.gitattributes

# Content that is embedded into the binary byte-verbatim (Askama templates,
# fluent locales, rust-embed text assets) must check out with LF everywhere:
# rendered output is compared against LF string literals in tests, and a CRLF
# Windows checkout fails every multi-line assertion of rendered HTML (#671).
# Binary assets (fonts, images) carry no attribute and stay untouched.
crates/ui/templates/** text eol=lf
crates/ui/assets/**/*.js text eol=lf
crates/ui/assets/**/*.css text eol=lf
crates/ui/tests/golden/** text eol=lf
crates/ui-chrome/templates/** text eol=lf
crates/hts-ui/templates/** text eol=lf
crates/ui-chrome/tests/golden/** text eol=lf
crates/hts-ui/templates/** text eol=lf
locales/** text eol=lf

# The SQL-on-FHIR conformance fixtures are kept as a byte-verbatim mirror of
# upstream (see crates/sof/tests/sql-on-fhir/README.md re-sync policy). A
# CRLF-converting Windows checkout would make `git diff` against a freshly
# re-synced upstream copy noisy with spurious line-ending churn, defeating the
# byte-verbatim comparison the re-sync workflow depends on.
crates/sof/tests/sql-on-fhir/tests/*.json text eol=lf

# Generated mongosh scripts are compared byte-for-byte against
# mongosh_script() output via include_str! in search_index_catalog.rs
# (#1059, #1084); a CRLF Windows checkout would fail that comparison.
crates/persistence/mongodb/*.mongosh.js text eol=lf

# SQL-on-FHIR SQL/bindings goldens are compared byte-for-byte against the
# generated SQL in crates/persistence/src/sof/golden_tests.rs; a CRLF Windows
# checkout would fail every case on line endings alone.
crates/persistence/tests/golden/** text eol=lf