Harvesting Fix Hints in the History of Bugs
This addresses the time-consuming and cost-sensitive task of bug fixing for software developers, offering a novel method to leverage historical data for recommendations.
The paper tackles the problem of automating bug fixes by proposing a new approach that recommends fix actions based on historical bug reports and fixes, using information retrieval and machine learning to compare new bugs to past ones and provide top-k hints for developers.
In software development, fixing bugs is an important task that is time consuming and cost-sensitive. While many approaches have been proposed to automatically detect and patch software code, the strategies are limited to a set of identified bugs that were thoroughly studied to define their properties. They thus manage to cover a niche of faults such as infinite loops. We build on the assumption that bugs, and the associated user bug reports, are repetitive and propose a new approach of fix recommendations based on the history of bugs and their associated fixes. In our approach, once a bug is reported, it is automatically compared to all previously fixed bugs using information retrieval techniques and machine learning classification. Based on this comparison, we recommend top-{\em k} fix actions, identified from past fix examples, that may be suitable as hints for software developers to address the new bug.