https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66190
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> --- That looks just wrong. ADDR_EXPR's operand isn't necessarily a decl, testing TREE_STATIC on random trees can give pretty random answers. I think what matters is where do we cp_genericize_r the DECL_INITIAL of a static variable, if that happens, perhaps we should track that case in cp_genericize_data and not instrument anything in there.
