Re: Question about rtrees (overleft replacing left in nodes)

Tom Lane wrote:
> I said:
>
>>However, if this is indeed wrong, why have we not heard bug reports
>>stating that rtree indexes don't work?  Can you generate a test case
>>in which it fails?
>
>
> Actually, it's not necessary to look very far: there's one rtree index
> defined in the regression database, and guess what: it gets << queries
> wrong.

OK, glad to know I'm not going crazy.  Oh well, if there's a way to
break something, trust me to find it within a week of using it. :(

Out of curiosity how many tuples are in that test?  I wasn't able to
invoke index scan with position operators even at 100K tuples.

Also worth noting that &< and &> queries fail (that's the test case I
just sent) as well; the << and >> failure is (if I understand correctly)
a result of &< and &> being used at the node level in the rtree.

Side question: is there a user contrib area for extensions?  The path to
this discovery started with a general interval "template class" to
support any interval type (open, half-open, or closed) on any scalar
data type (note: by "template class" read "C equivalent thereof using
Gnu cpp ## macro construction kluges to create another C file with data
types fileld in").  My original goal was [timestamp,timestamp) intervals
(or (t,t) or [t,t] or whatever) but it works with any scalar that's
internally numeric.  Someone else might as well benefit from the
frustrations I've had in figuring out how to handle operations on half-
and fully-open empty intervals. :)

Thanks,

-- Bill