Termination and Expressiveness of Execution Strategies for Networks of Bidirectional Model Transformations

When developers describe a software system with multiple models, such as architecture diagrams, deployment descriptions, and source code, these models must represent the system in a uniform way, i.e., they must be and stay consistent. One means to automatically preserve consistency after changes to models are model transformations, of which bidirectional transformations that preserve consistency between two models have been well researched. To preserve consistency between multiple models, such transformations can be combined to networks. When transformations are developed independently and reused modularly, the resulting network can be of arbitrary topology. For such networks, no universal strategy exists to orchestrate the execution of transformations such that the resulting models are consistent. In this article, we prove that termination of such a strategy can only be guaranteed if it is incomplete, i.e., if it is allowed to fail to restore consistency for some changes although an execution order of transformations exists that yields consistent models. We propose such a strategy, for which we prove termination and show that and why it makes it easier for users of model transformation networks to understand the reasons whenever the strategy fails. In addition, we provide a simulator for the comparison of different execution strategies. These findings help transformation developers and users in understanding when and why they can expect the execution of a transformation network to terminate and when they can even expect it to succeed. Furthermore, the proposed strategy guarantees them termination and supports them in finding the reason whenever it is not successful.