Timedeltas cast as datetime in DataFrames

[nbonnotte]

Code Sample, a copy-pastable example if possible

In [10]: pandas.DataFrame([[datetime.datetime(2017, 1, 1)]]) - datetime.datetime(2001, 1, 1)
Out[10]: 
           0
0 1986-01-01

In [11]: pandas.Series([datetime.datetime(2017, 1, 1)]) - datetime.datetime(2001, 1, 1)
Out[11]: 
0   5844 days
dtype: timedelta64[ns]

Problem description

For dataframe, the value "5844 days" is cast as a datetime, so it becomes "1970-01-01 + 5844 days = 1986-01-01"

Expected Output

In [10]: pandas.DataFrame([[datetime.datetime(2017, 1, 1)]]) - datetime.datetime(2001, 1, 1)
Out[10]:
0
0 5844 days

Output of pd.show_versions()

INSTALLED VERSIONS ------------------ commit: 211ecd5 python: 2.7.11.final.0 python-bits: 64 OS: Linux OS-release: 4.4.0-59-generic machine: x86_64 processor: x86_64 byteorder: little LC_ALL: None LANG: en_US.UTF-8 LOCALE: None.None

pandas: 0.19.0+527.g211ecd5
pytest: 3.0.6
pip: 9.0.1
setuptools: 22.0.0
Cython: 0.24
numpy: 1.11.0
scipy: 0.17.1
xarray: None
IPython: 4.2.0
sphinx: None
patsy: None
dateutil: 2.5.3
pytz: 2016.4
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: None
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: None
s3fs: None
pandas_gbq: None
pandas_datareader: None

[chris-b1]

Yep, looks buggy.

[jreback]

yes, but dupe of this: #8554

[jreback]

of course PR's to fix welcome!