Re: Primary and Foreign Keys

[Ronald Bourret <rpbourret@rpbourret.com>]

"Bullard, Claude L (Len)" wrote:
> 
> Umm, I can't seem to get my thinking past the need to
> be able to say (since I am reverse engineering
> backwards from the relational model) that
> the value for this is over there.  It is almost
> as if one needs to denormalize but that seems
> wasteful.  That is why links look attractive.

But denormalizing is a lot of the point of XML (and certainly one of the
selling points of native XML databases). The whole idea is to put all
the related data in one document.

How far to denormalize is an interesting question. If you keep the data
entirely normalized and in separate documents, you have just created an
incredibly inefficient relational database. If the data exists as XML
only between the database and an application, then denormalize all you
want. That's a selling point for XML -- you get all the data in one
place when and where you need it, rather than cobbling it together from
various result sets or piecing it out of a non-normal result set created
by joining multiple tables.

If the data exists as XML only to transfer it between databases, then
you need to be sure your data transfer software handles duplicate
records correctly, which is non-trivial. For example, if the same part
info is contained in two different sales orders, what does the software
do when it tries to insert the second part record? This argues toward
limited denormalization, such as putting sales order header and line
item info in one document, but linking to part info in another document.

-- Ron