///////////////////////////////////////////////////////////////////////////// // Name: thread.h // Purpose: interface of all thread-related wxWidgets classes // Author: wxWidgets team // RCS-ID: $Id$ // Licence: wxWindows license ///////////////////////////////////////////////////////////////////////////// /** See wxCondition. */ enum wxCondError { wxCOND_NO_ERROR = 0, wxCOND_INVALID, wxCOND_TIMEOUT, //!< WaitTimeout() has timed out wxCOND_MISC_ERROR }; /** @class wxCondition wxCondition variables correspond to pthread conditions or to Win32 event objects. They may be used in a multithreaded application to wait until the given condition becomes @true which happens when the condition becomes signaled. For example, if a worker thread is doing some long task and another thread has to wait until it is finished, the latter thread will wait on the condition object and the worker thread will signal it on exit (this example is not perfect because in this particular case it would be much better to just wxThread::Wait for the worker thread, but if there are several worker threads it already makes much more sense). Note that a call to wxCondition::Signal may happen before the other thread calls wxCondition::Wait and, just as with the pthread conditions, the signal is then lost and so if you want to be sure that you don't miss it you must keep the mutex associated with the condition initially locked and lock it again before calling wxCondition::Signal. Of course, this means that this call is going to block until wxCondition::Wait is called by another thread. @section condition_example Example This example shows how a main thread may launch a worker thread which starts running and then waits until the main thread signals it to continue: @code class MySignallingThread : public wxThread { public: MySignallingThread(wxMutex *mutex, wxCondition *condition) { m_mutex = mutex; m_condition = condition; Create(); } virtual ExitCode Entry() { ... do our job ... // tell the other(s) thread(s) that we're about to terminate: we must // lock the mutex first or we might signal the condition before the // waiting threads start waiting on it! wxMutexLocker lock(*m_mutex); m_condition->Broadcast(); // same as Signal() here -- one waiter only return 0; } private: wxCondition *m_condition; wxMutex *m_mutex; }; int main() { wxMutex mutex; wxCondition condition(mutex); // the mutex should be initially locked mutex.Lock(); // create and run the thread but notice that it won't be able to // exit (and signal its exit) before we unlock the mutex below MySignallingThread *thread = new MySignallingThread(&mutex, &condition); thread->Run(); // wait for the thread termination: Wait() atomically unlocks the mutex // which allows the thread to continue and starts waiting condition.Wait(); // now we can exit return 0; } @endcode Of course, here it would be much better to simply use a joinable thread and call wxThread::Wait on it, but this example does illustrate the importance of properly locking the mutex when using wxCondition. @library{wxbase} @category{threading} @see wxThread, wxMutex */ class wxCondition { public: /** Default and only constructor. The @a mutex must be locked by the caller before calling Wait() function. Use IsOk() to check if the object was successfully initialized. */ wxCondition(wxMutex& mutex); /** Destroys the wxCondition object. The destructor is not virtual so this class should not be used polymorphically. */ ~wxCondition(); /** Broadcasts to all waiting threads, waking all of them up. Note that this method may be called whether the mutex associated with this condition is locked or not. @see Signal() */ wxCondError Broadcast(); /** Returns @true if the object had been initialized successfully, @false if an error occurred. */ bool IsOk() const; /** Signals the object waking up at most one thread. If several threads are waiting on the same condition, the exact thread which is woken up is undefined. If no threads are waiting, the signal is lost and the condition would have to be signalled again to wake up any thread which may start waiting on it later. Note that this method may be called whether the mutex associated with this condition is locked or not. @see Broadcast() */ wxCondError Signal(); /** Waits until the condition is signalled. This method atomically releases the lock on the mutex associated with this condition (this is why it must be locked prior to calling Wait()) and puts the thread to sleep until Signal() or Broadcast() is called. It then locks the mutex again and returns. Note that even if Signal() had been called before Wait() without waking up any thread, the thread would still wait for another one and so it is important to ensure that the condition will be signalled after Wait() or the thread may sleep forever. @return Returns wxCOND_NO_ERROR on success, another value if an error occurred. @see WaitTimeout() */ wxCondError Wait(); /** Waits until the condition is signalled or the timeout has elapsed. This method is identical to Wait() except that it returns, with the return code of @c wxCOND_TIMEOUT as soon as the given timeout expires. @param milliseconds Timeout in milliseconds @return Returns wxCOND_NO_ERROR if the condition was signalled, wxCOND_TIMEOUT if the timeout elapsed before this happened or another error code from wxCondError enum. */ wxCondError WaitTimeout(unsigned long milliseconds); }; /** @class wxCriticalSectionLocker This is a small helper class to be used with wxCriticalSection objects. A wxCriticalSectionLocker enters the critical section in the constructor and leaves it in the destructor making it much more difficult to forget to leave a critical section (which, in general, will lead to serious and difficult to debug problems). Example of using it: @code void Set Foo() { // gs_critSect is some (global) critical section guarding access to the // object "foo" wxCriticalSectionLocker locker(gs_critSect); if ( ... ) { // do something ... return; } // do something else ... return; } @endcode Without wxCriticalSectionLocker, you would need to remember to manually leave the critical section before each @c return. @library{wxbase} @category{threading} @see wxCriticalSection, wxMutexLocker */ class wxCriticalSectionLocker { public: /** Constructs a wxCriticalSectionLocker object associated with given @a criticalsection and enters it. */ wxCriticalSectionLocker(wxCriticalSection& criticalsection); /** Destructor leaves the critical section. */ ~wxCriticalSectionLocker(); }; /** @class wxThreadHelper The wxThreadHelper class is a mix-in class that manages a single background thread, either detached or joinable (see wxThread for the differences). By deriving from wxThreadHelper, a class can implement the thread code in its own wxThreadHelper::Entry() method and easily share data and synchronization objects between the main thread and the worker thread. Doing this prevents the awkward passing of pointers that is needed when the original object in the main thread needs to synchronize with its worker thread in its own wxThread derived object. For example, wxFrame may need to make some calculations in a background thread and then display the results of those calculations in the main window. Ordinarily, a wxThread derived object would be created with the calculation code implemented in wxThread::Entry. To access the inputs to the calculation, the frame object would often need to pass a pointer to itself to the thread object. Similarly, the frame object would hold a pointer to the thread object. Shared data and synchronization objects could be stored in either object though the object without the data would have to access the data through a pointer. However with wxThreadHelper the frame object and the thread object are treated as the same object. Shared data and synchronization variables are stored in the single object, eliminating a layer of indirection and the associated pointers. Example: @code wxDECLARE_EVENT(wxEVT_COMMAND_MYTHREAD_UPDATE, wxThreadEvent); class MyFrame : public wxFrame, public wxThreadHelper { public: MyFrame(...) { ... } ~MyFrame() { // it's better to do any thread cleanup in the OnClose() // event handler, rather than in the destructor. // This is because the event loop for a top-level window is not // active anymore when its destructor is called and if the thread // sends events when ending, they won't be processed unless // you ended the thread from OnClose. // See @ref overview_windowdeletion for more info. } ... void DoStartALongTask(); void OnThreadUpdate(wxThreadEvent& evt); void OnClose(wxCloseEvent& evt); ... protected: virtual wxThread::ExitCode Entry(); // the output data of the Entry() routine: char m_data[1024]; wxCriticalSection m_dataCS; // protects field above DECLARE_EVENT_TABLE() }; wxDEFINE_EVENT(wxEVT_COMMAND_MYTHREAD_UPDATE, wxThreadEvent) BEGIN_EVENT_TABLE(MyFrame, wxFrame) EVT_COMMAND(wxID_ANY, wxEVT_COMMAND_MYTHREAD_UPDATE, MyFrame::OnThreadUpdate) EVT_CLOSE(MyFrame::OnClose) END_EVENT_TABLE() void MyFrame::DoStartALongTask() { // we want to start a long task, but we don't want our GUI to block // while it's executed, so we use a thread to do it. if (CreateThread(wxTHREAD_JOINABLE) != wxTHREAD_NO_ERROR) { wxLogError("Could not create the worker thread!"); return; } // go! if (GetThread()->Run() != wxTHREAD_NO_ERROR) { wxLogError("Could not run the worker thread!"); return; } } wxThread::ExitCode MyFrame::Entry() { // IMPORTANT: // this function gets executed in the secondary thread context! int offset = 0; // here we do our long task, periodically calling TestDestroy(): while (!GetThread()->TestDestroy()) { // since this Entry() is implemented in MyFrame context we don't // need any pointer to access the m_data, m_processedData, m_dataCS // variables... very nice! // this is an example of the generic structure of a download thread: char buffer[1024]; download_chunk(buffer, 1024); // this takes time... { // ensure noone reads m_data while we write it wxCriticalSectionLocker lock(m_dataCS); memcpy(m_data+offset, buffer, 1024); offset += 1024; } // VERY IMPORTANT: do not call any GUI function inside this // function; rather use wxQueueEvent(): wxQueueEvent(this, new wxThreadEvent(wxEVT_COMMAND_MYTHREAD_UPDATE)); // we used pointer 'this' assuming it's safe; see OnClose() } // TestDestroy() returned true (which means the main thread asked us // to terminate as soon as possible) or we ended the long task... return (wxThread::ExitCode)0; } void MyFrame::OnClose(wxCloseEvent&) { // important: before terminating, we _must_ wait for our joinable // thread to end, if it's running; in fact it uses variables of this // instance and posts events to *this event handler if (GetThread() && // DoStartALongTask() may have not been called GetThread()->IsRunning()) GetThread()->Wait(); Destroy(); } void MyFrame::OnThreadUpdate(wxThreadEvent& evt) { // ...do something... e.g. m_pGauge->Pulse(); // read some parts of m_data just for fun: wxCriticalSectionLocker lock(m_dataCS); wxPrintf("%c", m_data[100]); } @endcode @library{wxbase} @category{threading} @see wxThread, wxThreadEvent */ class wxThreadHelper { public: /** This constructor simply initializes internal member variables and tells wxThreadHelper which type the thread internally managed should be. */ wxThreadHelper(wxThreadKind kind = wxTHREAD_JOINABLE); /** The destructor frees the resources associated with the thread, forcing it to terminate (it uses wxThread::Kill function). Because of the wxThread::Kill unsafety, you should always wait (with wxThread::Wait) for joinable threads to end or call wxThread::Delete on detached threads, instead of relying on this destructor for stopping the thread. */ virtual ~wxThreadHelper(); /** This is the entry point of the thread. This function is pure virtual and must be implemented by any derived class. The thread execution will start here. You'll typically want your Entry() to look like: @code wxThread::ExitCode Entry() { while (!GetThread()->TestDestroy()) { // ... do some work ... if (IsWorkCompleted) break; if (HappenedStoppingError) return (wxThread::ExitCode)1; // failure } return (wxThread::ExitCode)0; // success } @endcode The returned value is the thread exit code which is only useful for joinable threads and is the value returned by @c "GetThread()->Wait()". This function is called by wxWidgets itself and should never be called directly. */ virtual ExitCode