Expressing instances of a recurring problem as sentences in a small, Domain-Specific Language (DSL) is a well-established software engineering technique for simplifying the development and maintenance of business applications.

In the Model-Driven Engineering (MDE) community, metamodels are used to specify the abstract syntax of DSLs in terms of domain concepts, properties, and relationships. Although modern language workbenches can generate much of the language infrastructure (e.g., APIs, (de)serialization, and editors) from a metamodel, support for collaborative model editing is typically delegated to external services, requiring additional integration effort from language engineers. In particular, local-first collaborative modeling, where modelers can seamlessly alternate between online and offline editing, with automatic reconciliation of changes upon reconnection, is poorly supported.

In this talk, we present an approach for automatically generating local-first, decentralized, collaborative runtimes for modeling languages directly from their specifications. We generate language implementations on top of Conflict-free Replicated Data Types (CRDTs): a metamodel is compiled into a composition of pure operation-based CRDTs that mirrors the structure of the metamodel. The generated runtimes support standard object-oriented metamodeling constructs, including objects, attributes, containment hierarchies, references, multiplicities, and subtyping. To achieve this, we extend the set of existing composable CRDTs with multisets, optional values, sum and product types, and typed graphs.

3rd year PhD student