Debugging C++ Release Projects - Finding the Lost Object Information

Debugging C++ Release Projects - Finding the Lost Object Information

By  Yochai Timmer15 Feb 2014
 

Introduction

When compiling a C++ project in release with optimizations, the compiler sometimes doesn't show the right object information.

The local variables are usually the first to go, and many times the this object's information are lost to the debugger.

The reason is that the compiler uses the available hardware registers to hold the information, and uses optimizations to avoid allocation of local variables.

In this tip, I'll show how to use Visual Studio's tools to see where the information is held.

Using the Code

I'll give a simple example of some code that throws an exception,

I'll create a simple class and a situation where one of the member's value will cause a division by zero exception.

I'll add an extra method just so the debugger will stop at the right line (due to optimizations, the debugger may not stop at the right line, because the actual code stack doesn't represent where you are in the code - just the return address. This issue may be explained in a future post.).

The extra method is important for this scenario, just so we can imitate the scenario I'm talking about.

  class SomeClass
    {
    public:
        int a,b;
        void MyMethod()
        {
            int i = 0;
            int j = i;
            DoStuff();
            a = this->a/this->b;
            DoStuff();
            DoStuff();
        }
 
        void DoStuff()
        {
            int i = 10;
            int j = 5;
            int k = 4;
            for ( ;i * j + k < 40000; i++)
            {
                int fs = 34;
            }
            printf("%d %d %d %d %d", i , j , k , 10,8);
        }
    };
int main(int argc, void* argv[])
{
    SomeClass a;
    a.a = 5;
    a.b = 0;
 
    a.MyMethod();
    return 0;
}

Now, if you run this little program in release mode, you'll get an exception complaining about a division by zero. This is because the member b is set to 0.

The debugger will conveniently stop at the right line, but unfortunately all the local variables and the this variable will be corrupted.

The information is obviously somewhere, but the debugger just doesn't seem to load the right information. If you look closely, then you can see that the this value is 01 which indicates that it's just reading the object's information from the wrong place/address.

So, how can we find the object's right address ?

Using the Call Stack

The first method, which is the easiest, is to go up the call stack and hope that at some level the debugger sees the right address. This is by far easier than the next method...

If we look at the call stack, we can see that the debugger correctly resolved that we are in the method MyMethod of the SomeClass class:

So, if we just go to the previous function in the stack (main), the debugger might still hold the right address to the object (because the object was known and used there). This will work as long as the compiler didn't decide to re-use the registers that held the object's address.

So, if we double click the line just below our current position, the debugger will jump to the code where the current method was called:

Here, we can clearly see that b equals 0, and we are dividing by b, so mystery solved - that's why we get an exception!

But... In deeper nested call stacks, with more complicated code, this doesn't always work, and the debugger may use the information throughout the whole call stack hierarchy.

Registry Values and Disassembly

The key is to understand that the CPU holds its information in its registers. Although the actual information may be stored in memory (stack or heap), the CPU itself only uses the registers.

The compiler adds assembly code that copies the relevant information from the memory to the registers for the CPU to actually do the calculations.

Knowing that, we can assume that most of the locally used variables and this address will be held in the registers (unless there is no farther use for them, in which case the register may be re-used to hold other information).

To know how exactly the variables are mapped to registers, we can use the Disassembly window and Registerswindow.

I'll not really go into assembly code, I'll just look at the register names.

Let's look at the Dissasembly window of our problematic code. We want to find a place where the code calls a local variable of the current object we want, the mapped assembly code can point out the register it's using:

We can see that when we call the local members, the matching assembly code calls a register called rdi.

This register probably holds the address of our object (object members are accessed by an offset from the object's starting point (address) in memory).

We will now want to see what value the register holds. We hope that it's an address (a large number). If we open the Registers window, we can see its value:

We can see that rdi holds the value 000000000025F950 it's a hexadecimal value that is pretty big, and this is what a memory address usually looks like. (Lower values usually represent counters or other local function members.)

Of course, this means absolutely nothing to me, or almost anyone else, so let's let the debugger help us figure it out.

We can now open the Watch window. It's a useful window which among other things allows you to enter specific addresses and show their value.

