has anybody else used r-tree indexes in 6.5?

has something changed with r-tree indexes in 6.5?  i had a database in
6.4.2 that had a box type and would do queries on it with the box
overlap operator ("&&").  the field i'm querying is indexed with an
r-tree index.  all this worked fine in 6.4.2.  then i did a dump &
restore from 6.4.2 to 6.5 and now the queries that were running ok in
6.4.2 aren't in 6.5.  if i drop the index, everything is fine (albeit
slow).  this is the error i'm getting:

ERROR:  Operator 500 must have a restriction selectivity estimator to be
used in a btree index

which doesn't make sense because it's an r-tree index, first of all. 
and second, i don't know what to look for (especially since this was
working great with 6.4.2).

i'm running on a linux system, compiled with a pretty recent egcs.  the
diffs in the regression tests didn't seem to be anything major.  the
only configuration option i selected was --with-perl when i compiled it,
so so everything should be pretty vanilla.  (i haven't even added in any
of my custom types yet, which have been known to mess with operators in
the past.)

it's really easy to reproduce the error:

create table test_table (area1 box);
insert into test_table values ( '(100,100,200,200)'::box );
create index test_table_index on test_table using rtree ( area1 box_ops
);
select * from test_table where area1 && '(0,0),(100,100)'::box;
drop index test_table_index;
select * from test_table where area1 && '(0,0),(100,100)'::box;

any ideas?

jeff

ps, is it just me, or do things seem a little dead around here today?