Towards predictor models for large libre software projects

Abstract

Libre (free/open source) software provides an ample range of publicly available data sources about its development, which can be retrieved and analyzed. Consequently, it offers a good opportunity to build predictive estimation and evolution models. The main challenge to understand libre software development is that its development nature is radically different from 'classical' in-house software development, common in industry in the last decades. Developers and other human resources are generally a mixture of a few hired developers and many volunteers whose contribution (in number of hours per week and in total time devoted to the project) is not foreseeable in advance. This paper is a first step in finding predictive models in the libre software world. We have studied three data repositories (versioning system, mailing lists and bug tracking system) of GNOME, a large libre software project with several thousand contributors and several millions of lines of code, measuring activity and participation in it during the last years. Results and correlations for these sources allow us to adventure some first estimations of how participation and activity will evolve in the future.