Visualizing GitHub Contributor Activity in ExplorViz

Abstract

As software projects grow, both their structure and their team dynamics become opaque. ExplorViz visualizes the structure and runtime behavior of software systems as interactive 3D software cities, but provides no view of contributor activity. Such a view could highlight development bottlenecks or reveal risks, such as code known to only a few contributors. We extract contributor data from Git commits, fetch issues and pull requests through the GitHub GraphQL API and persist both in an extended Neo4j data model. We compute eight formally defined metrics from this data and integrate them into the existing metrics system of ExplorViz, so they can be displayed as heat maps or mapped onto building geometry. We evaluate the extension through a performance analysis, a case study on two real repositories and a user study with eleven participants. Fetching issues and pull requests slowed down the analysis by a factor of 1.18 for a repository with few of them per commit and 11.54 for a repository with many. The reviewed scores are consistent with the underlying data, but the metrics do not always measure what they were designed for. Participants gave 63 correct answers out of 77, but eight of the eleven reported trouble telling what the metrics meant from their names alone. Commit-based metrics work on any Git repository, whereas the pull request and issue metrics additionally require a GitHub project that consistently links commits to pull requests and pull requests to issues.