Skip to content
girt

Overview

girt is a Rust library for applications that work with Git repositories. It opens a repository at a path you choose and works with its Git data directly: objects, trees, commits, tags, references, reflogs, the index and the working tree. Transfer is explicit. You fetch, clone and push over local paths, smart HTTP(S) or SSH, and your application decides when downloaded objects become reachable through a reference update.

girt is dual-licensed under MIT and Apache-2.0. It is an original implementation. Compatibility comes from Git’s format documentation and from observing the git executable, not from Git, gix or libgit2 source.

Who it’s for

Section titled “Who it’s for”

girt is for Rust programs that need Git storage and transfer as a library: version-control tools, forges, build systems and other programs that would otherwise shell out to git. Its first consumer is jj. An experimental jj branch replaces gix and the Git subprocess with girt and passes jj’s test suite. See girt as jj’s Git backend.

The girt CLI exercises the library on real repositories for inspection, experiments and comparisons with Git. It does not aim to reproduce Git’s full porcelain workflow.

What it supports

Section titled “What it supports”
  • Repositories: SHA-1 and SHA-256, ordinary and bare, linked worktrees, discovery, layered configuration, alternates and shallow history.
  • Objects: blobs, trees, commits and tags in loose and packed storage, pack reading and writing, delta compression, history walks, merge bases, tree comparison and content diffs.
  • References: the files backend (loose and packed) and reftable, conditional transactions, reflogs and tag peeling.
  • Working tree: index v2, v3 and v4, raw status and tree checkout on macOS and Linux.
  • Transfer: fetch, clone and push over local paths without running Git, over smart HTTP(S) (http feature) and through system OpenSSH (ssh feature).
  • Maintenance: retention planning, additive repacking, reflog expiry, pack retirement and loose object pruning.

Supported and refused lists what each area supports and what it rejects.

Where it stands

Section titled “Where it stands”

girt is experimental and its API still changes between minor releases. It runs jj’s full test suite and an end-to-end script against real Git. Performance is the main open work. Medium-sized repositories are mostly at parity with stock jj, while some operations on very large repositories are still several times slower. The benchmark shows where.

How this site relates to the API docs

Section titled “How this site relates to the API docs”

The API documentation on docs.rs is self-contained. Its crate and module pages explain each area, and every item documents its contract, limits and how to recover from failure. This site covers tasks and concepts that span several modules, the safety and compatibility model, and performance results. It links to docs.rs for details instead of repeating them.

Next steps

Section titled “Next steps”