Re: Operationalize orthogonality

"mAsterdam" <mAsterdam_at_vrijdag.org> wrote in message news:4482bb08$0$31651$e4fe514c_at_news.xs4all.nl...
> x wrote:
> > mAsterdam wrote:
> >
> >>Tony D wrote:
> >>
> >> > ... the relational model is orthogonal to type. That
> >> > means the relational model makes no demands and imposes no
> >> > restrictions on the types available (apart from the
> >> > obvious requirement for a boolean type).
> >
> >
> > I would say that there are some requirements on the types:
> > - equality is defined

> With a demarcated type base the equality need
> not only be defined, but also communicated/shared
> between the TB (type base) and (the rest of?) the DBMS.
> So it seems the DBMS and the TB at least need a common
> language for specifying and implementing operators.

The language of the interface (the "protocol") ?

> > - the values can be denoted in the "relational" language

You forgot this.

> >>What would the requirements for a dbms be to have
> >>this orthogonality implemented?
> >
> >
> >>ISTM the dbms should not have native types
> >>(except boolean and maybe type generators
> >>like relation() and attribute() ) but instead
> >>rely on a lower level, external to the dbms,
> >>type system.

> > What is "external" in software ? Crossing an interface ?

> Yes (I used the term 'demarcated').

> > That interface is then a set of requirements.

> A protocol.

> >>This poses the next question: what would
> >>be the requirements for this type system?
> >
> >
> > Whose requirements ?

> Good question :-)

> The users of the type system.

> Which categories of users do we see?

> The end-users of course would be the same
> as the end-users of the applications and the database.

> The database (via a hypothetical dbms conforming to
> the protocol).

> The aplication (via a language/app framework conforming
> to the protocol).

> Am I forgetting users?

> Both the application and the database would need some form
> of garantuee (thinking of 'leases' like in DHCP, but longer term)
> that a used type will remain valid.

How complicated! It would not be easyer to follow Mr. Codd advice ? Domains, not types. Received on Mon Jun 05 2006 - 08:04:19 CEST