Characterization of design discussions in modern code review.

Abstract

Modern Code Review (MCR) is an increasingly popular lightweight activity for improving software quality. As part of MCR, developers participate in a number of discussions which are stored in tools for supporting the process. By analyzing these discussions, researchers found that there is a considerable amount of design information within them. They also had positive results in their studies of techniques for automatically identifying design discussions in the reviews. However, most of this research is quantitative and has not thoroughly analyzed questions such as how developers conduct design discussions and what topics of design they discuss. To retrieve design information more effectively from code review discussions, it is necessary to know how developers discuss design during code reviews in order to be able to distinguish design information from the rest. Furthermore, it is necessary to know what kind of design information exist in these discussions. With the goal of better understanding MCR and the process of how design is discussed in MCR in order to fill the current knowledge gaps, we performed a qualitative study to characterize design information in MCR by applying Straussian Grounded Theory (GT) to a dataset of design discussions from open source software (OSS) projects of the Apache Software Foundation. As a result, we produced a model of how developers discuss design during code review, a classification of types of design information discussed in MCR and a dataset of design discussions. We believe our work will be of significant help in future research aiming to extract design information from MCR discussions in a manner that is useful to practitioners.