[LEAPSECS] USWP7A docs for 2013 September meetings

[Warner Losh]

On Aug 19, 2013, at 4:23 PM, Zefram wrote:


> Warner Losh wrote:

>> That is unless the API explicitly says "always hide leap seconds from

>> users in this way"...

> 

> That's basically what it says.  The API is based around a single base

> time scale, which in the present era is UTC-SLS.  The only way leap

> seconds would matter for a computation with JSR-310 would be if you

> converted between its time scale and some other time scale such as TAI.

> No such conversion is available through the API, so that doesn't arise.

> Computations within the time scale, which are the main business of the

> API, are unaffected by leap seconds, because they use the seconds of

> UTC-SLS without caring that some of them are of different lengths when

> looked at in a different way.

> 

> The only place where leap seconds directly affect the operation of

> the API is where you ask a Clock object what the current time is.

> It must reply in UTC-SLS, thus hiding leap seconds in the standard way.

> But implementations aren't obliged to provide an accurate clock.  If they

> don't advertise that a clock is sub-second accurate, then a POSIX-like

> discontinuity around leap seconds is acceptable, just as it's acceptable

> for the clock to give readings that are minutes off.


By catering to the sucky implementation, it precludes the precision one :(

Warner