Fail to emerge =x11-libs/qt-webkit-4.8.0 with: internal compiler error: in simplify_gen_subreg_concatn, at lower-subreg.c:489 Please submit a full bug report Reproducible: Always Steps to Reproduce: 1. emerge =x11-libs/qt-webkit-4.8.0 2. 3. Actual Results: Emerge failed. Expected Results: Emerged =x11-libs/qt-webkit-4.8.0 .
Created attachment 300439 [details] build.log
Created attachment 300441 [details] environment
Created attachment 300443 [details] info
Created attachment 300445 [details] pqv
Rerun the command with -save-temps and attach the resulting preprocessed source (*.ii).
Created attachment 300511 [details] Script*.ii & SerializedScriptValue.ii files All *.ii files is ~ 650Mb... I can attach another needed files.
I've found, that it's gcc bug - https://bugs.launchpad.net/gcc-linaro/+bug/795131
Oops, I meant run the g++ command line that caused the error. In any case SerializedScriptValue.ii is the right one. I don't see this reported upstream so I guess the Linaro guys never got around to it.
Workaround for qt-webkit: emerge it by =sys-devel/gcc-4.5.3-r2
Is this fixed in gcc-4.6.3?
I had try to build with gcc-4.6.3 from https://bugs.gentoo.org/show_bug.cgi?id=406571 , but got same error last month. Now I'm need to build gcc-4.6.3 from portage tree, and try again.
Same error through build =x11-libs/qt-webkit-4.8.2 ... build with gcc-4.5.3 - success.
According to original reporter on gcc bugzie it was fixed by: http://gcc.gnu.org/viewcvs?view=revision&revision=171520 Changset was not backported to gcc-4.6 branch
Any updates on this? Looks like there is only need of including that patch in our patchset, no?
(In reply to comment #14) > Any updates on this? Looks like there is only need of including that patch > in our patchset, no? I encountered this when building KDE for the Raspberry Pi. Applying that patch to our GCC 4.6.3 package results in the following ICE when rebuilding GCC: /var/tmp/portage/sys-devel/gcc-4.6.3/work/gcc-4.6.3/gcc/tree-ssa-sccvn.c: In function 'vn_nary_may_trap': /var/tmp/portage/sys-devel/gcc-4.6.3/work/gcc-4.6.3/gcc/tree-ssa-sccvn.c:3669:1: internal compiler error: in df_bb_refs_record, at df-scan.c:3578
That sounds more like a bug than a porting issue...
(In reply to comment #16) > That sounds more like a bug than a porting issue... a bug in gcc that is
(In reply to comment #15) > (In reply to comment #14) > > Any updates on this? Looks like there is only need of including that patch > > in our patchset, no? > > I encountered this when building KDE for the Raspberry Pi. Applying that > patch to our GCC 4.6.3 package results in the following ICE when rebuilding > GCC: > > /var/tmp/portage/sys-devel/gcc-4.6.3/work/gcc-4.6.3/gcc/tree-ssa-sccvn.c: In > function 'vn_nary_may_trap': > /var/tmp/portage/sys-devel/gcc-4.6.3/work/gcc-4.6.3/gcc/tree-ssa-sccvn.c: > 3669:1: internal compiler error: in df_bb_refs_record, at df-scan.c:3578 ryao, I can try at my end with armv7a. Just to be clear in reproducing this, you applied the patch from http://gcc.gnu.org/viewcvs?view=revision&revision=171520 to our in tree gcc-4.6.3 and hit that when rebuilding gcc.