Entry() = 0; /** @deprecated Use CreateThread() instead. */ wxThreadError Create(unsigned int stackSize = 0); /** Creates a new thread of the given @a kind. The thread object is created in the suspended state, and you should call @ref wxThread::Run "GetThread()->Run()" to start running it. You may optionally specify the stack size to be allocated to it (ignored on platforms that don't support setting it explicitly, e.g. Unix). @return One of the ::wxThreadError enum values. */ wxThreadError CreateThread(wxThreadKind kind = wxTHREAD_JOINABLE, unsigned int stackSize = 0); /** This is a public function that returns the wxThread object associated with the thread. */ wxThread* GetThread() const; /** Returns the last type of thread given to the CreateThread() function or to the constructor. */ wxThreadKind GetThreadKind() const; }; /** Possible critical section types */ enum wxCriticalSectionType { wxCRITSEC_DEFAULT, /** Recursive critical section under both Windows and Unix */ wxCRITSEC_NON_RECURSIVE /** Non-recursive critical section under Unix, recursive under Windows */ }; /** @class wxCriticalSection A critical section object is used for exactly the same purpose as a wxMutex. The only difference is that under Windows platform critical sections are only visible inside one process, while mutexes may be shared among processes, so using critical sections is slightly more efficient. The terminology is also slightly different: mutex may be locked (or acquired) and unlocked (or released) while critical section is entered and left by the program. Finally, you should try to use wxCriticalSectionLocker class whenever possible instead of directly using wxCriticalSection for the same reasons wxMutexLocker is preferrable to wxMutex - please see wxMutex for an example. @library{wxbase} @category{threading} @see wxThread, wxCondition, wxCriticalSectionLocker */ class wxCriticalSection { public: /** Default constructor initializes critical section object. By default critical sections are recursive under Unix and Windows. */ wxCriticalSection( wxCriticalSectionType critSecType = wxCRITSEC_DEFAULT ); /** Destructor frees the resources. */ ~wxCriticalSection(); /** Enter the critical section (same as locking a mutex): if another thread has already entered it, this call will block until the other thread calls Leave(). There is no error return for this function. After entering the critical section protecting a data variable, the thread running inside the critical section may safely use/modify it. Note that entering the same critical section twice or more from the same thread doesn't result in a deadlock; in this case in fact this function will immediately return. */ void Enter(); /** Leave the critical section allowing other threads use the global data protected by it. There is no error return for this function. */ void Leave(); }; /** The possible thread kinds. */ enum wxThreadKind { /** Detached thread */ wxTHREAD_DETACHED, /** Joinable thread */ wxTHREAD_JOINABLE }; /** The possible thread errors. */ enum wxThreadError { /** No error */ wxTHREAD_NO_ERROR = 0, /** No resource left to create a new thread. */ wxTHREAD_NO_RESOURCE, /** The thread is already running. */ wxTHREAD_RUNNING, /** The thread isn't running. */ wxTHREAD_NOT_RUNNING, /** Thread we waited for had to be killed. */ wxTHREAD_KILLED, /** Some other error */ wxTHREAD_MISC_ERROR }; /** Defines the interval of priority */ enum { WXTHREAD_MIN_PRIORITY = 0u, WXTHREAD_DEFAULT_PRIORITY = 50u, WXTHREAD_MAX_PRIORITY = 100u }; /** @class wxThread A thread is basically a path of execution through a program. Threads are sometimes called @e light-weight processes, but the fundamental difference between threads and processes is that memory spaces of different processes are separated while all threads share the same address space. While it makes it much easier to share common data between several threads, it also makes it much easier to shoot oneself in the foot, so careful use of synchronization objects such as mutexes (see wxMutex) or critical sections (see wxCriticalSection) is recommended. In addition, don't create global thread objects because they allocate memory in their constructor, which will cause problems for the memory checking system. @section thread_types Types of wxThreads There are two types of threads in wxWidgets: @e detached and @e joinable, modeled after the the POSIX thread API. This is different from the Win32 API where all threads are joinable. By default wxThreads in wxWidgets use the @b detached behavior. Detached threads delete themselves once they have completed, either by themselves when they complete processing or through a call to Delete(), and thus @b must be created on the heap (through the new operator, for example). Typically you'll want to store the instances of the detached wxThreads you allocate, so that you can call functions on them. Because of their nature however you'll need to always use a critical section when accessing them: @code // declare a new type of event, to be used by our MyThread class: wxDECLARE_EVENT(wxEVT_COMMAND_MYTHREAD_COMPLETED, wxThreadEvent); wxDECLARE_EVENT(wxEVT_COMMAND_MYTHREAD_UPDATE, wxThreadEvent); class MyFrame; class MyThread : public wxThread { public: MyThread(MyFrame *handler) : wxThread(wxTHREAD_DETACHED) { m_pHandler = handler } ~MyThread(); protected: virtual ExitCode Entry(); MyFrame *m_pHandler; }; class MyFrame : public wxFrame { public: ... ~MyFrame() { // it's better to do any thread cleanup in the OnClose() // event handler, rather than in the destructor. // This is because the event loop for a top-level window is not // active anymore when its destructor is called and if the thread // sends events when ending, they won't be processed unless // you ended the thread from OnClose. // See @ref overview_windowdeletion for more info. } ... void DoStartThread(); void DoPauseThread(); // a resume routine would be nearly identic to DoPauseThread() void DoResumeThread() { ... } void OnThreadUpdate(wxThreadEvent&); void OnThreadCompletion(wxThreadEvent&); void OnClose(wxCloseEvent&); protected: MyThread *m_pThread; wxCriticalSection m_pThreadCS; // protects the m_pThread pointer