Geography ST_Intersection -- check the brain-dead obvious before transformation

[robe]

Not sure if this is worthwhile to implement in 1.5 so setting as 2.0 milestone, but in the

"Think of the children" mindset, I really don't want to field questions like

I did this:

SELECT ST_AsText(ST_Intersection(ST_GeographyFromText('SRID=4326;POINT(1.2456 2.001)'), 
		ST_GeographyFromText('SRID=4326;POINT(1.2456 2.001)'))); 

and it gave me a different ("POINT(1.2456000000062 2.00099999999996)) which is a different point than what I started out with. If a point intersects an object, shouldn't the intersection be the point?

Makes me almost wonder if its better to create a geography _ST_Intersection function that handles the simple cases like POINTS, MULTIPOINTS before falling back on a transform. That would make it more changeable too without requiring an SQL script upgrade.

[pramsey]

There's two issues in this ticket: should we try to mask transform drift? No. Should we implement the transform wrapped functionality purely in C? Maybe, can we talk on the list about it?