Using attributed flow graph parsing to recognize Clichés in programs

Abstract

This paper presents a graph parsing approach to recognizing common, stereotypical computational structures, called clichés, in computer programs. Recognition is a powerful technique for efficiently reconstructing useful design information from existing software. We use a flow graph formalism, which is closely related to hypergraph formalisms, to represent programs and clichés and we use attributed flow graph parsing to automate recognition. The formalism includes mechanisms for tolerating variations in programs due to structure sharing (a common optimization in which a structural component is used to play more than one functional role). The formalism has also been designed to capture aggregation relationships on graph edges, which is used to encode aggregate data structure clichés and the abstract operations on them. A chart parsing algorithm is used to solve the problem of determining which clichés in a given cliché library are in a given program.