Benchmarks
These benchmarks compare hk, lefthook, pre-commit, and prek running 10 linters on a synthetic project. Since lefthook's parallel: true mode has race conditions when linters touch overlapping files, we run lefthook in safe (sequential) mode—the only correct option. pre-commit and prek both run hooks sequentially.
hk is the only tool that runs linters in parallel and safely.
See Why hk? for context on why these differences exist.
Setup
A synthetic project with ~6,000 files across multiple languages:
- 4000 Python, 500 JavaScript/TypeScript, 500 JSON, 500 Shell, 250 YAML, 200 CSS, 200 Markdown
Ten linters with overlapping file coverage:
| Linter | Files | How hk avoids write locks |
|---|---|---|
| prettier | *.{js,ts,css,md} | check_list_files — only locks files that need fixing |
| eslint | *.{js,ts} | Falls back to write lock (eslint has no diff/list mode) |
| black | *.py | check_diff — hk applies the diff itself |
| ruff check | *.py | Check only — read lock |
| ruff format | *.py | check_diff — hk applies the diff itself |
| jq | *.json | check_diff |
| yq | *.{yml,yaml} | check_diff |
| shfmt | *.{sh,bash} | check_diff |
| trailing-whitespace | * (all files) | check_diff via hk util (built-in Rust) |
| newlines | * (all files) | check_diff via hk util (built-in Rust) |
The trailing-whitespace and newlines linters use **/* globs—they overlap with every other linter. pre-commit and prek always run hooks sequentially. lefthook supports parallel: true but has no file-level coordination, so overlapping linters cause race conditions—we run it in safe (sequential) mode. hk is the only tool that runs everything in parallel safely, using file-level read/write locks.
Results
Reproducing
Everything is in the benchmark/ directory.
Prerequisites
mise use hyperfine lefthook prettier eslint shfmt jq yq
uv tool install pre-commit prek black ruffGenerate and run
# Generate a synthetic project (~700 files)
benchmark/generate-project.sh /tmp/hk-bench
# Run benchmarks
benchmark/run.sh /tmp/hk-bench
# Customize
NUM_JS=500 NUM_PY=500 benchmark/generate-project.sh /tmp/hk-bench
RUNS=20 WARMUP=3 benchmark/run.sh /tmp/hk-benchResults are saved as JSON in benchmark/results/ and both a chart and data file are generated in docs/public/.