What is it about?

Repairing bugs in a timely manner is a crucial aspect of code development. This work introduces VarPatch, an automated repair system utilizing multiple specialized large language model agents. The system targets bugs that must be fixed through variable addition. Each agent is provided with individualized prompts and various types of context information. The agents then collaborate to generate, verify, and select the best patch. VarPatch is evaluated on a Java bug dataset and outperforms baseline repair tools and GPT-4.

Featured Image

Why is it important?

This research summarizes and addresses the variable addition repair pattern, which many existing repair tools fail to patch due to the complex workflow of introducing, checking, and updating new variables. Moreover, a new multi-agent architecture for automated program repair is introduced. The results show that our system can patch more bugs than baseline repair tools and GPT-4 with reasonable time and cost.

Perspectives

In an era of learning-based repair, this work is an important step towards using large language models for more accurate and efficient automated program repair. In the future, this system can be modified and extended to cover more coding languages and bug repair patterns. I believe the multi-agent approach is a promising research direction that can be applied to automate not just code repair, but also other steps in the software development process. As large language models become increasingly powerful and prevalent, I hope this work inspires further research on how to leverage the capabilities of large language models for diverse tasks.

Elisa Zhang
Dougherty Valley High School

Read the Original

This page is a summary of: Poster: Repairing Bugs with the Introduction of New Variables: A Multi-Agent Large Language Model, December 2024, ACM (Association for Computing Machinery),
DOI: 10.1145/3658644.3691412.
You can read the full text:

Read

Contributors

The following have contributed to this page