Re: So let me get this right: (Was: NFNF vs 1NF ...)

Alfredo Asks:

>> "Where Codd said that relational valued attributes are not [OK]"

>From [1]

"The term relation is used here in its accepted mathematical sense. Given sets S1, S2, ... Sn (not necessarily distinct), R is a relation on these n sets of it is a set of n-tuples each of which has its first element from S1, its second from S2, and so on. {FN1}"

{FN1} "More concisely, R is a subset of the Cartesian product S1 x S2 x ... x Sn."

Now, what is the "accepted mathematical sense." of a "relation"? Well, to be sound the "mathematical understanding" should be free of paradoxes, and contradictions. As soon as you start saying "the set of domains includes sets of sets" or the like, you leap headlong into a twisted maze of paradoxes. Consequently I propose that in the original paper Codd ruled out the possibility that the values in an attribute could be considered as amenable to manipulation via the "Operations on Relations" he describes in Section 2.1.

Complex domains, on the other hand, are not ruled out. Plonk a matrix, or any other type constructor, in an attribute: fine! Plonk something that can be projected, restricted, joined, unioned, etc, and you've just complicated the crap out of things in a way Codd appears here to rule out.

> But bag data model makes rutime far less efficient.

 Evidence? (Besides a handful of example queries where this is obviously true, usually characterized by the way that the answer includes a Cartesian product of keys...)

What you've stated is the opposite of what early systems builders concluded. And given that that the contents of an XML document are 'strictly ordered', and therefore posess set properties (implicitly, every node in an XML document is preceeded by an implied value which distinguishes that node from all others) we're about to find out just how efficient a set algebra can be.

(Note: I ain't sayin' either way. I'm just sayin' the case is far from clear. This is an empirical question, not a theoretic one.)

>>Date et. al. said "When you define a domain, you gotta define order!"

>Where?

>I never readed that.

You are quite correct: I mis-remembered. The only requirement is "equals" as far back as [1]. However, the object-relational crowd allowed scalar domains which do not even have equals. That's just fine from a data modeling perspective, so long as those domains are never used in a key. But even that would be outlawed by the 3rdM "interpretation".

?>>Date et. al. said: "The relational model needs no extension, no
>>correction, no subsumption, and above all no perversion."

> But it needs correct interpretation.

Any my point is that you appear to believe that Date, Darwen & Pascal have a unique ability to "correctly interpret".

My challenge to you is to find one single, stinkin' thing anyone else has ever written about the relational model of which the Date / Darwen crowd approved. Just one citation where they say something like: "Fred's idea is actually kind of cool." There are lots of 'em: Zaniolo's GEM, Codd's additional relational operators, recursive operators, Stonebraker's "Queries as Domains", any systematic treatment of non-constraint (ie. active) rules.

(I was going to ask, "Failing that, just one place where they took someone else's idea and made a contribution.", but their ideas on inheritance and updateable views meet this criteria.)

[1] Codd, E. "A Relational Model of Data for Large Shared Data Banks" Communications of the ACM 13(6): 1970.

[2] Darwen, Hugh. and Date C. J. "The Third Manifesto" SIGMOD Record. ACM. March, 1995. (or
http://delivery.acm.org/10.1145/210000/202667/P039.pdf?key1=202667&key2=6886508011&coll=GUIDE&dl=GUIDE&CFID=38431871&CFTOKEN=35654682 for the library challenged). Received on Thu Feb 10 2005 - 19:29:06 CET