On Wed, 3 Nov 2021 23:42:33 -0500, Alan Altmark wrote: > >Architecturally, the TOD clock follows TAI, with an epoch (TOD = all zeros) of >Jan 1, 1900. It doesn't care about the 37 leap seconds that have been >introduced into UTC since the epoch.(hence, TAI) If you used a standard >TOD-to-civil time conversion (which all assume TOD is UTC), the result would >be 37 seconds ahead of UTC. If the system is leap second-aware, then it >subtracts the leap seconds when generating civil time. TAI -> UTC -> local >time. > I believe there have been 27 leap seconds. When UTC was established it was set to coincide with GMT which was then (1972?) 10 seconds behind TAI. The PoOps describes this pretty well and lists all 27 leap seconds. Also listed at: <http://ftp.iana.org/tz/tzdb/leap-seconds.list>. Note that the list begins with number 10, not 0.
>When you don't configure leap seconds, the old (and new) algorithms will >generate the correct civil time, but the observant person will discern that >the epoch has moved 37 seconds backward into December 31, 1899. (And with >each subsequent leap second, it moves backward an additional second.) > I believe the epoch of the TOD is 1900-01-01 00:00:10 (proleptic) TAI (somewhat mythical because neither TOD nor TAI existed then.) It's hard to get sign conventions correct. CVTLDTO is added to TOD; CVTLSO (by IETF convention) subtracted/ --gil ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
