Graph Construction for Imperative Programs Using Neural and Structural Methods
A study introduces a methodology for transforming imperative programming languages and their annotations into typed, attributed graphs, facilitating the reuse of verification artifacts. This technique combines abstract syntax tree parsing with semantic embeddings derived from SentenceTransformer and CodeBERT. The experiments utilize datasets such as C with ACSL, Java with JML, and Dafny for C#. Findings indicate uniform graph representations across various programming languages and annotation formats, laying the groundwork for future enhancements in semantics and approximate graph matching.
Key facts
- Pipeline converts imperative programs and annotations into typed, attributed graphs.
- Uses abstract syntax tree parsing and semantic embeddings from SentenceTransformer and CodeBERT.
- Experiments cover C with ACSL, Java with JML, and Dafny for C#.
- Consistent graph representations achieved across different languages and annotation styles.
- Work provides basis for semantic enrichment and approximate graph matching for verification artefact reuse.
Entities
Institutions
- arXiv