Abstract
Code changes propagate. Type, frequency, size of changes typically explain and even predict impact of changes in software products. What can changes tell about software processes? In this study, we propose a novel method to render software processes by graphs of linked commits as carriers of change information. Mining histories in such commit graphs allows to exploit techniques of graph analysis and coloring that can be used to understand activities in software processes. As application of our method, we analysed colored commit graphs to investigate the presence of large architectural changes and their likelihood of occurrence in bug fixing. For this, we introduced a new measure of architectural change based on hashing and a linear-time kernel for bit-labels graphs. We applied our approach to analyse the evolution of change of Eclipse JDT and Spring Framework.