Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!purdue!bouma
From: bouma@cs.purdue.EDU (William J. Bouma)
Newsgroups: comp.lang.forth
Subject: Re: Recursion (was Re: The most fantastically fun word...)
Summary: handling recursion in FORTH
Message-ID: <6730@medusa.cs.purdue.edu>
Date: 9 May 89 20:40:35 GMT
References: <6716@medusa.cs.purdue.edu> <5096@tekgvs.LABS.TEK.COM>
Distribution: usa
Organization: Department of Computer Science, Purdue University
Lines: 41

In article <5096@tekgvs.LABS.TEK.COM> toma@tekgvs.LABS.TEK.COM (Tom Almy) writes:
>In article <6716@medusa.cs.purdue.edu> bouma@cs.purdue.EDU (William J. Bouma) writes:
>>standard. It was really stupid , since all I had to do to make recursion
>>work in the thing was to put the word being defined into the dictionary as
>>soon as it was seen, rather than after the definition was complete. The
>
>
>There is a good reason *not* to make a word be visible in the dictionary
>before it is seen -- sometimes it is nice to be able to redefine a
>word in terms of itself.  As an example:
>
>: /  DUP 0= IF ." ATTEMPT TO DIVIDE " DROP . ." BY ZERO"  ABORT THEN  / ;
>

   I do not agree with this reasoning. It seems like redefining a word in
   terms of itself is a more unusual case than is recursion. Thus it should
   have to suffer the unnatural syntax rather than the recursion. In my
   system I do it like this:

: /  DUP 0= IF ." ATTEMPT TO DIVIDE " DROP . ." BY ZERO"  ABORT THEN  '' / ;

   where '' just grabs the next word and gets its old definition. Of course
   either way will work, but I see no "good reason" not to do it my way.

>There is a nice feature in Laboratory Microsystem's Forths (and others)
>that allows forward referencing of colon definitions and thus mutually
>recursive words.  It is used in this way:
>
>F: WORD-B
>: WORD-A   ... WORD-B ... ;
>R: WORD-B  ... WORD-A ... ;

   Durring a colon definition, I just put all undefined words in the dictionary
   with their code fields pointing at code that traps an error and prints
   " undefined word executed WORD ". This eliminates the need for any special
   mechanism for mutual recursion. When WORD-B is finaly defined, the code
   pointer is just changed to point at the current code (as any normal re-
   definition).

-- 
Bill <bouma@cs.purdue.edu>  ||  ...!purdue!bouma