In my opinion, computers would *MOSTLY* be better served by a clock that
executes the sequence described by the gregorian calendar exactly.
There have been recent proposals to just stop doing leap seconds
entirely and allow midnight to drift a minute over the next few decades.
My proposal was for a geometric time standard in which one 'geometric
second' is defined as the time required for Earth to rotate 1/86400 of a
full revolution relative to the sun. The exact time would be determined
with a telescope (or gyroscope of sufficient precision) observing the
angle between the surface of the earth and the stars. Which is, you
know, exactly how a second was determined for most of the time we've
been using seconds in the first place. It accepts that seconds are not
all exactly the same length, but no seconds differ by more than 1 part
in 15 million.
The obvious benefits include the clock never jumping and never having
more or less than 60 seconds in a minute. Which would eliminate clock
complexity as the cause of most of the basic errors that screw up
business and utility applications. My understanding of the matter is
that people who need a time standard in which all the seconds are the
same length, to a measurement more precise than 1 in 15 million, are
relatively rare, know exactly why they need such a standard, are willing
to sweat the extra complexity required, and will select a different time
standard as the basic system time in their computers. But for personal
and business purposes, including even most of the highly sophisticated
programming business purposes such as AI training clusters, geometric
time would be a better and simpler choice, less prone to the kind of
errors people actually care about.
Alas, the proposal was voted down. There's a particular mindset that
just can't abide the idea that seconds might not all be exactly the same
length. For anyone, anywhere, even to make life and ordinary
programming easier and less bug-prone.
Bear
Bear
On 10/13/22 10:31, Arthur A. Gleckler wrote:
On Wed, Oct 12, 2022 at 11:01 PM Marc Nieper-Wißkirchen
<marc.nie...@gmail.com> wrote:
You can add 86,400 UTC "seconds" per day if you want UTC dates.
That was the hack I was considering, but I thought that assuming
86,400 per day, even for UTC, was incorrect.
