[OpenLayers-Dev] Project this morning: Canvas Renderer

[Christopher Schmidt]

On Sun, Apr 13, 2008 at 04:20:12PM +0200, Eric Lemoine wrote:
> Hi Chris,
> 
> Thanks for this work and early release. One comment below.
> 
> On Sat, Apr 12, 2008 at 5:54 PM, Christopher Schmidt
> <crschmidt at metacarta.com> wrote:
> >  It implements feature selection by looping over its internal list of
> >  features, and doing intersection queries.
> 
> Given the potential performance problem this method brings, do we
> still plan to generalize its use to solve the "feature selection with
> multiple vector layers" issue? Maybe the performance could be made
> better using some geo index algorithm. Opinions on that?

My hope is that:
 * Our current SoC proposal for work in this area is accepted
 * Our SoC student finds a magical solution :)

More seriously, I think that the use case of "whole world in vectors"
and the use case of "I've got 3 point layers with 20 points each" are
radically different: the former is not just hundreds, but thousands, of
times more expensive.

A point-based intersection test with 150 features was able to execute
in only a handful of milliseconds, meaning that it could likely be a
realistic choice for people who need that kind of behavior. 

It was always my understanding that doing intersection-based selections
would be more expensive than determining the feature from the event
directly: the performance lost here didn't surprise me, but it is my
hope that there is some improvement that can be done to improve the
peformance, and that we can make use of SoC to do real investigation
into answering the questions related to it.

Regards,
-- 
Christopher Schmidt
MetaCarta