pd.NaT.date() returns datetime.date(1, 255, 255)

[s-celles]

Hello,

pd.NaT.date() returns datetime.date(1, 255, 255)
that's an odd results
I was expecting None (or something else) because that's really strange to have
date of year "1" in an Excel file

Kind regards

[shoyer]

Could we make NaT.date() just raise a ValueError? I think I would prefer that over None, which would usually just mean differing the error until later.

[s-celles]

Yes raising a ValueError is more understandable.

see http://stackoverflow.com/questions/18212247/convert-text-dates-to-dates-and-then-keeping-the-na-values

[shoyer]

If you'd like to put together a PR to fix this, that would be great!

[s-celles]

Yes but where is defined Not A Time.
What kind of behaviour should we have ? Does getting any attribute / applying any method should raise an error ?

[shoyer]

NaT is defined by the NaTType class here: https://github.com/pydata/pandas/blob/bceb342d76ca4061fc8d3a1089bc4e83a32da656/pandas/tslib.pyx#L568

NaTType is an indirect subclass of datetime, so it inherits the default date method. You simply need to supply the method on NaTType, e.g.,

def date(self):
    raise ValueError('some helpful message')

(and yes, this is Cython, but you can write pure Python code here.)

This will also need a test to verify that the error is raised, probably somewhere around here:
https://github.com/pydata/pandas/blob/bceb342d76ca4061fc8d3a1089bc4e83a32da656/pandas/tseries/tests/test_tslib.py#L589

[shoyer]

If there are other methods that are not returning appropriate values, feel free to override them to raise an error as well. But most of them should be supplying appropriate returns values already (e.g., nan for NaT.year).

[s-celles]

I directly edited code on GitHub and send PR this way (because I don't have a Python install here)... I just hope that Travis CI will be ok.

[s-celles]

Sorry but it seems that my PR fails with CI but I try locally it should works. Any idea ?

[s-celles]

An other PR #9523