| 1 | // Run: tsx src/__tests__/activate-topic-stale.test.tsx |
| 2 | // |
| 3 | // Locks in last-click-wins for single-surface topic activation (#6607): when |
| 4 | // a newer navigation starts while app.StartTopicActivation is still in |
| 5 | // flight, the stale completion must neither flip the visible tab away from |
| 6 | // the user's last click nor delete the newer surface's cached state (the |
| 7 | // single-surface prune removes every other tab state, blanking the visible |
| 8 | // transcript). |
| 9 | |
| 10 | import { JSDOM } from "jsdom"; |
| 11 | import React, { act } from "react"; |
| 12 | import { createRoot } from "react-dom/client"; |
| 13 | import type { AppBindings } from "../lib/bridge"; |
| 14 | import { enqueueNavigationRequest, type NavigationCoalescingRefs } from "../lib/openTopicCoalescing"; |
| 15 | import { useController } from "../lib/useController"; |
| 16 | import { historySliceFromMessages } from "./mockHistorySlice"; |
| 17 | import type { BalanceInfo, CheckpointMeta, ContextInfo, EffortInfo, HistoryMessage, HistorySliceRequest, JobView, Meta, TabMeta, TopicActivationEvent, TopicActivationRequest, WireEvent } from "../lib/types"; |
| 18 | import { installDesktopHostStub } from "./desktopHostStub"; |
| 19 | |
| 20 | let passed = 0; |
| 21 | let failed = 0; |
| 22 | |
| 23 | function ok(value: boolean, label: string) { |
| 24 | if (value) { |
| 25 | process.stdout.write(` PASS ${label}\n`); |
| 26 | passed += 1; |
| 27 | } else { |
| 28 | process.stdout.write(` FAIL ${label}\n`); |
| 29 | failed += 1; |
| 30 | } |
| 31 | } |
| 32 | |
| 33 | function eq(actual: unknown, expected: unknown, label: string) { |
| 34 | ok(actual === expected, `${label}${actual === expected ? "" : `: expected ${JSON.stringify(expected)}, got ${JSON.stringify(actual)}`}`); |
| 35 | } |
| 36 | |
| 37 | function flushPromises(): Promise<void> { |
| 38 | return new Promise((resolve) => setTimeout(resolve, 0)); |
| 39 | } |
| 40 | |
| 41 | function deferred<T>() { |
| 42 | let resolve!: (value: T) => void; |
| 43 | let reject!: (reason?: unknown) => void; |
| 44 | const promise = new Promise<T>((res, rej) => { |
| 45 | resolve = res; |
| 46 | reject = rej; |
| 47 | }); |
| 48 | return { promise, resolve, reject }; |
| 49 | } |
| 50 | |
| 51 | async function waitFor(label: string, predicate: () => boolean) { |
| 52 | for (let attempt = 0; attempt < 30; attempt += 1) { |
| 53 | await act(async () => { |
| 54 | await flushPromises(); |
| 55 | }); |
| 56 | if (predicate()) return; |
| 57 | } |
| 58 | throw new Error(`timed out waiting for ${label}`); |
| 59 | } |
| 60 | |
| 61 | function tabMeta(id: string, overrides: Partial<TabMeta> = {}): TabMeta { |
| 62 | const workspaceRoot = `/repo/${id}`; |
| 63 | return { |
| 64 | id, |
| 65 | scope: "project", |
| 66 | workspaceRoot, |
| 67 | workspaceName: id, |
| 68 | workspacePath: workspaceRoot, |
| 69 | gitBranch: "main", |
| 70 | topicId: `topic-${id}`, |
| 71 | topicTitle: id, |
| 72 | sessionPath: `${workspaceRoot}/sessions/${id}.jsonl`, |
| 73 | label: `model-${id}`, |
| 74 | ready: true, |
| 75 | running: false, |
| 76 | mode: "normal", |
| 77 | toolApprovalMode: "ask", |
| 78 | tokenMode: "full", |
| 79 | active: false, |
| 80 | cwd: workspaceRoot, |
| 81 | ...overrides, |
| 82 | }; |
| 83 | } |
| 84 | |
| 85 | function metaFor(tab: TabMeta): Meta { |
| 86 | return { |
| 87 | label: tab.label, |
| 88 | ready: tab.ready, |
| 89 | startupErr: tab.startupErr, |
| 90 | eventChannel: "agent:event", |
| 91 | cwd: tab.cwd || tab.workspaceRoot, |
| 92 | workspaceRoot: tab.workspaceRoot, |
