mutex_v2增加超时等待功能

修改无死锁锁定多个锁的算法，以便于支持mutex_v2

librf/librf.h

@@ -40,10 +40,10 @@
 
 #include "src/def.h"
 #include "src/macro_def.inl"
-#include "src/spinlock.h"
 #include "src/counted_ptr.h"
 #include "src/type_traits.inl"
 #include "src/type_concept.inl"
+#include "src/spinlock.h"
 #include "src/state.h"
 #include "src/future.h"
 #include "src/promise.h"
@@ -65,8 +65,8 @@
 #include "src/when.h"
 
 #include "src/_awaker.h"
-#include "src/mutex.h"
 #include "src/ring_queue.h"
 #include "src/intrusive_link_queue.h"
 #include "src/channel.h"
 #include "src/event.h"
+#include "src/mutex.h"

librf/src/def.h

@@ -1,6 +1,6 @@
 #pragma once
 
-#define LIB_RESUMEF_VERSION 20802 // 2.8.2
+#define LIB_RESUMEF_VERSION 20803 // 2.8.3
 
 #if defined(RESUMEF_MODULE_EXPORT)
 #define RESUMEF_NS export namespace resumef

librf/src/event_v2.h

@@ -38,12 +38,6 @@ RESUMEF_NS
 			timeout_awaiter wait_until(const std::chrono::time_point<_Clock, _Duration>& tp) const noexcept;
 
 
-			//以下wait_any/wait_all实现，借助when_any/when_all实现。
-			//其接口尽管兼容了v1版本的event_t，但是，其内部细节并没有兼容
-			//v1版本的event_t的wait_any，确保只触发了其中之一，或者超时
-			//而when_any会导致所有的event_t都被触发
-			//改日有空再补上
-
 			template<class _Iter>
 			struct [[nodiscard]] any_awaiter;
 

librf/src/event_v2.inl

@@ -69,7 +69,6 @@ RESUMEF_NS
 			virtual bool on_notify(event_v2_impl* eptr) = 0;
 			virtual bool on_timeout() = 0;
 
-			//将自己加入到通知链表里
 			template<class _PromiseT, typename = std::enable_if_t<traits::is_promise_v<_PromiseT>>>
 			scheduler_t* on_await_suspend(coroutine_handle<_PromiseT> handler) noexcept
 			{
@@ -83,8 +82,19 @@ RESUMEF_NS
 				return sch;
 			}
 
-			//为浸入式单向链表提供的next指针
+			inline void add_timeout_timer(std::chrono::system_clock::time_point tp)
+			{
+				this->_thandler = this->_scheduler->timer()->add_handler(tp, 
+					[st = counted_ptr<state_event_base_t>{ this }](bool canceld)
+					{
+						if (!canceld)
+							st->on_timeout();
+					});
+			}
+
+			//为侵入式单向链表提供的next指针
 			//counted_ptr<state_event_base_t> _next = nullptr;
+			timer_handler _thandler;
 		};
 
 		struct state_event_t : public state_event_base_t
@@ -96,10 +106,6 @@ RESUMEF_NS
 			virtual void on_cancel() noexcept override;
 			virtual bool on_notify(event_v2_impl* eptr) override;
 			virtual bool on_timeout() override;
-		public:
-			//typedef spinlock lock_type;
-
-			timer_handler _thandler;
 		protected:
 			//_value引用awaitor保存的值，这样可以尽可能减少创建state的可能。而不必进入没有state就没有value实体被用于返回。
 			//在调用on_notify()或on_timeout()任意之一后，置为nullptr。
@@ -118,12 +124,7 @@ RESUMEF_NS
 			virtual void on_cancel() noexcept override;
 			virtual bool on_notify(event_v2_impl* eptr) override;
 			virtual bool on_timeout() override;
-		public:
-			//typedef spinlock lock_type;
 
-			//为浸入式单向链表提供的next指针
-			//counted_ptr<state_event_t> _next = nullptr;
-			timer_handler _thandler;
 			std::atomic<intptr_t> _counter;
 		protected:
 			bool* _value;
@@ -202,7 +203,7 @@ RESUMEF_NS
 		struct event_t::timeout_awaitor_impl : public _Btype
 		{
 			template<class... Args>
-			timeout_awaitor_impl(clock_type::time_point tp, Args... args)
+			timeout_awaitor_impl(clock_type::time_point tp, Args&&... args) noexcept(std::is_nothrow_constructible_v<_Btype, Args&&...>)
 				: _Btype(std::forward<Args>(args)...)
 				, _tp(tp)
 			{}
@@ -211,29 +212,18 @@ RESUMEF_NS
 			{
 				if (!_Btype::await_suspend(handler))
 					return false;
-
-				_PromiseT& promise = handler.promise();
-				auto* parent_state = promise.get_state();
-				scheduler_t* sch = parent_state->get_scheduler();
-
-				this->_state->_thandler = sch->timer()->add_handler(_tp, [st = this->_state](bool canceld)
-				{
-					if (!canceld)
-						st->on_timeout();
-				});
-
+				this->_state->add_timeout_timer(_tp);
 				return true;
 			}
 		protected:
 			clock_type::time_point _tp;
 		};
 
