[DevTools] Dispatch messages to V8InspectorSession directly.

third_party/WebKit/Source/core/inspector/CodeGeneratorInstrumentation.py

Index: third_party/WebKit/Source/core/inspector/CodeGeneratorInstrumentation.py
diff --git a/third_party/WebKit/Source/core/inspector/CodeGeneratorInstrumentation.py b/third_party/WebKit/Source/core/inspector/CodeGeneratorInstrumentation.py
index aa4a56fbb83e381412b415be23a1a629deeb8186..fbfdc6c0c77dd4bbee5fbdb6a28ffebdba0c2605 100755
--- a/third_party/WebKit/Source/core/inspector/CodeGeneratorInstrumentation.py
+++ b/third_party/WebKit/Source/core/inspector/CodeGeneratorInstrumentation.py
@@ -97,10 +97,11 @@ namespace blink {
 
 ${forward_list}
 
-class CORE_EXPORT InstrumentingAgents : public GarbageCollected<InstrumentingAgents> {
+class CORE_EXPORT InstrumentingAgents : public GarbageCollectedFinalized<InstrumentingAgents> {

[sof, 2016/05/23 06:20:53]

Why this change? This object doesn't need to run any finalizer (and if it did,
the clang plugin would issue an error.)

[dgozman, 2016/05/23 15:47:02]

I was fixing a crash, and thought that maybe we don't destruct HeapHashMap
inside this object. That seemed to help at first, but not really.

I can revert this back in a follow-up.

[sof, 2016/05/23 18:00:30]

Thanks for explaining; the HeapHashMap backing store table will independently
finalize its elements, if needed.

The reason I fuss over this :) is that we're considering enabling a plugin
warning by default (warn-unneeded-finalizer) which checks for unnecessary uses
of *Finalized<>, so keeping the codebase warning-free until then makes some
sense. I don't think such a warning will be getting in anyone's way very much,
but that's the trade-off to make.

https://codereview.chromium.org/2003033002/ switches to GarbageCollected<>.

     WTF_MAKE_NONCOPYABLE(InstrumentingAgents);
 public:
     InstrumentingAgents();
+    ~InstrumentingAgents() { }
     DECLARE_TRACE();
 ${accessor_list}