[LEAPSECS] more poison in the NTP pool

Paul Sheer p at 2038bug.com
Tue Sep 4 12:19:17 EDT 2012

> >

> > previously i proposed that any code that measures a fail-timeout make

> > that timeout 1000ms longer whenever t % 86400 is near zero


> I assume that you mean near 0 or near 86400 since % is unsigned.



code example posted previously:


> > this pattern would duck most bugs


> With timeouts, yes. With other things, not so much...


as a thought-experiment, i conjecture that it would have prevented all
the system failures reported in the media in the last few leap seconds

am i close?

> > these bogus advertisements seem like another good reason to implement

> > this


> It says a lot about the robustness of leap seconds...




More information about the LEAPSECS mailing list