-		struct [[nodiscard]] event_t::timeout_awaiter : event_t::timeout_awaitor_impl<event_t::awaiter>
+		struct [[nodiscard]] event_t::timeout_awaiter : timeout_awaitor_impl<event_t::awaiter>
 		{
 			timeout_awaiter(clock_type::time_point tp, detail::event_v2_impl* evt) noexcept
-				: timeout_awaitor_impl(tp, evt)
-			{
-			}
+				: timeout_awaitor_impl<event_t::awaiter>(tp, evt)
+			{}
 		};
 
 		template<class _Rep, class _Period>

librf/src/mutex_v2.cpp

@@ -4,7 +4,7 @@ RESUMEF_NS
 {
 	namespace detail
 	{
-		void state_mutex_t::resume()
+		void state_mutex_base_t::resume()
 		{
 			coroutine_handle<> handler = _coro;
 			if (handler)
@@ -15,28 +15,115 @@ RESUMEF_NS
 			}
 		}
 
-		bool state_mutex_t::has_handler() const  noexcept
+		bool state_mutex_base_t::has_handler() const  noexcept
 		{
 			return (bool)_coro;
 		}
 		
-		state_base_t* state_mutex_t::get_parent() const noexcept
+		state_base_t* state_mutex_base_t::get_parent() const noexcept
 		{
 			return _root;
 		}
 
-		bool state_mutex_t::on_notify()
+
+		void state_mutex_t::on_cancel() noexcept
+		{
+			mutex_v2_impl** oldValue = _value.load(std::memory_order_acquire);
+			if (oldValue != nullptr && _value.compare_exchange_strong(oldValue, nullptr, std::memory_order_acq_rel))
+			{
+				*oldValue = nullptr;
+				_thandler.stop();
+
+				this->_coro = nullptr;
+			}
+		}
+
+		bool state_mutex_t::on_notify(mutex_v2_impl* eptr)
+		{
+			mutex_v2_impl** oldValue = _value.load(std::memory_order_acquire);
+			if (oldValue != nullptr && _value.compare_exchange_strong(oldValue, nullptr, std::memory_order_acq_rel))
+			{
+				*oldValue = eptr;
+				_thandler.stop();
+
+				assert(this->_scheduler != nullptr);
+				if (this->_coro)
+					this->_scheduler->add_generator(this);
+
+				return true;
+			}
+			return false;
+		}
+
+		bool state_mutex_t::on_timeout()
+		{
+			mutex_v2_impl** oldValue = _value.load(std::memory_order_acquire);
+			if (oldValue != nullptr && _value.compare_exchange_strong(oldValue, nullptr, std::memory_order_acq_rel))
+			{
+				*oldValue = nullptr;
+				_thandler.reset();
+
+				assert(this->_scheduler != nullptr);
+				if (this->_coro)
+					this->_scheduler->add_generator(this);
+
+				return true;
+			}
+			return false;
+		}
+
+
+		void state_mutex_all_t::on_cancel() noexcept
+		{
+			intptr_t oldValue = _counter.load(std::memory_order_acquire);
+			if (oldValue >= 0 && _counter.compare_exchange_strong(oldValue, -1, std::memory_order_acq_rel))
+			{
+				*_value = false;
+				_thandler.stop();
+
+				this->_coro = nullptr;
+			}
+		}
+
+		bool state_mutex_all_t::on_notify(event_v2_impl*)
 		{
-			assert(this->_scheduler != nullptr);
-			if (this->_coro)
+			intptr_t oldValue = _counter.load(std::memory_order_acquire);
+			if (oldValue <= 0) return false;
+
+			oldValue = _counter.fetch_add(-1, std::memory_order_acq_rel);
+			if (oldValue == 1)
 			{
-				this->_scheduler->add_generator(this);
+				*_value = true;
+				_thandler.stop();
+
+				assert(this->_scheduler != nullptr);
+				if (this->_coro)
+					this->_scheduler->add_generator(this);
+
 				return true;
 			}
 
+			return oldValue >= 1;
+		}
+
+		bool state_mutex_all_t::on_timeout()
+		{
+			intptr_t oldValue = _counter.load(std::memory_order_acquire);
+			if (oldValue >= 0 && _counter.compare_exchange_strong(oldValue, -1, std::memory_order_acq_rel))
+			{
+				*_value = false;
+				_thandler.reset();
+
+				assert(this->_scheduler != nullptr);
+				if (this->_coro)
+					this->_scheduler->add_generator(this);
+
+				return true;
+			}
 			return false;
 		}
 
