Search code examples
c++token-name-resolution

Declaration of method changes meaning of symbol


For the following code:

struct foo {};

struct A
{
    typedef foo foo_type;

    void foo();
};

GCC gives a compiler error:

test.cpp:7:14: error: declaration of 'void A::foo()' [-fpermissive]
     void foo();
              ^
test.cpp:1:8: error: changes meaning of 'foo' from 'struct foo' [-fpermissive]
 struct foo {};
        ^

But clang accepts it without compiler errors. Who is right?

Note that if the typedef is removed, or changed to typedef ::foo foo_type, both gcc and clang accept the code.


Solution

  • gcc is correct, but clang is not required to give a diagnostic (3.3.7):

    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.

    This is because of how class scope works. The foo of void foo(); is visible within the entire scope of the class A, so the declaration of void foo(); changes the meaning of foo in the typedef from referring to struct foo to the name of the function foo.