BUG: DataFrame attributes are lost after specifying dtype for select columns

[blaylockbk]

• 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 master branch of pandas.

Reproducible Example

# I add attributes to a DataFrame
df = pd.DataFrame(dict(column1=[1,2,3,4], column2=[1,2,3,4]))
df.attrs['name'] = "hello"
print(df)
print(df.attrs)

OUT:
   column1  column2
0        1        1
1        2        2
2        3        3
3        4        4
{'name': 'hello'}

# When I change the dtype, the attribute is still there (as expected)
df1 = df.astype(float)
print(df1)
print(df1.attrs)

OUT:
   column1  column2
0      1.0      1.0
1      2.0      2.0
2      3.0      3.0
3      4.0      4.0
{'name': 'hello'}

# But when I specify the dtype for specific columns, the attributes are lost
df2 = df.astype({'column1':float})
print(df2)
print(df2.attrs)

OUT:
column1  column2
0      1.0        1
1      2.0        2
2      3.0        3
3      4.0        4
{}

Issue Description

I assigned attributes to a DataFrame. When I change the DataFrame dtype with df.astype(float), the DataFrame attribute are retained, as expected. But when I change the dtype for specific columns with df.astype({'column1':float}), the attributes are lost.

Expected Behavior

I expected the DataFrame to retrain the attributes after specifying the dtype for specific columns as it does when I specify the dtype for the full DataFrame.

Installed Versions

INSTALLED VERSIONS

commit : 945c9ed
python : 3.9.7.final.0
python-bits : 64
OS : Linux
OS-release : 5.3.18-24.78-default
Version : #1 SMP Mon Aug 2 21:00:23 UTC 2021 (a13100d)
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.3.4
numpy : 1.21.3
pytz : 2021.3
dateutil : 2.8.2
pip : 21.3.1
setuptools : 58.2.0
Cython : None
pytest : 6.2.5
hypothesis : None
sphinx : 4.0.3
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.11.0
IPython : 7.28.0
pandas_datareader: None
bs4 : 4.10.0
bottleneck : None
fsspec : 2021.10.1
fastparquet : None
gcsfs : None
matplotlib : 3.4.3
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyxlsb : None
s3fs : 2021.10.1
scipy : 1.7.1
sqlalchemy : 1.4.26
tables : None
tabulate : None
xarray : 0.19.0
xlrd : None
xlwt : None
numba : None

[phofl]

This is related to #28283

[skwirskj]

I am new to the open-source community but would love to help and contribute to this issue!

[skwirskj]

take

[skwirskj]

@blaylockbk I have confirmed that this bug exists on the master branch of pandas

[skwirskj]

I have a quick question about how I am planning to fix this bug. I am thinking of adding a finalize call to propogate the metadata right after line 5877 in generic.py:
[5877] result.columns = self.columns

Can anyone think of any issues this may bring up with performance? I know finalize is noted as sometimes taking up a bit of time compared to others, but I believe it is necessary to fix this issue.