Shall We Test Service-Based Models or Generated Code?

Testing early reduces the cost of the validation process. Model-Driven Testing promotes the creation of the tests at the model level and their transformation into code which can be executed. However, we miss an empirical evaluation on the efficiency of generating tests designed on the service-based component models compared to implementing them based on the generated code. In this paper, we evaluate if one shall generate tests from platform-independent models or implement them on the generated code. We consider service-based component models and a case study modeling and generating the code of a platoon of vehicles. Our experimentation on that first case study confirms that generating tests from the service-based models is easier compared to implementing them on the generated code. Moreover, efficiency of the tests of both approaches is similar, considering code coverage and mutation analysis.