Building broken with CS 13282

Building broken with CS 13282

[Christoph Egger]

Hi!

Building CS 13282 is broken:

[...]
gmake[2]: Entering directory `/usr/src/xen-unstable.hg/xen'
gmake[2]: Leaving directory `/usr/src/xen-unstable.hg/xen'
[ -e include/asm ] || ln -sf asm-x86 include/asm
gmake -f /usr/src/xen-unstable.hg/xen/Rules.mk -C include
gmake[2]: Entering directory `/usr/src/xen-unstable.hg/xen/include'
gmake[2]: Nothing to be done for `all'.
gmake[2]: Leaving directory `/usr/src/xen-unstable.hg/xen/include'
gmake -f /usr/src/xen-unstable.hg/xen/Rules.mk -C arch/x86 asm-offsets.s
gmake[2]: Entering directory `/usr/src/xen-unstable.hg/xen/arch/x86'
gcc -O2 -fomit-frame-pointer -m64 -DNDEBUG -std=gnu99 -Wall 
-Wstrict-prototypes -Wno-unused-value -Wdeclaration-after-statement -nostdinc 
-fno-builtin -fno-common -fno-strict-aliasing -iwithprefix include -Werror 
-Wno-pointer-arith -pipe -I/usr/src/xen-unstable.hg/xen/include 
-I/usr/src/xen-unstable.hg/xen/include/asm-x86/mach-generic 
-I/usr/src/xen-unstable.hg/xen/include/asm-x86/mach-default -msoft-float 
-fno-stack-protector -mno-red-zone -fpic -fno-reorder-blocks 
-fno-asynchronous-unwind-tables -DGCC_HAS_VISIBILITY_ATTRIBUTE -g -D__XEN__ 
-DCRASH_DEBUG -S -o asm-offsets.s x86_64/asm-offsets.c
In file /usr/src/xen-unstable.hg/xen/include/xen/sched.h:9,
                 from x86_64/asm-offsets.c:9:
/usr/src/xen-unstable.hg/xen/include/xen/shared.h:8:24: error: compat/xen.h: 
No such file or directory
[...]

There is no xen/include/compat/ directory in my tree.

Re: Building broken with CS 13282

[Keir Fraser]

On 8/1/07 10:49, "Christoph Egger" <Christoph.Egger@amd.com> wrote:

> /usr/src/xen-unstable.hg/xen/include/xen/shared.h:8:24: error: compat/xen.h:
> No such file or directory
> [...]
> 
> There is no xen/include/compat/ directory in my tree.

It's supposed to be auto-generated by the gunk in xen/include/Makefile. It
works for me. :-)

 -- Keir

RE: Building broken with CS 13282

[Yu, Ping Y]

C/s 13282 build are OK in 32/pae/64 platform in my side. 

>-----Original Message-----
>From: xen-devel-bounces@lists.xensource.com [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Christoph Egger
>Sent: 2007年1月8日 18:49
>To: xen-devel@lists.xensource.com
>Subject: [Xen-devel] Building broken with CS 13282
>
>
>Hi!
>
>Building CS 13282 is broken:
>
>[...]
>gmake[2]: Entering directory `/usr/src/xen-unstable.hg/xen'
>gmake[2]: Leaving directory `/usr/src/xen-unstable.hg/xen'
>[ -e include/asm ] || ln -sf asm-x86 include/asm
>gmake -f /usr/src/xen-unstable.hg/xen/Rules.mk -C include
>gmake[2]: Entering directory `/usr/src/xen-unstable.hg/xen/include'
>gmake[2]: Nothing to be done for `all'.
>gmake[2]: Leaving directory `/usr/src/xen-unstable.hg/xen/include'
>gmake -f /usr/src/xen-unstable.hg/xen/Rules.mk -C arch/x86 asm-offsets.s
>gmake[2]: Entering directory `/usr/src/xen-unstable.hg/xen/arch/x86'
>gcc -O2 -fomit-frame-pointer -m64 -DNDEBUG -std=gnu99 -Wall
>-Wstrict-prototypes -Wno-unused-value -Wdeclaration-after-statement -nostdinc
>-fno-builtin -fno-common -fno-strict-aliasing -iwithprefix include -Werror
>-Wno-pointer-arith -pipe -I/usr/src/xen-unstable.hg/xen/include
>-I/usr/src/xen-unstable.hg/xen/include/asm-x86/mach-generic
>-I/usr/src/xen-unstable.hg/xen/include/asm-x86/mach-default -msoft-float
>-fno-stack-protector -mno-red-zone -fpic -fno-reorder-blocks
>-fno-asynchronous-unwind-tables -DGCC_HAS_VISIBILITY_ATTRIBUTE -g -D__XEN__
>-DCRASH_DEBUG -S -o asm-offsets.s x86_64/asm-offsets.c
>In file /usr/src/xen-unstable.hg/xen/include/xen/sched.h:9,
>                 from x86_64/asm-offsets.c:9:
>/usr/src/xen-unstable.hg/xen/include/xen/shared.h:8:24: error: compat/xen.h:
>No such file or directory
>[...]
>
>There is no xen/include/compat/ directory in my tree.
>
>
>
>
>_______________________________________________
>Xen-devel mailing list
>Xen-devel@lists.xensource.com
>http://lists.xensource.com/xen-devel

