What's a need for union { } in C++? [duplicate]

This question already has answers here: Closed 11 years ago.

Possible Duplicate:

C/C++: When would anyone use a union? Is it basically a remnant from the C only days?

Hi all.

What are the reasons for unions to exist in C++? And what are they, actually?

I found this code:

#include <iostream>开发者_如何学C
using namespace std;
union mixture {
    short number;
    char symbol[2];

main() {
    mixture m1, m2;
    cout << "Enter 2 symbols to form short number made of them: ";
    cin >> m1.symbol[0] >> m1.symbol[1];
    cout << "2 more to go..: ";
    cin >> m2.symbol[0] >> m2.symbol[1];
    cout.setf(ios::hex);
    cout << "Biggest received number: " <<  (m1.number > m2.number ? m1.number : m2.number)  << endl;
    system("pause");
    return 0;
}
    return 0;

But actually, what I win from using union { struct1, struct2 } instead of writing struct { struct2(struct1}: a(struct1.a), b(_struct1.b {}} struct2? to transparently support both types?

I do some embedding stuff (Arduino etc), but never seen the real usage for structs.

Examples, please.

---

The union lets you treat your data as either char or short without having to cast. Casting pointer between types can produce type-punning errors in the optimizer and generate incorrect output.

EDIT: For an example, I use unions to byte swap doubles and floats after reading them from a file. If you read byteswapped floating point numbers the numbers might get normalized or adjusted while bytes swapped, which of course results in junk once they're swapped:

union float_int
{
  float ff;
  int32_t ii;
};

float_int data;
read(fd, &data, 4);       // Read 4 bytes
byteswap(data.ii);        // Passing as a float can alter the bits.
float myvalue = data.ff;  // My float is now byteswaped and ready to go.

---

A union is fundamentally different from a struct in that only one of the members is guaranteed to be usable at a time -- but this is a white lie, see next. (The data may or may not overlap depending upon the compiler, target, and various packing rules and this overlap can be [ab]used in cases The overlap can be [ab]used depending upon compiler, target, types, etc).

Conceptually a union represents a "discrete value" (X has/is A or B but X is not/does not have A and B) and it also allows the compiler (or whatever uses this model) to represent the data more efficiently in cases.

Happy coding.

---

In hardware level code you have to be sure your bits are in consecutive memory, this is one way to make sure of it.

edit: You dont HAVE to be sure, but there are plenty of cases it does.

---

Firstly, they're inherited from C and there was no good reason to remove support, so C++ retained support.

It's one of the only widely-supported ways to do type-punning (but still likely illegal by the letter of the standard).

It can save space in certain cases.