How to initialize a structure with flexible array member

Sangeeth Saravanaraj picture Sangeeth Saravanaraj · Dec 31, 2011 · Viewed 27.6k times · Source

I have the following structure

typedef struct _person {
    int age;
    char sex;
    char name[];
}person;

I have done some basic internet search (but unsuccessful) on how to create an instance and initialize a structure with a flexible array member without using malloc().

For example: for normal structures like

struct a {
    int age; 
    int sex;
};

We can create an instance of struct a and initialize it like

struct a p1 = {10, 'm'};

But for structures with flexible array in it (like _person as mentioned above) how can we create an instance and initialize like how we do it for normal structures?

Is it even possible? If so, how do we pass the array size during the initialization and the actual value to be initialized?

(or)

Is it true that the only way to create a structure with flexible array is using malloc() as mentioned in C99 specification - 6.7.2.1 Structure and union specifiers - point #17?!

Answer

Jens Gustedt picture Jens Gustedt · Dec 31, 2011

No, flexible arrays must always be allocated manually. But you may use calloc to initialize the flexible part and a compound literal to initialize the fixed part. I'd wrap that in an allocation inline function like this:

typedef struct person {
  unsigned age;
  char sex;
  size_t size;
  char name[];
} person;

inline
person* alloc_person(int a, char s, size_t n) {
  person * ret = calloc(sizeof(person) + n, 1);
  if (ret) memcpy(ret,
                  &(person const){ .age = a, .sex = s, .size = n},
                  sizeof(person));
  return ret;
}

Observe that this leaves the check if the allocation succeeded to the caller.

If you don't need a size field as I included it here, a macro would even suffice. Only that it would be not possible to check the return of calloc before doing the memcpy. Under all systems that I programmed so far this will abort relatively nicely. Generally I think that return of malloc is of minor importance, but opinions vary largely on that subject.

This could perhaps (in that special case) give more opportunities to the optimizer to integrate the code in the surroundings:

#define ALLOC_PERSON(A,  S,  N)                                 \
((person*)memcpy(calloc(sizeof(person) + (N), 1),               \
                 &(person const){ .age = (A), .sex = (S) },     \
                 sizeof(person)))

Edit: The case that this could be better than the function is when A and S are compile time constants. In that case the compound literal, since it is const qualified, could be allocated statically and its initialization could be done at compile time. In addition, if several allocations with the same values would appear in the code the compiler would be allowed to realize only one single copy of that compound literal.