Markov Chain-based Test Data Adequacy Criteria: a Complete Family
Mohammed Al-Ghafees, James A. Whittaker · Informing Science and IT Education Conference · 2002
The idea of using white box data flow information to select test cases is well established and has proven an effective testing strategy. This paper extends the concept of data flow testing to the case in which the source code is unavailable and only black box information can be used to make test selection decisions. In such cases, data flow testing is performed by constructing a behavior model of the software under test to act as a surrogate for the program flow graph upon which white box data flow testing is based. The behavior model is a graph representation of externally-visible software state and input-induced state transitions. We first summarize the modeling technique and then define the new data flow selection rules and describe how they are used to generate test cases. Theoretical proof of concept is provided based on a characteristic we call transition variation. Finally, we present results from a laboratory experiments in which we compare the fault detection capability of black box data flow tests to other common techniques of test generation from graphs, including simple random sampling, operational profile sampling and state transition coverage.