https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100470
Jonathan Wakely <redi at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Keywords| |wrong-code
Last reconfirmed| |2022-09-20
--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2019/p1286r2.html says that
S" has a potentially-throwing move constructor.
It looks like the __is_nothrow_constructible built-in gets this wrong:
using size_t = decltype(sizeof(0));
void* operator new(size_t, void* p) noexcept { return p; }
namespace std {
template<typename T> T&& declval() noexcept;
template<typename T>
constexpr bool is_nothrow_move_constructible_v
= noexcept(new (declval<void*>()) T(declval<T>()));
}
struct S1{
S1(S1&&) noexcept(false);
};
struct S2{
S2(S2&&) noexcept(false) = default;
};
struct S3{
S3(S3&&) noexcept(false){}
};
struct S4{
S4(S4&&) = default;
};
static_assert(!std::is_nothrow_move_constructible_v<S1>); // OK
static_assert(!std::is_nothrow_move_constructible_v<S2>); // OK
static_assert(!std::is_nothrow_move_constructible_v<S3>); // OK
static_assert( std::is_nothrow_move_constructible_v<S4>); // OK
static_assert(!__is_nothrow_constructible(S1, S1)); // OK
static_assert(!__is_nothrow_constructible(S2, S2)); // failed
static_assert(!__is_nothrow_constructible(S3, S3)); // OK
static_assert( __is_nothrow_constructible(S4, S4)); // OK
This makes it a libstdc++ regression in GCC 11.1 and later, because we use
__is_nothrow_construcitble now, instead of a pure library implementation for
the traits.