On Mon, Jan 13, 2003 at 01:38:36PM -0500, Rich Bowen wrote: > On Mon, 13 Jan 2003, fglock wrote: > > > About "Rata Die" and "Julian Day": I'd prefer a seconds-based > > implementation, because leap seconds would make 'seconds' be a > > varying fraction, in a day-based calendar. > > Seems to me that there is a difference between representing a particular > date/time and attempting to represent accurately the number of seconds > since time began (wherever you are measuring from). The former being > more reasonably possible. If you have a day, and 90001 seconds to work > with, you can accurately represent the time of a particular event even > for days with leap seconds, and an extra hour, for any historical event. > > I'm unclear how this is any less accurate than having a strictly seconds > based method, and it means that we can store a wider range of dates in a > smaller number.
I'm not sure how you reach the conclusion that using seperate storage for jd and jseconds will get you a wider range of dates in a smaller number. a day has 86400 seconds. you need 17 bits for this. This leaves the other 47 bits of a TAI label for the date. how can you get a wider range of days in jd, which is smaller than a tai label so, assuming commonly used integer sizes, 32 bits? Martijn
