sparkrules-native (Rust Tier-1 scorer)

PyO3 extension implementing the Tier-1 scalar interpreter described in docs/CHOOSING_A_BACKEND.md. It reads rule AST JSON emitted by SparkRules Python (RulePack.to_native_json()) — the DRL lexer/parser stays on the Python side.

Build (local)

cd sparkrules_native
python -m venv .venv
.venv\\Scripts\\activate   # Windows
pip install maturin
maturin develop --release
python -c "import sparkrules_native as m; print(m.native_version())"

Parity

All scores must match LocalRuleExecutor.score() for the same fact dict / JSON-able structure. Run:

pytest tests/integration/test_native_parity.py -q

Consumers (no PyPI yet)

Until https://pypi.org/project/sparkrules-native/ resolves, use:

• native-wheels.yml artifacts (.whl per OS/Python), or
• maturin develop --release / maturin build --release locally.

See docs/NATIVE_TIER1.md (PyPI verification commands, Glue --extra-py-files, artifact names).

Publish to PyPI (maintainers)

The crate is CI-built (.github/workflows/native-wheels.yml) but not automatically published. publish-sparkrules-native.yml has no effect until the secret exists and the workflow run finishes green; then confirm the project on PyPI.

1. Create a PYPI_API_TOKEN PyPI token secret on the GitHub repo (maturin reads MATURIN_PYPI_TOKEN; the workflow wires secrets.PYPI_API_TOKEN into it).
2. Run publish-sparkrules-native.yml via Actions → workflow_dispatch.
3. Restore sparkrules[native] in root pyproject.toml (native = ["sparkrules-native>=0.1.0,<1.0"]) once pip install sparkrules-native works from PyPI.

Tier 2+

src/eval_columnar.rs is reserved for an Arrow batch path; not implemented in Tier-1.