ppc64 assistance
Hi All, I've having some issues with a PPC64 build and was wondering if someone a bit more ppc savy could have a poke. It builds fine on x86 and ppc32. git head also has the same issue as the current stable release so unfortunately that doesn't help much. PPC64 build http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 PPC32 build http://koji.fedoraproject.org/koji/taskinfo?taskID=1449147 Cheers, Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, Jul 2, 2009 at 6:29 AM, Peter Robinson wrote: > Hi All, > > I've having some issues with a PPC64 build and was wondering if > someone a bit more ppc savy could have a poke. A backtrace would be really useful here (remember to build with -ggdb -O0 for extra usefulness). The typelib code is fairly heavy on lowlevel C bit-twiddling so it's not unlikely this is an upstream issue. -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, Jul 2, 2009 at 11:49 AM, Colin Walters wrote: > On Thu, Jul 2, 2009 at 6:29 AM, Peter Robinson wrote: >> Hi All, >> >> I've having some issues with a PPC64 build and was wondering if >> someone a bit more ppc savy could have a poke. > > A backtrace would be really useful here (remember to build with -ggdb > -O0 for extra usefulness). > > The typelib code is fairly heavy on lowlevel C bit-twiddling so it's > not unlikely this is an upstream issue. OK, will do. I'm seeing the same issues on git master as well. Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, Jul 2, 2009 at 11:49 AM, Colin Walters wrote: > On Thu, Jul 2, 2009 at 6:29 AM, Peter Robinson wrote: >> Hi All, >> >> I've having some issues with a PPC64 build and was wondering if >> someone a bit more ppc savy could have a poke. > > A backtrace would be really useful here (remember to build with -ggdb > -O0 for extra usefulness). > > The typelib code is fairly heavy on lowlevel C bit-twiddling so it's > not unlikely this is an upstream issue. Interestingly with -ggdb it builds fine with out without -O0. With -O0 http://koji.fedoraproject.org/koji/taskinfo?taskID=1449368 Without. http://koji.fedoraproject.org/koji/taskinfo?taskID=1449389 Regards, Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, Jul 2, 2009 at 9:20 AM, Peter Robinson wrote: > > Interestingly with -ggdb it builds fine with out without -O0. That makes it a lot more likely to be a compiler flaw (though not guaranteed). Does 'make check' pass in the source tree you built with -O0? A backtrace would still be great though. Try setting DEBUG=fcatch in the environment. Some day hopefully ABRT will come and do this transparently and automatically... -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
>> Interestingly with -ggdb it builds fine with out without -O0. > > That makes it a lot more likely to be a compiler flaw (though not guaranteed). > > Does 'make check' pass in the source tree you built with -O0? > > A backtrace would still be great though. Try setting DEBUG=fcatch in > the environment. Some day hopefully ABRT will come and do this > transparently and automatically... A -O0 segfaults in x86 too so I just discovered :-( It seems also that the frysk package isn't built in rawhide ppc and I'm not sure if there's any other bits I need to add to it. The make clean is on this koji report. http://koji.fedoraproject.org/koji/getfile?taskID=1449535&name=build.log Cheers, Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
Peter Robinson wrote: > http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 Unrelated to this issue, but please use "make V=1" so we see the actual build command lines in the build.log (see the thread about the new automake). Kevin Kofler -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
>> http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 > > Unrelated to this issue, but please use "make V=1" so we see the actual > build command lines in the build.log (see the thread about the new > automake). With V=1 http://koji.fedoraproject.org/koji/getfile?taskID=1450335&name=build.log Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, 2009-07-02 at 16:28 +0100, Peter Robinson wrote: > >> http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 > > > > Unrelated to this issue, but please use "make V=1" so we see the actual > > build command lines in the build.log (see the thread about the new > > automake). > > With V=1 > > http://koji.fedoraproject.org/koji/getfile?taskID=1450335&name=build.log You know you can have access to a real box to test this on if you want it, right? You don't have to do it all in koji and look at the build logs. -- David WoodhouseOpen Source Technology Centre david.woodho...@intel.com Intel Corporation -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
Peter Robinson wrote: http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 Unrelated to this issue, but please use "make V=1" so we see the actual build command lines in the build.log (see the thread about the new automake). With V=1 http://koji.fedoraproject.org/koji/getfile?taskID=1450335&name=build.log AFAIS, your spec doesn't seem to pass RPM_OPT_FLAGS correctly, as well as does the package seems to play dirty games with CFLAGS. I haven't checked details, though. FWIW: Build breakdowns on ppc64 often are caused by not passing RPM_OPT_FLAGS correctly or a package playing dirty games with CFLAGS/CXXFLAGS. Ralf -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 >>> >>> Unrelated to this issue, but please use "make V=1" so we see the actual >>> build command lines in the build.log (see the thread about the new >>> automake). >> >> With V=1 >> >> http://koji.fedoraproject.org/koji/getfile?taskID=1450335&name=build.log > > AFAIS, your spec doesn't seem to pass RPM_OPT_FLAGS correctly, as well as > does the package seems to play dirty games with CFLAGS. I haven't checked > details, though. > > FWIW: Build breakdowns on ppc64 often are caused by not passing > RPM_OPT_FLAGS correctly or a package playing dirty games with > CFLAGS/CXXFLAGS. The dirty games were to try and work out this issue with PPC64 (it built fine on the rest of them) and it seems your right with the dirty CFLAGS game as it broke other previous working ones. This rev of the spec file was the one that worked on all but ppc64. http://cvs.fedoraproject.org/viewvc/rpms/gobject-introspection/devel/gobject-introspection.spec?revision=1.9&view=markup Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
