Memory position of elements in C/C++ union

Question

I have a union in C like this:

union AUnion {
  struct CharBuf {
    char *buf;
    size_t len;
  } charbuf;
  uint8_t num;
  double fp_num;
};

My question is, can I guarantee that if given the following:

union AUnion u;

Then the following are true:

&u == &u.num
&u == &u.fp_num
&u == &u.charbuf

I.e they all start at the beginning of the memory segment where u is stored.

In the case of this C program compiled with gcc version 5.3.0 and -std=c11 the above is true:

#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>

union AUnion {
    struct CharBuf {
        char *buf;
        size_t len;
    } charbuf;
    uint8_t num;
    double fp_num;
};

int main(void)
{
    union AUnion u;
    printf("%d\n", ((void*)&u) == ((void*)&u.charbuf));
    printf("%d\n", ((void*)&u.charbuf) == ((void*)&u.num));
    printf("%d\n", ((void*)&u.num) == ((void*)&u.fp_num));
}

As it prints:

1
1
1

Compiling the code above as C++11 with the same compiler results in the same output as compiling it as C11.

But is this standardized behaviour? Is it undefined? Can I rely on this behaviour with most C compilers? Can I expect this behaviour with C++ compilers as well?

Answer 1

In 6.7.2.1p16 the C standard guarantees that:

The size of a union is sufficient to contain the largest of its members. The value of at most one of the members can be stored in a union object at any time. A pointer to a union object, suitably converted, points to each of its members (or if a member is a bit- field, then to the unit in which it resides), and vice versa.

So, yes, you can rely on all members starting at the unions address (note this is the same for the first member of a struct).

The C++ standard includes a similar sentence with respect to C-style (i.e. only C-style members) unions/structs, because C++ allows to pass unions to C functions which does require this layout.The relevant section in the C++ standard is 9.5.

---

However, note there might be padding bits inside standard simple types (integers, floats). And their internal may vary (endianess). You also might violate strict aliasing rule (C: effective type).