为什么对嵌套类型的访问会影响C ++中的成员解析？

Question

Assume I have a simple structure...

struct A {
    std::vector<uint32_t> v;
};

And I want to find the size of the vector element type...

sizeof(A::v);                        // legal
sizeof(decltype(A().v)::value_type); // legal
sizeof(A::v::value_type)             // illegal 

Error message...

example.cpp:18:44: error: no member named 'v' in 'A'

Why does the third version fail? I don't understand how adding another layer of scope resolution causes the more broad scope resolution to fail. In fact, the second example seems equivalent to the failing example, but obviously the compiler disagrees.

This is a what-if/language question. I am not trying to solve a particular problem with vector sizes but rahter understand this oddity of the language.

Update

Here is another example from an answer below that seems to further my confusion...

sizeof(decltype(A::v)::value_type);  // legal  

How is decltype(A::v)::value_type different from A::v::value_type in this context?

Answer 1

I don't think you can get a typename from a variable using the :: token. Ignore sizeof and ignore A::v. The following doesn't work:

A a; // A is some type with a public typedef for value_type in its declaration
A::value_type b; // ok
a::value_type c; // error

GCC 4.7.1 gives me this error:

x.cpp:9:1: error: 'a' does not name a type

sizeof makes this confusing only because it is a compile time construct that accepts both types and variables as its parameter.

Answer 2

Correct me if I'm wrong but doesn't the last one fail because you are accessing an instance in the last one whereas in the second one you get the class name from the decltype(). If you wrap the third one in a decltype it works

sizeof(decltype(A::v)::value_type);