Jeff,

There is a DBMS design criteria which states that:
  All data and relations shall be visable to the user.   This means that
all keys and indexing schemes should be capable of being regenerated from
the visible data.

According to Date  the idea that all information must be derivable from the
visible values in the database has always been implicit in the relational
model.
I am confused because I read you to say that is not the case with RDBMS
systems.   By Date & Codd's definitions "true" RDBMS systems share this
feature with IMAGE.

I refer you to Codd's RM/V2  (1990)

RS-4  Information Portability
------------------------------------------

"If a row of a base R-table is moved in any kind of storage by the DBMS,
its information content as perceived by users remains unchanged   ...  The
entire information content of the DBMS as seen by users must not be
dependent upon the site or equipment in which any of the data is located."

I read the last sentence to mean that the entire "information content" of
the database should be capable of being recreated from the user-visible
data alone.

Codd comments that: "any hashing of data done by the system must not be
perceptible to users.    Similarly, if the primary key value is ever system
generated, that value cannot be a pointer or an address, and cannot be
location-dependent in any way."

- Cortlandt Wilson

Jeff Kell <[log in to unmask]> wrote in article
<[log in to unmask]>...
> [...previous answers snipped...]
>
> Here's another way of looking at a true RDBMS vs
> Indexed/Networked/Heirarchical database models:
>
> With IMAGE, you can have all of your internal pointers, indexes, and
> other "relational" information completely hosed by a catastrophic
> failure and you can fully recreate the database from the raw data (well,
> with the exception of the bitmaps that show active/free status of
> records within a block).
>
> With an RDBMS, the data is scattered around everywhere, and many
> "tables" (depending on the specific software) are simply pointers to
> the data and not "duplicates" of the data itself (as in Image "keys"
> actually duplicating in each affected dataset).  Lose your pointers here
> and pray you have a recent backup.
>
> Jeff Kell <[log in to unmask]>
>