+
 		
 		void mutex_v2_impl::lock_until_succeed(void* sch)
 		{
@@ -93,20 +180,20 @@ RESUMEF_NS
 			{
 				if (_counter.fetch_sub(1, std::memory_order_relaxed) == 1)
 				{
-					if (!_wait_awakes.empty())
+					_owner.store(nullptr, std::memory_order_relaxed);
+					while (!_wait_awakes.empty())
 					{
 						state_mutex_ptr state = _wait_awakes.front();
 						_wait_awakes.pop_front();
 
-						//锁定状态转移到新的state上
-						_owner.store(state->get_root(), std::memory_order_relaxed);
-						_counter.fetch_add(1, std::memory_order_relaxed);
+						if (state->on_notify(this))
+						{
+							//锁定状态转移到新的state上
+							_owner.store(state->get_root(), std::memory_order_relaxed);
+							_counter.fetch_add(1, std::memory_order_relaxed);
 
-						state->on_notify();
-					}
-					else
-					{
-						_owner.store(nullptr, std::memory_order_relaxed);
+							break;
+						}
 					}
 				}
 

librf/src/mutex_v2.h

@@ -22,25 +22,25 @@ RESUMEF_NS
 			~mutex_t() noexcept;
 
 			struct [[nodiscard]] awaiter;
-			awaiter lock() const noexcept;
-			awaiter operator co_await() const noexcept;
+			awaiter/*scoped_lock_mutex_t*/ lock() const noexcept;
+			awaiter/*scoped_lock_mutex_t*/ operator co_await() const noexcept;
 
 			struct [[nodiscard]] try_awaiter;
 			//co_await try_lock()获得是否加锁成功。此操作无论成功与否都会立即返回。
 			//如果加锁成功，则需要调用co_await unlock()解锁。或者使用unlock(root_state())解锁。
 			//如果加锁失败，且要循环尝试加锁，则最好调用co_await yield()让出一次调度。否则，可能造成本调度器死循环。
-			try_awaiter try_lock() const noexcept;
+			try_awaiter/*bool*/ try_lock() const noexcept;
 
 			//此操作立即返回
 			struct [[nodiscard]] unlock_awaiter;
-			unlock_awaiter unlock() const noexcept;
+			unlock_awaiter/*void*/ unlock() const noexcept;
 
 
 			struct [[nodiscard]] timeout_awaiter;
 			template <class _Rep, class _Period>
-			timeout_awaiter try_lock_for(const std::chrono::duration<_Rep, _Period>& dt) const noexcept;
+			timeout_awaiter/*bool*/ try_lock_for(const std::chrono::duration<_Rep, _Period>& dt) const noexcept;
 			template <class _Rep, class _Period>
-			timeout_awaiter try_lock_until(const std::chrono::time_point<_Rep, _Period>& tp) const noexcept;
+			timeout_awaiter/*bool*/ try_lock_until(const std::chrono::time_point<_Rep, _Period>& tp) const noexcept;
 
 
 			void lock(void* unique_address) const;

librf/src/mutex_v2.inl

@@ -4,13 +4,15 @@ RESUMEF_NS
 {
 	namespace detail
 	{
-		struct state_mutex_t : public state_base_t
+		struct state_mutex_base_t : public state_base_t
 		{
 			virtual void resume() override;
 			virtual bool has_handler() const  noexcept override;
 			virtual state_base_t* get_parent() const noexcept override;
 
-			bool on_notify();
+			virtual void on_cancel() noexcept = 0;
+			virtual bool on_notify(mutex_v2_impl* eptr) = 0;
+			virtual bool on_timeout() = 0;
 
 			inline scheduler_t* get_scheduler() const noexcept
 			{
@@ -24,12 +26,53 @@ RESUMEF_NS
 				this->_root = root;
 			}
 
-		private:
+			inline void add_timeout_timer(std::chrono::system_clock::time_point tp)
+			{
+				this->_thandler = this->_scheduler->timer()->add_handler(tp,
+					[st = counted_ptr<state_mutex_base_t>{ this }](bool canceld)
+					{
+						if (!canceld)
+							st->on_timeout();
+					});
+			}
+
+			timer_handler _thandler;
+		protected:
 			state_base_t* _root;
-			//friend mutex_v2::mutex_t;
 		};
 
-		//做成递归锁?
+		struct state_mutex_t : public state_mutex_base_t
+		{
+			state_mutex_t(mutex_v2_impl*& val)
+				: _value(&val)
+			{}
+
+			virtual void on_cancel() noexcept override;
+			virtual bool on_notify(mutex_v2_impl* eptr) override;
+			virtual bool on_timeout() override;
+		public:
+			timer_handler _thandler;
+		protected:
+			std::atomic<mutex_v2_impl**> _value;
+		};
+
+		struct state_mutex_all_t : public state_event_base_t
+		{
+			state_mutex_all_t(intptr_t count, bool& val)
+				: _counter(count)
+				, _value(&val)
+			{}
+
+			virtual void on_cancel() noexcept override;
+			virtual bool on_notify(event_v2_impl* eptr) override;
+			virtual bool on_timeout() override;
+		public:
+			timer_handler _thandler;
+			std::atomic<intptr_t> _counter;
+		protected:
+			bool* _value;
+		};
+
 		struct mutex_v2_impl : public std::enable_shared_from_this<mutex_v2_impl>
 		{
 			using clock_type = std::chrono::system_clock;
@@ -159,12 +202,18 @@ RESUMEF_NS
 				_PromiseT& promise = handler.promise();
 				auto* parent = promise.get_state();
 				_root = parent->get_root();
+				if (_root == nullptr)
+				{
+					assert(false);
+					_mutex = nullptr;
+					return false;
+				}
 
