Re: [HACKERS] Re: ORDBMS

Tom Lane wrote:

> For a lot of this older stuff, there isn't even any documentation
> (that I know of) on what it's *supposed* to do, let alone on how
> thorough the original implementation was. For instance, I have
> no idea what "classes as datatypes" actually means, in the sense
> of what you could do with them in POSTQUEL. Anyone remember?

I thought that the original berkeley doco covered this to some extent.
Basicly I think you're supposed to be able to go...

CREATE TABLE address (street TEXT, number TEXT, suburb TEXT, zip TEXT);
CREATE TABLE person (name TEXT, address ADDRESS);

SELECT name, address FROM person WHERE person.address.suburb = 'New
York';

And last time I bothered to look postgresql wouldn't reject these SQL
out of
hand, but I don't think it works properly. You should also be able to
write
constraints I guess for address so that it applies to any class that
has an address.

> First problem is to figure out what it should be doing ;-). If you
> can get consensus on that, I doubt anyone will object to making it
> work again.

Hmm. My guess is the original researchers didn't do too much wrong,
so getting the old stuff working shouldn't be too much controversy
if it's documented in the old papers. I
suspect they forgot a few things (like the doubt over getting
columns of sub-classes), but for that I don't think there's that much
controversy, either it does it or not.