Deriving Verification Conditions and Program Assertions to Support Software Inspection

Abstract

In order to trust a reusable software component, the correctness of its implementation with respect to its specification must be assured. Formal proof of correctness, while offering this assurance, is an often difficult, if not impractical, goal to achieve. Formal code reading techniques employed in software inspection have proved useful as a human reasoning process to verify correctness with a high degree of assuredness. Such techniques rely on the reader being able to abstract the semantics of a given component in order to reason about its correctness. The paper presents a method and supporting tools which yield formal semantic properties directly from the syntax of the component's code. The method includes an extension of existing algorithmic and heuristic invariant generation techniques. Although, in many cases, the semantic information derived is strong enough to be useful as program assertions in Hoare-style formal proofs, we focus on the generation of information to assist human reasoning based code reading processes. To this end, we use examples to illustrate the application of the method and prototype tools to yield semantic information directly from program code, such as first-order predicate calculus specifications, input-output diagrams, and constructive reports of loop termination conditions.