BUG: df.loc[[x], :] fails if df has zero rows

[RagBlufThim]

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

• I have confirmed this bug exists on the latest version of pandas.

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample, a copy-pastable example

import pandas as pd
df = pd.DataFrame({"A": [12,23,34,45]}, index = [list("aabb"), [0,1,2,3]])
print(df)
print("- - -")
print(df.loc[df.A < 30, :].loc[["b"], :])  # empty as expected
print("- - -")
print(df.loc[df.A < 10, :].loc[["b"], :])  # raises ValueError

Complete Output of Code Sample

      A
a 0  12
  1  23
b 2  34
  3  45
- - -
Empty DataFrame
Columns: [A]
Index: []
- - -
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\test\venv124\lib\site-packages\pandas\core\indexing.py", line 889, in __getitem__
    return self._getitem_tuple(key)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexing.py", line 1060, in _getitem_tuple
    return self._getitem_lowerdim(tup)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexing.py", line 791, in _getitem_lowerdim
    return self._getitem_nested_tuple(tup)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexing.py", line 865, in _getitem_nested_tuple
    obj = getattr(obj, self.name)._getitem_axis(key, axis=axis)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexing.py", line 1113, in _getitem_axis
    return self._getitem_iterable(key, axis=axis)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexing.py", line 1053, in _getitem_iterable
    keyarr, indexer = self._get_listlike_indexer(key, axis, raise_missing=False)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexing.py", line 1254, in _get_listlike_indexer
    indexer, keyarr = ax._convert_listlike_indexer(key)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexes\multi.py", line 2559, in _convert_listlike_indexer
    _, indexer = self.reindex(keyarr, level=level)
  File "C:\test\venv124\lib\site-packages\pandas\core\indexes\multi.py", line 2470, in reindex
    target, indexer, _ = self._join_level(
  File "C:\test\venv124\lib\site-packages\pandas\core\indexes\base.py", line 3924, in _join_level
    ngroups = 1 + new_lev_codes.max()
  File "C:\test\venv124\lib\site-packages\numpy\core\_methods.py", line 39, in _amax
    return umr_maximum(a, axis, None, out, keepdims, initial, where)
ValueError: zero-size array to reduction operation maximum which has no identity

Problem description

In both cases the df.loc[df.A...] returns a dataframe that doesn't contain any rows with an index value of "b".
Accordingly in the first case the result of .loc[["b"], :] is an empty dataframe, but in the second case a ValueError is raised. The difference between the cases is that in the first case df.loc[df.A...] returns a dataframe with some rows (though none with index value "b"), while in the second case df.loc[df.A...] returns a dataframe with zero rows.
I think that shouldn't make a difference.

In the original code .loc[df.A...] and .loc[["b"], :] are not directly combined in one expression, but the first one creates a selection of rows of the dataframe, this selection is processed further, and during this another expression uses the second .loc.

The traceback looks very similar to the one in #40235. Maybe both bugs have a common root cause.

Expected Output

df.loc[df.A < 10, :].loc[["b"], :] returns an empty dataframe like df.loc[df.A < 30, :].loc[["b"], :] does.

Output of pd.show_versions()

INSTALLED VERSIONS

commit : 2cb9652
python : 3.9.4.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.18362
machine : AMD64
...

pandas : 1.2.4
numpy : 1.20.2
pytz : 2021.1
dateutil : 2.8.1
...

[phofl]

Hi, thanks for your report. I think both cases should raise a KeyError not return an empty DataFrame. But we are not as consistent as we would like with MultiIndexes.

[phofl]

@jbrockmendel

We are running through def _get_listlike_indexer(self, key, axis: int, raise_missing: bool = False) here which still has raise_missing. Should this have been removed with 1.0?

pandas/pandas/core/indexing.py

Line 1269 in 88ce933

def _get_listlike_indexer(self, key, axis: int, raise_missing: bool = False):

[jreback]

@jbrockmendel

We are running through def _get_listlike_indexer(self, key, axis: int, raise_missing: bool = False) here which still has raise_missing. Should this have been removed with 1.0?

pandas/pandas/core/indexing.py

Line 1269 in 88ce933

def _get_listlike_indexer(self, key, axis: int, raise_missing: bool = False):

hmm yeah i think we did remove that (or thought)

[phofl]

Thanks, will check what impact this would have, then we can decide if we want to remove now or with 2.0

[jbrockmendel]

We are running through def _get_listlike_indexer(self, key, axis: int, raise_missing: bool = False) here which still has raise_missing. Should this have been removed with 1.0?

AFAICT raise_missing only affects the message that goes with the KeyError, not whether an exception is raised at all. Am I reading that wrong?

[phofl]

Yes you are right. I've only read the docstring, which says

raise_missing: bool
    Whether to raise a KeyError if some labels are not found. Will be
    removed in the future, and then this method will always behave as
    if raise_missing=True.

so I figured this would affect the actual error thrown. Should we remove this keyword and always raise with the same message? If we want to wait with this, I would update the docstring to reflect the actual behavior

[jbrockmendel]

No real opinion here, will trust your judgement. I'll be happy with just about anything that simplifies the code.