is_holiday function doesn't return correct result when pandas Timestamp is passed.

[yassineAlouini]

The is_holiday function expects a datetime.datetime format. However, it doesn't raise any error when a pandas Timestamp is passed (https://pandas.pydata.org/pandas-docs/version/0.23.4/generated/pandas.Timestamp.html).

Here is a code example (workalendar version '3.1.1'):

from workalendar import europe
import pandas as pd
import datetime


fr = europe.France()
# Should be True and returns True
print(fr.is_holiday(datetime.datetime(2017, 1, 1)))
# Should be True but returns False
print(fr.is_holiday(pd.to_datetime("2017-1-1")))
# To fix this, use the `to_pydatetime` method
print(fr.is_holiday(pd.to_datetime("2017-1-1").to_pydatetime()))

I expect the library to raise an error when the passed input isn't of type datetime.datetime.
What are your thoughts?
Thanks in advance. :)

[brunobord]

mmmm.... this is sad that padas Timestamp doesn't inherit from the stdlib types...
If there's a way to convert from pandas classes to standard types, that could be doable. Although I wouldn't like to add a dependency to pandas in the project (only in tests it would be fine).

[yassineAlouini]

@brunobord Yes, it is possible using the .to_pydatetime() method:

import pandas as pd
tms = pd.to_datetime('2017-1-1')
print(type(tms))
# pandas._libs.tslibs.timestamps.Timestamp
print(type(tms.to_pydatetime()))
# datetime.datetime

[brunobord]

ok, looks like it's doable.

Then my question would be: why should workalendar accept pandas Timestamps? I've had an afterthought and I'm not sure workalendar would support exotic date/datetime input types.
If we accept pandas timestamps, why not accepting strings like YYYY-MM-DD, and then integers (as a unix timestamp), and then a picture of a calendar as an input? (yes, I'm exaggerating, for the sake of the argument).
I'm not saying that I want to reject this possibility, I'm saying that I think I'd need a serious argument to accept non-standard types in our API.

[yassineAlouini]

I see your point.

I am not advising to accept additional types. Maybe a check of type + raising an exception would be enough to avoid dealing with other formats than datetime.datetime?

What do you think?

[johnberroa]

I had the same problem. Throwing an error would be nice because if I didn't double check I would have never known.

[brunobord]

turns out pandas' Timeout is a datetime subclass. Interesting, we may be able to support more types, as long as they are something like "duck dates" (that quack like a date)

[brunobord]

closed by #298 + released with the version 3.2.0