Improving Search-Based Automatic Program Repair With Neural Machine Translation
Open Access
- 4 April 2022
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 51167-51175
Abstract
The challenge of automatically repairing bugs in programs to reduce debugging expenses and increase program quality is known as automated program repair. To overcome this issue, test-suite-based repair techniques use a specified test suite as an oracle and alter the input faulty program to pass the full test suite. GenProg is a well-known example of this kind of repair, in which genetic programming is used to reorder the statements already present in the faulty program. However, recent practical experiments suggest that GenProg’s performance, notably for Java, is not sufficient. Improved program dependability necessitates the use of automatic program repair techniques. Template-based program repair techniques have recently been combined with search-based techniques to solve program issues automatically. Although intriguing, it has two fundamental drawbacks: Its search space often lacks the correct solution, and the technique disregards program expertise, such as precise code language. Compared with the template-based program repair approach, existing neural-machine-translation-based approaches are not limited by these constraints due to their ability to learn and generate new solutions. We propose an approach that combines a search-based automatic program repair technique with a neural-machine-translation-based approach. More specifically, we use both redundancy assumption and sequence-to-sequence learning of correct patches as the source for potential fix statements that feed into a multiobjective evolutionary search algorithm to find test-suite-adequate patches. In this work, a novel framework called ARJANMT is introduced for automatically repairing Java programs. Two sets of controlled experiments are conducted on 410 bugs from two benchmarks to investigate the repairability and correctness of our proposed framework. A comparison between state-of-the-art automatic program repair frameworks is made. The experimental results indicate that combining those two types of repair techniques (search-based and neural-machine-translation-based) produces better results or fixes bugs that they previously were unable to fix individually.Keywords
Funding Information
- USA National Science Foundation (1757828, 1822137, 2050869)
This publication has 13 references indexed in Scilit:
- CoCoNuT: combining context-aware neural translation models using ensemble for program repair2020
- SEQUENCER: Sequence-to-Sequence Learning for End-to-End Program RepairIEEE Transactions on Software Engineering, 2019
- Empirical review of Java program repair tools: a large-scale experiment on 2,141 bugs and 23,551 repair attempts2019
- A hybrid evolutionary system for automatic software repair2019
- ARJA: Automated Repair of Java Programs via Multi-Objective Genetic ProgrammingIEEE Transactions on Software Engineering, 2018
- Be more familiar with our enemies and pave the way forward: A review of the roles bugs played in software failuresJournal of Systems and Software, 2017
- DynaMoth2016
- Nopol: Automatic Repair of Conditional Statement Bugs in Java ProgramsIEEE Transactions on Software Engineering, 2016
- GenProg: A Generic Method for Automatic Software RepairIEEE Transactions on Software Engineering, 2011
- Recent Catastrophic Accidents: Investigating How Software was Responsible2010