菜单
Table hook registration and materialization/retrieval for different data frameworks. Use when working on register_table, PandasTableHook, PolarsTableHook, SQLAlchemyTableHook, PyArrow, or View resolution.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Rules for working on ParquetTableStore and its metadata synchronization layer. Use when touching duckdb_parquet.py, sync_views, metadata_store, schema_prefix, or S3/GCS/MinIO integration.
Transaction model using __odd/__even schema slots. Use when working on init_stage, commit_stage, Schema, Stage, or transaction handling.
Test instance configuration and pytest flag conventions. Use when running tests, adding new test instances, or debugging CI failures involving with_instances, skip_instances, conftest, or INSTANCE_MARKS.
| name | table-hooks |
| description | Table hook registration and materialization/retrieval for different data frameworks. Use when working on register_table, PandasTableHook, PolarsTableHook, SQLAlchemyTableHook, PyArrow, or View resolution. |
Each data framework (pandas, polars, SQLAlchemy, ibis, pydiverse.transform) has a table hook registered on the store via @Store.register_table(). Hooks implement materialize() and retrieve().
The ParquetTableStore overrides hooks to write/read parquet files instead of SQL tables. Key hooks:
PandasTableHook — uses PyArrow for parquet I/OPolarsTableHook — uses native polars parquet I/O with PyArrow fallback (GCS credential issues)SQLAlchemyTableHook — uses DuckDB COPY ... TO for parquet export_ParquetPyArrowMixin — shared PyArrow-based reading logic for view resolutionViews (View objects) support source unions, column renaming, sorting, and limits. The _resolve_view_paths() method resolves a table's view chain to actual parquet file paths for reading.