# Arrow connector `ArrowConnector` (`lawsynth_connectors.arrow`) ingests Apache Arrow data into the common batch/provenance pipeline. PyArrow is optional: importing the core package never imports it, and using this connector without `pyarrow` raises `DependencyUnavailableError` naming the `arrow` extra (`pip install lawsynth-connectors[arrow]`). ## Sources `table_from_arrow` accepts: - a `pyarrow.Table` (used as-is); - a `pyarrow.RecordBatch` (wrapped into a one-batch table); - a file path to an Arrow IPC file, tried as the random-access **file** format and falling back to the **stream** format, read via `memory_map`; - any object exposing `to_table(columns=...)` (e.g. a `pyarrow.dataset.Dataset`). Duplicate field names are rejected with `DataValidationError`. A projection that names columns not present in the table is also rejected. ## Reading ```python from lawsynth_connectors import ConnectorConfig, ReadRequest from lawsynth_connectors.arrow import ArrowConnector connector = ArrowConnector(ConnectorConfig(name="arrow", options={"data": table})) with connector: rows = connector.read_all(ReadRequest("in-memory", columns=["t", "x"])) ``` The source is resolved from the request options `data`, then the connector's configured `data`, then a `datasets` mapping keyed by resource name. Rows are produced with `records_from_arrow`, which honors `columns`, `offset`, and `limit` and yields plain `dict` records batch-by-batch through `to_pylist()`, so the result flows into the standard bounded `DataBatch` pipeline with fingerprints. ## Capabilities `read`, `snapshots`, and `projections`. The connector reads Arrow into records; it does not write Arrow and performs no LawSynth domain inference.