DECLARE_EVENT_TABLE() }; BEGIN_EVENT_TABLE(MyFrame, wxFrame) EVT_CLOSE(MyFrame::OnClose) EVT_MENU(Minimal_Start, MyFrame::DoStartThread) EVT_COMMAND(wxID_ANY, wxEVT_COMMAND_MYTHREAD_UPDATE, MyFrame::OnThreadUpdate) EVT_COMMAND(wxID_ANY, wxEVT_COMMAND_MYTHREAD_COMPLETED, MyFrame::OnThreadCompletion) END_EVENT_TABLE() wxDEFINE_EVENT(wxEVT_COMMAND_MYTHREAD_COMPLETED, wxThreadEvent) wxDEFINE_EVENT(wxEVT_COMMAND_MYTHREAD_UPDATE, wxThreadEvent) void MyFrame::DoStartThread() { m_pThread = new MyThread(this); if ( m_pThread->Create() != wxTHREAD_NO_ERROR ) { wxLogError("Can't create the thread!"); delete m_pThread; m_pThread = NULL; } else { if (m_pThread->Run() != wxTHREAD_NO_ERROR ) { wxLogError("Can't create the thread!"); delete m_pThread; m_pThread = NULL; } // after the call to wxThread::Run(), the m_pThread pointer is "unsafe": // at any moment the thread may cease to exist (because it completes its work). // To avoid dangling pointers OnThreadExit() will set m_pThread // to NULL when the thread dies. } } wxThread::ExitCode MyThread::Entry() { while (!TestDestroy()) { // ... do a bit of work... wxQueueEvent(m_pHandler, new wxThreadEvent(wxEVT_COMMAND_MYTHREAD_UPDATE)); } // signal the event handler that this thread is going to be destroyed // NOTE: here we assume that using the m_pHandler pointer is safe, // (in this case this is assured by the MyFrame destructor) wxQueueEvent(m_pHandler, new wxThreadEvent(wxEVT_COMMAND_MYTHREAD_COMPLETED)); return (wxThread::ExitCode)0; // success } MyThread::~MyThread() { wxCriticalSectionLocker enter(m_pHandler->m_pThreadCS); // the thread is being destroyed; make sure not to leave dangling pointers around m_pHandler->m_pThread = NULL; } void MyFrame::OnThreadCompletion(wxThreadEvent&) { wxMessageOutputDebug().Printf("MYFRAME: MyThread exited!\n"); } void MyFrame::OnThreadUpdate(wxThreadEvent&) { wxMessageOutputDebug().Printf("MYFRAME: MyThread update...\n"); } void MyFrame::DoPauseThread() { // anytime we access the m_pThread pointer we must ensure that it won't // be modified in the meanwhile; since only a single thread may be // inside a given critical section at a given time, the following code // is safe: wxCriticalSectionLocker enter(m_pThreadCS); if (m_pThread) // does the thread still exist? { // without a critical section, once reached this point it may happen // that the OS scheduler gives control to the MyThread::Entry() function, // which in turn may return (because it completes its work) making // invalid the m_pThread pointer if (m_pThread->Pause() != wxTHREAD_NO_ERROR ) wxLogError("Can't pause the thread!"); } } void MyFrame::OnClose(wxCloseEvent&) { { wxCriticalSectionLocker enter(m_pThreadCS); if (m_pThread) // does the thread still exist? { m_out.Printf("MYFRAME: deleting thread"); if (m_pThread->Delete() != wxTHREAD_NO_ERROR ) wxLogError("Can't delete the thread!"); } } // exit from the critical section to give the thread // the possibility to enter its destructor // (which is guarded with m_pThreadCS critical section!) while (1) { { // was the ~MyThread() function executed? wxCriticalSectionLocker enter(m_pThreadCS); if (!m_pThread) break; } // wait for thread completion wxThread::This()->Sleep(1); } Destroy(); } @endcode For a more detailed and comprehensive example, see @sample{thread}. For a simpler way to share data and synchronization objects between the main and the secondary thread see wxThreadHelper. Conversely, @b joinable threads do not delete themselves when they are done processing and as such are safe to create on the stack. Joinable threads also provide the ability for one to get value it returned from Entry() through Wait(). You shouldn't hurry to create all the threads joinable, however, because this has a disadvantage as well: you @b must Wait() for a joinable thread or the system resources used by it will never be freed, and you also must delete the corresponding wxThread object yourself if you did not create it on the stack. In contrast, detached threads are of the "fire-and-forget" kind: you only have to start a detached thread and it will terminate and destroy itself. @section thread_deletion wxThread Deletion Regardless of whether it has terminated or not, you should call Wait() on a @b joinable thread to release its memory, as outlined in @ref thread_types. If you created a joinable thread on the heap, remember to delete it manually with the @c delete operator or similar means as only detached threads handle this type of memory management. Since @b detached threads delete themselves when they are finished processing, you should take care when calling a routine on one. If you are certain the thread is still running and would like to end it, you may call Delete() to gracefully end it (which implies that the thread will be deleted after that call to Delete()). It should be implied that you should @b never attempt to delete a detached thread with the @c delete operator or similar means. As mentioned, Wait() or Delete() functions attempt to gracefully terminate a joinable and a detached thread, respectively. They do this by waiting until the thread in question calls TestDestroy() or ends processing (i.e. returns from wxThread::Entry). Obviously, if the thread does call TestDestroy() and does not end, the thread which called Wait() or Delete() will come to halt. This is why it's important to call TestDestroy() in the Entry() routine of your threads as often as possible and immediately exit when it returns @true. As a last resort you can end the thread immediately through Kill(). It is strongly recommended that you do not do this, however, as it does not free the resources associated with the object (although the wxThread object of detached threads will still be deleted) and could leave the C runtime library in an undefined state. @section thread_secondary wxWidgets Calls in Secondary Threads All threads other than the "main application thread" (the one running wxApp::OnInit() or the one your main function runs in, for example) are considered "secondary threads". These include all threads created by Create() or the corresponding constructors. GUI calls, such as those to a wxWindow or wxBitmap are explicitly not safe at all in secondary threads and could