https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99186
Jonathan Wakely <redi at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords|diagnostic |
--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Reduced to remove the library dependency:
template<int N, typename T, typename... U>
struct tuple_impl : tuple_impl<N + 1, U...>
{ };
template<int N, typename T>
struct tuple_impl<N, T>
{ };
template<typename T, typename U> struct tuple : tuple_impl<0, T, U> { };
template<typename T, int N, typename... U>
void
get(const tuple_impl<N, T, U...>&)
{ }
enum class E1 {a};
enum class E2 {b,c};
template < auto >
struct S
{
int i;
};
int main()
{
tuple<S<E1::a>,S<E2::b>> x;
get<S<E2::b>>(x); // does not compile
}
Clang and EDG compile this, G++ has a bogus ambiguity:
ambig.C: In function ‘int main()’:
ambig.C:28:18: error: invalid initialization of reference of type ‘const
tuple_impl<0, S<E2::b>, S<E2::b> >&’ from expression of type ‘tuple<S<E1::a>,
S<E2::b> >’
28 | get<S<E2::b>>(x); // does not compile
| ^
ambig.C:13:5: note: in passing argument 1 of ‘void get(const tuple_impl<N, T, U
...>&) [with T = S<E2::b>; int N = 0; U = {S<E2::b>}]’
13 | get(const tuple_impl<N, T, U...>&)
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The bug is that S<E1::a> and S<E2::b> are distinct types but the compiler gets
them confuses.
We have several bugs about template<auto> types that get confused.
