When Arrow appeared in the Python ecosystem like three years ago or so, it had the noble purpose of fixing the mess datetimes and timezones can be in Python. Here is its official description:
Arrow is a Python library that offers a sensible, human-friendly approach to creating, manipulating, formatting and converting dates, times, and timestamps. It implements and updates the datetime type, plugging gaps in functionality, and provides an intelligent module API that supports many common creation scenarios. Simply put, it helps you work with dates and times with fewer imports and a lot less code.
But noble principles do not always result in well executed code and API.
Unintuitive API
The API of Arrow is all but intuitive. By taking inspiration in the famous requests library, which has a pretty well crafted API, it became the opposite. Basically, to create an Arrow object, you use the get() method which accepts a lot of different types and it will try to guess what you wanted to return an instance (sometimes erroneous). And if get() makes sense in the context of requests (after all you are executing a GET request) if does not make sense here.
Let's see some official examples:
>>> arrow.get(1367900664)
# <Arrow [2013-05-07T04:24:24+00:00]>
>>> arrow.get('1367900664')
# <Arrow [2013-05-07T04:24:24+00:00]>
>>> arrow.get(1367900664.152325)
# <Arrow [2013-05-07T04:24:24.152325+00:00]>
>>> arrow.get('1367900664.152325')
# <Arrow [2013-05-07T04:24:24.152325+00:00]>
>>> arrow.get(datetime.utcnow())
# <Arrow [2013-05-07T04:24:24.152325+00:00]>
>>> arrow.get(datetime.now(), 'US/Pacific')
# <Arrow [2013-05-06T21:24:32.736373-07:00]>
>>> from dateutil import tz
>>> arrow.get(datetime.now(), tz.gettz('US/Pacific'))
# <Arrow [2013-05-06T21:24:41.129262-07:00]>
>>> arrow.get(datetime.now(tz.gettz('US/Pacific')))
# <Arrow [2013-05-06T21:24:49.552236-07:00]>
>>> arrow.get('2013-05-05 12:30:45', 'YYYY-MM-DD HH:mm:ss')
# <Arrow [2013-05-05T12:30:45+00:00]>
>>> arrow.get('2013-09-30T15:34:00.000-07:00')
# <Arrow [2013-09-30T15:34:00-07:00]>
>>> arrow.get(2013, 5, 5)
# <Arrow [2013-05-05T00:00:00+00:00]>
That's a lot. For newcomers it might seem practical, but it only encourages bad practices.
Another example of bad API design is the replace() method. The native method basically replaces some or all elements of a datetime object but with Arrow it's been overridden to also shift time. Singular units replaces them while plural units shift time.
>>> arw = arrow.utcnow()
>>> arw
# <Arrow [2013-05-12T03:29:35.334214+00:00]>
>>> arw.replace(hour=4, minute=40)
# <Arrow [2013-05-12T04:40:35.334214+00:00]>
>>> arw.replace(weeks=+3)
# <Arrow [2013-06-02T03:29:35.334214+00:00]>
Its confusing at best and prone to errors.
Bugs and strange behavior
But even more than bad API designs, Arrow is buggy and can behave unexpectedly.
Let's see some examples.
Parsing
>>> arrow.get('2016-1-17')
# <Arrow [2016-01-01T00:00:00+00:00]>
# Parsing of a date with wrong day
>>> arrow.get('2015-06-31')
# <Arrow [2015-06-01T00:00:00+00:00]>
Instantiation
# Wrong offset is set on instantation
>>> arrow.Arrow(1970, 1, 1, tzinfo=pytz.timezone('Europe/Paris'))
# <Arrow [1970-01-01T00:00:00+00:09]>
>>> arrow.Arrow.fromtimestamp(0, pytz.timezone('Europe/Paris'))
# <Arrow [1970-01-01T01:00:00+00:09]>
Timezones
Timezones and DST transitions handling is where arrow is particularly broken.
# Working with DST
>>> just_before = arrow.Arrow(2013, 3, 31, 1, 59, 59, 999999, 'Europe/Paris')
>>> just_after = just_before.replace(microseconds=1)
'2013-03-31T02:00:00+02:00'
# Should be 2013-03-31T03:00:00+02:00
>>> (just_after.to('utc') - just_before.to('utc')).total_seconds()
-3599.999999
# Should be 1e-06
in_utc = arrow.get('2016-10-30 00:00')
# <Arrow [2016-10-30T00:00:00+00:00]>
in_amsterdam = in_utc.to('Europe/Amsterdam')
# <Arrow [2016-10-30T02:00:00+01:00]>
# Should be 2016-10-30T02:00:00+02:00
With pendulum these cases are handled correctly
>>> pendulum.parse('2016-1-17')
# <Pendulum [2016-01-17T00:00:00+00:00]>
>>> pendulum.parse('2015-06-31')
# ValueError: day is out of range for month
>>> pendulum.Pendulum(1970, 1, 1, tzinfo=pytz.timezone('Europe/Paris'))
# <Pendulum [1970-01-01T00:00:00+01:00]>
>>> pendulum.fromtimestamp(0, pytz.timezone('Europe/Paris'))
# <Pendulum [1970-01-01T01:00:00+01:00]>
>>> just_before = pendulum.create(2013, 3, 31, 1, 59, 59, 999999, 'Europe/Paris')
>>> just_after = just_before.add(microseconds=1)
# <Pendulum [2013-03-31T03:00:00+02:00]>
>>> (just_after.in_timezone('utc') - just_before.in_timezone('utc')).total_seconds()
# 1e-06
>>> in_utc = pendulum.parse('2016-10-30 00:00')
# <Pendulum [2016-10-30T00:00:00+00:00]>
>>> in_amsterdam = in_utc.in_tz('Europe/Amsterdam')
# <Pendulum [2016-10-30T02:00:00+02:00]>
Arrow seems no longer maintained
As of the writing of this post, it's been four months since the last commit on the github project.
Issues are piling up, some critical, and pull requests are left without responses or comments.
So, it's safe to say that it is pretty much abandoned at this point.
Conclusion
Stop using Arrow.
There are alternatives to it, maintained and more complete, that you can use.
I want pendulum to be the most intuitive, complete and accurate library for Python but there are others you can choose.
But choose wisely.
