Automatic test oracle generation based on variable centrality analysis

Abstract

Nowadays, software has increased its complexity as it became essential part of our everyday life. While its pervasiveness and criticality grows as it has been incorporated in many fields, the task of ensuring that its behavior follows certain quality standards represents a crucial goal. Testing represents one of the most used tools to verify and validate the functionality of the software. The complexity of the software has increased the need for automated software testing. Most of the work has focused on the automation of test input data. The automation of the test oracles has not been explored exhaustively. Both elements are equally relevant in the testing process. While test input data allows to explore all possible states, oracles allow to visualize the faulty behaviour. In this work, it is proposed a novel technique to support the automatic generation of test oracles. The main idea is to rank program variables based on the observed interaction during program execution. To obtain a quantitative metric, the concept of network centrality is used. The only input that is required is the set of test input data. From this, the proposed approach performs the analysis during execution and recommends a set of assertions. The proposed approach was evaluated in an empirical study against a random oracle generation process and a mutation analysis based oracle generation process using a set of case examples from the critical systems domain. The results showed that the proposed approach performs better that the random based and the mutation based alternatives with increments in the effectiveness up to 25\% and 11\% respectively when using betweenness centrality and eigenvector centrality. The use of closeness centrality and degree centrality did not show significant improvement. The main contribution of th...