Peter Eisentraut wrote:
> I think putting some work and thought into inheritance and making it work
> right would make a lot of people very happy, and inheritance is one of the
> major ideas behind OO in any context. Another thing to expand upon would
> be using classes ("tables") as datatypes. I believe this is doesn't work
> all that well. But we're surely "ORDBMS material", if you like.
Yes, postgres pretends that classes as datatypes work, but if I remember
right it doesn't work in practice.
> Pure object-oriented databases (which is where the oid thing comes from)
> are somewhat separate though, they represent a paradigm shift similar to
> moving from, say, hierarchical or network databases to relational ones.
Yeh, but this need not be so. There is no necessary conflict between
the requirements of RDBMS and ODBMS. Postgres plus a couple of features
would quite fulfill both paradigms. Why no commercial vendor seems
to have done this very well I don't know.
> The research in that area is not at all complete and it lacks a
> standardized query language and a whole bunch of other stuff.
Not really true. There IS a standard object query language called OQL,
which
is supported by some ODBMSes. OQL is basicly SQL, except you don't have
to
specify WHERE criteria when it's obvious and a few bits and pieces.
There
is no reason you couldn't support SQL+OQL because they don't really
contradict.
> Since a
> major goal of this project is moving ever closer to SQL compliance,
> becoming an "OODB" is not in the near future.
I would have thought what was in the near future, is whatever people
choose to hack on. I take it no-one is going to reject sensible patches
along this line?