xen: Don't use -nostdinc flags with CLANG
PATCH 1/1 xen: Don't use -nostdinc flags with CLANG
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from /home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file not found with <angled> include; use "quotes" instead
^~~~~~~~~~
"stdarg.h"
In file included from sched_sedf.c:8:
/home/julieng/works/xen/xen/include/xen/lib.h:101:63: error: unknown type name 'va_list'
extern int vsnprintf(char *buf, size_t size, const char *fmt, va_list args)
^/home/julieng/works/xen/xen/include/xen/lib.h:105:64: error: unknown type name 'va_list' extern int vscnprintf(char *buf, size_t size, const char *fmt, va_list args)
I have the same errors on different version of clang:
- clang 3.0 on debian wheezy
- clang 3.3 on Fedora 20
- clang 3.5 build from trunk
Removing -nostdinc fix the build on clang.
--- xen/Rules.mk | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
xen/Rules.mk +2 -1
@@ -46,7 +46,8 @@ CFLAGS += -Werror -Wredundant-decls -Wno-pointer-arith CFLAGS += -pipe -g -D__XEN__ -include $(BASEDIR)/include/xen/config.h # Solaris puts stdarg.h &c in the system include directory. ifneq ($(XEN_OS),SunOS) -CFLAGS += -nostdinc -iwithprefix include +CFLAGS-y += -iwithprefix include +CFLAGS-$(gcc) += -nostdinc endif CFLAGS-$(XSM_ENABLE) += -DXSM_ENABLE
Signature · 1 line
1.8.5.3
Quoted 20 lines
Quoted 1 line
Quoted 1 line · Julien Grall
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from/home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file
not found with <angled> include; use "quotes" instead^~~~~~~~~~ "stdarg.h"In file included from sched_sedf.c:8:
/home/julieng/works/xen/xen/include/xen/lib.h:101:63: error: unknown type name 'va_list'
extern intvsnprintf(char *buf, size_t size, const char *fmt, va_list args)^
/home/julieng/works/xen/xen/include/xen/lib.h:105:64: error: unknown type name 'va_list'
extern intvscnprintf(char *buf, size_t size, const char *fmt, va_list args)
Quoted 6 lines
I have the same errors on different version of clang:
- clang 3.0 on debian wheezy
- clang 3.3 on Fedora 20
- clang 3.5 build from trunkRemoving -nostdinc fix the build on clang.
But does this also do the right thing? I.e. I doubt you're immune then against picking up headers you don't want to include in a hypervisor build, or properly failing if - for whatever reason, e.g. during development after having made a mistake - a header can't be found in the paths we want the compiler to look for them, but can be found in a "standard" include directory.
IOW I think it first needs to be understood/explained why and by how much clang behavior differs here.
Jan
Quoted 27 lines
Signed-off-by Julien Grall <[email protected]>--- xen/Rules.mk | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)xen/Rules.mk +2
-1@@ -46,7 +46,8 @@ CFLAGS += -Werror -Wredundant-decls -Wno-pointer-arith CFLAGS += -pipe -g -D__XEN__ -include $(BASEDIR)/include/xen/config.h # Solaris puts stdarg.h &c in the system include directory. ifneq ($(XEN_OS),SunOS) -CFLAGS += -nostdinc -iwithprefix include +CFLAGS-y += -iwithprefix include +CFLAGS-$(gcc) += -nostdinc endif CFLAGS-$(XSM_ENABLE) += -DXSM_ENABLESignature · 7 lines
1.8.5.3 _______________________________________________ Xen-devel mailing list [email protected] http://lists.xen.org/xen-devel
Quoted 9 lines · Julien Grall
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from/home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file not found with <angled> include; use "quotes" instead^~~~~~~~~~ "stdarg.h"
Looks like on your system stdarg.h doesn't live in a compiler-specific path, like we have for the BSDs. I think we should just go to using our own definitions for stdarg/stdbool everywhere; trying to chase the compiler-specific versions around is a PITA, and the pieces we actually need are trivial.
Tim.
Quoted 11 lines · Tim Deegan
Quoted 9 lines · Julien Grall
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from/home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file not found with <angled> include; use "quotes" instead^~~~~~~~~~ "stdarg.h"
Quoted 5 lines
Looks like on your system
stdarg.hdoesn't live in a compiler-specific path, like we have for the BSDs. I think we should just go to using our own definitions for stdarg/stdbool everywhere; trying to chase the compiler-specific versions around is a PITA, and the pieces we actually need are trivial.
