[LEAPSECS] Leap seconds have a larger context than POSIX

[Miroslav Lichvar]

On Wed, Feb 05, 2020 at 03:32:54PM -0800, Tom Van Baak wrote:
> Code should allow for a leap second event at the end of any month. The June
> / December thing is one of several guidelines for BIPM; it's not a rule that
> anyone writing UTC code should assume or depend on. Nor should code ever
> assume leap seconds are positive.

If the source can be trusted, yes, that would make sense. But in
reality, when working with not completely reliable information (due to
bugs in specifications, software, firmware, attacks, etc.), I think it
might be better to have the current June/December practice hardcoded.

For example, when a bug in an NTP implementation allowing leap seconds
to be inserted at any month caused them to sometimes get stuck and
repeat every month, the bogus leap seconds were spreading among
servers that didn't check what month it is.

How much energy is needed to slow down or speed up Earth enough for
UTC to require more than two leap seconds per year and how quickly can
it be released without destroying those that care about UTC following
UT1?

-- 
Miroslav Lichvar