On Fri, 2011-01-21 at 09:10 -0600, Andy Colson wrote:
> On 1/21/2011 2:16 AM, Nick Rudnick wrote:
> > Dear all,
> >
> > for the sake academic teaching, a colleague asked me in how far
> > PostgreSQL does support object functionality these days.
Although postgres at one time had "ORDBMS" as a goal, I haven't seen any
real interest or work in that area for a long time. The OO features are
primarily:
1. OIDs -- hidden columns with auto-generated number.
At one time, OIDs were a part of every table, now they must be specified
explicitly. They still remain on many of the system catalogs, but not
all. Mostly, they are now just an implementation detail that people only
notice when looking at the catalogs (and I am personally annoyed that
they are hidden, because when looking at the catalogs usually you want
to see the OIDs).
2. Inheritance
This feature is now used almost exclusively for physical partitioning
rather than logical design.
> You do realize that ORM sucks, right?
Andy, please try to keep the criticism constructive.
Nick, I think the problem with ORDBMS is that they essentially introduce
pointers in the data, and that adds a lot of complexity.
For instance, the relational algebra has nice closure properties. If you
join two relations, you get another relation.
But what if you introduce pointers? Well, then each table might start
out with OIDs, but then when you join with another table and do a GROUP
BY, you lose any meaningful OIDs. So what do you have? Something else.
Therefore, you've lost closure.
Now, maybe there is a good reason to sacrifice closure and other nice
properties of the relational model. But there is a perception among many
people (like Andy) that the case has not been made.
Regards,
Jeff Davis