For BSDs, we are using our own stdargs/stdbool. So we don't include the system <stdarg.h>.
Linux is using $(CC) -print-file-name=include to get the right path. It works with both gcc and clang on Linux distos, but not on FreeBSD.
Signature · 1 line
Julien Grall
Quoted 20 lines · Julien Grall
Quoted 11 lines · Tim Deegan
Quoted 9 lines · Julien Grall
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from/home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file not found with <angled> include; use "quotes" instead^~~~~~~~~~ "stdarg.h"Quoted 5 lines
Looks like on your system
stdarg.hdoesn't live in a compiler-specific path, like we have for the BSDs. I think we should just go to using our own definitions for stdarg/stdbool everywhere; trying to chase the compiler-specific versions around is a PITA, and the pieces we actually need are trivial.
Quoted 5 lines
For BSDs, we are using our own stdargs/stdbool. So we don't include the system <
stdarg.h>.Linux is using $(CC) -print-file-name=include to get the right path. It works with both gcc and clang on Linux distos, but not on FreeBSD.
Wait - is the error message you posted from clang on FreeBSD? That's surprising; on FreeBSD xen/stdarg.h shouldn't be trying to include <stdarg.h> at all. Is __FreeBSD__ not being defined?
Tim.
Quoted 24 lines · Tim Deegan
Quoted 23 lines · Julien Grall
Quoted 11 lines · Tim Deegan
Quoted 9 lines · Julien Grall
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from/home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file not found with <angled> include; use "quotes" instead^~~~~~~~~~ "stdarg.h"Quoted 5 lines
Looks like on your system
stdarg.hdoesn't live in a compiler-specific path, like we have for the BSDs. I think we should just go to using our own definitions for stdarg/stdbool everywhere; trying to chase the compiler-specific versions around is a PITA, and the pieces we actually need are trivial.For BSDs, we are using our own stdargs/stdbool. So we don't include the system <
stdarg.h>.
Quoted 6 lines
Quoted 2 lines
Linux is using $(CC) -print-file-name=include to get the right path. It works with both gcc and clang on Linux distos, but not on FreeBSD.
Wait - is the error message you posted from clang on FreeBSD? That's surprising; on FreeBSD
xen/stdarg.hshouldn't be trying to include <stdarg.h> at all. Is__FreeBSD__not being defined?
No it's from Linux (Debian Wheezy and Fedora). I just gave a try to the "-print-file-name" solution on FreeBSD.
Signature · 1 line
Julien Grall
Quoted 30 lines · Julien Grall
Quoted 27 lines · Tim Deegan
Quoted 25 lines · Julien Grall
Quoted 11 lines · Tim Deegan
Quoted 9 lines · Julien Grall
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from/home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file not found with <angled> include; use "quotes" instead^~~~~~~~~~ "stdarg.h"Quoted 5 lines
Looks like on your system
stdarg.hdoesn't live in a compiler-specific path, like we have for the BSDs. I think we should just go to using our own definitions for stdarg/stdbool everywhere; trying to chase the compiler-specific versions around is a PITA, and the pieces we actually need are trivial.For BSDs, we are using our own stdargs/stdbool. So we don't include the system <
stdarg.h>.Linux is using $(CC) -print-file-name=include to get the right path. It works with both gcc and clang on Linux distos, but not on FreeBSD.
Quoted 7 lines
Quoted 3 lines
Wait - is the error message you posted from clang on FreeBSD? That's surprising; on FreeBSD
xen/stdarg.hshouldn't be trying to include <stdarg.h> at all. Is__FreeBSD__not being defined?No it's from Linux (Debian Wheezy and Fedora). I just gave a try to the "-print-file-name" solution on FreeBSD.
Oh, OK. Yeah, we knew that didn't work there, because on *BSD the compiler-specific headers like stdarg.h actually live in /usr/include and can themselves include other system headers. That's why we have our own implementation of the bits we need, that we use on BSD.
Are you using a very old version of clang? As 06a9c7e points out, our current runes didn't work before clang v3.0.
