[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[no subject]
- To: henry at MIT-AI
- From: Guy Steele <GLS at SU-AI>
- Date: Tue, 12 Jun 79 21:04:00 GMT
- Cc: GLS at SU-AI, bug-lisp at MIT-AI, bug-lispm at MIT-AI
- Original-date: 12 Jun 1979 1404-PDT
11-Jun-79 1907 HENRY at MIT-AI (Henry Lieberman) CASEGO
Date: 11 JUN 1979 2209-EDT
From: HENRY at MIT-AI (Henry Lieberman)
Subject: CASEGO
To: GLS at SU-AI
What you really have in the examples you give to show
the use of CASEGO is a loop. I think it should be coded
as such, either by the use of DO or auxiliary functions
rather than your proposed CASEGO.
(DEFUN CASEGO-EXAMPLE (THE-FROB)
(CASEQ THE-FROB
((BAR) ...)
((FOO))
(T (... "I'll assume BAR") (CASEGO-EXAMPLE 'BAR))))
I think your proposal is like coding a loop using GO, which
is generally acknowledged to be less winning.
This is true, although what you have written is not semantically
equivalent (though functionally equivalent) in that it re-performs
the dispatch needlessly. This corresponds to the situation with the
famous Boehm-Jacopini "theorem", which shows that any program can
be made "structured" provided that a number of boolean variables
and redundant tests are introduced. Also, without LABEL or LABELS
or some such I can't write the above in MacLISP "in place".
A better way around the problem using functions would be just to do
(CASEQ THE-FROB
((BAR) (BAR-CASE))
((FOO) ...)
...
(T (... "assume BAR") (BAR-CASE)))
...
(DEFUN BAR-CASE ...)
Again, in MacLISP I can't really write this "in place" and take
advantage of local variables.
I just used the name GO by analogy with the behavior in a PROG.
Actually, I might phrase my request entirely without reference to
GO (because some people dislike GO), and say that I want a construct
CASERETRY such that
(CASEQ FROB
((FOO) FOO1)
((BAR) BAR1 (CASERETRY HUNOZ))
((BAZ QUUX) BAZQUUX1)
(T BARF (CASERETRY 'BAR))
expands into
(LABELS ((FOO-CASE (LAMBDA () FOO1))
(BAR-CASE (LAMBDA () BAR1 (DISPATCH HUNOZ)))
(BAZQUUX-CASE (LAMBDA () BAZQUUX1))
(DEFAULT-CASE (LAMBDA () BARF (BAR-CASE)))
(DISPATCH (LAMBDA (X)
(SELECTQ X
((FOO) (FOO-CASE))
((BAR) (BAR-CASE))
((BAZ QUUX) (BAZQUUX-CASE))
(T (DEFAULT-CASE))))))
(DISPATCH FROB))
where SELECTQ represents the more primitive simple-dispatch operation.
Now this uses no GOs, and therefore must be generally acknowledged to be
more winning, right?