Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!caip!nike!oliveb!glacier!navajo!avg
From: avg@navajo.STANFORD.EDU (Allen Van Gelder)
Newsgroups: net.database
Subject: Re: Problem with relational theory
Message-ID: <819@navajo.STANFORD.EDU>
Date: Fri, 5-Sep-86 19:25:56 EDT
Article-I.D.: navajo.819
Posted: Fri Sep  5 19:25:56 1986
Date-Received: Sun, 7-Sep-86 03:24:38 EDT
References: <595@ur-tut.UUCP> <452@hdsvx1.UUCP>
Reply-To: avg@navajo.UUCP (Allen Van Gelder)
Organization: Stanford University
Lines: 18

I disagree with Hoffman's attempted justification of the way
Eric Carleen's query is handled.  Having a calculus-based language
loses its point if the user has to know how the system is going
to translate it to relational algebra, possibly incorrectly.

His query, essentially:
	retrieve t1  where t1.name = t2.name  or t1.name = t3.name
has an obvious meaning in relational calculus, and he has a right
to expect the answer to be based on this meaning.

The correct way to translate an "or" in a selection is as a union.
This is shown in "Principles of DB Systems," by J.D. Ullman in
proving that relational algebra has the same power as relational
calculus.  If some system chooses a different route, it is that
system's responsibility to get it right.  Anything else is like
defining the meaning of a program to be the machine code produced
by the compiler -- easy to implement, but not very useful.
Maybe IBM can run this bluff, but I doubt RTI can.