BUG: wrong errors when indexing with list that includes pd.NA

[jorisvandenbossche]

Currently, indexing with a list including pd.NA (so the list version of indexing with a BooleanArray or IntegerArray) works on the array, but not on Series:

("works" = raising the correct error message)

In [21]: a = pd.array([1, 2, 3])  

In [22]: a[[True, False, pd.NA]]  
...
ValueError: Cannot mask with a boolean indexer containing NA values

In [23]: a[[0, 1, pd.NA]] 
...
ValueError: Cannot index with an integer indexer containing NA values

vs

In [24]: s = pd.Series(a)  

In [25]: s[[True, False, pd.NA]]    
...
KeyError: 'Passing list-likes to .loc or [] with any missing labels is no longer supported, see https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#deprecate-loc-reindex-listlike'

In [26]: s[[0, 1, pd.NA]]  
...
TypeError: boolean value of NA is ambiguous

Because the validation of the indexer isn't yet updated to handle listlikes that include pd.NA.

And similar problems for setitem.

[jorisvandenbossche]

Note that the version with an actual array or series of "boolean", this works already fine:

In [27]: s[pd.array([True, False, pd.NA])]  
...
ValueError: cannot mask with array containing NA / NaN values

but for integer it is actually the same issue as for the list:

In [28]: s[pd.array([0, 1, pd.NA])]  
...
TypeError: boolean value of NA is ambiguous

[lopof]

Update current behavior.
Array:

a = pd.array([1, 2, 3])
a[[True, False, pd.NA]]

<IntegerArray>
[1]
Length: 1, dtype: Int64

a[[0, 1, pd.NA]]

...
ValueError: Cannot index with an integer indexer containing NA values

Series:

s = pd.Series(a)
s[[True, False, pd.NA]]

...
KeyError: '[<NA>] not in index'

s[[0, 1, pd.NA]]

...
KeyError: '[<NA>] not in index'

s[pd.array([True, False, pd.NA])]

0    1
dtype: Int64

s[pd.array([0, 1, pd.NA])]

...
KeyError: '[nan] not in index'