Skip to content

ROADMAP.md

Latest commit

 

History

History
119 lines (89 loc) · 4.02 KB

ROADMAP.md

File metadata and controls

119 lines (89 loc) · 4.02 KB

Roadmap

Where this project is heading and how we plan to get there. Versions between today and 1.0 are intentionally rough — we'll cut releases as the work lands rather than holding up shipping for a perfect milestone boundary.

Today

v0.4.0 — released 2026-05-02. 596 of 794 vendored upstream shell tests pass (~75%) across 57 full-pass suites. Day-to-day flows (clean / smudge, fetch / pull / push, track, locking, migrate) work end-to-end. See tests/SCOREBOARD.md for the per-suite breakdown.

Path to 1.0

The bar for 1.0 is full feature parity with upstream Go git-lfs:

  • 100% pass on the vendored upstream shell test suite.
  • Comprehensive, polished documentation.
  • Polished --help output and man pages across every subcommand.
  • Packaged binary releases for Linux and macOS (Windows if there's demand).
  • A code-style and design audit, performed independently on each side.
  • A comparative audit against upstream Go git-lfs to surface any divergence the test suite doesn't catch.

Step releases

These are step targets, not contracts — the actual cuts will be opportunistic.

v0.4

Goals:

  • 75% test coverage (~+73 tests): Knock out the almost-done sweeps (7 partially-passing suites that are each one or two tests away from a full pass) and the cheap half of the long-tail polish (t-prune, t-ls-files, t-pointer, t-install, t-clone).
  • Initial sweep over man pages and help text. Make sure every help text and man page works, explains high-level concept at the right level of detail.

v0.5

Goals:

  • 85% test coverage (~+80 tests beyond v0.4): Networking + credentials cluster: t-credentials, t-credentials-no-prompt, t-credentials-protect, t-askpass, t-extra-header, t-content-type, t-fetch-recent, t-expired. Plus retry / Retry-After handling, which clears the t-batch-retries-* cluster.
  • Polished docs, man pages, help text. What this means:
  • Every man page should have the right sections (bug reporting, examples)
  • Help text should have arguments grouped into headings, where appropriate
  • Hosted docs should be comprehensive (installation instructions)
  • Style sweep over the prose

v0.6

Goals:

  • 90% test coverage
  • CI-driven binary builds for Linux (musl only?) and macOS (x86_64 + aarch64), tagged release artifacts on GitLab + mirrored on GitHub for discoverability.
  • A subset of the remaining greenfield commands and protocols: custom transfer adapters + tus, SSH (git-lfs-authenticate), merge-driver, smudge-side pointer extensions, dedup, completion, logs, update, standalone-file, t-multiple-remotes.

v0.7

Goals:

  • 95% test coverage
  • A the full remaining greenfield commands and protocols: custom transfer adapters + tus, SSH (git-lfs-authenticate), merge-driver, smudge-side pointer extensions, dedup, completion, logs, update, standalone-file, t-multiple-remotes.
  • An internal audit of the codebase: reviewing every flow, making sure that (a) the coding style legible and refactored, (b) invariants and algorithms are documented, and (c) there are no obvious bugs we can spot

v0.8

Goals:

  • 100% test coverage
  • A comparative audit with the Go git-lfs codebase, making sure that we did not miss any functionality that is not expressed through the test suite, or that our behaviour does not differ significantly (correctness).

v1.0

Goals:

  • Initial stable release
  • Full parity with Go git-lfs

Open questions

  • Windows support and packaging. Worth doing speculatively, or wait for someone to ask? Likely not worth the effort at this point.
  • Hot-path optimization. The current "shell out to git" model is great for parity, but a gix integration could be a meaningful speedup on large fetches/scans. Out of scope for 1.0; flag for a later major.
  • Crate API stability. The eight git-lfs-* library crates are published but their public API isn't pinned yet. Decide before 1.0 whether the binary and the libraries share a version.