Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Program Repair by Jointly Learning to Localize and Repair

Marko Vasic, Aditya Kanade, Petros Maniatis, David Bieber, Rishabh Singh

TL;DR

This work targets variable-misuse bugs by jointly learning to classify, localize, and repair using multi-headed pointer networks over program tokens. By predicting both the bug location and a repair variable in a single model, it overcomes limitations of prior enumerative methods that treat localization and repair separately. Across ETH-Py150 and MSR-VarMisuse datasets, the joint model achieves superior localization and repair accuracy while maintaining high classification and true-positive rates, and demonstrates robustness and practical relevance through industrial, real-world data. The results suggest that integrating token-level pointers with end-to-end optimization offers a promising direction for neural program repair, with potential extensions to include richer semantic information via graph-based approaches.

Abstract

Due to its potential to improve programmer productivity and software quality, automated program repair has been an active topic of research. Newer techniques harness neural networks to learn directly from examples of buggy programs and their fixes. In this work, we consider a recently identified class of bugs called variable-misuse bugs. The state-of-the-art solution for variable misuse enumerates potential fixes for all possible bug locations in a program, before selecting the best prediction. We show that it is beneficial to train a model that jointly and directly localizes and repairs variable-misuse bugs. We present multi-headed pointer networks for this purpose, with one head each for localization and repair. The experimental results show that the joint model significantly outperforms an enumerative solution that uses a pointer based model for repair alone.

Neural Program Repair by Jointly Learning to Localize and Repair

TL;DR

This work targets variable-misuse bugs by jointly learning to classify, localize, and repair using multi-headed pointer networks over program tokens. By predicting both the bug location and a repair variable in a single model, it overcomes limitations of prior enumerative methods that treat localization and repair separately. Across ETH-Py150 and MSR-VarMisuse datasets, the joint model achieves superior localization and repair accuracy while maintaining high classification and true-positive rates, and demonstrates robustness and practical relevance through industrial, real-world data. The results suggest that integrating token-level pointers with end-to-end optimization offers a promising direction for neural program repair, with potential extensions to include richer semantic information via graph-based approaches.

Abstract

Due to its potential to improve programmer productivity and software quality, automated program repair has been an active topic of research. Newer techniques harness neural networks to learn directly from examples of buggy programs and their fixes. In this work, we consider a recently identified class of bugs called variable-misuse bugs. The state-of-the-art solution for variable misuse enumerates potential fixes for all possible bug locations in a program, before selecting the best prediction. We show that it is beneficial to train a model that jointly and directly localizes and repairs variable-misuse bugs. We present multi-headed pointer networks for this purpose, with one head each for localization and repair. The experimental results show that the joint model significantly outperforms an enumerative solution that uses a pointer based model for repair alone.

Paper Structure

This paper contains 15 sections, 8 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Pointer Models for Localization and Repair of VarMisuse
  4. Problem Definition
  5. Multi-headed Pointer Models
  6. Training the Model
  7. Evaluation
  8. Benchmarks
  9. Joint Model vs. Enumerative Approach
  10. Performance Comparison:
  11. Effect of Incorrect Slot Placement
  12. Comparison of Graph and Pointer Networks
  13. Evaluation on Variable Misuse in Practice
  14. Conclusion
  15. Training Data Generation

Figures (4)

  • Figure 1: Enumerative solution to the VarMisuse problem.
  • Figure 2: Illustration of our model on a real-example (left), and design of the model (right).
  • Figure 3: The drop in repair accuracy of the repair model due to incorrect slot placement.
  • Figure 4: A sample of differing repair predictions (and prediction probabilities) for slots shown in blue and injected bugs enclosed in red.