.gitattributes

# W59 (ROADMAP-100): Windows-CI truthing — line endings are part of the
# language contract, not a checkout detail.
#
# Evidence: on windows-latest runners, git's default autocrlf converted LF to
# CRLF in the working tree, and `tests/string_literals.op` proof #1
# ("triple keeps quotes and newlines") burned because a triple-quoted
# string's verbatim content gained \r characters the SPEC says are not there.
# The differential harness compares bytes, the frozen expected/*.out files
# are byte-exact, and the oracle reads source — all three assume LF.
#
# Policy: the repository is LF everywhere, on every platform. Windows
# editors and Rust/Python toolchains all handle LF fine; CRLF creeping into
# working trees is what breaks things.
* text=auto eol=lf

# Frozen differential outputs must stay byte-identical everywhere.
*.out text eol=lf

# Red-team payloads are adversarial BYTES — rt_p7c (BOM+CRLF) and rt_p7i
# (CRLF) exist to prove the lexer tolerates hostile line endings; the
# payload IS the test. Mark the directory -text so normalization can never
# "fix" an attack into a non-attack.
tests/redteam/** -text

# GitHub Languages bar: github/linguist does not know the `.op` extension
# yet (linguist/operon.yml is the prepared upstream contribution, not
# merged), so every Operon source used to render as unrecognized and the
# language statistics showed NO Operon at all despite 3600+ `.op` files.
# linguist-language= is the documented override for exactly this state.
*.op linguist-language=Operon

# Machine-generated artifacts are data, not authorship — keep them out of
# the language bar (22 CodeQL SARIF reports would otherwise count as JSON;
# the fuzz crash corpus is machine-written crash reproduction input).
docs/codeql/*.sarif linguist-generated
fuzz_corpus/** linguist-generated