Make CategoricalColumn.__init__ strict (#16456)

This PR transfers some of the validation logic in `build_column` directly into `CategoricalColumn` just in case `CategoricalColumn` is called independently of `build_column`. Additionally adds stricter validation of `data`, `dtype` and `children` so the column doesn't represent an invalid state

xref #16469

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)
  - GALI PREM SAGAR (https://github.com/galipremsagar)

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

URL: #16456

python/cudf/cudf/_lib/column.pyx

@@ -86,7 +86,7 @@ cdef class Column:
         object mask=None,
         int offset=0,
         object null_count=None,
-        object children=()
+        tuple children=()
     ):
         if size < 0:
             raise ValueError("size must be >=0")
@@ -297,11 +297,11 @@ cdef class Column:
                 dtypes = [
                     base_child.dtype for base_child in self.base_children
                 ]
-                self._children = [
+                self._children = tuple(
                     child._with_type_metadata(dtype) for child, dtype in zip(
                         children, dtypes
                     )
-                ]
+                )
         return self._children
 
     def set_base_children(self, value):

python/cudf/cudf/core/column/categorical.py

@@ -465,6 +465,18 @@ def reorder_categories(
         )
 
 
+def validate_categorical_children(children) -> None:
+    if not (
+        len(children) == 1
+        and isinstance(children[0], cudf.core.column.numerical.NumericalColumn)
+        and children[0].dtype.kind in "iu"
+    ):
+        # TODO: Enforce unsigned integer?
+        raise ValueError(
+            "Must specify exactly one child NumericalColumn of integers for representing the codes."
+        )
+
+
 class CategoricalColumn(column.ColumnBase):
     """
     Implements operations for Columns of Categorical type
@@ -481,8 +493,7 @@ class CategoricalColumn(column.ColumnBase):
         respectively
     """
 
-    dtype: cudf.core.dtypes.CategoricalDtype
-    _codes: NumericalColumn | None
+    dtype: CategoricalDtype
     _children: tuple[NumericalColumn]
     _VALID_REDUCTIONS = {
         "max",
@@ -499,33 +510,37 @@ class CategoricalColumn(column.ColumnBase):
 
     def __init__(
         self,
+        data: None,
+        size: int | None,
         dtype: CategoricalDtype,
         mask: Buffer | None = None,
-        size: int | None = None,
         offset: int = 0,
         null_count: int | None = None,
-        children: tuple["column.ColumnBase", ...] = (),
+        children: tuple[NumericalColumn] = (),  # type: ignore[assignment]
     ):
+        if data is not None:
+            raise ValueError(f"{data=} must be None")
+        validate_categorical_children(children)
         if size is None:
-            for child in children:
-                assert child.offset == 0
-                assert child.base_mask is None
-            size = children[0].size
+            child = children[0]
+            assert child.offset == 0
+            assert child.base_mask is None
+            size = child.size
             size = size - offset
-        if isinstance(dtype, pd.api.types.CategoricalDtype):
-            dtype = CategoricalDtype.from_pandas(dtype)
         if not isinstance(dtype, CategoricalDtype):
-            raise ValueError("dtype must be instance of CategoricalDtype")
+            raise ValueError(
+                f"{dtype=} must be cudf.CategoricalDtype instance."
+            )
         super().__init__(
-            data=None,
+            data=data,
             size=size,
             dtype=dtype,
             mask=mask,
             offset=offset,
             null_count=null_count,
             children=children,
         )
-        self._codes = None
+        self._codes = self.children[0].set_mask(self.mask)
 
     @property
     def base_size(self) -> int:
@@ -558,13 +573,14 @@ def _process_values_for_isin(
         rhs = cudf.core.column.as_column(values, dtype=self.dtype)
         return lhs, rhs
 
-    def set_base_mask(self, value: Buffer | None):
+    def set_base_mask(self, value: Buffer | None) -> None:
         super().set_base_mask(value)
-        self._codes = None
+        self._codes = self.children[0].set_mask(self.mask)
 
-    def set_base_children(self, value: tuple[ColumnBase, ...]):
+    def set_base_children(self, value: tuple[NumericalColumn]) -> None:  # type: ignore[override]
         super().set_base_children(value)
-        self._codes = None
+        validate_categorical_children(value)
+        self._codes = value[0].set_mask(self.mask)
 
     @property
     def children(self) -> tuple[NumericalColumn]:
@@ -586,9 +602,7 @@ def categories(self) -> ColumnBase:
 
     @property
     def codes(self) -> NumericalColumn:
-        if self._codes is None:
-            self._codes = self.children[0].set_mask(self.mask)
-        return cast(cudf.core.column.NumericalColumn, self._codes)
+        return self._codes
 
     @property
     def ordered(self) -> bool:
@@ -1131,7 +1145,7 @@ def _mimic_inplace(
     ) -> Self | None:
         out = super()._mimic_inplace(other_col, inplace=inplace)
         if inplace and isinstance(other_col, CategoricalColumn):
-            self._codes = other_col._codes
+            self._codes = other_col.codes
         return out
 
     def view(self, dtype: Dtype) -> ColumnBase:

python/cudf/cudf/core/column/column.py

@@ -1578,19 +1578,14 @@ def build_column(
         return col
 
     if isinstance(dtype, CategoricalDtype):
-        if not len(children) == 1:
-            raise ValueError(
-                "Must specify exactly one child column for CategoricalColumn"
-            )
-        if not isinstance(children[0], ColumnBase):
-            raise TypeError("children must be a tuple of Columns")
         return cudf.core.column.CategoricalColumn(
+            data=data,  # type: ignore[arg-type]
             dtype=dtype,
             mask=mask,
             size=size,
             offset=offset,
             null_count=null_count,
-            children=children,
+            children=children,  # type: ignore[arg-type]
         )
     elif dtype.type is np.datetime64:
         return cudf.core.column.DatetimeColumn(