Random Generation of Git Graphs

Abstract

Version Control Systems, such as Git and Mercurial, manage the history of a project as a Directed Acyclic Graph encoding the various divergences and synchronizations happening in its life cycle. A popular workflow in the industry, called the feature branch workflow, constrains these graphs to be of a particular shape: a unique main branch, and non-interfering feature branches. Here we focus on the uniform random generation of those graphs with n vertices, including k on the main branch, for which we provide three algorithms, for three different use-cases. The first, based on rejection, is efficient when aiming for small values of k (more precisely whenever k = O($\sqrt$ n)). The second takes as input any number k of commits in the main branch, but requires costly precalculation. The last one is a Boltzmann generator and enables us to generate very large graphs while targeting a constant k/n ratio. All these algorithms are linear in the size of their outputs.