Search text and nested structures with readable patterns, typed captures and JSON output. PEGREP chooses an execution strategy for each pattern while preserving its matching semantics.

# Find matching lines across a source tree.
pegrep -F TODO -n -g '*.rs' crates

# Extract only server-error status codes.
printf '200 404 500 503\n' | pegrep 'status:int where status >= 500' -o
# 500
# 503

# Capture a call's arguments, including nested calls and quoted parentheses.
printf '%s' 'call(a, f("x)"))' |
  pegrep '"call" args:balanced("(", ")", lex="strings")' --json

Download an archive from GitHub Releases, extract it, and put pegrep (pegrep.exe on Windows) on your PATH. Each release includes SHA256SUMS.

Or build from source with Rust 1.97.1 and a C toolchain:

git clone https://github.com/copyleftdev/pegrep.git
cd pegrep
cargo install --locked --path crates/pegrep-cli
pegrep --help

Build from the complete workspace: it includes the vendored Unicode and traversal patches. See installation and checksum verification.

• Literal search with -F, regular expressions with --regex, and composable PEG patterns.
• Typed captures, comparisons, JSON strings, balanced delimiters and reusable nonrecursive patterns.
• Ignore-aware directory traversal, globs, context lines, counts, file lists and bounded parallel workers.
• JSON Lines with byte spans and typed values; versioned pattern IR and explainable execution plans.
• Explicit limits for memory, work, nesting and output; limit exhaustion reports an error.

pegrep --regex 'ERROR[0-9]+' -o application.log
printf 'answer=42\n' | pegrep 'key:identifier "=" value:int' --json
pegrep --explain '"ERROR" ws code:int'
pegrep -F TODO --count-matches -j 4 crates

PEG choice commits to the first successful alternative, and repetition does not backtrack. For example, ("a" | "ab") "c" does not match abc. Typed filters run after that choice commits.

Input defaults to strict UTF-8 with binary detection. --bytes accepts arbitrary bytes. JSON preserves invalid bytes using base64. Cross-file output order is unspecified unless you select --sort path or serial execution.

Exit codes: 0 matches, 1 no matches, 2 error, 130 interruption. Errors can follow partial output, so check the exit status. Resource reservations are not a total-process RSS ceiling, and cancellation is cooperative.

The v0.1 MVP includes balanced nesting, but not recursive user grammars, rewriting or language AST adapters. Performance depends on the pattern and workload; there is no universal speed advantage over ripgrep.

User guide · Pattern language · JSON/output contract · Examples · Release process

python3 -m pip install -r requirements-design.txt
bash scripts/gates.sh design
bash scripts/gates.sh dev

CI checks formatting, lint, default/all-feature tests, documentation, pattern/output fixtures and native Linux/macOS/Windows behavior. Tagged releases run those checks before publishing native archives and checksums. The full release gate also checks fuzzing, equivalence, controlled performance and isolated installation.

MIT OR Apache-2.0, at your option. Vendored dependencies retain their own notices.