[Python-ideas] Date/time literals

[Alexander Belopolsky]

On Tue, Jun 1, 2010 at 1:44 PM, MRAB <python at mrabarnett.plus.com> wrote:
..
>> but
>>>>>
>>>>> datetime(1985, 6, 30, 23, 59, 60) - datetime(1985, 7, 1, 0, 0, 0)
>>
>> datetime.timedelta(0)
>>
> Actually, that's wrong because there was a leap second. The clock went:
>
>    1985-06-30 23:59:59
>    1985-06-30 23:59:60
>    1985-07-01 00:00:00
>
> The following year, however, it went:
>
>    1986-06-30 23:59:59
>    1986-07-01 00:00:00

It is only wrong if you expect datetime difference to reflect the
actual duration between the corresponding UTC events.  The datetime
library does not do it even for dates.

For example, on my system

>>> date(1752, 9, 14) - date(1752, 9, 2)
datetime.timedelta(12)

even though calendar application on the same machine shows that
September 14 was the day after September 2 in 1752.

$ cal 9 1752
   September 1752
Su Mo Tu We Th Fr Sa
       1  2 14 15 16
17 18 19 20 21 22 23
24 25 26 27 28 29 30

This was a deliberate design choice to implement proleptic calendar
rather than a historically more accurate variant.  Similarly I see
nothing wrong with datetime difference not capturing leap seconds.  An
application interested in leap seconds effects, however should still
be able to use the basic datetime object and define its own duration
functions.