Hannu Krosing <[EMAIL PROTECTED]> writes:
> I think that placement of NULL's should be a property of ORDER BY and
> separated from opclass.
That would be an extremely bad idea, because it would immediately remove
index scans as one way to meet an ORDER BY. I'm thinking in terms of
NULL high/low as becoming a property of btree opclasses so that indexes
know what to do with nulls, and so that the planner can tell whether a
given index meets the required sort ordering or not.
Alternatively we could define an index's ordering as being specified by
both an opclass and a NULL direction, but that doesn't seem better to
me; especially since the null-direction concept doesn't seem meaningful
for non-btree indexes at all, but a structure like that would require us
to associate a null-direction with all indexes.
regards, tom lane
---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
choose an index scan if your joining column's datatypes do not
match
