Test data generation using genetic programming

Context: Typically, search-based test data generation methods search on a population of program input values. Program input values can be regarded as solutions to underlying path constraints over program input parameters. One way to discover these path constraints is to use the symbolic execution method. Search-based methods attempt to find input values which are solutions to these path constraints, without knowing the actual constraints.Objective: In this paper, we show that we can search for the underlying path constraints using search-based methods, without resorting to symbolic execution. Trying to discover the exact or a good enough approximation of the underlying constraints may lead to a more targeted search, compared to directly searching for program input values. Besides, the construction of approximate constraints by searching may help to avoid some problems of symbolic execution.Method: The proposed method uses genetic programming for learning constraints on program input parameters.Results: To evaluate the performance of the proposed approach, a series of experiments have been conducted on a number of different benchmark programs. For 91.8% of benchmark programs, the proposed method achieved the best efficiency among the competitive algorithms.Conclusion: The results show that, if constraint solving can be provided for some or all parameter types of the methods of programs under test, our approach can improve the efficiency and effectiveness of search-based test data generation.