Benchmarks

Accuracy and speed checks live under benchmarks/. For a human-readable speed table (bytesense vs chardet, representative µs means) and how to interpret workload-dependent cases (e.g. large UTF-8 buffers), see the Speed (pytest-benchmark) section in the project README on GitHub (MkDocs does not ship the repo root README.md, so this link targets the repository file).

Run accuracy gate (must pass):

pytest benchmarks/test_bench_detection.py -k accuracy -v

Run pytest-benchmark comparisons:

# Do not use -k bench_ — it matches the module name test_bench_detection.py and runs every test.
pytest benchmarks/test_bench_detection.py \
  -k "test_bench_bytesense_fast_path or test_bench_cn_fast_path or test_bench_bytesense_full or test_bench_cn_full or test_bench_chardet_full" \
  --benchmark-sort=mean -v \
  --benchmark-json=benchmarks/results/output.json

Results JSON is gitignored under benchmarks/results/.