protect against potential runtime race condition.

suspect there's potential for platform-specific racing on our singular bu_init_lock depending on how static initialization is ordered.  this would explain several issues observed consistently on freebsd (possibly obscure issues on other platforms), but still testing.  the solution is to not rely on static initialization of the mutex, instead resorting to runtime init on first use via pthread_once().

src/libbu/semaphore.c

@@ -85,9 +85,30 @@ struct bu_semaphores {
     pthread_mutex_t mu;
 };
 
-static pthread_mutex_t bu_init_lock = SEMAPHORE_INIT;
+static pthread_mutex_t bu_init_lock;
+static pthread_once_t bu_init_once = PTHREAD_ONCE_INIT;
 #  define DEFINED_BU_SEMAPHORES 1
 
+static void bu_init_lock_init(void)
+{
+    pthread_mutexattr_t attr;
+    pthread_mutexattr_init(&attr);
+    pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
+    int ret = pthread_mutex_init(&bu_init_lock, &attr);
+    pthread_mutexattr_destroy(&attr);
+    if (ret) {
+	fprintf(stderr, "bu_semaphore_acquire(): pthread_mutex_init() failed on init lock\n");
+        sem_bomb(ret);
+    }
+}
+
+static void ensure_bu_init_lock_initialized(void)
+{
+    /* help ensure thread-safe initialization */
+    pthread_once(&bu_init_once, bu_init_lock_init);
+}
+
+
 /*
  * multithread support based on the Windows kernel.
  */
@@ -112,7 +133,6 @@ static struct bu_semaphores bu_semaphores[SEMAPHORE_MAX] = {{SEMAPHORE_MAGIC, SE
 #endif
 
 
-
 void
 bu_semaphore_init(unsigned int nsemaphores)
 {
@@ -158,6 +178,8 @@ bu_semaphore_init(unsigned int nsemaphores)
     }
 
 #  elif defined(HAVE_PTHREAD_H)
+    ensure_bu_init_lock_initialized();
+
     int ret = pthread_mutex_lock(&bu_init_lock);
     if (ret) {
 	fprintf(stderr, "bu_semaphore_acquire(): pthread_mutex_lock() failed on init lock\n");