Re: PRIMARY KEY on a *group* of columns imply that each

Stephane Bortzmeyer wrote:
> On Tue, Apr 26, 2005 at 03:48:44PM -0500,
>  Scott Marlowe <smarlowe@g2switchworks.com> wrote
>  a message of 26 lines which said:
>
>
>>Here's a quote from the SQL1992 spec that's VERY clear:
>
>
> Yes, PostgreSQL is right and implement the standard. Now, what's the
> rationale for the standard? I understand it for a single column but,
> for several columns, it should be still possible to have different
> tuples, such as (3, NULL) and (5, NULL) for instance.

The value of (3,NULL) isn't well-defined. In particular, you can't say
that (3,NULL) = (3,NULL) since NULL means not-known. The fact that part
of the value is not known means the value as a whole is not known.

--
   Richard Huxton
   Archonet Ltd