Styjun

I'm experiencing strange behavior when attempting to convert between UTC and specific timezones. I'd love for someone to explain why I'm seeing this behavior and what the more "correct" way of getting timezone information might be.

Code:

import pytz
import datetime
from django.utils import timezone

print(timezone.now())
print(pytz.utc.localize(datetime.datetime.now()))
print('n')

def get_local_and_utc_date_ranges(days=1500, days_ago=2, local_timezone="America/Asuncion"):
 seller_timezone = pytz.timezone(local_timezone)
 utc_timezone = pytz.utc

 seller_today = timezone.now().astimezone(seller_timezone)
 seller_days_ago = seller_today - timezone.timedelta(days=days_ago)

 local_date_end = seller_days_ago.replace(hour=23, minute=59, second=59, microsecond=999999)
 local_date_start = (local_date_end - timezone.timedelta(days=days)).replace(hour=0, minute=0, second=0, microsecond=0)

 utc_date_end = local_date_end.astimezone(utc_timezone)
 utc_date_start = local_date_start.astimezone(utc_timezone)

 date_ranges = 
 "local_date_end": local_date_end,
 "local_date_start": local_date_start,
 "utc_date_end": utc_date_end,
 "utc_date_start": utc_date_start,
 

 return date_ranges


def get_utc_and_local_date_ranges(days=1500, days_ago=2, local_timezone='America/Asuncion'):
 seller_timezone = pytz.timezone(local_timezone)
 utc_timezone = pytz.utc

 utc_today = datetime.datetime.utcnow()
 utc_days_ago = utc_today - datetime.timedelta(days=days_ago)

 local_date_end = seller_timezone.localize(utc_days_ago).replace(
 hour=23, minute=59, second=59, microsecond=999999
 )
 local_date_start = (local_date_end - datetime.timedelta(days=days)).replace(
 hour=0, minute=0, second=0, microsecond=0
 )
 utc_date_end = local_date_end.astimezone(utc_timezone)
 utc_date_start = local_date_start.astimezone(utc_timezone)

 date_ranges = 
 'local_date_end': local_date_end,
 'local_date_start': local_date_start,
 'utc_date_end': utc_date_end,
 'utc_date_start': utc_date_start,
 

 return date_ranges



days = 1500
days_ago = 2

dates = get_local_and_utc_date_ranges(days=days, days_ago=days_ago)
dates2 = get_utc_and_local_date_ranges(days=days, days_ago=days_ago)


print('dates1:')
print('local_date_start:', dates['local_date_start'])
print('local_date_end:', dates['local_date_end'])
print('utc_date_start:', dates['utc_date_start'])
print('utc_date_end:', dates['utc_date_end'])
print('n')

print('dates2:')
print('local_date_start:', dates2['local_date_start'])
print('local_date_end:', dates2['local_date_end'])
print('utc_date_start:', dates2['utc_date_start'])
print('utc_date_end:', dates2['utc_date_end'])
print('n')

Output:

2019-03-25 18:57:55.929908+00:00
2019-03-25 18:57:55.930005+00:00


dates1:
local_date_start: 2015-02-12 00:00:00-04:00
local_date_end: 2019-03-23 23:59:59.999999-04:00
utc_date_start: 2015-02-12 04:00:00+00:00
utc_date_end: 2019-03-24 03:59:59.999999+00:00


dates2:
local_date_start: 2015-02-12 00:00:00-03:00
local_date_end: 2019-03-23 23:59:59.999999-03:00
utc_date_start: 2015-02-12 03:00:00+00:00
utc_date_end: 2019-03-24 02:59:59.999999+00:00

Note the inconsistent UTC offset (that particular timezone switched to DST on Mar 23rd). But when I try to replicate the issue using the following code:

import pytz 
import datetime 
from django.utils import timezone 

now1 = timezone.now() - datetime.timedelta(days=2)
now2 = pytz.utc.localize(datetime.datetime.now()) - datetime.timedelta(days=2)

seller_timezone = pytz.timezone('America/Asuncion')

print(now1.astimezone(seller_timezone).replace(
 hour=23, minute=59, second=59, microsecond=999999
 ))
print(now2.astimezone(seller_timezone).replace(
 hour=23, minute=59, second=59, microsecond=999999
 ))

The output is correct:

2019-03-23 23:59:59.999999-03:00
2019-03-23 23:59:59.999999-03:00

I'm hoping someone can understand why this behavior is happening and how I might avoid the inconsistencies if so.

Your get_local_and_utc_date_ranges() function is producing incorrect results because it's doing datetime arithmetic (i.e. subtracting a timedelta) with a localized time, which doesn't work.

seller_today = timezone.now().astimezone(seller_timezone)
seller_days_ago = seller_today - timezone.timedelta(days=days_ago)

This is noted in the datetime module documentation:

As for addition, the result [of subtracting a timedelta] has the same tzinfo attribute as the input datetime, and no time zone adjustments are done even if the input is aware.

This is also noted in the pytz documentation:

If you perform date arithmetic on local times that cross DST boundaries, the result may be in an incorrect timezone.

pytz offers a fix:

A normalize() method is provided to correct this.

So you could use:

seller_days_ago = seller_timezone.normalize(seller_today - timezone.timedelta(days=days_ago))
...
local_date_start = seller_timezone.normalize(local_date_end - timezone.timedelta(days=days)).replace(hour=0, minute=0, second=0, microsecond=0)

However, the documentation also notes that:

The preferred way of dealing with times is to always work in UTC.

So a better solution would be to only do arithmetic in UTC:

utc_today = datetime.datetime.utcnow()
utc_date_end = utc_today - datetime.timedelta(days=days_ago)
utc_date_start = utc_date_end - datetime.timedelta(days=days)

local_date_end = seller_timezone.localize(utc_date_end).replace(hour=23, minute=59, second=59, microsecond=999999)
local_date_start = seller_timezone.localize(utc_date_start).replace(hour=0, minute=0, second=0, microsecond=0)