Extending single- to multi-variant model transformations by trace-based propagation of variability annotations

Abstract

Model-driven engineering involves the construction of models on different levels of abstraction. Software engineers are supported by model transformations, which automate the transition from high-to low-level models. Product line engineering denotes a systematic process that aims at developing different product variants from a set of reusable assets. When modeldriven engineering is combined with product line engineering, engineers have to deal with multi-variant models. In annotative approaches to product line engineering, model elements are decorated with annotations, i.e., Boolean expressions that define the product variants in which model elements are to be included. In model-driven product line engineering, domain engineers require multi-variant transformations, which create multi-variant target models from multi-variant source models. We propose a reuse-based gray-box approach to realizing multi-variant model transformations. We assume that single-variant transformations already exist, which have been developed for model-driven engineering, without considering product lines. Furthermore, we assume that single-variant transformations create traces, which comprise the steps executed in order to derive target models from source models. Single-variant transformations are extended into multi-variant transformations by trace-based propagation: after executing a single-variant transformation, the resulting single-variant target model is enriched with annotations that are calculated with the help of the transformation's trace. This approach may be applied to single-variant transformations written in different languages and requires only access to the trace, not to the respective transformation definition. We also provide a correctness criterion for trace-based propagation, and a proof that this criterion is satisfied under the prerequisites of a formal computational model.

Towards Trace-Based Synchronization of Variability Annotations in Evolving Model-Driven Product Lines

“…𝑡 , where 𝑛, 𝑚 ∈ N + offer access to these elements. Accordingly, we assume a trace is rule-based and generationcomplete [50,51], i.e., it references all corresponding source and created target elements per rule application. In contrast, an incomplete trace records only pivotal elements [9] and a complete trace distinguishes context from created elements in the target model.…”

“…Currently, multi-variant model transformations serve well for propagating annotations in one direction [20,35,48,50,51]. However, in iterative and collaborative development scenarios, the developed product line evolves continuously.…”

“…Thus, a multi-variant model changes over time-including its annotations. While bidirectional model transformations synchronize the elements and their relations of two models, multi-variant model transformations [35,39,51] propagate the annotations from a source model to the target model regardless of present changes in the target model. For instance, when changing the annotation of a Java class to a more restrictive visibility (i.e., it is integrated in less variants) while the annotation of the corresponding element in the UML model remains unchanged, state-of-the art multi-variant transformations typically overwrite the annotation of the Java class provoking possible data leakage on Java side or an inconsistent, potentially broken set of products.…”

Towards Development with Multi-Version Models: Detecting Merge Conflicts and Checking Well-Formedness

“…Westfechtel and Greiner [22] present a solution to configuration management for product lines, where several configurations are represented in a unified feature model. Information about the presence of elements in certain configurations is encoded in logic formulas and propagated along model transformations.…”

“…Information about the presence of elements in certain configurations is encoded in logic formulas and propagated along model transformations. While this approach bears some similarity to the encoding of different versions and collective well-formedness checking in our solution, the technique in [22] focuses on the definition of product lines and hence does not consider merging.…”