From owner-freebsd-current  Mon Dec  8 11:25:37 1997
Return-Path: <owner-freebsd-current>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.7/8.8.7) id LAA01923
          for current-outgoing; Mon, 8 Dec 1997 11:25:37 -0800 (PST)
          (envelope-from owner-freebsd-current)
Received: from watermarkgroup.com (lor.watermarkgroup.com [207.202.73.33])
          by hub.freebsd.org (8.8.7/8.8.7) with SMTP id LAA01894
          for <current@freebsd.org>; Mon, 8 Dec 1997 11:25:27 -0800 (PST)
          (envelope-from luoqi@watermarkgroup.com)
Received: by watermarkgroup.com (4.1/SMI-4.1)
	id AA12202; Mon, 8 Dec 97 14:24:56 EST
Date: Mon, 8 Dec 97 14:24:56 EST
From: luoqi@watermarkgroup.com (Luoqi Chen)
Message-Id: <9712081924.AA12202@watermarkgroup.com>
To: jfieber@indiana.edu
Subject: Re: Oddity with netscape and current
Cc: current@freebsd.org
Sender: owner-freebsd-current@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

> For the record, the problem which this solves/works-around exists
> in FreeBSD-2.2.5/XFree86-3.3.1 as well as (I gather) current (and
> what version of X?). I'm using a 16 bit display and have not
> tried it on 8 or 24.  The reports on what works with what
> combinations of FreeBSD, Netscape, and X11 have been rather
> inconsistent--some curious methods of suppressing the symptoms
> have been discovered but does anyone have a clue about the
> problem? 
This has something to do with floating point exceptions. Netscape
uses the fbsd default exception mask (precision/underflow/denomrmals),
and *ignores* SIGFPE signal, which means when an unmasked exception
arises, the FPU register stack will be completely messed up, and sometime
this results in an infinite loop. To fix this problem, Netscape should either:
1. trap SIGFPE signals and handle the exceptions correctly, or
2. ignore SIGFPE, but use fpsetmask to mask ALL fp exceptions.

Before Netscape fixes this problem, all we can do is to try to keep
these unwanted exceptions from happening. Apparently, Netscape goes
into an infinite loop when it does some CMS color lookup during startup.
That's why I proposed to disable CMS by setting CMSDB to null.

-lq
> 
> -john
> 
>