Language Workbench
A language workbench is a development environment for building custom, usually domain-specific languages (DSLs). It provides everything that makes a language practically usable: language definition, an editor with code completion, validation, and code generation or interpretation. The term was coined in 2005 by an article by Martin Fowler.
What a language workbench provides
Defining a language is the smaller part of the work — it only becomes usable through tooling. This is exactly what a language workbench delivers as an integrated whole:
- Language definition: the grammar or structure of the language — the metamodel with concepts, relationships and constraints.
- Editor: syntax highlighting, code completion, navigation and refactoring for the new language.
- Validation: domain rules are checked as you type, not only at build or test time.
- Generation and processing: code generators, interpreters and model transformations produce code, configuration or documentation from the models.
Without a language workbench, each of these components would have to be developed by hand — which is why building external DSLs used to be uneconomical in most cases.
Parser-based vs. projectional
Two paradigms dominate:
| Paradigm | Principle | Typical representative |
|---|---|---|
| Parser-based | Models exist as text and are parsed. Text files can be diffed and merged and integrate seamlessly with version control and CI/CD pipelines. | Xtext |
| Projectional | Users edit the abstract syntax tree directly; the notation is merely a projection of it. This enables notations beyond text — tables, mathematical expressions, graphical elements — and the clean combination of several languages. | JetBrains MPS |
Well-known language workbenches
Xtext is the established open source workbench for textual DSLs based on Eclipse and the Eclipse Modeling Framework. Via the Language Server Protocol (LSP), Xtext languages can also be embedded into modern editors such as VS Code. itemis has been part of the Xtext founding community (since 2006) and provides active committers.
JetBrains MPS is the best-known projectional language workbench. Its particular strength: several domain languages can be combined cleanly, and notations are not limited to text. itemis is the main maintainer of central MPS extensions (iets3, mbeddr) and has been working with this technology as a JetBrains partner for over ten years.
Language workbenches in practice
The choice of workbench follows from two questions: Which notation does the domain need — plain text, or also tables, formulas, graphics? And where should the tool run — in VS Code, in the browser, as an extension of an existing toolchain? For textual DSLs with IDE integration, Xtext plus LSP is a proven path; for complex language composition and formal notations, MPS is the more flexible choice. Regardless of the tool, one thing holds: the success of a language is decided in the domain analysis that precedes it — a workbench accelerates language building, but does not replace an understanding of how the experts of a domain actually work.