Travis CI: issue with python 3.5 and doctest

[garawalid]

Travis CI with python 3.5 returns different results than expected in doctest.
Here is an example :

        >>> df = ks.DataFrame({'name': ['Raphael', 'Donatello'],
        ...                    'mask': ['red', 'purple'],
        ...                    'weapon': ['sai', 'bo staff']})
        >>> df.to_csv(index=False)
Expected:
    'name,mask,weapon\nRaphael,red,sai\nDonatello,purple,bo staff\n'
Got:
    'mask,name,weapon\nred,Raphael,sai\npurple,Donatello,bo staff\n'

I check doctest with Pandas 0.23.4 and python 3.6 locally and everything works well.

We had this issue in these PRs: #239 #238 #267

[HyukjinKwon]

I think this is an issue inherited from Pandas. You can use ... for now

[arisehkawamura]

Hope this helps

Are dictionaries ordered in Python 3.6+? - Stack Overflow

[HyukjinKwon]

Thanks, either way looks not specific to this library.

[ueshin]

We can't leave the flakiness.

I guess we can fix by explicitly reordering columns after df is created, or avoiding using dict.

df = ks.DataFrame({ ... })
df = df[['name', 'mask', 'weapon']]

or

df = ks.DataFrame([['Raphael', 'red', 'sal'],
                   ['Donatello', 'purple', 'bo staff']],
                  columns=['name', 'mask', 'weapon'])

Let me fix the existing tests.

[ueshin]

Seems columns=[...] works:

df = ks.DataFrame({ ... }, columns=['name', 'mask', 'weapon'])

[garawalid]

I agree columns=[...] works. Still, some cases where we have to skip them by # doctest: +SKIP.
Example DataFrame.to_json #238