bru

https://zieka.github.io/bru

Homebrew, reimagined in native code.

A fast, native implementation of common Homebrew commands written in Zig. Drop-in compatible — falls back to brew for anything not yet implemented.

Install

curl -fsSL https://zieka.github.io/bru/install.sh | bash

Usage

Use bru exactly like you'd use brew:

bru install ripgrep
bru search fd
bru upgrade
bru services start postgresql@14
bru bundle dump
bru rollback ripgrep

Commands

39 native commands including install, upgrade, search, list, info, deps, doctor, cleanup, plus:

• services — manage background services (start/stop/restart)
• bundle — export and install from Brewfiles
• rollback — revert a formula to its previous version
• self-update — update bru itself
• post_install execution — runs Homebrew post_install Ruby blocks via the system Ruby (/usr/bin/ruby on macOS), so packages like node, openssl, fontconfig, and ca-certificates are fully configured at install time. Disable per-invocation with --no-post-install, override the Ruby path with --use-system-ruby=<path>, or set BRU_NO_POST_INSTALL=1. Failures are logged to <cache>/post-install-logs/ and never block the install.

Anything not implemented natively falls back to the real brew binary.

How bru compares

vs Homebrew

	Homebrew	bru
Language	Ruby	Zig (single static binary)
Startup	~1.5s interpreter load	Near-zero
Formula lookups	Parse 25MB JSON	O(1) via mmap'd binary index
Auto-update	Runs on every install	Never — bru update is explicit
Rollback	Not supported	bru rollback <formula> with full history
Memory	Ruby GC pauses	Arena allocator (free is a no-op)
Downloads	Sequential	Parallel with worker threads

vs zerobrew

	zerobrew	bru
Drop-in compatible	No — separate /opt/zerobrew prefix, requires migration	Yes — same prefix, same Cellar, interchangeable with brew
Fallback	None — switch to brew manually	Transparent — unimplemented commands delegate to brew automatically
Formula index	Parses JSON API on every lookup	Binary index (BRUI format), mmap'd O(1) lookups
Dependencies	~30 Rust crates, SQLite, Tokio	Zero — Zig stdlib only
Binary size	7.9 MB	Single static binary, significantly smaller
Migration	Required (zb migrate)	Not needed — works on existing Homebrew installation
Casks	Not supported	.app casks (binary-and-app artifacts) work; brew falls back for .pkg/font casks
Correctness	Skips patching ffmpeg's libav*.pc pkg-config files — @@HOMEBREW_PREFIX@@ placeholders leak through (see Benchmark below)	Full pkg-config patching alongside Mach-O relocation

vs nanobrew

	nanobrew	bru
Drop-in compatible	No — separate /opt/nanobrew prefix	Yes — same prefix, same Cellar
Fallback	None — unsupported commands fail	Transparent — delegates to brew
Tar extraction	Shells out to system tar	Native Zig tar with parallel buffer pool
Tab files	Separate state tracking	Writes brew-compatible INSTALL_RECEIPT.json — brew sees bru-installed packages as its own
Mach-O patching	Patches dylibs but leaves *.pc files unpatched	Full relocation via install_name_tool + codesign for both binaries and pkg-config
Data files	Skips bundle data (e.g. tesseract tessdata/eng.traineddata doesn't extract) — tesseract --list-langs finds nothing	Extracts everything declared in the bottle
Migration	Required (nb migrate)	Not needed

vs malt

	malt	bru
Drop-in compatible	No — separate /opt/malt prefix	Yes — same prefix, same Cellar
Fallback	None	Transparent — delegates to brew
Warm-cache hot path	Extracts to content-addressed store, then APFS clonefile into the user-visible Cellar (skips re-relocation on re-install)	Same APFS clonefile strategy, with the relocation cached per-bottle in kegs/<sha>/
Mach-O patching	Patches dylibs but partial pkg-config coverage — ffmpeg's libavdevice.pc ships with @@HOMEBREW_PREFIX@@ placeholders	Full pkg-config patching
Data files	Same as nanobrew — tesseract tessdata/ is missing after install	Extracts data subdirectories declared in the bottle
Casks	Supported	Supported

The key difference

bru is designed as a transparent accelerator, not a parallel ecosystem. The other native alternatives chase raw install speed by skipping the unglamorous patching work — pkg-config files, language data, signature re-application — leaving subtly broken installs that pass <bin> --version but fail real consumers (compilers calling pkg-config --libs libsodium, tesseract --list-langs, etc). bru does the full patching work, keeps the brew-compatible Cellar layout, and falls back to brew for anything unimplemented. Alias brew=bru and everything works — faster for native commands, identical for the rest.

Benchmark (beta)

test/benchmark/ ships an end-to-end correctness × speed benchmark that compares bru against brew and three other native homebrew-alikes (zerobrew, nanobrew, malt) by actually installing real formulae and casks.

For each (tool, package) pair the bench:

1. Uninstalls the package and surgically evicts every cached blob keyed by the bottle's SHA256 (brew API → cache/blobs + store, across each tool's content-addressed prefix).
2. Times a cold install (no cached bottle on disk).
3. Uninstalls again, times a warm install (bottle in cache but not yet linked).
4. Runs a correctness probe while the install is on disk:
   • ffmpeg: ffprobe present, every libav*.pc patched, -filters lists 200+ plugins, transcode a generated test pattern to PNG.
   • libsodium: dylib has no @@PLACEHOLDER@@, lib/pkgconfig/libsodium.pc exists and is patched.
   • duckdb: libduckdb.dylib install_name patched, lib/cmake/DuckDB/*.cmake clean, statically-linked parquet round-trip + json_extract both return expected values.
   • tesseract: --list-langs finds eng/osd (catches missing tessdata/).
   • act: binary launches, --help prints the expected usage banner.
   • rectangle (cask): Rectangle.app is in /Applications/, executable inside the bundle, codesign --verify passes.

Run it:

bash test/benchmark/bench_contenders.sh   # ensure latest brew, zb, nb, mt; build bru
bash test/benchmark/bench_chart.sh        # measure + render

Each per-package result lands at test/benchmark/results-<package>.svg. Green ✓ = verified, red ! = passes --version but fails a deeper probe, grey dot = not supported by that tool.

Beta caveats:

• The deeper probes are formula-specific (hand-written per package). Adding a new package to PACKAGES / CASKS env vars works for timing but uses a --version-only fallback for correctness until a probe is written for it.
• "Cold" install timing depends on surgical cache eviction; the bench uses brew's API to derive per-tool cache paths, so a new tool with a custom content-addressing scheme would need its clear_cache function updated.
• Warm-cache numbers are sensitive to disk state and CPU load — run the bench on an otherwise-idle machine for trustworthy comparisons.

Building from source

Requires Zig 0.15+:

zig build -Doptimize=ReleaseFast

Run tests:

zig build test