"Gilles Bernard" <gbernard(at)matra-ms2i(dot)fr> writes:
> I tryed all you suggested me : (forme && '((...))'), drop the index and
> recreate it and do a vacuum analyze and it
> still do a sequential scan on the whole table.
Hm. You might try reducing the selectivity estimates returned by
areasel() and friends in src/backend/utils/adt/geo_selfuncs.c.
I thought they were small enough already, but perhaps not.
regards, tom lane