LLM-Assisted Style Customization of Xtext-Based Domain-Specific Languages
| Typ | Bachelorarbeit | |
|---|---|---|
| Aushang | LLM Assisted Style Customization of Xtext Based DSLs.pdf | |
| Betreuer | Wenden Sie sich bei Interesse oder Fragen bitte an: Weixing Zhang (E-Mail: weixing.zhang@kit.edu), Bowen Jiang (E-Mail: bowen.jiang@kit.edu) |
Xtext is one of the most widely used DSL development frameworks, capable of automatically generating textual grammars from Ecore meta-models. However, the auto-generated grammars are often redundant and poorly readable, and frequently require manual modification to reach the style desired by users (e.g., a Python-like style). Prior work proposed a semi-automated approach to address this issue, but that approach relies on predefined rules, offers limited flexibility, and requires users to have a certain level of Xtext knowledge. As large language models (LLMs) have demonstrated strong capabilities in code understanding and generation, a natural question arises: can LLMs be leveraged to support DSL style customization in a more flexible and intelligent way? Specifically, in the meta-model-driven scenario, an LLM can take the generated grammar and adapt it according to the user's style requirements; in the grammar-driven scenario, users only need to describe their domain knowledge and desired style in natural language, and the LLM can directly generate a grammar that conforms to the Xtext specification and reflects the intended style. This would significantly lower the barrier to DSL development while improving the usability and readability of the resulting grammars.