end your application prematurely. This is due to several reasons, including the underlying native API and the fact that wxThread does not run a GUI event loop similar to other APIs as MFC. A workaround for some wxWidgets ports is calling wxMutexGUIEnter() before any GUI calls and then calling wxMutexGUILeave() afterwords. However, the recommended way is to simply process the GUI calls in the main thread through an event that is posted by wxQueueEvent(). This does not imply that calls to these classes are thread-safe, however, as most wxWidgets classes are not thread-safe, including wxString. @section thread_poll Don't Poll a wxThread A common problem users experience with wxThread is that in their main thread they will check the thread every now and then to see if it has ended through IsRunning(), only to find that their application has run into problems because the thread is using the default behavior (i.e. it's @b detached) and has already deleted itself. Naturally, they instead attempt to use joinable threads in place of the previous behavior. However, polling a wxThread for when it has ended is in general a bad idea - in fact calling a routine on any running wxThread should be avoided if possible. Instead, find a way to notify yourself when the thread has ended. Usually you only need to notify the main thread, in which case you can post an event to it via wxQueueEvent(). In the case of secondary threads you can call a routine of another class when the thread is about to complete processing and/or set the value of a variable, possibly using mutexes (see wxMutex) and/or other synchronization means if necessary. @library{wxbase} @category{threading} @see wxThreadHelper, wxMutex, wxCondition, wxCriticalSection, @ref overview_thread */ class wxThread { public: /** The return type for the thread functions. */ typedef void* ExitCode; /** This constructor creates a new detached (default) or joinable C++ thread object. It does not create or start execution of the real thread - for this you should use the Create() and Run() methods. The possible values for @a kind parameters are: - @b wxTHREAD_DETACHED - Creates a detached thread. - @b wxTHREAD_JOINABLE - Creates a joinable thread. */ wxThread(wxThreadKind kind = wxTHREAD_DETACHED); /** The destructor frees the resources associated with the thread. Notice that you should never delete a detached thread -- you may only call Delete() on it or wait until it terminates (and auto destructs) itself. Because the detached threads delete themselves, they can only be allocated on the heap. Joinable threads should be deleted explicitly. The Delete() and Kill() functions will not delete the C++ thread object. It is also safe to allocate them on stack. */ virtual ~wxThread(); /** Creates a new thread. The thread object is created in the suspended state, and you should call Run() to start running it. You may optionally specify the stack size to be allocated to it (Ignored on platforms that don't support setting it explicitly, eg. Unix system without @c pthread_attr_setstacksize). If you do not specify the stack size,the system's default value is used. @warning It is a good idea to explicitly specify a value as systems' default values vary from just a couple of KB on some systems (BSD and OS/2 systems) to one or several MB (Windows, Solaris, Linux). So, if you have a thread that requires more than just a few KB of memory, you will have mysterious problems on some platforms but not on the common ones. On the other hand, just indicating a large stack size by default will give you performance issues on those systems with small default stack since those typically use fully committed memory for the stack. On the contrary, if you use a lot of threads (say several hundred), virtual adress space can get tight unless you explicitly specify a smaller amount of thread stack space for each thread. @return One of: - @b wxTHREAD_NO_ERROR - No error. - @b wxTHREAD_NO_RESOURCE - There were insufficient resources to create the thread. - @b wxTHREAD_NO_RUNNING - The thread is already running */ wxThreadError Create(unsigned int stackSize = 0); /** Calling Delete() gracefully terminates a @b detached thread, either when the thread calls TestDestroy() or when it finishes processing. @note This function works on a joinable thread but in that case makes the TestDestroy() function of the thread return @true and then waits for its completion (i.e. it differs from Wait() because it asks the thread to terminate before waiting). See @ref thread_deletion for a broader explanation of this routine. */ wxThreadError Delete(void** rc = NULL); /** Returns the number of system CPUs or -1 if the value is unknown. For multi-core systems the returned value is typically the total number of @e cores, since the OS usually abstract a single N-core CPU as N different cores. @see SetConcurrency() */ static int GetCPUCount(); /** Returns the platform specific thread ID of the current thread as a long. This can be used to uniquely identify threads, even if they are not wxThreads. */ static wxThreadIdType GetCurrentId(); /** Gets the thread identifier: this is a platform dependent number that uniquely identifies the thread throughout the system during its existence (i.e. the thread identifiers may be reused). */ wxThreadIdType GetId() const; /** Returns the thread kind as it was given in the ctor. @since 2.9.0 */ wxThreadKind GetKind() const; /** Gets the priority of the thread, between zero and 100. The following priorities are defined: - @b WXTHREAD_MIN_PRIORITY: 0 - @b WXTHREAD_DEFAULT_PRIORITY: 50 - @b WXTHREAD_MAX_PRIORITY: 100 */ unsigned int GetPriority() const; /** Returns @true if the thread is alive (i.e. started and not terminating). Note that this function can only safely be used with joinable threads, not detached ones as the latter delete themselves and so when the real thread is no longer alive, it is not possible to call this function because the wxThread object no longer exists. */ bool IsAlive() const; /** Returns @true if the thread is of the detached kind, @false if it is a joinable one. */ bool IsDetached() const; /** Returns @true if the calling thread is the main application thread. */ static