If not, rather than chasing this around any further, I think we should abandon trying to use the compiler-provided headers even on linux. Does this patch fix your build issue?
commit e7003f174e0df9192dde6fa8d33b0a20f99ce053
Author: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000
xen: stop trying to use the system <stdarg.h> and <stdbool.h>
We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in /usr/include.
On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
xen/include/xen/stdarg.h +14 -16
@@ -1,23 +1,21 @@#ifndef __XEN_STDARG_H__#define __XEN_STDARG_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -typedef __builtin_va_list va_list ; -# ifdef __GNUC__ -# define __GNUC_PREREQ__ (x, y) \ - ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ - (__GNUC__ > (x))) -# else -# define __GNUC_PREREQ__ (x, y) 0 -# endif -# if !__GNUC_PREREQ__ (4 ,5 ) -# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) -# endif -# define va_start (ap, last) __builtin_va_start((ap), (last)) -# define va_end (ap) __builtin_va_end(ap) -# define va_arg __builtin_va_arg +#ifdef __GNUC__ +# define __GNUC_PREREQ__ (x, y) \ + ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ + (__GNUC__ > (x)))#else -# include <stdarg.h> +# define __GNUC_PREREQ__ (x, y) 0#endif +#if !__GNUC_PREREQ__ (4 ,5 ) +# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) +#endif + +typedef __builtin_va_list va_list ; +#define va_start (ap, last) __builtin_va_start((ap), (last)) +#define va_end (ap) __builtin_va_end(ap) +#define va_arg __builtin_va_arg +#endif /* __XEN_STDARG_H__ */
xen/include/xen/stdbool.h +4 -8
@@ -1,13 +1,9 @@#ifndef __XEN_STDBOOL_H__#define __XEN_STDBOOL_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -# define bool _Bool -# define true 1 -# define false 0 -# define __bool_true_false_are_defined 1 -#else -# include <stdbool.h> -#endif +#define bool _Bool +#define true 1 +#define false 0 +#define __bool_true_false_are_defined 1#endif /* __XEN_STDBOOL_H__ */
Julien Grall
Tested-byQuoted 2 lines · Tim Deegan
Are you using a very old version of clang? As 06a9c7e points out, our current runes didn't work before clang v3.0.
I'm using clang 3.5 (which have other issue to compile Xen), but I have also tried 3.0 and 3.3.
Quoted 9 lines
If not, rather than chasing this around any further, I think we should abandon trying to use the compiler-provided headers even on linux. Does this patch fix your build issue?
commit
e7003f174e0df9192dde6fa8d33b0a20f99ce053Author: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>
With this patch, -iwithprefix include is not necessary now. I wondering if we can remove it from the command line.
Quoted 8 lines
We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Signed-off-by Tim Deegan <[email protected]>
This patch is working fine to build xen clang 3.0, 3.3. I have others issue to build with clang 3.5.
Thanks!
Signature · 1 line
Julien Grall
Quoted 13 lines · Julien Grall
Quoted 2 lines · Tim Deegan
Are you using a very old version of clang? As 06a9c7e points out, our current runes didn't work before clang v3.0.
I'm using clang 3.5 (which have other issue to compile Xen), but I have also tried 3.0 and 3.3.
Quoted 4 lines
If not, rather than chasing this around any further, I think we should abandon trying to use the compiler-provided headers even on linux. Does this patch fix your build issue?
Quoted 8 lines
Quoted 5 lines
commit
e7003f174e0df9192dde6fa8d33b0a20f99ce053Author: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>With this patch, -iwithprefix include is not necessary now. I wondering if we can remove it from the command line.
Yes, I think so.
Quoted 7 lines
Quoted 7 lines
We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Quoted 6 lines
Quoted 1 line
Signed-off-by Tim Deegan <[email protected]>This patch is working fine to build xen clang 3.0, 3.3. I have others issue to build with clang 3.5.
Tested-by Julien Grall <[email protected]>
Great! Assuming you'll have a series of patches to fix the clang-3.5 build, do you want to just take this into that series, and drop the -iwithprefix at the same time?
Cheers,
Tim.
(Add George as release manager)
Quoted 34 lines · Tim Deegan
Quoted 13 lines · Julien Grall
Quoted 2 lines · Tim Deegan
Are you using a very old version of clang? As 06a9c7e points out, our current runes didn't work before clang v3.0.
