[LEAPSECS] internet drafts about zoneinfo

[Warner Losh]

On 03/06/2011 18:55, Tony Finch wrote:

> On 5 Mar 2011, at 16:27, Joe Gwinn<joegwinn at comcast.net>  wrote:

>> At 2:27 PM +0000 3/5/11, Tony Finch wrote:

>>> On 4 Mar 2011, at 22:13, Joe Gwinn<joegwinn at comcast.net>  wrote:

>>>> As I have said many times, in theory POSIX time is a form of TAI, being a constant offset from TAI,

>>> Wrong in every respect.

>> Could you be more specific?

> There is no "in theory" since POSIX is a practical specification. The specification says time_t is based on UTC. POSIX time has never been a constant offset from TAI. What has been said many times is that it might be nice to change POSIX time to be based on TAI but if you try it you find that this breaks most timekeeping code.


POSIX time_t is a lossy mapping function from UTC second labels 
(complete with :60) to a counter.  Leap second ambiguously map to some 
time_t value that is the same as another second (they have to map to 
something during the leapsecond, if nothing else).  This mapping is both 
lossy (because you can't undo it unambiguously) and ambiguous (since the 
standard insists that leap seconds don't exist).

This makes it more like a drunken TAI that has 'episodes' every 18 
months or so where things are a little unstead and wind up a little 
further away from TAI after each one...

Warner