 				scoped_lock<detail::mutex_v2_impl::lock_type> lock_(_mutex->_lock);
 				if (_mutex->try_lock_lockless(_root))
 					return false;
 
-				_state = new detail::state_mutex_t();
+				_state = new detail::state_mutex_t(_mutex);
 				_state->on_await_suspend(handler, parent->get_scheduler(), _root);
 
 				_mutex->add_wait_list_lockless(_state.get());
@@ -174,7 +223,7 @@ RESUMEF_NS
 
 			scoped_lock_mutex_t await_resume() noexcept
 			{
-				mutex_impl_ptr mtx = _root ? _mutex->shared_from_this() : nullptr;
+				mutex_impl_ptr mtx = _mutex ? _mutex->shared_from_this() : nullptr;
 				_mutex = nullptr;
 
 				return { std::adopt_lock, mtx, _root };
@@ -203,6 +252,14 @@ RESUMEF_NS
 			{
 				assert(_mutex != nullptr);
 			}
+			~try_awaiter() noexcept(false)
+			{
+				assert(_mutex == nullptr);
+				if (_mutex != nullptr)
+				{
+					throw lock_exception(error_code::not_await_lock);
+				}
+			}
 
 			bool await_ready() noexcept
 			{
@@ -222,7 +279,9 @@ RESUMEF_NS
 
 			bool await_resume() noexcept
 			{
-				return _mutex != nullptr;
+				detail::mutex_v2_impl* mtx = _mutex;
+				_mutex = nullptr;
+				return mtx != nullptr;
 			}
 		protected:
 			detail::mutex_v2_impl* _mutex;
@@ -240,6 +299,14 @@ RESUMEF_NS
 			{
 				assert(_mutex != nullptr);
 			}
+			~unlock_awaiter() noexcept(false)
+			{
+				assert(_mutex == nullptr);
+				if (_mutex != nullptr)
+				{
+					throw lock_exception(error_code::not_await_lock);
+				}
+			}
 
 			bool await_ready() noexcept
 			{
@@ -258,6 +325,7 @@ RESUMEF_NS
 
 			void await_resume() noexcept
 			{
+				_mutex = nullptr;
 			}
 		protected:
 			detail::mutex_v2_impl* _mutex;
@@ -270,73 +338,32 @@ RESUMEF_NS
 
 
 
-		struct [[nodiscard]] mutex_t::timeout_awaiter
-		{
-			timeout_awaiter(detail::mutex_v2_impl* mtx, clock_type::time_point tp) noexcept
-				: _mutex(mtx)
-				, _tp(tp)
-			{
-				assert(_mutex != nullptr);
-			}
-
-			~timeout_awaiter() noexcept(false)
-			{
-				assert(_mutex == nullptr);
-				if (_mutex != nullptr)
-				{
-					throw lock_exception(error_code::not_await_lock);
-				}
-			}
-
-			bool await_ready() noexcept
-			{
-				return false;
-			}
-
-			template<class _PromiseT, typename = std::enable_if_t<traits::is_promise_v<_PromiseT>>>
-			bool await_suspend(coroutine_handle<_PromiseT> handler)
-			{
-				_PromiseT& promise = handler.promise();
-				auto* parent = promise.get_state();
-				_root = parent->get_root();
 
-				scoped_lock<detail::mutex_v2_impl::lock_type> lock_(_mutex->_lock);
-				if (_mutex->try_lock_lockless(_root))
-					return false;
-
-				_state = new detail::state_mutex_t();
-				_state->on_await_suspend(handler, parent->get_scheduler(), _root);
-
-				_mutex->add_wait_list_lockless(_state.get());
-
-				return true;
-			}
+		struct [[nodiscard]] mutex_t::timeout_awaiter : public event_t::timeout_awaitor_impl<awaiter>
+		{
+			timeout_awaiter(clock_type::time_point tp, detail::mutex_v2_impl * mtx) noexcept
+				: event_t::timeout_awaitor_impl<mutex_t::awaiter>(tp, mtx)
+			{}
 