| 93 | workspaceName: tab.workspaceName, |
| 94 | workspacePath: tab.workspacePath, |
| 95 | gitBranch: tab.gitBranch, |
| 96 | autoApproveTools: false, |
| 97 | bypass: false, |
| 98 | collaborationMode: tab.collaborationMode ?? "normal", |
| 99 | toolApprovalMode: tab.toolApprovalMode ?? "ask", |
| 100 | tokenMode: tab.tokenMode ?? "full", |
| 101 | goal: "", |
| 102 | goalStatus: "stopped", |
| 103 | }; |
| 104 | } |
| 105 | |
| 106 | function userMessage(content: string): HistoryMessage { |
| 107 | return { role: "user", content }; |
| 108 | } |
| 109 | |
| 110 | console.log("\nactivate topic stale completion"); |
| 111 | |
| 112 | const dom = new JSDOM("<!doctype html><html><body><div id=\"root\"></div></body></html>", { |
| 113 | pretendToBeVisual: true, |
| 114 | url: "http://localhost/", |
| 115 | }); |
| 116 | (globalThis as typeof globalThis & { IS_REACT_ACT_ENVIRONMENT: boolean }).IS_REACT_ACT_ENVIRONMENT = true; |
| 117 | globalThis.window = dom.window as unknown as Window & typeof globalThis; |
| 118 | globalThis.document = dom.window.document; |
| 119 | Object.defineProperty(globalThis, "navigator", { configurable: true, value: dom.window.navigator }); |
| 120 | globalThis.Node = dom.window.Node; |
| 121 | globalThis.HTMLElement = dom.window.HTMLElement; |
| 122 | globalThis.Event = dom.window.Event; |
| 123 | globalThis.CustomEvent = dom.window.CustomEvent; |
| 124 | globalThis.KeyboardEvent = dom.window.KeyboardEvent; |
| 125 | globalThis.MouseEvent = dom.window.MouseEvent; |
| 126 | globalThis.localStorage = dom.window.localStorage; |
| 127 | globalThis.requestAnimationFrame = dom.window.requestAnimationFrame.bind(dom.window); |
| 128 | globalThis.cancelAnimationFrame = dom.window.cancelAnimationFrame.bind(dom.window); |
| 129 | |
| 130 | const context: ContextInfo = { used: 12, window: 100, sessionTokens: 12 }; |
| 131 | const effort: EffortInfo = { supported: true, current: "auto", default: "auto", levels: ["auto"] }; |
| 132 | const balance: BalanceInfo = { available: false, display: "" }; |
| 133 | const jobs: JobView[] = []; |
| 134 | const checkpoints: CheckpointMeta[] = []; |
| 135 | const tabA = tabMeta("tab-a", { active: true }); |
| 136 | const tabX = tabMeta("tab-x"); |
| 137 | const tabY = tabMeta("tab-y"); |
| 138 | let backendActiveId = "tab-a"; |
| 139 | // Per-tab holds so any activation can be stalled mid-flight and released. |
| 140 | const activationHolds = new Map<string, Promise<void>>(); |
| 141 | const tabsById = new Map([tabA, tabX, tabY].map((tab) => [tab.id, tab])); |
| 142 | const replayTargets: string[] = []; |
| 143 | // The pending ticketed activation backend-side: a newer StartTopicActivation |
| 144 | // supersedes it (cancelled), exactly like the real generation protocol. |
| 145 | let mockPendingActivation: { requestId: string; tabId: string } | undefined; |
| 146 | |
| 147 | function emitTopicActivation(event: TopicActivationEvent): void { |
| 148 | desktopStub.emit("topic:activation", event); |
| 149 | } |
| 150 | |
| 151 | function currentTabs(): TabMeta[] { |
| 152 | return Array.from(tabsById.values()).map((tab) => ({ ...tab, active: tab.id === backendActiveId })); |
| 153 | } |
| 154 | |
| 155 | const appStubTable = ({ |
| 156 | main: { |
| 157 | App: { |
| 158 | RegisterNavigationIntent: async () => {}, |
| 159 | ListTabs: async () => currentTabs(), |
| 160 | MetaForTab: async (tabID: string) => metaFor(tabsById.get(tabID) ?? tabA), |
| 161 | ContextUsageForTab: async () => context, |
| 162 | EffortForTab: async () => effort, |
