CONICET | Buscador de Institutos y Recursos Humanos

There is a large body of work in the literature about automatic unit tests generation, and many successful results have been reported so far. However, current approaches target library classes, but not full applications.A major obstacle for testing full applications is that they interact with the environment. For example, they establish connections to remote servers. Thoroughly testing such applications requires tests that completely control the interactions between the application and its environment.Recent techniques based on mocking enable the generation of tests which include environment interactions; however, generating the right type of interactions is still an open problem.In this paper, we describe a novel approach which addresses this problem by enhancing search-based testing with complex test data generation. Experiments on an artificial system show that the proposed approach can generate effective unit tests. Compared with current techniques based on mocking, we generate more robust unit tests which achieve higher coverage and are, arguably, easier to read and understand.