Key classes in object-oriented systems

Abstract

Several object-oriented systems, such as Lucene, Tomcat, Javac have their respective design documented using key-classes, defined as important/central classes to understand the object-oriented design. Considering this fact, and considering that, in general, software architecture is not formally documented to help developers understanding and assessing software design, Keecle is proposed as an approach based on dynamic and static analysis for detection of key classes in a semi-automatic way. The application of filtering mechanisms on the search space of the dynamic data is proposed in order to obtain a reduced set of key classes. The approach is evaluated with fourteen proprietary and open source systems in order to verify that the found classes correspond to the key classes of the ground-truth, which is defined from the documentation or defined by the developers. The results were analyzed in terms of precision and recall, and have shown to be superior to the state-of-the-art approach. The role of key classes in assessing design has also been investigated. The organization of the key classes in a dependency graph, which highlights explicit dependency relations in the source code, was evaluated to be adequate for design comprehension and assessment. Key classes were evaluated whether they are more prone to bad smells, and whether specific types of bad smells are associated with different levels of cohesion and coupling metrics. In addition, the ownership of key classes was shown to be more concentrated in a reduced set of developers. Finally, we conducted an experimental study with students and a survey with developers to evaluate documentation based on key classes. The results indicate that the documentation based on key classes are a feasible alternative for use as complementary documentation to the existing one, or for use as main documentation in environments where documentation is not available.