mirror of
https://github.com/vitorpamplona/amethyst.git
synced 2026-10-05 19:28:25 +00:00
Claude
de34f392a9
A replaceable/addressable version that a stored one already beats is not written. Since18c576a068the SQLite store has reported it as SUPERSEDED, "duplicate: a newer version ...", and RelaySession acks every `duplicate:` reason with OK true. So a client was told its event is on the relay when no REQ will ever return it. NIP-01's third OK field is `true` when the event was accepted. classifyRowError now answers RejectionReason.REPLACED, `replaced: a newer version exists`, which RelaySession sends as OK false. That is strfry's answer too (`false, "replaced: have newer event"`). It also names the case correctly: the relay does not have THIS event, it has a newer one. The W09 work is untouched. Classification still asks the database rather than the driver's message, so Android's null-message constraint failures still classify correctly, and a byte-for-byte re-offer of a stored version is still DUPLICATE (OK true). The MDK `wn` loop18c576a068was fixing (a second KeyPackage minted in the same second) now gets `OK false replaced:` instead of raw constraint text, which was the unclassifiable part. Retrying a `replaced:` rejection cannot succeed, so that is for the client to stop doing, not for the relay to paper over with OK true. RejectionReason.SUPERSEDED stays, @Deprecated in favour of REPLACED, for source compatibility. NostrServerTest and InsertOutcomeClassificationTest flip their expectations, and the event-store-semantics skill updates W01/W02 and W09 and adds a changelog entry. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AdSGU88PYrRVyVHjiEc5sw