Re: Nullable columns or Separate table ?

Jeffrey,

Thanks for the link. It seems acm.org web site has a "bug" and wouldn't accept perfectly valid Canadian address and wouldn't accept my registration.

Anyhow, after some more thinking here is what I think. (Since from your post you appear to be in academia, please provide input).

This design issue can be resolved by extending the 4th Normalization form. (i.e. removing multi-valued dependencies). If you look at this closely, it really is mismatch between attribute dependencies. A relation has some attribute that are single-valued dependent on the primary key and some attributes who are *AT MOST* single-valued dependent. So,

Emp_ID -> Emp_Name -- always single-valued dependency

but

Emp_ID ->* Manages_Dept -- at most single-valued dependency

I used ->* to represent when Manages_Dept is single-valued or zero-valued dependent on Emp_ID.

If we call this as 4.5th normalization rule, "All attributes should be either always single-valued dependent on primary key or there should be only one multi-valued dependency." then this can lead to Approach (2) in my original post.

Thanks

P. Adhia

P.S. I'll ready your paper, when acm.org resolves issues around creating my ID.

Jeffrey Parsons wrote:
>
> A theoretical viewpoint arguing for your "Approach 2" is given in the
> following article:
>
> Jeffrey Parsons and Yair Wand, "Emancipating Instances from the Tyranny of
> Classes in Information Modeling," forthcoming in ACM Transactions on Database
> Systems (tentatively scheduled for June 2000). A preliminary version of the
> paper can be downloaded (by ACM members) at
> http://www.acm.org/tods/Upcoming.html
  Received on Fri Apr 28 2000 - 00:00:00 CEST