Mining Python fix patterns via analyzing fine-grained source code changes

Many code changes are inherently repetitive, and researchers employ repetitiveness of the code changes to generate bug fix patterns. Automatic Program Repair (APR) can automatically detect and fix bugs, thus helping developers to improve the quality of software products. As a critical component of A...

Full description

Saved in:
Bibliographic Details
Published in:Empirical software engineering : an international journal 2022-03, Vol.27 (2), Article 48
Main Authors: Yang, Yilin, He, Tianxing, Feng, Yang, Liu, Shaoying, Xu, Baowen
Format: Article
Language:English
Subjects:
Automation
Collective Knowledge in Software Engineering
Compilers
Computer Science
Critical components
Fixing
Interpreters
Java
Open source software
Programming Languages
Python
Research methodology
Software
Software engineering
Software Engineering/Programming and Operating Systems
Software quality
Source code
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Many code changes are inherently repetitive, and researchers employ repetitiveness of the code changes to generate bug fix patterns. Automatic Program Repair (APR) can automatically detect and fix bugs, thus helping developers to improve the quality of software products. As a critical component of APR, software bug fix patterns have been revealed by existing studies to be very effective in detecting and fixing bugs in different programming languages (e.g., Java/C++); yet the fix patterns proposed by these studies can not be directly applied to improve Python programs because of syntactic incompatibilities and lack of analysis of dynamic features. In this paper, we proposed a mining approach to identify fix patterns of Python programs by extracting fine-grained bug-fixing code changes. We first collected bug reports from GitHub repository and employed the abstract syntax tree edit distance to cluster similar bug-fixing code changes to generate fix patterns. We then evaluated the effectiveness of these fix patterns by applying them to single-hunk bugs in two benchmarks (BugsInPy and QuixBugs). The results show that 13 out of 101 real bugs can be fixed without human intervention; that is, the generated bug patch is identical or semantically equivalent with developer’s patches. Also, we evaluated the fix patterns in the wild. For each complex bug, 15% of the bug code could be fixed, and 37% of the bug code could be matched by fix patterns.
ISSN:1382-3256
1573-7616
DOI:10.1007/s10664-021-10087-1