From: KD

To:  Editor

Date: 21 Feb 2004

 

In Date and Darwin's DATABASE FOUNDATIONS paper #5 Multiple Assignment they use the PL/1 term "pseudovariable"  (aside:  elsewhere some DBMS vendors use the term ‘pseudocolumn’ to refer to such system-generated values as the system time).  Given their survey of "pseudovariable" language capabilities spanning Algol (a beautiful language), APL, Cobol, Fortran, POP-2 and SQL, the most powerful of these perspicuous variable value setting capabilities is omitted--Common Lisp.

 

In general (SETF x y) can be understood as saying 'see to it that x evaluates to y'.  As a macro SETF can look inside its arguments to see what needs to be done to make such a statement true.  If the first argument (after macroexpansion) is a symbol, the SETF just expands into a SETQ. But if the first argument is a query, the SETF expands into the corresponding assertion ... This transformation from a query to assertion is called inversion.  All the most frequently used Common Lisp access functions have predefined inversions.” 

--Paul Graham, On Lisp, pp. 165-6]

 

 

Graham proceeds in this chapter on Generalized Variables to do the following, in summary:

 

·    Build a 'toggle' macro atop SETF.

·    Discuss the 'multiple evaluation [assignment] problem' due to side effects.

·    Present various useful utilities built upon the SETF related support macros including memorization.

·    Describe how to create user-defined inversions using DEFSETF.

 

Although Graham's exposition does not address subtleties such as atomicity w.r.t. constraints, he does demonstrate some useful meta-programming capabilities that should be embraced in the database community.

 

A non-tutorial description of CL's SETF can be found at:

 

http://www-2.cs.cmu.edu/Groups/AI/html/cltl/cltl2.html

 

specifically: http://www-2.cs.cmu.edu/Groups/AI/html/cltl/clm/node80.html

 

I suggest that you consider deprecating the term ‘pseudovariable’ in favor of ‘Generalized Variable’.

 

 

From: Hugh Darwen

To: KD

 

I must say I have felt rather uncomfortable with the term pseudovariable, but Date was a teacher of PL/I before he discovered Codd's work, and I haven't thought of a better term.  When I think about it carefully, it's possibly not that bad really.  A pseudo-X is something that is not an X but is masquerading as one.  I could argue that a pseudovariable is a nontrivial expression masquerading as a variable.

 

 

Posted 04/16/04