Unless I'm going to do date arithmetic (e.g. 10 days from 20080526), I like yymmdd because I can tell at a glance what the thing is. For date and time I would use 2 integers, yyyymmdd hhmmss, which is the same amount of space as a single floating point number without the tolerance problem.
----- Original Message ----- From: Tracy Harms <[EMAIL PROTECTED]> Date: Friday, May 2, 2008 10:53 Subject: [Jprogramming] RE: [Jgeneral] NUB Problem To: [email protected] > Raul wrote: > > > ... those floating point numbers are times > > packed as YYYYMMDD.hhmmss > > ... > > Basically: convert to a more compact format > > ... > > An even better approach might be to capture > > that original time stamp as character data, > > rather than using #: > > > While I recognize that a textual representation may be > best, there may be several advantages to recording > time numerically. It seems to me that when such an > approach is taken Julian dating should be used. I > think 32-bit representation is adequate to support the > fractions-of-a-day required here, but even if it is > not the origin can be adjusted (by approx. 2.45e6 > days) to provide plenty of fractional accuracy. > > So, my recommendation: Continue to handle temporal > values with floating point, but change to use Julian > dating. ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
