BUG: DataFrame.copy mutates the source dataframe, discarding Series attributes listed in Series._metadata.

[selik]

• 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

"""
Metadata wipeout example.
"""

import pandas as pd


class S(pd.Series):
    _metadata = ['foo']
    @property
    def _constructor(self):
        return S
    @property
    def _constructor_expanddim(self):
        return DF


class DF(pd.DataFrame):
    def __repr__(self):
        foos = {k: getattr(v, 'foo', None) for k, v in self.items()}
        return super().__repr__() + f'\n{foos=}'
    @property
    def _constructor(self):
        return DF
    @property
    def _constructor_sliced(self):
        return S


if __name__ == '__main__':
    df = DF({'a': [1]})
    df['a'].foo = 'bar'
    df.copy()  # ``DataFrame.copy`` mutates the data!
    assert hasattr(df['a'], 'foo'), "Dataframe mutated by ``copy`` method"

Issue Description

This behavior is a regression. In version 1.0.x, Pandas supported Series extension by listing attribute names in _metadata. These attributes were preserved after DataFrame.copy. Currently, DataFrame.copy will discard the extended Series attributes, not only on the new dataframe returned, but on the original as well!

Expected Behavior

DataFrame.copy should preserve the metadata attributes of each member Series.

Installed Versions

In [2]: pd.show_versions()

INSTALLED VERSIONS

commit : 945c9ed
python : 3.9.7.final.0
python-bits : 64
OS : Darwin
OS-release : 21.1.0
Version : Darwin Kernel Version 21.1.0: Wed Oct 13 17:33:24 PDT 2021; root:xnu-8019.41.5~1/RELEASE_ARM64_T8101
machine : arm64
processor : arm
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.3.4
numpy : 1.21.4
pytz : 2021.3
dateutil : 2.8.2
pip : 21.3.1
setuptools : 58.5.3
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.0.2
IPython : 7.29.0
pandas_datareader: None
bs4 : None
bs4 : None
bottleneck : None
fsspec : None
fastparquet : None
gcsfs : None
matplotlib : 3.4.3
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyxlsb : None
s3fs : None
scipy : 1.7.1
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : None

[jreback]

_metadata is an internal notion - there is no guarantee about preserving this, that said

• would take a PR to fix
• generally you should use the public .attrs which has way more testing and guarantees

[simonjayhawkins]

This behavior is a regression. In version 1.0.x

first bad commit: [4fc1758] BUG: DataFrame._item_cache not cleared on on .copy() (#33299)

[selik]

@jreback if _metadata is an internal notion, the documentation about defining original properties should be revised to clarify that. Instead of mentioning attrs, it suggests using _metadata.

Define _metadata for normal properties which will be passed to manipulation results.

[jreback]

would take a PR removing that
_metadata has little support

[phofl]

The premise doesn't make sense with CoW

df["a"]
df["a"]

are two entirely different objects under CoW so changes to one won't propagate to the other

closing