Generalized behavior-based retrieval [from a software reuse library]

The user of a large reuse library faces the formidable discovery problem of searching for all and only those components useful in solving the current programming task. The author describes a retrieval technique that generalizes the simple idea of executing each component on test inputs, and reporting those that compute correct outputs. It exploits the precise extensional semantics of executable code to provide the canonicality in query formulation that is missing in information-retrieval approaches to reuse. One generalization improves recall by considering small programs that are constructible from library components, rather than just single components. Furthermore, functional modeling of components allows the technique to handle complex behaviors, such as side effects. The author analyzes the technique and presents a working prototype, which has been tested on two libraries: one containing general programming, and the other containing some Unix shell commands