Re: Examples of SQL anomalies?
Date: Fri, 04 Jul 2008 18:40:35 GMT
"Brian Selzer" <brian_at_selzer-software.com> wrote in message
> "David Cressey" <cressey73_at_verizon.net> wrote in message
> > "Brian Selzer" <brian_at_selzer-software.com> wrote in message
> > news:nFD9k.5753$LG4.2422_at_nlpi065.nbdc.sbc.com...
> >> "-CELKO-" <jcelko212_at_earthlink.net> wrote in message
> >> >>> The question is, if these issues are due to the SQL specification
> >> >>> simply due to a problem in a specific SQL product. Or could it be,
> > that
> >> >>> the definition is not precise enough in some points, so that
> >> >>> vendors implemented it differently? <<
> >> >
> >> > Nope, it is the specs. All aggregate (set) functions begin by
> >> > removing the NULLs from their parameter set, then if there is a
> >> > DISTINCT option on the parameter, they remove redundant duplicates
> >> > finally do the operation (MIN, MAX, AVG, SUM, COUNT on what is left.
> >> > Since an empty set has no elements upon which to apply an operation,
> >> > SQL returns a NULL (okay, it should be an "undefined" if we were
> >> > mathematically correct).
> >> >
> >> MIN, MAX and AVG are meaningless when applied to an empty bag, but it
> > seems
> >> to me that COUNT should always return 0 when the bag is empty, and
> >> similarly, SUM should return 0. SUM should only return NULL if one of
> >> the
> >> values to be summed is NULL.
> > By definition, none of the values to be summed are NULL.
> OK. OK. Yes, I know. NULL is not a value. I guess to be precise, I
> have said "if the cardinality of the set of rows targeted by the query is
> greater than the number of values to be summed, then SUM should return
> NULL," but I think that would have caused more confusion. Bottom line: if
> not all of the amounts are known, then the total amount is suspect.
This has to do with a continuing disagreement between you and me. In my view there is no such thing as the "set of rows targeted by the query". There is a set of data targeted by the query. That is all. Rows that do not contain any data with regard to the current query are not part of the target, by definition.
This same disagreement surfaces when you trot out your well worn argument that a row could have all its values updated and still be "the same row" in some meaningful sense. The row is just a container. That's all it is. It has no meaning in the universe of discourse. The data in a row preumably does have meaning in the Uof D.
> > I agree about COUNT and SUM on an empty bag. BTW, the DISTINCT keyword
> > can
> > be used with aggregates to obtain a projection: select count (distinct
> > zip_code) from employees .
Received on Fri Jul 04 2008 - 20:40:35 CEST