Compare commits

..
Author SHA1 Message Date
jackandClaude Fable 5 5e69a19cb3 Document the V3 transport architecture and remaining roadmap
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 00:58:20 +01:00
jackandClaude Fable 5 5c2903ec7c Extract mesh-ping diagnostics state into a pure engine-confined tracker
First slice of the feature-module direction: BLEMeshPingTracker owns the
outstanding-probe map and the per-link inbound response budget as pure
state (register/resolve/expire/reset), so the security invariants — a
pong only resolves against the probed peer, the budget keys on the
ingress link because claimed senders are forgeable, panic reset drops
probes and budget together — are now unit-tested without queues or
radios. The transport keeps only packet I/O, timers, and main-actor
delivery around it.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 00:57:26 +01:00
jackandClaude Fable 5 cbfdb9696e Wire gateway/bridge/panic features to capability ports, not BLEService
App wiring discovered mesh-only features by casting the Transport to
the concrete BLEService class in nine places. Those surfaces are now
three capability protocols — BluetoothStateReporting,
PanicResettingTransport, and MeshBridgingTransport — discovered with
as? like any optional capability, so the bootstrapper, panic flow, and
lifecycle coordinator no longer name the concrete transport at all. A
future second mesh transport picks up gateway/bridge wiring and the
panic lifecycle by conforming, and the remaining Transport god-protocol
requirements can migrate to the same pattern.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 00:54:49 +01:00
jackandClaude Fable 5 87cbf5a03c Unify the message and collections queues into one serial engine queue
The old model ran a concurrent message queue over a second concurrent
collections queue whose barrier flags served as the real mutual
exclusion — every field carried an ownership comment, and correctness
lived in per-site discipline. The message queue is now a single serial
engine queue that owns all mesh protocol state; the collections queue,
its 98 sync/async hops, and every barrier flag are gone. Cross-thread
callers go through onEngine, which documents and (in debug) enforces
the transport's sync-edge order: main and test threads may block on the
engine, the engine may block on bleQueue and the crypto/identity
queues, and nothing may block the other way.

The debug trap caught two latent inversions the leaf-lock structure had
been masking: the verified-announce rebind path re-resolved the ingress
link through the engine from inside its bleQueue critical section (it
now receives the already-resolved link), and the noise
session-generation closures sync-re-entered the engine from the noise
manager's queue while their own engine slot was blocked on it (they now
touch engine state directly, which the held slot makes exclusive).

BLE throughput is orders of magnitude below what one serial queue
sustains; the full suite runs at identical speed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 00:49:27 +01:00
jackandClaude Fable 5 d2a47d1ade Stop bleQueue maintenance and status paths from blocking on collectionsQueue
The traffic-burst tracker becomes a lock-backed monitor (written by the
receive pipeline, read by scan-duty adaptation and announce pacing on
bleQueue), the status-log peer summary and topology refresh read the
already lock-backed registry directly, and the stalled-fragment reap
moves to an async collections hop with the gossip resync request inside
it. bleQueue no longer sync-waits on the collections queue anywhere.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 00:39:41 +01:00
jackandClaude Fable 5 0bde6d21f7 Move BLE link egress/ingress buffers to bleQueue ownership
pendingPeripheralWrites, pendingNotifications, and pendingWriteBuffers
were collectionsQueue-guarded, but every producer and drain already runs
on bleQueue next to the CoreBluetooth objects they feed — each access
paid a cross-queue barrier for state that never leaves the radio thread,
and the notification drain even invoked peripheralManager.updateValue
from the collections queue. They are now bleQueue-confined like the link
state store: CB delegate callbacks and drains touch them directly, and
the few engine-side entry points hop to bleQueue (the direction the
transport's sync-edge order already allows). This clears most
bleQueue-to-collectionsQueue sync edges ahead of merging the collections
queue into the message queue.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 00:36:26 +01:00
jackandClaude Fable 5 f718097c74 Make peer registry and local announce state lock-backed
The main actor answered isPeerConnected/peerNickname/currentPeerSnapshots
and flipped runtime capability bits by blocking on collectionsQueue behind
whatever transport work was in flight. Peer state now lives in a
lock-backed BLEPeerRegistryStore (every registry mutation is a single
whole-transition method, so readers never observe a torn state), and the
runtime capability bits move into BLELocalIdentityStateStore next to the
identity they ride announces with. No transport entry point called from
the main actor blocks on a transport queue for peer state anymore.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-27 00:29:04 +01:00
12 changed files with 922 additions and 468 deletions
@@ -5,6 +5,20 @@ struct BLELocalIdentitySnapshot: Equatable, Sendable {
let peerID: PeerID
let peerIDData: Data
let nickname: String
/// Runtime-toggled capability bits (e.g. the internet-gateway toggle)
/// ORed into `PeerCapabilities.localSupported` for every announce.
