https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83394
--- Comment #2 from Martin Sebor <msebor at gcc dot gnu.org> ---
I can reproduce the missing diagnostic but I can't find anything wrong with the
code emitted for the test case. It's the same as in prior versions of GCC.
Can you clarify in what way the emitted object code is wrong and/or worse than
in previous versions of GCC? I do note that calls to the operator are rejected
with a hard error, same as in GCC 7, so maybe you meant accepts-invalid?
$ cat u.C && gcc -S -Wall u.C
struct A { __attribute__((__noinline__)) operator int (); };
__attribute__((__always_inline__))
A::operator int () { return 0; }
int f (A &a) { return a; }
.file "u.C"
u.C: In function ‘int f(A&)’:
u.C:4:1: error: inlining failed in call to always_inline ‘A::operator int()’:
function not inlinable
A::operator int () { return 0; }
^
u.C:6:23: note: called from here
int f (A &a) { return a; }
^
Regarding comment #1, I'm not attached to where in the C++ front end
declarations with the mutually exclusive attributes are detected, as long as it
doesn't compromise the design goals of the patch (see
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg00769.html and
https://gcc.gnu.org/ml/gcc-patches/2017-08/msg01371.html for details).
