On 08/24/2012 10:28 AM, Viktor Mihajlovski wrote: > > fwiw ... with the current GIT level I get the following build error on Ubuntu 10.04 32-bit > > CC libvirt_util_la-virnetdev.lo > cc1: warnings being treated as errors > /usr/include/netlink/object.h:58: error: inline function 'nl_object_priv' declared but never defined libvirt itself never uses that function. Also, in the version of /usr/include/netlink/object.h I'm working with (on Fedora 17) that function isn't declared as inline. Thomas, do you have any info on the libnl version used in Ubuntu 10.04, and/or what might be the cause of this warning.

Ah, that rings a bell, I knew I'd seen this issue before. I'm actually surprised it STILL hasn't been fixed in Ubuntu, and maybe it's time to apply the workaround below as a build-breaker fix (I think I've posted it on list before, but never committed it).

diff --git i/src/util/virnetlink.h w/src/util/virnetlink.h index 1982dae..82154de 100644 --- i/src/util/virnetlink.h +++ w/src/util/virnetlink.h @@ -26,7 +26,11 @@ # if defined(__linux__) && defined(HAVE_LIBNL) +/* Work around a bug where older libnl-1 headers expected older gcc + * semantics of 'extern inline' that conflict with C99 semantics. */ +# define inline # include <netlink/msg.h> +# undef inline

On 08/27/12 23:09, Eric Blake wrote: > On 08/24/2012 10:19 AM, Eric Blake wrote: > >> Ah, that rings a bell, I knew I'd seen this issue before. I'm actually >> surprised it STILL hasn't been fixed in Ubuntu, and maybe it's time to >> apply the workaround below as a build-breaker fix (I think I've posted >> it on list before, but never committed it). >> > >> >> diff --git i/src/util/virnetlink.h w/src/util/virnetlink.h >> index 1982dae..82154de 100644 >> --- i/src/util/virnetlink.h >> +++ w/src/util/virnetlink.h >> @@ -26,7 +26,11 @@ >> >> # if defined(__linux__) && defined(HAVE_LIBNL) >> >> +/* Work around a bug where older libnl-1 headers expected older gcc >> + * semantics of 'extern inline' that conflict with C99 semantics. */ >> +# define inline >> # include <netlink/msg.h> >> +# undef inline > > I've now pushed this under the build-breaker rule. > This breaks the build with libnl3: CC libvirt_util_la-virnetdevbridge.lo cc1: warnings being treated as errors /usr/include/libnl3/netlink/list.h:21:20: error: 'NL_INIT_LIST_HEAD' defined but not used [-Wunused-function] /usr/include/libnl3/netlink/list.h:37:20: error: 'nl_list_add_tail' defined but not used [-Wunused-function] /usr/include/libnl3/netlink/list.h:43:20: error: 'nl_list_add_head' defined but not used [-Wunused-function] /usr/include/libnl3/netlink/list.h:49:20: error: 'nl_list_del' defined but not used [-Wunused-function] /usr/include/libnl3/netlink/list.h:55:19: error: 'nl_list_empty' defined but not used [-Wunused-function] /usr/include/libnl3/netlink/object.h:60:23: error: 'nl_object_priv' defined but not used [-Wunused-function] cc1: error: unrecognized command line option "-Wno-suggest-attribute=const" cc1: error: unrecognized command line option "-Wno-suggest-attribute=pure" make[3]: *** [libvirt_util_la-virnetdev.lo] Error 1 That annoying libnl-1 bug exists also in other distros that for some reason refuse to fix it (e.g. gentoo) but upgrading to libnl3 helps and libvirt compiles cleanly. Libnl3 unfortunately has some inline functions in the header files where removing the inline keyword is catastrophic. I vote for reverting this patch and have distros apply that long-known trivial patch that fixes libnl-1 headers rather than breaking libnl3 support. Peter

On 27/08/2012, at 1:00 PM, Daniel Veillard wrote: > I tagged the rc2 in the git tree and pushed tarball and rpms to > the usual place: > ftp://libvirt.org/libvirt/ > > it should include the various fixes that were made on top of rc1 > and fix the compilation problem exposed on OS-X, as well as picking > libnl3 if available, > > please give it a try and report, I am still shooting for a final > release this Wednesday, Still a no go on OSX. Different error this time: ... libvirtd.c: In function 'main': libvirtd.c:1316: error: 'NETLINK_ROUTE' undeclared (first use in this function) libvirtd.c:1316: error: (Each undeclared identifier is reported only once libvirtd.c:1316: error: for each function it appears in.) libvirtd.c:1322: error: 'NETLINK_KOBJECT_UEVENT' undeclared (first use in this function) make[3]: *** [libvirtd-libvirtd.o] Error 1 make[3]: *** Waiting for unfinished jobs.... Any ideas? + Justin

On Mon, Aug 27, 2012 at 4:57 AM, Justin Clift <jclift(a)redhat.com&gt; wrote: > On 27/08/2012, at 1:00 PM, Daniel Veillard wrote: > >> I tagged the rc2 in the git tree and pushed tarball and rpms to >> the usual place: >> ftp://libvirt.org/libvirt/ >> >> it should include the various fixes that were made on top of rc1 >> and fix the compilation problem exposed on OS-X, as well as picking >> libnl3 if available, >> >> please give it a try and report, I am still shooting for a final >> release this Wednesday, >> > Still a no go on OSX. Different error this time: > > ... > libvirtd.c: In function 'main': > libvirtd.c:1316: error: 'NETLINK_ROUTE' undeclared (first use in this function) > libvirtd.c:1316: error: (Each undeclared identifier is reported only once > libvirtd.c:1316: error: for each function it appears in.) > libvirtd.c:1322: error: 'NETLINK_KOBJECT_UEVENT' undeclared (first use in this function) > make[3]: *** [libvirtd-libvirtd.o] Error 1 > make[3]: *** Waiting for unfinished jobs.... >

