include: co/thread.h.


The Thread class is an encapsulation of the thread. When a Thread object is created, the thread will be started. Once the thread is started, it will run until the thread function exits. The Thread class does not provide a method for forcing terminating the thread (something like pthread_cancel). This type of method is usually unsafe and may cause damage to user data.


explicit Thread(co::Closure* cb);

template<typename F>
explicit Thread(F&& f);

template<typename F, typename P>
Thread(F&& f, P&& p);

template<typename F, typename T, typename P>
Thread(F&& f, T* t, P&& p);
  • Constructor, the thread will start after the Thread object is created.

  • In the first version, the parameter cb is a pointer of Closure type. Generally, users are not recommended to use this version directly.

  • For the 2-4th version, a Closure object is constructed according to the passed in parameters, and then the first version is called.

  • In the second version, the parameter f is any runnable object, as long as we can call f() or (*f)().

  • In the third version, the parameter f is any runnable object, as long as we can call f(p), (*f)(p) or (p->*f)().

  • In the fourth version, the parameter f is a method with one parameter in class T, the parameter t is a pointer of type T, and p is the parameter of f.

  • Example

Thread x([](){});                  // lambda
Thread x(f);                       // void f();
Thread x(f, p);                    // void f(void*); void* p;
Thread x(f, 7);                    // void f(int v);
Thread x(&T::f, &t);               // void T::f(); T t;
Thread x(&T::f, &t, 7);            // void T::f(int v); T t;
Thread x(std::bind(&T::f, &t, 7)); // void T::f(int v); T t;


  • Destructor, call the join() method, wait for the thread to exit, and release system resources.


void detach();
  • Let the thread run independently of the Thread object. Once this method is called, the Thread object is useless. When the thread function exits, the system resources are automatically released.

  • Example

void f();
Thread(f).detach(); // run f() in a thread


void join();
  • Calling this method will block until the thread function exits, and then release system resources.
  • If the detach() method has been called before, nothing will happen when calling this method.


uint32 thread_id();
  • This function returns the id of the current thread.
  • In the internal implementation, TLS is used to save the thread id, and each thread only needs one system call.


uint32 current_thread_id();
  • Deprecated since v2.0.2, use co::thread_id() instead.

#Mutex (Mutex)

Mutex is a kind of mutual exclusion lock commonly used in multithreaded programming. At the same time, at most one thread holds the lock, and other threads must wait for the lock to be released.


  • Constructor, allocate system resources and initialize.


  • Destructor, release system resources.


void lock();
  • Acquire the lock, the method will block until the lock is successfully acquired.


bool try_lock();
  • Acquire the lock, will not block. If the lock is successfully acquired, it returns true, otherwise it returns false.


void unlock();
  • Release the lock, this method is generally called by the thread that held the lock.


The MutexGuard class is used to automatically acquire and release the lock in Mutex to prevent users from forgetting to release the lock.


explicit MutexGuard(Mutex& m);
explicit MutexGuard(Mutex* m);
  • The constructor, the parameter m is a reference or pointer of the Mutex. When it is a pointer, m MUST NOT be NULL.
  • Internally, m’s lock() method is called to acquire the lock.


  • Destructor, release the lock acquired in the constructor.

  • Example

Mutex m;
MutexGuard g(m);


SyncEvent is a synchronization mechanism between multiple threads, suitable for the producer-consumer model. The consumer calls the wait() method to wait for the synchronization signal, and the producer calls the signal() method to generate the synchronization signal. SyncEvent supports multiple producers and multiple consumers, but in actual applications, there is usually a single consumer.


explicit SyncEvent(bool manual_reset = false, bool signaled = false);
  • Constructor, the parameter manual_reset indicates whether to manually set the synchronization state to unsynchronized, and the parameter signaled indicates whether the initial state is synchronized.


  • Destructor, release system resources.


void reset();
  • This method sets SyncEvent to unsynchronized state.
  • When manual_reset is true in the constructor, the user needs to manually call this method after wait() returns, otherwise the SyncEvent may remain in a synchronized state forever.


void signal();
  • This method generates a synchronization signal and sets SyncEvent to a synchronized state.


void wait();
bool wait(uint32 ms);
  • The first version will wait until SyncEvent becomes synchronized.
  • The second version will wait until SyncEvent becomes synchronized or timed out. The parameter ms specifies the timeout period in milliseconds.
  • The second version returns true when the SyncEvent becomes synchronized, and returns false when it timed out.
  • When manual_reset is false in the constructor, SyncEvent will be automatically set to unsynchronized when wait() ends.

#Code example

bool manual_reset = false;
SyncEvent ev(manual_reset);

void f1() {
    if (!ev.wait(1000)) {
        LOG << "f1: timedout..";
    } else {
        LOG << "f1: event signaled..";
        if (manual_reset) ev.reset();

void f2() {
    LOG << "f2: send a signal..";


#TLS-based thread_ptr

template <typename T, typename D=std::default_delete<T>>
class thread_ptr;

The thread_ptr class is similar to std::unique_ptr, but uses the TLS mechanism internally, and each thread sets and owns its own ptr.


  • Constructor, allocate system resources and initialize.


  • Destructor, delete the private ptr of each thread, and release TLS related system resources.


T* get() const;
  • Return the ptr of the current thread.
  • If the thread has not set a ptr before, this method returns NULL.


void operator=(T* p);
  • Assignment, set the current thread’s ptr to p, which is equivalent to reset(p).


T* operator->() const;
  • Overload operator->, returns the ptr of the current thread.


T& operator*() const;
  • Overload operator*, returns the reference of the object pointed to by the ptr of the current thread.


bool operator==(T* p) const;
  • Determine whether the current thread’s ptr is equal to p.


bool operator!=(T* p) const;
  • Determine whether the current thread’s ptr is not equal to p.


bool operator!() const;
  • Determine whether the ptr of the current thread is NULL, return true if it is NULL, otherwise return false.

#thread_ptr::operator bool

explicit operator bool() const;
  • Convert thread_ptr to bool type. If the internal pointer is not NULL, return true, otherwise return false.


T* release();
  • Release the ptr of the current thread.
  • This method sets the ptr of the current thread to NULL and returns the previous ptr.


void reset(T* p = 0);
  • Reset the ptr of the current thread to p, p is 0 by default, and the previous ptr will be deleted by D()(x).

#Code example

struct T {
    T(int v): _v(v) {}
    void run() {
        LOG << current_thread_id() << "v:" << _v;
    int _v;

thread_ptr<T> pt;

// Each thread will set its own pointer, and different threads will not affect each other
void f(int v) {
    if (pt == NULL) {
        LOG << "new T(" << v << ")";
        pt.reset(new T(v));

Thread(f, 1).detach(); // start thread 1
Thread(f, 2).detach(); // start thread 2