.gitattributes

# Test inputs the parsers read verbatim. The tests build and match marker lines
# with "\n", so a CRLF checkout changes what they assert:
# `create_host_prefers_kvm_and_q35` replaces "kvm\n<domainCapabilities>" and,
# with CRLF, replaces nothing — the parser then sees the KVM the test expects
# to have been simulated away — and
# `vnc_console_without_a_password_or_refused_one` looks up "{marker}\n", finds
# no match, and unwraps the None. Pin the endings; the files hold no CR today,
# so nothing in history is rewritten.
#
# The binaries among them (`*.hive`, `*.bin`) carry CR bytes that belong to the
# data the storage-migration tests restore, and those fixtures cannot be
# regenerated. `text` would normalise them at checkout, so each extension is
# excluded — a later line wins, and the exclusions below come after. The
# rules that follow this block (`rootfs_manifest/*.sig`) are narrower still
# and override it for their own files.
crates/sbm_parser/tests/fixtures/** text eol=lf
test/fixtures/** text eol=lf
test/fixtures/**/*.hive -text
test/fixtures/**/*.bin -text

# sqlx hashes each migration file's bytes and records the checksum in the
# agent's `_sqlx_migrations` table. `migrate!` embeds those bytes at compile
# time, so a checkout that converted line endings produces a binary whose
# checksums disagree with the ones an earlier build already recorded, and the
# migrator refuses to start rather than run against a database it cannot
# account for.
#
# Windows checkouts are where this bites: `Rust tests (windows-latest)` failed
# `shipped_migrations_keep_their_checksums` because CRLF changed migration 1's
# SHA-384 from ac7a765b… to 1ca2d91b…. Pin the bytes so every platform hashes
# the same file.
monitor/migrations/*.sql text eol=lf

# Ed25519 verifies the original bytes, including line endings.
test/fixtures/rootfs_manifest/signed.json text eol=lf
test/fixtures/rootfs_manifest/*.sig -text

# No Git LFS: every tracked image and theme package is a regular blob. An LFS
# rule over them makes each file whose mtime changes (a build, an IDE) show as
# modified, and `git add` would turn it into a pointer. The theme store also
# cannot hold pointers: `git archive HEAD:store` (scripts/store-tarball.sh)
# reads no `.gitattributes` above the subtree, so no smudge runs. Moving to LFS
# means `git add --renormalize` on the files in the same change.