Tidy up the debugger.

services/tracing/tracing_app.h

Index: services/tracing/tracing_app.h
diff --git a/services/tracing/tracing_app.h b/services/tracing/tracing_app.h
index 19c88c21180ad5e34fa07a9f610efa927a352b49..7d50efb7bbdbc0cb6c4df2f562996812874a2d05 100644
--- a/services/tracing/tracing_app.h
+++ b/services/tracing/tracing_app.h
@@ -8,6 +8,7 @@
 #include "base/macros.h"
 #include "base/memory/scoped_ptr.h"
 #include "base/memory/scoped_vector.h"
+#include "mojo/common/binding_set.h"
 #include "mojo/common/interface_ptr_set.h"
 #include "mojo/public/cpp/application/application_delegate.h"
 #include "mojo/public/cpp/application/application_impl.h"
@@ -44,7 +45,7 @@ class TracingApp : public mojo::ApplicationDelegate,
   scoped_ptr<TraceDataSink> sink_;
   ScopedVector<TraceRecorderImpl> recorder_impls_;
   mojo::InterfacePtrSet<TraceProvider> provider_ptrs_;
-  mojo::Binding<TraceCollector> collector_binding_;
+  mojo::BindingSet<TraceCollector> collector_bindings_;

[ppi, 2015/12/20 15:36:21]

Collector instances are "stateful", it is incorrect to bind independent requests
to the same instance.

[jeffbrown, 2015/12/20 20:52:58]

Ahhh, then this whole class is somewhat problematic.  The service is trying to
behave like a singleton yet it crashes when multiple connections are made to it.

Specifically, if you start tracing when launching an app you are unable to
restart tracing later using the mojo_debug tool.

[ppi, 2015/12/21 09:39:50]

Yes, and the problems go deeper, down to C++ base:: implementation of tracing,
which can't do two overlapping tracing runs.

I think we should print a better error message saying that tracing can support
only one trace collection at a time when subsequent connection is made, and
close the passed message pipe.

   bool tracing_active_;
   mojo::String tracing_categories_;