Operator <<->> incorrectly calculates m distance

[dbaston]

When computing the m-distance between points and lines, the m ordinate of the first geometry is ignored.

SELECT 'POINT M (0 0 100000)'::geometry <<->> 'LINESTRING M (0 0 5, 0 1 6)'::geometry AS dist;
 dist 
------
    5


SELECT 'POINT M (0 0 1e9)'::geometry <<->> 'LINESTRING M (0 0 5, 0 1 6)'::geometry AS dist;
 dist 
------
    5

[dbaston]

Corrected in trunk at r14939, 2.2 at r14940

[dbaston]

Reopening, because the test case I added fails on Postgres < 95. Presumably the operator should produce identical results regardless of PG version?

[robe]

No. I think the <<->> in 9.5+ returns tru distance rather than centroid box distance. Though I see our docs don't mention it which they should:

​http://postgis.net/docs/manual-2.2/geometry_distance_centroid_nd.html

All the old centroid box KNN operators use recheck in 9.5+

[strk]

Yeah, that's going to become a very confusing issue, I'm afraid.

[strk]

See how the tests for <-> are handled. The current all-red build badges table is alarming

[dbaston]

Apparently fixed at ​https://github.com/postgis/postgis/commit/963ed4e3bc942ff1417906cb57a2c007ba763b3c

[strk]

That is r14942