I have been doing a bit of thinking about the Missing Information
problem in both SQL and TRDBM systems

and would like some feedback in relation (pun intended) to those
thoughts.



Assume a simple Customer - Order schema



Customer (Customer# INT) KEY (Customer#)

Order (Order# INT, Customer# INT) KEY (Order#) FOREIGN KEY
(Customer.Customer#)



The Customer# domain is a set of unsigned integers from 1 -> INT.MAX



Lets assume that the Customer relation has no tuples ie: is a empty set.



The domain for the Customer# attribute in the Customer relation is still
1 -> INT.MAX



Now here is where I sink or swim........



BUT the domain for the Customer# attribute in the Order relation is
effectively an empty set

due to the RI constraint. The domain "grows" as tuples are added to the
Customer table.



So from there I have a couple of questions



1) Is an empty set of type X a valid value of the domain X?

2) Does an empty set of type X violate 1NF?



So in context it boils down to this question...

Can I insert an empty set (with the relevant empty-set types) into the
Order table without violating RI?



If the answer is YES then could we not simply use an empty-set as the
Missing Information marker?



I got these thoughts from TABLE_DEE and TABLE_DUM as outlined in TTM but
unfortunately

my copy is currently 1200Km away and I am struggling to remember there
definitions correctly.



All criticism welcome however harsh.......


--
Posted via http://dbforums.com

"byrmol" <member27348 (AT) dbforums (DOT) com> wrote

No. It is a valid value from of the domain set of domain X.


No. Set values are values of set domains.


No. The order table does not have a set-valued or relation-valued attribute;
it has an integer-valued attribute.


Not applicable.


TABLE_DEE is a nullary relation with cardinality 1 and TABLE_DUM is a
nullary relation with cardinality 0.


Harsh?

Thanks Bob for succint reply..



I added harsh because I see some atrocious ad hominem attacks in
this forum...



In rugby terms.. Play the Ball not the man....



In reference to question 1..



If we extend every Domain to be a set domain would this allow us to use
an empty set as the Missing Information marker?



ie.. The INT Domain now becomes this {{1}, {2}...{n}}

instead of {1,2 .....n}


--
Posted via http://dbforums.com

"byrmol" <member27348 (AT) dbforums (DOT) com> wrote

The INT domain would become {{},{0],{0,1},{0,2}...{n,n,n,n,n...}} if we
arbitrarily allow a set of INT where we now allow INT. It does not seem like
a sufficiently cautious proposal to me.

Relation-valued attributes already allow us to represent optional data as a
constrained set where we can enforce a maximum cardinality of 1 by
specifying a nullary key.