As far as I understood ZoneInfo in python 3.9 fixes the issue of "The year of 2038" problem with DST time issue. But I can see that the timedelta is wrong. Below it should have been 2 hours, not 3. What is wrong?
from zoneinfo import ZoneInfo import datetime dt = datetime.datetime(2038,3,14,0,0, tzinfo = ZoneInfo(r'America/New_York')) print(dt) dt2 = datetime.datetime(2038,3,14,3,0, tzinfo = ZoneInfo(r'America/New_York')) print(dt2-dt) 
from zoneinfo import ZoneInfo import datetime dt = datetime.datetime(2038,3,14,0,0, tzinfo = ZoneInfo(r'America/New_York')) print(dt) # Output: 2038-03-14 00:00:00-05:00 dt2 = datetime.datetime(2038,3,14,3,0, tzinfo = ZoneInfo(r'America/New_York')) print(dt2) # Output: 2038-03-14 03:00:00-04:00 timedelta = dt2-dt1 print(timedelta) # Output: 3:00:00 The above is correct because it is made aware of the DST, aka in the DST aware context. For example, if you add 3 hours to 2038-03-14 00:00:00-05:00 like the code below, you get a consistent result:
from zoneinfo import ZoneInfo import datetime dt = datetime.datetime(2038,3,14,0,0, tzinfo = ZoneInfo(r'America/New_York')) print(dt) # Output: 2038-03-14 00:00:00-05:00 print(dt+datetime.timedelta(hours=3)) # Output: 2038-03-14 03:00:00-04:00 However, your thinking of 2 hours time difference is correct in the NO-DST-AWARE context, as shown below:
from zoneinfo import ZoneInfo import datetime dt = datetime.datetime(2038,3,14,0,0, tzinfo = ZoneInfo(r'America/New_York')) print(dt) # Output: 2038-03-14 00:00:00-05:00 dt2 = datetime.datetime(2038,3,14,3,0, tzinfo = ZoneInfo(r'America/New_York')) print(dt2) # Output: 2038-03-14 03:00:00-04:00 dt_isoformat = dt.isoformat() print(dt_isoformat) # Output: 2038-03-14T00:00:00-05:00 dt2_isoformat = dt2.isoformat() print(dt2_isoformat) # Output: 2038-03-14T03:00:00-04:00 print(datetime.datetime.fromisoformat(dt2_isoformat)-datetime.datetime.fromisoformat(dt_isoformat)) # Output: 2:00:00 
I think It’s 3 hours, from 00 (which is 12) to 3 that’s absolutely 3 hours.
timedeltas in Python show wall time difference if you're working with aware datetime. Between your example datetimes, there is a wall time difference of three hours although absolute time difference is only two hours (DST transition). More explanation can be found in this blog article by Paul Ganssle. Btw. this is not specific to the 2k38 problem (completely different story, see int32...)print(dt2.astimezone(ZoneInfo('UTC')) - dt.astimezone(ZoneInfo('UTC')))(most clear / readable option I think).