docs(changelog): 2026.3.2

CHANGELOG.md

@@ -2,6 +2,14 @@
 
 ## Unreleased
 
+## 2026.3.2 - 2026-10-01
+
+OpenClawKit 2026.3.2 is a patch release. Decoding and importing channel config
+now uses a fraction of the stack it did; before, it could overflow a 512 KB
+cooperative thread in debug builds. The test suites also stop failing on loaded
+CI runners. There are no breaking changes, and the upstream parity target stays
+OpenClaw `2026.9.6`.
+
 ### Added
 
 - `ConfigBoxed`, the non-optional counterpart of `ConfigIndirect`: a property
@@ -20,6 +28,34 @@
   a full `ChannelsConfig` decode from 110 KB to 38 KB. The public properties
   are unchanged.
 
+### Tests
+
+- Test waits no longer give up at a wall-clock deadline (#22, #25, #26, #29,
+  #30). The macOS CI job runs about 3,150 tests in parallel and can stall the
+  cooperative pool for 5–8 s, which failed waits whose condition was about to
+  hold.
+  - Waits now await an event, or poll until the test is cancelled, and every
+    suite or test that waits has `.timeLimit(.minutes(1))`.
+  - A wait that hangs records which wait it was (`AsyncWaitTimeoutError`).
+  - Product waits that ignore cancellation, such as `agent.wait`,
+    `EmbeddedAgentRuntime.wait(runID:)` and the approval and question brokers,
+    go through `awaitCancellable`, so the time limit can end them.
+  - Elapsed-time assertions became time limits or ordering checks.
+  - Product timeouts on passing paths are now past the time limit, so a
+    regression fails as a hang instead of passing when the timeout fires.
+- Each speech-synthesizer test uses its own `TalkSystemSpeechSynthesizer`, so
+  parallel suites no longer share its state (#23).
+- Chat view-model tests stamp canned replies after the send instead of at test
+  start plus a fixed offset. A stalled send no longer makes the reply look
+  older than the message it answers (#27).
+- `ConfigDecodeStackDepthTests` measures the peak stack of the channel-config
+  decode and the document import on a thread whose stack the test maps itself,
+  and bounds both (#28).
+- 3,755 macOS tests pass under `swift test` (Swift Testing: 3,150 in
+  `OpenClawKitTests`, 585 in `OpenClawLinuxRuntimeTests`, 20 E2E), and the
+  `OpenClawLinuxRuntimeTests` bundle no longer crashes with SIGBUS.
+- Release gates: all 13 required CI checks passed on the release tree.
+
 ## 2026.3.1 - 2026-09-30
 
 OpenClawKit 2026.3.1 adds [Sign in with ChatGPT](https://developers.openai.com/siwc)

Tests/OpenClawKitTests/CoreAIModelRuntimeTests.swift

@@ -296,17 +296,15 @@ struct CoreAIModelRuntimeTests {
         #expect(texts == ["aaaaa", "bbbbb"])
     }
 
-    @Test
+    @Test(.timeLimit(.minutes(1)))
     func cancelStopsOnlyTheRunningGeneration() async throws {
         let executor = SharedCacheExecutor(limit: 40)
         let engine = CoreAILocalModelEngine(tokenizer: ScalarTokenizer(), executorFactory: { _, _ in executor })
         var configuration = LocalModelConfig(enabled: true, runtime: CoreAILocalModelEngine.runtimeID, modelPath: "/m")
         configuration.temperature = 0
         try await engine.loadModel(path: "/m", configuration: configuration)
         let running = Task { try await engine.generate(prompt: "a", systemPrompt: nil, configuration: configuration, onToken: nil) }
-        for _ in 0..<500 where await executor.runs == 0 {
-            try await Task.sleep(nanoseconds: 1_000_000)
-        }
+        try await waitUntil("first generation running") { await executor.runs > 0 }
         let queued = Task { try await engine.generate(prompt: "b", systemPrompt: nil, configuration: configuration, onToken: nil) }
         await engine.cancelGeneration(token: nil)
         await #expect(throws: CoreAIRuntimeError.cancelled) {

