Advances in LLM-Based Program Repair and Code Maintenance

The field of program repair and code maintenance is rapidly advancing with the integration of Large Language Models (LLMs). Recent developments have focused on improving the precision and effectiveness of LLM-based approaches, enabling them to better support real-world software development workflows. Notably, researchers have explored the use of LLMs for vulnerability localization, automated program repair, and code refactoring, with promising results. These advancements have the potential to significantly reduce the time and effort required for code maintenance, while also improving the overall quality and reliability of software systems. Noteworthy papers in this area include: T2L-Agent, which achieves up to 58.0% detection and 54.8% line-level localization of vulnerabilities in open-source software. AuthFix, which successfully generates correct patches for 74% of OpenID bugs. Refine, which boosts the performance of AutoCodeRover by 14.67% and achieves state-of-the-art results in workflow-based program repair approaches.

Sources

From Trace to Line: LLM Agent for Real-World OSS Vulnerability Localization

Automated Repair of OpenID Connect Programs (Extended Version)

Abstain and Validate: A Dual-LLM Policy for Reducing Noise in Agentic Program Repair

LLM Agents for Automated Dependency Upgrades

REFINE: Enhancing Program Repair Agents through Context-Aware Patch Refinement

Refactoring with LLMs: Bridging Human Expertise and Machine Understanding

Improving IR-based Bug Localization with Semantics-Driven Query Reduction

JSON Whisperer: Efficient JSON Editing with LLMs

FreshBrew: A Benchmark for Evaluating AI Agents on Java Code Migration

Built with on top of