[LEAPSECS] Look Before You Leap ? The Coming Leap Second and AWS | Hacker News

[Warner Losh]

> On May 21, 2015, at 7:49 AM, Poul-Henning Kamp <phk at phk.freebsd.dk> wrote:
> 
> --------
> In message <20150521134322.GG10661 at ucolick.org>, Steve Allen writes:
> 
>> POSIX does not want to know geophysics, nor astrometry, nor politics.
>> POSIX does not care what is meant by "day".
>> POSIX wants someone else to decide what "day" means, and for all those
>> other details to be handled outside the kernel in the libraries and
>> applications.
> 
> POSIX is not just a kernel API, it *also* defines the functions for
> deciding what "day" means and all that.

POSIX has decided that a day is 86400 seconds. Thus you can’t implement
days with leap seconds with SI seconds. And you can’t have a system
that’s synchronized to UTC if you smear seconds or ‘drop’ that second.
The down stream effects are a bunch of bad choices, which bad choice
you make depends heavily on the application. Some (maybe most) second
smearing is good. Others, where stable elapsed time matters a lot, smeared
seconds are a disaster, but dropping a second is also bad. Having multiple
choices here leads to poor quality of implementation of leap seconds.

To do things moderately close to ‘right’ you have to invent your own
thing, and then translate your ‘right’ thing into the imperfect POSIX
thing and accept those folks that use purely POSIX interfaces can
never have the whole story and will have some aspect of their time
keeping disrupted around leap second events.

The real problem here is that POSIX defines time passing in such an
abstract way that you can’t use UTC to realize it without something
giving.

Warner

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 842 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <https://pairlist6.pair.net/pipermail/leapsecs/attachments/20150521/00e45af4/attachment.pgp>