Path: utzoo!attcan!uunet!lll-winken!lll-tis!ames!mailrus!bbn!rochester!pt.cs.cmu.edu!SKEF.SLISP.CS.CMU.EDU!skef
From: skef@SKEF.SLISP.CS.CMU.EDU (Skef Wholey)
Newsgroups: comp.lang.lisp
Subject: Re: Summing a list
Message-ID: <3467@pt.cs.cmu.edu>
Date: 2 Nov 88 05:37:40 GMT
References: <10794@srcsip.UUCP> <10813@srcsip.UUCP> <249@pitstop.UUCP> <200@ai.etl.army.mil>
Sender: netnews@pt.cs.cmu.edu
Organization: Carnegie-Mellon University, CS/RI
Lines: 27

In article <200@ai.etl.army.mil>, hoey@ai.etl.army.mil (Dan Hoey) writes:
> In article <29856@think.UUCP> barmar@kulla.think.com.UUCP (Barry Margolin)
> 	writes:
> >... I don't know why :KEY was left out ...
> ... The ``right'' way is
> 	(flet ((+-weight (x y)
> 		 (+ x (box-weight y))))
> 	  (reduce #'+-weight ... :initial-value 0))
> Guy has explained the omission of :KEY based on some feature of REDUCE,
> but I don't remember it being extremely convincing. ...

The reasoning is this: in all other sequence functions, the :KEY function
is applied to the operands of the test (either :TEST, :TEST-NOT, or the
PREDICATE argument to an -IF or -IF-NOT function) -- see p. 247 of CLtL.
To have :KEY do a kind of selection in REDUCE rather than a filtering
operation would have destroyed this consistency.  I don't know if this is
"extremely convincing", but there is something to be said for keeping what
consistency CL has...

Oh, my vote for doing what you'd want to do:

    (reduce #'+ (map 'list #'box-weight ...))

Any "good" compiler should "do the right thing" with this.  (Only 1/2 :-]
here -- it is not really difficult to generate good code for stuff like
that.)

--Skef