Inititialization by one thread

suggest change

In most cases all data that is accessed by several threads should be initialized before the threads are created. This ensures that all threads start with a clear state and no race condition occurs.

If this is not possible once_flag and call_once can be used

#include <threads.h>
#include <stdlib.h>

// the user data for this example
double const* Big = 0;

// the flag to protect big, must be global and/or static
static once_flag onceBig = ONCE_INIT;

void destroyBig(void) {
   free((void*)Big);
}

void initBig(void) {
    // assign to temporary with no const qualification
    double* b = malloc(largeNum);
    if (!b) {
       perror("allocation failed for Big");
       exit(EXIT_FAILURE);
    }
    // now initialize and store Big
    initializeBigWithSophisticatedValues(largeNum, b);
    Big = b;
    // ensure that the space is freed on exit or quick_exit
    atexit(destroyBig);
    at_quick_exit(destroyBig);
}

// the user thread function that relies on Big
int myThreadFunc(void* a) {
   call_once(&onceBig, initBig);
   // only use Big from here on
   ...
   return 0;
}

The once_flag is used to coordinate different threads that might want to initialize the same data Big. The call to call_once guarantees that

initBig is called exactly once
call_once blocks until such a call to initBig has been made, either by the same or another thread.

Besides allocation, a typical thing to do in such a once-called function is a dynamic initialization of a thread control data structures such as mtx_t or cnd_t that can’t be initialized statically, using mtx_init or cnd_init, respectively.

Found a mistake? Have a question or improvement idea? Let me know.

Table Of Contents