Add a ChannelEndpointClient abstraction.

mojo/edk/system/channel.h

 // Copyright 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 MOJO_EDK_SYSTEM_CHANNEL_H_
 #define MOJO_EDK_SYSTEM_CHANNEL_H_
 
 #include <stdint.h>
 
 #include "base/containers/hash_tables.h"
@@ skipping 183 matching lines @@
   // Note: The IDs generated by this should be checked for existence before use.
   LocalChannelEndpointIdGenerator local_id_generator_;
 
   typedef base::hash_map<ChannelEndpointId, scoped_refptr<MessagePipe>>
       IdToMessagePipeMap;
   // Map from local IDs to pending/incoming message pipes (i.e., those which do
   // not yet have a dispatcher attached).
   // TODO(vtl): This is a layering violation, since |Channel| shouldn't know
   // about |MessagePipe|. However, we can't just hang on to |ChannelEndpoint|s
   // (even if they have a reference to the |MessagePipe|) since their lifetimes
   // are tied to the "remote" side. When |ChannelEndpoint::OnDisconnect()|

[yzshen1, 2014/11/18 00:28:17]

I think there isn't OnDisconnect anymore

[viettrungluu, 2014/11/18 00:47:29]

Done.

-  // (eventually) results in |ChannelEndpoint::DetachFromMessagePipe()| being
-  // called. We really need to hang on to the "local" side of the message pipe,
-  // to which dispatchers will be "attached".
+  // (eventually) results in |ChannelEndpoint::DetachFromClient()| being called.
+  // We really need to hang on to the "local" side of the message pipe, to which
+  // dispatchers will be "attached".
   IdToMessagePipeMap incoming_message_pipes_;
   // TODO(vtl): We need to keep track of remote IDs (so that we don't collide
   // if/when we wrap).
   RemoteChannelEndpointIdGenerator remote_id_generator_;
 
   DISALLOW_COPY_AND_ASSIGN(Channel);
 };
 
 }  // namespace system
 }  // namespace mojo
 
 #endif  // MOJO_EDK_SYSTEM_CHANNEL_H_