Re: On view updating

Alfredo Novoa wrote:

> Costin Cozianu wrote:

>>* value types and reference types are distinct

Forgetting this (by DBMS vendors even) has led to many problems. Remember the pointer/FK "discussion" a while ago?

>>(l-values are distinct from r-values)

Both that deceptively simple looking overloading of the 'type'-concept, and (D&D's) in some ways similar attempts to reduce 'object' to 'type' lead to many vocabulary based, camp generating non-problems.

I said it a while ago, I'll say it differently now: IMHO only the *defining* operators should participate in any definiton of a type. Other operators do *not* belong there. This is in line with a well established tradition. A good example is Peano's definition of Natural numbers based on the successor operator.

'plus', 'minus' and what have you operate on natural numbers, but they are *not* part of the definition of natural numbers.

http://en.wikipedia.org/wiki/Peano_postulates

> ... With tables some of the assignments will fail because they violate
> some integrity constraints. With views some assignments will fail for
> the same reason and other will fail because the DBMS does not know how
> to resolve them. And perhaps other will fail because the DBMS knows
> several ways to translate the assignment but it can not decide which
> to apply.
>
> In case we can't know any of the translations we could return a value
> instead of a variable, but this is a choice for the implementors.
>

>>Therefore updatability of views cannot be viewed as a general mechanism 
>>in the model but rather as a synctactic shorthand that deals with 
>>special cases by translating them to real updates of real tables.

What if they are syntactic shorthands - as long as it is clearly defined what they are shorthands for, I don't see a problem?

> I would like to know what the other comp.databases.theory folks think
> about all this.
Received on Fri Sep 24 2004 - 13:39:08 CEST