private typedef as return type of private function

[Graham]

For information, really (I tried Comeau, and it doesn't complain), here's another VC++ non-standardism:

// A.h
#include <vector>

class A
{
    typedef std::vector<int>::iterator iterator;

public:
    A() : v(10, 1) {}

private:
    iterator f();
    std::vector<int> v;
};

// A.cpp
#include "A.h"

A::iterator A::f()
{
    return v.begin();
}

VC++6.0 rejects this, saying that you can't access the private typedef A::iterator. Basically it seems to be failing to get the context of the (private) function before deciding whether private access is allowed. As I said, Comeau has no problem with it.

[jwbarton]

I tried this example in my copy of Visual C++ 6.0 (SP5) and it compiled without any complaints. Are you missing something from this example?

Best regards,
John

[AnthonyMai]

It compiles with no problem on my machine, too. My VC++ don't even have SP5 installed.

Microsoft bashing has become a prominent part of the developer culture. Whenver something doesn't work people jump right onto Microsoft and assume M$ to be guilty until proven innocent.

[Graham]

John: possibly the SP5 part. Or it may be that the actual example (which involved nested classes and multiple files) is a bit more complex. I will admit to not trying that simple example - what I had was more along the lines of:

// A.h

class A
{
// blah, blah
private:
    class B;
    B* mem;
};

// AB.h

#include "A.h"

class A::B
{
     typedef /*whatever*/ type
//blah
private:
     type f();
};

//AB.cpp

A::B::type A::B::f()
{
}

Still, it may be the fact that my company has kept us at a lower SP that's the problem.

[jwbarton]

I tried the following test, and get the error C2248 that you described:

class A
{
public:
   A()
      {
      }

private:
   class B;
   B *pB;
};


class A::B
   {
   typedef B* ptrB;

   public:
      ptrB GetOne();
   };



A::B::ptrB A::B::GetOne()
   {
   return new B;
   }

Does this compile with the Comeau compiler? I am not sure whether this should compile, as I don't know what the rules are for typedefs being defined inside of a class.


Best regards,
John

[Graham]

It looks like I may have unjustly maligned VC++. Comeau throws the nested class version out, too. I had assumed that nesting in this case wouldn't (or shouldn't) make a difference - if it works one level deep, why not two levels? Mea culpa - just goes to show where assumptions, however apparently reasonable, get you.

For the record, this is what I threw at it:

class A
{
public:
    A() {}
private:
    class B;
    B* p;
};

class A::B
{
    typedef B* pB;

    pB f();
};

A::B::pB A::B::f()
{
    return new B;
}

and the message reported is:

"ComeauTest.c", line 17: error: class "A::B" is inaccessible
  A::B::pB A::B::f()
     ^

Which is not quite the same error, but in the same ball park

Anybody here able to work out what the standard says about this, and why there should be a difference in behaviour between outer-level and nested classes.

It all seems a bit odd, because you have to revert to the type that the typedef is an alias for in order to define the function.