Merge remote-tracking branch 'upstream/branch-25.02' into rm/scalar/d…

…evicescalar

python/cudf/cudf/_lib/CMakeLists.txt

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

python/cudf/cudf/_lib/column.pxd

@@ -1,4 +1,4 @@
-# Copyright (c) 2020-2024, NVIDIA CORPORATION.
+# Copyright (c) 2020-2025, NVIDIA CORPORATION.
 
 from typing import Literal
 
@@ -13,6 +13,8 @@ from pylibcudf.libcudf.column.column_view cimport (
 from pylibcudf.libcudf.types cimport size_type
 from rmm.librmm.device_buffer cimport device_buffer
 
+cdef dtype_from_lists_column_view(column_view cv)
+cdef dtype_from_column_view(column_view cv)
 
 cdef class Column:
     cdef public:

python/cudf/cudf/_lib/column.pyx

@@ -19,24 +19,25 @@ from cudf.core.buffer import (
     as_buffer,
     cuda_array_interface_wrapper,
 )
-from cudf.utils.dtypes import _get_base_dtype
+from cudf.utils.dtypes import (
+    _get_base_dtype,
+    dtype_to_pylibcudf_type,
+    PYLIBCUDF_TO_SUPPORTED_NUMPY_TYPES,
+)
 
 from cpython.buffer cimport PyObject_CheckBuffer
-from libc.stdint cimport uintptr_t
-from libcpp.memory cimport make_unique, unique_ptr
+from libc.stdint cimport uintptr_t, int32_t
+from libcpp.memory cimport make_shared, make_unique, shared_ptr, unique_ptr
 from libcpp.utility cimport move
 from libcpp.vector cimport vector
 
 from rmm.pylibrmm.device_buffer cimport DeviceBuffer
 
-from cudf._lib.types cimport (
-    dtype_from_column_view,
-    dtype_to_pylibcudf_type,
+from pylibcudf cimport (
+    DataType as plc_DataType,
+    Column as plc_Column,
+    Scalar as plc_Scalar,
 )
-
-from cudf._lib.types import dtype_from_pylibcudf_column
-
-from pylibcudf cimport DataType as plc_DataType, 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
@@ -45,6 +46,7 @@ from pylibcudf.libcudf.column.column_factories cimport (
     make_numeric_column
 )
 from pylibcudf.libcudf.column.column_view cimport column_view
+from pylibcudf.libcudf.lists.lists_column_view cimport lists_column_view
 from pylibcudf.libcudf.null_mask cimport null_count as cpp_null_count
 from pylibcudf.libcudf.scalar.scalar cimport scalar
 
@@ -60,6 +62,80 @@ cdef get_element(column_view col_view, size_type index):
     return pylibcudf.interop.to_arrow(plc_scalar).as_py()
 
