CC no longer working for unique_ptr

[killerbot]

I can confirm that the first problem is solved : foo.

But the second one still fails : bar->

[Loaden]

I can confirm that the first problem is solved : foo.

But the second one still fails : bar->

About the second issue: shared_ptr<Test> and auto_ptr<Test> works well.
So, There should has some thing changed with unique_ptr.

[Loaden]

I can't solved this issue. It's too complex.

Code Select Expand

/// 20.7.12.2 unique_ptr for single objects.
 template <typename _Tp, typename _Dp = default_delete<_Tp> >
   class unique_ptr
   {
     // use SFINAE to determine whether _Del::pointer exists
     class _Pointer
     {
template<typename _Up>
 static typename _Up::pointer __test(typename _Up::pointer*);

template<typename _Up>
 static _Tp* __test(...);

typedef typename remove_reference<_Dp>::type _Del;

     public:
typedef decltype( __test<_Del>(0)) type;
     };

     typedef std::tuple<typename _Pointer::type, _Dp>  __tuple_type;
     __tuple_type                                      _M_t;

   public:
     typedef typename _Pointer::type   pointer;
...
     pointer
     operator->() const
     {

_GLIBCXX_DEBUG_ASSERT(get() != pointer());
return get();
     }

See: typedef typename _Pointer::type   pointer;

I don't understood what's the meaning.

If change to:

Code Select Expand

      _Tp*
     operator->() const
     {

_GLIBCXX_DEBUG_ASSERT(get() != pointer());
return get();
     }

Every thing will works fine.

[ollydbg]

Code Select Expand

typedef typename _Pointer::type   pointer;
we should resolve the actual type of the string "_Pointer::type".

As the definition of the embedded class _Pointer, "_Pointer::type" is actually a type from the "decltype operator" deduced from the expression "__test<_Del>(0)".

Then "__test<_Del>(0)" is a function calling expression defined by

Code Select Expand

template<typename _Up>
 static _Tp* __test(...);


So, it's actual type is "_Tp*".

As a conclusion: C++'s type-deduction mechanism is heaven for programmer, but it is the hell for compiler/parser writer. , each step should involve the symbol table check and semantics check.
I don't think it can be implemented in CC, it was too complex.