test-corpus

Run python3 test_documents/scripts/fetch_corpus.py to download everything, or pass --include 'pdf/**' to narrow the fetch. Files come from a public GCS bucket over anonymous HTTPS with no credentials, and fetched files are gitignored inside the submodule.

The test_documents submodule's binary fixtures are not committed to git, so tests reading from it fail until you run the fetch script. These are missing-file errors, not assertion failures, so read the message before concluding the suite regressed.

No. Baking corpus bytes in at compile time turns a missing fixture into a build failure for the whole crate. Use the read_test_fixture helper instead, which prints a greppable SKIP line and returns None when the fixture is unavailable.

git worktree add does not populate submodules, so the control worktree has an empty test_documents and every fixture-guarded test skips silently. Symlink the main tree's test_documents into the worktree and use a dedicated CARGO_TARGET_DIR.

From inside the submodule, run publish_corpus.py with --dry-run first, then publish and verify against the bucket. Push the submodule commit before the superproject gitlink, or CI fails at checkout with a missing submodule SHA.

The fixtures/ directory at the repo root is committed to git and safe to use with include_bytes!. Only test_documents/ is bucket-fetched and uncommitted, so it requires the fetch script and the read_test_fixture guard.