Fix key bug, simplify some implementations

python/cudf/cudf/core/_base_index.py

@@ -421,7 +421,7 @@ def hasnans(self):
         raise NotImplementedError
 
     @property
-    def nlevels(self):
+    def nlevels(self) -> int:
         """
         Number of levels.
         """
@@ -1951,7 +1951,6 @@ def drop_duplicates(
         return self._from_columns_like_self(
             drop_duplicates(
                 list(self._columns),
-                keys=range(len(self._columns)),
                 keep=keep,
                 nulls_are_equal=nulls_are_equal,
             ),

python/cudf/cudf/core/frame.py

@@ -1014,19 +1014,6 @@ def to_arrow(self):
             }
         )
 
-    @_performance_tracking
-    def _positions_from_column_names(self, column_names) -> list[int]:
-        """Map each column name into their positions in the frame.
-
-        The order of indices returned corresponds to the column order in this
-        Frame.
-        """
-        return [
-            i
-            for i, name in enumerate(self._column_names)
-            if name in set(column_names)
-        ]
-
     @_performance_tracking
     def _copy_type_metadata(self: Self, other: Self) -> Self:
         """

python/cudf/cudf/core/indexed_frame.py

@@ -3055,21 +3055,21 @@ def _slice(self, arg: slice, keep_index: bool = True) -> Self:
         return result
 
     def _positions_from_column_names(
-        self, column_names, offset_by_index_columns=False
-    ):
+        self,
+        column_names: set[abc.Hashable],
+        offset_by_index_columns: bool = True,
+    ) -> list[int]:
         """Map each column name into their positions in the frame.
 
         Return positions of the provided column names, offset by the number of
         index columns if `offset_by_index_columns` is True. The order of
         indices returned corresponds to the column order in this Frame.
         """
-        num_index_columns = (
-            len(self.index._data) if offset_by_index_columns else 0
-        )
+        start = self.index.nlevels if offset_by_index_columns else 0
         return [
-            i + num_index_columns
-            for i, name in enumerate(self._column_names)
-            if name in set(column_names)
+            i
+            for i, name in enumerate(self._column_names, start=start)
+            if name in column_names
         ]
 
     def drop_duplicates(
@@ -4349,9 +4349,7 @@ def _drop_na_rows(self, how="any", subset=None, thresh=None):
             cudf.core._internals.stream_compaction.drop_nulls(
                 [*self.index._columns, *data_columns],
                 how=how,
-                keys=self._positions_from_column_names(
-                    subset, offset_by_index_columns=True
-                ),
+                keys=self._positions_from_column_names(subset),
                 thresh=thresh,
             ),
             self._column_names,
@@ -6282,17 +6280,16 @@ def ge(self, other, axis="columns", level=None, fill_value=None):  # noqa: D102
             other=other, op="__ge__", fill_value=fill_value, can_reindex=True
         )
 
-    def _preprocess_subset(self, subset):
+    def _preprocess_subset(self, subset) -> set[abc.Hashable]:
         if subset is None:
             subset = self._column_names
         elif (
-            not np.iterable(subset)
-            or isinstance(subset, str)
+            is_scalar(subset)
             or isinstance(subset, tuple)
             and subset in self._column_names
         ):
             subset = (subset,)
-        diff = set(subset) - set(self._data)
+        diff = set(subset) - set(self._column_names)
         if len(diff) != 0:
             raise KeyError(f"columns {diff} do not exist")
         return subset