[BUG] Allow where() to work with a Series and other=cudf.NA

[sarahyurick]

In Pandas, I can do this:

import pandas as pd

df = pd.DataFrame()
df['id'] = [0, 1, 1]
df['val'] = [0, 4, 2]

df.where(df['id']==1, pd.NA)
	id	val
0	<NA>	<NA>
1	1	4
2	1	2

however, when I try in cuDF, I get:

import cudf

df = cudf.DataFrame()
df['id'] = [0, 1, 1]
df['val'] = [0, 4, 2]

df.where(df['id']==1, cudf.NA)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/tmp/ipykernel_50448/3560034258.py in <module>
      5 df['val'] = [0, 4, 2]
      6 
----> 7 df.where(df['id']==1, cudf.NA)

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/frame.py in where(self, cond, other, inplace)
    878         """
    879 
--> 880         return cudf.core._internals.where(
    881             frame=self, cond=cond, other=other, inplace=inplace
    882         )

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/_internals/where.py in where(frame, cond, other, inplace)
    263             cond.columns = frame.columns
    264 
--> 265         (source_df, others,) = _normalize_columns_and_scalars_type(
    266             frame, other
    267         )

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/_internals/where.py in _normalize_columns_and_scalars_type(frame, other, inplace)
    158                     source_col,
    159                     other_scalar,
--> 160                 ) = _check_and_cast_columns_with_other(
    161                     source_col=source_df._data[col_name],
    162                     other=other

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/_internals/where.py in _check_and_cast_columns_with_other(source_col, other, inplace)
     44 
     45     if cudf.utils.dtypes.is_scalar(other):
---> 46         device_obj = _normalize_scalars(source_col, other)
     47     else:
     48         device_obj = other

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/_internals/where.py in _normalize_scalars(col, other)
     28         )
     29 
---> 30     return cudf.Scalar(other, dtype=col.dtype if other in None else None)
     31 
     32 

TypeError: argument of type 'NoneType' is not iterable

[beckernick]

Are we unable to handle the series itself or just the other=cudf.NA? I believe we recently did some work on this in #7383

[sarahyurick]

@beckernick Good point - I just tested it and I get:

import pandas as pd

df = pd.DataFrame()
df['id'] = [0, 1, 1]
df['val'] = [0, 4, 2]

df.where(df['id']==1, 28)
	id	val
0	28	28
1	1	4
2	1	2

versus

import cudf

df = cudf.DataFrame()
df['id'] = [0, 1, 1]
df['val'] = [0, 4, 2]

df.where(df['id']==1, 28)
	id	val
0	28	<NA>
1	1	<NA>
2	1	<NA>

so I'll have to make sure that case is handled, too.

EDIT: Actually, it looks like it's getting mangled with some of the code I've already written. I'll make sure to keep this case and PR #8747 in mind, though.