BUG: don't allow an empty dataframe to have scalar assignment succeed (GH5744)

[jreback]

closes #5744
related #5720

I think this is the correct behavior

In [1]: df = pd.DataFrame(columns=['a', 'b', 'c c'])

In [2]: df['d'] = 3
ValueError: Cannot set a frame with no defined index and a value that cannot be converted to a Series

In [3]: df['c c']
Out[3]: Series([], name: c c, dtype: object)

[socheon]

Can we just allow the scalar assignment for an empty dataframe?
It can be useful to add new columns dynamically.
A common use case for me is to read a data file using read_csv and then add some derived columns.
Checking that the read dataframe is not empty everytime before adding a new column could be tedious.

[jreback]

@socheon

in general working with a dataframe that does not have an index is very odd to begin with
if you have an index, then their is no problem

you can read_csv and then add a column with a scalar no problem

if you have an example of something you think ought to work, pls put it up

[socheon]

@jreback

At work, I receive daily many third party csv files (e.g. new orders etc) which I need to parse, add some derived columns, and then load into database. Occasionally, some of the files may be empty and just having the column names. For example

Date,ProductID,Quantity,UnitCost
01282013,1,10,3
01282013,2,5,6

To load into database, I simply use

from datetime import datetime
df = pd.read_csv('data.csv', dtype={'Date': 'object'})
df.Date = df.Date.map(lambda s: datetime.strptime(s, '%m%d%Y'))
df['TotalCost'] = df.Quantity * df.UnitCost
load_into_database(df)

But on some days when there are no orders i.e. the input file is just

Date,ProductID,Quantity,UnitCost

The program would break when adding new columns to empty dataframe. Of course, I could always
check if len(df) > 0 before adding new columns. But somehow, I would still need to add the TotalCost column to the empty dataframe before passing it to the function load_into_database.

Anyway, the intended change will not a problem for me. Probably this example is too specific and has easy workaround.

[jreback]

@socheon

I hear ya!

just trying to guard against have an operation which looks like it succeeds, but actually doesn't do anything.

but I see that in 0.12 that was ok....

let me see...maybe I will revert back..

[jreback]

@socheon alright...give a try with master again...and let me know (original behavior is restored)

[socheon]

Thanks. I tried again and the master is good.