Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM-Based Repair of C++ Implicit Data Loss Compiler Warnings: An Industrial Case Study

Chansong You, Hyun Deok Choi, Jingun Hong

TL;DR

The paper addresses implicit data loss warnings in large-scale C++ projects and presents an LLM-driven automated program repair pipeline that leverages LSP for context, Tree-sitter for code extraction, and a self-consistency mechanism to decide when range checks are needed. It demonstrates the approach on a real-world SAP HANA component, achieving high patch acceptance (102 of 110, about 92.7%) and reducing the overhead of fixes compared to a baseline, while approaching human-designed solutions. The work highlights practical viability for integrating automated warnings repair into development workflows and discusses current limitations, such as handling high-level design changes and macro-expanded code. Overall, the study suggests that LLM-based repair can meaningfully reduce manual effort in maintaining performance-sensitive C++ code while preserving correctness and efficiency.

Abstract

This paper presents a method to automatically fix implicit data loss warnings in large C++ projects using Large Language Models (LLMs). Our approach uses the Language Server Protocol (LSP) to gather context, Tree-sitter to extract relevant code, and LLMs to make decisions and generate fixes. The method evaluates the necessity of range checks concerning performance implications and generates appropriate fixes. We tested this method in a large C++ project, resulting in a 92.73% acceptance rate of the fixes by human developers during the code review. Our LLM-generated fixes reduced the number of warning fix changes that introduced additional instructions due to range checks and exception handling by 39.09% compared to a baseline fix strategy. This result was 13.56% behind the optimal solutions created by human developers. These findings demonstrate that our LLM-based approach can reduce the manual effort to address compiler warnings while maintaining code quality and performance in a real-world scenario. Our automated approach shows promise for integration into existing development workflows, potentially improving code maintenance practices in complex C++ software projects.

LLM-Based Repair of C++ Implicit Data Loss Compiler Warnings: An Industrial Case Study

TL;DR

The paper addresses implicit data loss warnings in large-scale C++ projects and presents an LLM-driven automated program repair pipeline that leverages LSP for context, Tree-sitter for code extraction, and a self-consistency mechanism to decide when range checks are needed. It demonstrates the approach on a real-world SAP HANA component, achieving high patch acceptance (102 of 110, about 92.7%) and reducing the overhead of fixes compared to a baseline, while approaching human-designed solutions. The work highlights practical viability for integrating automated warnings repair into development workflows and discusses current limitations, such as handling high-level design changes and macro-expanded code. Overall, the study suggests that LLM-based repair can meaningfully reduce manual effort in maintaining performance-sensitive C++ code while preserving correctness and efficiency.

Abstract

This paper presents a method to automatically fix implicit data loss warnings in large C++ projects using Large Language Models (LLMs). Our approach uses the Language Server Protocol (LSP) to gather context, Tree-sitter to extract relevant code, and LLMs to make decisions and generate fixes. The method evaluates the necessity of range checks concerning performance implications and generates appropriate fixes. We tested this method in a large C++ project, resulting in a 92.73% acceptance rate of the fixes by human developers during the code review. Our LLM-generated fixes reduced the number of warning fix changes that introduced additional instructions due to range checks and exception handling by 39.09% compared to a baseline fix strategy. This result was 13.56% behind the optimal solutions created by human developers. These findings demonstrate that our LLM-based approach can reduce the manual effort to address compiler warnings while maintaining code quality and performance in a real-world scenario. Our automated approach shows promise for integration into existing development workflows, potentially improving code maintenance practices in complex C++ software projects.
Paper Structure (18 sections, 4 figures, 1 algorithm)

This paper contains 18 sections, 4 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Compiler Warnings in C++
  4. Compiler Warnings in Large-Scale C++ Projects
  5. Automated Program Repair
  6. Large Language Models in APR
  7. Methodology
  8. Context Search
  9. Context Extraction
  10. Patch Generation
  11. Range Check Determination
  12. Final Patch Generation
  13. Patch Validation
  14. Experimental Results
  15. Range Check Decision Making
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Compiler Warning Distribution in SAP HANA
  • Figure 2: System Architecture for Automated Warning Fix Generation
  • Figure 3: Number of correct range check decisions with various LLMs and number of samples used for self-consistency
  • Figure 4: Acceptance and rejection rate of LLM generated warning fixes