Minor adjustment to coordinating thread spawns (catch seems to have problems…

Minor adjustment to coordinating thread spawns (catch seems to have problems with the other implementation).

ci_scripts/run_tests.sh

@@ -6,4 +6,6 @@ cmake .. -DCMAKE_BUILD_TYPE=RELEASE -DTHREAD_SANITIZER=OFF
 make
 
 # run the actual tests
+echo "Running tests..."
 ./bin/tests
+echo "...tests passed!"

lib/pls/include/pls/internal/scheduling/scheduler_impl.h

@@ -40,6 +40,7 @@ scheduler::scheduler(unsigned int num_threads,
   worker_threads_.reserve(num_threads);
   task_managers_.reserve(num_threads);
   thread_states_.reserve(num_threads);
+  std::atomic<unsigned> num_spawned{0};
   for (unsigned int i = 0; i < num_threads_; i++) {
     auto &this_task_manager =
         task_managers_.emplace_back(std::make_unique<task_manager>(i,
@@ -54,29 +55,22 @@ scheduler::scheduler(unsigned int num_threads,
 
     if (reuse_thread && i == 0) {
       worker_threads_.emplace_back();
+      num_spawned++;
       continue; // Skip over first/main thread when re-using the users thread, as this one will replace the first one.
     }
 
     auto *this_thread_state_pointer = this_thread_state.get();
-    worker_threads_.emplace_back([this_thread_state_pointer] {
+
+    worker_threads_.emplace_back([this_thread_state_pointer, &num_spawned] {
       thread_state::set(this_thread_state_pointer);
+      num_spawned++;
       work_thread_main_loop();
     });
   }
 
-  // Make sure all threads are created and touched their stacks.
-  // Executing a work section ensures one wakeup/sleep cycle of all workers
-  // and explicitly forcing one task per worker forces them to initialize their stacks.
-  std::atomic<unsigned> num_spawned;
-  this->perform_work([&]() {
-    for (unsigned i = 0; i < num_threads; i++) {
-      spawn([&]() {
-        num_spawned++;
-        while (num_spawned < num_threads) std::this_thread::yield();
-      });
-    }
-    sync();
-  });
+  while (num_spawned < num_threads) {
+    std::this_thread::yield();
+  }
 }
 
 class scheduler::init_function {
@@ -346,12 +340,12 @@ void scheduler::spawn_and_sync_internal(Function &&lambda) {
 
     // execute the lambda itself, which could lead to a different thread returning.
 #if PLS_PROFILING_ENABLED
-    spawning_state.get_scheduler().profiler_.task_finish_stack_measure(syncing_state.get_thread_id(),
-                                                                      last_task->stack_memory_,
-                                                                      last_task->stack_size_,
-                                                                      last_task->profiling_node_);
-    syncing_state.get_scheduler().profiler_.task_stop_running(spawning_state.get_thread_id(),
-                                                              last_task->profiling_node_);
+    spawning_state.get_scheduler().profiler_.task_finish_stack_measure(spawning_state.get_thread_id(),
+                                                                       last_task->stack_memory_,
+                                                                       last_task->stack_size_,
+                                                                       last_task->profiling_node_);
+    spawning_state.get_scheduler().profiler_.task_stop_running(spawning_state.get_thread_id(),
+                                                               last_task->profiling_node_);
     auto *next_dag_node =
         spawning_state.get_scheduler().profiler_.task_sync(spawning_state.get_thread_id(), last_task->profiling_node_);
     last_task->profiling_node_ = next_dag_node;