[Python] pa.Array.from_pandas converts empty timestamp[s][pyarrow, UTC] pandas Series to ChunkedArray, not TimestampArray

[Wainberg]

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

>>> pa.Array.from_pandas(pd.Series([], dtype=pd.ArrowDtype(pa.timestamp('s'))))  # correct
<pyarrow.lib.TimestampArray object at 0x7fb665f77fa0>
[]
>>> pa.Array.from_pandas(pd.Series([], dtype=pd.ArrowDtype(pa.timestamp('s'))).dt.tz_localize('UTC'))  # incorrect
<pyarrow.lib.ChunkedArray object at 0x7fb665fd8680>
[

]

Same issue with pa.array() instead of pa.Array.from_pandas().

Component(s)

Python

[jorisvandenbossche]

@Wainberg this is actually controlled by pandas itself, so can you report this to https://github.com/pandas-dev/pandas/issues/ ?

[mroeschke]

Ping-ponging back here: Looks like assume_timezone on empty pa.ChunkedArray[pa.TImestampArray] drops the empty pa.TimestampArray

In [3]: import pyarrow as pa

In [4]: arr = pa.chunked_array([pa.array([], type=pa.timestamp("ns"))])

In [5]: arr
Out[5]: 
<pyarrow.lib.ChunkedArray object at 0x10791b600>
[
  []
]

In [6]: import pyarrow.compute as pc

In [7]: pc.assume_timezone(arr, "UTC")
Out[7]: 
<pyarrow.lib.ChunkedArray object at 0x10789bc40>
[

]

In [8]: pa.__version__
Out[8]: '15.0.1'

In [9]: pc.assume_timezone(arr, "UTC").chunks
Out[9]: []

In [10]: arr.chunks
Out[10]: 
[<pyarrow.lib.TimestampArray object at 0x10797ee00>
 []]

[jorisvandenbossche]

To be honest I would also expect it to preserve the chunking structure, but it seems this something we don't do in general (I tried several other scalar kernels on a chunked array with one empty chunk, and they all return a chunked array with no chunks), so that is not specific to assume_timezone

[jorisvandenbossche]

But indeed it's because pandas returns a ChunkedArray with 0 chunks to pyarrow, that we don't unpack that to the single chunk Array, because of:

arrow/python/pyarrow/array.pxi

Lines 116 to 118 in 54ff758

	if isinstance(res, ChunkedArray) and res.num_chunks==1:
	res = res.chunk(0)
	return res

in the code handling objects with an __arrow_array__ method.