From: DT
To: Editor

I had a question about the missing-values suggestion in PRACTICAL ISSUES IN DATABASE MANAGEMENT, page 234. You write:

"Table operations would have to be modified to yield results with as many tables as there are types of propositions with only known values."

How would this be represented in a language like Tutorial D, where relvars are required to be strongly typed?

One possible idea is to make use of type inheritance. Suppose I had a domain of tuple values {x,a,b,c} (all integers, say) where x is not allowed to be missing but a, b, and c are allowed to be missing. Suppose we extended the domains of a, b, and c with an "imaginary" special value that we will never represent, which I will show for diagram purposes only as '?'. Then the domain can be split into parts:

XABC {x,a,b,c} possrep: {X: int, A: int, B: int, C: int}
XAB {x,a,b,'?'} possrep: {X: int, A: int, B: int}
XAC {x,a,'?',c} possrep: {X: int, A: int, C: int}
XBC {x,'?',b,c} possrep: {X: int, B: int, C: int}
XA {x,a,'?','?'} possrep: {X: int, A: int}
XB {x,'?',b,'?'} possrep: {X: int, B: int}
XC {x,'?','?',c} possrep: {X: int, C: int}
X {x,’?','?','?'} possrep: {X: int}

Using Mr. Date's specialization by constraint idea, we can inherit all the subtuple types from the main tuple type. Updates could make a tuple change type. A relation of relations of XABC type could be used to return results of a query. Each relation within the relation would contain one subtype.

However, the exponential explosion of possible subtypes would be very difficult to handle, practically speaking. As you admit in your book, a real DBMS might have to handle thousands of small subtables. This cannot be passed off as an "implementation detail" since table operations "yield results" at the user presentation level. No matter how efficient the underlying system might be, this seems unacceptable. Perhaps we have to fall back on default values after all.
 
 

Fabian Pascal Responds:  Tutorial D does not explicitly incorporate the concept of missing data as metadata expounded in my PRACTICAL ISSUES IN DATABASE MANAGEMENT, which originates with David McGoveran. THE THIRD MANIFESTO refers in an appendix to Chris Date's default value scheme included in his RELATIONAL DATABASE WRITINGS 1991-1994. He proposed it only as a better solution than NULLs, but it does not address the fundamental meta-data nature of missing data that I raise in my book. He subscribes to the meta-data approach as the theoretically correct solution and we both agree that the lack of elegant, simple solution is inherent in the nature of missing data, because it is outside the scope of the two-valued logic of the real world. Incidentally, the term "NULL value" is a contradiction in terms: NULLs are not values [and thus violate Codd’s Information Principle].

Posted 04/05/02

[ABOUT] [QUOTES] [LINKS]