Fix handle_watcher when start is called twice on the same handle.

mojo/common/handle_watcher.cc

 // 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.
 
 #include "mojo/common/handle_watcher.h"
 
 #include <map>
 
 #include "base/atomic_sequence_num.h"
 #include "base/bind.h"
@@ skipping 440 matching lines @@
 HandleWatcher::~HandleWatcher() {
 }
 
 void HandleWatcher::Start(const Handle& handle,
                           MojoHandleSignals handle_signals,
                           MojoDeadline deadline,
                           const base::Callback<void(MojoResult)>& callback) {
   DCHECK(handle.is_valid());
   DCHECK_NE(MOJO_HANDLE_SIGNAL_NONE, handle_signals);
 
+  // Need to clear the state before creating a new one.
+  state_.reset();

[viettrungluu, 2014/12/08 18:10:01]

I assume that this is because we need the dtor of the old state to run before
the ctor for the new state (for at least one of
Same/SecondaryThreadWatchingState).

This is pretty unintuitive though, and mostly speaks to the badness of doing
stuff in dtors.

[qsr, 2014/12/09 10:02:23]

 The destructor remove the handler, and neither of those can watch the same
handle multiple times.

   if (MessagePumpMojo::IsCurrent()) {
     state_.reset(new SameThreadWatchingState(
         this, handle, handle_signals, deadline, callback));
   } else {
     state_.reset(new SecondaryThreadWatchingState(
         this, handle, handle_signals, deadline, callback));
   }
 }
 
 void HandleWatcher::Stop() {
   state_.reset();
 }
 
 }  // namespace common
 }  // namespace mojo