Newsgroups: comp.lang.smalltalk
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornellcs!newsstand.cit.cornell.edu!portc01.blue.aol.com!news-peer.gsl.net!news.gsl.net!hunter.premier.net!feed1.news.erols.com!howland.erols.net!newsfeed.internetmci.com!in3.uu.net!xyzzy!nntp
From: thomas.h.robinson@boeing.com (Thomas H. Robinson)
Subject: Re: Another Way Of Looking At Smalltalk (2)
X-Nntp-Posting-Host: gandalf.rt.cs.boeing.com
Message-ID: <MPG.ce2a26163d44dba989680@xyzzy.cs.boeing.com>
Sender: nntp@news.boeing.com (NNTP News Access)
Organization: BI&SS Research & Technology
X-Newsreader: MicroPlanet Gravity v1.01 (30 Day Trial)
References: <54n17hINNmai@topdog.cs.umbc.edu> <54p8t4$1r3@newsbf02.news.aol.com> <3277ADB8.2633@inxpress.net>
Date: Thu, 31 Oct 1996 19:48:19 GMT
Lines: 56

In article <3277ADB8.2633@inxpress.net>, drs@inxpress.net says...

> The other issue is the practical viability of performance increase under the 
> current VM-runtime models. 
> As you say, "the compilers/interpreters catch up." How do you suppose that
> will happen? Answer: it won't. I have already stated, plainly, the two
> very reasonable options for progress--emulation of dynamic systems by what are 
> really static compilation algorithms (a la MT), and hardware support. 
> Neither of these seems to be of any interest at PPD. 

Recently, I reinstalled a system that ran on the kind of hardware you're
talking about.  The system was built in Digitalk Smalltalk/V 286, and ran
on a special processor board called a "Mach board", using a special 
version of the Digitalk V-286 image.  I don't know the name of the 
company that built the board, but it's sort of moot, since the product 
isn't available anymore.

The 286 vintage machine hardware failed and was disposed of, along with 
the board, for which there was no spare.  The system is now running (in 
V-286) on a Pentium.  

> I find it amusing that guys like James (senior sales manager at a company
> based solely on a niche language, selling to a strictly niche market) fail
> to grasp the simplicity of the hardware solution....

Even if Smalltalk was a niche language (and I don't think it is) niche 
software will, in my opinion, always be more viable than niche hardware.

> ....In correspondence, he has 
> already scoffed at the MT approach as a "pipe dream" that would require 
> taking an "economic hit" to develop. But he offers no alternatives. 
> The only conclusion I can draw is that PPD has no ideas about improving 
> performance ...

You should look at the developer's conference papers on the PPD web 
site.  They indicate an awareness of performance as an issue on the list 
to be worked, I think

> ....and no plans other than "waiting for better mainstream chips
> to come along." 
> 
> Personally, I'd rather write in C++ than wait.

If your application warrants it then fine, but Smalltalk has always 
delivered ENOUGH speed for the applications I've built with it (including 
the use of sockets to talk to factory machines during processing).  You 
MIGHT be able to make a real C++ application significantly faster than 
Smalltalk, but it has to be worth the extra cost in dollars.  

There are lots of applications where the delivery time frame is short, 
the problem domain is complicated and the application is custom or the 
market is small. Then C++ simply is not an option.

FWIW,

Tom 
