"David Allardyce" <dave@pod13.net> writes:
> ... If they had defined uniqueness as "all rows
> must be distinct" then two all null rows would violate the uniqueness
> constraint. Not the behavior I want at all.
Er, why not? You're essentially arguing that the UNIQUE constraint
should treat nulls as equal, so I'd think two all-null rows should be
considered equal too. I do not see how you can consistently maintain
that (x,NULL) equals (x,NULL) but (NULL,NULL) doesn't equal (NULL,NULL).
But we have been around the maypole on this question several times
before, with no permanent resolution --- the plain fact is that the
spec isn't very clearly written. Useful data would be tests
demonstrating how other systems (Oracle, DB2, etc) interpret the issue.
regards, tom lane