From owner-freebsd-current  Wed Jul 31 04:42:23 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id EAA24391
          for current-outgoing; Wed, 31 Jul 1996 04:42:23 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id EAA24383
          for <freebsd-current@FreeBSD.ORG>; Wed, 31 Jul 1996 04:42:13 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id VAA24793; Wed, 31 Jul 1996 21:34:37 +1000
Date: Wed, 31 Jul 1996 21:34:37 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199607311134.VAA24793@godzilla.zeta.org.au>
To: freebsd-current@FreeBSD.ORG, syssgm@devetir.qld.gov.au
Subject: Re: `const char rcsid[]' vs -traditional
Sender: owner-current@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>I find:
>	static const char rcsid[] = ...

>How long will it be before the ANSI compilers start discarding unreferenced
>static data?  Or is "rcsid" already a special case in gcc?  :-)

There's nothing to stop ANSI compilers from discarding unreferenced data,
but declaring static variables as `const' is a documented way of stopping
gcc from discarding them.  At least, the following macro is a documented
way for gcc-2.6.3:

	#define USE(var) \
	  static void *const use_##var = (&use_##var, &var, 0)

This method also works for functions.  It is used in MAKE_SET() in
<sys/kernel.h>.

Unfortunately, gcc with maximal warnings generates many warnings for
this macro.  E.g., when it is applied to

	static int x;
	USE(x);

z.c:4: warning: initialization makes pointer from integer without a cast
z.c:4: initializer element is not constant

There are even more warnings (which actually occur thousands of times for
compiling the kernel) for gcc-2.7.2.  The USE() macro is no longer documented
for gcc-2.7.2; you are supposed to use the `unused' attribute instead:

	static const int x __attribute__((__unused__));

The `unused' attribute doesn't work for gcc-2.6.3 so it will need to be
used in a macro that expands to the attribute version only for recent
versions of gcc.

The `unused' attribute doesn't work completely for functions:

	static void x(void) __attribute__((__unused__));
	static void x(void) {}

This works right for -O2 but the function still gets deleted for -O3.
The function probably needs to referenced via a pointer to keep it.
MAKE_SET() fails to do this for the function pointer arg because the
function pointer arg is only referenced in assembler.

>Should we switch to #ident?  It doesn't do anything useful on 2.1.5, but
>we could fix that.

>	#ident "$Id: frobble.c,v ..."

>What does the C standard say about #ident?  Has anyone put a copy of the
>standard on the web?

#ident is a syntax error.  The standard won't be put on the web until its
owners figure out how to charge for it there :-(.

Anyway, we shouldn't use #ident because it is impossible to hide the details
in a macro.

Bruce