NetBSD Problem Report #48303

From www@NetBSD.org  Sat Oct 12 10:03:30 2013
Return-Path: <www@NetBSD.org>
Received: from mail.netbsd.org (mail.netbsd.org [149.20.53.66])
	(using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits))
	(Client CN "mail.NetBSD.org", Issuer "Postmaster NetBSD.org" (verified OK))
	by mollari.NetBSD.org (Postfix) with ESMTPS id 7945F7263E
	for <gnats-bugs@gnats.NetBSD.org>; Sat, 12 Oct 2013 10:03:30 +0000 (UTC)
Message-Id: <20131012100329.3750A72642@mollari.NetBSD.org>
Date: Sat, 12 Oct 2013 10:03:29 +0000 (UTC)
From: justin@specialbusservice.com
Reply-To: justin@specialbusservice.com
To: gnats-bugs@NetBSD.org
Subject: Linux cross build fails on ppc
X-Send-Pr-Version: www-1.0

>Number:         48303
>Category:       toolchain
>Synopsis:       Linux cross build fails on ppc
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    toolchain-manager
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Oct 12 10:05:01 +0000 2013
>Last-Modified:  Mon Oct 28 13:10:01 +0000 2013
>Originator:     Justin Cormack
>Release:        cvs head
>Organization:
>Environment:
Linux pike 3.6.7-4.fc17.ppc64 #1 SMP Thu Dec 6 06:41:58 MST 2012 ppc64 ppc64 ppc64 GNU/Linux

>Description:
I am trying to cross build NetBSD on ppc:
./build.sh -m evbppc build

This fails with

#   compile  compat/raise_default_signal.lo
cc -O  -I. -I./include -I/home/justin/src/tools/compat -I/home/justin/src/tools/compat/sys  -DHAVE_NBTOOL_CONFIG_H=1 -D_FILE_OFFSET_BITS=64 -D__DBINTERFACE_PRIVATE -c -o raise_default_signal.lo.o    /home/justin/src/tools/compat/../../lib/libutil/raise_default_signal.c
In file included from /usr/include/signal.h:359:0,
                 from /home/justin/src/tools/compat/../../lib/libutil/raise_default_signal.c:42:
/usr/include/sys/ucontext.h:101:16: error: expected identifier or '(' before '[' token

*** Failed target:  raise_default_signal.lo


The issue in that that Linux header has a field called __unused:

typedef struct {
        unsigned long   __unused[4];

(which Linux does quite a bit); however __unused has been #defined to an empty string so it will not compile.

>How-To-Repeat:
Cross build from recent ppc linux.

>Fix:
I would be surprised if this hasnt been worked around elsewhere, as Linux uses __unused a fair amount in headers. It would seem best to use something else to avoid collisions.

>Release-Note:

>Audit-Trail:
From: Justin Cormack <justin@specialbusservice.com>
To: gnats-bugs@netbsd.org, tech-toolchain@netbsd.org
Cc: 
Subject: Re: install/48303: Linux cross build fails on ppc
Date: Mon, 28 Oct 2013 10:32:54 +0000

 I looked at where this was being redefined and it is in compat_defs.h
 and the following patch fixes this issue, which makes it somewhat
 simpler than I thought to fix. The only question is whether undefining
 __unused is necessary on some other platform, in which case the
 #define __unused could be wrapped eg in #ifndef __linux__ which should
 be safe.

 cvs diff -u tools/compat/compat_defs.h
 Index: tools/compat/compat_defs.h
 ===================================================================
 RCS file: /cvsroot/src/tools/compat/compat_defs.h,v
 retrieving revision 1.93
 diff -u -r1.93 compat_defs.h
 --- tools/compat/compat_defs.h 24 Oct 2013 13:59:47 -0000 1.93
 +++ tools/compat/compat_defs.h 28 Oct 2013 10:26:16 -0000
 @@ -130,7 +130,6 @@
  #undef __restrict
  #define __restrict
  #undef __unused
 -#define __unused
  #undef __arraycount
  #define __arraycount(__x) (sizeof(__x) / sizeof(__x[0]))
  #undef __USE



