Add _from_column_like_self factory (#10022)

Follow up to #9558

On a return trip from libcudf, it is a common pattern for cudf frame to apply its own metadata to the columns. This PR generalizes this procedure as a new factory function `_from_colums_like_self`

Authors:
  - Michael Wang (https://github.com/isVoid)

Approvers:
  - Ram (Ramakrishna Prabhu) (https://github.com/rgsl888prabhu)
  - Paul Taylor (https://github.com/trxcllnt)
  - Vyas Ramasubramani (https://github.com/vyasr)

URL: #10022

python/cudf/cudf/core/_base_index.py

@@ -1444,7 +1444,8 @@ def drop_duplicates(
         """
 
         # This utilizes the fact that all `Index` is also a `Frame`.
-        result = self.__class__._from_columns(
+        # Except RangeIndex.
+        return self._from_columns_like_self(
             drop_duplicates(
                 list(self._columns),
                 keys=range(len(self._data)),
@@ -1453,8 +1454,6 @@ def drop_duplicates(
             ),
             self._column_names,
         )
-        result._copy_type_metadata(self, include_index=False)
-        return result
 
     def dropna(self, how="any"):
         """
@@ -1476,12 +1475,10 @@ def dropna(self, how="any"):
             for col in self._columns
         ]
 
-        result = self.__class__._from_columns(
+        return self._from_columns_like_self(
             drop_nulls(data_columns, how=how, keys=range(len(data_columns)),),
             self._column_names,
         )
-        result._copy_type_metadata(self, include_index=False)
-        return result
 
     def _gather(self, gather_map, nullify=False, check_bounds=True):
         """Gather rows of index specified by indices in `gather_map`.
@@ -1501,14 +1498,11 @@ def _gather(self, gather_map, nullify=False, check_bounds=True):
         ):
             raise IndexError("Gather map index is out of bounds.")
 
-        result = self.__class__._from_columns(
+        return self._from_columns_like_self(
             gather(list(self._columns), gather_map, nullify=nullify),
             self._column_names,
         )
 
-        result._copy_type_metadata(self, include_index=False)
-        return result
-
     def take(self, indices, axis=0, allow_fill=True, fill_value=None):
         """Return a new index containing the rows specified by *indices*
 
@@ -1561,12 +1555,10 @@ def _apply_boolean_mask(self, boolean_mask):
         if not is_bool_dtype(boolean_mask.dtype):
             raise ValueError("boolean_mask is not boolean type.")
 
-        result = self.__class__._from_columns(
+        return self._from_columns_like_self(
             apply_boolean_mask(list(self._columns), boolean_mask),
             column_names=self._column_names,
         )
-        result._copy_type_metadata(self)
-        return result
 
     def _split_columns_by_levels(self, levels):
         if isinstance(levels, int) and levels > 0:

python/cudf/cudf/core/frame.py

@@ -163,6 +163,22 @@ def _from_columns(
 
         return cls._from_data(data, index)
 
+    def _from_columns_like_self(
+        self,
+        columns: List[ColumnBase],
+        column_names: List[str],
+        index_names: Optional[List[str]] = None,
+    ):
+        """Construct a `Frame` from a list of columns with metadata from self.
+
+        If `index_names` is set, the first `len(index_names)` columns are
+        used to construct the index of the frame.
+        """
+        frame = self.__class__._from_columns(
+            columns, column_names, index_names
+        )
+        return frame._copy_type_metadata(self, include_index=bool(index_names))
+
     def _mimic_inplace(
         self: T, result: Frame, inplace: bool = False
     ) -> Optional[Frame]:

python/cudf/cudf/core/indexed_frame.py

@@ -556,7 +556,7 @@ def _gather(
         ):
             raise IndexError("Gather map index is out of bounds.")
 
-        result = self.__class__._from_columns(
+        return self._from_columns_like_self(
             libcudf.copying.gather(
                 list(self._index._columns + self._columns)
                 if keep_index
@@ -568,9 +568,6 @@ def _gather(
             self._index.names if keep_index else None,
         )
 
-        result._copy_type_metadata(self, include_index=keep_index)
-        return result
-
     def _positions_from_column_names(
         self, column_names, offset_by_index_columns=False
     ):
@@ -628,7 +625,7 @@ def drop_duplicates(
         keys = self._positions_from_column_names(
             subset, offset_by_index_columns=not ignore_index
         )
-        result = self.__class__._from_columns(
+        return self._from_columns_like_self(
             libcudf.stream_compaction.drop_duplicates(
                 list(self._columns)
                 if ignore_index
@@ -640,8 +637,6 @@ def drop_duplicates(
             self._column_names,
             self._index.names if not ignore_index else None,
         )
-        result._copy_type_metadata(self)
-        return result
 
     def add_prefix(self, prefix):
         """
@@ -1354,7 +1349,7 @@ def _drop_na_rows(
                 for col in self._columns
             ]
 
-        result = self.__class__._from_columns(
+        return self._from_columns_like_self(
             libcudf.stream_compaction.drop_nulls(
                 list(self._index._data.columns) + data_columns,
                 how=how,
@@ -1366,8 +1361,6 @@ def _drop_na_rows(
             self._column_names,
             self._index.names,
         )
-        result._copy_type_metadata(self)
-        return result
 
     def _apply_boolean_mask(self, boolean_mask):
         """Apply boolean mask to each row of `self`.
@@ -1378,15 +1371,13 @@ def _apply_boolean_mask(self, boolean_mask):
         if not is_bool_dtype(boolean_mask.dtype):
             raise ValueError("boolean_mask is not boolean type.")
 
-        result = self.__class__._from_columns(
+        return self._from_columns_like_self(
             libcudf.stream_compaction.apply_boolean_mask(
                 list(self._index._columns + self._columns), boolean_mask
             ),
             column_names=self._column_names,
             index_names=self._index.names,
         )
-        result._copy_type_metadata(self)
-        return result
 
     def take(self, indices, axis=0):
         """Return a new frame containing the rows specified by *indices*.