Tests/OpenClawKitTests/EmbeddedAgentStackTests.swift

@@ -2,7 +2,7 @@ import Foundation
 import Testing
 @testable import OpenClawKit
 
-@Suite("Embedded agent stack")
+@Suite("Embedded agent stack", .timeLimit(.minutes(1)))
 struct EmbeddedAgentStackTests {
     struct ToolCallingProvider: ModelProvider {
         let id = "stack"
@@ -32,7 +32,7 @@ struct EmbeddedAgentStackTests {
             RequestFrame(type: "req", id: "1", method: "sessions.send", params: AnyCodable(["key": AnyCodable("agent:main:main"), "message": AnyCodable("hi")]))
         )
         let runID = try #require(sent.payload?.dictionaryValue?["runId"]?.stringValue)
-        #expect(await stack.runtime.wait(runID: runID, timeoutMs: 5_000)?.output == "stack says hi")
+        #expect(try await awaitCancellable("sent run finished") { await stack.runtime.wait(runID: runID) }?.output == "stack says hi")
 
         let history = await stack.server.handle(
             RequestFrame(type: "req", id: "2", method: "chat.history", params: AnyCodable(["sessionKey": AnyCodable("agent:main:main")]))

Tests/OpenClawKitTests/GatewayServerChatUICompatTests.swift

@@ -5,7 +5,7 @@ import Testing
 
 /// The in-process server's session rows, `sessions.changed`, `session.message` and protocol-v4
 /// `chat` payloads decode through the ChatUI transport models.
-@Suite("Gateway server ChatUI compatibility")
+@Suite("Gateway server ChatUI compatibility", .timeLimit(.minutes(1)))
 struct GatewayServerChatUICompatTests {
     private func makeStack(_ name: String, turns: [String]) async throws -> (GatewayServer, EmbeddedAgentRuntime, URL) {
         let root = FileManager.default.temporaryDirectory.appendingPathComponent("\(name)-\(UUID().uuidString)")
@@ -43,11 +43,8 @@ struct GatewayServerChatUICompatTests {
         private(set) var frames: [EventFrame] = []
         func record(_ frame: EventFrame) { self.frames.append(frame) }
 
-        func wait(_ predicate: @Sendable ([EventFrame]) -> Bool) async -> [EventFrame] {
-            for _ in 0..<500 {
-                if predicate(self.frames) { return self.frames }
-                try? await Task.sleep(nanoseconds: 10_000_000)
-            }
+        func wait(_ label: String, _ predicate: @escaping @Sendable ([EventFrame]) -> Bool) async throws -> [EventFrame] {
+            try await waitUntil(label) { predicate(await self.frames) }
             return self.frames
         }
     }
@@ -101,7 +98,7 @@ struct GatewayServerChatUICompatTests {
         let send = try GatewayPayloadCodec.decode(OpenClawChatSendResponse.self, from: ack.payload)
         #expect(send.runId == "ui-run-1")
 
-        let frames = await recorder.wait { frames in
+        let frames = try await recorder.wait("final chat, both session messages and the end change") { frames in
             frames.contains { $0.event == "chat" && $0.payload?.dictionaryValue?["state"] == AnyCodable("final") }
                 && frames.filter { $0.event == "session.message" }.count >= 2
                 && frames.contains { $0.event == "sessions.changed" && $0.payload?.dictionaryValue?["phase"] == AnyCodable("end") }

Tests/OpenClawKitTests/GatewayServerRegistryTests.swift

@@ -404,7 +404,7 @@ struct GatewayServerRegistryTests {
         #expect(await reloaded.listSecretKeys() == ["TOKEN", "legacy"])
     }
 
