On Apr 1, 2004, at 2:23 AM, Brent Pinkney wrote:

> This in turn removes the key monotonic functionality where
>
> 	(DateAndTime now) < (DateAndTime now)
>
> This is key to any application that needs sorted timestamps.
>
> I provided a version of 5839accurateDateAndTimeV2-avi as 
> accurateDateAndTime-brp which keeps Avi's fixes but preserves the 
> monotonicty.
>
> The only reason 5867DateTimeFactoredPrint-nk passes all the unit tests 
> is that is changes SchedultTest>>testMonotonicty so that is not longer 
> tests for strict monotonicty.
>
> I sent an email to Ned about this yesterday. I propably should have 
> used the list.
>
> It is a bit cheaky to change the specification of a kernel class like 
> this !

The only specification I know of for the DateAndTime class is the ANSI 
specification, and it certainly doesn't require "monotonicity" as you 
define it.  When we last discussed this on the list, there were a 
couple of voices opposed to that behavior and only yours in favor.  I'm 
happy to discuss it further but I think it's rather "cheeky" to 
consider it a specification set in stone without any community 
consensus on the matter.  As you point out, this is a fairly core class 
and affects most of us one way or another.

Brent, it's clear that you have strong opinions about the proper 
semantics of a date and time handling library, and that's great; it 
seems, however, that having Chronology in the base image is cramping 
your style, because the needs/opinions of the community as a whole 
don't necessarily match your vision.  Maybe we should pull it back out 
into a separate package, and include the more minimal DateAndTime 
implementation from the ANSI Compatibility package in the base instead?

If it looks like we'll want to do that in the long run, let's do it 
now.  I personally already have an application that uses Timespan 
heavily, and so it will either have to get refactored or depend on 
Chronology, but that just serves me right for using an alpha release... 
:)

Avi