I've been doing an experiment to see if I can design a simple managed
wrapper around an unmanaged class, but faced a small problem that may be
a compiler bug.

I created an unmanaged DLL first:

// unmanaged.h
#ifdef UNMANAGED_EXPORTS
#define UNMANAGED_IMPORT __declspec(dllexport)
#else
#define UNMANAGED_IMPORT __declspec(dllimport)
#endif
#ifdef _MANAGED
public
#endif
class UNMANAGED_IMPORT UnClass
{
public:
   int Square(int x) { return x * x; }
};

Then I created a managed assembly wrapper:

//managed.h
#pragma once
#include "unmanaged.h"
public ref class Class1
{
public:
   Class1() : p(new UnClass) { }
   ~Class1() { delete p;}
   UnClass* get() { return p; }
private:
   UnClass* p;
};

And finally a managed test application:

// test.cpp
#include "unmanaged.h"
#using "managed.dll"
using namespace System;
int main(array<System::String^>^ args)
{
   Class1 c;
   UnClass* p = c.get();
   p->Square(10); // this works
   // c.get()->Square(10); // ERROR!!! why?
   return 0;
}

I used the
#ifdef _MANAGED
public
#endif
trick to make my unmanaged class public, to avoid the "C3767:  candidate
function not available" error, which the original poster had. It works
nicely.

However, there's one glitch. The following doesn't work in test.cpp:

c.get()->Square(10);

I get a compiler error C2039: 'Square': is not a member of 'UnClass'. I
don't understand why, it obviously is a member. I'm forced to assign the
output of the c.get() call to temporary variable 'p', or explicitly
cast, like this:

static_cast<UnClass*>(c.get())->Square(10);

Since c.get() returns UnClass*, the cast is not necessary, and I
shouldn't be getting a compiler error. Is this expected behavior, or is
it a bug in Beta 2?

Tom