let runtimeCapabilities: PeerCapabilities
/// Rendezvous cell advertised while bridging; rides announces only
/// while the `.bridge` capability is enabled.
let bridgeGeohash: String?
var advertisedCapabilities: PeerCapabilities {
PeerCapabilities.localSupported.union(runtimeCapabilities)
}
var advertisedBridgeGeohash: String? {
runtimeCapabilities.contains(.bridge) ? bridgeGeohash : nil
}
}
/// Lock-backed local identity state shared by the transport's message,
@@ -12,8 +26,8 @@ struct BLELocalIdentitySnapshot: Equatable, Sendable {
///
/// `peerID` and its binary wire representation must change as one unit during
/// panic rotation. A snapshot also gives announce construction one consistent
/// view of the nickname and identity instead of reading three independently
/// mutable properties across queues.
/// view of the nickname, identity, and advertised capabilities instead of
/// reading independently mutable properties across queues.
final class BLELocalIdentityStateStore: @unchecked Sendable {
private let lock = NSLock()
private var state: BLELocalIdentitySnapshot
@@ -25,7 +39,9 @@ final class BLELocalIdentityStateStore: @unchecked Sendable {
state = BLELocalIdentitySnapshot(
peerID: peerID,
peerIDData: Data(hexString: peerID.id) ?? Data(),
nickname: nickname
nickname: nickname,
runtimeCapabilities: [],
bridgeGeohash: nil
)
}
@@ -38,7 +54,9 @@ final class BLELocalIdentityStateStore: @unchecked Sendable {
state = BLELocalIdentitySnapshot(
peerID: state.peerID,
peerIDData: state.peerIDData,
nickname: nickname
nickname: nickname,
runtimeCapabilities: state.runtimeCapabilities,
bridgeGeohash: state.bridgeGeohash
)
}
}
@@ -48,8 +66,48 @@ final class BLELocalIdentityStateStore: @unchecked Sendable {
state = BLELocalIdentitySnapshot(
peerID: peerID,
peerIDData: Data(hexString: peerID.id) ?? Data(),
nickname: state.nickname
nickname: state.nickname,
runtimeCapabilities: state.runtimeCapabilities,
bridgeGeohash: state.bridgeGeohash
)
}
}
/// Flips a runtime capability bit. Returns whether anything changed.
@discardableResult
func setCapability(_ capability: PeerCapabilities, enabled: Bool) -> Bool {
lock.withLock {
var capabilities = state.runtimeCapabilities
if enabled {
capabilities.insert(capability)
} else {
capabilities.remove(capability)
}
guard capabilities != state.runtimeCapabilities else { return false }
state = BLELocalIdentitySnapshot(
peerID: state.peerID,
peerIDData: state.peerIDData,
nickname: state.nickname,
runtimeCapabilities: capabilities,
bridgeGeohash: state.bridgeGeohash
)
return true
}
}
/// Sets the bridged rendezvous cell. Returns whether anything changed.
@discardableResult
func setBridgeGeohash(_ cell: String?) -> Bool {
lock.withLock {
guard cell != state.bridgeGeohash else { return false }
state = BLELocalIdentitySnapshot(
peerID: state.peerID,
peerIDData: state.peerIDData,
nickname: state.nickname,
runtimeCapabilities: state.runtimeCapabilities,
bridgeGeohash: cell
)
return true
}
}
}
@@ -0,0 +1,62 @@
import BitFoundation
import Foundation
struct BLEMeshPingProbe {
let peerID: PeerID
let sentAt: Date
let lifecycleGeneration: UInt64
let completion: @MainActor (MeshPingResult?) -> Void
let timeout: DispatchWorkItem
}
/// Engine-confined /ping diagnostics state: outstanding probes keyed by
/// their unguessable nonce, plus the inbound response budget.