> Any ideas? > > + Justin > > You can try something like the following to get you a little farther and see if it fails farther on. Just trying to see if we can get Mac OS X built and see what changes need to happen before 0.10.0 goes out final, not necessary "proper" patches just yet. diff --git a/src/util/virnetlink.h b/src/util/virnetlink.h index 1982dae..c3f3e81 100644 --- a/src/util/virnetlink.h +++ b/src/util/virnetlink.h @@ -34,6 +34,9 @@ struct nl_msg; struct sockaddr_nl; struct nlattr; +#define NETLINK_ROUTE 0 /* Routing/device hook */ +#define NETLINK_KOBJECT_UEVENT 15 /* Kernel messages to userspace */ + # endif /* __linux__ */ int virNetlinkStartup(void);

You can try something like the following to get you a little farther and see if it fails farther on. Just trying to see if we can get Mac OS X built and see what changes need to happen before 0.10.0 goes out final, not necessary "proper" patches just yet. diff --git a/src/util/virnetlink.h b/src/util/virnetlink.h index 1982dae..c3f3e81 100644 --- a/src/util/virnetlink.h +++ b/src/util/virnetlink.h @@ -34,6 +34,9 @@ struct nl_msg; struct sockaddr_nl; struct nlattr; +#define NETLINK_ROUTE 0 /* Routing/device hook */ +#define NETLINK_KOBJECT_UEVENT 15 /* Kernel messages to userspace */ + # endif /* __linux__ */ int virNetlinkStartup(void);

I would suggest a fix along those lines: Daniel diff --git a/daemon/libvirtd.c b/daemon/libvirtd.c index 19dd26b..352d4fe 100644 --- a/daemon/libvirtd.c +++ b/daemon/libvirtd.c @@ -1312,17 +1312,21 @@ int main(int argc, char **argv) { goto cleanup; } +#if defined(__linux__) && defined(NETLINK_ROUTE) /* Register the netlink event service for NETLINK_ROUTE */ if (virNetlinkEventServiceStart(NETLINK_ROUTE, 0) < 0) { ret = VIR_DAEMON_ERR_NETWORK; goto cleanup; } +#endif +#if defined(__linux__) && defined(NETLINK_KOBJECT_UEVENT) /* Register the netlink event service for NETLINK_KOBJECT_UEVENT */ if (virNetlinkEventServiceStart(NETLINK_KOBJECT_UEVENT, 1) < 0) { ret = VIR_DAEMON_ERR_NETWORK; goto cleanup; } +#endif /* Run event loop. */ virNetServerRun(srv);

On Tue, Aug 28, 2012 at 04:05:36PM +0800, Daniel Veillard thus spake: > I don't like too much the idea of adding kernel defines in our headers. >Basically this is all due to commits >d57567940153147f4d43875fb92c3030b3178b03 and >080bf330e3749d94ebe094f8deca0e3e67d3f2fe > > The fact that it breaks only in daemon.c, i.e. after the library > compiled is a good infocation that the guards >#if WITH_VIRTUALPORT in src/util/virnetdevvportprofile.c >#if WITH_MACVTAP in src/util/virnetdevmacvlan.c >and >#if defined(__linux__) && defined(HAVE_LIBNL) && defined(IFLA_VF_MAX) > in src/util/virnetdev.c > >seems sufficient to avoid the portability problem in those files >and the remaining problem is really about daemon/libvirtd.c and >I would suggest a fix along those lines: > >Daniel > >diff --git a/daemon/libvirtd.c b/daemon/libvirtd.c >index 19dd26b..352d4fe 100644 >--- a/daemon/libvirtd.c >+++ b/daemon/libvirtd.c >@@ -1312,17 +1312,21 @@ int main(int argc, char **argv) { > goto cleanup; > } > >+#if defined(__linux__) && defined(NETLINK_ROUTE) > /* Register the netlink event service for NETLINK_ROUTE */ > if (virNetlinkEventServiceStart(NETLINK_ROUTE, 0) < 0) { > ret = VIR_DAEMON_ERR_NETWORK; > goto cleanup; > } >+#endif > >+#if defined(__linux__) && defined(NETLINK_KOBJECT_UEVENT) > /* Register the netlink event service for NETLINK_KOBJECT_UEVENT */ > if (virNetlinkEventServiceStart(NETLINK_KOBJECT_UEVENT, 1) < 0) { > ret = VIR_DAEMON_ERR_NETWORK; > goto cleanup; > } >+#endif > > /* Run event loop. */ > virNetServerRun(srv); >-- >Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ &gt;daniel(a)veillard.com | Rpmfind RPM search engine http://rpmfind.net/ >http://veillard.com/ | virtualization library http://libvirt.org/ > This is building as expected with your new patch for FreeBSD. Here is a link to the build logs. https://redports.org/buildarchive/20120828164025-20179/