I'm using clang 3.5 (which have other issue to compile Xen), but I have also tried 3.0 and 3.3.
Quoted 4 lines
If not, rather than chasing this around any further, I think we should abandon trying to use the compiler-provided headers even on linux. Does this patch fix your build issue?
Quoted 8 lines
Quoted 5 lines
commit
e7003f174e0df9192dde6fa8d33b0a20f99ce053Author: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>With this patch, -iwithprefix include is not necessary now. I wondering if we can remove it from the command line.
Yes, I think so.
Quoted 9 lines
Quoted 8 lines
We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Signed-off-by Tim Deegan <[email protected]>
Quoted 8 lines
Quoted 4 lines
This patch is working fine to build xen clang 3.0, 3.3. I have others issue to build with clang 3.5.
Tested-by Julien Grall <[email protected]>Great! Assuming you'll have a series of patches to fix the clang-3.5 build, do you want to just take this into that series, and drop the -iwithprefix at the same time?
If it's possible I'd like this patch goes in Xen 4.4 to fix build with official version of clang (until 3.4).
Clang 3.5 is still under development, so I don't think it's important to have support for it in Xen 4.4.
Cheers,
Signature · 1 line
Julien Grall
Tim Deegan
Tested-by Julien GrallQuoted 46 lines · Julien Grall
(Add George as release manager)
Quoted 34 lines · Tim Deegan
Quoted 13 lines · Julien Grall
Quoted 2 lines · Tim Deegan
Are you using a very old version of clang? As 06a9c7e points out, our current runes didn't work before clang v3.0.
I'm using clang 3.5 (which have other issue to compile Xen), but I have also tried 3.0 and 3.3.
Quoted 4 lines
If not, rather than chasing this around any further, I think we should abandon trying to use the compiler-provided headers even on linux. Does this patch fix your build issue?
Quoted 8 lines
Quoted 5 lines
commit
e7003f174e0df9192dde6fa8d33b0a20f99ce053Author: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>With this patch, -iwithprefix include is not necessary now. I wondering if we can remove it from the command line.
Yes, I think so.
Quoted 9 lines
Quoted 8 lines
We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Signed-off-by Tim Deegan <[email protected]>Quoted 8 lines
Quoted 4 lines
This patch is working fine to build xen clang 3.0, 3.3. I have others issue to build with clang 3.5.
Tested-by Julien Grall <[email protected]>Great! Assuming you'll have a series of patches to fix the clang-3.5 build, do you want to just take this into that series, and drop the -iwithprefix at the same time?
Quoted 5 lines
If it's possible I'd like this patch goes in Xen 4.4 to fix build with official version of clang (until 3.4).
Clang 3.5 is still under development, so I don't think it's important to have support for it in Xen 4.4.
Fair enough. In that case it needs a release ack from George. It:
- fixes a compile issue on some version s of clang;
- might cause a regression with other compilers, but the regression is likely to be obvious (i.e. a compile-time failure).
And it needs an ack from Keir, for changing common code.
v2 is below, removing "-iwithprefix". I've kept your tested-by; hope that's OK.
Cheers,
Tim.
commit 1d62fcb9ad8d2b409ac2cf0e8a3824e19ca3313f
Author: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000
xen: stop trying to use the system <stdarg.h> and <stdbool.h>
We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in /usr/include.
