On lör, 2011-08-06 at 08:17 +0100, Dean Rasheed wrote:
> The current behaviour is inconsistent - the not-null property of a PK
> is sometimes inherited and sometimes not, depending on whether the PK
> is added at table-creation time or later. So a change in either
> direction is a change to some current behaviour, unless we leave it
> inconsistent, which seems unacceptable.
Yeah, OK, that's wrong then.