A machine learning based automatic folding of dynamically typed languages

Abstract

The popularity of dynamically typed languages has been growing strongly lately. Elegant syntax of such languages like javascript, python, PHP and ruby pays back when it comes to finding bugs in large codebases. The analysis is hindered by specific capabilities of dynamically typed languages, such as defining methods dynamically and evaluating string expressions. For finding bugs or investigating unfamiliar classes and libraries in modern IDEs and text editors features for folding unimportant code blocks are implemented. In this work, data on user foldings from real projects were collected and two classifiers were trained on their basis. The input to the classifier is a set of parameters describing the structure and syntax of the code block. These classifiers were subsequently used to identify unimportant code fragments. The implemented approach was tested on JavaScript and Python programs and compared with the best existing algorithm for automatic code folding.