> > In terms of relational data model you can think so.
>
> In that case, it doesn't fit into the relational model at all as it
> doesn't store "one value".

Then same applies to BLOB - it does not fit relational model too. I
think you're wrong here.

> There's nothing wrong with thousands of tables.

What about supporting them?

> > Because if you split your XML document apart it is no longer XML
> > document that satisfies your DTD/XSD. It is something completely
> > different, similar to extracting bits from your integer.
>
> Instead, you have your relational constraints to validate it. As
> long as you can recreate your XML document, what's wrong with it?

First, you have an answer in the original article - you loose
performance. Second, you try to access single "bits" of the atomic
entity with your queries. That is not natural, though perfectly legal.

> Yes, it perhaps might be _easier_ to store it as an XML document,
> but does that justify the current hype?

Sure. If people need to store and query XML documents, a native XML
store is the solution.

> > No, some you build with joins.
>
> I disagree here. There's no point in joining unrelated data.

I meant relationships like grandfather-grandchild. You do not have
explicit relation using some surrogate key.

> Well yes, but we can store the data from the complex structure
> without the need for the complex structure.

Sure, but you always want to compose/decompose that complex structure
into parts that from your point of view are atomic, execute query on
that parts and then assemble them again back into molecule. They
propose to treat molecule as an atom and define new query language
optimized to querying molecules.

> Performance has nothing to do with the relational model. Performance
> is implementation defined. The relation model is a logical model.

It has. What's wrong with hierarchical model, unless you consider how
many CPU cycles were burnt in order to extract some simple information?

> There's no argueing that one implementation of a certain model
> performs differently from another implementation.
>
> But to dismiss the relational model as a foundation for data storage
> because current implementations perform like a fat dog doesn't hold
> merit.

I dismiss it in certain cases as the model that is not applicable for
my information. You can try to describe a 1 m3 of our air with quantum
model, but you will not succeed (or maybe you would if you were
immortal and you were God). Though thermodynamical model describes it
perfectly (and introduces some new properties like temperature, that
in fact does not exist - there is no temperature of an atom, but you
can easily tell that it is cold outside).

> Go complain with the vendors to make better products.
>
> There's no need to go the OODBMS route ;-)

There is such need. :) Your claims are like claims of a person who
just discovered quantum theory and general relativity theory and
starts to claim that all other theories should be deprecated because
one can always describe something using that two. Sorry, but that does
not work in all cases. Maybe for God, but not for you. :)

Roman