Compositional and Transitive Reasoning During Class Abstraction

Authors: Alexander Egyed

Class diagrams are among the most widely used object-oriented design techniques. They are effective in modeling the structure of software systems at any stages of the software life cycle. Still, class diagrams can become as complex and overwhelming as the software systems they describe. This paper describes a technique for abstracting lower-level class structures into higher-level ones by ‘collapsing’ lower-level class patterns into single, higher-level classes and relationships. This paper is an extension to an existing technique that re-interprets the transitive meaning of lower-level classes into higher-level relationships (relational reasoning). The existing technique is briefly summarized. The extensions proposed in this paper are two-fold: This paper augments the set of abstraction rules to also collapse class patterns into higher- level classes (compositional reasoning). While this augmentation is simple and in sync with traditional views of refinement and abstraction, it has drawbacks in defining class features like methods and attributes. This paper thus also demonstrates how to filter low-level class features during abstraction. Our approach requires some human guidance in deciding when to use compositional or relational reasoning but is otherwise fully automated. Our approach is conservative in its results guaranteeing completeness but at the expense of some false positives (i.e., the filter errs in favor of not eliminating in case of doubt). The proposed technique is applicable to model understanding, inconsistency detection, and reverse engineering.

Download PDF here!