https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97986
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jsm28 at gcc dot gnu.org
--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
We end up gimplifying
int * input = SAVE_EXPR <n>, (int *) &VA_ARG_EXPR <&ap>;
I think this is garbage coming from the frontend which should either
reject this as comment#1 says or in case the rvalue should be produced
use a TARGET_EXPR for the purpose of address-taking and array-to-pointer
decay.
The case of a function type (which would decay to a pointer as well) is
rejected:
In file included from t.c:1:
t.c:7:29: error: second argument to ‘va_arg’ is a function type ‘int(void)’
7 | int *input = va_arg(ap, int()(int));
| ^~~
and clang says
t.c:7:32: error: function cannot return function type 'int (int)'
7 | int *input = va_arg(ap, int()(int));
| ^
but a function can also not return an array type.
