$T^3$: Multi-level Tree-based Automatic Program Repair with Large Language Models

TL;DR

This paper introduces $T^3$, a novel multi-level tree-based framework that leverages large language models and chain-of-thought reasoning to improve automated program repair accuracy and efficiency.

Contribution

The paper proposes an innovative $T^3$ framework combining LLMs with tree search, enhancing reasoning and repair precision in automated program repair tasks.

Findings

$T^3$ improves repair solution accuracy.

Effective guidance for sample selection and repair strategies.

Enhanced reasoning capabilities in APR with LLMs.

Abstract

Automatic Program Repair (APR) is a core technology in software development and maintenance, with aims to enable automated defect repair with minimal human intervention. In recent years, the substantial advancements in Large Language Models (LLMs) and the Chain-of-Thought (CoT) techniques have significantly enhanced the reasoning capabilities of these models. However, due to the complex logic and multi-step reasoning ability needed, the application of CoT techniques in the APR domain remains insufficient. This study systematically evaluates the performance of several common CoT techniques in APR tasks and proposes an innovative framework $T^3$, which integrates the powerful reasoning capabilities of LLMs with tree search, effectively improving the precision of generating candidate repair solutions. Furthermore, $T^3$ provides valuable guidance for optimizing sample selection and repair strategies in APR tasks, establishing a robust framework for achieving efficient automated debugging.