On Wed, 24 May 2006, David J. Fiander wrote:

> Since the collection of ISBNs can be treated as an equivalence class,
> can't any arbitrary member of the class be designated as the group
> identifier?  This eliminates the need to create a synthetic id, and it
> means that, for singular items, there's no need to create a separate
> group id.

I can see two drawbacks:

1) It could be confusing.  If someone doesn't read the documentation
properly they could be mislead to believe that the table is an
un-normalized one.

2) One could optimize by only storing those numbers that have more than
one in a group.  From Thom's description it seems like that is what xISBN
does.  (I could be wrong here, going from memory). Problem of course if
you don't store the 1 to 1 mapping is that you don't know if there really
is no known relationships or if that particular isbn hasn't been examined
for any relationships with other material yet.  Of course, even if you
examined it there might be a relationship that hasn't been caught, so
the difference might not be that huge.  In either case you could say it is
really saying we don't know of any relation with this isbn and other

Jon Gorman