System.Time.Clock Design Issues

[Ashley Yakeley]

In article 
<7DFF3BC6CA957441AEADA7F340BFAA340A029337 at GBLONEX11.lon.invesco.com>,
 "Bayley, Alistair" <Alistair_Bayley at ldn.invesco.com> wrote:

> I should read more...

My original message has a lot of background:
<http://www.haskell.org/pipermail/libraries/2005-January/002908.html>

> So UTC and UT1 seem to be more-or-less equivalent, in that they both include
> leap seconds.

Well, they don't drift apart if that's what you mean. There are no leap 
seconds in UT1, instead UT1 seconds are of variable length. 

> UT1 measures the actual earth day

Yes, with some corrections to even it out.

> (so I assume it has some fractional component for seconds),

No, the UT1 day is divided into exactly 86400 UT1 seconds. The length of 
these seconds varies, just as the length of the UT1 day varies. Although 
you're right if you mean the count of seconds is continuous rather than 
discrete.

> while UTC (which I assume counts integral
> seconds) tracks UT1 with an error of +/-0.9s. UTC "differs from TAI by an
> integral number of seconds".

Yes. UTC days are always an integer number of seconds. Most are 86400, 
some are 86401, and some could potentially be 86399.
 
> TAI ignores leap seconds, but does track time accurately by always ticking
> (i.e. not slowing down at a leap second). There are exactly 86400 seconds in
> a TAI day, so the day will drift w.r.t. the earth's actual day, whereas UTC
> tracks the actual day accurately by inserting leap seconds. So the TAI day
> isn't very useful.

Yes. Instead it's better to consider TAI as a count of seconds.

> POSIX is "idealised" time, where days are always 23:59:59, and a second
> sometimes takes two seconds.

Yes. POSIX is an encoding of UTC, but broken in this way, because it is 
a count of SI seconds that assumes 86400 to the day to make calculation 
easier.

-- 
Ashley Yakeley, Seattle WA