Using Program Families for Maintenance Experiments

In general, new modularization techniques require a significant intellectual investment from practitioners in order to adopt them. Before practitioners are willing to make such an investment, they want a careful scientific assessment of the technique for a number of properties (e.g., effects on reusability, reliability, and maintainability). Our work is concerned with conducting such empirical investigations for assessing how modularization techniques affect maintainability; in particular, this paper presents an approach to conducting formal experiments for assessing a technique's impact on perfective maintenance. We refer to such experiments as maintenance experiments in the sequel. We are currently using our program families-based approach in maintenance experiments to assess a technique we previously invented for modularizing synchronization concerns, called Szumo. The root program of each family implements a multi-threaded GUI browser, which reads and displays text from a network server. We are using three perfective maintenance tasks to populate the program families: one that adds network error handling, another that adds more settings and GUI controls, and another that reads data from multiple servers.