Re: [boost] [boost-users] [date_time] length of generic periods

From: Jeff Garland (jeff_at_[hidden])
Date: 2005-07-21 23:01:21

On Thu, 21 Jul 2005 22:28:08 +0200, Friedrich Wilckens wrote
> Hi,
> 
> the concept of the length of generic periods puzzles me. The "normal"
> case p.length() = p.end() - p.begin() makes sense to me, but why is a
> period with last() == begin() treated as a special case, having 
> length 0?

Long ago in a mailing list somewhere the issue of null intervals appeared and
that's where this logic came from...

> Having this special case make the length concept behave strangely. 
> For example, [2005-May-1/2005-May-1] has length 0, 

Well, that seems incorrect [2005-May-1/2005-May-1) has length zero (Note the
non-inclusive end).  But, then again, what is the duration of [1,1] -- zero I
believe.

>but [2005-May- 1/2005-May-2] has length 2! 

Hmm.

> Actually there is no period of length 1 at all.

Oh my...that doesn't seem right at all....

> As I understand from the documentation, the philosophical reason behind
> this seems to be that the end points of periods are considered "points"
> and that a point and therefore a period that consists of a single point
> has zero extension.
>
> This sounds plausible if you think of the periods like intervals on the
> real number line. But our situation is different, I think, since we
> always deal with discrete quantities that have a finite length (or
> "duration").

Actually, I don't think it is different from an 'integer numberline'.

> I would interpret the period [2005-May-1/2005-May-1] as the time period
> that lasts during the whole day 2005-May-1, from 00:00 up to
> 23:59:59.999... Therefore, I would give it the length (duration) 1.

Yes, that seems correct except that a date-period cannot represent anything
less than a single day.  Just as integers in a range [1,1] cannot represent
1.99999.

> [2005-May-1/2005-Apr-30] would accordingly be an empty period of length
> 0, and [2005-May-1/2005-Apr-29] would be an invalid period of length 
> -1.
> 
> The length of a date_period p would then be simply the number of 
> days in p, multiplied with the unit of the duration_type. The 
> template parameter duration_type can be used to defined different 
> measurements of the length. In the standard case (duration of a day 
> is 1) you measure the period length in days. But you could also 
> measure it in seconds, by using a duration_type with unit() = 86400.
> 
> What do you think about this?

I'm worried that there is a problem here, but I'm going to need some time to
look at it.  I won't be able to for a couple days...

Jeff