Unstable results with Prepared Geometry (ST_Within, ST_Intersects)

[strk]

This is pretty puzzling. Load attached dataset, then: (I suspect it affects all functions that used prepared geometry - does same with ST_Intersects)

-- This query says none of the 3 polygons contain the point
=# select gid, ST_Within(
ST_GeomFromText('POINT (-115.04252 36.05137)', -1), the_geom
) from foo_reload;

-- This query says polygon 3 does contain the point
=# select gid, ST_Within(
ST_GeomFromText('POINT (-115.04252 36.05137)', -1), the_geom
) from foo_reload where gid = 3; 

Storing the point geometry into a table makes no difference. Geometry number 3 really contains the point. The other two don't.

I've tested this with PostgreSQL 8.4.3, PostGIS 2.0.0SVN and GEOS 3.3.0dev

[robe]

Just to add to strk — me too, and see ST_DWithin works fine and since ST_Within works fine with just one record, and I don't think ST_DWithin uses prepared geometry, that rules out Geos and any weirdness in PostgreSQL. So I'm guessing prepared geometry hashing. Check this out:

select gid, ST_IsValid(the_geom), ST_Within(ST_GeomFromText('POINT (-115.04252 36.05137)',  -1), the_geom) from foo_reload where gid !=2;


 gid | st_within
-----+-----------
   1 | f
   3 | t

select gid, ST_NPoints(the_geom), ST_Distance(the_geom,ST_GeomFromText('POINT (-115.04252 36.05137)',  -1)) As dist , ST_DWithin(ST_GeomFromText('POINT (-115.04252 36.05137)',  -1), the_geom,0.0000001) from foo_reload

 gid | st_npoints |        dist         | st_dwithin
-----+------------+---------------------+------------
   1 |        111 |   0.149589748004329 | f
   2 |        342 | 0.00350800000001072 | f
   3 |       2916 |                   0 | t

[robe]

Oops wrong output - but still the same result

select gid, ST_IsValid(the_geom), 
ST_Within(ST_GeomFromText('POINT (-115.04252 36.05137)',  -1), the_geom) 
from foo_reload where gid !=2;


 gid | st_isvalid | st_within
-----+------------+-----------
   1 | t          | f
   3 | t          | t

[strk]

I think it has to do with prepared geometries too. You get true where gid > 2 and two falses where gid > 1.

[strk]

Reporter says this bug is in 1.5.2, calling for an 1.5.3 on bugfix here (but not planning to fix this myself, not in the short term at least)

[robe]

I think this highly points the finger at prepared geometry

select gid, ST_IsValid(the_geom), ST_Intersects(ST_GeomFromText('POINT (-115.04252 36.05137)',  -1), the_geom) from foo_reload;

 gid | st_isvalid | st_intersects
-----+------------+---------------
   1 | t          | f
   2 | t          | f
   3 | t          | f