-    @Test
+    @Test(.timeLimit(.minutes(1)))
     func builtinHandlersAcceptUpstreamWireShapes() async throws {
         let root = try Self.makeTempDirectory(named: "upstream-wire")
         defer { try? FileManager.default.removeItem(at: root) }
@@ -461,10 +461,14 @@ struct GatewayServerRegistryTests {
         #expect(missing.error?.errorCode == .invalidRequest)
 
         // agent.wait accepts the upstream `runId` key and the legacy `runID` key.
-        let upstreamWait = await server.handle(Self.frame("agent.wait", params: ["runId": AnyCodable("run-main"), "timeoutMs": AnyCodable(1_000)]))
+        let upstreamWait = try await awaitCancellable("upstream-keyed run finished") {
+            await server.handle(Self.frame("agent.wait", params: ["runId": AnyCodable("run-main")]))
+        }
         #expect(upstreamWait.payload?.dictionaryValue?["status"] == AnyCodable("ok"))
         #expect(upstreamWait.payload?.dictionaryValue?["runId"] == AnyCodable("run-main"))
-        let legacyWait = await server.handle(Self.frame("agent.wait", params: ["runID": AnyCodable("run-legacy")]))
+        let legacyWait = try await awaitCancellable("legacy-keyed run finished") {
+            await server.handle(Self.frame("agent.wait", params: ["runID": AnyCodable("run-legacy")]))
+        }
         #expect(try GatewayPayloadCodec.decode(GatewayAgentWaitResult.self, from: legacyWait.payload).output == "hi")
 
         // sessions.patch accepts upstream `agentId`/`model`/`fastMode` and null clears.

Tests/OpenClawKitTests/GatewayServerTests.swift

@@ -149,7 +149,7 @@ struct GatewayServerTests {
         #expect(deletedSession.deleted == true)
     }
 
-    @Test
+    @Test(.timeLimit(.minutes(1)))
     func gatewayServerSupportsAgentRunWaitTimeoutAndCleanup() async throws {
         let root = try self.makeTempDirectory(named: "gateway-server-agent")
         defer { try? FileManager.default.removeItem(at: root) }
@@ -191,12 +191,14 @@ struct GatewayServerTests {
         )
         #expect(!accepted.runID.isEmpty)
 
-        let waited = try await self.request(
-            server,
-            method: "agent.wait",
-            params: GatewayAgentWaitParams(runID: accepted.runID, timeoutMs: 5_000),
-            as: GatewayAgentWaitResult.self
-        )
+        let waited = try await awaitCancellable("first run finished") {
+            try await self.request(
+                server,
+                method: "agent.wait",
+                params: GatewayAgentWaitParams(runID: accepted.runID),
+                as: GatewayAgentWaitResult.self
+            )
+        }
         #expect(waited.status == "ok")
         #expect(waited.output == "first")
 
@@ -227,12 +229,14 @@ struct GatewayServerTests {
         )
         #expect(timedOut.status == "timeout")
 
-        let eventuallyCompleted = try await self.request(
-            server,
-            method: "agent.wait",
-            params: GatewayAgentWaitParams(runID: slow.runID, timeoutMs: 1_000),
-            as: GatewayAgentWaitResult.self
-        )
+        let eventuallyCompleted = try await awaitCancellable("slow run finished") {
+            try await self.request(
+                server,
+                method: "agent.wait",
+                params: GatewayAgentWaitParams(runID: slow.runID),
+                as: GatewayAgentWaitResult.self
+            )
+        }
         #expect(eventuallyCompleted.status == "ok")
         #expect(eventuallyCompleted.output == "slow")
     }
@@ -741,7 +745,7 @@ struct GatewayServerTests {
         }
     }
 
