Allocation audit from Android benchmark showed 19 allocs in signSchnorr
and 4 in verifySchnorr. Most were intermediate ByteArray/LongArray that
can be replaced with pre-allocated scratch buffers.
Changes:
- Add sha256Into() (expect/actual) that writes digest into existing buffer
instead of allocating a new ByteArray(32) per call. Uses
MessageDigest.digest(buf,off,len) on JVM/Android, CC_SHA256 on Apple.
- Add scratch byte buffers to PointScratch: hashBuf(256), bytesTmp1/2(32),
scalarTmp1/2/3 for intermediate scalar results.
- Add ScalarN.reduceTo() allocation-free variant.
- Rewrite signSchnorrInternal to reuse scratch buffers for:
- dBytes serialization (bytesTmp1 instead of U256.toBytes alloc)
- AUX_PREFIX+auxrand hash (hashBuf instead of array concatenation)
- auxHash XOR (scalarTmp1/2/3 instead of U256.fromBytes allocs)
- nonce/challenge hash inputs (hashBuf instead of ByteArray alloc)
- nonce scalar (scalarTmp1 instead of ScalarN.reduce alloc)
- challenge scalar (scalarTmp3 instead of allocs)
- e*d and k+e*d (splitK1/entryTmp2 instead of ScalarN.mul/add allocs)
- Rewrite verifySchnorr to use hashBuf and sha256Into for challenge hash.
Only the 64-byte output signature is allocated per sign call.
Verify allocates nothing for 32-byte messages (hashBuf is large enough).
https://claude.ai/code/session_01EMY5RnXb9rnsyU2KbXrSaY
Extends the ThreadLocal SHA256 optimization to the streaming hash function.
The pool is retained for EventHasherSerializer and HashingByteArrayBuilder
which use its acquire/release pattern for long-lived incremental hashing.
https://claude.ai/code/session_01BhU63WUe9AhikZxRdw3Lpg
The SHA256 pool used ArrayBlockingQueue.take()/put() which acquire a lock
on every call. Profiling showed ~10µs synchronization overhead per SHA256
for ~2µs of actual hashing — the lock cost 5× more than the hash itself.
Replaced with ThreadLocal<MessageDigest> which gives each thread its own
instance with zero synchronization. MessageDigest.digest() implicitly
resets state, so no explicit reset is needed.
The pool is kept available for streaming use cases (hashStream) that need
incremental hashing with update()/digest() across multiple calls.
Only affects jvmAndroid — Apple uses CC_SHA256 directly, Linux uses
whyoleg CryptographyProvider. Both already have no pool overhead.
https://claude.ai/code/session_01BhU63WUe9AhikZxRdw3Lpg
Added encrypt(data, aad) and decrypt(data, aad) overloads to the
AESGCM expect/actual class across all platforms:
- JVM/Android: Uses Cipher.updateAAD() with AES/GCM/NoPadding
- Apple: Uses whyoleg.cryptography encryptWithIvBlocking(iv, data, aad)
- Linux: Same as Apple via whyoleg.cryptography
Updated HPKE aeadSeal/aeadOpen and MlsCryptoProvider aeadEncrypt/aeadDecrypt
to use the AAD-aware methods instead of ignoring the AAD parameter.
The EncryptWithLabel test still fails due to an X25519 DH computation
discrepancy between Python reference and the Quartz JVM implementation.
The HPKE implementation is internally consistent (MlsGroupTest passes).
https://claude.ai/code/session_01NocQDWj2Y92FugjfgazzL3
Adds full NIP-BE implementation for BLE-based Nostr peer-to-peer
messaging and synchronization with KMP support across all targets.
