This routine generates a unique, thread-specific data key that
    is visible to all threads in the process. The variable key
    provided by this routine is an opaque object used to locate
    thread-specific data. Although the same key value can be used
    by different threads, the values bound to the key by pthread_
    setspecific() are maintained on a per-thread basis and persist
    for the life of the calling thread. The initial value of the key
    in all threads is NULL.

    The Threads Library imposes a maximum number of thread-specific
    data keys, equal to the symbolic constant PTHREAD_KEYS_MAX.

    Thread-specific data allows client software to associate "static"
    information with the current thread. For example, where a routine
    declares a variable static in a single-threaded program, a
    multithreaded version of the program might create a thread-
    specific data key to store the same variable.

    This routine generates and returns a new key value. The key
    reserves a cell within each thread. Each call to this routine
    creates a new cell that is unique within an application
    invocation. Keys must be generated from initialization code that
    is guaranteed to be called only once within each process. (See
    the pthread_once() description for more information.)

    When a thread terminates, its thread-specific data is
    automatically destroyed; however, the key remains unless
    destroyed by a call to pthread_key_delete(). An optional
    destructor function can be associated with each key. At thread
    exit, if a key has a non-NULL destructor pointer, and the thread
    has a non-NULL value associated with that key, the destructor
    function is called with the current associated value as its sole
    argument. The order in which thread-specific data destructors are
    called at thread termination is undefined.

    Before each destructor is called, the thread's value for the
    corresponding key is set to NULL. After the destructors have been
    called for all non-NULL values with associated destructors, if
    there are still some non-NULL values with associated destructors,
    then this sequence of actions is repeated. If there are still
    non-NULL values for any key with a destructor after four
    repetitions of this sequence, the thread is terminated. At this
    point, any key values that represent allocated heap will be lost.
    Note that this occurs only when a destructor performs some action
    that creates a new value for some key. Your program's destructor
    code should attempt to avoid this sort of circularity.