Abstract :
[en] Social coding platforms such as GitHub have profoundly transformed collaborative software development. Developers regularly contribute to numerous projects through mechanisms like pull requests, code reviews, and issue tracking. While this shift has fostered broader participation, it has also significantly increased the workload on maintainers, making repository management more complex and time-intensive. To mitigate this growing burden, developers adopt workflow configuration tools to automate repetitive and labour-intensive tasks such as testing, building, deploying, and monitoring code quality. Over the past two decades, tools like Travis, CircleCI, and Jenkins have played key roles in automating Continuous Integration and Continuous Deployment/Delivery (CI/CD) pipelines. In 2019, GitHub introduced GitHub Actions to help developers automate their software development workflows. While GitHub Actions quickly became the dominant CI/CD solution in GitHub, research remains relatively limited regarding its impact on collaborative development practices and the challenges developers face when integrating it into their workflows.
This dissertation therefore investigates the role of GitHub Actions on the collaborative development process. Through empirical studies combining both quantitative and qualitative methods, it has two core objectives: (1) understanding the adoption and usage patterns of GitHub Actions, and (2) examining the evolutionary dynamics of GitHub Actions workflows. These studies reveal important patterns related to workflow creation, maintenance, and long-term sustainability. A large-scale analysis based on software repository mining reveals that, while GitHub Actions enables powerful automation, its
integration into everyday development practices introduces new layers of complexity. Issues such as maintaining workflow correctness, adapting to changes in third-party Actions, along with the increasing complexity of managing configuration settings, emerge as recurring concerns. These findings shed light on the nuanced trade-offs developers face when relying on automation at scale and point to broader implications for the sustainability and maintainability of CI/CD workflows.
