> I never cease to be amazed that UNIX always seems to know when we go onto or
> off dayloght savings time. How?
Old systems use some timezone description in the TZ envvar. Not very powerful,
and requires cheezy handtuning.
Newer systems use the Olson zoneinfo system where each timezone is described
in a file with the DST rules (and how they change over time: those files had
to be updated when Europe changed its DST rules, but the old rules are still
applied for old times (and newer rules are already applied for future times)),
It's indeed pretty neat.
Quote:> I point out that the time for reverting to winter time here in Europe changed
> about a year ago, yet even our HP workstations seemed to know about it, even
> without an OS update.
The update that brought the new info was probably done "long" before, so you
didn't notice the relationship.
Quote:> Granted, we do use nntpd, yet nntpd refuses to modify the time if the local
I assume you mean "ntpd" because I fail to see why you'd suddenly switch to
discussing NetNews. NTP completely disreagrds the issue of local time: it only
deals with UTC time and nothing else so it's completely immune to any DST
issue (and quite rightly so: how would NTP servers running in different
timezones cooperate sanely else ?).
Quote:> time and that obtained remotely differ wildly; i.e., by more than a minute or
> two. The daemon merely exits quietly, without so much as a murmer (as i
> recently witnessed again after an OS update on one of our workstations, where
> the update obviously was confused about the actual time).
Yes, I've been bitten by this one also. The typical answer is "use ntpdate",
but it doesn't always cut it: if you're disconnected from the rest of the
world at boot time, ntpdate might fail (as well as hang stupidly because
although it is careful to quickly time out when the servers are unavailable, it
just uses the standard resolve library and hence might have to wait for
the DNS request to (oh so slowly) time out).
It really seems that xntpd should have a better failure mode. For instance, if
the local time is too far from UTC, instead of silently failing, it could do
the equivalent of what ntpdate does (and keep trying until it works).
Or it should at least be possible to ask ntpd to exit with an informative error
code, so that we get a chance to do something like
until ntpd; do ntpdate; done
The current behavior is very annoying since it's surprisingly hard to detect