The subject test includes this function:

void test_for_non_eager_instantiation() {

// Ensure we don’t accidentally instantiate std::basic_string<Nonsense>

// since it may not be well formed and can cause an error in the

// non-immediate context.

static_assert(!std::is_constructible<std::bitset<3>, Nonsense*>::value, “”);

static_assert(!std::is_constructible<std::bitset<3>, Nonsense*, size_t, Nonsense&, Nonsense&>::value, “”);


which requires nonstandard behavior. Should these assertions be libc+±specific? (Should someone file an LWG issue to standardize the expected behavior?)

They’re not necessary if you allow std::string<> to be instantiated with garbage.

We found it distasteful that std::bitset poisoned overload sets containing it.
For example, MSVC should accept this code but instead emits a hard compile error.

void f(std::bitset<101>);
void f(void*);
void* p;
f({p}); // Blows up.

So despite the standard not requiring std::bitset to SFINAE it’s constructors,
it probably means to require that they don’t poison overload resolution by blowing up.