Strengthening Inferred Specifications using Search Based Testing

Software specification is an important element of the software development process. However, in most cases the specifications are out-of-date or even missing. One solution for this kind of problem is to use some process that infers the specification automatically. Work by Ernst et al (2000 ; 2002) has shown how specifications can be generated using program execution traces. These approaches are dependent on the test suites used to produce the traces, which may lead to unreliable specifications being inferred. Such specification inference is highly useful, however. In this paper we show how search based testing techniques can challenge and identify erroneous elements of such inferred specifications. This leads to a much tighter (accurate) inferred specifications. Thus, specification inference and search based test data generation are shown to be complementary.