formal, the crucial step cannot be automated. This research thus investigates an alternative approach to logic program development, namely (fully) automated logic program synthesis. This requires formal specifications, and our choice went towards exploring synthesis from incomplete specifications. The resulting system, called SYNAPSE, is being integrated into the FOLON environment. Let R be the relation one has in mind when elaborating a specification of a procedure for predicate r. We call R the intended relation, in contrast to the relation actually specified, called the specified relation. This distinction is important in general, but crucial with incomplete specifications, where one deliberately admits a gap between the two. We assume R is known, even if we don’t have a formal definition of it. In our approach [5], incomplete specifications are expressed with examples and properties. More precisely, a specification by examples and properties of a relation r consists of: • a set of examples of r (ground atoms); and • a set of properties of r (non-recursive Horn clauses). The specified relation, that is the set of logical consequences of the given examples and properties, is assumed to be a subset of the intended relation R. Examples have been around for quite a while as an incomplete specification formalism. Their attractiveness lies in their naturalness and conciseness, but their weaknesses are ambi-