Automatic Mining of Code Fix Patterns from Code Repositories

The majority of conventional automated approaches to examination and correction of source code (such as static and dynamic analysis methods) often prove to be not fully adequate for modern software engineers. They generally demand the existence of preprescribed patterns of unwanted code and/or proposed patches. Thus, such solutions are limited in implementation to a relatively narrow set of widely known cases.The approach proposed and implemented in this work allows for the rapid adaptation of the developed tool to yet unknown and domain-specific cases, while offering more fine-granularity, by focusing on vertex-to-vertex matching of code modifications represented by abstract syntax trees and comparison of vertex data using word-embedding tools. The evaluation is based on real C, C++, and C# code repositories from open source OS Tizen, as well as synthetic tests.