[LEAPSECS] tinkering with time ?

[Stephen Colebourne]

On 1 February 2011 06:45, Rob Seaman <seaman at noao.edu> wrote:

> Successful programming environments model the behavior of the real world, not artificial constructs.

>

> Perhaps some progress could be made on the "predictable leap second scheduling" front?  How would Java (or any software systems) gracefully accommodate a published schedule with a long lead time?


The JSR-310 implementation is supplied with a list of known
leap-seconds. I hope to move to picking that up as part of picking up
the latest zoneinfo, so it would be as up to date as the zoneinfo
file.

Today, there is no leap second data, but zoneinfo data requires
running a script at the low JDK level and a JVM reboot. This is a
tricky system level operation that operators dislike. Note that the
leapsecs data would also need to be updated in the OS for that to
return an accurate time.

With a finished 310, the leap second data is moved into the zoneinfo
files, requiring only a new jar file to be present. This would just
require a system reboot.

However, 310 is written to allow continual updating of this data from
a network source of zoneinfo/leap data. That won't be my work, but if
it happens, then distributing leapsecs data won't be a problem for
many JVMs.

Stephen