Re: Relationals vs. Objects Databases I

Akmal B Chaudhri wrote:

> 
> On 18 Jan 1998, Richard Ronteltap wrote:
> 
> [snip]
> > There is no technical reason when ODBMS should not replace
> > RDMBS. An ODBMS can easily be made a superset of RDBMS
> > functionality. Some problems of RDBMS are:
> >
> > - No (object) identity for referencing
> > - No duplicate handling (In RDBMS, what looks the same, is the same)
> > - Bad complex data handling. Everything must be broken down into
> >   rectangular tables. The relations between these smashed bits are not
> >   part of the model.
> > - Bad programming language integration (impedance mismatch)
> >   SQL and relational algebra are computationally incomplete and
> >   there is no fitting alternative that is.
> > - No extensibility. You can't add new basic data types.

The list of RDBMS problems can be even extended. For example:

• neglecting (or naive) strong static typing
• naive approach to generic programming (dynamic SQL)
• naive approach to naming-scoping binding issues (embedded SQL does not obey PL scoping rules)
• naive approach to parameter passing
• naive approach to cursor processing (updating!)
• no uniform approach to persistent, volatile, individual and bulk data
• no orthogonality of type constructors (complex objects)
• no orthogonal persistence
• no orthogonality of a query language to persistence
• idiotic solutions wrt null values (see Date)
• no idea of type unions
• decorative mathematics that explains nothing and feeds a lot of theoretical parasites
• semantic reefs in a lot of SQL constructs (e.g. group by)
• very limited concept of a view (especially concerning view updating)
• etc. etc.

> So how about object-relational? Does it not "fix" some of these issues?

Yes, some issues are fixed and some new are introduced. The question is what means "relational" in "object-relational"? Does it mean they apply the relational theory? Does it mean they follow the Codd's relational model? This is a nonsense, the association with Codd and mathematics is lost more than 10 years ago. In "relational" SQL3 we have a lot of stuff that Codd never dreamed about: ADT-s, complex structures as elements of tuples, persistent tuple identifiers that can be used as pointers (the heresy in the relational model!), etc. Eclectic monster that swallows every buzzword flying in the air.

"Relational" in "object-relational" means nothing. (Hence we can shorten "object-relational" it to "object", if you like.) There is no ideological difference between object DBMS and object-relational DBMS. It is difficult to see a single technical point of one of them that cannot be adopted in another one. There are mostly marketing differences: object-relational DBMS are (usually) associated with big companies that previously offered relational systems, and object DBMS are associated with smaller companies that consider OO-ness as a black horse. (Perhaps, UniSQL is an exception.). The rest is the "newspeak" of commercial folders, including the compatibility with SQL and other advantages of the "object-relational" wave.

Regards
Kaz Received on Mon Jan 19 1998 - 00:00:00 CST