Keep running into issues where browser engineers write spec IDL that violates the IDL spec in whatever ways their binding implementation does not enforce. If we're lucky a second implementor, with a binding implementation with different holes, tries to implement before it ships
It's wrong in the spec, then doesn't get caught when implementing because the implementation doesn't enforce some of the IDL spec rules on what valid IDL is. And generally if the mental model is based on the bindings code not the IDL spec, it makes wrong IDL more likely...