-    @Test
+    @Test(.timeLimit(.minutes(1)))
     func sdkGatewayServerSupportsCatalogSkillsAndRuntimeHandlers() async throws {
         let root = try self.makeTempDirectory(named: "gateway-server-sdk")
         defer { try? FileManager.default.removeItem(at: root) }
@@ -812,10 +816,16 @@ struct GatewayServerTests {
                 modelProviderID: "openai"
             )
         )
-        let completed: GatewayAgentWaitResult = try await client.request(
-            "agent.wait",
-            params: GatewayAgentWaitParams(runID: accepted.runID, timeoutMs: 5_000)
-        )
+        // The client's own request timeout is pushed past the time limit too, so neither side can
+        // expire while the pool is stalled.
+        let completed = try await awaitCancellable("runtime run finished") {
+            try await client.request(
+                "agent.wait",
+                params: GatewayAgentWaitParams(runID: accepted.runID),
+                timeoutMs: 3_600_000,
+                as: GatewayAgentWaitResult.self
+            )
+        }
         #expect(completed.status == "ok")
         #expect(completed.output == "runtime-output")
     }

Tests/OpenClawKitTests/MediaPipelineTests.swift

@@ -7,7 +7,7 @@ import Glibc
 import Darwin
 #endif
 
-@Suite("Media pipeline", .serialized)
+@Suite("Media pipeline", .serialized, .timeLimit(.minutes(1)))
 struct MediaPipelineTests {
     @Test
     func normalizeRejectsEmptyMimeTypeAndOversizedPayloads() async throws {
@@ -415,22 +415,26 @@ struct MediaPipelineTests {
         }
 
         let baseURL = URL(string: "http://127.0.0.1:\(port)/")!
-        try await self.waitForHTTPServer(baseURL: baseURL)
+        try await self.waitForHTTPServer(baseURL: baseURL, process: process)
         try await body(baseURL)
     }
 
-    private func waitForHTTPServer(baseURL: URL) async throws {
-        var lastError: (any Error)?
-        for _ in 0..<20 {
-            do {
-                _ = try Data(contentsOf: baseURL)
+    /// Polls until the server answers. There is no deadline: a loaded runner only delays Python's
+    /// startup, and the suite's time limit bounds a server that never answers.
+    private func waitForHTTPServer(baseURL: URL, process: Process) async throws {
+        while !Task.isCancelled {
+            if (try? Data(contentsOf: baseURL)) != nil {
                 return
-            } catch {
-                lastError = error
-                try await Task.sleep(nanoseconds: 100_000_000)
             }
+            guard process.isRunning else {
+                throw OpenClawCoreError.unavailable("Local HTTP server exited with status \(process.terminationStatus)")
+            }
+            try? await Task.sleep(nanoseconds: 100_000_000)
         }
-        throw lastError ?? OpenClawCoreError.unavailable("Local HTTP server did not start")
+        // Swift Testing drops errors thrown after a time-limit cancellation, so record which wait hung.
+        let timeout = AsyncWaitTimeoutError(label: "local HTTP server answering")
+        Issue.record(timeout)
+        throw timeout
     }
 
     private func reserveTCPPort() throws -> Int {

Tests/OpenClawKitTests/ModelRoutingTests.swift

@@ -332,9 +332,11 @@ struct ModelRoutingTests {
                 StaticProvider(id: "primary", text: "primary-output"),
                 StaticProvider(id: "secondary", text: "secondary-output"),
             ],
+            // An hour-long window: the second request is inside it however slow the runner is, and a
+            // drop never waits for the window to pass.
             throttlePolicy: ModelProviderThrottlePolicy(
                 maxRequestsPerWindow: 1,
-                windowMs: 1_000,
+                windowMs: 3_600_000,
                 strategy: .drop
             ),
             diagnosticsSink: await pipeline.sink()
@@ -355,32 +357,49 @@ struct ModelRoutingTests {
         #expect(events.contains(where: { $0.name == "model.request.retry" }))
     }
 
