Differencing UML models: A domain-specific vs. a domain-agnostic method

Comparing software artifacts to identify their similarities and differences is a task ubiquitous in software engineering. Logical-design comparison is particularly interesting, since it can serve multiple purposes. When comparing the as-intended vs. the as-implemented designs, one can evaluate implementation-to-design conformance. When comparing newer code versions against earlier ones, one may better understand the development process of the system, recognize the refactorings it has gone through and the qualities motivating them, and infer high-order patterns in its history. Given its importance, design differencing has been the subject of much research and a variety of algorithms have been developed to compare different types of software artifacts, in support of a variety of different software-engineering activities. Our team has developed two different algorithms for differencing logical-design models of object-oriented software. Both algorithms adopt a similar conceptual model of UML logical designs (as containment trees); however, one of them is heuristic whereas the other relies on a generic tree-differencing algorithm. In this paper, we describe the two approaches and we compare them on multiple versions of an open-source software system.