Backport PR #45623: REGR: loc.setitem losing index name when df was empty before (#45650)

Co-authored-by: Patrick Hoefler <[email protected]>

Author: meeseeksmachine

doc/source/whatsnew/v1.4.1.rst

@@ -15,6 +15,7 @@ including other versions of pandas.
 Fixed regressions
 ~~~~~~~~~~~~~~~~~
 - Regression in :meth:`Series.mask` with ``inplace=True`` and ``PeriodDtype`` and an incompatible ``other`` coercing to a common dtype instead of raising (:issue:`45546`)
+- Regression in :meth:`DataFrame.loc.__setitem__` losing :class:`Index` name if :class:`DataFrame` was empty before (:issue:`45621`)
 -
 
 .. ---------------------------------------------------------------------------

pandas/core/indexing.py

@@ -2003,7 +2003,7 @@ def _setitem_with_indexer_missing(self, indexer, value):
                 # We will ignore the existing dtypes instead of using
                 #  internals.concat logic
                 df = value.to_frame().T
-                df.index = [indexer]
+                df.index = Index([indexer], name=self.obj.index.name)
                 if not has_dtype:
                     # i.e. if we already had a Series or ndarray, keep that
                     #  dtype.  But if we had a list or dict, then do inference

pandas/tests/frame/indexing/test_indexing.py

@@ -1281,6 +1281,13 @@ def test_setting_mismatched_na_into_nullable_fails(
         with pytest.raises(TypeError, match=msg):
             df2.iloc[:2, 0] = [null, null]
 
+    def test_loc_expand_empty_frame_keep_index_name(self):
+        # GH#45621
+        df = DataFrame(columns=["b"], index=Index([], name="a"))
+        df.loc[0] = 1
+        expected = DataFrame({"b": [1]}, index=Index([0], name="a"))
+        tm.assert_frame_equal(df, expected)
+
 
 class TestDataFrameIndexingUInt64:
     def test_setitem(self, uint64_frame):