Would the constraint not be satisfied if each combination (including NULL) were not also forced to be unique? I maintain that the constraint is still satisfied. So, it is satisfied if I stuff thousands of NULL values in there. And it is satisfied if I only allow a single NULL value. With multiple columns, it might be set such that each combination must be unique, like binary counting. Let me also state that I agree: allowing null values in a unique index is ludicrous. But if it is allowed, I think forcing the combinations to be single valued makes more sense than allowing any number of them. -----Original Message----- From: Stephan Szabo [mailto:sszabo@xxxxxxxxxxxxxxxxxxxxx] Sent: Thursday, January 20, 2005 11:14 AM To: Dann Corbit Cc: Tom Lane; Greg Stark; pgsql-general@xxxxxxxxxxxxxx Subject: Re: Unique Index On Thu, 20 Jan 2005, Dann Corbit wrote: > It is clear to me that only allowing a single null value will not > violate the explanation below. Given two rows in T with one column each (NULL), (NULL) Find two rows such that the value of each column in one row is non-null and equal to the value of the corresponding column in the other row according to 8.2. If there are no such rows the unique predicate returns true (ie the constraint is satisfied). ---------------------------(end of broadcast)--------------------------- TIP 8: explain analyze is your friend
