On Mon, 22 Dec 2008, Steve Allen wrote: > > I am aware of the interesting breakages that happened when zoneinfo > files were retroactively modified to be inconsistent with POSIX. > Clearly that change cannot be done for past history.
It can't be done for future history either, because it breaks invariants that are relied on all over the place. > > There is a pervasive assumption in Unix that midnight UT is when > > time_t % 86400 == 0. > > So if the broadcast time scale were to become TI then these operations > would be taking place at "atomic day midnight" instead of mean solar > day midnight. That's still a legitimate kind of day. This isn't about scheduling operations, it's about inter-converting between a count of seconds and broken-down civil time. There are lots of in-kernel functions and standard data formats (network protocols, file systems) that rely on the "trivial UT" assumption of 86400 seconds per day. POSIX time is one example amongst many. > The argument that "atomic time will drift by no more than a few > seconds from mean solar during this century has been used by those who > want to abandon leaps. In the presence of an internationally approved > atomic time scale I think that argument goes both ways. It doesn't > hurt anybody if certain technical operations drift a few seconds from > civil wall clock time. Disseminating a linear timescale instead of UTC doesn't address the problem if civil time continues to be UTC, because all these systems are more tightly coupled to civil time than they are to the timescale of their reference clock(s). Civil time is a pervasive dependency - i.e. lots of software has built-in assumptions about its structure - whereas time synchronization is a relatively self-contained function. Tony. -- f.anthony.n.finch <d...@dotat.at> http://dotat.at/ LUNDY FASTNET IRISH SEA: SOUTH OR SOUTHWEST 4 OR 5, OCCASIONALLY 6 IN IRISH SEA. MODERATE, OCCASIONALLY ROUGH IN LUNDY AND FASTNET. DRIZZLE. MODERATE OR GOOD, OCCASIONALLY POOR. _______________________________________________ LEAPSECS mailing list LEAPSECS@leapsecond.com http://six.pairlist.net/mailman/listinfo/leapsecs