[LEAPSECS] What's the point?

[Warner Losh]

On 02/08/2011 16:30, Rob Seaman wrote:

>> Even the olson database won't give you all the answers, but it will give you many of them.

> But you guys continue to reject Steve Allen's zoneinfo option...which represents a system layered on a relatively static timezone DB.  Punting to local governments is vastly more complex.


The basic problem with that approach is that you need to update your 
timezone files for every leap second, and you can never, ever, miss an 
update, or you are hozed.  NTP also does everything in UTC time, so if 
you come up with one set of data, then get updated timezone info that 
tells you that you missed a leap second, the underlying clock running in 
TAI time will suddenly be a second off.  This can be mitigated somewhat 
if you have a working GPS receiver and can afford to wait the 30 minute 
it may take to get the almanac data to startup, but there are still a 
large number of situations where you don't know until 'later' that 
you've made the wrong guess at startup and now everything you've based 
on a TAI or TAI-like time in your application is off by 1 or more seconds.

At least some versions of the timezone code also don't re-read the 
zonefiles if they change (stating these files every time operation is 
prohibitively expensive).  This means long-running control programs will 
have systematic errors depending on when they were started relative to 
the timezone files.  And having the program itself restart also has issues.

If time were always broadcast in TAI, and if the number of leapseconds 
was always available, then these issues would go away.  But it isn't.

As the number of these issues you try to code around grows, the 
complexity reaches a point where you say "geeze, this was a stupid idea."

Warner