///
/// The budget is keyed by the ingress link (the directly connected peer
/// that delivered the packet), never the packet-claimed sender: pings are
/// unsigned, so the claimed sender is attacker-controlled and rotating it
/// would reset the budget, turning a directed unencrypted probe into an
/// amplification primitive.
///
/// Pure state the transport owns packet I/O, timers, and main-actor
/// completion delivery around it.
struct BLEMeshPingTracker {
private var pendingProbes: [Data: BLEMeshPingProbe] = [:]
private var responseLimiter = SyncResponseRateLimiter(
maxResponses: TransportConfig.meshPingInboundMaxPerLink,
window: TransportConfig.meshPingInboundWindowSeconds
)
mutating func register(_ probe: BLEMeshPingProbe, nonce: Data) {
pendingProbes[nonce] = probe
}
/// Resolves a pong against its outstanding probe. The echoed nonce plus
/// the sender check bind the reply to the probed peer.
mutating func resolve(nonce: Data, from peerID: PeerID) -> BLEMeshPingProbe? {
guard pendingProbes[nonce]?.peerID == peerID else { return nil }
return pendingProbes.removeValue(forKey: nonce)
}
/// Removes a timed-out probe so its completion can fire once with nil.
mutating func expire(nonce: Data) -> BLEMeshPingProbe? {
pendingProbes.removeValue(forKey: nonce)
}
/// Whether an inbound ping delivered by this link is within budget.
mutating func shouldRespond(toLink linkPeerID: PeerID, now: Date) -> Bool {
responseLimiter.shouldRespond(to: linkPeerID, now: now)
}
/// Drops all probes and restores a fresh response budget (panic wipe).
/// Returns the orphaned timeout work items for the caller to cancel.
mutating func reset() -> [DispatchWorkItem] {
let timeouts = pendingProbes.values.map(\.timeout)
pendingProbes.removeAll()
responseLimiter = SyncResponseRateLimiter(
maxResponses: TransportConfig.meshPingInboundMaxPerLink,
window: TransportConfig.meshPingInboundWindowSeconds
)
return timeouts
}
}
@@ -0,0 +1,87 @@
import BitFoundation
import Foundation
/// Lock-backed shared ownership of the peer registry, readable from any
/// queue or the main actor without hopping onto a transport queue.
///
/// Mutations stay serialized by the transport (they only run on its
/// queues), so the lock's job is to let the main actor answer questions
/// like `isPeerConnected` without blocking behind in-flight transport
/// work. Every `BLEPeerRegistry` mutation is a single whole-transition
/// method, so a reader between two mutations always observes a valid
/// pre- or post-state, never a torn one.
///
/// Closures passed to `read`/`mutate` run under the (non-recursive) lock
/// and must not call back into the store.
final class BLEPeerRegistryStore: @unchecked Sendable {
private let lock = NSLock()
private var registry = BLEPeerRegistry()
/// One consistent view across multiple registry reads.
