# Line endings are normalised in the repository and on checkout. This project
# hashes its own installed sources to prove a wheel matches a baseline, and a
# CRLF checkout once made that hash disagree with a clean CI checkout of
# identical code -- sixteen jobs red over nothing. The hasher normalises now,
# but the file it reads should not have varied in the first place.
* text=auto eol=lf

*.whl binary
*.png binary
*.svg text

# Vendored, not authored. lme4's published fixtures, fetched verbatim from
# Rdatasets and kept byte for byte so the oracle tests compare against the
# same bytes lme4 published -- see data/README.md for the retrieval recipe.
# They are GPL-2 while this project is MIT, which is why the sdist excludes
# them; they should not be counted as this project's code either.
data/** linguist-vendored

# Vendored notice text from the crates the extension statically links.
THIRD-PARTY-LICENSES.md linguist-vendored

# Generated, not authored. Written by bench/make_charts.py and the recorders;
# editing any of them by hand decouples the artefact from the run it depicts.
docs/assets/*.svg linguist-generated
bench/baseline.json linguist-generated
bench/performance.json linguist-generated
bench/stress_sweep_results.json linguist-generated
