http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60906
--- Comment #8 from Harald van Dijk <harald at gigawatt dot nl> ---
Oh, based on the existing error for
void f();
void f() __attribute__((regparm(3)));
which is accepted on x86-64, but fails on x86-32 with
test.cc:2:36: error: new declaration ‘void f()’
void f() __attribute__((regparm(3)));
^
test.cc:1:6: error: ambiguates old declaration ‘void f()’
void f();
^
even though things like
void f();
void f() __attribute__((noreturn));
and
void f();
void f() __attribute__((deprecated));
are accepted, I assumed that the existing logic to check attribute
compatibility (which I suggested to reuse) was a reliable way of determining
which attributes are and which aren't safe. If that's incorrect, then sure, my
suggestion is not suitable, I can completely see how huge of a maintenance
burden that would become.
