Who's Who in SBSE

Available soon...

Search based testing approaches are efficient in test data generation; however they are likely to perform poorly when applied to programs with state variables. The problem arises when the target function includes guards that reference some of the program state variables whose values depend on previous function calls. Thus, merely considering the target function to derive test data is not sufficient. This paper introduces a testability transformation approach based on the analysis of control and data flow dependencies to bypass the state variable problem. It achieves this by eliminating state variables from guards and/ or determining which functions to call in order to satisfy guards with state variables. A number of experiments demonstrate the value of the proposed approach.

Automated test data generation through evolutionary testing (ET) is a topic of interest to the software engineering community. While there are many ET-based techniques for automatically generating test data from code, the problem of generating test data from an extended finite state machine (EFSMs) is more complex and has received little attention. In this paper, we introduce a novel approach that addresses the problem of generating input test sequences that trigger given feasible paths in an EFSM model by employing an ET-based technique. The proposed approach expresses the problem as a search for input parameters to be applied to a set of functions to be called sequentially. In order to apply ET-based technique, a new fitness function is introduced to cope with the case when a test target involves calls to a set of transitions sequentially. We evaluate our approach empirically using five sets of randomly generated paths through two EFSM case studies: INRES and class 2 transport protocols. In the experiments, we apply two search techniques: a random and an ET-based which utilizes our new fitness function. Experimental results show that the proposed approach produces input test sequences that trigger all the feasible paths used with a success rate of 100%, however, the random technique failed in most cases with a success rate of 20.8%.