Re: Predictable Internet Time

[Joe Touch <[e-mail]>]

Hi, all,

Please take this discussion to ART ([e-mail])

Joe


On 3/28/2017 10:39 AM, Nico Williams wrote:
> On Tue, Jan 03, 2017 at 06:35:03PM +0000, Stewart Bryant wrote:
>> Yes, the system should use a leap-second-free constant-duration-seconds time
>> for everything. It is only humans that need the variable jumpy version of
>> time presented to them, and that is a UI issue.
> I don't think that's quite right.
>
> Unix time is canonically a count of seconds since an epoch and admits no
> leap seconds.  A conversion of Unix time to adjusted (i.e., with leap
> seconds added) time since the same epoch would be nice, as well as:
>
>  - conversion from either to broken down time
>     - with seconds count up to 60 when converting from seconds-with-leaps
>  - formatting of dates from either as well as from broken down time
>     - but when converting from seconds-with-leaps the formatted string
>       can show seconds-in-minute values of 60
>     - whereas when converting from Unix seconds the formatted string can
>       never show seconds-in-minute values of 60
>  - parsing to broken down time and to seconds-since-epoch types
>     - ignoring leap seconds when converting to Unix time
>     - correctly accounting for all leap seconds prior to that time when
>       converting to seconds-with-leaps-since-epoch
>
> It's all about data typing.  In POSIX C API terms, we need new functions
> to deal with seconds-with-leaps-since-epoch, and obviously a lot of
> prayer because there's no way in C to distinguish one integer type from
> the other (well, the new type could be wrapped in a struct..).
>
> We will have to deal with multiple kinds of time as long as different
> applications/users need different definitions of time.  Astronomers may
> need corrected time, while bankers may want to ignore leap seconds --
> all on the same systems (same OSes, same code libraries).
>
> Functionally this is about data typing -- preferably strong data typing.
>
> Smearing is a third time representation that also needs conversion to
> the other types (more opportunity for bugs!); it leaves a bad taste.  It
> may work for some apps though, but all nodes had better smear in exactly
> the same way -- within your tolerances -- if your app depends on time
> for synchronization.
>
> Nico