Re: Double 59; 60; or double 00?

From: M. Warner Losh <imp_at_BSDIMP.COM>
Date: Wed, 04 Jan 2006 16:29:11 -0700 (MST)

In message: <004701c61173$f7721d40$9515f204_at_computer>
            "Tom Van Baak" <tvb_at_LEAPSECOND.COM> writes:
: With the surge of leap second captures this time
: around, are there any concerns over the growing(?)
: use of double :59 second or double :00 second
: instead of :59:60 for a positive leap second?
:
: Although not technically correct, they do seem a
: practical, perhaps even clever, alternative -- in some
: cases -- to the upstream parsing trouble that the
: exceptional :59:60 causes.
:
: Given a choice I would vote for double :59 over
: double :00 since :59 is still in the previous day
: (where the leap second occurs). Also, :59 offers
: more symmetry between positive and negative
: leap seconds (58 and 00 are constant; 59 is
: double or nothing depending on leap type).

ntp implements the double 59, and that works well when one can't
represent the number as :60 (which is impossible to do in POSIX
time_t).

POSIX, interestingly enough, doesn't define what happens at the
leapsecond very well. On the one hand, they officially don't exist.
On the other, if you convert 23:59:60 to a time_t using mktime, you
get 0:00:00 due to normalization that happens in mktime (a more or
less accidental effect). This leads some to conclude that the proper
POSIX sequence is with two 0's in a row. While this is a conclusion
from reading POSIX, it isn't the only conclusion.

Warner
Received on Wed Jan 04 2006 - 15:30:10 PST