40

I am trying to print the address of a virtual member function. If I know which class implements the function I can write:

print("address: %p", &A::func);

But I want to do something like this:

A *b = new B();

printf("address: %p", &b->func); 
printf("address: %p", &b->A::func);

However this does not compile. Is it possible to do something like this, perhaps looking up the address in the vtable at runtime?

Peter Cordes
  • 328,167
  • 45
  • 605
  • 847
hidayat
  • 9,493
  • 13
  • 51
  • 66

5 Answers5

20

Currently there is no standard way of doing this in C++ although the information must be available somewhere. Otherwise, how could the program call the function? However, GCC provides an extension that allows us to retrieve the address of a virtual function:

void (A::*mfp)() = &A::func;
printf("address: %p", (void*)(b->*mfp));

...assuming the member function has the prototype void func(). This can be pretty useful when you want to cache the address of a virtual function or use it in generated code. GCC will warn you about this construct unless you specify -Wno-pmf-conversions. It's unlikely that it works with any other compiler.

Boann
  • 48,794
  • 16
  • 117
  • 146
Lukas W
  • 216
  • 2
  • 2
6

Pointers to member functions are not always simple memory addresses. See the table in this article showing the sizes of member function pointers on different compilers - some go up to 20 bytes.

As the article outlines a member function pointer is actually a blob of implementation-defined data to help resolve a call through the pointer. You can store and call them OK, but if you want to print them, what do you print? Best to treat it as a sequence of bytes and get its length via sizeof.

AshleysBrain
  • 22,335
  • 15
  • 88
  • 124
2

I found a way to do this using a disassembler (https://github.com/vmt/udis86). The steps are:

  1. Get a pointer to the virtual function via normal C++ code

  2. Disassemble the jmp instruction at that address

  3. Parse the real address from the disassembled string

Here is how I did it:

// First get the raw pointer to the virtual function
auto myVirtualFuncPtr = &MyClass::myFunc;
void* myVirtualFuncPtrRaw = (void*&)myVirtualFuncPtr;

// Resolve the real function!
void* myFuncPtr = resolveVirtualFunctionAddress(myVirtualFuncPtrRaw);

...

static void* resolveVirtualFunctionAddress(void* address)
{
    const int jumpInstructionSize = 5;

    static ud_t ud_obj;
    ud_init(&ud_obj);
    ud_set_mode(&ud_obj, sizeof(void*) * 8);
    ud_set_syntax(&ud_obj, UD_SYN_INTEL);
    ud_set_pc(&ud_obj, (uint64_t)address);
    ud_set_input_buffer(&ud_obj, (unsigned uint8_t*)address, jumpInstructionSize);

    std::string jmpInstruction = "";

    if (ud_disassemble(&ud_obj))
    {
        jmpInstruction += ud_insn_asm(&ud_obj);
    }

    // TODO: Implement startsWith and leftTrim yourself
    if (startsWith(jmpInstruction, "jmp "))
    {
        std::string jumpAddressStr = leftTrim(jmpInstruction, "jmp ");
        return hexToPointer(jumpAddressStr);
    }

    // If the jmp instruction was not found, then we just return the original address
    return address;
}

static void* hexToPointer(std::string hexString)
{
    void* address;
    std::stringstream ss;

    ss << std::hex << hexString;
    ss >> address;

    return address;
}
Zachary Canann
  • 1,131
  • 2
  • 13
  • 23
1

From what I can tell in the standard, the only time you get dynamic binding is during a virtual function call. And once you've called a function, you're executing the statements within the function (i.e., you can't "stop halfway" into the call and get the address.)

I think it's impossible.

GManNickG
  • 494,350
  • 52
  • 494
  • 543
  • I suppose it's impossible portably, but knowing the implementation it should be possible to inspect the virtual table of an object at runtime. – Matthieu M. Jun 18 '10 at 13:03
0

Doesn't make a lot a of sense to me. If you have a normal function:

void f( int n ) {
}

then you can take its address:

f

but you cannot take the address of a function call, which is what you seem to want to do.

  • @GMan That's what I thought I said. Anyhow, I don't think this is possible. –  Jun 18 '10 at 08:44