datedelta 1.0.1

Usage

Install datedelta with:

pip install datedelta

Then you can add a datedelta to a date:

>>> import datetime
>>> import datedelta

>>> datetime.date(2016, 3, 23) + datedelta.datedelta(years=1, months=1, days=-1)
datetime.date(2017, 4, 22)

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(years=2)
datetime.date(2018, 3, 1)

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(years=2, days=-1)
datetime.date(2018, 2, 28)

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(years=2, months=6)
datetime.date(2018, 9, 1)

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(years=4)
datetime.date(2020, 2, 29)

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(years=4, days=1)
datetime.date(2020, 3, 1)

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(years=4, months=6)
datetime.date(2020, 8, 29)

You can also subtract a datedelta from a date:

>>> import datetime
>>> import datedelta

>>> datetime.date(2016, 3, 23) - datedelta.datedelta(years=-1, months=-1, days=1)
datetime.date(2017, 4, 22)

>>> datetime.date(2020, 2, 29) - datedelta.datedelta(years=2)
datetime.date(2018, 3, 1)

>>> datetime.date(2020, 2, 29) - datedelta.datedelta(years=2, days=1)
datetime.date(2018, 2, 28)

>>> datetime.date(2020, 2, 29) - datedelta.datedelta(years=2, months=-6)
datetime.date(2018, 9, 1)

>>> datetime.date(2020, 2, 29) - datedelta.datedelta(years=4)
datetime.date(2016, 2, 29)

>>> datetime.date(2020, 2, 29) - datedelta.datedelta(years=4, days=-1)
datetime.date(2016, 3, 1)

>>> datetime.date(2020, 2, 29) - datedelta.datedelta(years=4, months=-6)
datetime.date(2016, 8, 29)

These results may appear slightly surprising. However, they’re consistent, for reasons explained below.

Behavior

There are two date arithmetic traps in the Gregorian calendar:

1. Leap years. Problems arise when adding years to a February 29th gives a result in a non-leap year.

2. Variable number of days in months. Problems arise when adding months to a 29th, 30th or 31st gives a result in a month where that day doesn’t exist.

In both cases, the result must be changed to the first day of the next month.

Provided periods are represented by [start date inclusive, end date exclusive), this method gives consistent results. (This representation of periods is akin to 0-based indexing, which is the convention Python uses.)

For example, if someone subscribes for a year starting on 2016-02-29 inclusive, the end date must be 2017-03-01 exclusive. If it was 2016-02-28 exclusive, the subscription would be one day too short.

Operations are always performed on years, then months, then days. This order usually provides the expected behavior. It also minimizes loss of precision.

Limitations

Additions involving datedelta are neither associative not commutative in general.

Here are two examples where adding a datedelta then subtracting it doesn’t return the original value:

>>> import datetime
>>> import datedelta

>>> datetime.date(2020, 2, 29) + datedelta.datedelta(years=1)
datetime.date(2021, 3, 1)

>>> datetime.date(2021, 3, 1) - datedelta.datedelta(years=1)
datetime.date(2020, 3, 1)

>>> datetime.date(2020, 1, 31) + datedelta.datedelta(months=1)
datetime.date(2020, 3, 1)

>>> datetime.date(2020, 3, 1) - datedelta.datedelta(months=1)
datetime.date(2020, 2, 1)

Here are two examples where adding two datedelta gives a different result depending on the order of operations:

>>> import datetime
>>> import datedelta

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(months=6) + datedelta.datedelta(years=1)
datetime.date(2017, 8, 29)

>>> datetime.date(2016, 2, 29) + datedelta.datedelta(years=1) + datedelta.datedelta(months=6)
datetime.date(2017, 9, 1)

>>> datetime.date(2016, 1, 31) + datedelta.datedelta(months=2) + datedelta.datedelta(months=5)
datetime.date(2016, 8, 31)

>>> datetime.date(2016, 1, 31) + datedelta.datedelta(months=5) + datedelta.datedelta(months=2)
datetime.date(2016, 9, 1)

To avoid problems, you should always start from the same reference date and add a singe timedelta. Don’t chain additions or subtractions.

To minimize the risk of incorrect results, datedelta only implements operations that have unambiguous semantics:

• Adding a datedelta to a date

• Subtracting a datedelta from a date

• Adding a datedelta to a datedelta when components have the same sign

• Subtracting a datedelta from a datedelta when components have opposite signs

(PEP 20 says: “In the face of ambiguity, refuse the temptation to guess.”)

Alternatives

datedelta.datedelta is smarter than datetime.timedelta because it knows about years and months in addition to days.

datedelta.datedelta provides a subset of the features found in dateutil.relativedelta. Not only does it only support dates, but:

• It omits the “replace” behavior which is very error-prone.

• It doesn’t allow explicit control of leapdays.

• It uses keyword-only arguments.

• It requires Python 3.

Handling leap days automatically reduces the number of choices the programmer must make and thus the number of errors they can make.

Note that datedelta.datedelta adjusts non-existing days to the first day of the next month while dateutil.relativedelta adjusts them to the last day of the current month.

If you’re stuck with Python 2, just copy the code, make datedelta inherit from object, and remove the * in the signature of __init__.

If you’re comfortable with dateutil and don’t mind its larger footprint, there’s little to gain by switching to datedelta.

Changelog

1.0

• Initial stable release.