Implement RTN23 idle-transport detection and the recovery spec points around it

src/IO.Ably.Shared.MsgPack/CustomSerialisers/GeneratedSerializers/IO_Ably_ConnectionDetailsMessageSerializer.cs

@@ -35,7 +35,7 @@ public IO_Ably_ConnectionDetailsMessageSerializer(MsgPack.Serialization.Serializ
         }
 
         protected override void PackToCore(MsgPack.Packer packer, IO.Ably.ConnectionDetails objectTree) {
-            packer.PackMapHeader(7);
+            packer.PackMapHeader(8);
             this._serializer0.PackTo(packer, "clientId");
             this._serializer0.PackTo(packer, objectTree.ClientId);
             this._serializer0.PackTo(packer, "connectionKey");
@@ -44,6 +44,8 @@ protected override void PackToCore(MsgPack.Packer packer, IO.Ably.ConnectionDeta
             this._serializer1.PackTo(packer, objectTree.ConnectionStateTtl);
             this._serializer0.PackTo(packer, "maxFrameSize");
             this._serializer2.PackTo(packer, objectTree.MaxFrameSize);
+            this._serializer0.PackTo(packer, "maxIdleInterval");
+            this._serializer1.PackTo(packer, objectTree.MaxIdleInterval);
             this._serializer0.PackTo(packer, "maxInboundRate");
             this._serializer2.PackTo(packer, objectTree.MaxInboundRate);
             this._serializer0.PackTo(packer, "maxMessageSize");
@@ -112,6 +114,33 @@ protected override IO.Ably.ConnectionDetails UnpackFromCore(MsgPack.Unpacker unp
                     result.MaxFrameSize = nullable2.Value;
                 }
                 unpacked = (unpacked + 1);
+                System.Nullable<System.TimeSpan> nullableMaxIdleInterval = default(System.Nullable<System.TimeSpan>);
+                if ((unpacked < itemsCount)) {
+                    if ((unpacker.Read() == false)) {
+                        throw MsgPack.Serialization.SerializationExceptions.NewMissingItem(4);
+                    }
+                    if (((unpacker.IsArrayHeader == false)
+                                && (unpacker.IsMapHeader == false))) {
+                        nullableMaxIdleInterval = this._serializer1.UnpackFrom(unpacker);
+                    }
+                    else {
+                        MsgPack.Unpacker disposableMaxIdleInterval = default(MsgPack.Unpacker);
+                        disposableMaxIdleInterval = unpacker.ReadSubtree();
+                        try {
+                            nullableMaxIdleInterval = this._serializer1.UnpackFrom(disposableMaxIdleInterval);
+                        }
+                        finally {
+                            if (((disposableMaxIdleInterval == null)
+                                        == false)) {
+                                disposableMaxIdleInterval.Dispose();
+                            }
+                        }
+                    }
+                }
+                if (nullableMaxIdleInterval.HasValue) {
+                    result.MaxIdleInterval = nullableMaxIdleInterval;
+                }
+                unpacked = (unpacked + 1);
                 System.Nullable<long> nullable3 = default(System.Nullable<long>);
                 if ((unpacked < itemsCount)) {
                     nullable3 = MsgPack.Serialization.UnpackHelpers.UnpackNullableInt64Value(unpacker, typeof(IO.Ably.ConnectionDetails), "Int64 maxInboundRate");
@@ -230,7 +259,35 @@ protected override IO.Ably.ConnectionDetails UnpackFromCore(MsgPack.Unpacker unp
                                                 }
                                             }
                                             else {
-                                                unpacker.Skip();
+                                                if ((key == "maxIdleInterval")) {
+                                                    System.Nullable<System.TimeSpan> nullableMaxIdle = default(System.Nullable<System.TimeSpan>);
+                                                    if ((unpacker.Read() == false)) {
+                                                        throw MsgPack.Serialization.SerializationExceptions.NewMissingItem(i);
+                                                    }
+                                                    if (((unpacker.IsArrayHeader == false)
+                                                                && (unpacker.IsMapHeader == false))) {
+                                                        nullableMaxIdle = this._serializer1.UnpackFrom(unpacker);
+                                                    }
+                                                    else {
+                                                        MsgPack.Unpacker disposableMaxIdle = default(MsgPack.Unpacker);
+                                                        disposableMaxIdle = unpacker.ReadSubtree();
+                                                        try {
+                                                            nullableMaxIdle = this._serializer1.UnpackFrom(disposableMaxIdle);
+                                                        }
+                                                        finally {
+                                                            if (((disposableMaxIdle == null)
+                                                                        == false)) {
+                                                                disposableMaxIdle.Dispose();
+                                                            }
+                                                        }
+                                                    }
+                                                    if (nullableMaxIdle.HasValue) {
+                                                        result.MaxIdleInterval = nullableMaxIdle;
+                                                    }
+                                                }
+                                                else {
+                                                    unpacker.Skip();
+                                                }
                                             }
                                         }
                                     }

