DynaFix: Iterative APR with Execution-Level Dynamic Information
Published:Dec 31, 2025 05:13
•1 min read
•ArXiv
Analysis
This paper introduces DynaFix, an innovative approach to Automated Program Repair (APR) that leverages execution-level dynamic information to iteratively refine the patch generation process. The key contribution is the use of runtime data like variable states, control-flow paths, and call stacks to guide Large Language Models (LLMs) in generating patches. This iterative feedback loop, mimicking human debugging, allows for more effective repair of complex bugs compared to existing methods that rely on static analysis or coarse-grained feedback. The paper's significance lies in its potential to improve the performance and efficiency of APR systems, particularly in handling intricate software defects.
Key Takeaways
- •DynaFix is an execution-level dynamic information-driven APR method.
- •It iteratively leverages runtime information (variable states, control-flow paths, call stacks) to refine the repair process.
- •DynaFix achieves a 10% improvement over state-of-the-art baselines and repairs 38 previously unrepaired bugs.
- •It reduces the patch search space by 70% compared with existing methods.
Reference
“DynaFix repairs 186 single-function bugs, a 10% improvement over state-of-the-art baselines, including 38 bugs previously unrepaired.”