ST_ConvexHullAgg

[robe]

I've been thinking that we really need a ConvexHull agg.

Most of the cases where people use it they do stuff like

SELECT ST_ConvexHull(ST_Collect(geom)) 
....

or

SELECT ST_ConvexHull(ST_Union(geom)) 
..

Which when you think of it, seems inefficient to store all those bits of geometries just to squash them.

I'd have to do some tests, but I'm suspecting following how we do ST_Extent should be more efficient since any convexhull you compute for 2 geometries in a set is always going to be a subset or equal to what you compute for 3 geometries.

So we could also parallelize this since

the combine would be something like ST_ConvexHull(collect(ST_ConvexHull(worker1_geoms),ST_ConvexHull(worker2_geoms),ST_ConvexHull(worker3_geoms)))

Let me know if I'm missing anything here.