Repository navigation
pa.Table.from_struct_array fails for an empty array #48344
Description
Describe the bug, including details regarding any error messages, version, and platform.
This is the inverse of #46355, so I'll continue from there 😅
Repro
import pyarrow as pa
table = pa.table({"a": pa.array([]), "b": pa.array([], type=pa.float64())})
array = table.to_struct_array()
print(array)
print(array.type.fields)
pa.Table.from_struct_array(array)Output
<pyarrow.lib.ChunkedArray object at 0x...>
[
]
[pyarrow.Field<a: null>, pyarrow.Field<b: double>]
./pyarrow/table.pxi:4945, in pyarrow.lib.Table.from_struct_array()
./pyarrow/table.pxi:5032, in pyarrow.lib.Table.from_batches()
ValueError: Must pass schema, or at least one RecordBatchWorkaround
I'm doing quite a lot of version branching here already:
- https://github.com/narwhals-dev/narwhals/blob/e4f0f83a15808349f2f4970e5549b791671fe601/narwhals/_plan/arrow/series.py#L321-L330
- https://github.com/narwhals-dev/narwhals/blob/e4f0f83a15808349f2f4970e5549b791671fe601/narwhals/_plan/arrow/functions.py#L337-L334
In short, it looks like:
import pyarrow as pa
table = pa.table({"a": pa.array([]), "b": pa.array([], type=pa.float64())})
array = table.to_struct_array()
from_struct_array = pa.schema(array.type.fields).empty_table()
print(from_struct_array)pyarrow.Table
a: null
b: double
----
a: [0 nulls]
b: [[]]Component(s)
Python
Activity
apacheGH-48344: [Python] Fix Table.from_struct_array for empty Chunke…
apacheGH-48344: [Python] Fix Table.from_struct_array for empty Chunke…
AlenkaF commentedon Jun 22, 2026
@1fanwang can you add a comment under this issue, so I can assign it to you?
1fanwang commentedon Jun 22, 2026
Thanks @AlenkaF
PR opened #49869 to address this issue.
apacheGH-48344: Preserve TypeError for invalid empty ChunkedArray
apacheGH-48344: [Python] Fix Table.from_struct_array for empty Chunke…
apacheGH-48344: Preserve TypeError for invalid empty ChunkedArray
GH-48344: [Python] Fix Table.from_struct_array for empty ChunkedArray (…
AlenkaF commentedon Aug 28, 2026
Issue resolved by pull request 49869
#49869