Re: 1GB Tables as Classes, or Tables as Types, and all that refuted

"Alfredo Novoa" <alfredo_at_ncs.es> wrote in message news:41a60fe8.15975578_at_news.wanadoo.es...
> On Thu, 25 Nov 2004 16:43:49 +0100, "Rene de Visser"
> <Rene_de_Visser_at_hotmail.de> wrote:
>
> >> It's well known that the DBMSs that tried to treat relational
> >> variables as types were a fiasco. The problems derived from such
> >> fundamental mistake helped a lot.
> >>
> >I presume though, not for algebraically defined n-ary relations (n <> 1)?
>
> I don't understand you. What do you mean with algebraically defined
> relations?
>
> Something like this?:
>
> http://foldoc.doc.ic.ac.uk/foldoc/foldoc.cgi?discriminated+union

No. I mean for example the relation integer-binary-plus defined by the set of tuples

(x y z) s.t. x + y = z for all x, y, z that are integers.

i.e. this is a relation with an infinite number of tuples.

i.e. the tuples of the relation are defined algebracially as opposed to being stored.

As such the example is a constant relation. The tuples do not change with time.

a very simple example of an algebraically defined relation is 'Integer' consisting of the tuples (1-tuples in this case) (x) s.t. x is an integer.

>
> >Otherwise this would lead directly to algebraic types,
> >which would break the required uniformity of tuples within a relation.
>
> I don't understand you, but if we treat to use relvars as types then
> we would break the Relational Model and we would not have relations at
> all.
>
> See TTM page 28.
>
> >I also presume that such DBMS languages only supported global relational
> >variables?
>
> I don't see the reason for such presumption.
>
If we base types on relational variables in a language that supports lexical relation variables, then this leads to lexical types. i.e. type that would only be valid in the lexical closure of the variable. This doesn't seem very usefull in a 'normal' relational language which generally works with global constraints.

Over the previous few years there has been however work with langauges that support 1st class relations, and then I guess 1st class types would make sense as well. But as you were talking about previous DBMS systems, I guess these were before the idea of 1st class relations.

Rene. Received on Thu Nov 25 2004 - 18:17:43 CET