[postgis-devel] GeomUnion speedups

[strk at refractions.net]

A summary of my recent works in GeomUnion speedup.

Martin Davis suggested me that a buffer(0) of a collection
of input geometries has the same result then an incremental
union of them.

I tested this and we *really* get a big performance
improvement this way, as graphs for geometry components
are built only once!

You can test this easily using buffer(collect(the_geom), 0)
in place of geomunion(the_geom).

If there are no drawbacks in this I'd have geomunion directly
do this for us. The drawback I immediately see is that
while geomunion does not accept GEOMETRYCOLLECTION in input
this method does allow them. Should we see this a bug ? (we
can perform the check from within GeomUnion to avoid this).

Martin, are there other drawbacks ?

About memory we might implement a chunk-based incremental
approach whereas blocks of geometries (defined by memory
size maybe ?) would be buffer(0)'d to early dissolve 
vertexes and reduce memory occupation. Thoughts on this ?

--strk;