mojo: Make InterfacePtr<> testable in if() statements without .get().

mojo/public/cpp/bindings/interface_ptr.h

Index: mojo/public/cpp/bindings/interface_ptr.h
diff --git a/mojo/public/cpp/bindings/interface_ptr.h b/mojo/public/cpp/bindings/interface_ptr.h
index b868a0a5c36167c1885c16f597a600bfed4ecef2..80c09c7227effc2fb36a6049361703a4a27eb503 100644
--- a/mojo/public/cpp/bindings/interface_ptr.h
+++ b/mojo/public/cpp/bindings/interface_ptr.h
@@ -104,6 +104,16 @@ class InterfacePtr {
     return &internal_state_;
   }
 
+  // Allow InterfacePtr<> to be used in boolean expressions, but not

[yzshen1, 2014/07/18 19:17:43]

One thing that may worth considering:
I just landed a CL to make InterfacePtr<> do the setup work lazily: Bind() only
records the pipe and waiter; the internal ProxyWithStub/Rounter are constructed
only when the pointer is actually used. (Please see
https://code.google.com/p/chromium/issues/detail?id=394883 for why this is
needed.)

Calling get() is one way of "actually using" the InterfacePtr<> and doing all
the setup.

Therefore, it seems nice if we can avoid get() in Testable(), instead we test
whether it has been bound. What do you think?

[Elliot Glaysher, 2014/07/18 19:32:15]

What do you propose replacing get() with?

[yzshen1, 2014/07/18 19:38:45]

Maybe an IsBound() method on the internal state, which checks
(handle_.is_valid() || router).

+  // implicitly convertible to a real bool (which is dangerous).
+ private:
+  typedef internal::InterfacePtrState<Interface> InterfacePtr::*Testable;
+
+ public:
+  operator Testable() const {
+    return get() ? &InterfacePtr::internal_state_ : NULL;
+  }
+
  private:
   typedef internal::InterfacePtrState<Interface> State;
   mutable State internal_state_;