[linux-elitists] (from Pigdog) Feuerstein on Oracle

Paul J Collins sneakums@sto-kerrig.org
Mon Oct 23 12:07:11 PDT 2000

>>>>> "Bad" == Bad  <mr.bad@pigdog.org> writes:

>>>>> "PJC" == Paul J Collins <sneakums@sto-kerrig.org> writes:
    PJC> First normal form (1NF) is "each fact is present only once in
    PJC> the database".  This is a pain in the ass to do in an OODBMS
    PJC> without your objects becoming uncannily relation-like.

    Bad> Hmm. I'm having a hard time picturing what you mean, here. Would this
    Bad> be, say, two objects (like a parent and a child) that are mutually
    Bad> navigable? So "A is the parent of B" is a fact shown in both B's
    Bad> parent reference and A's child reference?

Er, I don't think that's what I'm driving at.

If you have a relation like so:

  Name       | Address
  Paul       | Ireland
  Evan       | USA
  Nick       | USA
  Bertie     | Ireland

Then it is not 1NF, since the fact of a location's name is duplicated
all over the place.  But if you replace it with two relations like

  Name       | Addr_id
  Paul       | 1
  Evan       | 2
  Nick       | 2
  Bertie     | 1

  Addr_id    | Address
  1          | Ireland
  2          | USA

You can then create a view based on these relations:

CREATE VIEW name_address
        AS SELECT Name, Address FROM names, addresses
                WHERE names.addr_id = addresses.addr_id;

which gives you a view that's much the same as your non-normalised
relation but that will remain consistent, since the underlying
relations can't help but be so (given appropirate
primary-key/foreign-key/phase-of-moon thingies, blah blah blah).

Note: I severely tested the limits of my SQL knowledge in the
construction of that query.  If it's wrong, bad or stupid, let me just
say that I am not surprised.

    Bad> That seems pretty valid, but I don't see a good reason that
    Bad> you couldn't just yank one of the refs. Like, I'd probably
    Bad> recommend doing this for regular in-memory objects, anyways.

    Bad> Maybe what you're saying is that A would contain a full copy
    Bad> of B, which would of course get out of synch with B. I can
    Bad> see this as another valid problem, but again the answer would
    Bad> be: don't do that.  It's bad object design.

Most certainly.

I don't believe that OODBMSes are a valid way to store and manipulate
data.  Not that I can back this opinion up, of course.

Paul Collins - - - - http://www.sto-kerrig.org/ - - - - [ A&P,a&f ]
GPG: 0A49 49A9 2932 0EE5 89B2  9EE0 3B65 7154 8131 1BCD
``Take care to avoid the heavy stuff,
  I give up, this literature is fluff.''

More information about the linux-elitists mailing list