QEventLoop 销毁时注意事项
(2011-01-02 13:44:03)
标签:
qeventloopqt源码venusit |
分类: It文章 |
查看原文:http://newfaction.net/2011/01/02/qeventloop-destroyed-notes.html
上篇文章对QT的QEventLoop 做了简单介绍, 通过对QEventLoop 源码的查阅,QEventLoop的实现无非是: 线程+PostEvent的强化。 有用到QEventLoop的童鞋可能会发现 "为什么QEventLoop在调用exit()以后,直接调用delete 销毁QEventLoop时将出现异常!!!下面做简单剖析:
上篇文章对QT的QEventLoop 做了简单介绍, 通过对QEventLoop 源码的查阅,QEventLoop的实现无非是: 线程+PostEvent的强化。 有用到QEventLoop的童鞋可能会发现 "为什么QEventLoop在调用exit()以后,直接调用delete 销毁QEventLoop时将出现异常!!!下面做简单剖析:
上篇文章中,关于QEventLoop::processEvents的实现:
bool QEventLoop::processEvents( ProcessEventsFlags flags )
{
// process events from the QWS server
int nevents = 0;
......
// handle gui and posted events
if (qt_is_gui_used ) {
//a.首先执行globalPostedEvents队列中的事件,这些事件都是通过QApplication::postEvent或QThread::postEvent发送的
QApplication::sendPostedEvents();
//b.检查QWSEvent队列:QApplication::Data的QPtrList<QWSEvent> queue队列
while ( qt_fbdpy->eventPending() ) { // also flushes output buffer
if ( d->shortcut ) {
return FALSE;
}
QWSEvent *event = qt_fbdpy->getEvent(); // get next event
nevents++;
bool ret = qApp->qwsProcessEvent( event ) == 1;
delete event;
if ( ret ) {
return TRUE;
}
}
}
if ( d->shortcut ) {
return FALSE;
}
//c.如果是QWSServer,需要检查QWSServer的QPtrList<QWSCommandStruct> commandQueue队列。
extern QPtrQueue<QWSCommand> *qt_get_server_queue();
if ( !qt_get_server_queue()->isEmpty() ) {
QWSServer::processEventQueue();
}
//d.如果上面的事件执行过程中产生postedEvent,再次执行,主是要一些moveEvent,resizeEvent,paintEvent
QApplication::sendPostedEvents();
// don't block if exitLoop() or exit()/quit() has been called.
bool canWait = d->exitloop || d->quitnow ? FALSE : (flags & WaitForMore);
// Process timers and socket notifiers - the common UNIX stuff
//e.检查timer事件队列中,确发时间最小的一个,确定为select的等待时间。
// return the maximum time we can wait for an event.
static timeval zerotm;
timeval *tm = qt_wait_timer(); // wait for timer or event
if ( !canWait ) {
if ( !tm )
tm = &zerotm;
tm->tv_sec = 0; // no time to wait
tm->tv_usec = 0;
}
//f.准备select调用所需要的fd队列。
int highest = 0;
if ( ! ( flags & ExcludeSocketNotifiers ) ) {
// return the highest fd we can wait for input on
if ( d->sn_highest >= 0 ) { // has socket notifier(s)
if ( d->sn_vec[0].list && ! d->sn_vec[0].list->isEmpty() )
d->sn_vec[0].select_fds = d->sn_vec[0].enabled_fds;
else
FD_ZERO( &d->sn_vec[0].select_fds );
if ( d->sn_vec[1].list && ! d->sn_vec[1].list->isEmpty() )
d->sn_vec[1].select_fds = d->sn_vec[1].enabled_fds;
else
FD_ZERO( &d->sn_vec[1].select_fds );
if ( d->sn_vec[2].list && ! d->sn_vec[2].list->isEmpty() )
d->sn_vec[2].select_fds = d->sn_vec[2].enabled_fds;
else
FD_ZERO( &d->sn_vec[2].select_fds );
}
//g.检查有没有注册的preselect_handler
if ( qt_preselect_handler ) {
QVFuncList::Iterator it, end = qt_preselect_handler->end();
for ( it = qt_preselect_handler->begin(); it != end; ++it )
(**it)();
}
//h,执行select调用。
...................
nsel = select( highest + 1,
&d->sn_vec[0].select_fds,
&d->sn_vec[1].select_fds,
&d->sn_vec[2].select_fds,
tm );
...............
//i.看看是否有thread发来的唤醒消息
// some other thread woke us up... consume the data on the thread pipe so that
// select doesn't immediately return next time
if ( nsel > 0 && FD_ISSET( d->thread_pipe[0], &d->sn_vec[0].select_fds ) ) {
char c;
::read( d->thread_pipe[0], &c, 1 );
}
//j.检查是否有注册的postselect_hander事件
if ( qt_postselect_handler ) {
QVFuncList::Iterator it, end = qt_postselect_handler->end();
for ( it = qt_postselect_handler->begin(); it != end; ++it )
(**it)();
}
//k.检查所有能进行操作的socket fd,并执行相应的QSocketNotifier::activated()
// activate socket notifiers
.....................
nevents += activateSocketNotifiers();
//l.检查timerList是否有timeout的timer并执行.
// activate timers
nevents += activateTimers();
// return true if we handled events, false otherwise
return (nevents > 0);
}
从源码中可以发现QEventLoop是由一个线程在后台间隔的调用processEvent的过程,在调用exit()时,做的工作如下:
1、设置exit状态标志
2、发出中断信号
第一步直接效果于: 调用isRunning等时,直接返回false;
第二步中断信号:是修改线程中某个标志位,让线程下次轮循调用到processEvent前退出工作流程! 因而调用exit()
并不能立刻使eventLoop退出,需要等待片刻,而这个片刻的时间大概多长呢? 答案是“一个processEvent”的工作周期。