RePaCA: LLM-Based Static Patch Correctness Assessment
A new technique called RePaCA has been introduced to address the problem of overfitting patches in Automated Program Repair (APR). APR aims to automatically fix software bugs, but often generates patches that pass tests without truly fixing the underlying issue. RePaCA is a static approach to Automated Patch Correctness Assessment (APCA), meaning it only requires the original and fixed code snippets. It leverages Large Language Models (LLMs) specialized in reasoning tasks. The model is prompted with both buggy and fixed code, and is guided to generate a Chain of Thought that analyzes code differences and reasons about how the patch addresses the bug. This approach aims to improve reliability, flexibility, and transparency compared to existing static techniques. The paper is available on arXiv with ID 2507.22580.
Key facts
- RePaCA is a novel static APCA technique.
- It leverages LLMs specialized in thinking tasks.
- The model is prompted with buggy and fixed code snippets.
- It generates a Chain of Thought to analyze code differences.
- The goal is to identify overfitting patches.
- The technique requires no additional information beyond code snippets.
- It addresses issues of reliability, flexibility, and transparency.
- The paper is on arXiv with ID 2507.22580.
Entities
—