Regression: concatenating `MultiIndex` with empty `RangeIndex` raises

[mlondschien]

• 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.

---

As also reported here: dask/dask#7610, PR #38671 introduced the following behaviour:

A reproducer with just pandas:

df1 = pd.DataFrame([[1, 2]], columns=pd.MultiIndex.from_tuples([('B', 1), ('C', 1)]))
df2 = pd.DataFrame(index=[0], columns=pd.RangeIndex(0))

pd.concat([df1, df2])

What triggers the error here is the columns=pd.RangeIndex(0) for the empty df2 (by default pandas creates a zero-length object dtype Index, which works fine, but if it's an empty RangeIndex, we now get this error).

This is a regression in itself in pandas. But I am also wondering a bit where the empty RangeIndex is coming from (it seems that the groupby operation in dask results in some empty partitions)

Originally posted by @jorisvandenbossche in dask/dask#7610 (comment)

IIUC the fix for this could possibly be as simple as replacing

pandas/pandas/core/indexes/base.py

Lines 2926 to 2928 in 0acbff8

	if isinstance(self, ABCMultiIndex) and not is_object_dtype(
	unpack_nested_dtype(other)
	):

with

            if isinstance(self, ABCMultiIndex) and not is_object_dtype(
                unpack_nested_dtype(other)
            ) and len(other) > 0:

[jorisvandenbossche]

cc @jbrockmendel