On Fri, 23 Feb 2018 07:24:36 -0800, Charles Mills wrote:
>There are two more or less redundant fields in the CVT: CVTTZ and CVTLDTO.
>(CVTLDTO is actually in CVT extension 2.) CVTLDTO contains the exact local
>time offset in timer units -- in other words, you can add it directly to an
>STCK value. CVTTZ is identical to the high order word (only) of CVTLDTO -- in
>other words, it is the *approximate* offset in units of 1.048576 seconds.
>(Approximate because you cannot represent an integral number of hours exactly
>in units of 1.048576 seconds.) Why 1.048576? One way of looking at that value
>is that it is 2^20 divided by 10^6. It is the ratio between mibi and mega.
>
An IBM Redbook has called this a megamicrosecond. Ugh! Humpty Dumpty!
>Are there possible timing windows? Yes, all over the place. (Between a STCK
>and inspecting CVTLDTO? Yes, a small window. What about someone else's STCK
>that is somehow passed to you and your inspecting CVTLDTO? Potentially a
>window of seconds, days or even years.) I have come to the conclusion that
>local time is something that it is impossible to deal with perfectly. You deal
>with it as well as you can.
>
It can be done a lot better than x/OS does it, and with no timing hazard:
https://en.wikipedia.org/wiki/Tz_database
Dismayingly, provincially, z/OS shuns the Olson database. NIH!?
-- gil
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
