The protocol keep track of view usage on the client side, so if the protocol still think that a view should be rendered but get destroyed on the side, when an update comes, it will try to render it.
We could probably auto detect that issue at the protocol level, but that was not implemented in that fashion.
You can see that logic here.