2013/9/5 Jonathan Wakely <jwakely.gcc@xxxxxxxxx>:
>
> For this code:
> $ cat b.cc
> inline void f() __attribute((always_inline)) { }
>
> $ g++ -std=gnu++0x b.cc -Wall -c -O3
> b.cc:1:17: error: attributes are not allowed on a function-definition
> $
>
> However, moving the attribute shows that actually it is allowed:
>
> $ cat b.cc
> inline __attribute((always_inline)) void f() { }
>
> $ g++ -std=gnu++0x b.cc -Wall -c -O3
> $
>
> If placing the attribute at that location is valid then the wording of
> the diagnostic is misleading, at best.
I am not sure it is relevant but, if I understand correctly,
given that a C source with a function definition and
three locations <X>, <Y>, and <Z>:
<X> void <Y> f () <Z>
{ ... }
you cannot place attribute at location <Z> because the C front end
may confuse attribute with old style C function definition:
void f (a, b)
int a;
int b;
{ ... }
(But for the declaration: <X> void <Y> f () <Z>;
All three locations are ok to use attribute.)
If the C++ front end parser uses some common parts of
C front end implementation, could it be a possible cause
having such misleading diagnostic?
Best regards,
jasonwucj
