Enhancement to PIP short circuit

[kneufeld]

The PIP short circuit in predicates like ST_Intersects and ST_Contains is currently defined for POINT - [MULTI]POLYGON pairs. This algorithm could be expanded to include MULTIPOINTs, esp. for one-member-multipoint cases and optionally in a few-member-multipoint cases (if performance testing warrants it).

[dbaston]

I started some experimentation on this: ​https://github.com/postgis/postgis/pull/75

[robe]

Wow Dan, you're cranking. Putting me to shame. Since you have something in works, changed this to 2.3.

[dbaston]

I've updated the pull request to extend the PIP code to cover MultiPoints in ST_Intersects, ST_Contains, ST_Covers, and ST_Within. Initially I was limiting this to MultiPoints with only a few members, but I've removed these restrictions - if the PostGIS PIP code is faster than GEOS for 1 PIP test (and it is), then it should be faster than GEOS for 100 PIP tests (and it is).

I think I'm misunderstanding something about the RTree side of this process, though. From profiling, the geos_intersects function is spending 50% of its time creating and destroying RTrees, but I'm having trouble getting it to ever _use_ those RTrees. What type of query is needed for the caching behavior to kick in?

[dbaston]

Committed to trunk at r14613. Hopefully worth the wait?

[robe]

(In [14618]) Document which functions Enhanced by PIP change. references #75 Also add section to show enhancments in xsl generation script