bool IsMain(); /** Returns @true if the thread is paused. */ bool IsPaused() const; /** Returns @true if the thread is running. This method may only be safely used for joinable threads, see the remark in IsAlive(). */ bool IsRunning() const; /** Immediately terminates the target thread. @b "This function is dangerous and should be used with extreme care" (and not used at all whenever possible)! The resources allocated to the thread will not be freed and the state of the C runtime library may become inconsistent. Use Delete() for detached threads or Wait() for joinable threads instead. For detached threads Kill() will also delete the associated C++ object. However this will not happen for joinable threads and this means that you will still have to delete the wxThread object yourself to avoid memory leaks. In neither case OnExit() of the dying thread will be called, so no thread-specific cleanup will be performed. This function can only be called from another thread context, i.e. a thread cannot kill itself. It is also an error to call this function for a thread which is not running or paused (in the latter case, the thread will be resumed first) -- if you do it, a @b wxTHREAD_NOT_RUNNING error will be returned. */ wxThreadError Kill(); /** Suspends the thread. Under some implementations (Win32), the thread is suspended immediately, under others it will only be suspended when it calls TestDestroy() for the next time (hence, if the thread doesn't call it at all, it won't be suspended). This function can only be called from another thread context. */ wxThreadError Pause(); /** Resumes a thread suspended by the call to Pause(). This function can only be called from another thread context. */ wxThreadError Resume(); /** Starts the thread execution. Should be called after Create(). Note that once you Run() a @b detached thread, @e any function call you do on the thread pointer (you must allocate it on the heap) is @e "unsafe"; i.e. the thread may have terminated at any moment after Run() and your pointer may be dangling. See @ref thread_types for an example of safe manipulation of detached threads. This function can only be called from another thread context. */ wxThreadError Run(); /** Sets the thread concurrency level for this process. This is, roughly, the number of threads that the system tries to schedule to run in parallel. The value of 0 for @a level may be used to set the default one. @return @true on success or @false otherwise (for example, if this function is not implemented for this platform -- currently everything except Solaris). */ static bool SetConcurrency(size_t level); /** Sets the priority of the thread, between 0 and 100. It can only be set after calling Create() but before calling Run(). The following priorities are defined: - @b WXTHREAD_MIN_PRIORITY: 0 - @b WXTHREAD_DEFAULT_PRIORITY: 50 - @b WXTHREAD_MAX_PRIORITY: 100 */ void SetPriority(unsigned int priority); /** Pauses the thread execution for the given amount of time. This is the same as wxMilliSleep(). */ static void Sleep(unsigned long milliseconds); /** This function should be called periodically by the thread to ensure that calls to Pause() and Delete() will work. If it returns @true, the thread should exit as soon as possible. Notice that under some platforms (POSIX), implementation of Pause() also relies on this function being called, so not calling it would prevent both stopping and suspending thread from working. */ virtual bool TestDestroy(); /** Return the thread object for the calling thread. @NULL is returned if the calling thread is the main (GUI) thread, but IsMain() should be used to test whether the thread is really the main one because @NULL may also be returned for the thread not created with wxThread class. Generally speaking, the return value for such a thread is undefined. */ static wxThread* This(); /** Waits for a @b joinable thread to terminate and returns the value the thread returned from Entry() or @c "(ExitCode)-1" on error. Notice that, unlike Delete(), this function doesn't cancel the thread in any way so the caller waits for as long as it takes to the thread to exit. You can only Wait() for @b joinable (not detached) threads. This function can only be called from another thread context. See @ref thread_deletion for a broader explanation of this routine. */ ExitCode Wait(); /** Give the rest of the thread's time-slice to the system allowing the other threads to run. Note that using this function is @b strongly discouraged, since in many cases it indicates a design weakness of your threading model (as does using Sleep() functions). Threads should use the CPU in an efficient manner, i.e. they should do their current work efficiently, then as soon as the work is done block on a wakeup event (wxCondition, wxMutex, select(), poll(), ...) which will get signalled e.g. by other threads or a user device once further thread work is available. Using Yield() or Sleep() indicates polling-type behaviour, since we're fuzzily giving up our timeslice and wait until sometime later we'll get reactivated, at which time we realize that there isn't really much to do and Yield() again... The most critical characteristic of Yield() is that it's operating system specific: there may be scheduler changes which cause your thread to not wake up relatively soon again, but instead many seconds later, causing huge performance issues for your application. With a well-behaving, CPU-efficient thread the operating system is likely to properly care for its reactivation the moment it needs it, whereas with non-deterministic, Yield-using threads all bets are off and the system scheduler is free to penalize them drastically, and this effect gets worse with increasing system load due to less free CPU resources available. You may refer to various Linux kernel @c sched_yield discussions for more information. See also Sleep(). */ static void Yield(); protected: /** This is the entry point of the thread. This function is pure virtual and must be implemented by any derived class. The thread execution will start here. The returned value is the thread exit code which is only useful for joinable threads and is the value returned by Wait(). This