Offline Delta-driven Model Transformation with Dependency Injection

When model transformations are used to implement consistency relations between very large models (VLMs), incrementality plays a cornerstone role in the realization of practical consistency maintainers. State-of-the-art model transformation engines with support for incrementality normally rely on a publish-subscribe model for linking model updates − deltas − to the application of model transformation rules, in so called dependencies, at run time. These deltas can then be propagated along an already executed model transformation. A small number of such engines use domain-specific languages (DSLs) for representing model deltas offline in order to enable their use in asynchronous, eventbased execution environments. The principal contribution of this work is the design of a forward delta propagation mechanism for incremental execution of model transformations, which decouples dependency tracking from delta propagation using two innovations. First, the publish-subscribe model is replaced with dependency injection, physically decoupling domain models from consistency maintainers. Second, a standardized representation of model deltas is reused, facilitating interoperability with EMF-compliant tools, both for defining deltas and for processing them asynchronously. This procedure has been implemented in a model transformation engine, whose performance has been evaluated empirically using the VIATRA CPS benchmark. In the experiments performed, the new transformation engine shows gains in the form of several orders of magnitude in the initial phase of the incremental execution of the benchmark model transformation and delta propagation is performed in real time, independently of the size of the models involved, whereas the up-to-now best-performant approach is dependent.