BUG: duplicate shifted holiday in California

[joooeey]

from workalendar.usa import California

cal = California()

cal.holidays(2019)
Out[13]: 
[(datetime.date(2019, 1, 1), 'New year'),
 (datetime.date(2019, 1, 21), 'Birthday of Martin Luther King, Jr.'),
 (datetime.date(2019, 2, 18), "Washington's Birthday"),
 (datetime.date(2019, 3, 31), 'Cesar Chavez Day'),
 (datetime.date(2019, 4, 1), 'Cesar Chavez Day (Observed)'),  # <===============
 (datetime.date(2019, 4, 1), 'Cesar Chavez Day (Observed)'),  # <===============
 (datetime.date(2019, 5, 27), 'Memorial Day'),
 (datetime.date(2019, 7, 4), 'Independence Day'),
 (datetime.date(2019, 9, 2), 'Labor Day'),
 (datetime.date(2019, 11, 11), 'Veterans Day'),
 (datetime.date(2019, 11, 28), 'Thanksgiving Day'),
 (datetime.date(2019, 11, 29), 'Thanksgiving Friday'),
 (datetime.date(2019, 12, 25), 'Christmas Day')]

Note that Cesar Chavez Day (Observed) is duplicated.

This does not occur for 2017, 2018, 2020, and 2021. I've just tested those five years. The same issue occurs with 2019 in the CaliforniaEducation calendar.

I noticed this issue because it breaks a version of my test for PR #527.

[brunobord]

thank you for the bug report, very "interesting" bug. I'll get into it as soon as I can.

[joooeey]

I just looked into this. The shift day is computed twice (and differently) in workalendar/usa/core.py. Saturday is shifted to Friday and Sunday to Monday in def shift and if the holiday falls on a Sunday it is shifted to Monday in def get_cesar_chavez_days.

Here's some info on how state holidays in California work. It says about the shift holidays:

*When a holiday falls on a Saturday, employees shall receive holiday credit.
**When a holiday falls on a Sunday, the holiday is observed on the following Monday

I could not find any source stating that Cesar Chavez Day is shifted any differently than other holidays.

Hence, I think this bug should be solved by not doing any special shifting for Cesar Chavez Day, i.e. have a def get_cesar_chavez_day function just like all the other holiday functions that returns a holiday on March 31 and no extra shift day.

However I noticed two other, separate issues:

• As per the link above, California state holidays falling on a Saturday are not actually shifted to Friday but result in holiday credit instead. In usa/core.py we shift them to Friday.
• Per Wikipedia Cesar Chavez Day is a state holiday in a few states, in workalendar we only have it in California. So we're inconsistent. Also, it doesn't seem to be a holiday for most workers in California.

I'm not gonna touch these two bullet points as this looks like a job for an American.

[brunobord]

Seems like Cesar Chavez day is an optional holiday in Colorado. That might be the reason why, at the time, I introduced it as a "core" holiday, with an "include_" flag to allow integrators to activate it or not depending on their context.

I need to dig a bit further on the duplicates you've found, though.

[brunobord]

ok, I think I have a lead, here.

Effectively, Chavez Day seems to be an exception when it comes to weekend-shifting.
But if your calendar includes the Cesar Chavez Day and if you're not in California, it may follow the same pattern as other holidays in other states (Shift to FRI if it happens on SAT, and shift to MON if it happens on SUN).

I think it's possible to have a satisfying procedure that would keep the usual behaviour and take the Californian exception the way it should be.

[brunobord]

Bug fixed, you can use the latest (10.4.0) release:

• PyPI: https://pypi.org/project/workalendar/10.4.0/
• Release notes: https://github.com/peopledoc/workalendar/releases/tag/10.4.0

I'm still working my way on the iCal export feature, soon to be ready.

[joooeey]

Effectively, Chavez Day seems to be an exception when it comes to weekend-shifting.

It would be sweet to document the source for this statement. To me it looked like this was an implementation mistake. I couldn't find any official info that shifting works differently for Cesar Chavez Day.