Re: Guessing?

From: Brian Selzer <>
Date: Sun, 25 May 2008 15:39:28 -0400
Message-ID: <gfj_j.595$>

"paul c" <toledobysea_at_ac.ooyah> wrote in message news:79h_j.164958$Cj7.18026_at_pd7urf2no...
> Thanks to Brian S for reminding about Codd's book. On the question of
> view updateability, I'm interested in comments about this quote from page
> 294:
>> Now for a second example, this one involving union and a view based
>> on two relations, not just one as in the first example. Suppose that two
>> of
>> the base relations in the database are SE and SW, where SE provides the
>> identification and immediate properties of suppliers east of the
>> Mississippi
>> River, while SW provides similar information about suppliers west of the
>> Mississippi. Suppose also that SE and SW are union-compatible and that
>> neither SE nor SW contains a column that indicates directly by its values
>> whether the supplier is east or west of the Mississippi.
>> Base: SE ( S# SNAME CITY STATE... )
>> SBase: W ( S# SNAME CITY STATE... )
>> Now, suppose that a view S is created as the union of SE and SW.
>> Suppose also that a user is authorized to enter a new row into the view
>> S.
>> Such a request must be reflected in some change applied to the base
>> relations,
>> which are the only relations that reflect the true state of the database.
>> How
>> does the DBMS decide which of the two base relations SE and SW is to be
>> the recipient of this row? Even if two of the immediate properties of
>> suppliers
>> recorded in SE and SW are the city and state in which each supplier is
>> located, it is not appropriate to assume that the DBMS or the database
>> has
>> any knowledge about geography, and in particular about which cities and
>> states are on which side of the river.
> >
>> It is worth noting that, in this second example, the view S is actually
>> the disjoint union of SE and SW, a reasonably simple case; still,
>> however,
>> entry of new rows into the view is not admissible. Nevertheless, whatever
>> it does, the DBMS would be guessing the user's or program's intent, and
>> such behavior is unacceptable in managing a shared database.
> I'm fastening particularly on where he says it is unacceptable for the
> DBMS to "guess" at intent. I'm inclined to call his attitude mystical. If
> the DBMS is ordered to insert the tuple and no information it has been
> given, such as constraints, countermands that, and it has a consistent
> method for doing so, why the dickens shouldn't it?

Well, that's just it. It has no consistent method for doing so. If it just guesses at the intent, 50% of the time it will guess wrong, and you'll end up with garbage in the database. As a consequence, queries like, "How many suppliers are west of the Mississippi?" will return the wrong answer.

> It is just as much a guess for the DBMS to give the impression that the
> request/order is ambiguous (a word Codd uses earlier on). In this case,
> it just doesn't know, ie., hasn't been told and should keep its figurative
> mouth shut!

But again, it should not allow the update if it has to guess, because inevitably, it will guess wrong at least some of the time and the database will end up corrupt. That it has to guess is not a guess.

> (My attitude about such inserts is for the DBMS to use its UNION feature
> to all base relations in the definition, then apply any defined
> constraints to the results. In the purest implementation some results
> might be seen by programmers as inconvenient, for example where a primary
> key was involved and the request implied a contradiction, the result might
> be an empty relation, so an implementation might support the raising of
> exceptions for convenience, but this is outside the RDM's scope.)
> Rather than guessing, I think he is really talking about the Information
> Principle, which is an idea that concerns designers, not DBMS
> implementations.
Received on Sun May 25 2008 - 21:39:28 CEST

Original text of this message