September 13, 2024
The article titled “PyBugHive: A Comprehensive Database of Manually Validated, Reproducible Python Bugs” written by Gábor Antal, Norbert Vándor, István Kolláth, Balázs Mosolygó, Péter Hegedűs, and Rudolf Ferenc (Department of Software Engineering) has been accepted into the Q1 journal IEEE Access.
The article introduces PyBugHive, a manually curated database of 149 reproducible Python bugs from 11 projects, designed to aid empirical research in areas like testing and automated program repair. Each entry in the database includes a bug summary, patch, and test cases, accessible via a command-line interface for easy comparison of buggy and fixed versions.
The study also demonstrates PyBugHive's utility by evaluating GPT-3.5's performance in bug detection and automatic program repair using the database, finding that while the model detected 45% of the bugs, it was far less successful in generating accurate fixes, only being able to fix one of the detected issues.
The article can be read on IEEE Xplore.