(From bug #32612 comment #6):
> +private_tubes_factory_tube_close_cb (
> + if (!tube_msg_checks (self, msg, node, NULL, &tube_id))
> + return FALSE;
> Er, this function allows Alice to close tubes between us and Bob, if she can
> guess or brute-force the tube ID. Pre-existing bug?
> + DEBUG ("tube ID already in use; do not open the offered tube and close "
> + "the existing tube if it's to the same contact");
> Not a merge blocker and presumably not your fault, but these semantics are
> crazy. We should have a separate tube ID "namespace" per peer, and store tubes
> in the hash table by (handle, id) tuples or something.