Atropos-XR ~ # emerge x11-libs/qt-webkit -vat1 These are the packages that would be merged, in reverse order: Calculating dependencies... done! [ebuild N ] x11-libs/qt-webkit-4.8.4:4 USE="exceptions gstreamer icu jit (-aqua) -debug -pch" 0 kB [ebuild N ] x11-libs/qt-xmlpatterns-4.8.4:4 USE="(-aqua) -c++0x -debug -pch" 0 kB Total: 2 packages (2 new), Size of downloads: 0 kB Would you like to merge these packages? [Yes/No] yes >>> Verifying ebuild manifests >>> Starting parallel fetch >>> Emerging (1 of 2) x11-libs/qt-xmlpatterns-4.8.4 >>> Installing (1 of 2) x11-libs/qt-xmlpatterns-4.8.4 >>> Emerging (2 of 2) x11-libs/qt-webkit-4.8.4 >>> Installing (2 of 2) x11-libs/qt-webkit-4.8.4 >>> Jobs: 2 of 2 complete Load avg: 2.27, 4.09, 4.70 >>> Auto-cleaning packages... >>> No outdated packages were found on your system. * GNU info directory index is up-to-date. Atropos-XR ~ # gcc-config -l [1] armv7a-hardfloat-linux-gnueabi-4.5.4 [2] armv7a-hardfloat-linux-gnueabi-4.6.3 * Successfully builds with current gcc and current portage tree. We sure this is a bug and not just someone running out of resources or something? Or perhaps it affects <4.8.4?
(In reply to comment #19) > Successfully builds with current gcc and current portage tree. > > We sure this is a bug and not just someone running out of resources or > something? Or perhaps it affects <4.8.4? Try compiling SerializedScriptValue.ii that's attached to this bug. The fact that newer versions of qt-webkit doesn't trigger ICE, doesn't mean that it was fixed in gcc.
(In reply to comment #20) > (In reply to comment #19) > > Successfully builds with current gcc and current portage tree. > > > > We sure this is a bug and not just someone running out of resources or > > something? Or perhaps it affects <4.8.4? > > Try compiling SerializedScriptValue.ii that's attached to this bug. The fact > that newer versions of qt-webkit doesn't trigger ICE, doesn't mean that it > was fixed in gcc. um, yeah, but if nothing in the tree triggers this ICE why are we holding stabilization? Mind you, if I'm the only guy randomly lucky enough to build successfully then we should hold off, but if we start holding off stabilization of gcc because of bugs that are already worked around we will never stabilize gcc...
> Try compiling SerializedScriptValue.ii that's attached to this bug. The fact > that newer versions of qt-webkit doesn't trigger ICE, doesn't mean that it > was fixed in gcc. I grabbed the minimal test case on the gcc bug and the ICE is still present, but I cannot trigger it on qt-webkit or anything else I've tested. Atropos-XR ii # g++ -O2 -mfpu=vfpv3 -mfloat-abi=hard -march=armv7-a -mthumb -fpermissive -w -c webkit_testcase-min.i webkit_testcase-min.i: In member function 'bool CloneSerializer::serialize(JSValue)': webkit_testcase-min.i:80:14: internal compiler error: in simplify_gen_subreg_concatn, at lower-subreg.c:489 Please submit a full bug report, with preprocessed source if appropriate. See <http://bugs.gentoo.org/> for instructions. Atropos-XR ii # g++ -O2 -mfpu=vfpv3 -mfloat-abi=hard -march=armv7-a -mthumb -w -c webkit_testcase-min.i webkit_testcase-min.i: In member function 'void Vector<T, inlineCapacity>::append(const U&) [with U = long long int, T = unsigned char, unsigned int inlineCapacity = 0u]': webkit_testcase-min.i:40:19: instantiated from 'void writeLittleEndian(Vector<unsigned char>&, T) [with T = long long int]' webkit_testcase-min.i:66:50: instantiated from here webkit_testcase-min.i:28:22: error: invalid conversion from 'const void*' to 'void*' [-fpermissive] webkit_testcase-min.i:8:30: error: initializing argument 2 of 'void* operator new(std::size_t, void*)' [-fpermissive] On an interesting note, if I remove -fpermissive it doesn't ICE, it just produces an error. Not a toolchain guy so maybe this is worthless information, but I'm sharing it in case it's not worthless.
