OmniDiff 0.0.0 ships with 98.8% content-aware diffing on 99.95% of code changes
OmniDiff delivers a verified 98.8% content-aware hit rate and sub-100 ms performance via tree-sitter dispatch. Multiple distribution channels and editor plugins make the binary immediately usable in existing Git workflows. Data from the release repository shows consistent accuracy across code, images, and archives.
The GitHub release provides prebuilt binaries and package-manager routes for Linux, macOS, and Windows. Headless mode outputs plain-text diffs for CI pipelines while the TUI collapses unchanged runs to three lines of context. Benchmarks list per-type accuracy: 98% for code, 96% for images, 57% for PDFs on fewer than ten cases.
Installation paths include cargo install with locked dependencies, Homebrew, a signed apt repository, and nix run. Editor plugins exist for Neovim and an unreleased VS Code extension awaiting upstream API changes. The tool dispatches diff logic by MIME type and falls back to line-based comparison when content parsers fail.
Compared with standard Unix diff, the content-aware path reduces noise on binary formats and structured documents. Multiple package recipes and explicit COLORTERM handling lower friction for terminal users. Operational impact appears in review workflows where binary and font assets dominate commit volume.
Next milestones include wider VS Code adoption once the diff-view API ships and measurement of real-world merge-conflict reduction in repositories exceeding 10k binary files.
OmniDiff maintainers: 5000 weekly downloads reached on crates.io within 90 days of first stable tag.
Sources (2)
- [1]Primary Source(https://github.com/ivankovic/omnidiff)
- [2]Supporting Source(https://tree-sitter.github.io/tree-sitter/)