On the scope of static checking in definitional languages

Abstract

The paper concerns the use in software development of a class of very high level languages characterized as definitional, where a specification of a program consists of assertions expressed as conditional equations. As compared to logic programming, where assertion are expressed in the form of Horn clauses, definitional languages are more amenable to efficient compilation. Also, use of equations is a natural way of expressing mathematical models in science, engineering and economics, where computer simulation of such models is often required. Finally, definitional languages are well suited for programming dataflow machines, another important building block in a Fifth Generation Project. Thus, in many applications definitional languages are preferred choice for programming new generation computers. In using definitional languages to develop programs, a user relies heavily on static analysis during the compilation stage of development, as compared to the traditional practice of relying much more extensively on dynamic analysis performed by execution of the program with test data. As this seems to be the common trend in development of all software tools for Fifth Generation Computers, the limits of static checking are investigated for definitional languages in this paper. A comprehensive approach to selecting and implementing automatic checks in the compiler is proposed. The checking methodology consists of representing the specification of a program by a directed graph and propagating various attributes throughout the entire graph. The choice of attributes and checks depends greatly on the types of errors that users are prone to make. This approach is described in the context of the MODEL language and compiler. The paper reports also on an experiment to evaluate the effectiveness of various checking methods incorporated in the compiler and on reliance of static vs dynamic methods of checking. The types of checking constructed were: syntax analysis, ambiguity in naming, completeness of definitions, data type consistency, dimensionality (corresponds to nesting of loops), ranges of dimensions (corresponds to loop terminations) and circular logic.