task complete callback

Mutex waiting for task complete. Waiting for a task to be finished, The destructor for the thread and thread pool now take less than 1% usage. 😎

include/MultiThreading/ConcurrentQueue.h

@@ -0,0 +1,123 @@
+#pragma once
+
+#include <queue>
+#include <atomic>
+#include <mutex>
+#include <condition_variable>
+#include <type_traits>
+
+/// A simple C++11 Concurrent Queue based on std::queue.
+/// Supports waiting operations for retrieving an element when it's empty.
+/// It's interrupted by calling interrupt.
+
+namespace MultiThreading {
+
+    template<typename T, typename Container = std::deque<T> >
+    class ConcurrentQueue {
+    public:
+        typedef typename std::queue<T>::size_type size_type;
+        typedef typename std::queue<T>::reference reference;
+        typedef typename std::queue<T>::const_reference const_reference;
+
+        ~ConcurrentQueue() {
+            interrupt();
+        }
+
+        void interrupt() {
+            interrupted = true;
+            condition_variable.notify_one();
+        }
+
+        bool contains(const T &value) {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            std::queue<T, Container> copy = queue;
+            while (!copy.empty())
+                if (copy.front() == value)
+                    return true;
+                else
+                    copy.pop();
+            return false;
+        }
+
+        void push(const T &e) {
+            std::unique_lock<std::mutex> lock(mutex);
+            queue.push(e);
+            condition_variable.notify_one();
+        }
+
+        template<typename... Args>
+        void emplace(Args &&... args) {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            queue.emplace(std::forward<Args>(args)...);
+            condition_variable.notify_one();
+        }
+
+        bool empty() {
+            //std::unique_lock<std::mutex> lock(this->mutex);
+            return queue.empty();
+        }
+
+        void pop() {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            if (!queue.empty())
+                queue.pop();
+        }
+
+        void front_pop(T &ret) {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            wait(lock);
+            ret = queue.front();
+            queue.pop();
+        }
+
+        size_type size() const {
+            //std::unique_lock<std::mutex> lock(this->mutex);
+            return queue.size();
+        }
+
+        reference front() {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            wait(lock);
+            return queue.front();
+        }
+
+        /*const_reference front() const {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            wait(lock);
+            return queue.front();
+        }*/
+
+        reference back() {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            wait(lock);
+            return queue.back();
+        }
+
+        /*const_reference back() const {
+            std::unique_lock<std::mutex> lock(this->mutex);
+            wait(lock);
+            return queue.back();
+        }*/
+
+        void swap(ConcurrentQueue &q) {
+            throw std::runtime_error("Not supported");
+        }
+
+    protected:
+        std::queue<T, Container> queue;
+        std::mutex mutex;
+        std::condition_variable condition_variable;
+        std::atomic_bool interrupted;
+
+
+    private:
+        void wait(std::unique_lock<std::mutex> &lock) {
+            interrupted = false;
+            while (queue.empty()) {
+                condition_variable.wait(lock);
+                if (interrupted)
+                    throw std::runtime_error("Interrupted");
+            }
+        }
+    };
+}

include/MultiThreading/Task.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <condition_variable>
 #include <functional>
 #include <memory>
 #include <atomic>
@@ -14,13 +15,22 @@ namespace MultiThreading {
 
 class MultiThreading::TaskBase {
 protected:
+    std::function<void()> complete_callback = nullptr;
+    std::condition_variable completed_condition;
     std::atomic<bool> complete = false;
+    std::mutex mtx;
+    void MarkTaskComplete() { std::lock_guard<std::mutex> lock(mtx); complete = true; completed_condition.notify_all(); }
 public:
     /// @returns whether the task is finished.
     [[nodiscard]] bool Complete() const { return complete; }
+
+    /// Condition variable waits for the task to be complete
+    /// @note There is no way to know if the task has been assigned, If you do this and that never happens the calling thread will wait forever.
+    void WaitComplete() { std::unique_lock<std::mutex> lock(mtx); completed_condition.wait(lock, [this]() { return complete.load(); }); }
+
     virtual void Run() = 0;
 public:
-    TaskBase() = default;
+    explicit TaskBase(std::function<void()> task_complete_callback = nullptr) : complete_callback(std::move(task_complete_callback)) {};
 };
 
