On Fri, 15 Jul 2005, Hannu Krosing wrote:
> What comes to Toms's May 20 argument that people would be surprised when
> they select form a table whith check(i>0) constraint and get out i<0
> then I think this is a question of education.
I potentially disagree. What are we trying to model here? Systems which
allow children to have values that are not in the domain of the parent are
useful for some things, but they also potentially lose some other useful
properties like substitutability since a child isn't a parent (or more
specifically f(parent) cannot assume parent's invarients are valid on its
argument which may be a child unless they are applied to the child at call
time).