Search-Based Propagation of Regression Faults in Automated Regression Testing
Sina Shamshiri, Gordon Fraser, Phil McMinn, Alessandro Orso · 2013
Over the lifetime of software programs, developers make changes by adding, removing, enhancing functionality or by refactoring code. These changes can sometimes result in undesired side effects in the original functionality of the software, better known as regression faults. To detect these, developers either have to rely on an existing set of test cases, or have to create new tests that exercise the changes. However, simply executing the changed code does not guarantee that a regression fault manifests in a state change, or that this state change propagates to an observable output where it could be detected by a test case. To address this propagation aspect, we present EVOSUITER, an extension of the EVOSUITE unit test generation tool. Our approach generates tests that propagate regression faults to an observable difference using a search-based approach, and captures this observable difference with test assertions. We illustrate on an example program that EVOSUITER can be effective in revealing regression errors in cases where alternative approaches may fail, and motivate further research in this direction.