func read<T>(_ body: (BLEPeerRegistry) -> T) -> T {
lock.withLock { body(registry) }
}
func mutate<T>(_ body: (inout BLEPeerRegistry) -> T) -> T {
lock.withLock { body(&registry) }
}
// MARK: - Single-question reads
var isEmpty: Bool { read { $0.isEmpty } }
var count: Int { read { $0.count } }
var peerIDs: [PeerID] { read { $0.peerIDs } }
var connectedCount: Int { read { $0.connectedCount } }
var connectedPeerIDs: [PeerID] { read { $0.connectedPeerIDs } }
var connectedRoutingData: [Data] { read { $0.connectedRoutingData } }
var snapshotByID: [PeerID: BLEPeerInfo] { read { $0.snapshotByID } }
func info(for peerID: PeerID) -> BLEPeerInfo? {
read { $0.info(for: peerID) }
}
func isConnected(_ peerID: PeerID) -> Bool {
read { $0.isConnected(peerID) }
}
func isReachable(_ peerID: PeerID, now: Date) -> Bool {
read { $0.isReachable(peerID, now: now) }
}
func nickname(for peerID: PeerID, connectedOnly: Bool) -> String? {
read { $0.nickname(for: peerID, connectedOnly: connectedOnly) }
}
func fingerprint(for peerID: PeerID) -> String? {
read { $0.fingerprint(for: peerID) }
}
func capabilities(for peerID: PeerID) -> PeerCapabilities {
read { $0.capabilities(for: peerID) }
}
func capabilitiesWereExplicitlyAdvertised(for peerID: PeerID) -> Bool {
read { $0.capabilitiesWereExplicitlyAdvertised(for: peerID) }
}
func advertisedBridgeGeohash() -> String? {
read { $0.advertisedBridgeGeohash() }
}
func displayNicknames(selfNickname: String) -> [PeerID: String] {
read { $0.displayNicknames(selfNickname: selfNickname) }
}
func transportSnapshots(selfNickname: String) -> [TransportPeerSnapshot] {
read { $0.transportSnapshots(selfNickname: selfNickname) }
}
/// Peers advertising `capability` that are reachable now, in one
/// consistent view.
func reachablePeers(advertising capability: PeerCapabilities, now: Date) -> [PeerID] {
read { registry in
registry.peers(advertising: capability)
.filter { registry.isReachable($0, now: now) }
}
}
}
@@ -77,6 +77,26 @@ struct BLEReceivePipeline {
}
}
/// Lock-backed traffic-level signal: the receive pipeline records packets,
/// and the radio layer (maintenance and scan-duty adaptation on bleQueue)
/// reads the level without crossing onto a transport queue.
final class BLERecentTrafficMonitor: @unchecked Sendable {
private let lock = NSLock()
private var tracker = BLERecentTrafficTracker()
func recordPacket(at now: Date) {
lock.withLock { tracker.recordPacket(at: now) }
}
func hasTraffic(within seconds: TimeInterval, now: Date) -> Bool {
lock.withLock { tracker.hasTraffic(within: seconds, now: now) }
}
func removeAll() {
lock.withLock { tracker.removeAll() }
}
}
struct BLERecentTrafficTracker: Equatable {
private var packetTimestamps: [Date] = []
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,61 @@
import BitFoundation
import CoreBluetooth
import Foundation
/// Optional transport capabilities, discovered with `as?` instead of casting
/// to a concrete transport class. `Transport` stays the contract every
/// transport genuinely implements; a capability protocol here is the
/// contract for one mesh-only feature surface, so app wiring depends on the
/// feature it needs rather than on `BLEService` itself.
/// Radio-state reporting for transports backed by a local radio.
protocol BluetoothStateReporting: AnyObject {
func getCurrentBluetoothState() -> CBManagerState
}
/// Panic-mode lifecycle for transports that own durable identity state.
/// A transport implementing this owns its own restart sequencing:
/// `completePanicReset` decides whether services come back, so generic
/// `startServices()` calls after a panic belong only to transports that
/// don't implement it.
protocol PanicResettingTransport: AnyObject {
/// Quiesces the radio and drains in-flight work ahead of a panic wipe.
func suspendForPanicReset()
/// Finishes a panic wipe, optionally restarting services.
func completePanicReset(restartServices: Bool)
/// Rotates the transport identity as part of a panic reset.
func resetIdentityForPanic(currentNickname: String, restartServices: Bool)
}
/// Internet-gateway and geohash-bridge wiring surface (BLE mesh today).
/// Everything the gateway/bridge/courier services need from the mesh
/// transport, so their bootstrap wiring never touches the concrete class.
protocol MeshBridgingTransport: AnyObject {
// Runtime-advertised capability bits
func setLocalCapability(_ capability: PeerCapabilities, enabled: Bool)
func setLocalBridgeGeohash(_ cell: String?)
func advertisedBridgeGeohash() -> String?
