BUG: ensure reindex / getitem to select columns properly copies data for extension dtypes (#51197)

Author: jorisvandenbossche

doc/source/whatsnew/v2.0.0.rst

@@ -1395,6 +1395,7 @@ ExtensionArray
 - Bug in :meth:`api.types.is_integer_dtype`, :meth:`api.types.is_unsigned_integer_dtype`, :meth:`api.types.is_signed_integer_dtype`, :meth:`api.types.is_float_dtype` where a custom :class:`ExtensionDtype` would not return ``True`` if ``kind`` returned the corresponding NumPy type (:issue:`50667`)
 - Bug in :class:`Series` constructor unnecessarily overflowing for nullable unsigned integer dtypes (:issue:`38798`, :issue:`25880`)
 - Bug in setting non-string value into ``StringArray`` raising ``ValueError`` instead of ``TypeError`` (:issue:`49632`)
+- Bug in :meth:`DataFrame.reindex` not honoring the default ``copy=True`` keyword in case of columns with ExtensionDtype (and as a result also selecting multiple columns with getitem (``[]``) didn't correctly result in a copy) (:issue:`51197`)
 
 Styler
 ^^^^^^

pandas/core/internals/managers.py

@@ -827,7 +827,7 @@ def _slice_take_blocks_ax0(
                     # only one item and each mgr loc is a copy of that single
                     # item.
                     for mgr_loc in mgr_locs:
-                        newblk = blk.copy(deep=False)
+                        newblk = blk.copy(deep=not only_slice)
                         newblk.mgr_locs = BlockPlacement(slice(mgr_loc, mgr_loc + 1))
                         blocks.append(newblk)
 

pandas/tests/frame/methods/test_reindex.py

@@ -133,6 +133,21 @@ def test_reindex_copies(self):
         result2 = df.reindex(columns=cols, index=df.index, copy=True)
         assert not np.shares_memory(result2[0]._values, df[0]._values)
 
+    def test_reindex_copies_ea(self):
+        # https://github.com/pandas-dev/pandas/pull/51197
+        # also ensure to honor copy keyword for ExtensionDtypes
+        N = 10
+        df = DataFrame(np.random.randn(N * 10, N), dtype="Float64")
+        cols = np.arange(N)
+        np.random.shuffle(cols)
+
+        result = df.reindex(columns=cols, copy=True)
+        assert not np.shares_memory(result[0].array._data, df[0].array._data)
+
+        # pass both columns and index
+        result2 = df.reindex(columns=cols, index=df.index, copy=True)
+        assert not np.shares_memory(result2[0].array._data, df[0].array._data)
+
     @td.skip_array_manager_not_yet_implemented
     def test_reindex_date_fill_value(self):
         # passing date to dt64 is deprecated; enforced in 2.0 to cast to object