Automatic test data generation to improve fault-localization based on causal-statistical analysis

The statistical-based software fault localization approaches highly depend on the program inputs and become unstable as input data changes. Therefore, generating appropriate test data plays an essential role in the quality of the software fault-localization process. This paper presents an approach to improving program fault localization by generating new test data. The minimized test suite determines the faulty execution path and the fault suspiciousness of each statement in the path is generated. First, the suspicious statements in the faulty path are determined. To this aim, the conditions of the faulty execution path are contradicted from the end to the beginning, and test data is created for the desired path using the Z3 solver. Afterward, the program under test is executed with the generated test data using the Concolic testing technique. The fault-suspicious branch is determined depending on the passing or failing of the program execution. As a result, the region of statements for applying the causal-statistical approach is minimized. The proposed approach is evaluated on the four projects in the Defects4J benchmark. The results show that 75% of faults are localized by examining a maximum of 1% of the program's source code. Compared to the related work, the results have improved by 17.98%. Moreover, the mean number of sentences examined for fault localization decreases by 16.78% in the worst case.