Column information lost in to_json() if data frame is empty

[aizvorski]

Create empty dataframe with a column and write it out with to_json(), then read it back in:

In [55]: s = pandas.DataFrame({'test': [] }, index=[] ).to_json(orient='columns')
In [56]: s
Out[56]: '{}'
In [57]: str( pandas.read_json(s, orient='columns') )
Out[57]: 'Empty DataFrame\nColumns: []\nIndex: []'

I think the expected string is '{"test":{}}'

This situation arises when code filters data frames and writes them out, while not being very careful to make sure the filtered frames contain at least one row.

[jreback]

cc @Komnomnomnom

[Komnomnomnom]

Thanks @aizvorski I can confirm this bug, and I agree about the expected output.

The JSON handling is producing its output by looping over the dataframe values array, which is why in this case the labels end up being ignored during serialisation.

@jreback I'm on vacation until mid July but I'll see if I can get this fixed for 0.15. There's also the other JSON issues / features I've been meaning to do since pre 0.14...I'll see if I can knock them out at the same time.

[behzadnouri]

closed by #9805

>>> pd.DataFrame({'test':[]}, index=[]).to_json(orient='columns')
'{"test":{}}'