 
+def dtype_from_pylibcudf_column(plc_Column col not None):
+    type_ = col.type()
+    tid = type_.id()
+
+    if tid == pylibcudf.TypeId.LIST:
+        child = col.list_view().child()
+        return cudf.ListDtype(dtype_from_pylibcudf_column(child))
+    elif tid == pylibcudf.TypeId.STRUCT:
+        fields = {
+            str(i): dtype_from_pylibcudf_column(col.child(i))
+            for i in range(col.num_children())
+        }
+        return cudf.StructDtype(fields)
+    elif tid == pylibcudf.TypeId.DECIMAL64:
+        return cudf.Decimal64Dtype(
+            precision=cudf.Decimal64Dtype.MAX_PRECISION,
+            scale=-type_.scale()
+        )
+    elif tid == pylibcudf.TypeId.DECIMAL32:
+        return cudf.Decimal32Dtype(
+            precision=cudf.Decimal32Dtype.MAX_PRECISION,
+            scale=-type_.scale()
+        )
+    elif tid == pylibcudf.TypeId.DECIMAL128:
+        return cudf.Decimal128Dtype(
+            precision=cudf.Decimal128Dtype.MAX_PRECISION,
+            scale=-type_.scale()
+        )
+    else:
+        return PYLIBCUDF_TO_SUPPORTED_NUMPY_TYPES[tid]
+
+
+cdef dtype_from_lists_column_view(column_view cv):
+    # lists_column_view have no default constructor, so we heap
+    # allocate it to get around Cython's limitation of requiring
+    # default constructors for stack allocated objects
+    cdef shared_ptr[lists_column_view] lv = make_shared[lists_column_view](cv)
+    cdef column_view child = lv.get()[0].child()
+
+    if child.type().id() == libcudf_types.type_id.LIST:
+        return cudf.ListDtype(dtype_from_lists_column_view(child))
+    else:
+        return cudf.ListDtype(dtype_from_column_view(child))
+
+
+cdef dtype_from_column_view(column_view cv):
+    cdef libcudf_types.type_id tid = cv.type().id()
+    if tid == libcudf_types.type_id.LIST:
+        return dtype_from_lists_column_view(cv)
+    elif tid == libcudf_types.type_id.STRUCT:
+        fields = {
+            str(i): dtype_from_column_view(cv.child(i))
+            for i in range(cv.num_children())
+        }
+        return cudf.StructDtype(fields)
+    elif tid == libcudf_types.type_id.DECIMAL64:
+        return cudf.Decimal64Dtype(
+            precision=cudf.Decimal64Dtype.MAX_PRECISION,
+            scale=-cv.type().scale()
+        )
+    elif tid == libcudf_types.type_id.DECIMAL32:
+        return cudf.Decimal32Dtype(
+            precision=cudf.Decimal32Dtype.MAX_PRECISION,
+            scale=-cv.type().scale()
+        )
+    elif tid == libcudf_types.type_id.DECIMAL128:
+        return cudf.Decimal128Dtype(
+            precision=cudf.Decimal128Dtype.MAX_PRECISION,
+            scale=-cv.type().scale()
+        )
+    else:
+        return PYLIBCUDF_TO_SUPPORTED_NUMPY_TYPES[<int32_t>(tid)]
+
+
 cdef class Column:
     """
     A Column stores columnar data in device memory.
@@ -357,7 +433,7 @@ cdef class Column:
             col = self
             data_dtype = col.dtype
 
-        cdef plc_DataType dtype = dtype_to_pylibcudf_type(data_dtype)
+        cdef plc_DataType dtype = <plc_DataType?>dtype_to_pylibcudf_type(data_dtype)
         cdef libcudf_types.size_type offset = self.offset
         cdef vector[mutable_column_view] children
         cdef void* data
@@ -420,7 +496,7 @@ cdef class Column:
             col = self
             data_dtype = col.dtype
 
-        cdef plc_DataType dtype = dtype_to_pylibcudf_type(data_dtype)
+        cdef plc_DataType dtype = <plc_DataType?>dtype_to_pylibcudf_type(data_dtype)
         cdef libcudf_types.size_type offset = self.offset
         cdef vector[column_view] children
         cdef void* data

python/cudf/cudf/core/_base_index.py

@@ -10,7 +10,6 @@
 from typing_extensions import Self
 
 import cudf
-from cudf._lib.types import size_type_dtype
 from cudf.api.extensions import no_default
 from cudf.api.types import is_integer, is_list_like, is_scalar
 from cudf.core._internals import copying
@@ -24,7 +23,11 @@
 from cudf.core.copy_types import GatherMap
 from cudf.errors import MixedTypeError
 from cudf.utils import ioutils
-from cudf.utils.dtypes import can_convert_to_column, is_mixed_with_object_dtype
+from cudf.utils.dtypes import (
+    SIZE_TYPE_DTYPE,
+    can_convert_to_column,
+    is_mixed_with_object_dtype,
+)
 from cudf.utils.utils import _is_same_name
 
 if TYPE_CHECKING:
@@ -2047,7 +2050,7 @@ def _gather(self, gather_map, nullify=False, check_bounds=True):
         # TODO: For performance, the check and conversion of gather map should
         # be done by the caller. This check will be removed in future release.
         if gather_map.dtype.kind not in "iu":
-            gather_map = gather_map.astype(size_type_dtype)
+            gather_map = gather_map.astype(SIZE_TYPE_DTYPE)
 
         GatherMap(gather_map, len(self), nullify=not check_bounds or nullify)
         return self._from_columns_like_self(

python/cudf/cudf/core/_internals/aggregation.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, Literal
@@ -8,9 +8,9 @@
 import pylibcudf as plc
 
 import cudf
-from cudf._lib.types import SUPPORTED_NUMPY_TO_PYLIBCUDF_TYPES
 from cudf.api.types import is_scalar
 from cudf.utils import cudautils
+from cudf.utils.dtypes import SUPPORTED_NUMPY_TO_PYLIBCUDF_TYPES
 
 if TYPE_CHECKING:
     from collections.abc import Callable