[AFS3-std] Re: A call for consensus on draft-deason-afs3-type-time-02

[Steve Simmons]

On Aug 1, 2011, at 5:30 PM, Russ Allbery wrote:

> . . . One of the interesting questions is whether conversion of
> the first and last calendar times above are, when represented as
> timestamps, one second apart or two seconds apart.
>=20
> Note that, in POSIX, they're one second apart, because POSIX time =
contains
> no leap seconds by definition (which means that it's not really =
possible
> to accurately represent those dates).
>=20
> On UNIX systems, using mktime on those dates will generally convert as
> follows:
>=20
>    1972-06-30 23:59:59         78821999
>    1972-06-30 23:59:60         78822000
>    1972-07-01 00:00:00         78822000
>=20
> There's not really a "right" answer here; we just need to say what the
> answer is.

The last paragraph says it, both w/r/t the POSIX issues above and the =
various other time issues discussed in this thread. We don't need to be =
absolutely right*, nor do we necessarily need to be fully in agreement =
with any other vendor or standards system. We just need to say what we =
are doing.

I proposed we be broken in the same way POSIX is broken, rather than =
being any other brand of broken or inventing our own. :-)

Steve

* IMHO 'absolutely right' would involved knowing what is right, and =
presenting to all the other broken implementations their expected broken =
data. The former is beyond our scope and abilities, the latter . . . =
well, that way lies madness. Let's face it, 99.99% of the time all we're =
talking about are filesystem timestamps. If someone wants to touch a =
file with the date Sept 10 1752 and not specify anything else about how =
to interpret that date, they deserve the wrongness they get. And anybody =
who's doing work where leap seconds and calendar type transitions matter =
is not doing it using POSIX, Windows, or any other OS-time =
representation.**

** No, I have no freaking idea what they're using.=