Best-first Strategies for Incremental Transformations of Logic Programs

Abstract

We propose some techniques for mechanizing a class of rule-based transformations of logic programs. Analogous techniques can be applied also in the case of functional programs for which the program transformation methodology was first introduced [4].

We consider the following abstract Transformation Problem: given a program P and a property Φ, we are required to find a program, say TransfP, equivalent to P such that Φ holds for TransfP. The property Φ characterizes the desired ‘shape’ of the program to be derived (for instance, linear recursive, tail recursive, without intermediate variables, etc.).

We propose some restrictions on the property Φ which make it possible to perform the desired transformation in an incremental way, in the sense that, if a clause which does not satisfy Φ is transformed into clauses which do satisfy Φ, then these resulting clauses need not be considered again during the transformation process.

The core of our technique is a procedure, called Incremental Transformation, which is a generalization of the procedure for eliminating unnecessary variables introduced in [13]. The generalization basically consists in the fact that we have parametrized our transformation procedure w.r.t. the property Φ which the transformed program is required to satisfy.

The proposed procedure is nondeterministic and it generates a possibly infinite tree of programs which are all derivable from the given program P. We study the use of best-first search strategies for traversing this tree of programs. We finally provide a sufficient condition for solving a given transformation problem by using a best-first strategy, in case our Incremental Transformation procedure can find a solution by using any other strategy.

This work has been partially supported by the ‘Progetto Finalizzato Sistemi Informatici e Calcolo Parallelo’ of the CNR and MURST 40%, Italy under grant n. 89.00026.69.