Re: [Phys-l] leap second limbo

[John Denker <jsd@av8n.com>]

On 01/20/2012 08:09 AM, Marc "Zeke" Kossover wrote:
> That exists, more or less. http://en.wikipedia.org/wiki/Unix_time

Specifically, the following subsection is the sort of thing I was 
asking about:
  http://en.wikipedia.org/wiki/Unix_time#International_Atomic_Time-based_variant

Alas it fails to cite any specific examples.

Additional information is here
  http://cr.yp.to/proto/utctai.html

which mentions:
> >   Arthur David Olson's popular time library uses an epoch of 1970-01-01 00:00:10 TAI. 

However, I don't see how a runtime library can solve the whole problem.  In
particular, consider a disk containing files with timestamps.  The timestamps
are put there by the kernel, not by some userland library.  We have a dilemma:
 -- If the timestamps are based on UTC-seconds, then in some cases it is
  impossible to tell the age of a file with one-second accuracy, because 
  some timestamps are reused. In particular, in some cases it is impossible
  to reliably determine which of two files is older.
 -- OTOH if the timestamps are based on TAI-seconds, then I cannot mount that
  disk on a POSIX system and expect the timestamps to be interpreted correctly.

In principle (and ?eventually? in practice) this dilemma could be overcome if 
the TAI-based disk were flagged as such, perhaps by a bit in the superblock 
somewhere, but have been unable to find any factual details about TAI-based 
unix.  A wikipedia allusion to "some systems" doesn't cut it.

Some additional information is here:
  http://thedjbway.b0llix.net/clockspeed/leapsecs.html

but it doesn't seem to address the timestamp-on-disk issue.