That's true. If this is an issue you could avoid using a signaling server altogether, and instead exchange offer/answer via e.g. GPG-encrypted email.
I wonder if there are any designs for a signaling server/system that doesn't have this loss of privacy of connection information, or if you just very quickly end up reimplementing Tor when you try..
Good point. If any sufficiently complicated NAT traversal solution ends up containing an ad hoc, informally-specified, bug-ridden, slow implementation of half of the hidden services protocol, might as well put our efforts into building out Tor itself and get the privacy for "free".
