CARDS (Corpus of Acyclic Repositories and Dependency Systems) is a collection of directed graphs which express dependency relations, extracted from diverse real-world sources such as package managers, version control systems, and event graphs. Each graph contains anywhere from thousands to hundreds of millions of nodes and edges, which are normalized into a simple, unified format. Both cyclic and acyclic variants are included (as some graphs, such as citation networks, are not entirely acyclic). The dataset is suitable for studying the structure of different kinds of dependencies, enabling the characterization and distinction of various dependency graph types. It has been utilized for developing and testing efficient algorithms which leverage the specificities of source version control graphs. The collection is publicly available at doi.org/10.5281/zenodo.14245890.

International audience

Version Control Systems, used by developers to keep track of the evolution of their code, model repositories as Merkle graphs of revisions. In order to synchronize efficiently between different instances of a repository, they need to determine the common knowledge that they share. This process is called discovery. In this paper, we provide theoretical definitions for the problem of discovery, establish some universal upper and lower bounds on the amount of data that needs to be exchanged, as well as NP-hardness for a restricted variant (with only 2 round-trips). We also present and analyze some algorithms that are used in extant VCSs, such as Mercurial and Git, and propose an algorithm based on chain-decomposition.

Previous work has shown that a lower bound on the number of words moved between large, slow memory and small, fast memory of size M by any conventional (non-Strassen like) direct linear algebra algorithm (matrix multiply, the LU, Cholesky, QR factorizations,...) is Ω(# flops / √ (M)). This holds for dense or sparse matrices. There are analogous lower bounds for the number of messages, and for parallel algorithms instead of sequential algorithms.