Qt源码分析之QPointer

时间:2022-01-27 00:09:10

QPointer是一个指针封装类,其作用类似于智能指针,但是它最大的特点应该是在指针的控制上,它希望一个Qt的指针(当然是从QObject派生的)可以同时被多个类拥有,这在
界面编程中当然是很常见的事情了,但是当这个指针被删除时,我们不希望再找到那两个界面类然后通知它们,相反我们希望这两个界面类可以直接判断QPointer中的isNull方法
很自然的知道原始指针已经不存在了

1.试验代码:
#include <QApplication>
#include <QPushButton>
#include <QPointer>

int main(int argc, char *argv[])
{
 QApplication app(argc, argv);

QPushButton* pButton = new QPushButton("wgs");
 QPointer<QPushButton> button = pButton;
 delete pButton;
 if (!button.isNull())
 {
  button->setText("www");
 }
    
    return app.exec();
}

一段很短的代码,这里需要的注意的是QPointer指针的使用
template <class T>
class QPointer
{
    QObject *o;
public:
    inline QPointer() : o(0) {}
    inline QPointer(T *p) : o(p)
        { QMetaObject::addGuard(&o); }
    inline QPointer(const QPointer<T> &p) : o(p.o)
        { QMetaObject::addGuard(&o); }
    inline ~QPointer()
        { QMetaObject::removeGuard(&o); }
    inline QPointer<T> &operator=(const QPointer<T> &p)
        { if (this != &p) QMetaObject::changeGuard(&o, p.o); return *this; }
    inline QPointer<T> &operator=(T* p)
        { if (o != p) QMetaObject::changeGuard(&o, p); return *this; }

inline bool isNull() const
        { return !o; }

inline T* operator->() const
        { return static_cast<T*>(const_cast<QObject*>(o)); }
    inline T& operator*() const
        { return *static_cast<T*>(const_cast<QObject*>(o)); }
    inline operator T*() const
        { return static_cast<T*>(const_cast<QObject*>(o)); }
};

QPointer只是一个简单的模板,对QMetaObject的相关操作做了简单的封装,这里的基本思想是
在QPointer构造的时候调用QMetaObject::addGuard(&o),把T的指针加入QMetaObject内的一个哈希表中,
在QPointer析构的时候调用QMetaObject::removeGuard(&o),把T的指针从哈希表中删除
这是一个QMetaObject中静态成员,该哈希表的定义如下:
typedef QMultiHash<QObject *, QObject **> GuardHash;
这个哈希表存储的是指针的值和指针的地址,因此加入的代码如下:
void QMetaObject::addGuard(QObject **ptr)
{
    if (!*ptr)
        return;
    GuardHash *hash = guardHash();
    if (!hash) {
        *ptr = 0;
        return;
    }
    QWriteLocker locker(guardHashLock());
    hash->insert(*ptr, ptr);
}

为什么不是只保存一个指针呢,原来是为了防止误删除,看看删除的代码:
void QMetaObject::removeGuard(QObject **ptr)
{
    if (!*ptr)
        return;
    GuardHash *hash = guardHash();
    if (!hash)
        return;
    QWriteLocker locker(guardHashLock());
    GuardHash::iterator it = hash->find(*ptr);
    const GuardHash::iterator end = hash->end();
    for (; it.key() == *ptr && it != end; ++it) {
        if (it.value() == ptr) {
            (void) hash->erase(it);
            break;
        }
    }
}
只有在it.value() == ptr的时候才会删除

但是等等,当删除普通指针时又如何更新这个哈希表呢?
delete pButton;如何通知QMetaObject中的哈希表更新?
答案是在QObject中,请注意QObject的析构函数
QObject::~QObject()
{
    Q_D(QObject);
    if (d->wasDeleted) {
#if defined(QT_DEBUG)
        qWarning("Double QObject deletion detected");
#endif
        return;
    }
    d->wasDeleted = true;

d->blockSig = 0; // unblock signals so we always emit destroyed()

// set all QPointers for this object to zero
    GuardHash *hash = ::guardHash();
    if (hash) {
        QWriteLocker locker(guardHashLock());
        GuardHash::iterator it = hash->find(this);
        const GuardHash::iterator end = hash->end();
        while (it.key() == this && it != end) {
            *it.value() = 0;
            it = hash->erase(it);
        }
    }

emit destroyed(this);

QConnectionList *list = ::connectionList();
    if (list) {
        QWriteLocker locker(&list->lock);
        list->remove(this);
    }

if (d->pendTimer) {
        // have pending timers
        QThread *thr = thread();
        if (thr || d->thread == 0) {
            // don't unregister timers in the wrong thread
            QAbstractEventDispatcher *eventDispatcher = QAbstractEventDispatcher::instance(thr);
            if (eventDispatcher)
                eventDispatcher->unregisterTimers(this);
        }
    }

d->eventFilters.clear();

// delete children objects
    if (!d->children.isEmpty()) {
        qDeleteAll(d->children);
        d->children.clear();
    }

{
        QWriteLocker locker(QObjectPrivate::readWriteLock());
        ::qt_removeObject(this);

/*
          theoretically, we cannot check d->postedEvents without
          holding the postEventList.mutex for the object's thread,
          but since we hold the QObjectPrivate::readWriteLock(),
          nothing can Go into QCoreApplication::postEvent(), which
          effectively means noone can post new events, which is what
          we are trying to prevent. this means we can safely check
          d->postedEvents, since we are fairly sure it will not
          change (it could, but only by decreasing, i.e. removing
          posted events from a differebnt thread)
        */
        if (d->postedEvents > 0)
            QCoreApplication::removePostedEvents(this);
    }

if (d->parent)        // remove it from parent object
        d->setParent_helper(0);

delete d;
    d_ptr = 0;
}
这里做了很多很多的事情,其中的代码
    // set all QPointers for this object to zero
    GuardHash *hash = ::guardHash();
    if (hash) {
        QWriteLocker locker(guardHashLock());
        GuardHash::iterator it = hash->find(this);
        const GuardHash::iterator end = hash->end();
        while (it.key() == this && it != end) {
            *it.value() = 0;
            it = hash->erase(it);
        }
    }

就是更新QMetaObject中哈希表的

这也就是单基类的好处,可以在这里控制很多事情

http://blog.csdn.net/oowgsoo/article/details/1529424