| 163 | BalanceForTab: async () => balance, |
| 164 | JobsForTab: async () => jobs, |
| 165 | CheckpointsForTab: async () => checkpoints, |
| 166 | ForkTargetsForTab: async () => ({ targets: [], verifiable: false }), |
| 167 | HistoryForTab: async (tabID: string) => { |
| 168 | if (tabID === "tab-x") return [userMessage("history X")]; |
| 169 | if (tabID === "tab-y") return [userMessage("history Y")]; |
| 170 | return [userMessage("history A")]; |
| 171 | }, |
| 172 | HistoryPageForTab: async (tabID: string) => { |
| 173 | const messages = await appStubTable.HistoryForTab(tabID); |
| 174 | return { messages, startTurn: 0, endTurn: messages.length, totalTurns: messages.length, hasOlder: false }; |
| 175 | }, |
| 176 | HistorySliceForTab: async (tabID: string, req: HistorySliceRequest) => |
| 177 | historySliceFromMessages(tabID, await appStubTable.HistoryForTab(tabID), req), |
| 178 | HistoryCheckpointTurnsForTab: async () => [], |
| 179 | ActivateTopic: async (_scope: string, workspaceRoot: string, topicId: string) => { |
| 180 | const target = Array.from(tabsById.values()).find((tab) => tab.workspaceRoot === workspaceRoot && tab.topicId === topicId) ?? tabA; |
| 181 | const hold = activationHolds.get(target.id); |
| 182 | if (hold) await hold; |
| 183 | backendActiveId = target.id; |
| 184 | return { ...target, active: true }; |
| 185 | }, |
| 186 | StartTopicActivation: async (req: TopicActivationRequest) => { |
| 187 | const target = Array.from(tabsById.values()).find((tab) => tab.workspaceRoot === req.workspaceRoot && tab.topicId === req.topicId) ?? tabA; |
| 188 | const hold = activationHolds.get(target.id); |
| 189 | if (hold) await hold; |
| 190 | backendActiveId = target.id; |
| 191 | const requestId = req.requestId || `mock-activation-${target.id}`; |
| 192 | const previous = mockPendingActivation; |
| 193 | mockPendingActivation = { requestId, tabId: target.id }; |
| 194 | if (previous && previous.requestId !== requestId) { |
| 195 | emitTopicActivation({ requestId: previous.requestId, tabId: previous.tabId, phase: "cancelled" }); |
| 196 | } |
| 197 | emitTopicActivation({ requestId, tabId: target.id, phase: "starting" }); |
| 198 | window.setTimeout(() => { |
| 199 | if (mockPendingActivation?.requestId !== requestId) return; |
| 200 | mockPendingActivation = undefined; |
| 201 | emitTopicActivation({ requestId, tabId: target.id, phase: "ready" }); |
| 202 | }, 0); |
| 203 | return { requestId, tabId: target.id, meta: { ...target, active: true } }; |
| 204 | }, |
| 205 | SetActiveTab: async (tabID: string) => { |
| 206 | const hold = activationHolds.get(tabID); |
| 207 | if (hold) await hold; |
| 208 | backendActiveId = tabID; |
| 209 | }, |
| 210 | ReplayPendingPrompts: async () => {}, |
| 211 | ReplayPendingPromptsForTab: async (tabID: string) => { |
| 212 | replayTargets.push(tabID); |
| 213 | if (!tabsById.get(tabID)?.pendingPrompt) return; |
| 214 | desktopStub.emit("agent:event", { |
| 215 | kind: "ask_request", |
| 216 | tabId: tabID, |
| 217 | ask: { id: `pending-${tabID}`, questions: [{ id: "choice", prompt: "Keep me through A-X-A", options: [] }] }, |
| 218 | }); |
| 219 | }, |
| 220 | } as Partial<AppBindings> as AppBindings, |
| 221 | }, |
| 222 | }).main.App; |
| 223 | const desktopStub = installDesktopHostStub(appStubTable); |
| 224 | |
| 225 | type Controller = ReturnType<typeof useController>; |
| 226 | let controller: Controller | undefined; |
| 227 | |
| 228 | function Probe() { |
| 229 | controller = useController(); |
| 230 | return null; |
| 231 | } |
| 232 | |
| 233 | const rootEl = document.getElementById("root"); |
