> After further thought I think the goal of making explicit sort order
> always match btree index results is unreachable, because the explicit
> sort hasn't got enough information. All it has is an operator ID, and
> that's about all it can possibly have, at least in the "USING operator"
> case. But btree ordering doesn't depend on an operator ID, it depends
> on an opclass. The counterexample goes like this: I could easily make
> two different opclasses, "int_forward" and "int_reverse", that both work
> on int4 data but produce opposite btree sort orders. They're even built
> from the same operators, just lined up differently. Now, how shall an
> explicit sort decide which btree ordering to conform to? Indeed, if
> I make two indexes on the same table using the two opclasses, it's not
> even predictable which ordering an index-driven sort will return.
>
> This counterexample is a bit farfetched of course, but it shows that
> there is no theoretically-pure answer. We have to make some unprovable
> assumptions about what to do.
But if we have an operator, shouldn't we just look for < and > and use
them if we have those symbols? Seems like it would be better than what
we have, if not perfect.
-- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610)
853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill,
Pennsylvania19026