ST_ConvexHullAgg #464
ST_ConvexHullAgg #464
Conversation
|
☝️ @robe2 |
|
OK, ran some tests against this, and the performance is terrible (10x worse than
|
| /* Single point is always convex */ | ||
| case POINTTYPE: | ||
| return LW_TRUE; | ||
|
|
There was a problem hiding this comment.
is TRIANGLE convex in context of this function?
| LANGUAGE 'c' IMMUTABLE STRICT _PARALLEL; | ||
|
|
||
| -- Availability: 3.0.0 | ||
| CREATE AGGREGATE ST_ConvexHullAgg (geometry) ( |
There was a problem hiding this comment.
Can the
final func be "ST_ConvexHull(geometry[])",
stype = internal (LWGEOM* geometrycollection),
sfunc = try{ append(LWGEOM) } except Out of Memory { ST_ConvexHull(LWGEOM), append(LWGEOM) },
serialfunc -> geometry_serialize(ST_ConvexHull(LWGEOM)),
deserialfunc -> gserialized parse?
This should make it catch up with ST_ConvexHull(ST_Collect()) speed.
|
Should we give up on this as hopeless? Only real reason for having it is if it could be faster than ST_ConvexHull(ST_Collect(..)). If it can't then the extra keystrokes is not worth the addition of a new agg. |
|
I'm closing this one out as I think we decided it's hopeless |
As described in https://trac.osgeo.org/postgis/ticket/3770,
Take in a set of geometry and return convex full of full set.