Dynamic memory allocation in C functions called from R language

Question

I'm writing an R package that calls a C script. The C script uses a structure with a dynamically determined length: the length of the array forest->edges depends on the data passed from R.

typedef struct {
  unsigned int n_edge;
  ...
  unsigned int max_node;
  unsigned int edges[];
} forest;

forest * forest_new (unsigned int *n_edge) {
  forest *f = malloc(sizeof(forest) + (2 * *n_edge * sizeof(int)));
  f->n_edge     = *n_edge;
  ...
  f->max_node   = 0;
  return f;
}

The code runs successfully in C, but crashes when an R call triggers forest_new . My hunch is that the crash results from memory allocation, and indeed the R manual mentions alternative means of allocating memory (eg R_alloc , Calloc ), which threads elsewhere seems to suggest should be used in place of malloc / calloc .

So part 1 of the question is when calls to malloc / calloc should, or must, be replaced by R-safe equivalents (perhaps they are irrelevant to my problem?). Part 2 of the question is how the R-safe functions can handle structures whose length is dynamically determined.

Answer 1

You can use malloc/calloc and free in packages to allocate and free memory, but you have to handle out of memory errors like in any C application. You can use Calloc/Free/Realloc provided by R and then you will get the error handled in the "R way" (an R error when out of memory). The R_alloc function allows you to allocate temporary data that is freed automatically when your external function exits (when you get back to R, it is stack-based allocation).

R does not care about whether your C code uses structures with dynamically determined length (flexible arrays), R does not access your structure at all.

If you need more help please post a self-contained example or provide more information about the error. Perhaps it is also worth checking n_edge has a correct/sane value in forest_new .