Only check for on-heap Member<T>s iff T is in scope.

Source/platform/heap/Handle.h

Index: Source/platform/heap/Handle.h
diff --git a/Source/platform/heap/Handle.h b/Source/platform/heap/Handle.h
index 10aeace32db59e50b21e725505b4eed274530250..d67733188b30df10864bd8e35353f5961c741518 100644
--- a/Source/platform/heap/Handle.h
+++ b/Source/platform/heap/Handle.h
@@ -789,9 +789,10 @@ protected:
         //   (b) a pointer to the head of an on-heap mixin object.
         //
         // We can check it by calling Heap::isHeapObjectAlive(m_raw),
-        // but we cannot call it here because it requres to include T.h.
-        // So we currently implement only the check for (a).
-        if (!IsGarbageCollectedMixin<T>::value)
+        // but we cannot call it here because it requires to include T.h.
+        // So we currently only try to implement the check for (a), but do
+        // not insist that T's definition is in scope.
+        if (IsFullyDefined<T>::value && !IsGarbageCollectedMixin<T>::value)

[Nico, 2015/06/30 20:04:45]

If this assert fires now depends on .h include order, right? If so, that seems
kind of bad too.

(If not:
https://code.google.com/p/chromium/codesearch#chromium/src/third_party/WebKit...
just inlines the static_assert – not sure this IsFullyDefined class is needed)

Did you try my suggestion of out-of-lining a few destructors?

[sof, 2015/06/30 20:10:59]

That's entirely to be expected. e.g., TrackBase doesn't #include
HTMLMediaElement, because it cannot without stepping into a cyclic dependency. 

So, I don't know what your concern is wrt this particular check.

             ASSERT(HeapObjectHeader::fromPayload(m_raw)->checkHeader());
 #endif
     }