 On Sat, Oct 12, 2013 at 11:05 AM,  <gnats-admin@netbsd.org> wrote:
 > Thank you very much for your problem report.
 > It has the internal identification `install/48303'.
 > The individual assigned to look at your
 > report is: install-manager.
 >
 >>Category:       install
 >>Responsible:    install-manager
 >>Synopsis:       Linux cross build fails on ppc
 >>Arrival-Date:   Sat Oct 12 10:05:01 +0000 2013
 >

Responsible-Changed-From-To: install-manager->toolchain-manager
Responsible-Changed-By: martin@NetBSD.org
Responsible-Changed-When: Mon, 28 Oct 2013 10:49:09 +0000
Responsible-Changed-Why:
Not an installer issue


From: Alan Barrett <apb@cequrux.com>
To: gnats-bugs@netbsd.org
Cc: 
Subject: Re: install/48303: Linux cross build fails on ppc
Date: Mon, 28 Oct 2013 14:57:38 +0200

 On Mon, 28 Oct 2013, Justin Cormack wrote:
 >I looked at where this was being redefined and it is in compat_defs.h
 >and the following patch fixes this issue, which makes it somewhat
 >simpler than I thought to fix. The only question is whether undefining
 >__unused is necessary on some other platform, in which case the
 >#define __unused could be wrapped eg in #ifndef __linux__ which should
 >be safe.

 Is the Linux use of "__unused" as a struct member name something new?

 Did you test a complete build with your patch?  The reason for the
 "#define __unused", which has been in compat_defs.h since 2006-10-12, is
 to deal with code that uses __unused in the NetBSD way, like

 	sometype varname __unused;

 and if __unused is not defined then such code would cause a syntax
 error.  There seems to be such code in src/usr.sbin/makefs, which would
 be built from src/tools/makefs.  There might also be more uses of
 __unused in the tools build.

 --apb (Alan Barrett)

From: Justin Cormack <justin@specialbusservice.com>
To: gnats-bugs@netbsd.org
Cc: toolchain-manager@netbsd.org, gnats-admin@netbsd.org, 
	netbsd-bugs@netbsd.org
Subject: Re: install/48303: Linux cross build fails on ppc
Date: Mon, 28 Oct 2013 13:07:39 +0000

 On Mon, Oct 28, 2013 at 1:00 PM, Alan Barrett <apb@cequrux.com> wrote:
 > The following reply was made to PR toolchain/48303; it has been noted by GNATS.
 >
 > From: Alan Barrett <apb@cequrux.com>
 > To: gnats-bugs@netbsd.org
 > Cc:
 > Subject: Re: install/48303: Linux cross build fails on ppc
 > Date: Mon, 28 Oct 2013 14:57:38 +0200
 >
 >  On Mon, 28 Oct 2013, Justin Cormack wrote:
 >  >I looked at where this was being redefined and it is in compat_defs.h
 >  >and the following patch fixes this issue, which makes it somewhat
 >  >simpler than I thought to fix. The only question is whether undefining
 >  >__unused is necessary on some other platform, in which case the
 >  >#define __unused could be wrapped eg in #ifndef __linux__ which should
 >  >be safe.
 >
 >  Is the Linux use of "__unused" as a struct member name something new?

 It is generally only found in non-x86 architectures, although it is
 also in some versions of struct stat on x86.

 >  Did you test a complete build with your patch?  The reason for the
 >  "#define __unused", which has been in compat_defs.h since 2006-10-12, is
 >  to deal with code that uses __unused in the NetBSD way, like
 >
 >         sometype varname __unused;
 >
 >  and if __unused is not defined then such code would cause a syntax
 >  error.  There seems to be such code in src/usr.sbin/makefs, which would
 >  be built from src/tools/makefs.  There might also be more uses of
 >  __unused in the tools build.

 Yes, apologies this patch will not work, the build eventually fails.

 I have filed a patch against glibc
 https://sourceware.org/ml/libc-alpha/2013-10/msg00870.html not sure if
 they will respond. It will still take a few years to filter to end
 user machines, so perhaps at least a note on the README in the compat
 directory could be helpful if there is really no way to fix this.

 Justin

>Unformatted:

Home	PR Database Search