Since Darcs, Git and Mercurial were published 15 years ago as the first generation of distributed version control systems, distributed computing has seen exciting progress, in particular with mathematical formalisations of "ideal" distributed datastructures (CRDTs for example).

In this talk, I'll show our work on Pijul, a version control system with sound mathematical properties, making it easy and intuitive to use for non-coders, as well as scalable to arbitrarily large monorepos and binary files.

I'll explain how the core datastructures were "found" rather than designed, why we had to write a new open source key-value store to fork tables efficiently (zero-copy forks), and how that key-value store ended up breaking performance records.