On The Effectiveness of Dynamic Reduction Techniques in Automated Program Repair
Omar I. Al-Bataineh
TL;DR
This work addresses the scalability bottlenecks of automated program repair (APR) on large-scale software by introducing dynamic reduction through observation-based slicing (ORBS). By integrating ORBS with a state-of-the-art template-based APR (TBar), the approach reduces test-suite size and suspicious-list size while preserving faulty behavior, thereby speeding up fault localization, patch generation, and patch validation. The authors formalize soundness criteria for APR reductions, demonstrate reliable configurations, and provide empirical results on Defects4J showing substantial improvements in repair time and search space without degrading patch quality. Overall, the method offers a practical pathway to scalable APR for industrial complexity and lays groundwork for broader evaluation and refinement.
Abstract
Repairing a large-scale buggy program using current automated program repair (APR) approaches can be a time-consuming operation that requires significant computational resources. We describe a program repair framework that effectively handles large-scale buggy programs of industrial complexity. The framework exploits program reduction in the form of program slicing to eliminate parts of the code irrelevant to the bug being repaired without adversely affecting the capability of the repair system in producing correct patches. Observation-based slicing is a recently introduced, language-independent slicing technique that shows a good effectiveness in a wide range of applications. In this work, we show how ORBS can be effectively integrated with APR to improve all aspects of the repair process including the fault localization step, patch generation step, and patch validation step. The presented repair framework indeed enhances the capability of APR by reducing the execution cost of a test suite and the search cost for the appropriate faulty statement corresponding to the bug being repair. Our empirical results on the widely used Defects4J dataset reveal that a substantial improvement in performance can be obtained without any degradation in repair quality.