 template <typename T>
@@ -29,15 +39,15 @@ private:
     T* result = nullptr;
     std::function<T()> callable = nullptr;
 private:
-    explicit Task(std::function<T()> callable, T* result = nullptr) : TaskBase(), result(result), callable(callable) {}
+    explicit Task(std::function<T()> callable, std::function<void()> complete_callback = nullptr, T* result = nullptr) : TaskBase(complete_callback), result(result), callable(std::move(callable)) {}
 public:
-    void Run() final { result ? *result = callable() : callable(); complete = true; }
+    void Run() final { result ? *result = callable() : callable(); MarkTaskComplete(); if (complete_callback) complete_callback();  }
 public:
     /// Create a Task. This is non-negotiable, This ensures that there's no issues related to the stack frame the task is on being popped before or while the job runs.
     /// @param callable The function to run, *usually a lambda or std::bind*
     /// @param result If your function returns a value, This is where you want it to go. nullptr for no return value or you don't want it.
     /// @note this is shared_ptr so you don't have to delete it.
-    static std::shared_ptr<Task<T>> Create(std::function<T()> callable, T* result = nullptr) { return std::shared_ptr<Task<T>>(new Task<T>(callable, result)); };
+    static std::shared_ptr<Task<T>> Create(std::function<T()> callable, const std::function<void()>& complete_callback = nullptr, T* result = nullptr) { return std::shared_ptr<Task<T>>(new Task<T>(callable, complete_callback, result)); };
     ~Task() = default;
 };
 
@@ -46,13 +56,13 @@ template <>
 class MultiThreading::Task<void> : public TaskBase {
 private:
     std::function<void()> callable = nullptr;
-private:
-    explicit Task(std::function<void()> callable) : TaskBase(), callable(std::move(callable)) {}
+    explicit Task(std::function<void()> callable, std::function<void()> complete_callback = nullptr) : TaskBase(std::move(complete_callback)), callable(std::move(callable)) {}
 public:
-    void Run() final { callable(); complete = true; }
+    void Run() final { callable(); MarkTaskComplete(); if (complete_callback) complete_callback(); }
 
     /// Create a Task. This is non-negotiable, This ensures that there's no issues related to the stack frame the task is on being popped before or while the job runs.
     /// @param callable The function to run, *usually a lambda or std::bind*
     /// @note this is shared_ptr so you don't have to delete it.
-    static std::shared_ptr<Task<void>> Create(const std::function<void()>& callable) { return std::shared_ptr<Task<void>>(new Task<void>(callable)); }
+    static std::shared_ptr<Task<void>> Create(const std::function<void()>& callable, const std::function<void()>& complete_callback = nullptr) { return std::shared_ptr<Task<void>>(new Task<void>(callable, complete_callback)); }
+    ~Task() = default;
 };

include/MultiThreading/Thread.h

@@ -1,10 +1,6 @@
 #pragma once
 
 #include <thread>
-#include <functional>
-#include <stdexcept>
-#include <condition_variable>
-#include <mutex>
 #include <MultiThreading/Task.h>
 
 namespace MultiThreading {
@@ -13,22 +9,39 @@ namespace MultiThreading {
 
 class MultiThreading::Thread {
 private:
-    std::thread worker;
     std::shared_ptr<TaskBase> current_task = nullptr;
-    std::atomic<bool> busy = false;
-    std::mutex mtx;
-    std::condition_variable cv;
     std::atomic<bool> stop = false;
+    std::condition_variable cv;
+    std::thread worker;
+    bool busy = false;
+    std::mutex mtx;
 private:
     void Runner();
 public:
-    [[nodiscard]] bool SetTask(const std::function<void()>& task);
+    /// @returns false if the thread is busy.
+    /// @param task The task to run.
+    /// @note Task is passed by value on purpose so that the original shared_ptr to the task is still good.
     [[nodiscard]] bool SetTask(std::shared_ptr<TaskBase> task);
+
+    /// @returns false if the thread is busy.
+    /// @param task The task to run.
+    /// @note Task is passed by reference because the frame after next is going to copy it anyway.
+    [[nodiscard]] bool SetTask(const std::function<void()>& task);
+
+    /// @returns true if the thread is currently doing work.
+    /// @note SetTask will return false if the thread is busy.
     [[nodiscard]] bool Busy() const { return busy; }
-    void Join() { if (worker.joinable()) worker.join(); };
+
+    /// Blocks the thread join is called from until execution of this thread exits.
+    /// @note Joining this thread from this thread will raise an exception.
+    void Join();
 public:
+    /// Create a thread which will initialize and then wait for a task.
     Thread() { worker = std::thread([this] { this->Runner(); }); }
+    /// Crete a thread which will immediately run a task and then wait for another.
     explicit Thread(std::shared_ptr<TaskBase> task);
+    /// Crete a thread which will immediately run a task and then wait for another.
     explicit Thread(const std::function<void()>& task);
+    /// Waits for the current task to finish (if there is one) and destroys the thread.
     ~Thread();
 };

include/MultiThreading/ThreadPool.h

@@ -3,27 +3,51 @@
 #include <MultiThreading/Thread.h>
 #include <vector>
 #include <queue>
+
 namespace MultiThreading {
     class ThreadPool;
 }
 