On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
xen/Rules.mk +1 -4
@@ -44,10 +44,7 @@ ALL_OBJS-$(x86) += $(BASEDIR)/crypto/built_in.o CFLAGS += -fno-builtin -fno-common CFLAGS += -Werror -Wredundant-decls -Wno-pointer-arith CFLAGS += -pipe -g -D__XEN__ -include $(BASEDIR)/include/xen/config.h -# Solaris puts stdarg.h &c in the system include directory. -ifneq ($(XEN_OS),SunOS) -CFLAGS += -nostdinc -iwithprefix include -endif +CFLAGS += -nostdinc CFLAGS-$(XSM_ENABLE) += -DXSM_ENABLE CFLAGS-$(FLASK_ENABLE) += -DFLASK_ENABLE -DXSM_MAGIC=0xf97cff8c
xen/include/xen/stdarg.h +14 -16
@@ -1,23 +1,21 @@#ifndef __XEN_STDARG_H__#define __XEN_STDARG_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -typedef __builtin_va_list va_list ; -# ifdef __GNUC__ -# define __GNUC_PREREQ__ (x, y) \ - ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ - (__GNUC__ > (x))) -# else -# define __GNUC_PREREQ__ (x, y) 0 -# endif -# if !__GNUC_PREREQ__ (4 ,5 ) -# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) -# endif -# define va_start (ap, last) __builtin_va_start((ap), (last)) -# define va_end (ap) __builtin_va_end(ap) -# define va_arg __builtin_va_arg +#ifdef __GNUC__ +# define __GNUC_PREREQ__ (x, y) \ + ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ + (__GNUC__ > (x)))#else -# include <stdarg.h> +# define __GNUC_PREREQ__ (x, y) 0#endif +#if !__GNUC_PREREQ__ (4 ,5 ) +# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) +#endif + +typedef __builtin_va_list va_list ; +#define va_start (ap, last) __builtin_va_start((ap), (last)) +#define va_end (ap) __builtin_va_end(ap) +#define va_arg __builtin_va_arg +#endif /* __XEN_STDARG_H__ */
xen/include/xen/stdbool.h +4 -8
@@ -1,13 +1,9 @@#ifndef __XEN_STDBOOL_H__#define __XEN_STDBOOL_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -# define bool _Bool -# define true 1 -# define false 0 -# define __bool_true_false_are_defined 1 -#else -# include <stdbool.h> -#endif +#define bool _Bool +#define true 1 +#define false 0 +#define __bool_true_false_are_defined 1#endif /* __XEN_STDBOOL_H__ */
Keir Fraser
Acked-byQuoted 23 lines · Tim Deegan
Fair enough. In that case it needs a release ack from George. It:
- fixes a compile issue on some version s of clang;
- might cause a regression with other compilers, but the regression is likely to be obvious (i.e. a compile-time failure).And it needs an ack from Keir, for changing common code.
v2 is below, removing "-iwithprefix". I've kept your tested-by; hope that's OK.
Cheers,
Tim.
commit
1d62fcb9ad8d2b409ac2cf0e8a3824e19ca3313fAuthor: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.
Quoted 6 lines
On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Signed-off-by Tim Deegan <[email protected]>Tested-by Julien Grall <[email protected]>
I'm fine with the principle of it. I don't know about how risky it is for 4.4.
George: ping.
Quoted 122 lines · Tim Deegan
Quoted 5 lines · Julien Grall
If it's possible I'd like this patch goes in Xen 4.4 to fix build with official version of clang (until 3.4).
Clang 3.5 is still under development, so I don't think it's important to have support for it in Xen 4.4.
Fair enough. In that case it needs a release ack from George. It:
- fixes a compile issue on some version s of clang;
- might cause a regression with other compilers, but the regression is likely to be obvious (i.e. a compile-time failure).And it needs an ack from Keir, for changing common code.
v2 is below, removing "-iwithprefix". I've kept your tested-by; hope that's OK.
Cheers,
Tim.
