Research Of Automatic Generation Technology For MC/DC Test Data Based On Genetic Algorithm

Generation of test data is a key for software testing. As modern software system becomes larger, and more and more complicated, traditional search technology can hardly be applied in the field of test data generation. Genetic algorithm indicates unique superiority when solving big space, nonlinear problems. Therefore, the research of combing genetic test and automatic test data generation is rewarding.From the application perspective of test data generation, this paper contains basic principles of genetic algorithm and test data automatic generation, and analysis of coverage criterion for current test data generation and construction limitation of the fitness function of the corresponding genetic algorithm. The core is to generate test data which meets for MC/DC, the design realizes MC/DC based test data generation.First of all, a deep research of relevant principle concerning test data automatic generation was performed, which focuses on test data automatic generation from the perspective of structural, basic principle and implemented process of genetic algorithm, and also analyzed applicability of applied genetic algorithm for automatic generation of test data.Secondly, this paper includes work process of genetic algorithm based automatic generation of test data. And a detailed analysis of current test data to generate widely used coverage criterion and structure of corresponding genetic algorithm fitness function. It contains deep analysis of disadvantages for the application and proposes solutions for the two problems.Then, this paper also illustrates the implementation of automatic generation for MC/DC based test data. Regarding the specialty of MC/DC test data, research was performed by parsing code, obtaining method of MC/DC test cases set as a search goal for genetic algorithm. Based on current situation, coding method is already determined. In the design of fitness function, by combining traditional fitness function based chaining approach, method of collecting control nodes either directly or indirectly affecting the traversal of the problem node via data dependencies was proposed to optimize the approach level evaluation.To assess the feasibility of our approach, the paper implemented a prototype of our approach for code and applied it to the well-known Triangle programs. It reported that the evidence of the superiority of the new fitness function that is able to avoid plateau leading to the degradation of the optimisation search techniques to a random search as in the case of traditional white box fitness functions and the impact to the Parameters'Domain Space and the number of Search Iterations was discussed.In the end, our approach was applied o the actual software system, the result was satisfied the required.