Difference between time addition using timestamp vs using timedelta?

Question

I'm working through Exercism.io gigasecond problem:

"Calculate the moment when someone has lived for 10^9 seconds."

My method was to convert the datetime input to a timestamp, add 10**9, then convert back. My answers are very close, but the test suite (provided by Exercism) fails because the hour arg is off by one. All other args are correct.

After a bit of searching, I found an answer that uses timedelta; the structure of the answer is almost identical to mine, but it generates the full correct answer.

My question is, why is there a difference in the output of these two methods? As you can see below, the problem only manifests when the input arg for hour is the default value of 0.

Code:

def add_gigasecond(birth_date):
    gigadate = birth_date.timestamp() + 10**9
    print(datetime.fromtimestamp(gigadate).__repr__())

    gigadate = birth_date + timedelta(seconds=10**9)
    print(gigadate.__repr__())

Inputs:

tests = [datetime(2011, 4, 25),
         datetime(1977, 6, 13),
         datetime(1959, 7, 19),
         datetime(2015, 1, 24, 22, 0, 0),
         datetime(2015, 1, 24, 23, 59, 59),]

Results:

test 1
datetime.datetime(2043, 1, 1, 0, 46, 40)
datetime.datetime(2043, 1, 1, 1, 46, 40)

test 2
datetime.datetime(2009, 2, 19, 0, 46, 40)
datetime.datetime(2009, 2, 19, 1, 46, 40)

test 3
datetime.datetime(1991, 3, 27, 0, 46, 40)
datetime.datetime(1991, 3, 27, 1, 46, 40)

test 4
datetime.datetime(2046, 10, 2, 23, 46, 40)
datetime.datetime(2046, 10, 2, 23, 46, 40)

test 5
datetime.datetime(2046, 10, 3, 1, 46, 39)
datetime.datetime(2046, 10, 3, 1, 46, 39)

As you can see, the top answer in each test (using timestamps) is off by 1, for all inputs where year, month, date are specified. The correct answer is achieved when the hour, minute, second, and microsecond args are included in the input.

Answer 1

Daylight Savings

Some detective work can shed light on these problems. For this investigation, I pick the first date in your list, 25-Apr-2011.

from datetime import datetime, timedelta

def add_seconds(birth_date, n):
    x = datetime.fromtimestamp(birth_date.timestamp() + n)
    y = birth_date + timedelta(seconds=n)
    return (x, y)

for i in range(365):
    ts, td = add_seconds(datetime(2011, 4, 25), i*60**2*24)  # add i days via seconds
    if ts-td:  # test for difference in timestamp vs timedelta methods
        print(i, ts, td)
        break

Result:

189 2011-10-30 23:00:00 2011-10-31 00:00:00

As you can see, there is a discrepancy on exactly 30th October. This is, for many countries, the second date in the year where we see a daylight savings adjustment.

Update: A similar question is asked here: Difference between evaluating timestamp and total_seconds. The underlying reason for the difference is that datetime.timedelta does not take into account DST, while timestamp does take this into account.