Step-by-Step Reasoning for Explainable Fault Localization

In the quest to streamline the identification of erroneous code segments, a novel integration of Large Language Models (LLMs) is being leveraged to provide developers with clear explanations for faults within programs. The introduction of FuseFL—an approach combining multiple data sources—significantly enhances LLM’s fault localization capabilities, as evidenced by the Refactory dataset results.

• LLMs assist developers in reasoning about code.
• The new approach, FuseFL, integrates fault localization results, test case outcomes, and code descriptions.
• Benchmarking demonstrates over a 30% improvement at Top-1 in fault localization.
• A dataset of human explanations coupled with human studies validates the FuseFL-generated explanations.

The ability of LLMs to simplify complex fault localization tasks not only boosts developer productivity but also paves the way for more intuitive debugging tools. Read more.