Repository navigation
[Python] .slice methods should accept Arrow scalar types #51145
Description
Describe the enhancement requested
The .slice methods take two integers as arguments, but cannot take arrow integer types.
import pyarrow as pa
a = pa.array([0, 1, 2, 3, 4, 5])
a.slice(0, 4)
# <pyarrow.lib.Int64Array object at 0x798d8fec2ce0>
# [
# 1,
# 2,
# 3,
# 4
# ]
a.slice(a[0], a[4])
# ---------------------------------------------------------------------------
# TypeError Traceback (most recent call last)
# Cell In[12], line 1
# ----> 1 a.slice(a[0], a[4])
#
# File pyarrow/array.pxi:1639, in pyarrow.lib.Array.slice()
# -> 1639 'Could not get source, probably due dynamically evaluated source code.'
#
# TypeError: '<' not supported between instances of 'pyarrow.lib.Int64Scalar' and 'int'The same is true for tables:
t = pa.Table.from_pydict({"a": a})
t.slice(a[0], a[4])
# ---------------------------------------------------------------------------
# TypeError Traceback (most recent call last)
# Cell In[13], line 1
# ----> 1 t.slice(a[0], a[4])
# File pyarrow/table.pxi:4299, in pyarrow.lib.Table.slice()
# -> 4299 'Could not get source, probably due dynamically evaluated source code.'
# TypeError: '<' not supported between instances of 'pyarrow.lib.Int64Scalar' and 'int'This is not generally true of other methods. For example, .take is fine with getting an arrow array. I can even do a[slice(a[0], a[0] + a[4])], where that slice contains pyarrow scalars.
I think it is kinda silly that if I have offsets for an arrow table in another arrow table I am having to convert those offsets to python integers (e.g. t.slice(a[0].as_py(), a[4].as_py())) or converting to indexing with a slice object.
Component(s)
Python
Activity
[-][python] `.slice` methods should accept Arrow scalar types[/-][+][Python][python] `.slice` methods should accept Arrow scalar types[/+][-][Python][python] `.slice` methods should accept Arrow scalar types[/-][+][Python] `.slice` methods should accept Arrow scalar types[/+]YusefSyed commentedon Sep 4, 2026
take
AlenkaF commentedon Sep 10, 2026
Thanks for the issue @ivirshup
The difference between
.takeand.sliceis that they take two different code paths..takegoes through the compute module code path, where_pack_compute_argsdoes the type mapping, then C++ compute functions are all called witharrow::Datumarguments. Whereas the.slicecode path goes to the C++Array::Slicewhich expectsint64_t offsetandint64_t length.I agree,
.slicebehavior could be improved on the Python side. Will look at the PR now.apacheGH-51145: [Python] Document Arrow scalar slice parameters
GH-51145: [Python] Accept Arrow integer scalars in slice methods (#51160
AlenkaF commentedon Sep 25, 2026
Issue resolved by pull request 51160
#51160