Files
buzz/mobile/lib
Tom Brow a93f3282cc perf(mobile): retain preference crypto across reconnects (#8442)
Foreground reconnects rebuild preference managers and repeatedly derive
the same NIP-44 keys on the UI isolate. Retain the six crypto objects in
providers scoped to relay configuration and identity while preserving
manager refresh, subscription, and publication behavior. Community
changes, key rotation, and sign-out invalidate the retained objects.

RelayConfig normalizes the signing key once at its shared accessor, so
identity, theme encryption, event signing, and NIP-42 authentication
receive the same value, including when an active configuration key has
surrounding whitespace. The identity-export authorization snapshot uses
the same normalized key for the active-identity check and the outgoing
recovery payload. The branch incorporates main's relay-scoped preference
managers.

### Validation
- Regression exercises six production crypto providers across
reconnects, configuration/key changes, and sign-out. Reintroducing a
session dependency makes it fail.
- Added a loopback WebSocket regression that exercises actual
RelaySocket NIP-42 authentication with a padded key and verifies its
challenge, relay, pubkey and signature. It fails before shared key
normalization.
- Added a connected-theme regression for whitespace-padded keys,
encryption round-trip, and a valid published event signature. It fails
before the normalization fix.
- Added padded-key identity export coverage through authorization, SAS
confirmation, exactly one normalized payload, and successful completion.
Separate relay-change and key-rotation tests verify both fresh and
expired grants fail closed. The padded-key regression failed before the
fix.
- Current head: all 48 focused pairing/crypto/auth tests and mobile
analysis passed. All [CI
checks](https://github.com/block/buzz/actions/runs/39497098787) passed,
including mobile test shards, Android, native Swift tests and release
builds. Codex feedback is addressed; Jude approved this exact head after
independent regression and mutation checks. Fresh [Codex security
review](https://github.com/block/buzz/pull/8442#issuecomment-6300273942)
found no concrete issues.
- Previous head: 122 config/socket/crypto/theme tests, 44 read-state
tests, and all [GitHub CI
checks](https://github.com/block/buzz/actions/runs/39491550425) passed,
including mobile test shards, Android, native Swift tests and release
builds.
- Original implementation: native iOS Debug build and full local
repository checks passed. A paired simulator background/resume capture
had no samples in the targeted crypto constructors or key derivation; an
82 ms UI frame remained. Debug/JIT, profiler overhead and shared-host
activity prevent an overall speedup claim. The later normalization
change is covered by the production-provider regression above.
- Two image decode timing failures occurred in the initial local full
suite; direct baseline/patch runs, the subsequent full repository run,
and the latest full mobile suite passed.

Cold-start credential validation is unchanged: padded persisted keys are
rejected by AuthNotifier before RelayConfig is constructed. The
normalization regressions cover active configurations and identity
export, not broader persisted-key acceptance.

<!-- buzz-review-completed -->

---------

Signed-off-by: Tom Brow <tomb@block.xyz>
2026-10-20 14:57:53 -07:00
..