commit
1d62fcb9ad8d2b409ac2cf0e8a3824e19ca3313fAuthor: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Signed-off-by Tim Deegan <[email protected]>Tested-by Julien Grall <[email protected]>xen/Rules.mk +1
-4@@ -44,10 +44,7 @@ ALL_OBJS-$(x86) += $(BASEDIR)/crypto/built_in.o CFLAGS += -fno-builtin -fno-common CFLAGS += -Werror -Wredundant-decls -Wno-pointer-arith CFLAGS += -pipe -g -D__XEN__ -include $(BASEDIR)/include/xen/config.h -# Solaris puts stdarg.h &c in the system include directory. -ifneq ($(XEN_OS),SunOS) -CFLAGS += -nostdinc -iwithprefix include -endif +CFLAGS += -nostdinc CFLAGS-$(XSM_ENABLE) += -DXSM_ENABLE CFLAGS-$(FLASK_ENABLE) += -DFLASK_ENABLE -DXSM_MAGIC=0xf97cff8cxen/include/xen/stdarg.h +14
-16@@ -1,23 +1,21 @@#ifndef __XEN_STDARG_H__#define __XEN_STDARG_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -typedef __builtin_va_list va_list ; -# ifdef __GNUC__ -# define __GNUC_PREREQ__ (x, y) \ - ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ - (__GNUC__ > (x))) -# else -# define __GNUC_PREREQ__ (x, y) 0 -# endif -# if !__GNUC_PREREQ__ (4 ,5 ) -# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) -# endif -# define va_start (ap, last) __builtin_va_start((ap), (last)) -# define va_end (ap) __builtin_va_end(ap) -# define va_arg __builtin_va_arg +#ifdef __GNUC__ +# define __GNUC_PREREQ__ (x, y) \ + ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ + (__GNUC__ > (x)))#else -# include <stdarg.h> +# define __GNUC_PREREQ__ (x, y) 0#endif +#if !__GNUC_PREREQ__ (4 ,5 ) +# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) +#endif + +typedef __builtin_va_list va_list ; +#define va_start (ap, last) __builtin_va_start((ap), (last)) +#define va_end (ap) __builtin_va_end(ap) +#define va_arg __builtin_va_arg +#endif /* __XEN_STDARG_H__ */ xen/include/xen/stdbool.h +4
-8@@ -1,13 +1,9 @@#ifndef __XEN_STDBOOL_H__#define __XEN_STDBOOL_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -# define bool _Bool -# define true 1 -# define false 0 -# define __bool_true_false_are_defined 1 -#else -# include <stdbool.h> -#endif +#define bool _Bool +#define true 1 +#define false 0 +#define __bool_true_false_are_defined 1#endif /* __XEN_STDBOOL_H__ */ _______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Quoted 13 lines · Tim Deegan
George: ping.
Quoted 10 lines · Tim Deegan
Quoted 5 lines · Julien Grall
If it's possible I'd like this patch goes in Xen 4.4 to fix build with official version of clang (until 3.4).
Clang 3.5 is still under development, so I don't think it's important to have support for it in Xen 4.4.
Fair enough. In that case it needs a release ack from George. It:
- fixes a compile issue on some version s of clang;
- might cause a regression with other compilers, but the regression is likely to be obvious (i.e. a compile-time failure).
So the main risk would be if stgarg.h contained something like the "__GNUC_PREREQ__(4, 5)" #ifdef-ery that we missed. Without this patch, we know it doesn't compile on the latest version of clang. I think that's probably worse than the risk of potentially not compiling on some compiler that doesn't end up being tested before the release:
Quoted 111 lines
Quoted 111 lines
And it needs an ack from Keir, for changing common code.
v2 is below, removing "-iwithprefix". I've kept your tested-by; hope that's OK.
Cheers,
Tim.
commit
1d62fcb9ad8d2b409ac2cf0e8a3824e19ca3313fAuthor: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Signed-off-by Tim Deegan <[email protected]>Tested-by Julien Grall <[email protected]>xen/Rules.mk +1
-4@@ -44,10 +44,7 @@ ALL_OBJS-$(x86) += $(BASEDIR)/crypto/built_in.o CFLAGS += -fno-builtin -fno-common CFLAGS += -Werror -Wredundant-decls -Wno-pointer-arith CFLAGS += -pipe -g -D__XEN__ -include $(BASEDIR)/include/xen/config.h -# Solaris puts stdarg.h &c in the system include directory. -ifneq ($(XEN_OS),SunOS) -CFLAGS += -nostdinc -iwithprefix include -endif +CFLAGS += -nostdinc CFLAGS-$(XSM_ENABLE) += -DXSM_ENABLE CFLAGS-$(FLASK_ENABLE) += -DFLASK_ENABLE -DXSM_MAGIC=0xf97cff8cxen/include/xen/stdarg.h +14
-16@@ -1,23 +1,21 @@#ifndef __XEN_STDARG_H__#define __XEN_STDARG_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -typedef __builtin_va_list va_list ; -# ifdef __GNUC__ -# define __GNUC_PREREQ__ (x, y) \ - ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ - (__GNUC__ > (x))) -# else -# define __GNUC_PREREQ__ (x, y) 0 -# endif -# if !__GNUC_PREREQ__ (4 ,5 ) -# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) -# endif -# define va_start (ap, last) __builtin_va_start((ap), (last)) -# define va_end (ap) __builtin_va_end(ap) -# define va_arg __builtin_va_arg +#ifdef __GNUC__ +# define __GNUC_PREREQ__ (x, y) \ + ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ + (__GNUC__ > (x)))#else -# include <stdarg.h> +# define __GNUC_PREREQ__ (x, y) 0#endif +#if !__GNUC_PREREQ__ (4 ,5 ) +# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) +#endif + +typedef __builtin_va_list va_list ; +#define va_start (ap, last) __builtin_va_start((ap), (last)) +#define va_end (ap) __builtin_va_end(ap) +#define va_arg __builtin_va_arg +#endif /* __XEN_STDARG_H__ */ xen/include/xen/stdbool.h +4
-8@@ -1,13 +1,9 @@#ifndef __XEN_STDBOOL_H__#define __XEN_STDBOOL_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -# define bool _Bool -# define true 1 -# define false 0 -# define __bool_true_false_are_defined 1 -#else -# include <stdbool.h> -#endif +#define bool _Bool +#define true 1 +#define false 0 +#define __bool_true_false_are_defined 1#endif /* __XEN_STDBOOL_H__ */ _______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Quoted 8 lines · George Dunlap
Quoted 7 lines · Tim Deegan
George: ping.
