I guess the problem was that we were passing memcpy the pointer &*R.begin(), which might be null if the string_view is empty?

C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.39.33519\include\xstring

template <class _Traits>
class _String_view_iterator {
public:
    _NODISCARD constexpr reference operator*() const noexcept {
#if _ITERATOR_DEBUG_LEVEL >= 1
        _STL_VERIFY(_Mydata, "cannot dereference value-initialized string_view iterator");
        _STL_VERIFY(_Myoff < _Mysize, "cannot dereference end string_view iterator");  // <<< EXCEPTION HERE
        return _Mydata[_Myoff];
#else // ^^^ _ITERATOR_DEBUG_LEVEL >= 1 / _ITERATOR_DEBUG_LEVEL == 0 vvv
        return *_Myptr;
#endif // ^^^ _ITERATOR_DEBUG_LEVEL == 0 ^^^
    }

Ah this was failing in the new test I added. I guess we never previously called prepend with an empty string_view in our demangling test-suite. Are we sure we want to now support that use-case (as opposed to removing the problematic test)? . If so, we also need to make the change on the libcxxabi side

Ah this was failing in the new test I added. I guess we never previously called prepend with an empty string_view in our demangling test-suite. Are we sure we want to now support that use-case (as opposed to removing the problematic test)? . If so, we also need to make the change on the libcxxabi side

I think the size checking is useful. I will add it to libcxxabi.