DataFrame.__init__(..., dtype=dt) makes unnecessary copies

[njsmith]

If I have a DataFrame df with homogenous type, then DataFrame(df) returns a view on the original dataframe. DataFrame(df, dtype=current_type) should be identical; but, instead, it makes an unnecessary copy.

>>> import pandas
>>> pandas.__version__
'0.8.0'
>>> df = pandas.DataFrame([[1, 2]])
>>> df
   0  1
0  1  2
>>> df[0].dtype
dtype('int64')
>>> view = pandas.DataFrame(df)
>>> view
   0  1
0  1  2
>>> view[0][0] = 100
>>> view
     0  1
0  100  2
>>> df
     0  1
0  100  2
>>> should_be_view = pandas.DataFrame(df, dtype=df[0].dtype)
>>> should_be_view
     0  1
0  100  2
>>> should_be_view[0][0] = 99
>>> should_be_view
    0  1
0  99  2
>>> df
     0  1
0  100  2

The same thing seems to happen in the input is an ndarray -- DataFrame(arr) returns a view, DataFrame(arr, dtype=arr.dtype) returns a copy.

[wesm]

Thanks for the report. fixed this