Protocol layer (commonMain):
- BleConfig: NIP-BE constants (service UUID, characteristic UUIDs)
- BleMessageChunker: DEFLATE compression + chunk splitting/joining
- BleRole/assignRole: Role assignment based on UUID comparison
- BleTransport: Platform-agnostic BLE transport interface
- BleNostrClient: IRelayClient implementation over BLE
- BleNostrServer: Relay server over BLE with notification support
- BleMeshManager: High-level mesh manager with auto-discovery,
role assignment, and event broadcasting
- BleChunkAssembler: Thread-safe chunk reassembly
Platform support:
- Deflate expect/actual for jvmAndroid, apple, and linux targets
- AndroidBleTransport: Full Android BLE implementation using GATT
server/client APIs with scanning, advertising, and MTU negotiation
Tests: Deflate compression, message chunking, role assignment,
and chunk assembly (25 tests passing).
https://claude.ai/code/session_01Tz5E73Rj7tL48A3qUGS5DT
Defers string construction until after the level check, avoiding
allocation when debug logging is filtered in release/benchmark builds.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Deprecated APIs in Quartz are kept for library consumers but internal usages
now carry @Suppress annotations so the module builds warning-free. Also
replaces redundant .toInt() on hex Int literals in ChaCha20Core and migrates
deprecated java.net.URL to URI.resolve() in ServerInfoParser.
The spotless license header delimiter is updated to recognize @file: annotations.
https://claude.ai/code/session_01EwS56YAGGnnac5EuwhaUSs
Add support for NIP-77 (Negentropy Syncing) protocol messages using
the negentropy-kmp library for efficient set reconciliation between
client and relay.
New files:
- NegOpenCmd, NegMsgCmd, NegCloseCmd: Client-to-relay commands
- NegMsgMessage, NegErrMessage: Relay-to-client messages
- NegentropySession: Client-side reconciliation orchestrator
- NegentropyServerSession: Server-side reconciliation handler
- NegentropyManager: High-level sync manager with listener callbacks
Updated serializers:
- Jackson MessageSerializer/Deserializer for NEG-MSG, NEG-ERR
- Jackson CommandSerializer/Deserializer for NEG-OPEN, NEG-MSG, NEG-CLOSE
- Kotlin MessageKSerializer/CommandKSerializer for all NIP-77 types
Compatible with strfry relay's negentropy implementation via the
negentropy-kmp library which implements the same Protocol V1 spec.
https://claude.ai/code/session_01Dc6W1G1jURAAR9kzyVvk6g
The field no longer means 'trust all certificates' — since the Samsung
One UI 7 fix, it means 'use the pinned trust store (hardcoded +
TOFU-pinned certs + system CAs) instead of system-only trust.'
The old name was actively misleading and could cause a future
contributor to interpret the flag literally, potentially reintroducing
the trust-all pattern that Samsung Knox and GrapheneOS reject.
Renamed across all 4 files:
- ElectrumxServer.kt: field definition + updated KDoc
- ElectrumXClient.kt: createSocket() usage + comments
- NamecoinSettings.kt: parseServerString() usage + comments
- NamecoinSettingsTest.kt: test assertions
Address compatibility and security issues identified by reviewing
GrapheneOS source (hardened Conscrypt, strict TLS enforcement):
1. .onion: prefer pinned factory, fall back to trust-all
GrapheneOS patches Conscrypt with stricter TLS enforcement that may
reject no-op X509TrustManagers even for proxied sockets. The .onion
server's cert is already in PINNED_ELECTRUMX_CERTS (same operator as
electrumx.testls.space), so we now use cachedPinnedSslFactory() as
the primary path. onionSslFactory() (trust-all) is kept as a fallback
on SSLHandshakeException only — this handles cert rotation or unknown
.onion servers gracefully.
2. TOFU: require explicit user confirmation before pinning
Previously, Test Connection auto-pinned every cert on success. An
attacker performing MITM during first test would get their cert
permanently trusted. Now each new cert triggers an AlertDialog showing
the full SHA-256 fingerprint. Users must explicitly accept ('Trust')
or reject each cert — aligning with GrapheneOS's philosophy of
explicit trust decisions.
