Invalid geometries: equal and disjoint?

[mwtoews]

It seems that GEOS and JTS think that invalid geometries are disjoint. For example, GEOS 3.5.0 from shapely:

from shapely import geos, wkt
geos.geos_version_string  # 3.5.0-CAPI-1.9.0 r4084
g = wkt.loads('LINESTRING(1 1, 1 1)')
g.is_valid  # False, Too few points in geometry component at or near point 1 1
g.relate(g)   # FFFFFFFF2
g.equals(g)   # False
g.disjoint(g) # True

And same from JTS. This is also the same behaviour of PostGIS up to version 2.1

SELECT PostGIS_Version(), PostGIS_GEOS_Version(),
    ST_Relate(g, g), ST_Equals(g, g), ST_Disjoint(g, g)
FROM (SELECT 'LINESTRING(1 1,1 1)'::geometry g) g;
-[ RECORD 1 ]--------+--------------------------------------
postgis_version      | 2.1 USE_GEOS=1 USE_PROJ=1 USE_STATS=1
postgis_geos_version | 3.4.2-CAPI-1.8.2 r3924
st_relate            | FFFFFFFF2
st_equals            | f
st_disjoint          | t

However, since PostGIS 2.2, the invalid geometry may also be equal:

SELECT PostGIS_Version(), PostGIS_GEOS_Version(),
    ST_Relate(g, g), ST_Equals(g, g), ST_Disjoint(g, g)
FROM (SELECT 'LINESTRING(1 1,1 1)'::geometry g) g;
-[ RECORD 1 ]--------+--------------------------------------
postgis_version      | 2.2 USE_GEOS=1 USE_PROJ=1 USE_STATS=1
postgis_geos_version | 3.5.0-CAPI-1.9.0 r4090
st_relate            | FFFFFFFF2
st_equals            | t
st_disjoint          | t

Unless this change was intentional, I'd expect invalid geometries to always be disjoint, and never equal. Disjoint and equal are opposite, so it is surprising to see this result.

[strk]

Results of GEOS/JTS operations on invalid geometries are undefined by design. The only results you can count on are the ones from ST_IsValid*() and ST_MakeValid()

[robe]

Do we have plans to do anything about this. Only peculiar thing to me is the ST_Equals bit. Seems we might not be strictly relying on GEOS for that. Which is not a bad thing.

I'm marking as won't fix for now but feel free to push if you think there is something serious like a memory leak.