Re: Ada.Calendar and NTP (and Unix Epoch)

[erlo.haugen@gmail.com]

> Where is a problem then?
> 
>    E1 + T1 vs. E2 + T2
> 
> <=>
> 
>    T1 vs. (E2 - E1) + T2
> 
> Where E2 - E1 is the time offset between the epochs.

That's true, the issue here is that I was fooled by Ada.Calendar.Conversion spec
into thinking that I can use To_Ada_Time to convert the NTP epoch base (1-1-1900 0:00) to a variable of type Time. This would make it easy to calculate
the offset between the two epochs:
epoch_diff := To_Ada_Time(70, 0, 1, 0, 0 ,0 0) - To_Ada_Time(00, 0, 1, 0, 0 ,0 0)

Snip from Ada.Calendar.Conversion spec:

  function To_Ada_Time
     (tm_year  : Interfaces.C.int;
      tm_mon   : Interfaces.C.int;
      tm_day   : Interfaces.C.int;
      tm_hour  : Interfaces.C.int;
      tm_min   : Interfaces.C.int;
      tm_sec   : Interfaces.C.int;
      tm_isdst : Interfaces.C.int) return Time;
   --  Convert a time value expressed in Unix-like fields of struct tm into
   --  a Time value relative to the Ada Epoch. The ranges of the formals are
   --  as follows:

   --     tm_year   --  years since 1900 *this one fooled me*
   --     tm_mon    --  months since January [0 .. 11]

I had the idea that zero would be a legal value for tm_year...

> 
> This presumes that they represent *same* time though maybe using different
> epochs. You might need to read the UNIX and NTP documentation to learn what
> exactly times they do represent. That would give you E2 - E1. To put it
> simple: same time = synchronized clocks.
> 

This is the exact case.

> Note that if any of two times is political, E2 - E1 does not exist as it
> becomes a function of time. This where all starts really complicated and is
> the reason why one should never ever mess up with the political time for
> the purpose of time stamping. And this is just a relatively simple case
> when clocks are synchronized but one of them is periodically skewed.
> 
> When clocks are not synchronized, time stamps are fundamentally
> incomparable without the history of synchronization, which, well, means
> that they were rather synchronized. (:-))
> 
> -- 
> Regards,
> Dmitry A. Kazakov
> http://www.dmitry-kazakov.de

Erlo