[postgis-devel] GeomUnion Performance Info

[strk at refractions.net]

I thought this would be of interest for the GeomUnion performance thread.

In particular, last question suggest to implement the disjoint-bbox
short-circuit in JTS directly.

--strk;

----- Forwarded message from Michaël Michaud <michael.michaud at free.fr> -----

Date: Tue, 21 Jun 2005 23:57:57 +0200
From: Michaël Michaud <michael.michaud at free.fr>
Subject: [jts-devel] Graph computation in JTS
Reply-To: JTS Topology Suite Development <jts-devel at lists.jump-project.org>
To: jts-devel at lists.jump-project.org
X-BeenThere: jts-devel at lists.jump-project.org

Hi,

Some slow operations in JUMP are operations involving union on big 
geometries (or iterative unions over a big set of geometries). I don't 
say JUMP is slower than... (this operation is slow with all the GIS I know).
I tried to figure out which part of the code may be involved (union --> 
OverlayOp ? --> PlanarGraph ?), but the graph part of JTS sounds a 
little bit complex to me and I have not a clear idea of how a planar 
graph is computed.
Here are three questions related to graph computation in JTS :

- is there a document presenting how a graph is modelized and how a 
graph is computed in JTS ?

- some classes are repeated in several packages (ex. PlanarGraph). Is it 
a programmer's choice, or does it need to be cleaned ? When one add JUMP 
packages and JCS packages, it may be quite difficult to find the good 
class for the good job !

- while computing a graph involving big geometries (for a union, an 
overlay, or any graph-based operation), are there any short-circuit to 
avoid computing segment intersections when geometry envelopes are 
completely disjoined, or when envelope1 and candidate segment of 
geometry 2 are completely disjoint ? (where in the code may I find 
answers to these questions ?)

Sorry for just speaking about what should be better
JTS is really a great library and I love it

Michael
_______________________________________________
jts-devel mailing list
jts-devel at lists.jump-project.org
http://lists.refractions.net/mailman/listinfo/jts-devel

----- End forwarded message -----