BUG: Copying an empty DataFrame with index results in a broken BlockManager

[filmor]

The following line illustrates the problem:

df = pd.DataFrame(index=[0]).copy()
df["a"] = 0

This results in an AttributeError since the function BlockManager.make_empty initialises df._data.blocks with a numpy array object instead of an empty list. Thus subsequent calls to blocks.pop and blocks.append fail.

I have a patch coming up.

[filmor]

Maybe not, the patch is at filmor@2c9f7d3, but a test fails (https://travis-ci.org/filmor/pandas/jobs/16920606). What is the rationale behind having a dtype for the BlockManager?

[jreback]

@filmor thanks for reporting......BlockManager does not have a dtype. BUT SingleBlockManager provides the property to refer to the dtype of its single block (its the implementation for a series)