Re: [Ntp] Leap second draft

[Daniel Franke <[e-mail]>]

On Sat, Mar 30, 2019 at 10:57 AM Kurt Roeckx <[e-mail]> wrote:
> In some other mail you wrote:
> > However, I do note that the current draft implements only part of what
> > I recommended, and I think it would be further improved by doing the
> > rest of it as well. To reiterate what I recommended, in place of the
> > current LEAP tag, provide the following three tags:
> >
> > MRLS - a TAI timestamp representing the time at which the
> > most-recently published leap second ends. This may be in either the
> > future or the past.
> >
> > LSDI - A bit indicating whether the leap second given in MRLS is an
> > insertion or a deletion.
> >
> > UTCO - What the TAI-UTC offset was or will be as of the time given
> > in MRLS.
>
> So I'm wondering why this MRLS is not part of this draft and that
> ELI is 2 bits instead of the 1 bit LSDI.
>
> Using that combination would at least avoid the comments about it
> being possible to have a leap second every 1 or 6 months.

The MRLS/LSDI representation I recommended before is better and easier
to work with when you're using TAI timestamps everywhere, and when
that's the case you have a better world overall. So for Roughtime
where the existing deployed base is negligible so we can pretty much
do what we want, that was my recommendation. But NTP uses (pseudo-)UTC
timestamps and I don't want to go back and mess with the semantics of
existing fields. This draft is what I see as the cleanest way to
retrofit better leap second handling onto the existing protocol and
existing buggy implementations of it.