Regression in replace raising ValueError for bytes object

doc/source/whatsnew/v1.2.1.rst

@@ -26,6 +26,7 @@ Fixed regressions
 - Bug in :meth:`read_csv` with ``float_precision="high"`` caused segfault or wrong parsing of long exponent strings. This resulted in a regression in some cases as the default for ``float_precision`` was changed in pandas 1.2.0 (:issue:`38753`)
 - Fixed regression in :meth:`Rolling.skew` and :meth:`Rolling.kurt` modifying the object inplace (:issue:`38908`)
 - Fixed regression in :meth:`read_csv` and other read functions were the encoding error policy (``errors``) did not default to ``"replace"`` when no encoding was specified (:issue:`38989`)
+- Fixed regression in :meth:`DataFrame.replace` raising ValueError when :class:`DataFrame` has dtype ``bytes`` (:issue:`38900`)
 
 .. ---------------------------------------------------------------------------
 

pandas/core/internals/blocks.py

@@ -2293,7 +2293,7 @@ class ObjectBlock(Block):
     _can_hold_na = True
 
     def _maybe_coerce_values(self, values):
-        if issubclass(values.dtype.type, str):
+        if issubclass(values.dtype.type, (str, bytes)):
             values = np.array(values, dtype=object)
         return values
 

pandas/tests/frame/methods/test_replace.py

@@ -1636,3 +1636,10 @@ def test_replace_unicode(self):
         result = df1.replace(columns_values_map)
         expected = DataFrame({"positive": np.ones(3)})
         tm.assert_frame_equal(result, expected)
+
+    def test_replace_bytes(self, frame_or_series):
+        # GH#38900
+        obj = frame_or_series(["o"]).astype("|S")
+        expected = obj.copy()
+        obj = obj.replace({None: np.nan})
+        tm.assert_equal(obj, expected)