Re: [boost] range, multi-dimensional

From: Phil Endecott (spam_from_boost_dev_at_[hidden])
Date: 2007-04-13 12:46:49

Neal Becker wrote:
> I have been using boost::range heavily, and find it useful for generic
> interfaces.  I think it is also useful to have a multi-dimensional
> extension to range.  In particular, a 2-d extension would help in creating
> algorithms that can accept a variety of 2-d structures.
>
> For example, range has size, begin, end.
>
> range2d would have size1, size2, begin1, begin2.
>
> Other useful members might include row, col.

Yes, I have been thinking along the same lines.  In fact I thought 
about this earlier today after Janek Kozicki posted asking what the 
focus of the proposed SoC geometry library is.  I would very much like 
to see some 2D containers, e.g. set<point2d> and map<point2d,T>, 
implemented using R-trees or similar.  For Boost, I presume that any 
such code will be expected to be "standard library like", and I was 
wondering how similar a set<point2d> can be to a set<int>.  In 
particular, when I want to iterate over all the points within a 
rectangle, I would love to be able to write:

   for(set<point2d>::const_iterator i = points.lower_bound(point2d(x0,y0));
       i != points.upper_bound(point2d(x1,y1)); ++i)

Unfortunately it can't be quite that similar to the 1D case.  Currently 
I have some very crude code where I define what I call a "window" into 
the set of points:

   window w(points,point(x0,y0),point(x1,y1));
   for(set<point2d>::const_iterator i = w.begin();
       i != w.end(); ++i)

Of course, having seen Neal's message, I realise that my "window" is 
just a 2D range.

Clearly, it would be a good plan to get basics like this (and things 
like a point type) thought about before Huseyin Akcan starts on his SoC 
work.  (Huseyin, I hope you're reading this!  Do please let us know 
what exactly you intend to do.)

I have been considering writing something to implement set<point> using 
space filling curves.  I haven't thought about it much yet and would 
value any suggestions.  My hope is that using this technique I can 
adapt the standard associative containers to work with 2D points, and 
thereby avoid having to implement any tree stuff.

In my case I'm dealing with latitude/longitude values and I think that 
the first thing I need is a fixed-point type, so that I can do the bit 
interleaving that the space filling curves need.  That could make a 
simple self-contained Boost library - is anyone interested?

Phil.