feature-sql-tool 0.1.0

Tooling for SQL feature lineage extraction and unified vector SQL generation

feature-sql-tool

feature-sql-tool is a draft Python package for two related tasks:

1. analyzing SQL scripts for model features and extracting dependency lineage;
2. building a unified SQL query for a model input vector from multiple feature SQL files.

The package is designed around sqlglot and uses a src layout with modern pyproject.toml packaging, which is the recommended approach in the Python Packaging User Guide. The packaging guide recommends defining build metadata in pyproject.toml, and the tool recommendations guide recommends building distributions with python -m build rather than calling setup.py directly. citeturn721430search0turn721430search2turn721430search6turn721430search7turn721430search9

Features in this first version

• FeatureSpec points to a .sql file instead of storing long SQL inline.
• SQL loading, parsing, scope registration, lineage extraction, graph classification, and unified SQL generation are split into separate modules.
• A first MVP service API is included.

Install locally

python -m pip install -U pip
pip install -e .

For development extras:

pip install -e .[dev]

Build distributions

python -m pip install -U build
python -m build

This creates:

• dist/*.tar.gz — source distribution
• dist/*.whl — wheel

Check distributions

python -m pip install -U twine
twine check dist/*

Upload to PyPI

python -m pip install -U twine
twine upload dist/*

Install from a downloaded archive

After publishing to PyPI:

pip download feature-sql-tool
pip install feature_sql_tool-0.1.0-py3-none-any.whl

Minimal usage example

from pathlib import Path

from feature_sql_tool import FeatureSqlTool, FeatureSpec

features = [
    FeatureSpec(
        feature_name="avg_payment_30d",
        sql_file_path=Path("sql/avg_payment_30d.sql"),
        final_alias="avg_payment_30d",
        entity_key="client_id",
        dialect="spark",
    ),
]

tool = FeatureSqlTool()
results = tool.analyze_features(features)
print(results[0].source_columns)

Notes

This package is still an MVP scaffold. Deep recursive CTE resolution, UNION-aware lineage, and real common-subgraph optimization are not fully implemented yet.