Emmanuel Charpentier wrote:
> I *strongly* suggest not to change anything in the default behaviour,
> which is what is expected from an SQL-compliant system, even if the
> database in question uses inheritance internally.
Can I assure you that these changes have NO EFFECT on anybody who
does not use inheritance. i.e. Postgres will remain as SQL compliant
as it was before.
> The reason for that plea is that a modification would crash any
> program not explicitly written for inheritance features.
No it won't. If you don't use inheritance, you will not be effected in
any way.
> : such
> features might be used by, say, the administrator and coere
> programmers of a database, who are not necessarily publish this
> internal use of inheritance to end-users. Furthermore, such a change
> would forbid evolution of a database from a pure-relational to an
> object-orien,ted one : the two representations would be incompatible.
>
> It should also pointed out that most interface programs (such as ODBC
> or JDBC drivers) are not and will not in a foreseeable future be
> designed for use of these features. Modifying the default behaviour
> would break them.
In my opinion, this change will give users of ODBC and such tools MORE
useful defaults. Of course if you are using a non-OO interface to an OO
database there will always be things you can't do. But IMHO, this gives
a more useful set of defaults as a trasition phase.
For example, currently if I have student and employee inheriting from
person, ODBC query of SELECT * from person will return... NOTHING! After
these changes the query will return all the persons (which happen to
be students and employees).
> Apart from that, I am, after 17 years of exposure to the concepts of
> object-oriented programming, still to be convinced of the value of
> this paradigm. This is *not* to suggest that these developments should
> be left over ! However, I *feel* that the real issues behind this
> concept are not yet fully understood, and that some deep theoretical
> work remains to be done (in logic, for example : while the
> well-understood relational theory directly relates to set theory, I
> think that a mathematically correct objects-and-types theory shoud
> emanate from category theory but remains to be created ...).
Well, the fact is people are using OO now, and it's hard for me to
explain the development advantages of an OO database to someone who
is not coding. But if you really want to find out why an OO database
is good, head on over to versant.com or odi.com, download the database
and write a small application. Apart from anything else, some people
need the improved performance NOW, and can't wait for the academics
to give their stamp of approval. And OO database coding simplicity
is saving millions of $$$ NOW.