BUG: Outer merge casts EA dtype to object

[adaitche]

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

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

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample, a copy-pastable example

d1 = pd.DataFrame([("a",)], columns=["id"], dtype="string")
d2 = pd.DataFrame([("b",)], columns=["id"], dtype="string")
result = d1.merge(d2, on="id", how="outer")
dtype = result.dtypes["id"]
print(dtype) # expected 'string', but is 'object'

Problem description

When doing an outer merge on a column of type string, the result contains this column with the type object. This makes it more difficult to use the new string type.

The problem doesn't occur with the types Int32, Float32, boolean. It seems to be specific to the string type.

The problem doesn't occur with a left, right or inner-merge.

Output of pd.show_versions()

INSTALLED VERSIONS

commit : 7d32926
python : 3.7.3.final.0
python-bits : 64
OS : Darwin
OS-release : 18.7.0
Version : Darwin Kernel Version 18.7.0: Tue Nov 10 00:07:31 PST 2020; root:xnu-4903.278.51~1/RELEASE_X86_64
machine : x86_64
processor : i386
byteorder : little
LC_ALL : en_US.UTF-8
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.2.2
numpy : 1.20.1
pytz : 2021.1
dateutil : 2.8.1
pip : 20.2.4
setuptools : 52.0.0.post20210125
Cython : None
pytest : 6.0.1
hypothesis : 5.30.1
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.11.3
IPython : 7.20.0
pandas_datareader: None
bs4 : None
bottleneck : None
fsspec : None
fastparquet : None
gcsfs : None
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : 1.3.23
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : None

[mzeitlin11]

Thanks for the report @adaitche! Can confirm this is still a bug on master, investigations to fix are welcome!

[mzeitlin11]

Issue here looks similar to #40148. An Index call loses the ExtensionDType, so either that call would need to be avoided, or #39133 solved.

[simonjayhawkins]

The problem doesn't occur with a left, right or inner-merge.

on master the problem now occurs with how="right"

>>> print(pd.__version__)
1.3.0.dev0+1567.g67c9385787
>>> d1 = pd.DataFrame([("a",)], columns=["id"], dtype="string")
>>> d2 = pd.DataFrame([("b",)], columns=["id"], dtype="string")
>>> result = d1.merge(d2, on="id", how="right")
>>> dtype = result.dtypes["id"]
>>> print(dtype)
object

first bad commit: [08ff65f] BUG: merge between partial index and index fails when result is empty (#34414) cc @phofl

The problem doesn't occur with the types Int32, Float32, boolean. It seems to be specific to the string type.

it appears this is also an issue for other EA dtypes (and now on master also for how="right")

>>> print(pd.__version__)
1.3.0.dev0+1567.g67c9385787
>>> d1 = pd.DataFrame([(1,)], columns=["id"], dtype="Int64")
>>> d2 = pd.DataFrame([(2,)], columns=["id"], dtype="Int64")
>>> result = d1.merge(d2, on="id", how="outer")
>>> dtype = result.dtypes["id"]
>>> print(dtype)
object
>>>