src/IO.Ably.Shared/AblyAuth.cs

@@ -305,11 +305,27 @@ public virtual async Task<TokenDetails> RequestTokenAsync(TokenParams tokenParam
                 bool shouldCatch = true;
                 try
                 {
-                    var callbackResult = await authOptions.AuthCallback(tokenParams);
+                    // RSA4c bounds an auth attempt by realtimeRequestTimeout. This one is awaited on
+                    // the realtime workflow's single reader thread, so an authCallback that never
+                    // returns would stall channel state, inbound messages and the idle monitor.
+                    //
+                    // Task.Run rather than awaiting the delegate directly: TimeoutAfter extends an
+                    // already-created Task, so a callback whose body runs synchronously - the most
+                    // ordinary shape in C# - would block before there is anything to bound. It
+                    // cannot cancel the callback; an abandoned one runs to completion.
+                    var callbackResult = await Task.Run(() => authOptions.AuthCallback(tokenParams))
+                        .TimeoutAfter(Options.RealtimeRequestTimeout, null);
 
                     if (callbackResult == null)
                     {
-                        throw new AblyException("AuthCallback returned null", ErrorCodes.ClientAuthProviderRequestFailed);
+                        // Covers a callback that returned null and one that timed out - TimeoutAfter
+                        // yields null for both, and the wrapper below replaces the message anyway.
+                        // ClientCallbackError, not ClientAuthProviderRequestFailed: the wrapper
+                        // rethrows as 80019 per RSA4c1 with this as the cause, so 80019 here would
+                        // duplicate its own parent. ably-js uses 40170 for the same case.
+                        throw new AblyException(
+                            $"AuthCallback returned null or did not return within {Options.RealtimeRequestTimeout.TotalSeconds}s",
+                            ErrorCodes.ClientCallbackError);
                     }
 
                     if (callbackResult is TokenDetails)
@@ -341,12 +357,17 @@ public virtual async Task<TokenDetails> RequestTokenAsync(TokenParams tokenParam
                             : HttpStatusCode.Unauthorized;
                     }
 
+                    // RSA4c1 wants the cause "set to the underlying cause". The four argument overload
+                    // assigns InnerException instead, which is not the spec's field and is not
+                    // serialised as one.
                     throw new AblyException(
                         new ErrorInfo(
                             "Error calling AuthCallback, token request failed. See inner exception for details.",
                             ErrorCodes.ClientAuthProviderRequestFailed,
                             statusCode,
-                            ex),
+                            href: null,
+                            cause: (ex as AblyException)?.ErrorInfo ?? new ErrorInfo(ex.Message),
+                            innerException: ex),
                         ex);
                 }
             }

src/IO.Ably.Shared/AuthOptions.cs

@@ -13,6 +13,13 @@ public class AuthOptions
         /// <summary>
         /// The callback used to get a new <see cref="IO.Ably.TokenDetails"/> or <see cref="IO.Ably.TokenRequest"/>.
         /// AuthCallback is used by internally by <see cref="IO.Ably.AblyAuth"/>.RequestTokenAsync.
+        /// <para>
+        /// The callback is bounded by <see cref="ClientOptions.RealtimeRequestTimeout"/> per RSA4c,
+        /// but cannot be cancelled: one that overruns is abandoned and keeps running, so a later auth
+        /// attempt may invoke it again concurrently and implementations must tolerate that. A result
+        /// returned after the bound is discarded, and each abandoned invocation holds a thread pool
+        /// worker until it returns.
+        /// </para>
         /// </summary>
         public Func<TokenParams, Task<object>> AuthCallback { get; set; }
 

src/IO.Ably.Shared/ClientOptions.cs

@@ -295,6 +295,51 @@ public bool UseBinaryProtocol
         /// </summary>
         public TimeSpan ChannelRetryTimeout { get; set; } = Defaults.ChannelRetryTimeout;
 