// Peers currently advertising bridging roles
func reachableGatewayPeers() -> [PeerID]
func reachableBridgePeers() -> [PeerID]
// Gateway carrier packets (mesh <-> Nostr uplink/downlink)
@discardableResult
func sendNostrCarrier(_ payload: Data, to gatewayPeer: PeerID) -> Bool
func broadcastNostrCarrier(_ payload: Data)
/// Sink for received carrier packets (set once by app wiring; called on
/// the main actor after transport-level checks).
var onNostrCarrierPacket: (@MainActor (_ payload: Data, _ from: PeerID, _ directedToUs: Bool) -> Void)? { get set }
// Bridge courier drops (sealed envelopes carried across the bridge)
func sealBridgeCourierEnvelope(_ content: String, messageID: String, recipientNoiseKey: Data) -> CourierEnvelope?
@discardableResult
func openBridgedCourierEnvelope(_ envelope: CourierEnvelope) -> Bool
@discardableResult
func deliverBridgedEnvelope(_ envelope: CourierEnvelope, to peerID: PeerID) -> Bool
func myNoiseStaticPublicKey() -> Data
func verifiedPeersWithNoiseKeys() -> [(peerID: PeerID, noiseKey: Data)]
/// Fired (off-main) when a signature-verified announce is processed.
var onVerifiedPeerAnnounce: ((_ peerID: PeerID) -> Void)? { get set }
}
+3
View File
@@ -450,3 +450,6 @@ extension BitchatDelegate {
}
extension BLEService: Transport {}
extension BLEService: BluetoothStateReporting {}
extension BLEService: PanicResettingTransport {}
extension BLEService: MeshBridgingTransport {}
@@ -106,8 +106,8 @@ extension ChatViewModel: ChatLifecycleContext {
}
func refreshBluetoothState() {
if let bleService = meshService as? BLEService {
updateBluetoothState(bleService.getCurrentBluetoothState())
if let radio = meshService as? BluetoothStateReporting {
updateBluetoothState(radio.getCurrentBluetoothState())
}
}
+9 -8
View File
@@ -1564,8 +1564,8 @@ final class ChatViewModel: ObservableObject, BitchatDelegate, SynchronousMessage
// Quiesce the mesh before clearing stores. Identity replacement below
// deliberately stays stopped until media deletion and marker commit.
if let bleService = meshService as? BLEService {
bleService.suspendForPanicReset()
if let panicTransport = meshService as? PanicResettingTransport {
panicTransport.suspendForPanicReset()
} else {
meshService.emergencyDisconnectAll()
}
@@ -1700,8 +1700,8 @@ final class ChatViewModel: ObservableObject, BitchatDelegate, SynchronousMessage
// Replace the BLE identity while keeping the radio stopped. It may
// reopen only after the durable panic transaction commits.
if let bleService = meshService as? BLEService {
bleService.resetIdentityForPanic(
if let panicTransport = meshService as? PanicResettingTransport {
panicTransport.resetIdentityForPanic(
currentNickname: nickname,
restartServices: false
)
@@ -1746,18 +1746,19 @@ final class ChatViewModel: ObservableObject, BitchatDelegate, SynchronousMessage
guard panicCompleted else { return false }
if let bleService = meshService as? BLEService {
if let panicTransport = meshService as? PanicResettingTransport {
// Startup recovery reopens admission but leaves actual service
// start to the bootstrapper immediately after this method.
bleService.completePanicReset(
panicTransport.completePanicReset(
restartServices: restartServices
)
}
if restartServices {
// All persistent state and media are gone. Bring each service back
// only now, under the new identity.
if !(meshService is BLEService) {
// only now, under the new identity a panic-resetting transport
// owns its own restart sequencing above.
if !(meshService is PanicResettingTransport) {
meshService.startServices()
}
@@ -195,9 +195,8 @@ private extension ChatViewModelBootstrapper {
DispatchQueue.main.asyncAfter(deadline: .now() + 0.1) { [weak viewModel] in
guard let viewModel,
let bleService = viewModel.meshService as? BLEService else { return }
let state = bleService.getCurrentBluetoothState()
viewModel.updateBluetoothState(state)
let radio = viewModel.meshService as? BluetoothStateReporting else { return }
viewModel.updateBluetoothState(radio.getCurrentBluetoothState())
}
viewModel.nostrRelayManager = NostrRelayManager.shared
@@ -331,7 +330,7 @@ private extension ChatViewModelBootstrapper {
func configureGateway() {
// Gateway mode bridges BLE mesh <-> Nostr; a mock transport (tests)
// has no carrier packets to bridge.
