Re: [Geotools-devel] ExentMosaic problem

[Martin D. <mar...@te...>]

Hello

> In general it is more useful to get back rectangles rather than a single 
> shape, since searches based on rectangles are much faster on any 
> practical implementation I know of.

Maybe, but I wonder if we should benchmark it to be sure. For example, 
calling "difference" many time may result in a quick explosion of the 
amount of rectangles. See attached figure for an example. Since Area try 
to do an intelligent work, it is not obvious to me that the Rectangle 
approach would be faster when the amount of Rectangles is big.

If performance is the prime concern, than a more efficient algorithm 
with rectangles would be:

R1 = the orange rectangle
N1 = the first white rectangle
N2 = the second white rectangle
if (R1.contains(P) && !N1.contains(P) && !N2.contains(P))


It would probably be more efficient than

if (R1.contains(P) || R2.contains(P) || R3.contains(P) ||
     R4.contains(P) || R5.contains(P) || R6.contains(P))


If we want to implement the (R1,N1,N2) algorithm instead of the 
(R1,R2,R3,R4,R5,R6) one, then we are going closer to Rob architecture 
with his Filter (AND, OR, NOT operations, etc.). It bring us to the 
point Rob made some time ago, that Extends are just special kind of Filters.

In anycase, my point is that Extend.difference shoud not returns an 
array of Extent. It should returns a single Extent instead. This single 
Extent may be a package-private implementation wrappind an array of 
other Extents, or a combinaison of AND/OR/NOT operations among the line 
of Rob's filters.



> But you are right we do need to revisit the ogc envelope and our extent 
> objects, to see what simplifications we can make. As we develop more 
> extents this interface will mature.


Two questions here:

1) Are Extent always two-dimensional?
2) Are Extent always rectangular in shape?

If the answer is true for 1 and 2, then Extent is basically a 
java.awt.geom.Rectangle2D. If 1 is false and 2 is true, then Extent is 
basically a org.opengis.pt.PT_Envelope. If both 1 and 2 are false, then 
I see no direct equivalence. In this last case, the Extent existence is 
easier to justify.

	Regards,

		Martin.