Storage tasks are getting lost

[Alexey Proskuryakov]

With WebCore SQL logging enabled, open <http://bugs.webkit.org/attachment.cgi?id=18455>. It runs without assertion failures now, but after a while, it just stops (it doesn't freeze Safari, but no more SQL requests are being served).

[Alexey Proskuryakov]

The database thread loop looks like this:

    while (!m_terminationRequested) {
        m_threadMutex.unlock();
        AutodrainedPool pool;

        LOG(StorageAPI, "Iteration of main loop for DatabaseThread %p", this);

        bool result;
        do {
            result = dispatchNextTaskIdentifier();
            if (m_terminationRequested)
                break;
        } while(result);

        if (m_terminationRequested)
            break;

        pool.cycle();
        m_threadMutex.lock();
        m_threadCondition.wait(m_threadMutex);
    }

It's pretty clear how the wakeup is lost - if a new task is dispatched after the inner loop has decided that it has nothing else to do, but before we're stopped on the condition, the condition is signaled in vain.

[Alexey Proskuryakov]

<rdar://problem/5729619>

[Alexey Proskuryakov]

Fixed in <http://trac.webkit.org/projects/webkit/changeset/30172>.