LLMs as Continuous Learners: Improving the Reproduction of Defective Code in Software Issues
This addresses the challenge of adapting to unique, evolving errors in individual code repositories for software developers, representing an incremental improvement over prior methods.
The paper tackles the problem of reproducing buggy code in software issues, which is crucial for identifying and fixing bugs, by proposing EvoCoder, a multi-agent continuous learning framework that improves issue reproduction rates by 20% over existing state-of-the-art methods.
Reproducing buggy code is the first and crucially important step in issue resolving, as it aids in identifying the underlying problems and validating that generated patches resolve the problem. While numerous approaches have been proposed for this task, they primarily address common, widespread errors and struggle to adapt to unique, evolving errors specific to individual code repositories. To fill this gap, we propose EvoCoder, a multi-agent continuous learning framework for issue code reproduction. EvoCoder adopts a reflection mechanism that allows the LLM to continuously learn from previously resolved problems and dynamically refine its strategies to new emerging challenges. To prevent experience bloating, EvoCoder introduces a novel hierarchical experience pool that enables the model to adaptively update common and repo-specific experiences. Our experimental results show a 20\% improvement in issue reproduction rates over existing SOTA methods. Furthermore, integrating our reproduction mechanism significantly boosts the overall accuracy of the existing issue-resolving pipeline.