Re: [AD] use double for al_current_time and al_rest?

[Andrei Ellman <ae-a-alleg@xxxxxxxxxx>]

Elias Pschernig wrote:
> Should we change
>
> unsigned long al_current_time(void)
> void al_rest(long msecs)
>
> to
>
> double al_current_time(void)
> void al_rest(double seconds)
>
> and also have them both use seconds?
>
>  [...]
>
> And I don't see any disadvantages right now..

Using a floating point value for timers is not a good idea because once 
the exponent gets too large, the precision of the value to al_rest() for 
decreases, and eventually, the current_time value will become so large 
that the amount to rest() for will underflow. While underflow may not 
happen for a long time, if someone wants to al_rest() for a 
high-precision ammount of time, the actual time rested will be inconsistent.


> The advantages would be:
>
> - No implicit guarantee of providing milliseconds accuracy - I don't
> think we always can do that.
>
> - No limit to milliseconds resolution, when we actually often can do
> much better with all those high precision timers the OS provides these
> days.
>
> - Seconds as unit seems to be the least surprise, instead of arbitrarily
> using one of milli, micro or nano seconds requiring to check docs each
> time.


Why not just used a fixed-point number of seconds?


AE.