+        private TimeSpan _realtimeRequestTimeout = Defaults.RealtimeRequestTimeout;
+
+        /// <summary>
+        /// How long the library waits for Ably to answer before treating a realtime request as
+        /// having failed. Applies while establishing a connection, while awaiting a response to a
+        /// Heartbeat, Connect, Attach, Detach or Close, and as part of the RTN23a idle timeout.
+        /// Default: 10s. Must be a positive interval, and values beyond a minute or so are rarely
+        /// useful. Disabling the timeout is not supported - the timeouts this value drives are all
+        /// required to fire - so both Timeout.InfiniteTimeSpan and TimeSpan.MaxValue are rejected.
+        /// TO3l11 - https://sdk.ably.com/builds/ably/specification/main/features/#TO3l11.
+        /// </summary>
+        /// <exception cref="ArgumentOutOfRangeException">when set to zero, a negative value, or an
+        /// interval too large for the underlying timers to fire.</exception>
+        public TimeSpan RealtimeRequestTimeout
+        {
+            get => _realtimeRequestTimeout;
+
+            set
+            {
+                // A non-positive value fails quietly rather than loudly: zero turns the RTN14c
+                // connect timeout into a hot reconnect loop, and a negative one stops the timer
+                // firing at all. Timeout.InfiniteTimeSpan is -1ms, so the lower bound catches it -
+                // which matters, because both Task.Delay and System.Threading.Timer accept it as
+                // genuine infinity and it would disable RTN14c, RTN12b, RTL4f, RTL5f and RSA4c.
+                //
+                // The upper bound is the tightest limit across the sinks this value reaches on every
+                // framework shipped: Task.Delay allows uint.MaxValue - 1 ms on .NET 6+ but only
+                // Int32.MaxValue on .NET Framework, Mono and Xamarin, and CountdownTimer casts to
+                // int for System.Threading.Timer. So Int32.MaxValue ms - a bound on the arithmetic,
+                // not a supported configuration, which is why the message names the useful range.
+                if (value <= TimeSpan.Zero || value.TotalMilliseconds > int.MaxValue)
+                {
+                    throw new ArgumentOutOfRangeException(
+                        nameof(RealtimeRequestTimeout),
+                        value,
+                        "RealtimeRequestTimeout must be a positive interval. The default is 10s; " +
+                        "values beyond a minute or so are rarely useful. Disabling the timeout is " +
+                        "not supported - the connect, close, attach, detach and auth timeouts it " +
+                        "drives are all required to fire.");
+                }
+
+                _realtimeRequestTimeout = value;
+            }
+        }
+
         /// <summary>
         /// Timeout for opening an http request.
         /// Default: 4s.
@@ -436,8 +481,6 @@ internal Func<DateTimeOffset> NowFunc
 
         internal bool SkipInternetCheck { get; set; }
 
-        internal TimeSpan RealtimeRequestTimeout { get; set; } = Defaults.RealtimeRequestTimeout;
-
         /// <summary>
         /// Default constructor for ClientOptions.
         /// </summary>

src/IO.Ably.Shared/Realtime/Connection.cs

@@ -181,10 +181,13 @@ private void HandleNetworkStateChange(NetworkState state)
         /// <returns>recoveryKey.</returns>
         public string CreateRecoveryKey()
         {
+            // RTN16g3, which replaces RTN16g2 as of specification 6.1.0 - null in CLOSED, CLOSING
+            // and FAILED, and SUSPENDED is deliberately not among them. RTN8d and RTN9d keep the
+            // key through SUSPENDED because RTN14h always attempts a resume, so the connection is
+            // still recoverable there and the key has to be available to hand over.
             if (Key.IsEmpty() || InnerState.State == Realtime.ConnectionState.Closing
                               || InnerState.State == Realtime.ConnectionState.Closed
-                              || InnerState.State == Realtime.ConnectionState.Failed
-                              || InnerState.State == Realtime.ConnectionState.Suspended)
+                              || InnerState.State == Realtime.ConnectionState.Failed)
             {
                 return string.Empty;
             }

src/IO.Ably.Shared/Realtime/Presence.cs

@@ -792,11 +792,22 @@ private void SendQueuedMessages()
 
         private void FailQueuedMessages(ErrorInfo reason)
         {
+            // RTL11 wants "an ErrorInfo indicating the failure". TaskWrapper has no branch for
+            // (false, null) and faults the task with a bare Exception, so the reason is coalesced
+            // here rather than at each call site.
+            var error = reason ?? ErrorInfo.ReasonUnknown;
+
             while (!PendingPresenceQueue.IsEmpty)
             {
                 if (PendingPresenceQueue.TryDequeue(out var queuedPresenceMessage))
                 {
-                    queuedPresenceMessage.Callback?.Invoke(false, reason);
+                    // Guarded like every other callback site: a throwing application callback would
+                    // otherwise abort the rest of the queue and skip the RTP5a map clearing in
+                    // ChannelDetachedOrFailed, which RealtimeChannels then swallows per channel.
+                    ActionUtils.SafeExecute(
+                        () => queuedPresenceMessage.Callback?.Invoke(false, error),
+                        Logger,
+                        nameof(FailQueuedMessages));
                 }
             }
         }