Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sun-barr!cronkite!exodus!sbb
From: sbb@laplace.eng.sun.com (Steve Byrne)
Newsgroups: comp.lang.smalltalk
Subject: Re: GNU Smalltalk-1.1 problems
Message-ID: <SBB.91Jun20190008@laplace.eng.sun.com>
Date: 21 Jun 91 02:00:08 GMT
References: <PETERS.91Jun19104522@yates.UVic.CA>
Sender: news@exodus.Eng.Sun.COM
Distribution: comp.lang.smalltalk
Organization: FSF hackers, Smalltalk division
Lines: 87
In-reply-to: peters@sirius.UVic.CA's message of 19 Jun 91 17:44:19 GMT

In article <PETERS.91Jun19104522@yates.UVic.CA> peters@sirius.UVic.CA (Doug  Peters) writes:


   From: peters@sirius.UVic.CA (Doug  Peters)
   Newsgroups: comp.lang.smalltalk
   Date: 19 Jun 91 17:44:19 GMT

   Granted, I am almost a novice at Smalltalk, but I thought that I was
   asking GNU Smalltalk-1.1 to do some pretty basic things...

   For example, I have a Complex class (I chose to make this a subclass
   of Number because I wanted "isKindOf: Number" to still work)
   So I have a file called "Complex.st", which looks like:
   "------------------------------------------------>8"
   Number variableSubclass: #Complex
	   instanceVariableNames: 're im '
	   classVariableNames: ''
	   poolDictionaries: ''
	   category: ''!

   !Complex methodsFor:'accessing'!

   re: aNumber im: anotherNumber
     re := aNumber.
     im := anotherNumber! !

   !Complex class methodsFor: 'instance creation'!

   re:aNumber im:anotherNumber "<- this is line 15"
     | a |
     a := super new.
     a re:aNumber im:anotherNumber. "<- this is line 18"
     ^a! !
   "------------------------------------------------>8"

   st> FileStream fileIn:'Complex.st' !
   produces:

   "Complex.st", line 15: parse error
   "Complex.st", line 15: Invalid message pattern
   "Complex.st", line 18: Error in expression
   "Complex.st", line 18: parse error
   "Complex.st", line 18: Invalid message pattern
   "Complex.st", line 19: Error in expression

The problem here is that GNU Smalltalk's lexer is greedy: it really thinks
that if it's seen "foo:bar" that it's processing a keyword symbol and is
expecting to see a colon following it.  Unfortunately, there is no
colon at the end, so it yells.  You can argue with whether this is the
right behavior or not; I think the Blue Book indicated that this was the way
things were, but it could probably be a little smarter in the future (like,
say the 2.0 timeframe when the compiler code is written in Smalltalk). 

In this case, if you just put a space between the keyword and the argument
name, you should win.

   st> ArrayedCollection variableSubclass: #Matrix
   st> 	instanceVariableNames: 'nrow ncol '
   st> 	classVariableNames: ''
   st> 	poolDictionaries: ''
   st>	category: ''!

   and I get:

   ArrayedCollection error: cannot create a variable subclass from a
   non-pointer variable parent class

I've changed things related to this in 1.2.  The basic problem is that
isWords and isBytes are lying: isWords should only return true if it's not
pointers AND is has the proper bit set.  isBytes should not be a pointer
and should not be words.  These two methods are used by variableSubclass
for error detection, although the way they are implemented now, they're
more involved with error creation :-)

Change the definitions in Behavior.st of isWords and isBytes thus:

    isBytes
	^self isPointers not & self isWords not
    !

    isWords
	^self isPointers not & ((self instanceSpec bitAt: 29) ~= 0)
    !

And you'll win.

Steve