Created attachment 348232 [details] emerge --info '=sys-devel/autogen-5.17.3' When install gentoo-x86_64 on vda (kvm) with latest stage3 and latest porteges -> next bug appears: grub emerging depend on autogen emerging; but autogen compilation ends with error /bin/sh ../libtool --tag=CC --mode=link x86_64-pc-linux-gnu-gcc -std=gnu99 -O2 -march=native -pipe -fomit-frame-pointer -Wl,-O1 -Wl,--as-needed -o columns cols.o ../autoopts/libopts.la -pthread -L/usr/lib64 -lguile -lltdl -lgmp -lcrypt -lm -lltdl -lm -ldl libtool: link: x86_64-pc-linux-gnu-gcc -std=gnu99 -O2 -march=native -pipe -fomit-frame-pointer -Wl,-O1 -o .libs/columns cols.o -pthread -Wl,--as-needed ../autoopts/.libs/libopts.so -L/usr/lib64 /usr/lib64/libguile.so -lgmp -lcrypt /usr/lib64/libltdl.so -lm -ldl -pthread top_builddir=.. \ /var/tmp/portage/sys-devel/autogen-5.17.3/work/autogen-5.17.3/agen5/autogen -L../autoopts/tpl -L../autoopts/tpl -MFstamp-agtexi -Tagtexi-cmd.tpl -DLEVEL=section --timeout=`expr 11 '*' 3` ./opts.def AutoGen aborting on signal 4 (Illegal instruction) in state OPTIONS make[2]: *** [stamp-agtexi] Аварийный останов make[2]: Выход из каталога `/var/tmp/portage/sys-devel/autogen-5.17.3/work/autogen-5.17.3/columns' make[1]: *** [all-recursive] Ошибка 1 make[1]: Выход из каталога `/var/tmp/portage/sys-devel/autogen-5.17.3/work/autogen-5.17.3' make: *** [all] Ошибка 2 * ERROR: sys-devel/autogen-5.17.3 failed (compile phase): * emake failed It is very regrettable.
I bet replacing -march=native with something else (or nothing) will get you around the bug. My guess is this is a problem in kvm (or, less likely, gcc), which means you're probably not getting a fix anytime soon. By the way, -march=native is known to cause (rare) compilation failures even without virtualization.
Yes, that was cause. I changed march according to CPU-model of the host system, and emerging became successful. (Don't remember source, but there was recommended to use "march=native" for guest systems in kvm-virtulization.)
(In reply to comment #1) > I bet replacing -march=native with something else (or nothing) will get you > around the bug. My guess is this is a problem in kvm (or, less likely, gcc), > which means you're probably not getting a fix anytime soon. > > By the way, -march=native is known to cause (rare) compilation failures even > without virtualization. Thanks, Roman.
(In reply to comment #3) > Thanks, Roman. No problem.