[Python ] Table.to_batches() loses schema information when table has zero rows

[alexowens90]

Describe the bug, including details regarding any error messages, version, and platform.

Calling to_batches on a zero-row table returns an empty list, losing the schema information:

Python 3.13.5 (main, Jun 12 2025, 08:56:14) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import pyarrow as pa
>>> pa.__version__
'23.0.1'
>>> table = pa.table({"col": pa.array([], pa.int64())})
>>> table.to_batches()
[]

I would expect the result to be a single-element list, containing a RecordBatch with zero rows, and the same schema as the original table.

Component(s)

Python

[pitrou]

Hmm, is that an actual problem? You can get the schema from table.schema, I'm not sure producing a zero-length RecordBatch is useful here.

[pitrou]

Or this could be an optional argument to to_batches. @jorisvandenbossche What @raulcd do you think?

[alexowens90]

Yes, I have a workaround to use table.schema like so:

        if table.num_rows == 0:
            pa_record_batches = [
                pa.RecordBatch.from_arrays(
                    [chunked_array.chunk(0) for chunked_array in table.itercolumns()], schema=table.schema
                )
            ]
        else:
            pa_record_batches = table.to_batches()

I guess the unexepcted behaviour from my perspective was that the following could does not work:

>>> import pyarrow as pa
>>> table = pa.table({"col": pa.array([], pa.int64())})
>>> new_table = pa.Table.from_batches(table.to_batches())
Traceback (most recent call last):
  File "<python-input-4>", line 1, in <module>
    new_table = pa.Table.from_batches(batches)
  File "pyarrow/table.pxi", line 5032, in pyarrow.lib.Table.from_batches
ValueError: Must pass schema, or at least one RecordBatch