-    @Test
+    @Test(.timeLimit(.minutes(1)))
     func routerThrottleDelayAppliesCooldownAndEmitsDiagnostics() async throws {
-        let pipeline = RuntimeDiagnosticsPipeline(eventLimit: 50)
+        let request = ModelGenerationRequest(sessionKey: "main", prompt: "hello", providerID: "primary")
         let router = ModelRouter(
             defaultProviderID: "primary",
             providers: [StaticProvider(id: "primary", text: "primary-output")],
             throttlePolicy: ModelProviderThrottlePolicy(
                 maxRequestsPerWindow: 1,
                 windowMs: 70,
                 strategy: .delay
-            ),
-            diagnosticsSink: await pipeline.sink()
+            )
         )
 
+        // The second request proceeds once the window has passed. A lower bound: a slow runner only
+        // adds to it (and may let the window pass before the second request arrives).
         let startedAt = Date()
-        _ = try await router.generate(
-            ModelGenerationRequest(sessionKey: "main", prompt: "hello", providerID: "primary")
-        )
-        _ = try await router.generate(
-            ModelGenerationRequest(sessionKey: "main", prompt: "hello", providerID: "primary")
-        )
-        let elapsed = Date().timeIntervalSince(startedAt)
+        _ = try await router.generate(request)
+        _ = try await router.generate(request)
+        #expect(Date().timeIntervalSince(startedAt) >= 0.06)
 
-        #expect(elapsed >= 0.06)
-        let events = await pipeline.recentEvents(limit: 50)
-        #expect(events.contains(where: { $0.name == "model.throttle.delay" }))
+        // With an hour-long window the second request always lands inside it and waits.
+        let pipeline = RuntimeDiagnosticsPipeline(eventLimit: 50)
+        let slowRouter = ModelRouter(
+            defaultProviderID: "primary",
+            providers: [StaticProvider(id: "primary", text: "primary-output")],
+            throttlePolicy: ModelProviderThrottlePolicy(
+                maxRequestsPerWindow: 1,
+                windowMs: 3_600_000,
+                strategy: .delay
+            ),
+            diagnosticsSink: await pipeline.sink()
+        )
+        _ = try await slowRouter.generate(request)
+        let delayed = Task { try await slowRouter.generate(request) }
+        try await waitUntil("model.throttle.delay emitted") {
+            await pipeline.recentEvents(limit: 50).contains { $0.name == "model.throttle.delay" }
+        }
+        let delay = try #require(await pipeline.recentEvents(limit: 50).first { $0.name == "model.throttle.delay" })
+        #expect((Int(delay.metadata["delayMs"] ?? "") ?? 0) > 3_000_000)
+        delayed.cancel()
+        await #expect(throws: CancellationError.self) {
+            _ = try await delayed.value
+        }
     }
 
     @Test

Tests/OpenClawKitTests/SystemStateReportingTests.swift

@@ -191,16 +191,17 @@ struct SystemStateReportingTests {
         #expect(recorder.reports.last == .volatile(.gateway, ["backoffMs": 4000]))
     }
 
-    @Test
+    @Test(.timeLimit(.minutes(1)))
     func coalescingReporterSchedulesTrailingFlush() async throws {
         let recorder = RecordingStateReporter()
-        let reporter = CoalescingSystemStateReporter(wrapping: recorder, minimumInterval: 0.05)
+        // The clock stands still, so the second update always lands inside the interval; only the
+        // trailing flush can forward it.
+        let clock = TestClock()
+        let reporter = CoalescingSystemStateReporter(wrapping: recorder, minimumInterval: 0.05, now: { clock.now })
         reporter.reportVolatileUpdate(.talk, ["level": 1])
         reporter.reportVolatileUpdate(.talk, ["level": 2])
         #expect(recorder.reports == [.volatile(.talk, ["level": 1])])
-        for _ in 0..<50 where recorder.reports.count < 2 {
-            try await Task.sleep(nanoseconds: 20_000_000)
-        }
+        try await waitUntil("trailing flush forwarded the held update") { recorder.reports.count >= 2 }
         #expect(recorder.reports == [.volatile(.talk, ["level": 1]), .volatile(.talk, ["level": 2])])
     }