https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123811
Patrick Palka <ppalka at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|Two conversion operators |member function
|that differ only in |template-head requirements
|constraints cause ICE with |not included in mangling
|checking |
--- Comment #2 from Patrick Palka <ppalka at gcc dot gnu.org> ---
(In reply to Patrick Palka from comment #1)
> Seems we're neglecting to mangle constraints on user-defined conversion
> functions, leading to the logically distinct conversion functions having the
> same same mangling. Clang mangles the constrained one as
> _ZNK1ScvT_IiQ11always_trueIS0_EEEv.
This bug actually seems to be more general: we don't include template-head
requirements of member functions at all, only their trailing requirements:
template<typename T> concept always_true = true;
template<class T> static void foo() { }
template<class T> requires always_true<T> static void foo() { }
template<class T> static void bar() { }
template<class T> static void bar() requires always_true<T> { }
struct B {
template<class T> static void foo() { }
template<class T> requires always_true<T> static void foo() { }
template<class T> static void bar() { }
template<class T> static void bar() requires always_true<T> { }
};
int main(){
foo<int>(); // _Z3fooIiQ11always_trueIT_EEvv
bar<int>(); // _Z3barIiEvvQ11always_trueIT_E
B::foo<int>(); // _ZN1B3fooIiEEvv ???
B::bar<int>(); // _ZN1B3barIiEEvvQ11always_trueIT_E
}
So a workaround for your testcase is to use a trailing-requirement:
template<typename T>explicit operator T() requires always_true<T>
