After thinking a bit more, I decided to reply in a more constructive
way.
Thomas Lockhart wrote:
> These "table links" seem to controvert the ability for a RDBMS to mix
> and match tables in ways which are not hardcoded beforehand.
Certainly links are only of use to their intended purpose, and to
nothing more. But you should be aware that real life relationships
are exactly ot this kind. The drawback of general relational model
is that links (=joins) are built from scratch at the moment of join.
This may seem an advantage, but really this is often an unnecessary
redundant feature whose design allows to build a swarm of relationships
which never existed and will never be used.
Keeping all that in mind, we might consider building a subsystem in
SQL server which is carefully optimized for such real life tasks.
There is no need to put any restrictions on general SQL, the only
thing proposed is enhancement of a particular side of the server.
> Regardless of whether "there exist some real servers that offer such
> features I am talking", a departure from the relation model in a
> relational database is likely to lead to undesireable constraints and
> restrictions in our future development.
>
You have already done a heroic deed of implementing MVCC, it seems
the most interfered with thing. I can see no serious interference
with any SQL feature which you might implement.
--
Leon.
