Fixed problems with task dependency tracking

And problems with HPX annotated functions.  However, still need
some work to resolve memory leaks of the task_wrapper objects.

src/apex/apex.cpp

@@ -145,9 +145,9 @@ static void finalize_hpx_runtime(void) {
         }
     }
     // Tell other localities to shutdown APEX
-    apex_schedule_shutdown();
+    //apex_schedule_shutdown();
     // Shutdown APEX
-    finalize();
+    //finalize();
     hpx_finalized = true;
     FUNCTION_EXIT
 }
@@ -459,7 +459,7 @@ profiler* start(const std::string &timer_name)
         APEX_UTIL_REF_COUNT_APEX_INTERNAL_START
         return profiler::get_disabled_profiler(); // don't process our own events - queue scrubbing tasks.
     }
-#ifdef APEX_HAVE_HPX
+#ifdef APEX_HAVE_HPX_disabled
     // Finalize at the _start_ of HPX shutdown so that we can stop any
     // outstanding hpx::util::interval_timer instances. If any are left
     // running, HPX shutdown will never complete.
@@ -880,7 +880,7 @@ void stop(task_wrapper * tt_ptr) {
         }
     }
     // TODO - need to think about how to safely do this.  The parent may finish before the child.
-    delete(tt_ptr);
+    //delete(tt_ptr);
 }
 
 void yield(profiler* the_profiler)

src/apex/profiler_listener.cpp

@@ -742,8 +742,7 @@ node_color * get_node_color(double v,double vmin,double vmax)
 #ifdef APEX_HAVE_HPX
     num_worker_threads = num_worker_threads - num_non_worker_threads_registered;
 #endif
-    double total_main = main_timer->elapsed() *
-        fmin(hardware_concurrency(), num_worker_threads);
+    //double total_main = main_timer->elapsed() * fmin(hardware_concurrency(), num_worker_threads);
 
     // output nodes with  "main" [shape=box; style=filled; fillcolor="#ff0000" ];
     unordered_map<task_identifier, profile*>::const_iterator it;
@@ -901,9 +900,7 @@ node_color * get_node_color(double v,double vmin,double vmax)
           profiler_listener * pl = inst->the_profiler_listener;
           if (pl != nullptr) {
 #ifdef APEX_TRACE_APEX
-      profiler * p = start("apex::process_profiles");
-              pl->process_profiles();
-      stop(p);
+              scoped_timer p("apex::process_profiles");
 #else
               pl->process_profiles();
 #endif
@@ -1285,7 +1282,6 @@ if (rc != 0) cout << "PAPI error! " << name << ": " << PAPI_strerror(rc) << endl
       // start the profiler object, which starts our timers
       //std::shared_ptr<profiler> p = std::make_shared<profiler>(tt_ptr, is_resume);
       // get the right task identifier, based on whether there are aliases
-      task_identifier * id = tt_ptr->get_task_id();
       profiler * p = new profiler(tt_ptr, is_resume);
       p->guid = thread_instance::get_guid();
       thread_instance::instance().set_current_profiler(p);

src/apex/thread_instance.cpp

@@ -66,7 +66,6 @@ std::unordered_set<std::string> thread_instance::open_profilers;
 // Global static unordered map of parent GUIDs to child GUIDs
 // to handle "overlapping timer" problem.
 std::unordered_map<uint64_t, std::vector<profiler*>* > thread_instance::children_to_resume;
-static std::mutex _profiler_stack_mutex;
 
 thread_instance& thread_instance::instance(bool is_worker) {
   if( _instance == nullptr ) {