On Tue, Nov 1, 2016 at 4:00 PM, Bernd Edlinger <bernd.edlin...@hotmail.de> wrote: > On 11/01/16 20:48, Jason Merrill wrote: >>>>>>> else if ((DECL_EXTERN_C_P (newdecl) >>>>>>> && DECL_EXTERN_C_P (olddecl)) >>>>>>> || compparms (TYPE_ARG_TYPES (TREE_TYPE (newdecl)), >>>>>>> TYPE_ARG_TYPES (TREE_TYPE (olddecl)))) >>>> >>>> So I was thinking to drop the "else" and the compparms test. >>> >>> Yes. But then we must somehow avoid: >>> >>> else >>> /* Discard the old built-in function. */ >>> return NULL_TREE;
>>> It maybe easier, just to copy the warning to the DECL_ANTICIPATED case? >> >> Or even move it there; removing the existing warning doesn't change >> anything in the testsuite, and I'm having trouble imagining how to >> trigger it. >> > > Nice. It must be something, which does not anticipate a declaration. > > like: > > cat test.cc > int __builtin_abort(void*); > > g++ -c test.cc > test.cc:1:5: warning: new declaration 'int __builtin_abort(void*)' > ambiguates built-in declaration 'void __builtin_abort()' > int __builtin_abort(void*); > ^~~~~~~~~~~~~~~ > > Intersting, the warning comes even though I forgot to add the > extern "C". Looks like anything starting with __builtin is implicitly extern "C" (set in grokfndecl). If we remove the DECL_ANTICIPATED check, I see some failures in builtin* tests due to missing extern "C". That seems appropriate at file scope, but I'm not sure it's right for namespace std. Jason
