impl/worker_thread, client/app: Smooth post() processing of tasks

builtin/voxelworld/client_lua/module.lua

@@ -8,7 +8,7 @@ local cereal = require("buildat/extension/cereal")
 local dump = buildat.dump
 local M = {}
 
-local UPDATE_TIME_FRACTION = 0.20
+local UPDATE_TIME_FRACTION = 0.10
 
 --local LOD_DISTANCE = 140
 --local LOD_DISTANCE = 100

doc/todo.txt

@@ -49,3 +49,7 @@ Buildat TODO
 	- Allow connecting the input systems of multiple windows, which is simple
 	  but less automatic and most people don't know how to get rid of window
 	  borders
+- A way for module reloading to interrupt world generation
+- Maintain a global budget of background processing per frame so that
+  a long woxelworld:update and ThreadPool::run_post() will not be run on the
+  same frame

src/client/app.cpp

@@ -34,7 +34,7 @@
 #include <FileSystem.h>
 #include <PhysicsWorld.h>
 #include <DebugRenderer.h>
-//#include <Profiler.h>
+#include <Profiler.h>
 #pragma GCC diagnostic pop
 extern "C" {
 #include <lua.h>
@@ -535,7 +535,11 @@ struct CApp: public App, public magic::Application
 		if(m_state)
 			m_state->update();
 
-		m_thread_pool->run_post();
+		{
+			magic::AutoProfileBlock profiler_block(
+					GetSubsystem<magic::Profiler>(), "Buildat|ThreadPool::post");
+			m_thread_pool->run_post();
+		}
 	}
 
 	void on_post_render_update(

src/impl/worker_thread.cpp

@@ -4,6 +4,7 @@
 #include "interface/mutex.h"
 #include "interface/semaphore.h"
 #include "core/log.h"
+#include <c55/os.h>
 #include <deque>
 #ifdef _WIN32
 	#include "ports/windows_compat.h"
@@ -127,16 +128,48 @@ struct CThreadPool: public ThreadPool
 
 	void run_post()
 	{
-		std::deque<up_<Task>> output_queue;
-		{
-			interface::MutexScope ms(m_mutex);
-			output_queue.swap(m_output_queue);
+		int64_t t1 = get_timeofday_us();
+		size_t queue_size = 0;
+		size_t post_count = 0;
+		for(;;){
+			// Pop an output task
+			up_<Task> task;
+			{
+				interface::MutexScope ms(m_mutex);
+				if(!m_output_queue.empty()){
+					queue_size = m_output_queue.size();
+					task = std::move(m_output_queue.front());
+					m_output_queue.pop_front();
+				}
+			}
+			if(!task)
+				break;
+			// run post() until too long has passed
+			bool done = false;
+			for(;;){
+				post_count++;
+				done = task->post();
+				// If done, take next output task
+				if(done)
+					break;
+				int64_t t2 = get_timeofday_us();
+				int64_t max_t = 2000;
+				if(queue_size > 4)
+					max_t += (queue_size - 4) * 5000;
+				if(t2 - t1 >= max_t)
+					break;
+			}
+			// If still not done, push task to back to front of queue and stop
+			// processing
+			if(!done){
+				interface::MutexScope ms(m_mutex);
+				m_output_queue.push_front(std::move(task));
+				break;
+			}
 		}
-		// TODO: Limit task->post() execution time per frame
-		for(auto &task : output_queue){
-			while(!task->post());
-		}
-		// Done with tasks; discard them
+		/*int64_t t2 = get_timeofday_us();
+		log_v(MODULE, "output post(): %ius (%zu calls; queue size: %zu)",
+				(int)(t2 - t1), post_count, queue_size);*/
 	}
 };