Controlled generation of models with defined properties

Abstract

Test models are required to evaluate and benchmark algorithms and tools which support model driven development. In many cases, test models are not readily available from real projects and they must be generated. Using existing model generators leads to test models of poor quality because they randomly apply graph operations on graph representations of models. Some approaches do not even guarantee the basic syntactic correctness of the created models. This paper presents the SiDiff Model Generator, which can generate models and model histories and which can modify existing models. The resulting models are syntactically correct, contain complex structures, and have specified statistical properties, e.g. the frequencies of model element types.