[RFC] First 3.5 project: don't assign _S_terminal in string::c_str()

[Paolo Carlini]

Hi,

in the current basic_string implementation we are quite careful
in keeping the internal representation null-terminated: I have
spotted only one place which should be fixed: _M_clone.

So, I would like to fix that and then remove the assignment:

const _CharT*
c_str() const
{
  // MT: This assumes concurrent writes are OK.
  const size_type __n = this->size();                     <=====
  traits_type::assign(_M_data()[__n], _Rep::_S_terminal); <=====
  return _M_data();
}

This is a bit risky (-> a tiny purely mainline project), because
if we do something wrong the user may call c_str() and have back
a non-null-terminated entity, completely unusable as a C-string.

On the other hand, removing the assign makes c_str way faster.

I have already checked that apparently we are already doing the
right thing for empty strings too, thanks to this:

  // The following storage is init'd to 0 by the linker, resulting
  // (carefully) in an empty string with one reference.
  static size_type _S_empty_rep_storage[];

Paolo.