mirror of
https://github.com/vitorpamplona/amethyst.git
synced 2026-10-06 11:48:24 +00:00
Claude
f52b07865e
An audit of this branch turned up ten findings. These are the six that are small and unambiguous; the other four are design-level and are written up separately rather than fixed in passing. - The fixture coordinator's `join_request_store` stored `at = clock++` and returned `clock`, so the value a caller acked with was always one past the value stored. Nothing could ever be retired. - `join_request_take_many` ignored `consumed` outright, which made retirement a no-op in every fixture-backed test: an answered request came back forever and no test could notice. It now retires exactly as `welcome_take` beside it does. - `pendingWelcomes` passed `drainRetirements()` inline, emptying the queue before the call it rode on. A failing `takeWelcomes` dropped those acks permanently and the coordinator kept re-serving consumed Welcomes. The join-request path already re-queued on failure; this matches it. - A cordn last-resort KeyPackage carries its marker inside `app_data_dictionary` (0x0006) but advertised leaf capabilities that did not include it. `CordnGroupPolicy.lastResortLeafCapabilities()` existed for precisely this and was called from nowhere. - `OpenStreamReceiver.lastActivityMs` started at 0, so against a real clock `needsProbe()` was true from construction — thirty-odd years idle before the first frame. Only a test calls it today, which is why it went unnoticed. - `KeyStoreCordnBlobCipher` serialised every cordn KeyStore encrypt and decrypt behind a lock whose justification no longer holds: `KeyStoreEncryption` keeps its `Cipher` in a `ThreadLocal` now, so the init/doFinal pair cannot interleave and the lock only cost concurrency. The KDoc explaining the old hazard is replaced by one explaining why there is no longer a lock. Incidental, from the merge: `EXTERNAL_SENDERS_EXTENSION_TYPE` was 0x0004, the same value as `EXTERNAL_PUB_EXTENSION_TYPE`. RFC 9420 §13.3 assigns it 0x0005. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012BfD4txdnsaPRXmNXbup9n