mirror of
https://github.com/jmcorgan/fips.git
synced 2026-10-05 19:18:25 +00:00
Johnathan Corgan
2236371537
After a traversal attempt, both sides published a NIP-09 deletion request signed by the node's routing key and naming the attempt's offer and answer wraps. Each request put the node's public identity next to the ids of its traversal signals on every relay it reached, which the one-time signing keys on the wraps are meant to avoid. What the requests deleted differed by side. A relay honouring NIP-59 deletes a gift wrap when the request is signed by the wrap's p-tagged recipient, so the initiator's request, sent after a successful punch, did delete the answer wrap addressed to it; strfry in the NAT lab logged that deletion. The offer it also named is addressed to the responder, and the responder's requests named only the answer it had sent to the initiator, so those deleted nothing. Drop all three traversal calls (the initiator after a punch, the responder on refusing an offer, and the responder after its punch attempt), and accept that an answer wrap now stays on a relay that stores it until its NIP-40 expiration. The wraps are ephemeral kinds, so relays that do not store ephemeral events never held them. The advertisement retraction keeps its deletion request, since it names an event the routing key signed itself. The signal envelope's event id had no other reader and is removed with its import. The NAT lab now asserts that its relay holds no kind 5 event after the cone, symmetric and lan scenarios, scanning the relay's own store and failing if the scan cannot run or be parsed. Before the fix each scenario failed it: cone and lan held two requests (the initiator's naming offer and answer, the responder's naming the answer) and symmetric held the responder's one. After it, all three held none. The responder's refusal path is not reached by any lab scenario and is covered by reading only.