#include
union u1 {
struct {
int *i;
} s1;
struct {
int i, j;
} s2;
};
union u2 {
struct {
int *i, j;
That's because the compiler needs to keep the entire struct (as well as the union) aligned to 8 bytes - due to the fact that you have a pointer inside. (which is 8-bytes in your case)
So even though you add only 4 bytes with the extra int
, struct-alignment forces everything to be aligned to 8 bytes - hence the +8 to bring the total size to 16 bytes.
The result of this is that:
struct {
int *i, j;
} s1;
has a size of 16 bytes. Since a union must be at least as large as the largest element, it is also forced up to 16.
http://en.wikipedia.org/wiki/Data_structure_alignment
Since a pointer is 8 bytes on your platform, it most probably also requires an alignment of 8 bytes. So when adding another 4 bytes, the struct cannot just have a size of 12 bytes, in which case the individual elements of say an array of u2
s would not be properly aligned at 8 byte boundaries, which is neccessary for the pointer member. So you need to increase its size to the next multiple of 8, which is 16. The additional 4 bytes are just unused/undefined.
The struct u2.s2
is 16 bytes because of alignment constraints. The compiler is guaranteeing that if you make an array of such structs, each pointer will be aligned on an 8-byte boundary. The field *i
takes 8 bytes, then j
takes 4 bytes, and the compiler inserts 4 bytes of padding. Because the struct is 16 bytes, the union containing it is also 16 bytes.