function is called by wxWidgets itself and should never be called directly. */ virtual ExitCode Entry() = 0; /** This is a protected function of the wxThread class and thus can only be called from a derived class. It also can only be called in the context of this thread, i.e. a thread can only exit from itself, not from another thread. This function will terminate the OS thread (i.e. stop the associated path of execution) and also delete the associated C++ object for detached threads. OnExit() will be called just before exiting. */ void Exit(ExitCode exitcode = 0); private: /** Called when the thread exits. This function is called in the context of the thread associated with the wxThread object, not in the context of the main thread. This function will not be called if the thread was @ref Kill() killed. This function should never be called directly. */ virtual void OnExit(); }; /** See wxSemaphore. */ enum wxSemaError { wxSEMA_NO_ERROR = 0, wxSEMA_INVALID, //!< semaphore hasn't been initialized successfully wxSEMA_BUSY, //!< returned by TryWait() if Wait() would block wxSEMA_TIMEOUT, //!< returned by WaitTimeout() wxSEMA_OVERFLOW, //!< Post() would increase counter past the max wxSEMA_MISC_ERROR }; /** @class wxSemaphore wxSemaphore is a counter limiting the number of threads concurrently accessing a shared resource. This counter is always between 0 and the maximum value specified during the semaphore creation. When the counter is strictly greater than 0, a call to wxSemaphore::Wait() returns immediately and decrements the counter. As soon as it reaches 0, any subsequent calls to wxSemaphore::Wait block and only return when the semaphore counter becomes strictly positive again as the result of calling wxSemaphore::Post which increments the counter. In general, semaphores are useful to restrict access to a shared resource which can only be accessed by some fixed number of clients at the same time. For example, when modeling a hotel reservation system a semaphore with the counter equal to the total number of available rooms could be created. Each time a room is reserved, the semaphore should be acquired by calling wxSemaphore::Wait and each time a room is freed it should be released by calling wxSemaphore::Post. @library{wxbase} @category{threading} */ class wxSemaphore { public: /** Specifying a @a maxcount of 0 actually makes wxSemaphore behave as if there is no upper limit. If @a maxcount is 1, the semaphore behaves almost as a mutex (but unlike a mutex it can be released by a thread different from the one which acquired it). @a initialcount is the initial value of the semaphore which must be between 0 and @a maxcount (if it is not set to 0). */ wxSemaphore(int initialcount = 0, int maxcount = 0); /** Destructor is not virtual, don't use this class polymorphically. */ ~wxSemaphore(); /** Increments the semaphore count and signals one of the waiting threads in an atomic way. Returns @e wxSEMA_OVERFLOW if the count would increase the counter past the maximum. @return One of: - wxSEMA_NO_ERROR: There was no error. - wxSEMA_INVALID : Semaphore hasn't been initialized successfully. - wxSEMA_OVERFLOW: Post() would increase counter past the max. - wxSEMA_MISC_ERROR: Miscellaneous error. */ wxSemaError Post(); /** Same as Wait(), but returns immediately. @return One of: - wxSEMA_NO_ERROR: There was no error. - wxSEMA_INVALID: Semaphore hasn't been initialized successfully. - wxSEMA_BUSY: Returned by TryWait() if Wait() would block, i.e. the count is zero. - wxSEMA_MISC_ERROR: Miscellaneous error. */ wxSemaError TryWait(); /** Wait indefinitely until the semaphore count becomes strictly positive and then decrement it and return. @return One of: - wxSEMA_NO_ERROR: There was no error. - wxSEMA_INVALID: Semaphore hasn't been initialized successfully. - wxSEMA_MISC_ERROR: Miscellaneous error. */ wxSemaError Wait(); /** Same as Wait(), but with a timeout limit. @return One of: - wxSEMA_NO_ERROR: There was no error. - wxSEMA_INVALID: Semaphore hasn't been initialized successfully. - wxSEMA_TIMEOUT: Timeout occurred without receiving semaphore. - wxSEMA_MISC_ERROR: Miscellaneous error. */ wxSemaError WaitTimeout(unsigned long timeout_millis); }; /** @class wxMutexLocker This is a small helper class to be used with wxMutex objects. A wxMutexLocker acquires a mutex lock in the constructor and releases (or unlocks) the mutex in the destructor making it much more difficult to forget to release a mutex (which, in general, will promptly lead to serious problems). See wxMutex for an example of wxMutexLocker usage. @library{wxbase} @category{threading} @see wxMutex, wxCriticalSectionLocker */ class wxMutexLocker { public: /** Constructs a wxMutexLocker object associated with mutex and locks it. Call IsOk() to check if the mutex was successfully locked. */ wxMutexLocker(wxMutex& mutex); /** Destructor releases the mutex if it was successfully acquired in the ctor. */ ~wxMutexLocker(); /** Returns @true if mutex was acquired in the constructor, @false otherwise. */ bool IsOk() const; }; /** The possible wxMutex kinds. */ enum wxMutexType { /** Normal non-recursive mutex: try to always use this one. */ wxMUTEX_DEFAULT, /** Recursive mutex: don't use these ones with wxCondition. */ wxMUTEX_RECURSIVE }; /** The possible wxMutex errors. */ enum wxMutexError { /** The operation completed successfully. */ wxMUTEX_NO_ERROR = 0, /** The mutex hasn't been initialized. */ wxMUTEX_INVALID, /** The mutex is already locked by the calling thread. */ wxMUTEX_DEAD_LOCK, /** The mutex is already locked by another thread. */ wxMUTEX_BUSY, /** An attempt to unlock a mutex which is not locked. */ wxMUTEX_UNLOCKED, /** wxMutex::LockTimeout() has timed out. */ wxMUTEX_TIMEOUT, /** Any other error */ wxMUTEX_MISC_ERROR }; /** @class wxMutex A mutex object is a synchronization object whose state is set to signaled when it is not owned by any thread, and nonsignaled when it is owned. Its name comes from its usefulness in coordinating mutually-exclusive access to a shared resource as only one thread at a time can own a mutex object. Mutexes may be recursive in the sense that a thread can lock a mutex which it had already locked before (instead of dead