https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88134
--- Comment #32 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Segher Boessenkool from comment #31) > A straightforward backport of that gives > > In file included from /home/segher/src/gcc/gcc/config/rs6000/rs6000.cc:28765: > ./gt-rs6000.h:143:6: error: 'atomic_update_decl' was not declared in this > scope > 143 | &atomic_update_decl, > | ^~~~~~~~~~~~~~~~~~ > ./gt-rs6000.h:145:13: error: 'atomic_update_decl' was not declared in this > scope > 145 | sizeof (atomic_update_decl), > | ^~~~~~~~~~~~~~~~~~ > ./gt-rs6000.h:150:6: error: 'atomic_clear_decl' was not declared in this > scope > 150 | &atomic_clear_decl, > | ^~~~~~~~~~~~~~~~~ > ./gt-rs6000.h:152:13: error: 'atomic_clear_decl' was not declared in this > scope > 152 | sizeof (atomic_clear_decl), > | ^~~~~~~~~~~~~~~~~ > ./gt-rs6000.h:157:6: error: 'atomic_hold_decl' was not declared in this scope > 157 | &atomic_hold_decl, > | ^~~~~~~~~~~~~~~~ > ./gt-rs6000.h:159:13: error: 'atomic_hold_decl' was not declared in this > scope > 159 | sizeof (atomic_hold_decl), > | ^~~~~~~~~~~~~~~~ > > Help? The issue is that the decls are inside #ifdef RS6000_GLIBC_ATOMIC_FENV ... #endif but gengtype will unconditionally register the root, refering to the variables. I suggest to remove the #ifdef guard and instead wrap the decls inside #pragma GCC diagnostic push #pragma GCC diagnostic ignored "-Wunused" .. #pragma GCC idagnostic pop The C++ FE doesn't seem to allow ATTRIBUTE_UNUSED anywhere here and I guess "fixing" gengtype isn't a thing you want to do ;)
