On 02/04/2023 00:00, Jonathan Wakely wrote: > > > On Sat, 1 Apr 2023, 17:58 Jonny Grant, <jg@xxxxxxxx <mailto:jg@xxxxxxxx>> wrote: > > > > On 31/03/2023 23:13, Xi Ruoyao wrote: > > On Fri, 2023-03-31 at 23:12 +0100, Jonny Grant wrote: > >> > >> > >> On 31/03/2023 22:58, Xi Ruoyao wrote: > >>> On Fri, 2023-03-31 at 22:54 +0100, Jonny Grant wrote: > >>>> If the error or warning attribute is used on a function > >>>> declaration > >>>> and a call to such a function is not eliminated through dead code > >>>> elimination or other optimizations, an error or warning > >>>> (respectively) > >>>> that includes message is diagnosed. > >>> > >>> In this example the "call to such a function" is clearly "eliminated > >>> through" inlining (one of "other optimizations"). > >>> > >>>> https://godbolt.org/z/n849GPTjj <https://godbolt.org/z/n849GPTjj> > >>> > >> > >> ok yes, now I understand. The compile_abort() got inlined as abort(). > >> > >> compile_abort(): > >> pushq %rax > >> call abort > >> main: > >> pushq %rax > >> call abort > >> > >> > >> So if I implement it, I must avoid it being optimized (using pragma > >> etc as below) > > > > Or just __attribute__((noipa)). > > That's much clearer. > > It does feel a shame the optimizer inlines the function, discarding the error("message"), before the attribute error("message") can be triggered. But we can just put that __attribute__((noipa)) at least. > > > Why provide a definition? Why do you want to define a function that can never be called, because calling it gives an error? You make a good point. There's no need to define it, the declaration is enough to get the build error. Without the definition, it doesn't get inlined too. Jonny
