SyncLock

// This may look like C code, but it's really -*- C++ -*-
/*
 * Copyright (C) 2010 Emweb bvba, Kessel-Lo, Belgium.
 *
 * See the LICENSE file for terms of use.
 */
#ifndef WT_SYNC_LOCK_H_
#define WT_SYNC_LOCK_H_

#include <Wt/WGlobal>
#include <Wt/WApplication>
#include <boost/thread.hpp>

namespace Wt {

/*! \class SyncLock Wt/SyncLock Wt/SyncLock
 *  \brief An dead-lock avoidance adaptor for a Boost mutex lock.
 *
 * A %SyncLock adapts a Boost mutex lock (such as boost::mutex::scoped_lock),
 * and provides the same API as the original lock (it derives from it).
 *
 * Just as can be expected by a lock, a call to lock() will block until the
 * thread has exclusive access to the mutex.
 *
 * While waiting to acquire the lock, however, the current
 * Wt::WApplication lock, may be temporarily released in favor of
 * another helper thread trying to acquire the application lock. Thus,
 * you need to be aware that when the lock is taken, some other thread
 * may have modified the current application state.
 *
 * A sync lock is useful in the context of a deployment where multiple
 * (or all) WApplication instances are running in a single process and
 * communicating with each other (such as the built-in httpd). Then,
 * this lock adaptor helps in avoiding a dead-lock situation when
 * applications wants to access a shared resource protected by a
 * mutex, which communicates to other applications while taking their
 * update lock. In that case, it is likely that you will also want to
 * iterate over all "registered" applications while holding the global
 * mutex, and this creates a natural dead-lock scenario because you
 * have two mutexes (global mutex, application mutex) which are
 * sequentially aquired in a different order:
 *  - application mutex -> global mutex, during an application request
 *  - global mutex -> application mutex, while propagating events to applications
 *
 * The altered behaviour of a call by application <i>A</i> to lock()
 * is that this application <i>A</i>'s state may be updated by another
 * thread (e.g. serving application <i>B</i>), which tries to take
 * <i>A</i>'s update lock.
 *
 * The following conventional locking code:
 * \code
 *   boost::recursive_mutex::scoped_lock lock(mutex_);
 * \endcode
 * is updated to a sync lock, by doing:
 * \code
 *   Wt::SyncLock<boost::recursive_mutex::scoped_lock> lock(mutex_);
 * \endcode
 */
template <class Lock>
class SyncLock : public Lock
{
public:
  /*! \brief Creates a RIIA lock.
   *
   * Initializes and acquires exclusive access to the \p mutex.
   */
  template <class Mutex>
  SyncLock(Mutex& mutex)
    : Lock(mutex, boost::defer_lock)
  {
    lock();
  }

  /*! \brief Creates a RIIA lock, but defers taking the lock.
   *
   * Initialize the lock but defers taking the lock.
   *
   * \sa lock()
   */
  template <class Mutex>
  SyncLock(Mutex& mutex, boost::defer_lock_t)
    : Lock(mutex, boost::defer_lock)
  { }

  /*! \brief Tries to acquire the lock, blocking while waiting.
   *
   * While an application <i>A</i> is waiting for the lock, its state
   * may be updated by another application: another application
   * <i>B</i> can succesfully take <i>A's</i> \link
   * WApplication::UpdateLock update lock\endlink.
   *
   * Therefore you need to be prepared to deal with application state
   * changes while waiting for the lock.
   */
  void lock() {
    WApplication *app = WApplication::instance();
    if (app) {
      int id = app->startWaitingAtLock();
      Lock::lock();
      app->endWaitingAtLock(id);
    } else
      Lock::lock();
  }
};

}

#endif // WT_SYNC_LOCK_H_