Re: Building broken with CS 13282

[John Levon]

On Mon, Jan 08, 2007 at 11:07:39AM +0000, Keir Fraser wrote:

> > There is no xen/include/compat/ directory in my tree.
> 
> It's supposed to be auto-generated by the gunk in xen/include/Makefile. It
> works for me. :-)

We have a seemingly related problem:

gmake[5]: Entering directory `/export/johnlev/xen/xen-unstable-child/xen.hg/xen/include'
mkdir -p compat
grep -v 'DEFINE_XEN_GUEST_HANDLE(long)' public/acm.h | \
sed -e 's,^[[:space:]]*#[[:space:]]*include[[:space:]]\+,__InClUdE__ ,' \
    -e 's,^[[:space:]]*#[[:space:]]*define[[:space:]]\+\([[:upper:]_]*_GUEST_HANDLE\),#define HIDE_\1,' \
    -e 's,^[[:space:]]*#[[:space:]]*define[[:space:]]\+\([[:lower:]_]*_guest_handle\),#define hide_\1,' \
    -e 's,XEN_GUEST_HANDLE\(_[[:xdigit:]]\+\)\?,COMPAT_HANDLE,g' \
     \
    >compat/acm.c.new
mv -f compat/acm.c.new compat/acm.c
gcc -E -O1 -fno-omit-frame-pointer -Wa,--divide -D_POSIX_C_SOURCE=200112L -D__EXTENSIONS__ -m64 -g -std=gnu99 -Wall -Wstrict-prototypes -Wno-unused-value -Wdeclaration-after-statement -fno-builtin -fno-common -fno-strict-aliasing -iwithprefix include -Werror -Wno-pointer-arith -pipe -I/export/johnlev/xen/xen-unstable-child/xen.hg/xen/include -I/export/johnlev/xen/xen-unstable-child/xen.hg/xen/include/asm-x86/mach-generic -I/export/johnlev/xen/xen-unstable-child/xen.hg/xen/include/asm-x86/mach-default -msoft-float -mno-red-zone -fpic -fno-reorder-blocks -fno-asynchronous-unwind-tables -g -D__XEN__ -DVERBOSE -include public/xen-compat.h -m32 -o compat/acm.i compat/acm.c
compat/acm.c:29:17: xen.h: No such file or directory

We've probably never built any ACM stuff at all before... surely #include "xen.h" isn't right?

regards
john

Re: Building broken with CS 13282

[Keir Fraser]

On 8/1/07 12:24, "John Levon" <levon@movementarian.org> wrote:

> We've probably never built any ACM stuff at all before... surely #include
> "xen.h" isn't right?

xen/include/compat/xen.h is auto-generated. The question is: why has it been
skipped in your build environment?

