[v3] libstdc++/16612

[Paolo Carlini]

Benjamin Kosnik wrote:

>Ok, here we go again. I believe this to now be conformant. Hurray.
>
>I'll wait 24 hrs for comments, and then check in a backport. (Which I
>believe will be possible for 3.4/4.0).
>  
>
Hi Benjamin.

Frankly, I cannot support this kind of approach:

*************** namespace std
*** 614,620 ****
        operator[] (size_type __pos) const
        {
      _GLIBCXX_DEBUG_ASSERT(__pos <= size());
!     return _M_data()[__pos];
        }
 
        /**
--- 621,631 ----
        operator[] (size_type __pos) const
        {
      _GLIBCXX_DEBUG_ASSERT(__pos <= size());
!     _CharT* __p = _M_data();
!     if (__p)
!       return __p[__pos];
!     else
!       return _Rep::_S_terminal;
        }

Having a conditional for the most fundamental operation, that is
character access, doesn't seem a price that can be paid in order to
easily support shared memory.

At least, we have to carry out complete benchmarks, and I want to ear
Nathan about this: in my opinion, the very idea of the empty_rep was
meant to avoid this kind of performance issues in his original design.

If really, we cannot avoid this approach, then we should use it *only*
in the shared memory case/making it configurable at build time.

Paolo.