[postgis-users] Comparing Geometries with Different SRIDs

[Kevin Neufeld]

I might as well give my 2 cents to this discussion as well...

Personally, I don't find this surprising at all.  You're comparing the 
bounding boxes of two geometries in completely different planar 
projections.  This is almost like doing:

SELECT 1::integer
UNION
SELECT true::boolean

and expect to get a single row back.  This doesn't work - type casting 
is clearly called for.  Obviously, I'm exaggerating here to make my 
point, but I prefer stricter type casting by the user than magically 
casting one to the other behind the scenes.  As Brent pointed out this 
morning, even transforming one geometry to the other results in 
coordinate drift and the two bounding boxes won't likely be the same 
anyway.  If it's left to magically transform the geometries behind the 
scenes, it'll be *far* more difficult to track down those application 
bugs that make certain assumptions about their geometries only to find 
out that = is failing due to precision round off errors via transform. 

In my opinion, throwing an ERROR is the better choice.

Cheers,
Kevin


Charlie Savage wrote:
> This looks like a related issue:
>
> select 'srid=4326;point(3 3)'::geometry
> union
> select 'srid=4325;point(3 3)'::geometry
>
> ERROR:  Operation on two GEOMETRIES with different SRIDs
>
> I find that surprising, I would expect to get a result table with two 
> rows.
>
> Charlie
>