We'll cast the address to a pointer of our object's type, and add a 0x prefix to the number so the debugger knows it's a hexadecimal value:

(SomeClass*)(0x000000000025F950)  

The debugger will now show us the object like we're used to seeing it:

Points of Interest

  1. Call Stack: Debug -> Windows -> Call Stack (alt+7)
  2. Locals: Debug -> Windows -> Locals (alt+4)
  3. Disassembly window: Debug -> Windows -> Disassembly (alt+8)
  4. Registers window: Debug -> Windows -> Registers (alt+5)
  5. Watch window: Debug -> Windows -> Watch -> Watch 1 (ctrl + alt + w, 1)

History

  • 15th February, 2014: Initial version

License

This article, along with any associated source code and files, is licensed under The Code Project Open License (CPOL)

  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
目标检测(Object Detection)是计算机视觉领域的一个核心问题,其主要任务是找出图像中所有感兴趣的目标(物体),并确定它们的类别和位置。以下是对目标检测的详细阐述: 一、基本概念 目标检测的任务是解决“在哪里?是什么?”的问题,即定位出图像中目标的位置并识别出目标的类别。由于各类物体具有不同的外观、形状和姿态,加上成像时光照、遮挡等因素的干扰,目标检测一直是计算机视觉领域最具挑战性的任务之一。 二、核心问题 目标检测涉及以下几个核心问题: 分类问题:判断图像中的目标属于哪个类别。 定位问题:确定目标在图像中的具体位置。 大小问题:目标可能具有不同的大小。 形状问题:目标可能具有不同的形状。 三、算法分类 基于深度学习的目标检测算法主要分为两大类: Two-stage算法:先进行区域生成(Region Proposal),生成有可能包含待检物体的预选框(Region Proposal),再通过卷积神经网络进行样本分类。常见的Two-stage算法包括R-CNN、Fast R-CNN、Faster R-CNN等。 One-stage算法:不用生成区域提议,直接在网络中提取特征来预测物体分类和位置。常见的One-stage算法包括YOLO系列(YOLOv1、YOLOv2、YOLOv3、YOLOv4、YOLOv5等)、SSD和RetinaNet等。 四、算法原理 以YOLO系列为例,YOLO将目标检测视为回归问题,将输入图像一次性划分为多个区域,直接在输出层预测边界框和类别概率。YOLO采用卷积网络来提取特征,使用全连接层来得到预测值。其网络结构通常包含多个卷积层和全连接层,通过卷积层提取图像特征,通过全连接层输出预测结果。 五、应用领域 目标检测技术已经广泛应用于各个领域,为人们的生活带来了极大的便利。以下是一些主要的应用领域: 安全监控:在商场、银行
目标检测(Object Detection)是计算机视觉领域的一个核心问题,其主要任务是找出图像中所有感兴趣的目标(物体),并确定它们的类别和位置。以下是对目标检测的详细阐述: 一、基本概念 目标检测的任务是解决“在哪里?是什么?”的问题,即定位出图像中目标的位置并识别出目标的类别。由于各类物体具有不同的外观、形状和姿态,加上成像时光照、遮挡等因素的干扰,目标检测一直是计算机视觉领域最具挑战性的任务之一。 二、核心问题 目标检测涉及以下几个核心问题: 分类问题:判断图像中的目标属于哪个类别。 定位问题:确定目标在图像中的具体位置。 大小问题:目标可能具有不同的大小。 形状问题:目标可能具有不同的形状。 三、算法分类 基于深度学习的目标检测算法主要分为两大类: Two-stage算法:先进行区域生成(Region Proposal),生成有可能包含待检物体的预选框(Region Proposal),再通过卷积神经网络进行样本分类。常见的Two-stage算法包括R-CNN、Fast R-CNN、Faster R-CNN等。 One-stage算法:不用生成区域提议,直接在网络中提取特征来预测物体分类和位置。常见的One-stage算法包括YOLO系列(YOLOv1、YOLOv2、YOLOv3、YOLOv4、YOLOv5等)、SSD和RetinaNet等。 四、算法原理 以YOLO系列为例,YOLO将目标检测视为回归问题,将输入图像一次性划分为多个区域,直接在输出层预测边界框和类别概率。YOLO采用卷积网络来提取特征,使用全连接层来得到预测值。其网络结构通常包含多个卷积层和全连接层,通过卷积层提取图像特征,通过全连接层输出预测结果。 五、应用领域 目标检测技术已经广泛应用于各个领域,为人们的生活带来了极大的便利。以下是一些主要的应用领域: 安全监控:在商场、银行
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值