Received: (at maintonly) by bugs.debian.org; 25 May 1997 17:05:26 +0000
Received: (qmail 6057 invoked from network); 25 May 1997 17:05:24 -0000
Received: from nhv-ct4-24.ix.netcom.com (HELO sauron.fashion-technology.com) (mail@205.184.152.152)
  by 206.190.143.161 with SMTP; 25 May 1997 17:05:24 -0000
Received: from klee by sauron.fashion-technology.com with local (Exim 1.62 #1)
	id 0wVgjN-0004rJ-00 (Debian); Sun, 25 May 1997 13:05:17 -0400
From: Klee Dienes <klee@mit.edu>
Reply-To: Klee Dienes <klee@mit.edu>
To: maintonly@bugs.debian.org
Subject: gcc issues
Message-Id: <E0wVgjN-0004rJ-00@sauron.fashion-technology.com>
Sender: Klee Dienes <klee@debian.org>
Date: Sun, 25 May 1997 13:05:17 -0400


Package: gcc

I don't have time to track this down before I leave, unfortunately,
but:

Using the latest version of gcc with libc5 causes it to build programs
that dump core immediately, at least with c++.  I like the fact that
gcc doesn't depend on a particular libc-dev (sometimes one really does
use a compiler without a libc, if rarely), but this is still a serious
trap for the unwary.  Perhaps gcc should be configured to recommend
libc6-dev or maybe conflict with libc5-dev?  Or at least a runtime
check to prevent it from making bad binaries?

In addition, it's easy to end up with a version of cpp that doesn't
match the version of gcc, leading to a gcc that can't find 'stddef.h',
among other things.  This should probably be fixed by changing gcc to
explicitly pass all directory arguments to cpp rather than have it
know where gcc keeps its header files.