| 234 | if (!rootEl) throw new Error("missing root"); |
| 235 | const root = createRoot(rootEl); |
| 236 | |
| 237 | await act(async () => { |
| 238 | root.render(<Probe />); |
| 239 | await flushPromises(); |
| 240 | }); |
| 241 | await waitFor("initial active tab", () => controller?.activeTabId === "tab-a"); |
| 242 | |
| 243 | // Click topic X: the backend call hangs (slow prune / disk). |
| 244 | const activateXGate = deferred<void>(); |
| 245 | activationHolds.set("tab-x", activateXGate.promise); |
| 246 | let activateX: Promise<TabMeta> | undefined; |
| 247 | await act(async () => { |
| 248 | activateX = controller?.activateTopic("project", tabX.workspaceRoot, tabX.topicId ?? ""); |
| 249 | await flushPromises(); |
| 250 | }); |
| 251 | eq(controller?.activeTabId, "tab-a", "held activation does not flip the tab early"); |
| 252 | |
| 253 | // The user clicks topic Y before X's backend call returns; Y resolves first. |
| 254 | await act(async () => { |
| 255 | await controller?.activateTopic("project", tabY.workspaceRoot, tabY.topicId ?? ""); |
| 256 | await flushPromises(); |
| 257 | }); |
| 258 | await waitFor("Y is active with its history", () => |
| 259 | controller?.activeTabId === "tab-y" && controller.state.items.some((item) => item.kind === "user" && item.text === "history Y")); |
| 260 | |
| 261 | // X's stale completion lands after Y applied. Last click must win. |
| 262 | await act(async () => { |
| 263 | activateXGate.resolve(); |
| 264 | activationHolds.delete("tab-x"); |
| 265 | await activateX; |
| 266 | await flushPromises(); |
| 267 | }); |
| 268 | await act(async () => { |
| 269 | await flushPromises(); |
| 270 | }); |
| 271 | eq(controller?.activeTabId, "tab-y", "stale activation must not flip the visible tab"); |
| 272 | ok(controller?.state.items.some((item) => item.kind === "user" && item.text === "history Y") === true, |
| 273 | "stale activation must not delete the visible tab's cached state"); |
| 274 | |
| 275 | // A fresh activation afterwards still applies normally (guard is not sticky). |
| 276 | await act(async () => { |
| 277 | await controller?.activateTopic("project", tabX.workspaceRoot, tabX.topicId ?? ""); |
| 278 | await flushPromises(); |
| 279 | }); |
| 280 | await waitFor("X activates cleanly on a fresh click", () => controller?.activeTabId === "tab-x"); |
| 281 | |
| 282 | // --- Through the REAL production navigation queue (#6613 review P1) --- |
| 283 | // |
| 284 | // App.enqueueNavigation serializes clicks: a click made while another request |
| 285 | // runs only becomes a pending queue entry — it does NOT run activateTopic, so |
| 286 | // the controller epoch does not advance by itself. The App wiring must bump |
| 287 | // the epoch at ENQUEUE time (noteNavigationIntent), otherwise the running |
| 288 | // stale activation passes the guard, flips the tab, and prunes cached state. |
| 289 | type NavInput = { workspaceRoot: string; topicId: string }; |
| 290 | const navRefs: NavigationCoalescingRefs<NavInput> = { |
| 291 | seqRef: { current: 0 }, |
| 292 | runningRef: { current: false }, |
| 293 | pendingRef: { current: null }, |
| 294 | }; |
| 295 | const enqueueNav = (workspaceRoot: string, topicId: string): Promise<void> => { |
| 296 | controller?.noteNavigationIntent(); // the App.tsx wiring under test |
| 297 | return enqueueNavigationRequest(navRefs, { workspaceRoot, topicId }, async (request) => { |
| 298 | await controller?.activateTopic("project", request.workspaceRoot, request.topicId); |
| 299 | }); |
| 300 | }; |
| 301 | |