3. Clarify trustAllCerts semantics
The field name is misleading post-refactor: it now means 'use pinned
trust store' not 'trust all certificates'. Added TODO to rename to
usePinnedTrustStore and updated inline comments to prevent future
misinterpretation.
Note: hostname verification on raw SSLSocket is a pre-existing gap
(not introduced by the Samsung fix PR) — SSLSocket.createSocket() uses
the host parameter for SNI only, not hostname verification. A follow-up
should add endpointIdentificationAlgorithm='HTTPS' or fingerprint-based
verification for pinned certs.
Connected to the .onion hidden service via Tor SOCKS proxy and
confirmed it serves the SAME certificate as electrumx.testls.space:
SHA-256: 53:65:D5:BB:26:19:F5:40:1C:D8:8E:FC:AF:FB:A5:B2:...
The .onion cert is already covered by the first pinned cert entry.
Added comments documenting this relationship and instructions for
fetching .onion certs via Tor for future updates.
The .onion still uses onionSslFactory() (trust-all) as the primary
path — this is correct since Tor provides its own authentication.
The pinned cert serves as defense-in-depth.
Tor hidden services are authenticated by their onion address (the
public key hash), making TLS certificate verification redundant.
The .onion server in TOR_ELECTRUMX_SERVERS was going through
pinnedSslFactory() but its cert isn't in the pinned list (and can't
be fetched without Tor). This would cause .onion connections to fail
with SSLHandshakeException when Tor mode is active.
Fix: .onion addresses now use a dedicated onionSslFactory() with
trust-all, which is safe because:
1. Tor provides end-to-end encryption
2. The onion address IS the server identity proof
3. Samsung Knox trust-all rejection doesn't apply to proxied sockets
When a user adds a custom ElectrumX server and runs Test Connection,
the server's TLS certificate is automatically captured and pinned
(Trust On First Use). This allows custom servers with self-signed
certificates to work on Samsung/Xiaomi/OnePlus devices.
Changes:
- ElectrumXClient: addPinnedCert(), setDynamicCerts() for runtime
cert management; testServer() now captures server cert PEM and
SHA-256 fingerprint from SSL session
- NamecoinSharedPreferences: persist pinned certs to DataStore
- AppModules: load pinned certs on startup, sync to ElectrumXClient
- NamecoinSettings: custom servers always set trustAllCerts=true
(ElectrumX servers almost universally use self-signed certs)
- UI: shows cert fingerprint in test results, auto-pins on success
- ServerTestResult: new serverCertPem + certFingerprint fields
Flow: Add server → Test Connection → cert auto-captured and pinned →
future connections trust that cert even on Samsung Knox devices.
Add per-server connection testing with detailed error reporting:
- Test Connection button tests each ElectrumX server individually
- Shows streaming results: ✅ success with response time, ❌ failure
with human-readable error (TLS handshake failed, connection refused,
timeout, DNS failed, invalid response, etc.)
- Diagnostic card shows: last test timestamp + success count, device
info (manufacturer/model/Android/API), TLS version negotiated
- ElectrumXClient.testServer() method for single-server diagnostics
with TLS version capture from SSL session
This gives users (and bug reporters) immediate visibility into why
Namecoin resolution may be failing silently on their device.
Samsung One UI 7 (Android 16) silently rejects TLS connections that use
a no-op X509TrustManager which accepts all certificates. This breaks
Namecoin resolution on all Samsung devices running One UI 7, including
Galaxy A15 (SM-A156E) and Galaxy S24 Ultra (SM-S938B).