locking the entire process in this situation by starting to wait on a mutex which will never be released while the thread is waiting) but using them is not recommended under Unix and they are @b not recursive by default. The reason for this is that recursive mutexes are not supported by all Unix flavours and, worse, they cannot be used with wxCondition. For example, when several threads use the data stored in the linked list, modifications to the list should only be allowed to one thread at a time because during a new node addition the list integrity is temporarily broken (this is also called @e program @e invariant). @code // this variable has an "s_" prefix because it is static: seeing an "s_" in // a multithreaded program is in general a good sign that you should use a // mutex (or a critical section) static wxMutex *s_mutexProtectingTheGlobalData; // we store some numbers in this global array which is presumably used by // several threads simultaneously wxArrayInt s_data; void MyThread::AddNewNode(int num) { // ensure that no other thread accesses the list s_mutexProtectingTheGlobalList->Lock(); s_data.Add(num); s_mutexProtectingTheGlobalList->Unlock(); } // return true if the given number is greater than all array elements bool MyThread::IsGreater(int num) { // before using the list we must acquire the mutex wxMutexLocker lock(s_mutexProtectingTheGlobalData); size_t count = s_data.Count(); for ( size_t n = 0; n < count; n++ ) { if ( s_data[n] > num ) return false; } return true; } @endcode Notice how wxMutexLocker was used in the second function to ensure that the mutex is unlocked in any case: whether the function returns true or false (because the destructor of the local object @e lock is always called). Using this class instead of directly using wxMutex is, in general, safer and is even more so if your program uses C++ exceptions. @library{wxbase} @category{threading} @see wxThread, wxCondition, wxMutexLocker, wxCriticalSection */ class wxMutex { public: /** Default constructor. */ wxMutex(wxMutexType type = wxMUTEX_DEFAULT); /** Destroys the wxMutex object. */ ~wxMutex(); /** Locks the mutex object. This is equivalent to LockTimeout() with infinite timeout. Note that if this mutex is already locked by the caller thread, this function doesn't block but rather immediately returns. @return One of: @c wxMUTEX_NO_ERROR, @c wxMUTEX_DEAD_LOCK. */ wxMutexError Lock(); /** Try to lock the mutex object during the specified time interval. @return One of: @c wxMUTEX_NO_ERROR, @c wxMUTEX_DEAD_LOCK, @c wxMUTEX_TIMEOUT. */ wxMutexError LockTimeout(unsigned long msec); /** Tries to lock the mutex object. If it can't, returns immediately with an error. @return One of: @c wxMUTEX_NO_ERROR, @c wxMUTEX_BUSY. */ wxMutexError TryLock(); /** Unlocks the mutex object. @return One of: @c wxMUTEX_NO_ERROR, @c wxMUTEX_UNLOCKED. */ wxMutexError Unlock(); }; // ============================================================================ // Global functions/macros // ============================================================================ /** @addtogroup group_funcmacro_thread */ //@{ /** This macro declares a (static) critical section object named @a cs if @c wxUSE_THREADS is 1 and does nothing if it is 0. @header{wx/thread.h} */ #define wxCRIT_SECT_DECLARE(cs) /** This macro declares a critical section object named @a cs if @c wxUSE_THREADS is 1 and does nothing if it is 0. As it doesn't include the @c static keyword (unlike wxCRIT_SECT_DECLARE()), it can be used to declare a class or struct member which explains its name. @header{wx/thread.h} */ #define wxCRIT_SECT_DECLARE_MEMBER(cs) /** This macro creates a wxCriticalSectionLocker named @a name and associated with the critical section @a cs if @c wxUSE_THREADS is 1 and does nothing if it is 0. @header{wx/thread.h} */ #define wxCRIT_SECT_LOCKER(name, cs) /** This macro combines wxCRIT_SECT_DECLARE() and wxCRIT_SECT_LOCKER(): it creates a static critical section object and also the lock object associated with it. Because of this, it can be only used inside a function, not at global scope. For example: @code int IncCount() { static int s_counter = 0; wxCRITICAL_SECTION(counter); return ++s_counter; } @endcode Note that this example assumes that the function is called the first time from the main thread so that the critical section object is initialized correctly by the time other threads start calling it, if this is not the case this approach can @b not be used and the critical section must be made a global instead. @header{wx/thread.h} */ #define wxCRITICAL_SECTION(name) /** This macro is equivalent to @ref wxCriticalSection::Leave "critical_section.Leave()" if @c wxUSE_THREADS is 1 and does nothing if it is 0. @header{wx/thread.h} */ #define wxLEAVE_CRIT_SECT(critical_section) /** This macro is equivalent to @ref wxCriticalSection::Enter "critical_section.Enter()" if @c wxUSE_THREADS is 1 and does nothing if it is 0. @header{wx/thread.h} */ #define wxENTER_CRIT_SECT(critical_section) /** Returns @true if this thread is the main one. Always returns @true if @c wxUSE_THREADS is 0. @header{wx/thread.h} */ bool wxIsMainThread(); /** This function must be called when any thread other than the main GUI thread wants to get access to the GUI library. This function will block the execution of the calling thread until the main thread (or any other thread holding the main GUI lock) leaves the GUI library and no other thread will enter the GUI library until the calling thread calls wxMutexGuiLeave(). Typically, these functions are used like this: @code void MyThread::Foo(void) { // before doing any GUI calls we must ensure that // this thread is the only one doing it! wxMutexGuiEnter(); // Call GUI here: my_window->DrawSomething(); wxMutexGuiLeave(); } @endcode This function is only defined on platforms which support preemptive threads and only works under some ports (wxMSW currently). @note Under GTK, no creation of top-level windows is allowed in any thread but the main one. @header{wx/thread.h} */ void wxMutexGuiEnter(); /** This function is only defined on platforms which support preemptive threads. @see wxMutexGuiEnter() @header{wx/thread.h} */ void wxMutexGuiLeave(); //@}