Abstract

As software development is becoming increasingly complex, developers often need to reuse others’ code or knowledge made available online to tackle problems encountered during software development and maintenance. This phenomenon of using others' code or knowledge, often found on online forums, is referred to as crowdsourcing. A good example of crowdsourcing is posting a coding question on the Stack Overflow website and having others contribute code that solves that question. Recently, the phenomenon of crowdsourcing has attracted much attention from researchers and practitioners and recent studies show that crowdsourcing improves productivity and reduces time-to-market. However, like any solution, crowdsourcing brings with it challenges such as quality, maintenance, and even legal issues.

The research presented in this thesis presents the result of a series of large-scale empirical studies involving some of the most popular crowdsourcing platforms such as Stack Overflow, Node Package Manager (npm), and Python Package Index (PyPI). The focus of these empirical studies is to investigate the role of reusing crowdsourcing knowledge and more particularly crowd code in the software development process.

We first present two empirical studies on the reuse of knowledge from crowdsourcing platforms namely Stack Overflow. We found that reusing knowledge from this crowdsourcing platform has the potential to assist software development practices, specifically through source code reuse.
However, relying on such crowdsourced knowledge might also negatively affect the quality of the software projects. Second, we empirically examine the type of development knowledge constructed on crowdsourcing platforms. We examine the use of trivial packages on npm and PyPI platforms. We found that trivial packages are common and developers tend to use them because they provide them with well tested and implemented code. However, developers are concerned about the maintenance overhead of these trivial packages due to the extra dependencies that trivial packages introduce. Finally, we used the gained knowledge to propose a pragmatic solution to improve the efficiency of relying on the crowd in software development. We proposed a rule-based technique that automatically detects commits that can skip the continuous integration process. We evaluate the performance of the proposed technique on a dataset of open-source Java projects. Our results show that continuous integration can be used to improve the efficiency of the reused code from crowdsourcing platforms.

Among the findings of this thesis are that the way software is developed has changed dramatically. Developers rely on crowdsourcing to address problems encountered during software development and maintenance. The results presented in this thesis provides new insights on how knowledge from these crowdsourced platforms is reused in software systems and how some of this knowledge can be better integrated into current software development processes and best practices.