Replace trustAllSslFactory() with pinnedSslFactory() that:
- Pins the actual self-signed certificates of known ElectrumX servers
- Also includes system CA certificates for servers with real certs
- Uses a proper TrustManagerFactory chain that Samsung Knox accepts
Additional OEM compatibility hardening:
- Cache the SSLSocketFactory (avoid expensive rebuild per connection)
- Request TLSv1.2 explicitly (Xiaomi MIUI/HyperOS and OnePlus ColorOS
Conscrypt forks may default to TLS 1.0 for raw socket upgrades)
- Enforce TLSv1.2+ enabled protocols on the SSLSocket
- KeyStore fallback to PKCS12 if default type fails (Xiaomi)
- Defensive try/catch on system CA cert re-insertion (some OEMs return
certs that cannot be added to a new KeyStore)
Tested on Android 16 (API 36) emulator — all ElectrumX servers connect
(hostname, IP-address, factory reuse) and .bit resolution works E2E.
Pinned certs:
- electrumx.testls.space:50002 (expires 2027-05-04)
- nmc2.bitcoins.sk:57002 (expires 2030-10-22)
ChessEngine.makeMove() now normalizes SAN notation before passing to
chesslib, fixing games against jester.nyo.dev and other clients that
use different notation:
- Castling: 0-0 → O-O, 0-0-0 → O-O-O (zeros to letters)
- Annotations: strips !, ?, !!, ??, !?, ?! suffixes
Previously, chesslib rejected 0-0/0-0-0 causing ChessStateReconstructor
to mark games as desynced and skip remaining moves. Games appeared
permanently stuck, even after app restart.
Also fixes discoverUserGames() silently dropping games classified as
spectator — now adds them to spectatingGames instead.
23 new interop tests covering SAN variants, reconstruction, spectator
detection, and missing start event handling.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add get_budget, sign_message, create_connection request/response types
- Add NwcTransactionType (incoming/outgoing) and NwcTransactionState
(PENDING/SETTLED/FAILED/ACCEPTED) constants
- Add missing error codes: BAD_REQUEST, NOT_FOUND, EXPIRED
- Add settle_deadline field to NwcTransaction
- Add total_count to ListTransactionsResult
- Add metadata and lud16 to GetInfoResult
- Add unpaid_outgoing and unpaid_incoming to ListTransactionsParams
- Update Jackson deserializers for new method types
- Add AlbyInteropTest with 25+ tests verifying compatibility with
Alby Hub's JSON formats for all request/response types
- Update existing tests for new constants and error codes
https://claude.ai/code/session_01JFogiwyR4CPVP8cRznqahJ
Two bugs in quartz's NIP-46 remote signer:
1. FixedKey.getEncoded() returned the raw byte array reference. javax.crypto.Mac
zeroes the key via destroy() after HMAC computation, corrupting the NIP-44
saltPrefix ("nip44-v2") after the first computeConversationKey call. All
subsequent NIP-44 decryptions with different key pairs fail with Invalid Mac.
Fix: return key.copyOf() instead of key.
2. NostrSignerRemote.connect() only handled pubkey responses. Amber (and other
signers) may respond with "ack" or "already connected" error. Added
ConnectResponse parser that maps these to success and falls back to
getPublicKey() to retrieve the actual pubkey.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* 'main' of https://github.com/vitorpamplona/amethyst: (35 commits)
New Crowdin translations by GitHub Action
fix chess tests
Replace subsequent checks with '!isNullOrEmpty()' call
use forEach: no intermediate list created.
Use filterIsInstance to avoid dual iteration
Lambda argument should be moved out of parentheses
replace null checks with Elvis
merge call chain with flatMap (slight performance improvement)
New Crowdin translations by GitHub Action
New Crowdin translations by GitHub Action
update cz, pt, de, sv
was this meant to be withContext?
add names to boolean params
remove unused imports
Explicit type arguments can be inferred
remove redundant modifiers
correct package name
If-Null return/break/... foldable to '?:'
Declaration has type inferred from a platform call, which can lead to unchecked nullability issues. Specify type explicitly as nullable or non-nullable.
Declaration has type inferred from a platform call, which can lead to unchecked nullability issues. Specify type explicitly as nullable or non-nullable.
...
