I just submitted a re-build of ODB for GCC 4.9.2 in rawhide ( http://koji.fedoraproject.org/koji/taskinfo?taskID=8051892 ) and it ran into an issue ( http://koji.fedoraproject.org/koji/taskinfo?taskID=8051895 ).
I don't have access to a machine with rawhide to do any more debugging, but I've included the pertinent output below and what's the recommend path for working on getting this resolved?
Thanks, Dave
libtool: compile: g++ -DHAVE_CONFIG_H -I.. -I.. -I/usr/lib/gcc/x86_64-redhat-linux/4.9.2/plugin/include -O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -m64 -mtune=generic -c validator.cxx -fPIC -DPIC -o .libs/validator.o
validator.cxx:1575:1: internal compiler error: in possible_polymorphic_call_targets, at ipa-devirt.c:1557 } ^ Please submit a full bug report, with preprocessed source if appropriate.
On Thu, Nov 06, 2014 at 07:58:01AM -0700, Dave Johansen wrote:
I just submitted a re-build of ODB for GCC 4.9.2 in rawhide ( http://koji.fedoraproject.org/koji/taskinfo?taskID=8051892 ) and it ran into an issue ( http://koji.fedoraproject.org/koji/taskinfo?taskID=8051895 ).
I don't have access to a machine with rawhide to do any more debugging, but I've included the pertinent output below and what's the recommend path for working on getting this resolved?
Perhaps http://gcc.gnu.org/PR60871 , but without a preprocessed testcase it is hard to tell. You could either try local mock, or you could hack for a temporary build small *.spec file change that would BuildRequire: sharutils and run: make ... || tar cjf - /tmp/cc*.out | uuencode cc.tar.bz2 (that way you could download build.log and uudecode the tarball with the preprocessed source from it).
As a workaround, supposedly -fno-devirtualize or -fno-devirtualize-speculatively options could help.
Jakub
On Thu, Nov 6, 2014 at 8:19 AM, Jakub Jelinek jakub@redhat.com wrote:
On Thu, Nov 06, 2014 at 07:58:01AM -0700, Dave Johansen wrote:
I just submitted a re-build of ODB for GCC 4.9.2 in rawhide ( http://koji.fedoraproject.org/koji/taskinfo?taskID=8051892 ) and it ran into an issue (
http://koji.fedoraproject.org/koji/taskinfo?taskID=8051895
).
I don't have access to a machine with rawhide to do any more debugging,
but
I've included the pertinent output below and what's the recommend path
for
working on getting this resolved?
Perhaps http://gcc.gnu.org/PR60871 , but without a preprocessed testcase it is hard to tell. You could either try local mock, or you could hack for a temporary build small *.spec file change that would BuildRequire: sharutils and run: make ... || tar cjf - /tmp/cc*.out | uuencode cc.tar.bz2 (that way you could download build.log and uudecode the tarball with the preprocessed source from it).
As a workaround, supposedly -fno-devirtualize or -fno-devirtualize-speculatively options could help.
I had actually turned off the use of devirtualization with GCC 4.9.0, but I guess that that is has cropped back up or wasn't fully resolved. http://pkgs.fedoraproject.org/cgit/odb.git/commit/?id=b9ca412b135ea0ef16d3b5...
For ODB, I can continue turning of devirtualization, but I just wanted to make sure that the right thing was done from the GCC end of things and make sure that an issue is left around.
-
Dave Johansen -
Jakub Jelinek