A Context-Sensitive Coverage Criterion for Test Suite Reduction

Thumbnail Image


umi-umd-5239.pdf (964.75 KB)
No. of downloads: 1244

Publication or External Link






Modern software is increasingly developed using multi-language implementations, large supporting libraries and frameworks, callbacks, virtual function calls, reflection, multithreading, and object- and aspect-oriented programming. The predominant example of such software is the graphical user interface (GUI), which is used as a front-end to most of today's software applications. The characteristics of GUIs and other modern software present new challenges to software testing. Because recently developed techniques for automated test case generation can generate more tests than are practical to regularly execute, one important challenge is test suite reduction. Test suite reduction seeks to decrease the size of a test suite without overly compromising its original fault detection ability. This research advances the state-of-the-art in test suite reduction by empirically studying a coverage criterion which considers the context in which program concepts are covered. Conventional approaches to test suite reduction were developed and evaluated on batch-style applications and, due to the aforementioned considerations, are not always easily applicable to modern software. Furthermore, many existing techniques fail to consider the context in which code executes inside an event-driven paradigm, where programs wait for and interactively respond to user- and system-generated events. Consequently, they yield reduced test suites with severely impaired fault detection ability. The novel feature of this research is a test suite reduction technique based on the call stack coverage criterion which addresses many of the challenges associated with coverage-based test suite reduction in modern applications. Results show that reducing test suites while maintaining call stack coverage yields good tradeoffs between size reduction and fault detection effectiveness compared to traditional techniques. The output of this research includes models, metrics, algorithms, and techniques based upon this approach.