Path: archiver1.google.com!news1.google.com!newsfeed.stanford.edu!canoe.uoregon.edu!logbridge.uoregon.edu!tethys.csu.net!csulb.edu!news.service.uci.edu!not-for-mail
From: tb+use...@becket.net (Thomas Bushnell, BSG)
Newsgroups: comp.lang.lisp,comp.lang.scheme
Subject: Re: self-hosting gc, narrowed
Date: 13 Mar 2002 14:58:16 -0800
Organization: University of California, Irvine
Lines: 25
Sender: t...@becket.becket.net
Message-ID: <878z8woz1z.fsf@becket.becket.net>
References: <87y9h24r1u.fsf@becket.becket.net> <876644n9h7.fsf@becket.becket.net> <LGQi8.10344$44.2580687@typhoon.ne.ipsvc.net> <87r8msrrlg.fsf@becket.becket.net> <1015951132.108200@haldjas.folklore.ee> <874rjlogn0.fsf@becket.becket.net> <1015964338.871766@haldjas.folklore.ee> <87lmcxovhy.fsf@becket.becket.net> <1016026028.852976@haldjas.folklore.ee> <joLj8.20461$44.4681068@typhoon.ne.ipsvc.net> <1016044316.907708@haldjas.folklore.ee> <87d6y8qhyf.fsf@becket.becket.net> <1016059644.126156@haldjas.folklore.ee>
NNTP-Posting-Host: vp190174.reshsg.uci.edu
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Reply-Permission: Posted or emailed replies to this message constitute
		 permission for an emailed response.
X-PGP-Fingerprint: 1F0A1E51  63 28 EB DA E6 44 E5 5E  EC F3 04 26 4E BF 1A 92
X-Windows: A terminal disease.
User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.1

Sander Vesik <san...@haldjas.folklore.ee> writes:

> In comp.lang.scheme Thomas Bushnell, BSG <tb+use...@becket.net> wrote:
> > Sander Vesik <san...@haldjas.folklore.ee> writes:

> >> Actually, if the OS cannot trust (but needs to), then allowing
> >> for pointers outside of that area is a bug. Because the user can
> >> then effectively change the data dispite the fact that the OS
> >> supposedly secured it.
> > 
> > No, no, you don't really understand the permission model.  The OS
> > *knows* that the user can "change the data", and that's not harmful at
> > all, as long as the OS exercises proper care when reading it.
> 
> How can I understand a permission model that has not been presented? 8-)

Here's the real clincher.

The gc needs access to primitives that can destroy the memory model of
the system if misused.

If everyone is sharing memory, then the gc is able to hose everybody.

Accordingly, the gc must be privileged.