 -- Keir

Re: Building broken with CS 13282

[John Levon]

On Mon, Jan 08, 2007 at 01:19:58PM +0000, Keir Fraser wrote:

> On 8/1/07 12:24, "John Levon" <levon@movementarian.org> wrote:
> 
> > We've probably never built any ACM stuff at all before... surely #include
> > "xen.h" isn't right?
> 
> xen/include/compat/xen.h is auto-generated. The question is: why has it been
> skipped in your build environment?

For one thing I can't see a dependency such that compat/xen.[ch] will be
built before compat/acm.[ch], and there don't seem to be gcc-generated
deps either ?

regards
john

Re: Building broken with CS 13282

[John Levon]

On Mon, Jan 08, 2007 at 02:13:32PM +0000, John Levon wrote:

> For one thing I can't see a dependency such that compat/xen.[ch] will be
> built before compat/acm.[ch], and there don't seem to be gcc-generated
> deps either ?

And even if I do:

compat/xen.c:30:24: xen-compat.h: No such file or directory
compat/xen.c:33:26: arch-x86/xen.h: No such file or directory

Ading -I public helps until it tries to make 'archa64.h', which looks
awfully like a mangled arch-ia64.h, though I don't follow how that's
happened.

regards
john

Re: Building broken with CS 13282

[Christoph Egger]

On Monday 08 January 2007 15:30, John Levon wrote:
> On Mon, Jan 08, 2007 at 02:13:32PM +0000, John Levon wrote:
> > For one thing I can't see a dependency such that compat/xen.[ch] will be
> > built before compat/acm.[ch], and there don't seem to be gcc-generated
> > deps either ?
>
> And even if I do:
>
> compat/xen.c:30:24: xen-compat.h: No such file or directory
> compat/xen.c:33:26: arch-x86/xen.h: No such file or directory
>
> Ading -I public helps until it tries to make 'archa64.h', which looks
> awfully like a mangled arch-ia64.h, though I don't follow how that's
> happened.

Perhaps this has to do with different sed implementations.

OpenBSD sed doesn't support alternatives as used in this
snippet from xen/include/Makefile:

        sed -e 's,__InClUdE__,#include,' \
            -e 's,"xen-compat.h",<public/xen-compat.h>,' \
            -e 's,\(struct\|union\|enum\)[[:space:]]\+\(xen_\?\)\?
\([[:alpha:]_]\),\1 compat_\3,g' \

So that's at least one reason for the build failure on OpenBSD.

On Linux with GNU sed, it builds for me.

Christoph

Re: Building broken with CS 13282

[John Levon]

On Mon, Jan 08, 2007 at 03:40:32PM +0100, Christoph Egger wrote:

> snippet from xen/include/Makefile:
> 
>         sed -e 's,__InClUdE__,#include,' \
>             -e 's,"xen-compat.h",<public/xen-compat.h>,' \
>             -e 's,\(struct\|union\|enum\)[[:space:]]\+\(xen_\?\)\?
> \([[:alpha:]_]\),\1 compat_\3,g' \

It's very unlikely that Solaris sed does too, which I suppose explains
the xen-compat.h issue at least.

regards
john

Re: Building broken with CS 13282

[Keir Fraser]

On 8/1/07 14:50, "John Levon" <levon@movementarian.org> wrote:

>> snippet from xen/include/Makefile:
>> 
>>         sed -e 's,__InClUdE__,#include,' \
>>             -e 's,"xen-compat.h",<public/xen-compat.h>,' \
>>             -e 's,\(struct\|union\|enum\)[[:space:]]\+\(xen_\?\)\?
>> \([[:alpha:]_]\),\1 compat_\3,g' \
> 
> It's very unlikely that Solaris sed does too, which I suppose explains
> the xen-compat.h issue at least.

