https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109353
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
So it seems to be
_M_insert_aux(iterator __position, bool __x)
{
if (this->_M_impl._M_finish._M_p != this->_M_impl._M_end_addr())
{
std::copy_backward(__position, this->_M_impl._M_finish,
this->_M_impl._M_finish + 1);
*__position = __x;
++this->_M_impl._M_finish;
}
else
{
const size_type __len =
_M_check_len(size_type(1), "vector<bool>::_M_insert_aux");
_Bit_pointer __q = this->_M_allocate(__len);
iterator __start(std::__addressof(*__q), 0);
iterator __i = _M_copy_aligned(begin(), __position, __start); // <--
*__i++ = __x;
where the compiler fails to optimize the case of no existing allocation.
Since that's from the push_back () itself no caching of the null start
in libstdc++ is possible.
Disabling early threading avoids the diagnostic because we then inline
very differently, not exposing most of the above.