-			scoped_lock_mutex_t await_resume() noexcept
+			bool await_resume() noexcept
 			{
-				mutex_impl_ptr mtx = _root ? _mutex->shared_from_this() : nullptr;
-				_mutex = nullptr;
-
-				return { std::adopt_lock, mtx, _root };
+				detail::mutex_v2_impl* mtx = this->_mutex;
+				this->_mutex = nullptr;
+				return mtx != nullptr;
 			}
-		protected:
-			detail::mutex_v2_impl* _mutex;
-			clock_type::time_point _tp;
-			counted_ptr<detail::state_mutex_t> _state;
-			state_base_t* _root = nullptr;
 		};
 
 		template <class _Rep, class _Period>
 		inline mutex_t::timeout_awaiter mutex_t::try_lock_until(const std::chrono::time_point<_Rep, _Period>& tp) const noexcept
 		{
-			return { _mutex.get(), tp };
+			return { tp, _mutex.get() };
 		}
 
 		template <class _Rep, class _Period>
 		inline mutex_t::timeout_awaiter mutex_t::try_lock_for(const std::chrono::duration<_Rep, _Period>& dt) const noexcept
 		{
 			auto tp = clock_type::now() + std::chrono::duration_cast<clock_type::duration>(dt);
-			return { _mutex.get(), tp };
+			return { tp, _mutex.get() };
 		}
 
 

librf/src/spinlock.h

@@ -88,172 +88,127 @@ RESUMEF_NS
 	};
 
 #endif
-	
+
 	namespace detail
 	{
-		template<class _Ty>
-		void _Lock_ref(_Ty& _LkN)
-		{
-			_LkN.lock();
-		}
-		template<class _Ty>
-		void _Lock_ref(std::reference_wrapper<_Ty> _LkN)
-		{
-			_LkN.get().lock();
-		}
-		template<class _Ty>
-		void _Unlock_ref(_Ty& _LkN)
-		{
-			_LkN.unlock();
-		}
-		template<class _Ty>
-		void _Unlock_ref(std::reference_wrapper<_Ty> _LkN)
-		{
-			_LkN.get().unlock();
-		}
-		template<class _Ty>
-		bool _Try_lock_ref(_Ty& _LkN)
-		{
-			return _LkN.try_lock();
-		}
-		template<class _Ty>
-		bool _Try_lock_ref(std::reference_wrapper<_Ty> _LkN)
+#if RESUMEF_ENABLE_CONCEPT
+		template<typename T>
+		concept _LockAssembleT = requires(T && v)
 		{
-			return _LkN.get().try_lock();
-		}
-
-		template<class _Ty>
-		void _Lock_from_locks(const int _Target, std::vector<_Ty>& _LkN) { // lock _LkN[_Target]
-			_Lock_ref(_LkN[_Target]);
-		}
-		// FUNCTION TEMPLATE _Try_lock_from_locks
-		template<class _Ty>
-		bool _Try_lock_from_locks(const int _Target, std::vector<_Ty>& _LkN) { // try to lock _LkN[_Target]
-			return _Try_lock_ref(_LkN[_Target]);
-		}
+			{ v.size() };
+			{ v[0] };
+			{ v._Lock_ref(v[0]) } ->void;
+			{ v._Try_lock_ref(v[0]) } ->bool;
+			{ v._Unlock_ref(v[0]) } ->void;
+			{ v._Yield() };
+			{ v._ReturnValue() };
+			{ v._ReturnValue(0) };
+			requires std::is_integral_v<decltype(v.size())>;
+		};
+#else
+#define _LockAssembleT typename
+#endif
 
-		// FUNCTION TEMPLATE _Unlock_locks
 		template<class _Ty>
-		void _Unlock_locks(int _First, int _Last, std::vector<_Ty>& _LkN) noexcept /* terminates */ {
-			for (; _First != _Last; ++_First) {
-				_Unlock_ref(_LkN[_First]);
+		struct _LockVectorAssembleT
+		{
+		private:
+			std::vector<_Ty>& _Lks;
+		public:
+			_LockVectorAssembleT(std::vector<_Ty>& _LkN)
+				: _Lks(_LkN)
+			{}
+			size_t size() const
+			{
+				return _Lks.size();
 			}
-		}
-
-		// FUNCTION TEMPLATE try_lock
-		template<class _Ty>
-		int _Try_lock_range(const int _First, const int _Last, std::vector<_Ty>& _LkN) {
-			int _Next = _First;
-			try {
-				for (; _Next != _Last; ++_Next) {
-					if (!_Try_lock_from_locks(_Next, _LkN)) { // try_lock failed, backout
-						_Unlock_locks(_First, _Next, _LkN);
-						return _Next;
-					}
-				}
+			_Ty& operator[](int _Idx)
+			{
+				return _Lks[_Idx];
+			}
+			void _Lock_ref(_Ty& _LkN) const
+			{
+				_LkN.lock();
 			}
-			catch (...) {
-				_Unlock_locks(_First, _Next, _LkN);
-				throw;
+			bool _Try_lock_ref(_Ty& _LkN) const
+			{
+				return _LkN.try_lock();
 			}
-
-			return -1;
-		}
-
-		// FUNCTION TEMPLATE lock
-		template<class _Ty>
-		int _Lock_attempt(const int _Hard_lock, std::vector<_Ty>& _LkN) {
-			// attempt to lock 3 or more locks, starting by locking _LkN[_Hard_lock] and trying to lock the rest
-			_Lock_from_locks(_Hard_lock, _LkN);
-			int _Failed = -1;
-			int _Backout_start = _Hard_lock; // that is, unlock _Hard_lock
-
-			try {
-				_Failed = _Try_lock_range(0, _Hard_lock, _LkN);
-				if (_Failed == -1) {
-					_Backout_start = 0; // that is, unlock [0, _Hard_lock] if the next throws
-					_Failed = _Try_lock_range(_Hard_lock + 1, (int)_LkN.size(), _LkN);
-					if (_Failed == -1) { // we got all the locks
-						return -1;
-					}
-				}
+			void _Unlock_ref(_Ty& _LkN) const
+			{
+				_LkN.unlock();
 			}
-			catch (...) {
-				_Unlock_locks(_Backout_start, _Hard_lock + 1, _LkN);
-				throw;
+			void _Yield() const
+			{
+				std::this_thread::yield();
 			}
-
-			// we didn't get all the locks, backout
-			_Unlock_locks(_Backout_start, _Hard_lock + 1, _LkN);
-			std::this_thread::yield();
-
-			return _Failed;
-		}
+			void _ReturnValue() const noexcept {}
+			template<class U>
+			U _ReturnValue(U v) const noexcept
+			{
+				return v;
+			}
+		};
 
