BUG: Inconsistency in converting Series to Dataframe when columns specified

[artemyk]

The dtypes don't match:

In [16]: pd.DataFrame(pd.Series([], dtype="int64"), columns=["D"]).dtypes
Out[16]: 
D    object
dtype: object

In [17]: pd.DataFrame(pd.Series([], dtype="int64")).dtypes
Out[17]: 
0    int64
dtype: object

This came up in #9983 .

When this is fixed, the check_dtype flag should also be removed in TestDataFrame.test_mode in pandas/tests/test_frame.py .

[TomAugspurger]

Maybe related: #10006

Haven't looked closely but they sound similar. I can't remember where I was at on that PR

[artemyk]

@TomAugspurger I think you meant a different PR ;-)

[TomAugspurger]

😆 #9237 is the one

[artemyk]

#9237 looks relevant indeed. Perhaps this case could also be tested/fixed in there?

[TomAugspurger]

Probably. Does your PR in #9983 depend on this being fixed?

[artemyk]

I would say not. In #9983 we are adding an extra check_dtype=False in pandas/tests/test_frame.py to get around this issue, but that can be removed in the commit for #9237 . @jreback -- thoughts?

[jreback]

no this is independent

[jreback]

actually this is not a bug; #9237 maybe will address the 'issue'. But this is correct. Passing columns reindexes the frame; since the Series is is not named (well it actually is de-facto named 0), it is effectively removed, and thus the dtype is object which is correct.

Pls remove the comment in the original issue w.r.t. to this.

[jreback]

sorry meant #9237