> um, yeah, but if nothing in the tree triggers this ICE why are we holding > stabilization? Because I don't want to have to rush a fix into stable two months down the line when something else in the tree starts triggering it. Because putting stuff I know is broken into stable makes me sad. etc. I see other distros using this patch so I'd like to assume ryao's error is unrelated.
(In reply to comment #23) every gcc we stabilize has bugs (ices/bad code gen/etc...). it's purely a trade off scheme. if there aren't any big blockers in the tree that we know of, then let's move forward.
(In reply to comment #22) > > Try compiling SerializedScriptValue.ii that's attached to this bug. The fact > > that newer versions of qt-webkit doesn't trigger ICE, doesn't mean that it > > was fixed in gcc. > > I grabbed the minimal test case on the gcc bug and the ICE is still present, > but I cannot trigger it on qt-webkit or anything else I've tested. Now, if you could rebuild your gcc with patch from c#13 and see if it helps or gives another problem like described in c#14, that'd be much appreciated. Maybe get in touch with blueness, so that both of you don't waste cycles simultaneously...
My comment on 19 was misleading, I have more familiarity with the bug now and the reason I can build everything is because of my -mfpu setting: Atropos-XR ii # g++ -O2 -mfpu=vfpv3 -mfloat-abi=hard -march=armv7-a -mthumb -fpermissive -w -c webkit_testcase-min.i webkit_testcase-min.i: In member function 'bool CloneSerializer::serialize(JSValue)': webkit_testcase-min.i:80:14: internal compiler error: in simplify_gen_subreg_concatn, at lower-subreg.c:489 Please submit a full bug report, with preprocessed source if appropriate. See <http://bugs.gentoo.org/> for instructions. Atropos-XR ii # g++ -O2 -mfpu=neon -mfloat-abi=hard -march=armv7-a -mthumb -fpermissive -w -c webkit_testcase-min.i Atropos-XR ii # I'm testing the patches from c#13 now sidenote, <3 neon
I still run into this problem here. However my gcc-4.6.3 was built on 2012-09-10, so it does not include some patches which were added later. I will rebuild gcc (with and without the patch from comment 13) and see if there is a problem afterwards.
(In reply to comment #13) > According to original reporter on gcc bugzie it was fixed by: > > http://gcc.gnu.org/viewcvs?view=revision&revision=171520 > > Changset was not backported to gcc-4.6 branch this patch resolves the ICE in the simplified test case. it will be several hours for me to test qt-webkit, but looks promising.
Two test platforms, same results build qt-webkit with -mfpu=neon and no patch, pass build qt-webkit with -mfpu=vfpv3 and patch from c13, pass generic test case failed before patch, passes after, both on my chromebook a15 and my odroid a9. Patch is good, please add it and let's get 4.6 stable.
(In reply to comment #18) > (In reply to comment #15) > > (In reply to comment #14) > > > Any updates on this? Looks like there is only need of including that patch > > > in our patchset, no? > > > > I encountered this when building KDE for the Raspberry Pi. Applying that > > patch to our GCC 4.6.3 package results in the following ICE when rebuilding > > GCC: > > > > /var/tmp/portage/sys-devel/gcc-4.6.3/work/gcc-4.6.3/gcc/tree-ssa-sccvn.c: In > > function 'vn_nary_may_trap': > > /var/tmp/portage/sys-devel/gcc-4.6.3/work/gcc-4.6.3/gcc/tree-ssa-sccvn.c: > > 3669:1: internal compiler error: in df_bb_refs_record, at df-scan.c:3578 > > ryao, I can try at my end with armv7a. Just to be clear in reproducing > this, you applied the patch from > > http://gcc.gnu.org/viewcvs?view=revision&revision=171520 > > to our in tree gcc-4.6.3 and hit that when rebuilding gcc. Yes.
I am no longer able to reproduce the ICE that I encountered when building GCC with this patch applied. I have no idea why, but I see no reason to block this patch from being merged into the patchset.
Fixed in 4.6.3 p1.10.
