Automated test data generation is always a hot topic in software engineering, and evolutionary testing (ET) is an emerging and promising technology for this purpose. However, in structural testing, the presence of flag variables lead evolutionary testing degenerate to random testing. All previous work only focused on the flag problem in structural programs, and no attention has been paid to unstructured programs with flag conditions, although numerous industrial real-world programs are of this kind. In this paper, as a further step of the author's research, a fitness calculation rule for flag conditions in unstructured programs is proposed. The experiments on exemplifications recurrent in industrial real-world programs, such as Linux and NS2, show that our new fitness calculation rule could effectively guide evolutionary search to successfully find the required test data at low cost, while all previous approaches failed.

Evolutionary testing (ET), automatically generating test data with good quality, is an effective technique based on evolutionary algorithm. However, the presence of flag variables will make it degenerate to random testing in structural testing. Much of previous work has addressed this problem, but all can be characterized as program-specific. In this paper, flag cost function is introduced as the main component of fitness function, whose value changes with the variation of flag problem. Based on this, a unified fitness calculation rule for flag conditions is proposed. The experiments on programs with flag problems, once considered as inextricable in previous work, and the Traffic Alert and Collision Avoidance System (TCAS) code showed the effectiveness of our unified approach.