BUG: series with dtype "Int64" don't work correctly with replace() method

[MarcusJellinghaus]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas. ( Version 1.0.4)

---

I get an AssertionError from the following code:

Code Sample

Int_series = pd.Series(data=[1, 2, 3], dtype="Int64")
Int_series.replace("", "ABC", inplace=True)  # throws exception

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
~/sandbox/pandas/pandas/core/internals/blocks.py in replace(self, to_replace, value, inplace, regex, convert)
    731         try:
--> 732             blocks = self.putmask(mask, value, inplace=inplace)
    733             # Note: it is _not_ the case that self._can_hold_element(value)

~/sandbox/pandas/pandas/core/internals/blocks.py in putmask(self, mask, new, inplace, axis, transpose)
   1608
-> 1609         new_values[mask] = new
   1610         return [self.make_block(values=new_values)]

~/sandbox/pandas/pandas/core/arrays/masked.py in __setitem__(self, key, value)
     93             value = [value]
---> 94         value, mask = self._coerce_to_array(value)
     95

~/sandbox/pandas/pandas/core/arrays/integer.py in _coerce_to_array(self, value)
    417     def _coerce_to_array(self, value) -> Tuple[np.ndarray, np.ndarray]:
--> 418         return coerce_to_array(value, dtype=self.dtype)
    419

~/sandbox/pandas/pandas/core/arrays/integer.py in coerce_to_array(values, dtype, mask, copy)
    241     elif not (is_integer_dtype(values) or is_float_dtype(values)):
--> 242         raise TypeError(f"{values.dtype} cannot be converted to an IntegerDtype")
    243

TypeError: <U3 cannot be converted to an IntegerDtype

During handling of the above exception, another exception occurred:

AssertionError                            Traceback (most recent call last)
<ipython-input-3-36a959f691c3> in <module>
      1 Int_series = pd.Series(data=[1, 2, 3], dtype="Int64")
----> 2 Int_series.replace("", "ABC", inplace=True)  # throws exception

~/sandbox/pandas/pandas/core/series.py in replace(self, to_replace, value, inplace, limit, regex, method)
   4449             limit=limit,
   4450             regex=regex,
-> 4451             method=method,
   4452         )
   4453

~/sandbox/pandas/pandas/core/generic.py in replace(self, to_replace, value, inplace, limit, regex, method)
   6662                 elif not is_list_like(value):  # NA -> 0
   6663                     new_data = self._mgr.replace(
-> 6664                         to_replace=to_replace, value=value, inplace=inplace, regex=regex
   6665                     )
   6666                 else:

~/sandbox/pandas/pandas/core/internals/managers.py in replace(self, value, **kwargs)
    592     def replace(self, value, **kwargs) -> "BlockManager":
    593         assert np.ndim(value) == 0, value
--> 594         return self.apply("replace", value=value, **kwargs)
    595
    596     def replace_list(

~/sandbox/pandas/pandas/core/internals/managers.py in apply(self, f, align_keys, **kwargs)
    400                 applied = b.apply(f, **kwargs)
    401             else:
--> 402                 applied = getattr(b, f)(**kwargs)
    403             result_blocks = _extend_blocks(applied, result_blocks)
    404

~/sandbox/pandas/pandas/core/internals/blocks.py in replace(self, to_replace, value, inplace, regex, convert)
    743                 raise
    744
--> 745             assert not self._can_hold_element(value), value
    746
    747             # try again with a compatible block

AssertionError: ABC

Problem description

The code does not work on Pandas Version 1.0.4. It throws an exception. However, it works on Pandas Version 1.0.1, and also with dtype="int64.

Expected Output

No crash. function replace() itself does nothing, since there is no "" in the series.

Output of pd.show_versions()

INSTALLED VERSIONS

commit : None
python : 3.7.0.final.0
python-bits : 64
OS : Windows
OS-release : 10
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 10, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.None
pandas : 1.0.4
numpy : 1.18.4
pytz : 2020.1
dateutil : 2.8.1
pip : 19.0.3
setuptools : 40.8.0
Cython : None
pytest : 5.4.2
hypothesis : None
sphinx : 3.0.3
blosc : None
feather : None
xlsxwriter : None

[TomAugspurger]

Added the traceback to the original post.

The line it raises on is #27768 (cc @jbrockmendel).

Not sure what to do here. I think the choices are

1. remove the assert
2. only assert for non extension blocks
3. push the _can_hold_element into the EA interface

[jbrockmendel]

IIUC the call to self.putmask (L810 in the linked PR) raises and then the assert not self._can_hold_element(value) (L823) fails?

I think _can_hold_element or something like it will need to be in the EA interface eventually. The same method would also be used for validating a fill_value for e.g. take or fillna or...