On Wed, 19 Jan 2005, Dann Corbit wrote:
> Even at that, I think that being able to insert more than one null value
> into a unique index should be considered as a bug (or diagnosed as an
> error).
AFAICT the UNIQUE constraint that it's used to model explicitly allows
multiple NULLs in the spec so I don't see making it error as being
terribly workable.