Question

Is the following code valid?

struct B{ using X=int; };
struct D1:B{ using X=X; };   // (1)
struct D2:B{ typedef X X; }; // (2)

I would expect the point of declaration of D2::X to be between the two Xs in (2), but it seems both gcc 4.8 and clang 3.2 accept it.
Is this standard behavior ? References to the working draft/standard will be appreciated.

Answer 1

There was a debate on whether using X = X should pick up the being defined X or the X that potentially already was in scope. To avoid “unknown types” and to make it similar to typedef, it was ruled that the being-defined X is not visible in its to-be-assigned type expression (so rather than being similar to int x = x, it is similar to typedef x x;).

Recall that typedef is just a normal declaration with the typedef keyword prepended. The first mentioning of X does not declare anything, it just says what type will be aliased. That’s the major difference with using X = X which could declare X earlier, if the committee decided that way.

Note however your code is has effectively undefined behavior, because it violates a rule that has no required diagnostic. 3.3.7p1b2

A name N used in a class S shall refer to the same declaration in its context and when re-evaluated in the completed scope of S. No diagnostic is required for a violation of this rule.