TY - BOOK

T1 - Cognitive-support code review tools

T2 - improved efficiency of change-based code review by guiding and assisting reviewers

AU - Baum, Tobias

N1 - Doctoral thesis

PY - 2019

Y1 - 2019

N2 - Code reviews, i.e., systematic manual checks of program source code by other developers, have been an integral part of the quality assurance canon in software engineering since their formalization by Michael Fagan in the 1970s. Computer-aided tools supporting the review process have been known for decades and are now widely used in software development practice. Despite this long history and widespread use, current tools hardly go beyond simple automation of routine tasks. The core objective of this thesis is to systematically develop options for improved tool support for code reviews and to evaluate them in the interplay of research and practice. The starting point of the considerations is a comprehensive analysis of the state of research and practice. Interview and survey data collected in this thesis show that review processes in practice are now largely change-based, i.e., based on checking the changes resulting from the iterative-incremental evolution of software. This is true not only for open source projects and large technology companies, as shown in previous research, but across the industry. Despite the common change-based core process, there are various differences in the details of the review processes. The thesis shows possible factors influencing these differences. Important factors seem to be the process variants supported and promoted by the used review tool. In contrast, the used tool has little influence on the fundamental decision to use regular code reviews. Instead, the interviews and survey data suggest that the decision to use code reviews depends more on cultural factors. Overall, the analysis of the state of research and practice shows that there is a potential for developing better code review tools, and this potential is associated with the opportunity to increase efficiency in software development. The present thesis argues that the most promising approach for better review support is reducing the reviewer's cognitive load when reviewing large code changes. Results of a controlled experiment support this reasoning. The thesis explores various possibilities for cognitive support, two of these in detail: Guiding the reviewer by identifying and presenting a good order of reading the code changes being reviewed, and assisting the reviewer through automatic determination of change parts that are irrelevant for review. In both cases, empirical data is used to both generate and test hypotheses. In order to demonstrate the practical suitability of the techniques, they are also used in a partner company in regular development practice. For this evaluation of the cognitive support techniques in practice, a review tool which is suitable for use in the partner company and as a platform for review research is needed. As such a tool was not available, the code review tool "CoRT" has been developed. Here, too, a combination of an analysis of the state of research, support of design decisions through scientific studies and evaluation in practical use was employed. Overall, the results of this thesis can be roughly divided into three blocks: Researchers and practitioners working on improving review tools receive an empirically and theoretically sound catalog of requirements for cognitive-support review tools. It is available explicitly in the form of essential requirements and possible forms of realization, and additionally implicitly in the form of the tool "CoRT". The second block consists of contributions to the fundamentals of review research, ranging from the comprehensive analysis of review processes in practice to the analysis of the impact of cognitive abilities (specifically, working memory capacity) on review performance. As the third block, innovative methodological approaches have been developed within this thesis, e.g., the use of process simulation for the development of heuristics for development teams and new approaches in repository and data mining.

AB - Code reviews, i.e., systematic manual checks of program source code by other developers, have been an integral part of the quality assurance canon in software engineering since their formalization by Michael Fagan in the 1970s. Computer-aided tools supporting the review process have been known for decades and are now widely used in software development practice. Despite this long history and widespread use, current tools hardly go beyond simple automation of routine tasks. The core objective of this thesis is to systematically develop options for improved tool support for code reviews and to evaluate them in the interplay of research and practice. The starting point of the considerations is a comprehensive analysis of the state of research and practice. Interview and survey data collected in this thesis show that review processes in practice are now largely change-based, i.e., based on checking the changes resulting from the iterative-incremental evolution of software. This is true not only for open source projects and large technology companies, as shown in previous research, but across the industry. Despite the common change-based core process, there are various differences in the details of the review processes. The thesis shows possible factors influencing these differences. Important factors seem to be the process variants supported and promoted by the used review tool. In contrast, the used tool has little influence on the fundamental decision to use regular code reviews. Instead, the interviews and survey data suggest that the decision to use code reviews depends more on cultural factors. Overall, the analysis of the state of research and practice shows that there is a potential for developing better code review tools, and this potential is associated with the opportunity to increase efficiency in software development. The present thesis argues that the most promising approach for better review support is reducing the reviewer's cognitive load when reviewing large code changes. Results of a controlled experiment support this reasoning. The thesis explores various possibilities for cognitive support, two of these in detail: Guiding the reviewer by identifying and presenting a good order of reading the code changes being reviewed, and assisting the reviewer through automatic determination of change parts that are irrelevant for review. In both cases, empirical data is used to both generate and test hypotheses. In order to demonstrate the practical suitability of the techniques, they are also used in a partner company in regular development practice. For this evaluation of the cognitive support techniques in practice, a review tool which is suitable for use in the partner company and as a platform for review research is needed. As such a tool was not available, the code review tool "CoRT" has been developed. Here, too, a combination of an analysis of the state of research, support of design decisions through scientific studies and evaluation in practical use was employed. Overall, the results of this thesis can be roughly divided into three blocks: Researchers and practitioners working on improving review tools receive an empirically and theoretically sound catalog of requirements for cognitive-support review tools. It is available explicitly in the form of essential requirements and possible forms of realization, and additionally implicitly in the form of the tool "CoRT". The second block consists of contributions to the fundamentals of review research, ranging from the comprehensive analysis of review processes in practice to the analysis of the impact of cognitive abilities (specifically, working memory capacity) on review performance. As the third block, innovative methodological approaches have been developed within this thesis, e.g., the use of process simulation for the development of heuristics for development teams and new approaches in repository and data mining.

M3 - Doctoral thesis

CY - Hannover

ER -