https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78188
--- Comment #21 from rguenther at suse dot de <rguenther at suse dot de> --- On November 4, 2016 3:30:55 PM GMT+01:00, "dje at gcc dot gnu.org" <gcc-bugzi...@gcc.gnu.org> wrote: >https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78188 > >--- Comment #18 from David Edelsohn <dje at gcc dot gnu.org> --- >Changing pass_ipa_comdats::gate to > >return HAVE_COMDAT_GROUP && optimize; > >does not experience the tree-vrp.c bootstrap failure and does not >generate the >numerous additional testsuite failures. > >There is one additional call to set_comdat_group() in cp/optimize.c, >although >that never triggered gcc_assert. And the testsuite failures are not >ICEs. > >The call to set_comdat_group() in varasm.c:make_decl_one_only() >apparently has >some additional effect that allows the additional symbols to appear in >libstdc++. Hmm, so claiming decl-one-only w/o comdat support is likely broken. That is, something assumes comdat groups if make_decl_one_only succeeds. Gating IPA comdat should be fine.
