Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!tut.cis.ohio-state.edu!ucbvax!agate!shelby!neon!cm-next-9!grd
From: grd@cm-next-9.Stanford.EDU (glen diener)
Newsgroups: comp.sys.next
Subject: Re: When is Smalltalk coming to the NeXT?
Message-ID: <1990Apr10.052100.10683@Neon.Stanford.EDU>
Date: 10 Apr 90 05:21:00 GMT
References: <6461@blake.acs.washington.edu> <34620@brunix.UUCP> <10639@portia.Stanford.EDU> <4470@stpstn.UUCP>
Sender: news@Neon.Stanford.EDU (USENET News System)
Distribution: comp.sys.next
Organization: CCRMA, Stanford University
Lines: 43

In article <4470@stpstn.UUCP> cox@stpstn.UUCP (Brad Cox) writes:
>In article <10639@portia.Stanford.EDU> grd@portia.Stanford.EDU (Glendon Diener) writes:
>>What can Smalltalk do that Objective-C can't? Incrementally compile.
>>IMHO, fast, incremental compilation is a *must* for exploratory
>>programming  ---stuff deleted---
>Objective-C was never designed as a panacea, but as a tool. It was never
>intended to 'compete' with Smalltalk (IMHO, tools *never* compete,
 --stuff deleted--

     Yes, but incrementally compiling ObjC is possible
already...the proof: Common Lisp can do it already, through its
ObjC interface and foreign loading function.

   I've hacked together an ObjC app which is foreign-loaded into a 
lisp image, then launched from lisp. The app then launches
a text window and, on command, writes the text to a .m file,
creates the .h file automatically (a 7 line awk script),
and sends the .m off to cc to make the .o. Finally, a simple
lisp call foreign-loads the .o into the currently executing image.
One can define and redefine classes at will, on the fly.
There are, however, several problems with it, chiefly:
-  there is no way to remove an already linked class:
   NeXT has supplied run-time functions to 
   add classes and methods, but none to remove them.
   If one redefines a class, all objects created from that
   time on are instances of the redefined class, but the old
   class lives on, and all previously created objects
   are still instances of it. (It is possible to go through these
   old objects and redirect their 'isa' pointers to the
   new class, but I haven't figured out what  pointers 
   to change in order to redefine a class somewhere up 
   the superclass chain of one of these objects--any suggestions?).
-  the application is tied to lisp.  This has its pros and cons,
   but I find the greatest disadvantage is that I can't run
   the monster into gdb.

  Anyway, given that Franz have proven incremental compilation of
ObjC is `do-able', wouldn't you like to see the efficiency of
ObjC delivered in a Smalltalk-80 or LISP flavors-style environment?
That's the tool I'd like to have.

-glen diener
grd@cm-next-9.stanford.edu