Quoted 4 lines · Tim Deegan
Quoted 3 lines · Julien Grall
If it's possible I'd like this patch goes in Xen 4.4 to fix build with official version of clang (until 3.4).
Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines
Clang 3.5 is still under development, so I don't think it's important to have support for it in Xen 4.4.
Fair enough. In that case it needs a release ack from George. It:
- fixes a compile issue on some version s of clang;
- might cause a regression with other compilers, but the regression is likely to be obvious (i.e. a compile-time failure).So the main risk would be if
stgarg.hcontained something like the "__GNUC_PREREQ__(4, 5)" #ifdef-ery that we missed.
Sorry, realized the grammar was ambiguous here. For the record, I meant, the risk would be a system where the stdarg.h contained something of that type that our own copy didn't.
-G
PATCH 1/1 xen: Don't use __builtin_stdarg_start().
In reply to Tim DeeganCset fca49a00 ("netbsd: build fix with gcc 4.5") changed the definition of va_start() to use __builtin_va_start() rather than __builtin_stdarg_start() for GCCs >= 4.5, but in fact GCC dropped __builtin_stdarg_start() before v3.3.
---
George: this should fix the build issues Roger was having with GCC 4.5 after my last stdarg patch.
xen/include/xen/stdarg.h +0 -12
@@ -1,18 +1,6 @@#ifndef __XEN_STDARG_H__#define __XEN_STDARG_H__ -#ifdef __GNUC__ -# define __GNUC_PREREQ__ (x, y) \ - ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ - (__GNUC__ > (x))) -#else -# define __GNUC_PREREQ__ (x, y) 0 -#endif - -#if !__GNUC_PREREQ__ (4 ,5 ) -# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) -#endif -typedef __builtin_va_list va_list ;#define va_start (ap, last) __builtin_va_start((ap), (last))#define va_end (ap) __builtin_va_end(ap)
Re: [PATCH] xen: Don't use __builtin_stdarg_start().
Quoted 11 lines · Tim Deegan
Cset
fca49a00("netbsd: build fix with gcc 4.5") changed the definition ofva_start()to use__builtin_va_start()rather than__builtin_stdarg_start()for GCCs >= 4.5, but in fact GCC dropped__builtin_stdarg_start()before v3.3.Signed-off-by Tim Deegan <[email protected]>---
George: this should fix the build issues Roger was having with GCC 4.5 after my last stdarg patch.
Once it has Roger's tested-by:
Re: [PATCH] xen: Don't use __builtin_stdarg_start().