 		template<class _Ty>
-		void _Lock_nonmember3(std::vector<_Ty>& _LkN) {
-			// lock 3 or more locks, without deadlock
-			int _Hard_lock = 0;
-			while (_Hard_lock != -1) {
-				_Hard_lock = _Lock_attempt(_Hard_lock, _LkN);
-			}
-		}
-					
-		template <class _Lock0, class _Lock1>
-		bool _Lock_attempt_small2(_Lock0& _Lk0, _Lock1& _Lk1)
+		struct _LockVectorAssembleT<std::reference_wrapper<_Ty>>
 		{
-			// attempt to lock 2 locks, by first locking _Lk0, and then trying to lock _Lk1 returns whether to try again
-			_Lock_ref(_Lk0);
-			try {
-				if (_Try_lock_ref(_Lk1))
-					return false;
+		private:
+			std::vector<std::reference_wrapper<_Ty>>& _Lks;
+		public:
+			_LockVectorAssembleT(std::vector<std::reference_wrapper<_Ty>>& _LkN)
+				: _Lks(_LkN)
+			{}
+			size_t size() const
+			{
+				return _Lks.size();
 			}
-			catch (...) {
-				_Unlock_ref(_Lk0);
-				throw;
+			std::reference_wrapper<_Ty> operator[](int _Idx)
+			{
+				return _Lks[_Idx];
 			}
-
-			_Unlock_ref(_Lk0);
-			std::this_thread::yield();
-			return true;
-		}
-
-		template <class _Lock0, class _Lock1>
-		void _Lock_nonmember2(_Lock0& _Lk0, _Lock1& _Lk1)
-		{
-			// lock 2 locks, without deadlock, special case for better codegen and reduced metaprogramming for common case
-			while (_Lock_attempt_small2(_Lk0, _Lk1) && _Lock_attempt_small2(_Lk1, _Lk0)) { // keep trying
+			void _Lock_ref(std::reference_wrapper<_Ty> _LkN) const
+			{
+				_LkN.get().lock();
 			}
-		}
-
-		template<class _Ty>
-		void _Lock_range(std::vector<_Ty>& lockes)
-		{
-			if (lockes.size() == 0)
+			void _Unlock_ref(std::reference_wrapper<_Ty> _LkN) const
 			{
+				_LkN.get().unlock();
 			}
-			else if (lockes.size() == 1)
+			bool _Try_lock_ref(std::reference_wrapper<_Ty> _LkN) const
 			{
-				_Lock_ref(lockes[0]);
+				return _LkN.get().try_lock();
 			}
-			else if (lockes.size() == 2)
+			void _Yield() const
 			{
-				_Lock_nonmember2(lockes[0], lockes[1]);
+				std::this_thread::yield();
 			}
-			else
+			void _ReturnValue() const noexcept {}
+			template<class U>
+			U _ReturnValue(U v) const noexcept
 			{
-				_Lock_nonmember3(lockes);
+				return v;
 			}
-		}
+		};
 	}
+}
+
+#define LOCK_ASSEMBLE_AWAIT(a) (a)
+#define LOCK_ASSEMBLE_RETURN(a) return (a)
+#include "without_deadlock_assemble.inl"
+#undef LOCK_ASSEMBLE_AWAIT
+#undef LOCK_ASSEMBLE_RETURN
 
