greg

Parallel walk-and-search architecture. PCRE2 JIT. High-performance design.

---

Performance Comparison

Below is a benchmark comparison demonstrating the execution time of greg alongside other popular code search tools. The benchmark measures search execution time for the pattern "ls" across a home directory structure (~1,141 files) under Linux / ARM64.

All stdout output was redirected to /dev/null to measure engine processing latency rather than terminal rendering speeds.

Tool	Average Execution Time (ms)	Relative Performance
ugrep	53 ms	1.00x (Baseline)
greg	54 ms	1.02x
ripgrep (rg)	63 ms	1.19x
The Silver Searcher (ag)	662 ms	12.49x

Average execution times show that greg achieves processing speeds comparable to standard high-performance search utilities.

---

Features

• Parallel walk-and-search — worker threads handle both directory traversal and file searching simultaneously (work-stealing design)
• PCRE2 JIT — full regex with JIT compilation, same engine as ripgrep
• Smart-case — case-insensitive when pattern is all lowercase (like ripgrep)
• Slab-allocated queue — zero per-file malloc/free overhead in the hot path
• gitignore support — respects .gitignore and .ignore files with a lock-free, reference-counted ignore tree supporting standard * and multi-level ** wildcards matched iteratively
• mmap file reading — memory-mapped I/O with MADV_SEQUENTIAL for large files
• Buffered output — 1 MiB stdout buffer, single fwrite per file result under mutex, printing no trailing newlines/blank space at the end of the results
• JIT compilation — uses the compiled JIT execution path (pcre2_jit_match) for maximum throughput

---

Installation

git clone <repo>cd greg
make
# Binary is ./greg

Dependencies:libpcre2-8 (dev package), a C99 compiler, pthreads.

# Debian/Ubuntu/Termux
apt install libpcre2-dev
# macOS
brew install pcre2

---

Usage

greg [options] <pattern> [path]

Options

Flag	Long form	Description
-i	--ignore-case	Search case-insensitively
-S	--smart-case	Case-insensitive if pattern is all lowercase (default)
-v	--invert-match	Select non-matching lines
-F	--fixed-strings	Treat pattern as a literal string
-n	--line-number	Always show line numbers
-N	--no-line-number	Never show line numbers
-l	--files-with-matches	Print only filenames containing matches
-c	--count	Print match count per file
-m NUM	--max-count NUM	Stop after NUM matches per file
-j NUM	--threads NUM	Number of worker threads (default: CPU cores)
-a	--text	Search binary files
	--raw	Machine-readable output format (<filepath>\n<line_num>:<match_start>:<match_end>\n using 1-based line numbers)
	--no-ignore	Ignore .gitignore / .ignore rules
	--follow	Follow symbolic links
	--hidden	Search hidden files and directories
	--heading	Group results by filename (default on TTY)
	--no-heading	Disable grouped output
	--color always|never	Control ANSI color output
-h	--help	Print help
	--version	Print version

Examples

# Search for "TODO" in current directory
greg TODO
# Case-insensitive search
greg -i "error" /var/log
# Only filenames
greg -l "main" src/
# Fixed string (no regex)
greg -F "fn main()".# Limit to 5 matches per file
greg -m 5 "panic".# Search hidden files too
greg --hidden "secret"~

---

Architecture

main thread
└─ pushes root path as WORK_DIR onto shared queue
worker threads (N = CPU cores)
├─ pop WORK_DIR → opendir/readdir → push child files (WORK_FILE) + subdirs (WORK_DIR)
└─ pop WORK_FILE → mmap → PCRE2 JIT match → buffered output

Key design decisions:

• Slab allocator for queue nodes — 256 nodes per slab, recycled via free-list; no per-item malloc
• Reference-counted ignore tree — parent ignore contexts shared across threads with atomic refcount, no copies
• Iterative Wildcard Matcher — safe backtrack-based iterative glob wildcard matching (supporting * and **), preventing stack overflows on complex nested directory patterns
• Task tracking in queue — automatic termination when active_tasks reaches zero; no explicit "deactivate" signal needed for the normal path

---

License

MIT