bulkcopy_arrow rejects polars DataFrames: Utf8View string columns unsupported #708

Describe the bug

The bulkcopy_arrow docstring says any object exposing arrow_c_stream works as a source, and names polars DataFrames as an example. In practice any polars ≥ 1.0 DataFrame with a string column fails.

Polars stores strings natively in Arrow's string-view layout, so its zero-copy arrow_c_stream export produces Utf8View columns, which the writer doesn't support. Frames without string columns work fine.

Workaround: cursor.bulkcopy_arrow("dbo.T", df.to_arrow()) — polars re-encodes strings to large_string for compatibility. That costs an extra copy of all string data, which is the overhead this API otherwise avoids.

Worth supporting Utf8View (and presumably BinaryView) in the writer — pandas is heading the same way with Arrow-backed strings, so view types will increasingly be what arrow_c_stream hands you. Failing that, the docstring shouldn't cite polars as a working example.

Exception message:

ValueError: Cannot map Arrow column 'name' (Utf8View) to SQL column 'name' (VarChar):
Usage Error: type combination is not supported by the Arrow row-major writer

To reproduce

import polars as pl

df = pl.DataFrame({"id": pl.Series([1, 2], dtype=pl.Int32), "name": ["a", "b"]})
cursor.bulkcopy_arrow("dbo.T", df)  # T: (id INT, name VARCHAR(50))

Expected behavior

That the above code just works.

Further technical details

mssql-python 1.13.0, polars 1.39.3, pyarrow 23.0.1, Linux x86_64.