Remove cudf._lib.scalar in favor of pylibcudf (#17701)

This PR changes `cudf.Scalar.device_scalar` to be a `pylibcudf.Scalar` object instead of a `cudf._lib.scalar.DeviceScalar`.
 
Most of the conversion logic previously in `cudf._lib.scalar.DeviceScalar` now lives in `python/cudf/cudf/core/scalar.py`

Some tests that exercised behaviors of `cudf.Scalar.device_scalar` when it was a `cudf._lib.scalar.DeviceScalar` were modified/removed.

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)

Approvers:
  - Vyas Ramasubramani (https://github.com/vyasr)

URL: #17701

python/cudf/cudf/_lib/CMakeLists.txt

@@ -12,7 +12,7 @@
 # the License.
 # =============================================================================
 
-set(cython_sources column.pyx scalar.pyx strings_udf.pyx)
+set(cython_sources column.pyx strings_udf.pyx)
 set(linked_libraries cudf::cudf)
 
 rapids_cython_create_modules(

python/cudf/cudf/_lib/column.pyx

@@ -33,7 +33,11 @@ from libcpp.vector cimport vector
 
 from rmm.pylibrmm.device_buffer cimport DeviceBuffer
 
-from pylibcudf cimport DataType as plc_DataType, Column as plc_Column
+from pylibcudf cimport (
+    DataType as plc_DataType,
+    Column as plc_Column,
+    Scalar as plc_Scalar,
+)
 cimport pylibcudf.libcudf.copying as cpp_copying
 cimport pylibcudf.libcudf.types as libcudf_types
 cimport pylibcudf.libcudf.unary as libcudf_unary
@@ -45,8 +49,6 @@ from pylibcudf.libcudf.column.column_view cimport column_view
 from pylibcudf.libcudf.lists.lists_column_view cimport lists_column_view
 from pylibcudf.libcudf.scalar.scalar cimport scalar
 
-from cudf._lib.scalar cimport DeviceScalar
-
 
 cdef get_element(column_view col_view, size_type index):
 
@@ -55,10 +57,8 @@ cdef get_element(column_view col_view, size_type index):
         c_output = move(
             cpp_copying.get_element(col_view, index)
         )
-
-    return DeviceScalar.from_unique_ptr(
-        move(c_output), dtype=dtype_from_column_view(col_view)
-    )
+    plc_scalar = plc_Scalar.from_libcudf(move(c_output))
+    return pylibcudf.interop.to_arrow(plc_scalar).as_py()
 
 
 def dtype_from_pylibcudf_column(plc_Column col not None):
@@ -767,7 +767,7 @@ cdef class Column:
                     base_nbytes = 0
                 else:
                     chars_size = get_element(
-                        offset_child_column, offset_child_column.size()-1).value
+                        offset_child_column, offset_child_column.size()-1)
                     base_nbytes = chars_size
 
         if data_ptr:
@@ -908,6 +908,6 @@ cdef class Column:
     def from_scalar(py_val, size_type size):
         return Column.from_pylibcudf(
             pylibcudf.Column.from_scalar(
-                py_val.device_value.c_value, size
+                py_val.device_value, size
             )
         )

python/cudf/cudf/api/types.py

@@ -16,6 +16,8 @@
 import pyarrow as pa
 from pandas.api import types as pd_types
 
+import pylibcudf as plc
+
 import cudf
 from cudf.core._compat import PANDAS_LT_300
 from cudf.core.dtypes import (  # noqa: F401
@@ -143,8 +145,8 @@ def is_scalar(val):
         val,
         (
             cudf.Scalar,
-            cudf._lib.scalar.DeviceScalar,
             cudf.core.tools.datetimes.DateOffset,
+            plc.Scalar,
             pa.Scalar,
         ),
     ) or (

python/cudf/cudf/core/_internals/binaryop.py

@@ -50,10 +50,10 @@ def binaryop(
         plc.binaryop.binary_operation(
             lhs.to_pylibcudf(mode="read")
             if isinstance(lhs, Column)
-            else lhs.device_value.c_value,
+            else lhs.device_value,
             rhs.to_pylibcudf(mode="read")
             if isinstance(rhs, Column)
-            else rhs.device_value.c_value,
+            else rhs.device_value,
             plc.binaryop.BinaryOperator[op],
             dtype_to_pylibcudf_type(dtype),
         )

python/cudf/cudf/core/_internals/copying.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2020-2024, NVIDIA CORPORATION.
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
 from __future__ import annotations
 
 from typing import TYPE_CHECKING
@@ -67,7 +67,7 @@ def scatter(
     plc_tbl = plc.copying.scatter(
         plc.Table([col.to_pylibcudf(mode="read") for col in sources])  # type: ignore[union-attr]
         if isinstance(sources[0], cudf._lib.column.Column)
-        else [slr.device_value.c_value for slr in sources],  # type: ignore[union-attr]
+        else [slr.device_value for slr in sources],  # type: ignore[union-attr]
         scatter_map.to_pylibcudf(mode="read"),
         plc.Table([col.to_pylibcudf(mode="read") for col in target_columns]),
     )