[libc++] Make sure __clear_and_shrink() maintains string invariants

[ldionne]

Summary

__clear_and_shrink() was added in D41976, and a test was added alongside
it to make sure that the string invariants were maintained. However, it
appears that the test never ran under UBSan before, which would have
highlighted the fact that it doesn't actually maintain the string
invariants.

[ldionne]

Adding folks who worked on D41976.

[vsk]

Looks reasonable to me, but I'm not comfortable lgtm'ing as I know too little about the string ABI.

It's interesting that UBSan flagged the issue. What diagnostic did you see?

libcxx/include/string

3962

__invariants checks that data()[0] == value_type(0) -- is this guaranteed to be equal to value_type()?

[ldionne]

Use value_type() when checking invariants.

[ldionne]

libcxx/include/string

3962

It looks to me like the invariants check is the odd one out here, since we use value_type() everywhere else.

libcxx/include/string

 basic_string<_CharT, _Traits, _Allocator>::__invariants() const
 {
     if (size() > capacity())
         return false;
     if (capacity() < __min_cap - 1)
         return false;
     if (data() == 0)
         return false;
-    if (data()[size()] != value_type(0))
+    if (data()[size()] != value_type())
         return false;
     return true;
 }
 
 // __clear_and_shrink
 
 template<class _CharT, class _Traits, class _Allocator>
 inline
 void
 basic_string<_CharT, _Traits, _Allocator>::__clear_and_shrink() _NOEXCEPT
 {
     clear();
     if(__is_long())
     {
         __alloc_traits::deallocate(__alloc(), __get_long_pointer(), capacity() + 1);
         __set_long_cap(0);
         __set_short_size(0);
+        traits_type::assign(*__get_short_pointer(), value_type());
     }
 }
 
 // operator==
 
 template<class _CharT, class _Traits, class _Allocator>
 inline _LIBCPP_INLINE_VISIBILITY
 bool