feat: fetch and clone from git bundle files

[rajdeepyadav004]

The need

Git bundles are the standard way to transfer repository data offline — they package refs + pack data into a single file. Regular git supports git fetch <bundle> and git clone <bundle> to pull objects and refs from them. Gitoxide currently has no equivalent.

This is relevant to the gix towards 1.0 roadmap (clone & fetch are core), and to C-integrate-gitoxide efforts where tools like eza/zed need offline-capable cloning.

Proposed API

A new GitBundle parser in gix-pack that reads the # v2 git bundle format, plus two methods on Repository:

// Fetch objects + refs from a bundle into an existing repo
let outcome = repo.fetch_from_bundle(path, &mut progress, &interrupt, opts)?;

// Create a new repo from a bundle
let repo = Repository::clone_from_bundle(dest, kind, path, &mut progress, &interrupt, None)?;

Design decisions

• Mirrors git behavior: no special --bundle flag — the CLI auto-detects .bundle files when passed as the remote argument
• Streaming: pack data streams directly from disk through gix's parallel pack processor, O(1) memory
• Incremental bundles: prerequisite refs are parsed and exposed (validation is left to the caller for now)
• Zero existing code touched: entirely additive, gated behind existing feature flags

Scope

What's included:

• Parse # v2 git bundle header format (refs, prerequisites, pack data)
• fetch_from_bundle() and clone_from_bundle() on Repository
• CLI auto-detection in clone and fetch commands
• is_bundle() helper for bundle file detection

What's intentionally deferred:

• Bundle creation (git bundle create equivalent)
• Bundle URI protocol (network bundle retrieval)
• Prerequisite auto-validation during fetch
• git bundle verify equivalent

Prior art

• Git's bundle.c — the reference implementation
• gix-pack already has Bundle::write_to_directory() for pack processing — this reuses it

Testing

• 26 unit tests for the parser (full, incremental, empty, CRLF, prerequisites, edge cases)
• 12 integration tests including an 89MB real-world shallow nixpkgs bundle (full and incremental, fetch and clone, bare and non-bare)
• All 360 existing integration tests pass unchanged

Disclosure

This proposal was developed with AI assistance. A working implementation exists on a feature branch and can be submitted as a PR if the approach looks right.

[Byron]

Thanks a lot for reporting and for the initial analysis.

This proposal was developed with AI assistance. A working implementation exists on a feature branch and can be submitted as a PR if the approach looks right.

It probably doesn't hurt to take a look at the implementation at some point, it sounds like it's more than 1k which seems outside of my review capacity for now.