Greg Stark <gsstark@mit.edu> writes:
> I have what appears to be a corrupt RTREE index.
I wonder if it's actually corrupt, or if it's just that the index
semantics don't truly match the operator. If the latter, REINDEXing
won't fix it.
As for the first theory, have you had any database crashes lately?
If so I'd write this off as a failure caused by the lack of WAL-logging
support in rtree.
As for the second theory, in this thread
http://archives.postgresql.org/pgsql-general/2004-03/msg01135.php
we concluded that the existing mapping of geometric operators onto
rtree indexes is wrong; see in particular
http://archives.postgresql.org/pgsql-general/2004-03/msg01143.php
However that discussion dealt with << and related operators, not @.
I didn't think @ was broken ... but I might have missed something.
(I was expecting bwhite to come back with a patch to fix the rtree
problems he'd identified, but he never did, so it's still an open
issue.)
regards, tom lane