+/// A group of threads to run tasks on.
 class MultiThreading::ThreadPool {
 private:
     std::vector<MultiThreading::Thread*> threads;
     std::queue<std::shared_ptr<TaskBase>> queue;
+    std::condition_variable queue_condition;
     std::mutex queue_mutex;
 private:
     /// @returns nullptr if the queue is empty.
     std::shared_ptr<TaskBase> Dequeue();
     void Runner(const std::shared_ptr<TaskBase>& task);
 public:
+    /// Set a task to be run on the thread-pool.
+    /// @param task The task to run.
     void Enqueue(const std::shared_ptr<MultiThreading::TaskBase>& task);
+
+    /// Set a task to be run on the thread-pool.
+    /// @param task The task to run.
     void Enqueue(const std::function<void()>& task);
 public:
+    /// @returns The number of threads in the thread pool.
     [[nodiscard]] unsigned int ThreadCount() const { return threads.size(); }
-    [[nodiscard]] unsigned int QueueSize();
+
+    /// @returns the number of tasks in the queue
+    /// @note this excludes the tasks currently being run.
+    [[nodiscard]] unsigned int PendingTasks();
+
+    /// @returns Whether a task you enqueue would have to wait for something else to finish.
+    [[nodiscard]] bool Busy();
+
+    /// Uses a condition variable to wait the calling thread until the queue is empty.
+    void WaitQueueEmpty();
 public:
-    ThreadPool();
-    explicit ThreadPool(unsigned int thread_count);
+    /// Constructs a thread-pool with a given number of threads (or hardware_concurrency as the default)
+    /// @param thread_count The number of threads.
+    /// @note If you do a *lot* of work on the main thread and hardware_concurrency >= 2, You should use hardware_concurrency -1.
+    explicit ThreadPool(unsigned int thread_count = std::thread::hardware_concurrency());
+
+    /// Waits for the threads to empty the queue and destroys the thread pool.
+    /// @note This should be one of the very last things your program does before exiting.
+    // TODO make Enqueue fail during destruction to avoid stalls.
     ~ThreadPool();
 };

main.cpp

@@ -5,13 +5,15 @@
 #include <iostream>
 
 using namespace MultiThreading;
+
+void cb() { std::cout << "hi" << std::endl; }
 int32_t some_test_func(int32_t hello) {
-    for (unsigned int i = 0; i < 100; i++)
-        std::cout << i << std::endl;
-    return hello;
+    for (unsigned int i = 0; i < 1000000000; i++) {}
+    std::cout << "task " << hello << " finishes." << std::endl;
+    return rand();
 }
 
-
+/*
 int main() {
     // Each task you create can be run by a thread only once. It's marked as complete after.
     // If you're running a lambda or std::function directly on the thread, It can be used multiple times.
@@ -21,8 +23,14 @@ int main() {
     // You can start threads in place like this, but you have to wait for the amount of time
     // it takes for the thread to start up which in many cases is longer than the job. Use thread pool.
     Thread some_thread;
+    //some_thread.SetTaskCompletionCallback(cb);
     some_thread.SetTask(some_task);
 
+    //while (!some_task->Complete()) {}
+
+    //std::this_thread::sleep_for(std::chrono::milliseconds(1));
+    std::cout << some_thread.Busy() << std::endl;
+
     // Some work running concurrently with the worker thread.
     //for (unsigned int i = 0; i < 10; i++) {}
 
@@ -32,26 +40,23 @@ int main() {
 
     //std::cout << a << std::endl;
 }
+*/
 
 
 
-/*
 int main() {
-    ThreadPool thread_pool(1);
+    srand(time(nullptr));
 
-    auto some_task_1 = Task<void>::Create(([] { return some_test_func(1); }));
-    auto some_task_2 = Task<void>::Create(([] { return some_test_func(1); }));
-    auto some_task_3 = Task<void>::Create(([] { return some_test_func(1); }));
-    auto some_task_4 = Task<void>::Create(([] { return some_test_func(1); }));
+    int32_t task_completion_count = 0;
+    auto* thread_pool = new ThreadPool(16);
 
-    thread_pool.Enqueue(some_task_1);
-    thread_pool.Enqueue(some_task_2);
-    thread_pool.Enqueue(some_task_3);
-    thread_pool.Enqueue(some_task_4);
 
-    while (!some_task_4->Complete()) {}
-    std::cout << thread_pool.ThreadCount() << std::endl;
+    for (unsigned int i = 0; i < 128; i++) {
+        auto some_task = Task<int32_t>::Create(([i] { return some_test_func(i); }), cb, &task_completion_count);
+        thread_pool->Enqueue(some_task);
+    }
 
