The languages of current dynamic constraint detection techniques are often specified by fixed grammars of universal properties. These properties may not be sufficient to express more subtle facts that describe the essential behavior of a given program. In an effort to make the dynamically recovered specification more expressive and program-specific we propose the state space partitioning technique as a solution which effectively adds program-specific disjunctive properties to the language of dynamic constraint detection. In this paper we present ContExt, a prototype implementation of the state space partitioning technique which relies on Daikon for dynamic constraint inference tasks.In order to evaluate recovered specifications produced by ContExt, we develop a methodology which allows us to measure quantitatively how well a particular recovered specification approximates the essential specification of a program's behavior. The proposed methodology is then used to comparatively evaluate the specifications recovered by ContExt and Daikon on two examples.

No abstract

No abstract

No abstract