Re: Proposal: 6NF

Brian Selzer <brian_at_selzer-software.com> wrote:
> I think you're making the assumption that an attribute doesn't apply if its
> value isn't known. That is not always true. It is often the case that not
> all of the information about something is received at the same time. The
> interval during which the first bit of information is received and the final
> bit may be sufficiently long that it makes sense to record parts of that
> information at different times rather than wait until all of the relevant
> information is available and then recording it en mass.

I think you're confusing tuples with things. In the relational model, properly understood, a tuple represents a fact, not a thing. What you are saying is that sometimes there are two facts about the same subject, and that you can find out one, and then later perhaps find out another. If you think of tuples as representing things, then you'd logically expect that all the simple attributes of that thing would be in one relation schema; and if some are unknown, you'd look for a way to represent that. If you understand tuples as facts, though, you are no longer forced into the presumption that all the basic facts about a thing need to be expressed in the same relation schema, and that allows you to consider designs where you can store each fact as you discover it.

This confusion -- that tuples in relations represent things, not facts -- is very common. There are, in fact, numerous database design tutorials, references, etc. that adopt it. I could make the case that, in fact, SQL is designed for it and that this thing-oriented pseudorelational  database design is more common in practice than relational database deasign. It is, however, not relational design.

-- 
Chris Smith