Assigned conversion via loc to Int64 fails under peculiar conditions

[shippy]

Code Sample, a copy-pastable example if possible

test = pd.DataFrame({'a': ['test'], 'b': [np.nan]})
test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64')
# *** AttributeError: 'IntegerArray' object has no attribute 'size'

Problem description

Assignment of .astype('Int64') sometimes fails with AttributeError: 'IntegerArray' object has no attribute 'size'.

Initially, I thought this was a regression from #25584, but it only manifests under very specific conditions:

1. There is exactly one null value in the converted column

   # works
   test = pd.DataFrame({'a': ['test', 'test'], 'b': [np.nan, 1.0]})
   test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64')
   
   # works
   test = pd.DataFrame({'a': ['test', 'test'], 'b': [np.nan, np.nan]})
   test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64')
   
   # fails
   test = pd.DataFrame({'a': ['test'], 'b': [np.nan]})
   test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64')
   # *** AttributeError: 'IntegerArray' object has no attribute 'size'

2. There is another column in the DataFrame

   # works
   test = pd.DataFrame({'b': [np.nan]})
   test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64')

3. Assignment happens via df.loc[:, colname] - without assignment, or with assignment to df[colname], things work.

   test = pd.DataFrame({'a': ['test'], 'b': [np.nan]})
   
   # fails
   test.loc[:, 'b'] = test['b'].astype('Int64')
   # *** AttributeError: 'IntegerArray' object has no attribute 'size'
   
   # fails
   test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64')
   # *** AttributeError: 'IntegerArray' object has no attribute 'size'
   
   # works
   test.loc[:, 'b'].astype('Int64')
   
   # works
   test['b'] = test['b'].astype('Int64')

Expected Output

test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64') shouldn't raise an AttributeError.

print(test) after assignment should read along the lines of

0    <NA>
dtype: Int64

Output of pd.show_versions()

INSTALLED VERSIONS

commit : None
python : 3.6.9.final.0
python-bits : 64
OS : Linux
OS-release : 4.15.0-46-generic
machine : x86_64
processor :
byteorder : little
LC_ALL : None
LANG : C.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.0.1
numpy : 1.16.1
pytz : 2018.9
dateutil : 2.8.0
pip : 19.3.1
setuptools : 41.4.0
Cython : None
pytest : 3.10.1
hypothesis : None
sphinx : 1.8.5
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.3.0
html5lib : None
pymysql : 0.9.3
psycopg2 : None
jinja2 : 2.10
IPython : 7.2.0
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : 4.3.0
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pytables : None
pytest : 3.10.1
pyxlsb : None
s3fs : None
scipy : 1.2.0
sqlalchemy : 1.2.17
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
xlsxwriter : None
numba : None

[shippy]

I'm happy to report that this works when using pd.NA instead of np.nan:

# works
test = pd.DataFrame({'a': ['test'], 'b': [pd.NA]})
test.loc[:, 'b'] = test.loc[:, 'b'].astype('Int64')

[Dr-Irv]

This works fine in pandas 1.1.1.

A PR that adds a test for this would be welcome.

[TAJD]

take

[hardikpnsp]

@TAJD, are you still working on the test? I would like to contribute if otherwise.

[TAJD]

Please contribute! I'm no longer working on this.

[hardikpnsp]

Take

[hardikpnsp]

So far I have added a test for functionality similar to the one provided in the issue. Basically, test astype via loc on np.nan on Int64, Int32 and Int16. should I add tests for pd.NA too? Was adding a test for Int32 and Int16 unnecessary? Also, do we need similar tests for Series? As I am quite new to this, I would appreciate some guidance.