Re: RTREE on points

Jeff Hoffmann <jeff@propertykey.com> writes:
> yes, it does seem like a little more work, but there doesn't seem to be
> a lot of usage of the geometric functions by the developers to look at
> missing features -- they're mostly just reactive to problems.

Jeff's correct, none of the core developers have much time to spend on
adding features to the geometric datatypes.  If someone else wants to
step up to the plate, though, contributions are welcome ;-)

The procedure for adding a new index opclass is somewhat documented for
btree opclasses.  To add an rtree opclass you'd be adding a different
set of operators and support functions, which set isn't documented
anywhere that I know of; you'd have to look at the existing examples
to figure out what is needed.

BTW, you should also look at the GIST stuff and figure out whether
it might not be better to develop a GIST opclass instead of rtree.
In the long run I suspect GIST will be better supported than rtree,
since it's more general.
        regards, tom lane