BUG: fillna downcast object to int even downcast=False/None

[yinweisu]

Pandas version checks

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

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

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd
df = pd.Series([1,2,3], dtype='object')
df.fillna('', downcast=False)  # will get int64 instead

Issue Description

fillna will convert an object dataframe to int64 dataframe even when downcast is specified to be False or None

Expected Behavior

The dataframe should keep the original dtype

Installed Versions

NSTALLED VERSIONS

commit : bb1f651
python : 3.9.7.final.0
python-bits : 64
OS : Darwin
OS-release : 20.6.0
Version : Darwin Kernel Version 20.6.0: Wed Nov 10 22:23:07 PST 2021; root:xnu-7195.141.14~1/RELEASE_X86_64
machine : x86_64
processor : i386
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.4.0
numpy : 1.21.4
pytz : 2021.3
dateutil : 2.8.2
pip : 21.3.1
setuptools : 57.4.0
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : None
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : 3.5.1
numba : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.7.3
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
zstandard : None

[phofl]

This is already known, see #44241

[jbrockmendel]

I think downcast=False is separate from #44241

[phofl]

The note inline says:

            # GH#44241 We downcast regardless of the argument;
            #  respecting 'downcast=None' may be worthwhile at some point,
            #  but ATM it breaks too much existing code.
            # split and convert the blocks

So this sounded related to me.

[jbrockmendel]

definitely related, but not identical. when we did #44241 we were focused on "None" and "infer", and False was an afterthought. IIRC there was only one place where we used downcast=False and it was in the io code so we just changed that, so there are no tests that use it.

[phofl]

Got you, thx