> Reece Hart <reece(at)harts(dot)net> writes:
>> However, it's not clear that you've considered a clause like 'ORDER BY
>> (foo IS NULL), foo', which I believe is not implementation dependent.
>
> Yeah, that should work reasonably portably ... where "portable" means
> "equally lousy performance in every implementation", unfortunately :-(.
> I rather doubt that many implementations will see through that to decide
> that they can avoid an explicit sort.
Well, an index on ((foo IS NULL), foo) might improve the performance
when sorting along these columns, but sure it's not a cure-all. And you
still have to modify the SQL and the database schema ...
regards
TV