guard let bleService = viewModel.meshService as? BLEService else { return }
guard let bleService = viewModel.meshService as? MeshBridgingTransport else { return }
let gateway = GatewayService.shared
gateway.publishToRelays = { event, geohash in
@@ -410,7 +409,7 @@ private extension ChatViewModelBootstrapper {
/// transport, the relay manager, location, and the public timeline. Same
/// closure-injection style as `configureGateway`.
func configureBridge() {
guard let bleService = viewModel.meshService as? BLEService else { return }
guard let bleService = viewModel.meshService as? MeshBridgingTransport else { return }
let bridge = BridgeService.shared
let idBridge = viewModel.idBridge
@@ -545,7 +544,7 @@ private extension ChatViewModelBootstrapper {
/// manager, the mesh transport's sealing/opening primitives, the courier
/// store, and the message router's deposit path.
func configureBridgeCourier() {
guard let bleService = viewModel.meshService as? BLEService else { return }
guard let bleService = viewModel.meshService as? MeshBridgingTransport else { return }
let courier = BridgeCourierService.shared
courier.bridgeEnabled = { BridgeService.shared.isEnabled }
@@ -0,0 +1,84 @@
import BitFoundation
import Foundation
import Testing
@testable import bitchat
struct BLEMeshPingTrackerTests {
private func makeProbe(peerID: PeerID) -> BLEMeshPingProbe {
BLEMeshPingProbe(
peerID: peerID,
sentAt: Date(timeIntervalSince1970: 1_000),
lifecycleGeneration: 1,
completion: { _ in },
timeout: DispatchWorkItem {}
)
}
@Test func resolveReturnsProbeOnlyForTheProbedPeer() {
var tracker = BLEMeshPingTracker()
let nonce = Data([1, 2, 3, 4, 5, 6, 7, 8])
let probed = PeerID(str: "aaaaaaaaaaaaaaaa")
tracker.register(makeProbe(peerID: probed), nonce: nonce)
// A pong claiming the right nonce from the wrong peer must not
// consume the probe.
let wrongPeer = tracker.resolve(nonce: nonce, from: PeerID(str: "bbbbbbbbbbbbbbbb"))
#expect(wrongPeer == nil)
let rightPeer = tracker.resolve(nonce: nonce, from: probed)
#expect(rightPeer != nil)
// Consumed exactly once.
let secondResolve = tracker.resolve(nonce: nonce, from: probed)
#expect(secondResolve == nil)
}
@Test func expireConsumesTheProbeSoResolveCannotFireTwice() {
var tracker = BLEMeshPingTracker()
let nonce = Data([9, 9, 9, 9, 9, 9, 9, 9])
let probed = PeerID(str: "aaaaaaaaaaaaaaaa")
tracker.register(makeProbe(peerID: probed), nonce: nonce)
let firstExpire = tracker.expire(nonce: nonce)
#expect(firstExpire != nil)
let secondExpire = tracker.expire(nonce: nonce)
#expect(secondExpire == nil)
let resolveAfterExpire = tracker.resolve(nonce: nonce, from: probed)
#expect(resolveAfterExpire == nil)
}
@Test func inboundBudgetIsPerLinkAndBounded() {
var tracker = BLEMeshPingTracker()
let now = Date(timeIntervalSince1970: 2_000)
let linkA = PeerID(str: "aaaaaaaaaaaaaaaa")
let linkB = PeerID(str: "bbbbbbbbbbbbbbbb")
var allowedOnA = 0
for _ in 0..<(TransportConfig.meshPingInboundMaxPerLink + 5) {
if tracker.shouldRespond(toLink: linkA, now: now) { allowedOnA += 1 }
}
#expect(allowedOnA == TransportConfig.meshPingInboundMaxPerLink)
// One saturated link must not consume another link's budget.
