Literal grep in pure Zig

Ripgrep speed, one static binary, zero excuses.

Recursive search scans every byte on every query. ziggygrep cuts the cost per byte instead of indexing around it. First byte prefilter skips what cannot match, line bounding runs on verified hits only. Output matches rg -F on all 7 test patterns.

Build in 30 seconds Get v0.1.0 binary

Free MIT · 656 KB Windows · 326 KB Linux · 22/22 benches at or below rg

ziggygrep bench console

$ ziggygrep HashMap . > /dev/null

200 files / 53 MB · median of 9

ziggygrep 37ms · rg 40ms · delta -3ms

$ ziggygrep x . > /dev/null

ziggygrep 66ms · rg 105ms · delta -39ms

$ ziggygrep ZZZNOMATCHXYZ . > /dev/null

ziggygrep 24ms · rg 24ms · tie

$ ziggygrep -c "pub fn" . > /dev/null

ziggygrep 33ms · rg 54ms · delta -21ms

$ ziggygrep -c e . > /dev/null

ziggygrep 31ms · rg 48ms · delta -17ms

totals only, no line text alloc per hit

$ ziggygrep -l fn src/ > /dev/null

ziggygrep 15ms · rg 20ms · delta -5ms

first 64 KB read, short circuit on first hit

22/22 at or below rgbyte identical on 7/7 patternsrerun: tools/normcmp.py

Why it earns a slot

Less work per byte, not work avoided.

No index to build, no daemon to feed. Each stage removes cost that regex pipelines pay on every file.

Literal-first engineFirst byte prefilter plus eql verify. Positions that cannot start the pattern cost one memchr step, zero slicing.
Threaded file sweepFixed fetchAdd(8) batches from one file list, per worker scratch buffers, files sorted at merge. Lists of 8 or fewer files run direct.
Fused scan plus formatHeaders built once on the stack, written once into the buffer. No span array, no second pass.
Flat memoryStreaming reads, no mmap, 256 KB accum with one flush, 60 KB pipe-safe chunks. Works on pipes.

Measured bench

Every cell is ziggygrep ms vs rg ms.

200 file 53 MB Rust corpus, ReleaseFast x86_64-v3, stdout piped to null, median of 9. Baseline rg 15.2.0 --no-config -F --no-heading --line-number.

he 112051/112051 HashMap 112426/112426 e 561539/561539 in, pub fn, x, fn 600000/600000

Pattern	Full	-c	-l
he	38 vs 44	28 vs 35	17 vs 21
in	85 vs 96	40 vs 55	17 vs 20
HashMap	37 vs 40	33 vs 43	16 vs 20
pub fn	66 vs 92	33 vs 54	15 vs 23
x	66 vs 105	33 vs 51	19 vs 20
fn	79 vs 110	38 vs 55	15 vs 20
e	87 vs 102	31 vs 48	15 vs 18
miss	24 vs 24 tie	-	-

GNU grep spot check on the same box: miss 128.7ms vs 17.9ms, HashMap 46.5ms vs 25.8ms. Speedups are ratios of medians on that fixture; your disk and cache shape your numbers.

Internals

Three stages, zero wasted bytes.

Prefilter, then confirm

Vector first byte scan skips 32 bytes per step. Line bounding runs on verified hits only. Same line hits dedupe, finished lines jump past.

Batch steal, sorted merge

Atomic index hands out fixed batches. Each worker owns scratch, carry, and chunk buffers. Merge sorts by file for deterministic output.

Buffer, then one flush

Headers built on the stack, written once. Pipes cap transport at 60 KB chunks. No mmap, flat memory, pipe safe.

Quickstart

One clone, one build, first hit in a minute.

Needs Zig 0.16.0 only. No packages, no services, no accounts.

git clone https://github.com/AkashPriyadarshii/ziggygrep.git
cd ziggygrep
zig build -Doptimize=ReleaseFast
./zig-out/bin/ziggygrep HashMap .

Search one pattern
Run ziggygrep TODO src/main.zig. Matches print as path:lineno:line.
Count or list
Run ziggygrep -c HashMap . for counts, ziggygrep -l fn . for file names.
Pipe it
No paths means stdin. Exit 0 match, 1 no match, 2 error. Alias zgg if you type it daily.

Limits

What it refuses to be.

No regexPatterns are literal bytes. Use ripgrep when you need regex.
No .gitignore in v0.1Dotfiles and .git are skipped; full ignore support is planned.
No colors, context, JSONNot in v0.1. Pipe to less when you want paging.
No index, no daemonEvery query scans. That is the design.

Ecosystem

Same author, same bar.

jev-seoFree SEO and GEO audit toolkit jev-superpowersSystematic engineering gates for agents jev-curateCurated lists with measured benchmarks jev-gitSemantic git reflex gates tdlib-androidTelegram TDLib builds for Android kharchaOffline expense tracking

TOOL AUTHOR

Akash Priyadarshi

Patna, Bihar, India

Builds fast single purpose CLI tools in Zig and Rust. ziggygrep is the literal search half; archived rustygrep keeps the regex, JSON, and MCP server modes.

GitHub Portfolio LinkedIn Resume