Pandas version checks

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

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

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

Reproducible Example

importpandasaspdimportnumpyasnpleft_data=np.random.rand(1000,5)left_index=pd.date_range(start='20240101 09:00:00',periods=1000,freq='min')left_columns=pd.MultiIndex.from_tuples([    ('price','A'),# Tuple[str, str]    ('price','B'),# Tuple[str, str]    ('price','C'),# Tuple[str, str]    ('diff', ('high','low')),# Tuple[str, Tuple[str, str]]    ('diff', ('open','close'))# Tuple[str, Tuple[str, str]]])left_df=pd.DataFrame(data=left_data,index=left_index,columns=left_columns)right_data=np.random.rand(990,3)right_index=pd.date_range(start='20240101 12:00:00',periods=990,freq='min')right_columns=pd.MultiIndex.from_tuples([    ('X',1),    ('X',2),    ('X',3),])right_df=pd.DataFrame(data=right_data,columns=right_columns,index=right_index)df1=pd.concat([left_df,right_df],axis=1,sort=False)print(df1)# df2 = pd.merge(left_df, right_df, left_index=True, right_index=True)# print(df2)# df3 = left_df.join(right_df)# print(df3)

Issue Description

Let's say we have two dataframes,left_df andright_df both of which have multilevel columns, or columns of typepandas.MultiIndex. The columns are of different types, in particular, one of the dataframes has a column which is of typeTuple[str, Tuple[str, str]] and the other has a column of typeTuple[str, int]. My goal is to concatenate these two dataframes along the columns. To this, I tried out usingpd.concat with theaxis=1 argument and experimented around with some other arguments as well.

Further experiments

1. To avoid this warning, I tried to usepd.DataFrame.join andpd.merge as well but they all return the same warning (given in the code)
2. I tried out the same thing with aRangeIndex instead of aDatetimeIndex and I get the same error. The error doesn't seem to depend on the indices of the dataframes.
3. I tried the same thing with single-level columns, or usingleft_columns=pd.Index([('A', 'A'), 'B', 'high', 'low']) andright_columns=pd.Index([('X', 'X'), 1, 'a']) to see if this happens only when we have multilevel columns.I do not get any warnings in this case, which confirms that this is an issue only withMultiIndex columns and not regular columns.
4. I tried various combinations of types in theMultiIndex columns and this issue doesn't arise when columns are just of typeTuple[str, str] andTuple[str, int] orTuple[str, Tuple[str, str]]. It might happen with other pairs of data types as well, I haven't checked all combinations.

Expected Behavior

It is hard to understand why is any kind of a sorting operations performed in this case. There should not be any instance of comparison. There is some common low level function being called in all 3 ofconcat,merge andjoin which is comparing the column values. It is expected that no warning be thrown in this case. To reproduce the output one gets (handling the correct type of join) without a warning, the code which can be run is:

new_index=pd.MultiIndex.from_tuples(np.concatenate([left_df.keys(),right_df.keys()]))df=pd.concat([left_df,right_df],axis=1,sort=False,ignore_index=True).reindex(new_index,axis=1)

Source

Installed Versions