let allowedOnB = tracker.shouldRespond(toLink: linkB, now: now)
#expect(allowedOnB)
}
@Test func resetDropsProbesRestoresBudgetAndHandsBackTimeouts() {
var tracker = BLEMeshPingTracker()
let now = Date(timeIntervalSince1970: 3_000)
let link = PeerID(str: "aaaaaaaaaaaaaaaa")
let nonce = Data([4, 4, 4, 4, 4, 4, 4, 4])
tracker.register(makeProbe(peerID: link), nonce: nonce)
for _ in 0..<TransportConfig.meshPingInboundMaxPerLink {
_ = tracker.shouldRespond(toLink: link, now: now)
}
let saturated = tracker.shouldRespond(toLink: link, now: now)
#expect(!saturated)
let timeouts = tracker.reset()
#expect(timeouts.count == 1)
let resolveAfterReset = tracker.resolve(nonce: nonce, from: link)
#expect(resolveAfterReset == nil)
let allowedAfterReset = tracker.shouldRespond(toLink: link, now: now)
#expect(allowedAfterReset)
}
}
+149
View File
@@ -0,0 +1,149 @@
# BLE Transport Architecture V3
The plan of record for restructuring `BLEService` from an 8.3k-line god
object into a layered mesh stack. ARCHITECTURE_V2 rebuilt the app layer
above the transport and deliberately deferred the transport itself; this
document covers that remainder: what already landed, the target shape, and
the order for the rest.
## Why the satellite strategy stalled
V2's transport approach was to peel pure policies and closure-driven
handlers out of `BLEService` while the class kept coordinating. The ~30
pure policy structs were a clear win. The five big handler extractions
were not: each needed an "environment" of 2030 closures that weakly
capture the service and hop queues back into its state. Logic left, but
state ownership and synchronization never moved, so extraction paid a
plumbing tax that grew as fast as the logic shrank — the five
`make*HandlerEnvironment()` factories alone were ~1.5k lines. The file
held ~60 mutable fields across four concurrency domains whose ownership
lived in comments, and every new feature added Transport requirements,
state maps, and switch cases to the same class.
Two chronic costs came straight from that structure: queue-order
deadlocks (the July 9 main↔bleQueue ABBA freeze), and timing-dependent
tests (correctness only observable through real queues and real time).
## Target shape
A packet-radio stack with one rule per layer about state and threads:
1. **`BLELinkLayer`** — the only CoreBluetooth import. Owns both managers,
scanning/advertising, duty cycle, connection scheduling, MTU, write
and notification backpressure buffers, state restoration. Speaks
`LinkEvent` up (link up/down, bytes in, writable) and `LinkCommand`
down (send bytes on link, scan/advertise policy). Knows nothing about
packets, peers, or Noise. bleQueue-confined. A `SimulatedLinkLayer`
implementing the same port gives multi-node tests real topologies with
no radios and no wall-clock waits.
2. **Mesh engine** — one serial queue owning all protocol state: wire
codec, fragmentation, dedup, relay policy, peer registry, topology,
gossip sync, Noise orchestration. Synchronous single-writer logic; the
pure policy satellites slot in unchanged. Endgame: the engine core
becomes `handle(event, now) -> [Effect]` (sans-I/O), which makes the
whole mesh property-testable and fuzzable in simulation.
3. **Feature modules** — courier, board, prekeys, private media, file
transfer, voice, diagnostics, groups, verify/vouch each own their
state and register for their message types. A new feature is a new
module, not edits to the engine.
4. **App boundary** — a small `Transport` core both transports genuinely
implement, plus capability protocols discovered with `as?`
(`MeshBridgingTransport` etc.), replacing the ~90-requirement
god-protocol and its inert defaults.
### Concurrency contract
State is owned one of three ways:
- **Engine-confined** — mutated only on the serial engine queue
(`mesh.message`). Cross-thread callers use `onEngine`.
- **bleQueue-confined** — link-layer state next to CoreBluetooth objects
(link store, write/notification buffers, link-auth maps).
