PracRepair: LLM-Empowered Automated Program Repair Inspired by Human-Like Debugging Practices

2026-06-17 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, quick

Summary

PracRepair is a novel LLM-based automated program repair (APR) framework. It mimics human debugging practices, addressing limitations of existing LLM-based APR methods that underutilize dynamic information. Current approaches often rely on static context, error messages, and coarse validation, overlooking crucial failure-execution and patch-validation dynamics. PracRepair constructs an on-demand static-dynamic context from buggy programs and failure executions. It then employs question-driven failure diagnosis to generate explicit repair hypotheses. Patches are iteratively refined using validation diagnostics and trace-level behavioral changes. Experimental evaluations on Defects4J V1.2 and V2.0 demonstrate its superior performance. With GPT-3.5, PracRepair fixed 139 bugs on V1.2 and 136 on V2.0. Using GPT-4o, these numbers improved to 162 and 171 bugs, respectively. The framework also generalizes effectively to Real-World Bugs (RWB), achieving top performance across various foundation models.

Key takeaway

For Machine Learning Engineers developing automated program repair solutions, integrate dynamic execution and validation feedback into your LLM-based frameworks. PracRepair demonstrates that utilizing on-demand static-dynamic context and iterative patch refinement significantly boosts bug-fixing capabilities. This human-inspired debugging approach improves performance on Defects4J and real-world bugs, suggesting a path to more robust and effective repair systems.

Key insights

PracRepair enhances LLM-based automated program repair by integrating dynamic execution and validation feedback, mimicking human debugging.

Principles

• Human-like debugging improves LLM-based APR.
• Dynamic execution context is crucial for failure diagnosis.
• Iterative patch refinement uses trace-level changes.

Method

PracRepair constructs on-demand static-dynamic context, performs question-driven failure diagnosis for hypotheses, and iteratively refines patches using validation diagnostics and trace-level behavioral changes.

In practice

• Integrate dynamic traces for bug localization.
• Use question-driven diagnosis for repair hypotheses.
• Refine patches with validation feedback.

Topics

• Automated Program Repair
• Large Language Models
• Dynamic Debugging
• Software Testing
• Defects4J
• GPT-4o

Best for: AI Scientist, Machine Learning Engineer, Software Engineer

Related on AIssential

• Automated Root-Cause Subclassification and No-Code Fix Generation for Invalid Bug Reports — Automating invalid bug report subclassification and no-code fix generation significantly reduces software maintenance overhead.
• DDOR: Delta Debugging for Explainable Overrefusal Testing and Repair — DDOR uses delta debugging to explain and mitigate LLM overrefusal in black-box settings, improving usability and safety.
• Beyond Problem Solving: UOJ-Bench for Evaluating Code Generation, Hacking, and Repair in Competitive Programming — LLMs can actively verify and debug competitive programming code, but current inference costs limit practical deployment.
• AutoRPA: Efficient GUI Automation through LLM-Driven Code Synthesis from Interactions — AutoRPA efficiently automates repetitive GUI tasks by converting LLM agent logic into robust, token-saving RPA functions.
• 🧠 Community Wisdom: Claude Code tips for ADHD users, resources for managing up, going from corporate to startup, tiring of being your company’s AI evangelist, and more — Community discussions offer practical strategies for LLM fine-tuning, evaluation, and inference optimization.
• Engineering a human-aligned LLM evaluation workflow with Prodigy and DSPy — Human-aligned LLM evaluation requires iterative workflows that integrate granular human feedback to engineer context-specific metrics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.