| Test data generation is a difficult, time consuming, costly phase in the software life cycle. Software engineers address this problem by decomposing it into three phases: unit test, integration test, and system test. For each phase, testers use abstract representations of the software product to define test objectives, specify test case design strategies, and generate tests. At the system test level, we find few general purpose test data generation methods, little use of abstract representations of the system under test, and application specific test generation schemes. This research shows one way to generalize system level tests by viewing an application through its user interface. We focus on command-based systems or command language user interfaces. A test case for a command-based system is a list of fully parameterized commands. Each command in the test case is issued to the system under test and the system is examined for its response. We capture command language syntax and semantics in a domain model. The result is a test data generation method called domain based testing (DBT). Testers guide test generation by defining test criteria, and map the test criteria to the domain model. The result is a test subdomain from which the test generator creates tests. To evaluate DBT and the quality of its test cases, this research uses a neural network classifier to assess test case effectiveness. The neural net classifies test case attributes/metrics into fault severity levels. Tests with low predicted effectiveness need not be run. The DBT test generation method and the neural net effectiveness prediction are applied to a command language for an industrial robot tape library. |