- **Lock-backed store** — state with legitimate cross-domain readers
(peer registry, local identity/capabilities, traffic monitor). Writes
still come from one domain; the lock exists so readers never block on
a queue. Every mutation is a single whole-transition method, so
readers never observe torn state.
Sync-edge order (deadlock freedom by construction, debug-enforced in
`onEngine`):
```
main / test threads ──sync──▶ engine ──sync──▶ bleQueue
└──sync──▶ noise / identity queues (leaves)
```
Nothing may sync-wait in the reverse direction: bleQueue and the crypto
queues reach the engine only via `async`, and nothing sync-dispatches to
main. Two subtleties worth knowing:
- A closure executed inside a noise-manager critical section entered
*from* an engine slot may touch engine state directly (the blocked
slot makes it exclusive) but must never sync-re-enter the engine —
that is a self-deadlock.
- bleQueue critical sections (e.g. the verified-announce link rebind)
must receive engine-derived values as arguments rather than fetching
them through `onEngine`.
## What landed in this pass
- **Lock-backed peer state** (`BLEPeerRegistryStore`): every main-actor
Transport read (`isPeerConnected`, nicknames, snapshots, capability
queries) reads a lock, not a queue. Runtime capability bits moved into
`BLELocalIdentityStateStore` beside the identity they ride announces
with.
- **bleQueue owns the link buffers**: `pendingPeripheralWrites`,
`pendingNotifications`, `pendingWriteBuffers` are bleQueue-confined
(their producers and drains already ran there); the notification drain
no longer invokes CoreBluetooth from a transport queue.
- **One serial engine queue**: the concurrent message queue and the
collections queue it guarded state with are one serial domain; every
barrier flag and per-field ownership comment deleted; ~98 cross-queue
hops removed. `onEngine` documents and debug-enforces the sync-edge
order — and its trap caught two latent inversions during migration
(the announce-rebind path and the noise session-generation closures).
- **Capability ports**: gateway/bridge/courier wiring, the panic
lifecycle, and radio-state reads go through `MeshBridgingTransport`,
`PanicResettingTransport`, and `BluetoothStateReporting`; no app code
casts to `BLEService` anymore.
- **First feature-owned state**: `BLEMeshPingTracker` holds the /ping
probe map and per-link response budget as pure state with unit tests —
the template for peeling the remaining features.
Full suite green throughout (1,953 tests), identical wall-clock — BLE
throughput is nowhere near what one serial queue sustains.
## Remaining roadmap (in order)
1. **Feature peeling.** Move each feature's state maps and handlers into
a module in the `BLEMeshPingTracker` mold: private media (six
generation-keyed maps + policy resolution — its main-actor reads
become a lock-backed store inside the module), courier, prekeys,
board, voice, file transfer, groups. The engine keeps a registry of
handled message types instead of a giant switch. Each module lands as
its own PR with its state's invariants unit-tested.
2. **Transport protocol split.** Continue what the capability ports
started: `Transport` shrinks to lifecycle + identity + snapshots +
basic messaging; files/voice/courier/board/diagnostics/verification
become capability protocols; `NostrTransport` drops its inert stubs;
coordinators declare the capability they need instead of receiving
the whole god-protocol (~48 call sites across 14 files).
3. **Link-layer extraction.** With the file slimmed, move the CB
delegates, scheduling, duty cycle, and buffers behind
`LinkEvent`/`LinkCommand` ports. Decide the link-auth boundary here:
`noiseAuthenticatedLinkOwners` and the rebind containment rules are
mesh security state that currently lives on bleQueue for atomicity
with bindings — the port design must keep "binding + auth check" one
critical section or make bindings engine-owned.
4. **Sans-I/O engine core + simulator.** Make the engine formally
`handle(event) -> [Effect]`, feed it from a `SimulatedLinkLayer`, and
move the multi-node E2E suite onto deterministic simulation (no
`waitUntil`, no timing hygiene battles). Property tests become
possible: relay-storm bounds, partition-heal convergence, dedup
soundness under duplicate floods.
## What this is not
No wire changes: packet formats, signing (padding is signed), the
peerID identity binding, and courier tag construction are untouched —
see the wire-landmines notes before assuming any of that is local.