Either we need to stipulate GNU sed, or re-code some of this stuff in Python
or Perl (probably Python, since that's already a dependency for the tools).
My hope would be that it would magically become a bit more readable. :-)

 -- Keir

Re: Building broken with CS 13282

[John Levon]

On Mon, Jan 08, 2007 at 04:17:02PM +0000, Keir Fraser wrote:

> >>         sed -e 's,__InClUdE__,#include,' \
> >>             -e 's,"xen-compat.h",<public/xen-compat.h>,' \
> >>             -e 's,\(struct\|union\|enum\)[[:space:]]\+\(xen_\?\)\?
> >> \([[:alpha:]_]\),\1 compat_\3,g' \
> > 
> > It's very unlikely that Solaris sed does too, which I suppose explains
> > the xen-compat.h issue at least.
> 
> Either we need to stipulate GNU sed, or re-code some of this stuff in Python
> or Perl (probably Python, since that's already a dependency for the tools).
> My hope would be that it would magically become a bit more readable. :-)

gsed would be a problem for us, it's not (yet) shipped in Solaris.
Python would be nice.

Though I'm still confused how this is supposed to work. Even if I force
xen to be made first:

gcc -E -O2 -fno-omit-frame-pointer -Wa,--divide
-D_POSIX_C_SOURCE=200112L -D__EXTENSIONS__ -m32 -march=i686 -DNDEBUG
-std=gnu99 -Wall -Wstrict-prototypes -Wno-unused-value
-Wdeclaration-after-statement -fno-builtin -fno-common
-fno-strict-aliasing -iwithprefix include -Werror -Wno-pointer-arith
-pipe -I/export/johnlev/xen/xen-unstable-child/xen.hg/xen/include
-I/export/johnlev/xen/xen-unstable-child/xen.hg/xen/include/asm-x86/mach-generic
-I/export/johnlev/xen/xen-unstable-child/xen.hg/xen/include/asm-x86/mach-default
-msoft-float -g -D__XEN__ -include public/xen-compat.h -m32 -o
compat/xen.i compat/xen.c
compat/xen.c:30:24: xen-compat.h: No such file or directory
compat/xen.c:33:26: arch-x86/xen.h: No such file or directory

We're still missing an -I public/ for the arch-x86/xen.h header?

regards
john

Re: Building broken with CS 13282

[Keir Fraser]

On 8/1/07 17:26, "John Levon" <levon@movementarian.org> wrote:

> compat/xen.i compat/xen.c
> compat/xen.c:30:24: xen-compat.h: No such file or directory
> compat/xen.c:33:26: arch-x86/xen.h: No such file or directory
> 
> We're still missing an -I public/ for the arch-x86/xen.h header?

No, you are supposed to get one auto-generated at
include/compat/arch-x86/xen.c.

If there are dependencies among the auto-generated headers then they need to
be listed explicitly. Multiple dependencies listed in a single rule can be
made in parallel.

Have you already thought about this, Jan?

 -- Keir

Re: Building broken with CS 13282

[John Levon]

On Mon, Jan 08, 2007 at 06:48:00PM +0000, Keir Fraser wrote:

> > compat/xen.i compat/xen.c
> > compat/xen.c:30:24: xen-compat.h: No such file or directory
> > compat/xen.c:33:26: arch-x86/xen.h: No such file or directory
> > 
> > We're still missing an -I public/ for the arch-x86/xen.h header?
> 
> No, you are supposed to get one auto-generated at
> include/compat/arch-x86/xen.c.

Oh, right, I see.

> If there are dependencies among the auto-generated headers then they need to
> be listed explicitly. Multiple dependencies listed in a single rule can be
> made in parallel.

I'm actually a little surprised this compiles on Linux...

regards
john

Re: Building broken with CS 13282

[Jan Beulich]

>If there are dependencies among the auto-generated headers then they need to
>be listed explicitly. Multiple dependencies listed in a single rule can be
>made in parallel.

No, that's not the case - each header can be built independently, since the
#include-s are being substituted before handing to the preprocessor (not the
compiler, hence incomplete definitions don't matter), and substituted back
afterwards.

John, in order to understand what's wrong on your end I'll need more than just
the failing compiler invocation, specifically all the commands run while building in
xen/include (invoked from xen/Makefile:73). Since that's a series of $(MAKE)
invocations there, failure here should prevent making $(TARGET). Hence I
have to assume that something fails in the process of handling xen/include, but
doesn't indicate so through the process exit status (for that knowing what
actually got generated in xen/include/compat/ might also provide a clue).

Jan