BUG: Regression in assert_frame_equal when using check_like

[galipremsagar]

• 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

In 1.2.1:

>>> import pandas as pd
>>> one = pd.Index([], dtype='object')
>>> two = pd.RangeIndex(start=0, stop=0, step=1)
>>> df_one = pd.DataFrame(index=one)
>>> df_two = pd.DataFrame(index=two)
>>> pd.testing.assert_frame_equal(df_one, df_two, check_like=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/nvme/0/pgali/envs/cudfdev/lib/python3.7/site-packages/pandas/_testing.py", line 1655, in assert_frame_equal
    obj=f"{obj}.index",
  File "/nvme/0/pgali/envs/cudfdev/lib/python3.7/site-packages/pandas/_testing.py", line 773, in assert_index_equal
    _check_types(left, right, obj=obj)
  File "/nvme/0/pgali/envs/cudfdev/lib/python3.7/site-packages/pandas/_testing.py", line 740, in _check_types
    assert_class_equal(left, right, exact=exact, obj=obj)
  File "/nvme/0/pgali/envs/cudfdev/lib/python3.7/site-packages/pandas/_testing.py", line 868, in assert_class_equal
    raise_assert_detail(obj, msg, repr_class(left), repr_class(right))
  File "/nvme/0/pgali/envs/cudfdev/lib/python3.7/site-packages/pandas/_testing.py", line 1073, in raise_assert_detail
    raise AssertionError(msg)
AssertionError: DataFrame.index are different

DataFrame.index classes are not equivalent
[left]:  Index([], dtype='object')
[right]: RangeIndex(start=0, stop=0, step=1)

In 1.1.5:

>>> import pandas as pd
>>> one = pd.Index([], dtype='object')
>>> two = pd.RangeIndex(start=0, stop=0, step=1)
>>> df_one = pd.DataFrame(index=one)
>>> df_two = pd.DataFrame(index=two)
>>> pd.testing.assert_frame_equal(df_one, df_two, check_like=True)

Problem description

Upto 1.1.5, both df_one and df_two were considered to be same dataframe when check_like=True, but in latest pandas this is broken.

Expected Output

>>> pd.testing.assert_frame_equal(df_one, df_two, check_like=True)

Output of pd.show_versions()

INSTALLED VERSIONS

commit : 9d598a5
python : 3.7.9.final.0
python-bits : 64
OS : Linux
OS-release : 4.15.0-76-generic
Version : #86-Ubuntu SMP Fri Jan 17 17:24:28 UTC 2020
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.2.1
numpy : 1.19.5
pytz : 2021.1
dateutil : 2.8.1
pip : 21.0.1
setuptools : 49.6.0.post20210108
Cython : 0.29.21
pytest : 6.2.2
hypothesis : 6.1.1
sphinx : 3.4.3
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 : 0.8.5
fastparquet : None
gcsfs : None
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 1.0.1
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : 0.52.0

[phofl]

This was caused by b867e21

[simonjayhawkins]

This was caused by b867e21

#37479 cc @amilbourne

[simonjayhawkins]

can confirm

first bad commit: [b867e21] ENH: Fix output of assert_frame_equal if indexes differ and check_like=True (#37479)

[amilbourne]

Apologies if I have broken something here.
I can see that the behaviour described above was changed by my PR, but I didn't think that ignoring index types was supposed to be part of what check_like=True did. The docs say:

If True, ignore the order of index & columns.
Note: index labels must match their respective rows
(same as in columns) - same labels must be with the same data.

I haven't actually checked this to confirm that it works yet, but if you want to ignore index type, could you use check_index_type=False?

[amilbourne]

Sorry it has taken me so long to come back on this, but I have just verified that my suggestion above (check_index_type=False) does indeed fix the problem:
In 1.2.1:

>>> import pandas as pd
>>> one = pd.Index([], dtype='object')
>>> two = pd.RangeIndex(start=0, stop=0, step=1)
>>> df_one = pd.DataFrame(index=one)
>>> df_two = pd.DataFrame(index=two)
>>> pd.testing.assert_frame_equal(df_one, df_two, check_like=True, check_index_type=False)

Obviously others may differ, but it seems to me that the current behaviour in 1.2.1 is correct and the previous behaviour in 1.1.5 was accidental. Therefore this is not a bug.

[rhshadrach]

Thanks for digging into this @amilbourne, I agree with your assessment.

Do @galipremsagar @phofl @simonjayhawkins agree?

[phofl]

Agree

[simonjayhawkins]

to ensure the behavior does not change again, we should have a test to catch this.

[phofl]

put something up

[amilbourne]

Phew!
Glad I didn't break anything.
@phofl - Thanks for the test