A Hybrid Regression Test Selection Technique for Object-Oriented Programs

We propose a regression test selection technique that is based on analysis of both the source code of an object-oriented program as well as the UML state machine models of the affected classes. We first construct a dependency graph model of the original program from the source code. When the program is suitably modified, the constructed model is updated to reflect the changes. Our model in addition to capturing control and data dependencies represents the dependencies arising from object-relations. To find the model elements affected due to a program change, we construct a forward slice of the constructed graph model, where each changed model element is used as a slicing criterion. Subsequently, we determine the affected methods from an analysis of the state machine models based on the changed statements. The test cases that exercise the affected model elements in the program model as well as the transitions caused by the affected methods in state machine models are selected for regression testing. Empirical studies carried out by us show that our technique selects on an average of 27.89 % more fault-revealing test cases compared to a purely dependence graph model-based technique while incurring about 38.21% increase in regression test suite size.