Abstract

Modern release engineering has moved from longer release cycles and separate development and release teams to a continuous and integrated process. However, release engineering practices include not only integration, build and test execution but also a better management of features. The goal of this research is to investigate the modern release engineering practices which cover four milestones in the field of release engineering, i. understanding rapid release by measuring the time and effort involved in release cycles, ii. feature management based on feature toggles iii. the impact of toggles on the system architecture, and iv. the quality of builds that contain ignored failing and flaky tests. This thesis is organized as a “manuscript” thesis whereby each milestone constitutes an accepted or submitted paper.
First, we investigate the rapid release model for two major open source software projects. We quantify the time and effort which is involved in both the development and stabilization phases of a release cycle where, we found that despite using the rapid release process, both the Chrome Browser and the Linux Kernel have a period where developers rush changes to catch the current release.
Second, we examine feature management based on feature toggles which is a widely used technique in software companies to manage features by turning them on/off during development as well as release periods. Developers typically isolate unrelated/unreleased changes on branches. However, large companies, such as Google and Facebook do their development on single branch. They isolate unfinished features using feature toggles that allow them to disable unstable code. Third, feature toggles provide not only a better management of features but also keep modules isolated and feature oriented which makes the architecture underneath the source code readable and iiieasily extractable. As the project grows, modules keep accepting features and features cross-cut into the modules. We found that the architecture can be easily extracted based on feature toggles and provides a different view compared to the traditional modular representations of software architecture.
Fourth, we investigate the impact of failing tests on the quality of builds where we consider browser-crash as a quality factor. In this study we found that ignoring failing and flaky tests leads to dramatically more crashes than builds with all tests passing.