-    //delete thread_pool;
-}
- */
+    /// do stuff after the job.
+    delete thread_pool;
+    std::cout << "The returned random value was: " << task_completion_count << std::endl;
+}

src/Thread.cpp

@@ -1,25 +1,30 @@
 #include <MultiThreading/Thread.h>
-#include <sstream>
-#include <iostream>
 #include <utility>
 
 using namespace MultiThreading;
 
 Thread::~Thread() {
     if (current_task)
-        while (!current_task->Complete()) {}
+        current_task->WaitComplete();
     stop = true;
+
+    cv.notify_one();
     Join();
     // Thread exits gracefully.
 }
 
 bool Thread::SetTask(std::shared_ptr<TaskBase> task) {
-    if (busy)
+    std::lock_guard<std::mutex> lock(mtx);
+
+    if (Busy())
         return false;
-    {
-        std::lock_guard<std::mutex> lock(mtx);
-        current_task = std::move(task);
-    }
+
+    if (current_task)
+        return false;
+
+    busy = true;
+    current_task = std::move(task);
+
     cv.notify_all();
     return true;
 }
@@ -30,17 +35,16 @@ bool Thread::SetTask(const std::function<void()>& task) {
 
 void Thread::Runner() {
     std::unique_lock<std::mutex> lock(mtx);
+
     while (!stop) {
         cv.wait(lock, [this] { return stop || current_task != nullptr; });
 
         if (stop)
             break;
 
-        busy = true;
-
         lock.unlock();
-        if (current_task)
-            current_task->Run();
+
+        current_task->Run();
 
         lock.lock();
 
@@ -62,3 +66,8 @@ Thread::Thread(const std::function<void()>& task) {
     if (!SetTask(task))
         throw std::runtime_error("Thread constructor failure.");
 }
+
+void Thread::Join() {
+    if (worker.joinable())
+        worker.join();
+}

src/ThreadPool.cpp

@@ -3,12 +3,8 @@
 using namespace MultiThreading;
 
 ThreadPool::ThreadPool(unsigned int thread_count) {
-    for (unsigned int i = 0; i < thread_count; i++)
-        threads.push_back(new Thread());
-}
 
-ThreadPool::ThreadPool() {
-    for (unsigned int i = 0; i < std::thread::hardware_concurrency(); i++)
+    for (unsigned int i = 0; i < thread_count; i++)
         threads.push_back(new Thread());
 }
 
@@ -17,12 +13,12 @@ void ThreadPool::Enqueue(const std::shared_ptr<MultiThreading::TaskBase>& task)
 
     // Assign it immediately if there's no wait and a thread open.
     if (queue.empty()) {
-        for (auto *t: threads) {
+        for (auto* t: threads) {
             if (t->Busy())
                 continue;
 
             if (!t->SetTask( [this, task] (){ Runner(task); } ))
-                throw std::runtime_error("There was an error while setting up the task to run on the thread.");
+                throw std::runtime_error("There was a collision while putting the task to the thread.");
             return;
         }
     }
@@ -38,6 +34,10 @@ std::shared_ptr<TaskBase> ThreadPool::Dequeue() {
 
     auto task = queue.front();
     queue.pop();
+
+    if (queue.empty())
+        queue_condition.notify_all();
+
     return task;
 }
 
@@ -46,25 +46,42 @@ void ThreadPool::Enqueue(const std::function<void()>& task) {
 }
 
 void ThreadPool::Runner(const std::shared_ptr<TaskBase>& task) {
-    if (!task->Complete())
-        task->Run();
+    auto running_task = task;
 
-    auto next_task = Dequeue();
-    if (!next_task)
-        return;
-    Runner(next_task);
+    while (running_task) {
+        if (!running_task->Complete())
+            running_task->Run();
+
+        running_task = Dequeue();
+    }
 }
 
-unsigned int ThreadPool::QueueSize() {
+unsigned int ThreadPool::PendingTasks() {
     std::lock_guard<std::mutex> lock(queue_mutex);
     return queue.size();
 }
 
 ThreadPool::~ThreadPool() {
-    // Wait for all tasks to be running.
-    while (QueueSize() != 0) {}
+    // Wait for all queued tasks to be running.
+    WaitQueueEmpty();
 
     // delete t waits for the thread to exit gracefully.
     for (auto* t: threads)
         delete t;
 }
+
+bool ThreadPool::Busy() {
+    if (PendingTasks() != 0)
+        return true;
+
+    bool all_busy = true;
+    for (auto* t : threads)
+        if (!t->Busy()) { all_busy = false; break; }
+
+    return all_busy;
+}
+
+void ThreadPool::WaitQueueEmpty() {
+    std::unique_lock<std::mutex> lock(queue_mutex);
+    queue_condition.wait(lock, [this]{ return queue.empty(); });
+}