[mojo-edk] Fix uninitialized reads in EDK dispatchers

mojo/edk/system/data_pipe_consumer_dispatcher.cc

Index: mojo/edk/system/data_pipe_consumer_dispatcher.cc
diff --git a/mojo/edk/system/data_pipe_consumer_dispatcher.cc b/mojo/edk/system/data_pipe_consumer_dispatcher.cc
index 7bf0b0177acdbd0a2a86891d6f6473cbbfab6752..2ed5917dd6166b59241a1cde9e001bf4e660393b 100644
--- a/mojo/edk/system/data_pipe_consumer_dispatcher.cc
+++ b/mojo/edk/system/data_pipe_consumer_dispatcher.cc
@@ -525,9 +525,7 @@ void DataPipeConsumerDispatcher::UpdateSignalsStateNoLock() {
              << " [control_port=" << control_port_.name() << "]";
 
     peer_closed_ = true;
-  }
-
-  if (port_status.has_messages && !in_transit_) {
+  } else if (port_status.has_messages && !in_transit_) {

[Anand Mistry (off Chromium), 2016/05/04 00:50:03]

I'm still pushing back on this construct because I think it's terrible. It may
be fine for us that are very familiar with C++ and this code base, but it's
going to confuse anyone that's new to this.

Basically, my opinion is that it should be possible to evaluate any branch of an
if block independently. There shouldn't be a sequential dependency in the
evaluation of the branch conditions. If you need a result in multiple branches,
pull it out of the if block.

Sorry, but that's my opinion on this.

[Ken Rockot(use gerrit already), 2016/05/04 02:57:05]

done

     ports::ScopedMessage message;
     do {
       int rv = node_controller_->node()->GetMessageIf(control_port_, nullptr,