https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114784
--- Comment #7 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Jakub Jelinek <ja...@gcc.gnu.org>: https://gcc.gnu.org/g:aa73eb97a1e3c84564fa71158d09f9c5582c4d2e commit r14-10086-gaa73eb97a1e3c84564fa71158d09f9c5582c4d2e Author: Jakub Jelinek <ja...@redhat.com> Date: Tue Apr 23 08:36:15 2024 +0200 c++: Copy over DECL_DISREGARD_INLINE_LIMITS flag to inheriting ctors [PR114784] The following testcase is rejected with error: inlining failed in call to 'always_inline' '...': call is unlikely and code size would grow errors. The problem is that starting with the r14-2149 change we try to copy most of the attributes from the inherited to inheriting ctor, but don't copy associated flags that decl_attributes sets. Now, the other clone_attrs user, cp/optimize.cc (maybe_clone_body) copies over DECL_COMDAT (clone) = DECL_COMDAT (fn); DECL_WEAK (clone) = DECL_WEAK (fn); if (DECL_ONE_ONLY (fn)) cgraph_node::get_create (clone)->set_comdat_group (cxx_comdat_group (clone)); DECL_USE_TEMPLATE (clone) = DECL_USE_TEMPLATE (fn); DECL_EXTERNAL (clone) = DECL_EXTERNAL (fn); DECL_INTERFACE_KNOWN (clone) = DECL_INTERFACE_KNOWN (fn); DECL_NOT_REALLY_EXTERN (clone) = DECL_NOT_REALLY_EXTERN (fn); DECL_VISIBILITY (clone) = DECL_VISIBILITY (fn); DECL_VISIBILITY_SPECIFIED (clone) = DECL_VISIBILITY_SPECIFIED (fn); DECL_DLLIMPORT_P (clone) = DECL_DLLIMPORT_P (fn); DECL_DISREGARD_INLINE_LIMITS (clone) = DECL_DISREGARD_INLINE_LIMITS (fn); The following patch just copies DECL_DISREGARD_INLINE_LIMITS to fix this exact bug, not really sure which other flags should be copied and which shouldn't. Plus there are tons of other flags, some of which might need to be copied too, some of which might not, perhaps in both places, like: DECL_UNINLINABLE, maybe DECL_PRESERVE_P, TREE_USED, maybe DECL_USER_ALIGN/DECL_ALIGN, maybe DECL_WEAK, maybe DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT, DECL_NO_LIMIT_STACK. TREE_READONLY, DECL_PURE_P, TREE_THIS_VOLATILE (for const, pure and noreturn attributes) probably makes no sense, DECL_IS_RETURNS_TWICE neither (returns_twice ctor?). What about TREE_NOTHROW? DECL_FUNCTION_SPECIFIC_OPTIMIZATION, DECL_FUNCTION_SPECIFIC_TARGET... Anyway, another problem is that if inherited_ctor is a TEMPLATE_DECL, as also can be seen in the using D<T>::D; case in the testcase, then DECL_ATTRIBUTES (fn) = clone_attrs (DECL_ATTRIBUTES (inherited_ctor)); attempts to copy the attributes from the TEMPLATE_DECL which doesn't have them. The following patch copies them from STRIP_TEMPLATE (inherited_ctor) which does. E.g. DECL_DECLARED_CONSTEXPR_P works fine as the macro itself uses STRIP_TEMPLATE too, but not 100% sure about other macros used on inherited_ctor earlier. 2024-04-23 Jakub Jelinek <ja...@redhat.com> PR c++/114784 * method.cc (implicitly_declare_fn): Call clone_attrs on DECL_ATTRIBUTES on STRIP_TEMPLATE (inherited_ctor) rather than inherited_ctor. Also copy DECL_DISREGARD_INLINE_LIMITS flag from it. * g++.dg/cpp0x/inh-ctor39.C: New test.
