Re: candidate keys in abstract parent relations

"paul c" <toledobythesea_at_oohay.ac> wrote in message news:ryeAf.295927$2k.205270_at_pd7tw1no...

> Sometimes I think modellers go too far trying to ascribe dbms attributes
> to the natural world. Certainly big biz and governments do.

Spot on!!

When we record a driver's licence number or an employee number, and name, all we are saying is that we believe that number was assigned to a person who will claim to have that name. Most of the time that also happens to work as identification for the person, but IMO that is an accidental side-effect of most people being honest. But I have a colleague who works on a criminal justice system, and they tie themselves in fantastic knots because they refuse to understand this.

I have never met a database designer yet, including myself, who doesn't struggle to remember that the database only has to assert what we are told, NOT what is objectively true. It doesn't "matter" if the database contains lies as long as it (and the application) doesn't invent or derive new lies. (Obviously there is a burden on the person entering the data to be diligent about establishing the truth as best they can before they enter it, but that's not a database design problem. Also obviously, it is nice to test the internal consistency of the assertions we record, if we can. And just as obviously there has to be a way to remove lies and all their false derivations when they are discovered.)

I am in danger of turning the conversation full-circle here by remarking on how surrogate keys are helpful, so I will stop now. (The problem IMO is how to discourage using them, not the opposite!)

Roy Received on Mon Jan 23 2006 - 10:16:22 CET