+RESUMEF_NS
+{
 	template<class _Ty>
 	class scoped_lock_range { // class with destructor that unlocks mutexes
 	public:
 		explicit scoped_lock_range(std::vector<_Ty>& locks_)
 			: _LkN(locks_)
 		{
-			detail::_Lock_range(locks_);
+			detail::_LockVectorAssembleT<_Ty> LA{ _LkN };
+			detail::_Lock_range(LA);
 		}
 
 		explicit scoped_lock_range(std::adopt_lock_t, std::vector<_Ty>& locks_)
@@ -263,7 +218,8 @@ RESUMEF_NS
 
 		~scoped_lock_range() noexcept
 		{
-			detail::_Unlock_locks(0, (int)_LkN.size(), _LkN);
+			detail::_LockVectorAssembleT<_Ty> LA{ _LkN };
+			detail::_Unlock_locks(0, (int)_LkN.size(), LA);
 		}
 
 		scoped_lock_range(const scoped_lock_range&) = delete;

librf/src/type_concept.inl

@@ -2,9 +2,9 @@
 
 #ifndef RESUMEF_ENABLE_CONCEPT
 #ifdef __cpp_lib_concepts
-#define RESUMEF_ENABLE_CONCEPT	0
+#define RESUMEF_ENABLE_CONCEPT	1
 #else
-#define RESUMEF_ENABLE_CONCEPT	0
+#define RESUMEF_ENABLE_CONCEPT	1
 #endif	//#ifdef __cpp_lib_concepts
 #endif	//#ifndef RESUMEF_ENABLE_CONCEPT
 

librf/src/without_deadlock_assemble.inl

@@ -0,0 +1,134 @@
+
+RESUMEF_NS
+{
+	namespace detail
+	{
+		// FUNCTION TEMPLATE _Unlock_locks
+		template<_LockAssembleT _LA>
+		auto _Unlock_locks(int _First, int _Last, _LA& _LkN) noexcept /* terminates */
+			->decltype(_LkN._ReturnValue())
+		{
+			for (; _First != _Last; ++_First) {
+				LOCK_ASSEMBLE_AWAIT(_LkN._Unlock_ref(_LkN[_First]));
+			}
+		}
+
+		// FUNCTION TEMPLATE try_lock
+		template<_LockAssembleT _LA>
+		auto _Try_lock_range(const int _First, const int _Last, _LA& _LkN) 
+			->decltype(_LkN._ReturnValue<int>(0))
+		{
+			int _Next = _First;
+			try {
+				for (; _Next != _Last; ++_Next)
+				{
+					if (!LOCK_ASSEMBLE_AWAIT(_LkN._Try_lock_ref(_LkN[_Next])))
+					{ // try_lock failed, backout
+						LOCK_ASSEMBLE_AWAIT(_Unlock_locks(_First, _Next, _LkN));
+						LOCK_ASSEMBLE_RETURN(_Next);
+					}
+				}
+			}
+			catch (...) {
+				LOCK_ASSEMBLE_AWAIT(_Unlock_locks(_First, _Next, _LkN));
+				throw;
+			}
+
+			LOCK_ASSEMBLE_RETURN(-1);
+		}
+
+		// FUNCTION TEMPLATE lock
+		template<_LockAssembleT _LA>
+		auto _Lock_attempt(const int _Hard_lock, _LA& _LkN)
+			->decltype(_LkN._ReturnValue<int>(0))
+		{
+			// attempt to lock 3 or more locks, starting by locking _LkN[_Hard_lock] and trying to lock the rest
+			LOCK_ASSEMBLE_AWAIT(_LkN._Lock_ref(_LkN[_Hard_lock]));
+			int _Failed = -1;
+			int _Backout_start = _Hard_lock; // that is, unlock _Hard_lock
+
+			try {
+				_Failed = LOCK_ASSEMBLE_AWAIT(_Try_lock_range(0, _Hard_lock, _LkN));
+				if (_Failed == -1)
+				{
+					_Backout_start = 0; // that is, unlock [0, _Hard_lock] if the next throws
+					_Failed = LOCK_ASSEMBLE_AWAIT(_Try_lock_range(_Hard_lock + 1, (int)_LkN.size(), _LkN));
+					if (_Failed == -1) { // we got all the locks
+						LOCK_ASSEMBLE_RETURN(-1);
+					}
+				}
+			}
+			catch (...) {
+				LOCK_ASSEMBLE_AWAIT(_Unlock_locks(_Backout_start, _Hard_lock + 1, _LkN));
+				throw;
+			}
+
+			// we didn't get all the locks, backout
+			LOCK_ASSEMBLE_AWAIT(_Unlock_locks(_Backout_start, _Hard_lock + 1, _LkN));
+			LOCK_ASSEMBLE_AWAIT(_LkN._Yield());
+
+			LOCK_ASSEMBLE_RETURN(_Failed);
+		}
+
+		template<_LockAssembleT _LA>
+		auto _Lock_nonmember3(_LA& _LkN) ->decltype(_LkN._ReturnValue())
+		{
+			// lock 3 or more locks, without deadlock
+			int _Hard_lock = 0;
+			while (_Hard_lock != -1) {
+				_Hard_lock = LOCK_ASSEMBLE_AWAIT(_Lock_attempt(_Hard_lock, _LkN));
+			}
+		}
+					
+		template<_LockAssembleT _LA>
+		auto _Lock_attempt_small2(_LA& _LkN, const int _Idx0, const int _Idx1)
+			->decltype(_LkN._ReturnValue<bool>(false))
+		{
+			// attempt to lock 2 locks, by first locking _Lk0, and then trying to lock _Lk1 returns whether to try again
+			LOCK_ASSEMBLE_AWAIT(_LkN._Lock_ref(_LkN[_Idx0]));
+			try {
+				if (LOCK_ASSEMBLE_AWAIT(_LkN._Try_lock_ref(_LkN[_Idx1])))
+					LOCK_ASSEMBLE_RETURN(false);
+			}
+			catch (...) {
+				LOCK_ASSEMBLE_AWAIT(_LkN._Unlock_ref(_LkN[_Idx0]));
+				throw;
+			}
+
+			LOCK_ASSEMBLE_AWAIT(_LkN._Unlock_ref(_LkN[_Idx0]));
+			LOCK_ASSEMBLE_AWAIT(_LkN._Yield());
+
+			LOCK_ASSEMBLE_RETURN(true);
+		}
+
+		template<_LockAssembleT _LA>
+		auto _Lock_nonmember2(_LA& _LkN) ->decltype(_LkN._ReturnValue())
+		{
+			// lock 2 locks, without deadlock, special case for better codegen and reduced metaprogramming for common case
+			while (LOCK_ASSEMBLE_AWAIT(_Lock_attempt_small2(_LkN, 0, 1)) && 
+				LOCK_ASSEMBLE_AWAIT(_Lock_attempt_small2(_LkN, 1, 0)))
+			{ // keep trying
+			}
+		}
+
+		template<_LockAssembleT _LA>
+		auto _Lock_range(_LA& lockes) ->decltype(lockes._ReturnValue())
+		{
+			if (lockes.size() == 0)
+			{
+			}
+			else if (lockes.size() == 1)
+			{
+				LOCK_ASSEMBLE_AWAIT(lockes._Lock_ref(lockes[0]));
+			}
+			else if (lockes.size() == 2)
+			{
+				LOCK_ASSEMBLE_AWAIT(_Lock_nonmember2(lockes));
+			}
+			else
+			{
+				LOCK_ASSEMBLE_AWAIT(_Lock_nonmember3(lockes));
+			}
+		}
+	}
+}

