ak.to_cudf fails with TypeError on cuDF 24.12 due to internal API changes #3783
Description
Version of Awkward Array
2.8.11
Description and code to reproduce
Description:
Calling ak.to_cudf on jagged or nested arrays fails when using cudf>=24.12.00. The internal cuDF column constructors (NumericalColumn and StructColumn) have changed their signatures, leading to TypeError when awkward attempts to pass the data keyword argument.
Reproducer:
import awkward as ak
import cudf
# 1. Jagged array failure (NumericalColumn)
arr_jagged = ak.Array([[[1, 2, 3], [], [3, 4]], []])
out_jagged = ak.to_cudf(arr_jagged)
# 2. Nested record failure (StructColumn)
arr_nested = ak.Array([{"a": {"b": 1}}, {"a": {"b": 2}}])
out_nested = ak.to_cudf(arr_nested)Error Traceback:
For Jagged Arrays:
E TypeError: NumericalColumn.__init__() got an unexpected keyword argument 'data'
...
/awkward/contents/listoffsetarray.py:2015: in _to_cudf
> ind_buf = cudf.core.column.numerical.NumericalColumn(
data=buf, dtype=index.dtype, mask=None, size=len(index)
)
For Nested Records:
E TypeError: StructColumn.__init__() got an unexpected keyword argument 'data'
...
/awkward/contents/recordarray.py:1150: in _to_cudf
> return cudf.core.column.struct.StructColumn(
data=None,
children=children,
dtype=dt,
mask=m,
size=length,
offset=0,
)