In reply to Tim DeeganQuoted 6 lines · Tim Deegan
Cset
fca49a00("netbsd: build fix with gcc 4.5") changed the definition ofva_start()to use__builtin_va_start()rather than__builtin_stdarg_start()for GCCs >= 4.5, but in fact GCC dropped__builtin_stdarg_start()before v3.3.Signed-off-by Tim Deegan <[email protected]>
With gcc version 4.4.5 (Debian 4.4.5-8)
Quoted 131 lines · Tim Deegan
Quoted 30 lines · Julien Grall
Quoted 27 lines · Tim Deegan
Quoted 25 lines · Julien Grall
Quoted 11 lines · Tim Deegan
Quoted 9 lines · Julien Grall
Commit 06a9c7e "xen: move -nostdinc into common Rules.mk." breaks compilation with clang:
In file included from sched_sedf.c:8:
In file included from/home/julieng/works/xen/xen/include/xen/lib.h:5:/home/julieng/works/xen/xen/include/xen/stdarg.h:20:12: error: 'stdarg.h' file not found with <angled> include; use "quotes" instead^~~~~~~~~~ "stdarg.h"Quoted 5 lines
Looks like on your system
stdarg.hdoesn't live in a compiler-specific path, like we have for the BSDs. I think we should just go to using our own definitions for stdarg/stdbool everywhere; trying to chase the compiler-specific versions around is a PITA, and the pieces we actually need are trivial.For BSDs, we are using our own stdargs/stdbool. So we don't include the system <
stdarg.h>.Linux is using $(CC) -print-file-name=include to get the right path. It works with both gcc and clang on Linux distos, but not on FreeBSD.
Quoted 7 lines
Quoted 3 lines
Wait - is the error message you posted from clang on FreeBSD? That's surprising; on FreeBSD
xen/stdarg.hshouldn't be trying to include <stdarg.h> at all. Is__FreeBSD__not being defined?No it's from Linux (Debian Wheezy and Fedora). I just gave a try to the "-print-file-name" solution on FreeBSD.
Oh, OK. Yeah, we knew that didn't work there, because on *BSD the compiler-specific headers like
stdarg.hactually live in/usr/includeand can themselves include other system headers. That's why we have our own implementation of the bits we need, that we use on BSD.Are you using a very old version of clang? As 06a9c7e points out, our current runes didn't work before clang v3.0.
If not, rather than chasing this around any further, I think we should abandon trying to use the compiler-provided headers even on linux. Does this patch fix your build issue?
commit
e7003f174e0df9192dde6fa8d33b0a20f99ce053Author: Tim Deegan <[email protected]> Date: Tue Feb 11 12:44:09 2014 +0000xen: stop trying to use the system <
stdarg.h> and <stdbool.h>We already have our own versions of the stdarg/stdbool definitions, for systems where those headers are installed in
/usr/include.On linux, they're typically installed in compiler-specific paths, but finding them has proved unreliable. Drop that and use our own versions everywhere.
Signed-off-by Tim Deegan <[email protected]>xen/include/xen/stdarg.h +14
-16@@ -1,23 +1,21 @@#ifndef __XEN_STDARG_H__#define __XEN_STDARG_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -typedef __builtin_va_list va_list ; -# ifdef __GNUC__ -# define __GNUC_PREREQ__ (x, y) \ - ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ - (__GNUC__ > (x))) -# else -# define __GNUC_PREREQ__ (x, y) 0 -# endif -# if !__GNUC_PREREQ__ (4 ,5 ) -# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) -# endif -# define va_start (ap, last) __builtin_va_start((ap), (last)) -# define va_end (ap) __builtin_va_end(ap) -# define va_arg __builtin_va_arg +#ifdef __GNUC__ +# define __GNUC_PREREQ__ (x, y) \ + ((__GNUC__ == (x) && __GNUC_MINOR__ >= (y)) || \ + (__GNUC__ > (x)))#else -# include <stdarg.h> +# define __GNUC_PREREQ__ (x, y) 0#endif +#if !__GNUC_PREREQ__ (4 ,5 ) +# define __builtin_va_start (ap, last) __builtin_stdarg_start((ap), (last)) +#endif + +typedef __builtin_va_list va_list ; +#define va_start (ap, last) __builtin_va_start((ap), (last)) +#define va_end (ap) __builtin_va_end(ap) +#define va_arg __builtin_va_arg +#endif /* __XEN_STDARG_H__ */ xen/include/xen/stdbool.h +4
-8@@ -1,13 +1,9 @@#ifndef __XEN_STDBOOL_H__#define __XEN_STDBOOL_H__ -#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) -# define bool _Bool -# define true 1 -# define false 0 -# define __bool_true_false_are_defined 1 -#else -# include <stdbool.h> -#endif +#define bool _Bool +#define true 1 +#define false 0 +#define __bool_true_false_are_defined 1#endif /* __XEN_STDBOOL_H__ */
Signature · 1 line
Julien Grall