tutorial/test_async_mutex.cpp

@@ -76,6 +76,24 @@ static future_t<> test_mutex_try_push(size_t idx)
 	}
 }
 
+static future_t<> test_mutex_timeout_push(size_t idx)
+{
+	for (size_t i = 0; i < N; ++i)
+	{
+		{
+			while (!co_await g_lock.try_lock_for(10ms))
+				co_await yield();
+
+			++g_counter;
+			std::cout << "push:" << g_counter << " on " << idx << std::endl;
+
+			co_await 50ms;
+			co_await g_lock.unlock();
+		}
+		co_await 50ms;
+	}
+}
+
 //🔒-50ms-🗝-50ms-🔒-50ms-🗝-50ms-|
 //---------........---------.......
 static std::thread test_mutex_async_push(size_t idx)
@@ -103,7 +121,7 @@ static std::thread test_mutex_async_push(size_t idx)
 
 static void resumable_mutex_synch()
 {
-	go test_mutex_try_push(0);
+	go test_mutex_timeout_push(0);
 	go test_mutex_pop(1);
 
 	this_scheduler()->run_until_notask();

vs_proj/librf.cpp

@@ -43,6 +43,7 @@ int main(int argc, const char* argv[])
 	//test_ring_queue<resumef::ring_queue_spinlock<int, false, uint32_t>>();
 	//test_ring_queue<resumef::ring_queue_lockfree<int, uint64_t>>();
 
+	resumable_main_event();
 	resumable_main_mutex();
 	return 0;
 

vs_proj/librf.vcxproj

@@ -263,6 +263,7 @@
     <None Include="..\librf\src\channel_v2.inl" />
     <None Include="..\librf\src\event_v2.inl" />
     <None Include="..\librf\src\exception.inl" />
+    <None Include="..\librf\src\without_deadlock_assemble.inl" />
     <None Include="..\librf\src\macro_def.inl" />
     <None Include="..\librf\src\mutex_v2.inl" />
     <None Include="..\librf\src\promise.inl" />

vs_proj/librf.vcxproj.filters

@@ -276,5 +276,8 @@
     <None Include="..\librf\src\mutex_v2.inl">
       <Filter>librf\src</Filter>
     </None>
+    <None Include="..\librf\src\without_deadlock_assemble.inl">
+      <Filter>librf\src</Filter>
+    </None>
   </ItemGroup>
 </Project>