Re: [boost] How to represent underspecified Date in boost::date_time library?

From: Jeff Garland (jeff_at_[hidden])
Date: 2007-07-11 15:11:51

On Wed, 11 Jul 2007 13:34:22 -0500, Martin Min wrote
> Jeff:
> Thanks for your reply.
> First, by "You can use gregorian::greg_year which is the first 
> parameter to a date", do you mean a date can be constructed by just 
> passing greg_year? or still I have to provide a month and day?

What I mean is you can just use greg_year to represent the year instead of the
date class.  However, given your requirements below it probably won't be of
much help.

>  I am working in  Natural  Language  Understanding area,  in which
> underspecifed Temporal Expressions abound in typical news text, and
> reasoning on these type of knowledge often has to deal with 
> incomplete information.  The granularity (or resolution) of time 
> often changes in a discourse, sometimes year, or month, or date, or 
> time, even century.

I see.

> What I like most boost::date_time are the various arithematic 
> operations on durations and date periods, and timezone support. Of 
> course, as well as the robustness.
> 
> Probably, what I can do is to define another Date class, which have the
> boost::date_time::date has a member, and add members I may need?

Possibly.  It seems like you might need something a bit different -- a
multi-resolution time point.  Or you need to have series of different time
points with different resolutions.  For example you could have something like
this:

  boost::gregorian::date; //time point with 1 day resolution
  class month_year;       //time point with 1 month resolution
  class quarter;          //time point with 3 month resolution
  class year;             //time point with 1 year resolution
  class decade;           //time point with 10 year resolution
  class century;          //time point with 100 year resolution

So, simple usage would look something like:

  month_year my(2007, Jul);
  quarter q(2007, 2);
  year y(2007)
  decade d(21, 1); //21st century, 1st decade
  ....

Of course the alternative is to have a single type that provides the something
like this:

   //pseudo code
   class general_time_point 
   {
     enum TIMEPOINT_RESOLUTION { DAY, MONTH, YEAR, QUARTER, DECADE, CENTURY}
     general_time_point(year, month, day);
     general_time_point(year, month);
     general_time_point(value, TIMEPOINT_RESOLUTION); //general constructor
   };

  
So if you had one of these 2 kinds of types, what would you like to do with
them?  There a conceptual issues w.r.t the meaning of arithmetic operations on
these kinds of types.  For example:
 
    decade c(21, 1);
    c += days(1);  //meaningless 
    year y(2007)
    y += years(2); //ok

The advantage of the multi-type solution is that the meaningless operations
can be prevented at compile time.  However, the multi-type solution is likely
more difficult in your particular case.

Jeff