I'm new to C language. The following code snippet is from a book I've been reading.

我是C语言的新手。下面的代码片段来自我一直在读的一本书。

struct S {
  int i; double d; char c;
};

int main(void) {
  unsigned char bad_buff[sizeof(struct S)];
  _Alignas(struct S) unsigned char good_buff[sizeof(struct S)];

  struct S *bad_s_ptr = (struct S *)bad_buff;   // wrong pointer alignment
  struct S *good_s_ptr = (struct S *)good_buff; // correct pointer alignment
}

The author states that bad_buff may have incorrect alignment for member-access expressions. I have difficulty in understanding this statement. What can go wrong if we use bad_buff ?

作者指出，bad_buff可能对成员访问表达式进行了不正确的对齐。我很难理解这句话。如果我们使用bad_buff，会出现什么问题？

The correct answer is: you don't. And if you are a C beginner, you are out on deep waters since this is a rather advanced topic. My rule of thumb is that a beginner should never use the cast operator for any purpose, since there are so many pitfalls when doing so.

正确的答案是：你没有。如果你是一个C级的初学者，你就陷入了困境，因为这是一个相当高级的话题。我的经验法则是，初学者永远不应该出于任何目的使用cast运算符，因为这样做会有很多陷阱。

There are two problems going from a raw character byte array to a struct type:

从原始字符字节数组到结构类型有两个问题：

• First there is alignment which the code addresses somewhat. But _Alignas etc only ensures that the buffer and the struct have the same alignment, ie start address. It does nothing for alignment of individual struct members and any padding the struct may contain internally. Such things are non-portable.

  首先是代码在一定程度上解决的对齐问题。但是_Alignas等只能确保缓冲区和结构具有相同的对齐方式，即起始地址。它对单个结构成员的对齐以及结构内部可能包含的任何填充都没有任何作用。这样的东西是不可携带的。

  
  



• Second, these kind of dirty casts are undefined behavior in the C language, meaning anything can happen. Including incorrect code getting generated, hardware exceptions getting thrown etc. Or it might just seem to work fine for now/this time/forever/until next full moon. What is the strict aliasing rule?

  其次，这种脏类型的强制转换在C语言中是未定义的行为，这意味着任何事情都可能发生。包括生成不正确的代码，抛出硬件异常等。或者它可能只是现在/这次/永远/直到下一次满月都很好。什么是严格的混叠规则？

  
  

  (As it happens, you can go from a struct to a character byte array safely, but not the other way around.)

  （碰巧的是，您可以安全地从结构转到字符字节数组，但不能反过来。）

  
  

There's no quick fix to address both of these issues. If you are aware that a struct has padding bytes and where they are, then you could do something like:

没有快速解决这两个问题的方法。如果您知道一个结构有填充字节以及它们在哪里，那么您可以执行以下操作：

typedef union
{
  some_struct s;
  unsigned char buf [sizeof(some_struct)];
} some_union;

This solves the strict aliasing problems and initial alignment, but it doesn't solve the issue of individual struct members potentially having padding bytes. For that you'd need some non-standard means like #pragma pack(1). Which in turn is problematic since padding is there for a reason and if you pack the struct you might not be able to access certain members as intended.

这解决了严格的混叠问题和初始对齐，但并不能解决单个结构成员可能具有填充字节的问题。为此，您需要一些非标准的方法，如#pragma pack（1）。这反过来又是有问题的，因为填充是有原因的，如果你打包了结构，你可能无法按预期访问某些成员。

Structs are overall non-portable and unsuitable for things like close to the metal constructs like data communication protocols or register maps. The only truly portable way is to write a serialization/deserialization function which copies in/out of individual struct members one by one.

结构总体上是不可移植的，不适合于数据通信协议或寄存器映射等接近金属的结构。唯一真正可移植的方法是编写一个序列化/反序列化函数，逐个复制单个结构成员的入/出。