Increase edge case check strictness when cols_droplevel contains empty values (#62527)

Author: matiaslindgren

pandas/core/frame.py

@@ -107,6 +107,7 @@
  is_list_like,
  is_scalar,
  is_sequence,
+ is_string_dtype,
  needs_i8_conversion,
  pandas_dtype,
 )
@@ -4454,8 +4455,12 @@ def _set_item_frame_value(self, key, value: DataFrame) -> None:
  cols_droplevel = maybe_droplevels(cols, key)
  if (
  not isinstance(cols_droplevel, MultiIndex)
+ and is_string_dtype(cols_droplevel.dtype)
  and not cols_droplevel.any()
  ):
+ # if cols_droplevel contains only empty strings,
+ # value.reindex(cols_droplevel, axis=1) would be full of NaNs
+ # see GH#62518 and GH#61841
  return
  if len(cols_droplevel) and not cols_droplevel.equals(value.columns):
  value = value.reindex(cols_droplevel, axis=1)

pandas/tests/indexing/multiindex/test_multiindex.py

@@ -271,3 +271,24 @@ def test_multiindex_assign_aligns_as_implicit_tuple(self):
  df1["C"] = s1
  tm.assert_frame_equal(df1, df2)
  tm.assert_frame_equal(df1, df3)
+
+ def test_multiindex_assign_alignment_with_non_string_dtype(self):
+ # GH 62518
+ columns = MultiIndex.from_arrays(
+ [["a", "a", "z", "z"], pd.Categorical([1, 2, 1, 2])]
+ )
+
+ meta = DataFrame(columns=columns, dtype=object)
+ meta["z"] = meta["z"].astype("int64")
+
+ result = DataFrame(
+ data={
+ ("a", 1): Series([], dtype=object),
+ ("a", 2): Series([], dtype=object),
+ ("z", 1): Series([], dtype="int64"),
+ ("z", 2): Series([], dtype="int64"),
+ },
+ columns=columns,
+ )
+
+ tm.assert_frame_equal(meta, result)