Metamorphic Relation Recognition Method Based on Control Flow Graph Features

Abstract

Abstract Metamorphic testing is an effective method to solve Oracle problems. It is used to test software which is no expected output or the expected output is difficult to construct. However, finding a suitable and effective metamorphic relation in the program is still a very difficult task, usually obtained by domain experts through manual analysis, which greatly reduces the efficiency of software testing. In this work we define a function parameter metamorphic relation from the operation of the function parameter sequence, and preset 10 function parameter metamorphic relation. Because functions with the same structure usually have the same function parameter metamorphic relation, we extract node features and independent path features from the control flow graph, and then create a classification model to judge the similarity between functions. Our experiment includes the evaluation of the prediction model and the prediction of the metamorphic relation on the real function. Good experimental results shows that this method can provides a new way to automated discovery of metamorphic relation.