What is structure alignment in C?

Answer 1

Structure alignment relates to the way objects in memory must be allocated according to the underlying architecture. For instance, a machine might expect an int to be aligned upon a 4-byte boundary. If that is the case then a structure with a member int must be aligned upon a 4-byte boundary and the int member offset from that boundary by some multiple of 4 bytes.

Consider the following:

struct X_ { char c; int i;

} X;

assert (sizeof(X)==sizeof(int)+sizeof(char)); /* oops */

Assuming sizeof(int) is 4 and sizeof(char) is 1 and that an int must be aligned on a 4-byte boundary, the actual length of this structure must be 8 bytes, not the 5 we expected. If it were anything other than 8, then an array of such structures would have misalignments in 3 out of every 4 elements. Note that in order to correctly align this structure, the compiler must insert 3 padding bytes immediately after the char member. This ensures that the int member is offset 4 bytes from the start of the structure. So long as the structure itself is aligned upon a 4-byte boundary (which it now will be because 8 is a multiple of 4), the int will always be correctly aligned.

Inserting padding bytes between members can often increase a structure's size far more than we might expect, particularly if there are many members. To avoid this, it is best to declare members in order of size, largest first. In this way, whenever padding is required, the padding is more likely to be placed at the end of the structure rather than between two or more members.

struct X_ {

char c;

int i;

char c2;

} X;

Given the same criteria as before, this structure would be 12 bytes in length. But by changing the order of the members, we can reduce this to just 8 bytes:

struct X_ {

int i;

char c;

char c2;

} X;

The reason this works is because a char can always be aligned on a 1-byte boundary, so two consecutive char members will easily fit inside a 4-byte word. Thus we only require 2 bytes of padding instead of 4.

Note that it is never safe to assume the lengths of any data type in C other than char which is always sizeof(char)==1. The lengths of all other types are measured in terms of a char, but will vary from system to system according to the underlying architecture. Always use the sizeof() operator to determine type lengths. Note that the sizeof() operator computes data type lengths at compile time, so there is no runtime cost, nor is there any need to store the return value.