COMPAT: should an empty string match a format (or just be NaT)

[jreback]

after #13033

I left this alone, but should an empty string just be a NaT (like we do elsewhere), even when a formatis present?

In [1]:  td = pd.Series(['May 04', 'Jun 02', ''], index=[1, 2, 3])

In [2]: td
Out[2]: 
1    May 04
2    Jun 02
3          
dtype: object

In [3]: pd.to_datetime(td, format='%b %y', errors='coerce')
Out[3]: 
1   2004-05-01
2   2002-06-01
3          NaT
dtype: datetime64[ns]

In [4]: pd.to_datetime(td, format='%b %y', errors='raise')
ValueError: time data '' does not match format '%b %y' (match)

related is that we don't coerce empty strings here either (and we have an odd error message).

In [1]: pd.to_datetime([1, ''], unit='s', errors='coerce')
Out[1]: DatetimeIndex(['1970-01-01 00:00:01', 'NaT'], dtype='datetime64[ns]', freq=None)

In [2]: pd.to_datetime([1, ''], unit='s')
ValueError: invalid literal for long() with base 10: ''

[jreback]

@sinhrks

[atulagrwl]

This seems to be fixed now. I can see same output in to_datetime whether errors='coerce' or without any errors param

In [1]: import pandas as pd

In [2]: pd.__version__
Out[2]: '0.23.0'

In [3]: pd.to_datetime([1, ''], unit='s', errors='coerce')
Out[3]: DatetimeIndex(['1970-01-01 00:00:01', 'NaT'], dtype='datetime64[ns]', freq=None)

In [4]: pd.to_datetime([1, ''], unit='s')
Out[4]: DatetimeIndex(['1970-01-01 00:00:01', 'NaT'], dtype='datetime64[ns]', freq=None)

@jreback

[fgebhart]

take