Re: Temporal operations

On Sat, 06 Sep 2003 13:53:39 -0700, Paul G. Brown wrote:

> No reference to user-defined types

        When the relational model says domains, user-defined types are implied.

        After all, you don't expect the DBMS to ship with all the domains you may ever need?

> Bob: it would appear either that you are a little lacking in
> 'intellectual honesty', or that you continue to struggle with basic
> written comprehension.

        Beware, your description of Bob looks like you're projecting yourself.

> BTW, D&D in the rest of their book promptly *extend* the model with
> domain inheritance (not a concept from mathematical logic and, as D&D
> themselves say, 'independent of' to the relational model)

        Again you are taking things out of context... the inheritance is orthogonal and optional, it has even been taken out of D4 v2 with D&D's blessing.

> *correct*
> the model by introducing a distinction between relations and and
> entirely new thing called a relvar

        What is the problem in making clear a distinction that was already implicit? Or don't you think the distinction between value and variable is worth making?

> subsume the core of the relational
> model (33 prescriptions) within an edifice that includes 26 other
> inessential rules.

        If something is a suggestion, how can it subsume what is prescripted?

> But to broaden the theme: why is it illegitimate to 'extend' or
> 'pervert'
> the relational model with 3VL but OK to add 'domain inheritance' (OO
> Prescriptions 2 & 3)?

        3VL is easily the most contentious point in the relational field, I wouldn't spend much arguing this in the current loaded context.

        But, even assuming 3VL to be OK, you still have to acknowledge it is fundamental if the system uses 3 or 2VL; while domain inheritance is just an optional accretion.

        How can you compare both, unless you're letting emotions take over your thinking?

> Why does a description of a logical model include
> two Prescriptions about transactions? (RP# 17 and OO # 5)?

        Read carefully... you will see that TTM title's _Foundations for Future Database Systems_, that is, it is meant not only to set the record straight on the current state of the relational model but also to describe a possible implementation. Transactions, or an alternative way of dealing with concurrency, may not be part of the relational model but are essential to a database system... their inclusion, you may notice, is not about prescribing transactions per se but about preventing perversions of transactions that are frequently claimed for by naïve programmers.

> And if (as
> they may) D&D remove transactions from their view of the model does
> this constitute a 'correction'? Using words like 'pervert' amps up the
> rhetoric but adds nothing of substance to the argument.

        Again, if a system will support transactions or simply simultaneous operations (D&D's new proposition) is hardly as essential as making the distinction between values and variables, domains and objects, and logical and physical...

> Leaving aside for a moment the point that if the U_ operators are
> shorthand for simpler operators why go to the trouble

        Ooops... I won't go into details now for lack of the book by my side now, but it seems you are misrepresenting stuff here.

> There are alternatives. Why not recognize that 'value identity' is
> not
> the same thing as 'value equality' yet 'value identity' fulfills all
> the requirements of the other aspects of the model (functional
> dependencies, set algebra, basis for proving operator re-ordering
> rules, etc). This is how Codd deals with the question in RM/V2[1] (in
> RT-1).

        Will try to study this tomorrow, luckly have found a (expensive) used copy. At first sight it seems nonsensical, but then I may be just naïve...

> Another question: why is D&D's approach to view updateability
> preferred to Codd's (or to Dayal & Bernstein's, or Keller's)?

        Perhaps by virtue of its reasonability, clarity, comprehensiveness?

        Or can you expand on Codd's advantages?

> Doing so means that we don't need to suddenly add to the model
> propositions like 'The following relations *must* exist", and makes the
> 'Relational Model' capable of capturing a much broader range of
> predicates.

        Again, can you expand please?

> I would also do away with domain operators, domain
> inheritanc (the whole ghastly edifice of the POSREP)

        Ops... again, POSREPs have little to do with domain inheritance... they certainly don't depend on inheritance, and are useful in its absence.  What do you have against them?

> [1] Codd, E. F. "The Relational Model for Database Management: Version
> 2"
> Addison-Wesley. 1990.
>
> (Note that this work pre-dates the publication of the Manifesto by a
> whole
> four years, pre-empts many of TTM's novelties (though not all)

        Which ones?

        BTW, TTM is not about novelties. It is about reaffirming the RM as a solid foundation against the OO Newspeak, and applying it to the needs that supposedly gave birth to OODBs.

-- 
 _   Leandro Guimarães Faria Corsetti Dutra     +41 (21) 648 11 34
/ \  http://br.geocities.com./lgcdutra/         +41 (78) 778 11 34
\ /  Answer to the list, not to me directly!    +55 (11) 5686 2219
/ \  Rate this if helpful: http://svcs.affero.net/rm.php?r=leandro