Adapting Unit Tests by Generating Combinatorial Test Data

Conventional unit tests are still mainly handcrafted. Generalizing conventional unit tests to parameterized unit tests supports automatic test data generation. Methods that were introduced to instantiate parameterized unit tests with concrete values as test data are based on search based approaches, dynamic symbolic execution, or property based testing. In this work, we introduce an approach that retrofits existing conventional unit tests into parameterized unit tests by generalization, and generate test data by combinatorial valuation to adapt existing conventional unit test suites. We conduct an empirical study to investigate whether our test suite adaption approach is beneficial in terms of additional fault detection capabilities and code coverage. Our results show that mutation score and condition coverage increase with feasible effort compared to existing conventional unit tests.