Contains gives the error answer, but ContainsPrep gives the correct one

[cuteDen-ECNU]

Consider the statement:

DROP TABLE IF EXISTS t0;
CREATE TABLE t0 (id int, geom geometry, valid boolean);
INSERT INTO t0 (id, geom) VALUES (1,ST_GeomFromText('LINESTRING(1 2,1 1)',0));
INSERT INTO t0 (id, geom) VALUES (2,ST_GeomFromText('GEOMETRYCOLLECTION(POINT(2 2),POINT(1 0),LINESTRING(1 2,1 1))',0));
INSERT INTO t0 (id, geom) VALUES (3,ST_GeomFromText('POLYGON((1 0,0 4,2 2,1 0))',0));
SELECT a1.id, a2.id FROM t0 As a1 JOIN t0 As a2 ON ST_Contains(a1.geom, a2.geom);      
-- result: {1,1; 2,1; 2,2; 3,1; 3,2; 3, 3}   
SELECT  ST_Contains(a1.geom, a2.geom) FROM t0 as a1, t0 as a2 WHERE a1.id = 3 and a2.id = 2;
-- expected: true; actual: false

g3 contains g2 because POINTs of g2 lie in the boundary of g3 and LINESTRING of g2 lies in the interior of g3, which satisfied the sematics of contains:

ST_Contains(A, B) ⇔ (A ⋂ B = B) ∧ (Int(A) ⋂ Int(B) ≠ ∅)

I consider the issue different from #982 because the boundary of g3 is not ambiguous.

The predicate of ContainsPrep gives the correct answer. A simple statement reproduced in Geos:

bin/geosop -a 'POLYGON((1 0,0 4,2 2,1 0))' -b 'GEOMETRYCOLLECTION(POINT(2 2),POINT(1 0),LINESTRING(1 2,1 1))' contains
# false
bin/geosop -a 'POLYGON((1 0,0 4,2 2,1 0))' -b 'GEOMETRYCOLLECTION(POINT(2 2),POINT(1 0),LINESTRING(1 2,1 1))' containsPrep
# true

The version of Geos is the latest one:
6f70b63

[pramsey]

Interesting, same result in covers/coversPrep.

[pramsey]

This is a problem relatively deep in the IM calculation, so we are going to instead focus on the RelateNG work, which hopefully removes this inconsistency in future releases (and is faster).

[cuteDen-ECNU]

Looking forward to the new releases! 😊