G ++和__attribute __（（优化））不改变调试器行为

Question

I am trying to play with __attribute__ to allow a function to be essentially compiled with different flags from the rest of the code. For example:

#include <iostream>
#include <vector>

void MyNormalFunction();

void MyDebugabbleFunction() __attribute__((optimize(0)));

void MyNormalFunction()
{
  std::cout << "Test" << std::endl;

  std::vector<int> a;

  for(unsigned int i = 0; i < 10; ++i)
  {
    a.push_back(i);
  }
}

void MyDebugabbleFunction()
{
  std::cout << "Test" << std::endl;

  std::vector<int> a;

  for(unsigned int i = 0; i < 10; ++i)
  {
    a.push_back(i);
  }
}

int main()
{
  MyNormalFunction();
  MyDebugabbleFunction();
  return 0;
}

I am building with -g -O2, but I want to be able to sanely debug MyDebugabbleFunction() — so I used the __attribute__((optimize(0))) on its declaration. However, I can't really tell any difference when stepping through these two functions with a debugger. I would expect the "seemingly erratic" behavior that I usually see when trying to step through optimized code in MyNormalFunction , but the standard "-g"-only debugger behavior in MyDebuggableFunction .

Is it that I have done something wrong with __attribute__ ? Or that I have used bad demo code (ie code that doesn't get "optimized a lot") inside the two functions? Or am I misinterpreting what the difference is supposed to be in the debugger?

I am using gcc 4.6.

---

EDIT based on GManNickG's suggestion

I used this code instead, and built with -O2 -g:

#include <iostream>
#include <vector>

int MyNormalFunction();

int MyDebugabbleFunction() __attribute__((optimize(0)));

int MyNormalFunction()
{
  int val = 0; // breakpoint here - debugger does NOT stop here
  val = 1;
  val = 2;
  return val;
} // debugger stops here instead

int MyDebugabbleFunction()
{
  int val = 0;  // breakpoint here - debugger stops here and steps through the next 3 lines as if it were built with only -g
  val = 1;
  val = 2;
  return val;
}

int main()
{
  int a = MyNormalFunction();
  std::cout << a << std::endl;

  int b = MyDebugabbleFunction();
  std::cout << b << std::endl;

  return 0;
}

Answer 1

Try a test like this instead:

int MyNormalFunction()
{
    int val = 0;
    val = 1;
    val = 2;

    // should optimize to return 2
    return val;
}

int MyDebuggableFunction() __attribute__((optimize(0)));
{
    int val = 0;
    val = 1;
    val = 2;

    // could optimize to return 2, but attribute blocks that
    return val;
}

int main()
{
    // we need to actually output the return values,
    // or main itself could be optimized to nothing
    std::cout << MyNormalFunction() << std::endl;
    std::cout << MyDebuggableFunction() << std::endl;
}

It'll make it much easier to follow.

---

Note that you should start in main , when stepping through, because it most likely will be reduced to:

int main()
{
    std::cout << 2 << std::endl;
    std::cout << MyDebuggableFunction() << std::endl;
}

If you're up for it, looking at the disassembly makes this task much easier.

Answer 2

After fixing the errors in the code, so it compiles:

g++ -S x.c

_Z16MyNormalFunctionv:
.LFB1255:
    .cfi_startproc
    movl    $2, %eax
    ret

_Z20MyDebuggableFunctionv:
.LFB1256:
    .cfi_startproc
    movl    $0, -4(%rsp)
    movl    $1, -4(%rsp)
    movl    $2, -4(%rsp)
    movl    -4(%rsp), %eax
    ret

As you can see the optimization attribute worked fine.