# Conflicts:
# amethyst/src/main/java/com/vitorpamplona/amethyst/ui/screen/loggedIn/settings/ReactionsSettingsScreen.kt
Add NamecoinLookupException sealed class to distinguish:
- NameNotFound: name queried successfully but doesn't exist on blockchain
- NameExpired: name exists but expired (>36000 blocks since last update)
- ServersUnreachable: all ElectrumX servers failed with connection errors
Previously, all three cases returned null from nameShowWithFallback,
making it impossible for the UI to show meaningful feedback.
Changes:
- ElectrumxClient.connectAndQuery: throws NameNotFound/NameExpired
instead of returning null for definitive blockchain answers
- ElectrumxClient.nameShow: lets NamecoinLookupException propagate
(only catches connection/IO errors as null)
- ElectrumxClient.nameShowWithFallback: short-circuits on NameNotFound/
NameExpired (no point trying other servers for definitive answers),
throws ServersUnreachable when all servers fail with connection errors
- SearchBarViewModel: new namecoinSearchState flow with typed states
(Idle/Loading/Resolved/NotFound/Expired/ServersUnreachable) for UI
to show loading spinners, resolved profiles, and specific error
messages. Derived namecoinResolvedUser from this flow.
Discovered while porting to notedeck (damus-io/notedeck#1314).
Fixes discovered while porting Namecoin NIP-05 to Primal Android
(PrimalHQ/primal-android-app#934):
1. Race condition: stale intermediate lookups overwrite final result
- Changed map to mapLatest in SearchBarViewModel so that previous
in-flight ElectrumX lookups are cancelled when the query changes
- Without this, typing 'd/testls' character by character causes
stale results from 'd/test', 'd/testl' etc. to arrive after
the correct 'd/testls' result and overwrite it with null
2. CancellationException swallowed in catch block
- mapLatest cancels previous coroutines via CancellationException
- The generic catch was catching it and emitting null, wiping the
valid result. Now re-throws CancellationException
3. Root lookup fails when names object has no underscore key
- extractFromDomainValue tried names[localPart] then names[underscore]
which is the same thing for root lookups
- Now falls back to first available entry for root lookups when
no underscore key exists
- Non-root lookups still fail correctly (no false matches)
4. Global Mutex serializes all lookups
- Single Mutex in ElectrumxClient blocked unrelated lookups to
different servers. A 25s timeout on one server stalled everything
- Now uses per-server mutexes via ConcurrentHashMap
5. customServers set but never read
- NamecoinNameService.setCustomServers() stored the list but the
resolver was constructed with the default serverListProvider
- Now wires customServers through to the resolver
6. resolveLive uses orphaned CoroutineScope
- Now accepts an optional external scope parameter so callers can
tie resolution to their own lifecycle
Adds 2 new tests: root fallback to first entry, non-root no-fallback.
- ChessPosition: include all fields in equals/hashCode
- AcceptedGamesRegistry: add thread safety with synchronized
- ChessEventBroadcaster: use valid filter for connection trigger
- ChessEventBroadcaster: fix misleading relayResults
- Remove println debug logging from ChessSubscription/Broadcaster
- UserProfileScreen: use proper JSON parsing via MetadataEvent
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add ChessLobbyState with completedGames, replaceGameState(), moveToCompleted()
- Add challengerAvatarUrl to ChessChallenge
- Add CompletedGame data class for game history
Platform Adapters:
- AndroidChessAdapter: AndroidChessPublisher, AndroidRelayFetcher, AndroidMetadataProvider
- DesktopChessAdapter: DesktopChessPublisher, DesktopRelayFetcher, DesktopMetadataProvider
Shared Infrastructure:
- ChessRelayFetchHelper for one-shot relay queries
- IUserMetadataProvider interface for platform-specific metadata
- ChessFilterBuilder with simple filter methods for fetchers
- ChessSubscriptionController interface for subscription management
Fix ChessSubscription.kt to remove broken dataSources().chess reference
(subscriptions now managed by ChessLobbyLogic)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Claude