Re: ID field as logical address

From: paul c <>
Date: Fri, 29 May 2009 16:39:55 GMT
Message-ID: <vdUTl.29814$PH1.1299_at_edtnps82>

Just a couple of comments about a post that is unusually vague and fuzzy coming from you, it's tedious to dismantle every sentence, so I mention only a couple:

Walter Mitty wrote:

> ... Now let's say
> you have a table that records values that look like (ID, value). In an
> actual case, value would probably ityself be a tuple, ...

Sounds like you are talking about a tuple within a tuple, the tuple that 'value' stands for being somehow 'within' the tuple that '(ID, value)' stands for. I don't think there is a relational notion that matches this, maybe you actually have RVA's in mind.

> So, Brian's claim that deleting a value and inserting a new one is somehow
> really different from updating an existing value works in this model, if
> you regard ID as somehow "special", and not "part of the value recorded in
> the table".
> ...

He is not the only one who talks about one thing and means something diifferent. Here, at least one of you is dreaming of a table that doesn't stand for one of Codd's relations. (Apparently SQL is capable of the same thing, for reasons such as unnamed columns. According to Date another flaw was that SQL is based on 'bags', not sets so it wouldn't be a surprise if Codd shunned the System R department when he worked for IBM.)

The first thing the mystics forget is that they can't make their claims without admitting that they are talking about something other than conventional relational theory. Their claims should be dismissed until they develop their own formal theory for whatever machinations they have in mind and make their assumptions and variations precise. Relational theory is still developing but people who refuse to give substitutes for the very few formal concepts only get in the way of progress.

Meanwhile, aspects of Codd's theory have loopholes such as whether some views can be 'updated', constraint theory is mostly undeveloped, etc. The majority of posts to c.d.t. are about physical questions. Regarding the topic at hand, I've never seen it asked here whether there is any logical reason why views can't have generated attribute values If a view can have D&D-style extend-ed attributes, is there a use for generating values? One possible use might be to simplify quota queries.

(Depending on how you read Codd's 1970 paper, he might have left the door slightly ajar when he described his reduction algorithm, because one can imagine hierarchies where arbitrary attributes need to be introduced. If there are hierarchies that can't be shown as relations without introduced attributes, are there relations that can't be shown as tables without introduced attributes? I think Codd might have said that if there are such relations, they should be avoided, but that's just a guess. I'm certain he would have said such situations weren't the problem he was after solving. Nevertheless, Codd never precluded such attributes as fas as I know, eg. he didn't forbid an attribute's value being generated by the machine.) Received on Fri May 29 2009 - 18:39:55 CEST

Original text of this message