On 05/02/2012 09:14 PM, Even Rouault wrote:
I've also read your note in the RFC about performance. It is interesting to see that envelope computation is the limiting factor when the method layer is in memory. It is an interesting track for a later optimization where envelope would be cached instead of being computed each time. Spatial indexing would also do the trick. It reminds me that a few years ago I've added a cpl_quad_tree.cpp implementation that is a direct transposition of the shapelib implementation, but not tied with shapefiles. It has never been seriously used AFAIR, but it could perhaps be a starting point to do in-memory spatial indexing. Even
I'll prepare a new patch later to address the issues I clipped away from this reply.
About the performance. I believe that at least in my test case a pre-computation of the envelope of the whole method layer (the effective part in fact, as there was an attribute filter) could speed up the computation into half and I was intending to test that. I'll take a look at the quad tree code too - it would be useful when the method layer has more effective features.
The performance is anyway an internal issues and the main thing is to get the API and logic acceptable.
Ari _______________________________________________ gdal-dev mailing list [email protected] http://lists.osgeo.org/mailman/listinfo/gdal-dev
