Does Genetic Programming Work Well on Automated Program Repair?

Qi, Yuhua; Mao, Xiaoguang; Lei, Yan; Dai, Ziying; Wang, Chengsong

doi:10.1109/iccis.2013.490

Cited by 16 publications

(12 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our early work [31], we have presented our insight with limited experiment. In this paper we will introduce our work in detail with the analysis to more experiment size.…”

Section: Introductionmentioning

confidence: 98%

The strength of random search on automated program repair

Mao

Lei

et al. 2014

Proceedings of the 36th International Conference on Software Engineering

Self Cite

267

240

View full text Add to dashboard Cite

Automated program repair recently received considerable attentions, and many techniques on this research area have been proposed. Among them, two genetic-programmingbased techniques, GenProg and Par, have shown the promising results. In particular, GenProg has been used as the baseline technique to check the repair effectiveness of new techniques in much literature. Although GenProg and Par have shown their strong ability of fixing real-life bugs in nontrivial programs, to what extent GenProg and Par can benefit from genetic programming, used by them to guide the patch search process, is still unknown.To address the question, we present a new automated repair technique using random search, which is commonly considered much simpler than genetic programming, and implement a prototype tool called RSRepair. Experiment on 7 programs with 24 versions shipping with real-life bugs suggests that RSRepair, in most cases (23/24), outperforms GenProg in terms of both repair effectiveness (requiring fewer patch trials) and efficiency (requiring fewer test case executions), justifying the stronger strength of random search over genetic programming. According to experimental results, we suggest that every proposed technique using optimization algorithm should check its effectiveness by comparing it with random search.

show abstract

“…In our early work [31], we have presented our insight with limited experiment. In this paper we will introduce our work in detail with the analysis to more experiment size.…”

Section: Introductionmentioning

confidence: 98%

The strength of random search on automated program repair

Mao

Lei

et al. 2014

Proceedings of the 36th International Conference on Software Engineering

Self Cite

267

240

View full text Add to dashboard Cite

show abstract

“…We motivate our study by observing that the impressive results achieved to date using genetic methods for improving software leverage the power of evolutionary search to only a limited extent. For example, most approaches rely primarily on mutation [14-16, 23-25, 35], use small population sizes 2 [14,23,24], and search for a small number of generations [14,[23][24][25]. Further, many of the repairs found using these methods can be reduced to a single mutation [18,24,25,35], suggesting that a large part of the successes achieved to date can be reduced to a form of random search.…”

Section: Introductionmentioning

confidence: 99%

Neutrality and epistasis in program space

Renzullo

Weimer

Moses³

et al. 2018

Proceedings of the 4th International Workshop on Genetic Improvement Workshop

View full text Add to dashboard Cite

Neutral networks in biology often contain diverse solutions with equal fitness, which can be useful when environments (requirements) change over time. In this paper, we present a method for studying neutral networks in software. In these networks, we find multiple solutions to held-out test cases (latent bugs), suggesting that neutral software networks also exhibit relevant diversity. We also observe instances of positive epistasis between random mutations, i.e. interactions that collectively increase fitness. Positive epistasis is rare as a fraction of the total search space but significant as a fraction of the objective space: 9% of the repairs we found to look (and 4.63% across all programs analyzed) were produced by positive interactions between mutations. Further, the majority (62.50%) of unique repairs are instances of positive epistasis. CCS CONCEPTS • Applied computing → Biological networks; • Software and its engineering → Software evolution; Search-based software engineering;

show abstract

“…We compare CURE with 25 APR techniques [1], [3], [5], [8]- [15], [18]- [20], [33], [34], [48]- [56]. Table I shows the comparison results.…”

Section: Ev a L U A T Io N A N D Re S U L T Smentioning

confidence: 99%

“…Type of bugs that CURE is applicable for: Similar to most state-of-the-art G&V APR techniques [1], [3], [5], [8]- [10], [12]- [15], [18]- [20], [33], [34], [49], [51]- [55], CURE is designed to fix single-hunk bugs (i.e., the buggy lines and patches are single code segments, and each buggy hunk has separate test cases).…”

Section: A Rq1: How Does Cure Perform Against State-of-the-art Apr Techniques?mentioning

confidence: 99%

CURE: Code-Aware Neural Machine Translation for Automatic Program Repair

Jiang

Lutellier

Tan

2021

2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE)

172

181

View full text Add to dashboard Cite

Automatic program repair (APR) is crucial to improve software reliability. Recently, neural machine translation (NMT) techniques have been used to fix software bugs automatically. While promising, these approaches have two major limitations. Their search space often does not contain the correct fix, and their search strategy ignores software knowledge such as strict code syntax. Due to these limitations, existing NMT-based techniques underperform the best template-based approaches. We propose CURE, a new NMT-based APR technique with three major novelties. First, CURE pre-trains a programming language (PL) model on a large software codebase to learn developer-like source code before the APR task. Second, CURE designs a new code-aware search strategy that finds more correct fixes by focusing on compilable patches and patches that are close in length to the buggy code. Finally, CURE uses a subword tokenization technique to generate a smaller search space that contains more correct fixes.Our evaluation on two widely-used benchmarks shows that CURE correctly fixes 57 Defects4J bugs and 26 QuixBugs bugs, outperforming all existing APR techniques on both benchmarks.

show abstract

Does Genetic Programming Work Well on Automated Program Repair?

Cited by 16 publications

References 13 publications

The strength of random search on automated program repair

The strength of random search on automated program repair

Neutrality and epistasis in program space

CURE: Code-Aware Neural Machine Translation for Automatic Program Repair

Contact Info

Product

Resources

About