Scaling-up domain-specific modelling languages through modularity services

Abstract

Context. Model-driven engineering (MDE) promotes the active use of models in all phases of software development. Even though models are at a high level of abstraction, large or complex systems still require building monolithic models that prove to be too big for their processing by existing tools, and too difficult to comprehend by users. While modularization techniques are well-known in programming languages, they are not the norm in MDE. Objective. Our goal is to ease the modularization of models to allow their efficient processing by tools and facilitate their management by users. Method. We propose five patterns that can be used to extend a modelling language with services related to modularization and scalability. Specifically, the patterns allow defining model fragmentation strategies, scoping and visibility rules, model indexing services, and scoped constraints. Once the patterns have been applied to the meta-model of a modelling language, we synthesize a customized modelling environment enriched with the defined services, which become applicable to both existing monolithic legacy models and new models. Results. Our proposal is supported by a tool called EMF-Splitter, combined with the Hawk model indexer. Our experiments show that this tool improves the validation performance of large models. Moreover, the analysis of 224 metamodels from OMG standards, and a public repository with more than 300 metamodels, demonstrates the applicability of our patterns in practice. Conclusions. Modularity mechanisms typically employed in programming IDEs can be successfully transferred to MDE, leading to more scalable and structured domain-specific modelling languages and environments.

Model Sensemaking Strategies: Exploiting Meta-Model Patterns to Understand Large Models

“…Additionally, solutions based on model partitioning have been devised to handle existing large models. These include model fragmentation according to user-defined strategies [14], [15]; partial model loading [16]; and model decomposition into smaller valid submodels [17]. Moreover, indexers [18] (similar to those in relational databases) and caching techniques for queries [19], [20] have been proposed for faster model element retrieval.…”

Language Composition via Kind-Typed Symbol Tables.

“…In metamodels, consistency is typically achieved in different ways than by using symbol tables. One could compare different, mainly metamodelbased approaches such as (Leduc et al 2017), (Garmendia et al 2019), (Voelter 2013), or (Voelter et al 2013) more precisely with our approach, however, this would require a large, comprehensive study that tests each approach with examples.…”