Identifying Cross-Cutting Concerns from History

As object-oriented programs evolve, they may suffer from the “tyranny of dominant decomposition”: The program can be modularised only one way at a time, leaving cross-cutting concerns scattered across many modules and tangled with one another. Aspectoriented programming (AOP) tries to remedy this by encapsulating these concerns into aspects. Aspect mining identifies such cross-cutting concerns and thus helps to migrate existing software to an aspectoriented program. Aspect mining of large software system like Eclipse is a problem: dynamic approaches depend on test cases and have trouble covering all code and thus detecting all cross-cutting functionality. And static approaches simply cannot analyse systems this big unless they work incrementally. We offer a new approach based on the observation that cross-cutting functionality does not exist from the beginning. Instead, it is introduced over time. More specifically, we speculate that considerable cross-cutting functionality is introduced within short periods of time. To find these, we analyse code additions from development tasks as they are recorded in a software repository like CVS. Since we analyse one task at a time, our approach is independent of a project’s total size. This enables us to report crosscutting functionality for Eclipse, a 1.6MLOC Java program. In this paper we sketch the basic idea of history-based aspect mining and some initial results.