Adding key press detection in the browser process.

media/audio/key_press_monitor.h

Index: media/audio/key_press_monitor.h
diff --git a/media/audio/key_press_monitor.h b/media/audio/key_press_monitor.h
new file mode 100644
index 0000000000000000000000000000000000000000..1454b5e1760955f148a6996eb93a8f900878c8cc
--- /dev/null
+++ b/media/audio/key_press_monitor.h
@@ -0,0 +1,55 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef MEDIA_AUDIO_KEY_PRESS_MONITOR_H_
+#define MEDIA_AUDIO_KEY_PRESS_MONITOR_H_
+
+#include "base/basictypes.h"
+#include "base/observer_list.h"
+#include "base/synchronization/lock.h"
+
+namespace base {
+class MessageLoopProxy;
+}  // namespace base
+
+namespace media {
+
+// This class monitors the system-wide key presses and notifies the listeners.
+// Not thread safe. All methods should be called on the thread whose message
+// loop is passed to the ctor.

[DaleCurtis, 2013/08/02 21:04:47]

What about construction and destruction?  Should destruction check
listeners_.size() == 0?  Can it be done in a thread safe manner?

The way this is setup KeyPressMonitor will be destroyed before AudioManager
finishes processing all of the messages on the audio message loop.  I suspect
that will lead to some crashes...

You'll need to be careful to make sure KeypressMonitor outlives AudioManager and
that KeypressMonitor avoids spamming the audio loop in a way that would prevent
shutdown.  I.e., if I can hang browser shutdown by typing quickly, we've got a
problem :)

[jiayl, 2013/08/02 21:45:45]

Changed it to RefCountedThreadSafe. AudioInputController now holds a reference
to KeyPressMonitor to make sure it's valid to call AddListener/RemoveListener.
When AudioInputController goes away, it releases KeyPressMonitor, which release
the message loop. There should be no circular reference or hang. Sounds right?

[DaleCurtis, 2013/08/02 21:55:29]

Ref counting is a last resort and likely overkill here.  The ownership semantics
are now quite messy.  If you just destruct KeypressMonitor after AudioManager,
you shouldn't have any problems.  Is there a reason you chose not to do that?

+class KeyPressMonitor {
+ public:
+  class KeyPressListener {
+   public:
+    // This is called on |message_loop_|'s thread.
+    virtual void OnKeyPressed() = 0;
+   protected:
+    virtual ~KeyPressListener() {}
+  };
+
+  // |message_loop| is the message loop that this object runs on.
+  KeyPressMonitor(base::MessageLoopProxy* message_loop);
+  ~KeyPressMonitor();
+
+  // Adds |listener| to receive event notifications. A listener should not be
+  // added more than once.
+  // The callers should make sure to call RemoveKeyPressListener before
+  // |listener| is destroyed.
+  void AddKeyPressListener(KeyPressListener* listener);
+  // Removes |listener| from receiving event notifications. Does nothing if
+  // |listener| has not been added or has already been removed.
+  void RemoveKeyPressListener(KeyPressListener* listener);
+
+ private:
+  // Raw pointers of the listeners.
+  ObserverList<KeyPressListener> listeners_;
+  // The message loop of the thread that this object runs on.
+  scoped_refptr<base::MessageLoopProxy> message_loop_;
+
+  DISALLOW_COPY_AND_ASSIGN(KeyPressMonitor);
+};
+
+}  // namespace media
+
+#endif  // MEDIA_AUDIO_KEY_PRESS_MONITOR_H_