>> >> http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 >> > >> > Unrelated to this issue, but please use "make V=1" so we see the actual >> > build command lines in the build.log (see the thread about the new >> > automake). >> >> With V=1 >> >> http://koji.fedoraproject.org/koji/getfile?taskID=1450335&name=build.log > > You know you can have access to a real box to test this on if you want > it, right? You don't have to do it all in koji and look at the build > logs. No I didn't. Are the details published anywhere of how to request access? Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, Jul 02, 2009 at 04:51:50PM +0100, Peter Robinson wrote: >>> >> http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 >>> > >>> > Unrelated to this issue, but please use "make V=1" so we see the actual >>> > build command lines in the build.log (see the thread about the new >>> > automake). >>> >>> With V=1 >>> >>> http://koji.fedoraproject.org/koji/getfile?taskID=1450335&name=build.log >> >> You know you can have access to a real box to test this on if you want >> it, right? You don't have to do it all in koji and look at the build >> logs. > >No I didn't. Are the details published anywhere of how to request access? I'm working on it. Basically, email dwmw2 or me an ssh public key. josh -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
Peter Robinson wrote: http://koji.fedoraproject.org/koji/taskinfo?taskID=1449113 Unrelated to this issue, but please use "make V=1" so we see the actual build command lines in the build.log (see the thread about the new automake). With V=1 http://koji.fedoraproject.org/koji/getfile?taskID=1450335&name=build.log AFAIS, your spec doesn't seem to pass RPM_OPT_FLAGS correctly, as well as does the package seems to play dirty games with CFLAGS. I haven't checked details, though. FWIW: Build breakdowns on ppc64 often are caused by not passing RPM_OPT_FLAGS correctly or a package playing dirty games with CFLAGS/CXXFLAGS. The dirty games were to try and work out this issue with PPC64 (it built fine on the rest of them) Well, this is expected ;) AFAICT, on the ppc64, the default/implicit -mXXX flags in GCC diverge from what is in RPM_OPT_FLAGS, so you end up with incorrectly compiled binaries, if not correctly passing through RPM_OPT_FLAGS. Conversely, if not passing RPM_OPT_FLAGS, one often gets away with "no build-breakdowns" on all targets, but is facing the kind of issue your are facing on the ppc64. Furthermore, several GCC flags people are using in dirty CFLAGS tricks often actually are target-specific, which will cause build-breakdowns on more exotic target. -ggdb is one of these non-portable options, but I am not sufficiently familiar with RH's ppc64 to be able to judge if -ggdb is valid on RH-ppc64. Ralf -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu July 2 2009, Josh Boyer wrote: > On Thu, Jul 02, 2009 at 04:51:50PM +0100, Peter Robinson wrote: > >No I didn't. Are the details published anywhere of how to request access? > > I'm working on it. Basically, email dwmw2 or me an ssh public key. I added this information to the PPC Sig wiki page: https://fedoraproject.org/wiki/Architectures/PowerPC#PPC_Shell_access_for_debugging Regards Till signature.asc Description: This is a digitally signed message part. -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, Jul 02, 2009 at 06:30:50PM +0200, Till Maas wrote: >On Thu July 2 2009, Josh Boyer wrote: >> On Thu, Jul 02, 2009 at 04:51:50PM +0100, Peter Robinson wrote: > >> >No I didn't. Are the details published anywhere of how to request access? >> >> I'm working on it. Basically, email dwmw2 or me an ssh public key. > >I added this information to the PPC Sig wiki page: >https://fedoraproject.org/wiki/Architectures/PowerPC#PPC_Shell_access_for_debugging > Thanks. That's where I was going to add it too after doing some rework to that page. I'll be sure not to accidentally delete it when I do :) josh -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Thu, Jul 2, 2009 at 9:30 AM, Colin Walters wrote: > On Thu, Jul 2, 2009 at 9:20 AM, Peter Robinson wrote: >> >> Interestingly with -ggdb it builds fine with out without -O0. > > That makes it a lot more likely to be a compiler flaw (though not guaranteed). Of course, it turns out this is very likely not a compiler flaw. Compiler authors everywhere are shocked I'm sure. See: http://bugzilla.gnome.org/show_bug.cgi?id=587823 And http://git.gnome.org/cgit/gobject-introspection/commit/?id=a1f5af4683b08892e87288ef4906782f4094703d -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
>>> Interestingly with -ggdb it builds fine with out without -O0. >> >> That makes it a lot more likely to be a compiler flaw (though not >> guaranteed). > > Of course, it turns out this is very likely not a compiler flaw. > Compiler authors everywhere are shocked I'm sure. > > See: http://bugzilla.gnome.org/show_bug.cgi?id=587823 > And > http://git.gnome.org/cgit/gobject-introspection/commit/?id=a1f5af4683b08892e87288ef4906782f4094703d Doing a scrratch build it gets further but there's another failure, although I haven't tried this with git head. http://koji.fedoraproject.org/koji/getfile?taskID=1457149&name=build.log Peter -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: ppc64 assistance
On Mon, Jul 6, 2009 at 12:07 PM, Peter Robinson wrote: Interestingly with -ggdb it builds fine with out without -O0. >>> >>> That makes it a lot more likely to be a compiler flaw (though not >>> guaranteed). >> >> Of course, it turns out this is very likely not a compiler flaw. >> Compiler authors everywhere are shocked I'm sure. >> >> See: http://bugzilla.gnome.org/show_bug.cgi?id=587823 >> And >> http://git.gnome.org/cgit/gobject-introspection/commit/?id=a1f5af4683b08892e87288ef4906782f4094703d > > Doing a scrratch build it gets further but there's another failure, > although I haven't tried this with git head. > > http://koji.fedoraproject.org/koji/getfile?taskID=1457149&name=build.log This looks relevant: http://sourceware.org/ml/libffi-discuss/2009/msg00185.html -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
