From owner-freebsd-current  Tue Dec 19 23:16:23 1995
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id XAA03437
          for current-outgoing; Tue, 19 Dec 1995 23:16:23 -0800 (PST)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id XAA03428
          for <current@freebsd.org>; Tue, 19 Dec 1995 23:16:14 -0800 (PST)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id SAA21287; Wed, 20 Dec 1995 18:15:48 +1100
Date: Wed, 20 Dec 1995 18:15:48 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199512200715.SAA21287@godzilla.zeta.org.au>
To: current@freebsd.org, nate@rocky.sri.MT.net
Subject: Re: Mis-feature in -current
Sender: owner-current@freebsd.org
Precedence: bulk

>This one was a strange one, but I finally figured it out.  I'm trying to
>build kernels which contain all of the necessary 'debug' symbols in
>them, so I'm not compiling the code with '-O'.

Kernels compiled with any level of optimization (and -g) should contain
all the necessary debug symbols.

>However, I've noticed that if you don't compile the code with '-O', gcc
>no longer optimizes out the calls to strlen.  What this means is that
>the kernel will no longer link if you compile any modules which uses
>strlen().

I found another interesting case related to this.  ipx_input() uses an
inefficient auto initializer for ipx_ipx.  gcc-2.7.0 makes a static copy
of the initial data and copies it to the variable a word at a time, but
gcc-2.7.2 pessimizes the initialization to two assignments and a memset
call.  Linking fails because memset isn't in the library.

Bruce