Just-in-time code duplicates extraction

Refactoring is a critical task in software maintenance, and usually performed to enforce better design coding practices, while coping with defects. The Extract Method refactoring widely used for merging duplicate code fragments into single new method. Several studies attempted recommend opportunities using different techniques, including program slicing, dependency graph analysis, change history structural similarity, feature extraction. However, irrespective of the method, most existing approaches interfere developer’s workflow: they require developer stop analyze suggested opportunities, also consider all suggestions entire project without focusing on development context. To increase adoption refactoring, this paper, we aim investigate effectiveness machine learning deep algorithms its recommendation maintaining workflow developer. proposed approach relies mining prior applied refactorings extracting their features train classifier that detects them user’s code. We implemented our as plugin IntelliJ IDEA called AntiCopyPaster. develop approach, trained evaluated various popular models dataset 18,942 from 13 Open Source Apache projects. results show best model Convolutional Neural Network (CNN), which recommends appropriate an F-measure 0.82. conducted qualitative study 72 developers evaluate usefulness developed plugin. tend appreciate idea are satisfied aspects plugin’s operation.