Expected: sys-devel/gcc-3.4.6-r1 to compile cleanly on PPC32 Actual: sys-devel/gcc-3.4.6-r1 fails to compile on PPC32 Reproducible: Always Observations: Ebuild appears to configure with '--disable-multilib', but then to attempt to compile multilib. Details: # sudo emerge -pv gcc [ebuild U ] sys-devel/gcc-3.4.6-r1 [3.4.5-r1] +altivec -bootstrap -boundschecking -build +doc* +fortran +gcj +gtk -hardened (-ip28) -ip32r10k (-multilib) -multislot (-n32) (-n64) +nls -nocxx -nopie -nossp +objc -vanilla 0 kB
Created attachment 88306 [details] Log of failed emerge =gcc-3.4.6-r1
Created attachment 88307 [details] Output of emerge info
this looks like the correct build error ... try emerging with MAKEOPTS=-j1 ./genflags /var/tmp/portage/gcc-3.4.6-r1/work/gcc-3.4.6/gcc/config/rs6000/rs6000.md > tmp-flags.h ./genflags /var/tmp/portage/gcc-3.4.6-r1/work/gcc-3.4.6/gcc/config/rs6000/rs6000.md > tmp-flags.h /bin/sh /var/tmp/portage/gcc-3.4.6-r1/work/gcc-3.4.6/gcc/move-if-change tmp-flags.h insn-flags.h insn-flags.h is unchanged echo timestamp > s-flags ./genconfig /var/tmp/portage/gcc-3.4.6-r1/work/gcc-3.4.6/gcc/config/rs6000/rs6000.md > tmp-config.h /bin/sh /var/tmp/portage/gcc-3.4.6-r1/work/gcc-3.4.6/gcc/move-if-change tmp-flags.h insn-flags.h mv: cannot stat `tmp-flags.h': No such file or directory make[4]: *** [s-flags] Error 1 make[4]: Leaving directory `/var/tmp/portage/gcc-3.4.6-r1/work/build/gcc' make[3]: *** [crtbegin.o] Error 2
Yes, disabling parallel make was the solution. Just to note, I've had success parallel-building gcc on both gentoo/x86 and gentoo/ppc for many years now. Perhaps, a notice, or automatic override of MAKEOPTS=-j1, should be added to the ebuild (or eclass)? A notice would be friendlier. (Parallel build '-j2' did work on i686.)
I can't reproduce this in my stable chroot with -j2. Can anyone else?
ive always used -j6 or higher w/out problems
After installing Portage 2.1, I had need to re-emerge sys-devel/gcc-3.4.6-r1 . The parallel build (-j2) fails again; same place in the compilation process. Serial build (-j1) does not fail. It occurred to me that this might be related to the filesystem I am using. Having 1.5 GB of RAM, I mount tmpfs to /var/tmp/portage for compilations that should take less than 1.0 GB of space. Perhaps the timing of a RAM fs exacerbates incomplete dependencies? (I cannot reproduce the issue on an x86.)
do not build on tmpfs, it is broken atm *** This bug has been marked as a duplicate of 114722 ***