http://sourceware.org/bugzilla/show_bug.cgi?id=13229
--- Comment #9 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-10-01 15:19:27 UTC --- OK, I now understand the problem. It is partly GNU LD issue. What happens is 1) nsGNOMEShellService includes header that define gfxUnknownSurface. The class is obviously uunused but it has virtual destructor and thus GCC decides to keep it around until after interprocedural optimization for possible devirtualization. Note that even with -O0 the class goes trivially away. This is stupid but should not affect code quality 2) Other shared library, libxul.so also included some modules that use&define gfxUnknownSurface. It is linked with LTO info and resolution is: 11903 34650d5e PREVAILING_DEF_IRONLY_EXP _ZTV17gfxUnknownSurface This consequentely makes GCC to privatize it and rename it to static _ZTV17gfxUnknownSurface.local.<somerandomnumbers> jh@evans:/abuild/jh/build-mozilla-new14-gnu-ld/toolkit/library> grep _ZTV17gfxUnknownSurface *.s libxul.so.ltrans18.s: leaq 16+_ZTV17gfxUnknownSurface.local.102757(%rip), %rax libxul.so.ltrans18.s: .hidden _ZTV17gfxUnknownSurface.local.102757 libxul.so.ltrans23.s: .hidden _ZTV17gfxUnknownSurface.local.102757 libxul.so.ltrans23.s: .globl _ZTV17gfxUnknownSurface.local.102757 libxul.so.ltrans23.s: .type _ZTV17gfxUnknownSurface.local.102757, @object libxul.so.ltrans23.s: .size _ZTV17gfxUnknownSurface.local.102757, 144 libxul.so.ltrans23.s:_ZTV17gfxUnknownSurface.local.102757: 3) Despite the fact that original _ZTV17gfxUnknownSurface has been fully optimized out from libxul.so, GNU LD incorrectly assigns a dynamic linker entry for it: jh@evans:/abuild/jh/build-mozilla-new14-gnu-ld/toolkit/library> nm -D libxul.so| grep _ZTV17gfxUnknownSurface 0000000000000000 W _ZTV17gfxUnknownSurface 4) Finally at a time we link the new shared library -llibxul is used. This makes linker to notice dynamic linker entry for _ZTV17gfxUnknownSurface and consequentely it changes prevailance info to PREVAILING_DEF as quoted above. This makes GCC to no longer optimize the symbol out and finally we die on undefined reference. So in addition to your patch, I think there are two bugs: 1) GNU LD should not produce dynamic linker entries for PREVAILING_DEF_IRONLY_EXP that has been optimized out. This is very important since in C++ code this is a mjority and size of linker table would grow a lot 2) The class is probably by some .h trickery hidden visibility for libxul and default visibility for the other library. Is it correct for gold&GNU LD to make the resolution PREVAILING_DEF because the symbol is mentioned but not used by some other shared library here? -- Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug. _______________________________________________ bug-binutils mailing list bug-binutils@gnu.org https://lists.gnu.org/mailman/listinfo/bug-binutils