| 302 | const gateY = deferred<void>(); |
| 303 | activationHolds.set("tab-y", gateY.promise); |
| 304 | const gateA = deferred<void>(); |
| 305 | activationHolds.set("tab-a", gateA.promise); |
| 306 | |
| 307 | let queuedFirst: Promise<void> | undefined; |
| 308 | let queuedSecond: Promise<void> | undefined; |
| 309 | await act(async () => { |
| 310 | queuedFirst = enqueueNav(tabY.workspaceRoot, tabY.topicId ?? ""); // runs, held mid-flight |
| 311 | await flushPromises(); |
| 312 | }); |
| 313 | await act(async () => { |
| 314 | queuedSecond = enqueueNav(tabA.workspaceRoot, tabA.topicId ?? ""); // queued, does not run yet |
| 315 | await flushPromises(); |
| 316 | }); |
| 317 | |
| 318 | // The first (now stale) activation resolves while the second is still queued. |
| 319 | await act(async () => { |
| 320 | gateY.resolve(); |
| 321 | activationHolds.delete("tab-y"); |
| 322 | await queuedFirst; |
| 323 | await flushPromises(); |
| 324 | }); |
| 325 | eq(controller?.activeTabId, "tab-x", "queued click invalidates the running activation (no flip to tab-y)"); |
| 326 | ok(controller?.state.items.some((item) => item.kind === "user" && item.text === "history X") === true, |
| 327 | "queued click keeps the visible tab's cached state intact"); |
| 328 | |
| 329 | // The queued request then runs and lands on the user's last click. |
| 330 | await act(async () => { |
| 331 | gateA.resolve(); |
| 332 | activationHolds.delete("tab-a"); |
| 333 | await queuedSecond; |
| 334 | await flushPromises(); |
| 335 | }); |
| 336 | await waitFor("queued last click applies once it runs", () => controller?.activeTabId === "tab-a"); |
| 337 | |
| 338 | // A prompt can arrive on A before a slow A→X→A switch completes. The final A |
| 339 | // activation must preserve its card, and X's stale completion must reassert A |
| 340 | // without clearing or replaying a sibling tab's prompt. |
| 341 | const promptBlockedA = { ...tabA, running: true, pendingPrompt: true, cancellable: true }; |
| 342 | tabsById.set(tabA.id, promptBlockedA); |
| 343 | await act(async () => { |
| 344 | desktopStub.emit("agent:event", { |
| 345 | kind: "ask_request", |
| 346 | tabId: tabA.id, |
| 347 | ask: { id: "pending-tab-a", questions: [{ id: "choice", prompt: "Keep me through A-X-A", options: [] }] }, |
| 348 | }); |
| 349 | await flushPromises(); |
| 350 | }); |
| 351 | eq(controller?.state.ask?.id, "pending-tab-a", "A starts the rapid switch with a visible ask"); |
| 352 | |
| 353 | const slowSwitchGate = deferred<void>(); |
| 354 | activationHolds.set(tabX.id, slowSwitchGate.promise); |
| 355 | let slowSwitch: Promise<TabMeta[] | undefined> | undefined; |
| 356 | await act(async () => { |
| 357 | slowSwitch = controller?.switchTab(tabX.id, tabX); |
| 358 | await flushPromises(); |
| 359 | }); |
| 360 | eq(controller?.activeTabId, tabX.id, "A→X renders the slow target optimistically"); |
| 361 | |
| 362 | await act(async () => { |
| 363 | await controller?.switchTab(tabA.id, promptBlockedA); |
| 364 | await flushPromises(); |
| 365 | }); |
| 366 | eq(controller?.activeTabId, tabA.id, "A→X→A returns to the prompt owner"); |
| 367 | eq(controller?.state.ask?.id, "pending-tab-a", "returning to A preserves its ask"); |
| 368 | ok(replayTargets.includes(tabA.id), "post-activation replay is scoped to A"); |
| 369 | |
| 370 | await act(async () => { |
| 371 | slowSwitchGate.resolve(); |
| 372 | activationHolds.delete(tabX.id); |
| 373 | await slowSwitch; |
| 374 | await flushPromises(); |
| 375 | }); |
| 376 | eq(controller?.activeTabId, tabA.id, "late X activation cannot replace A"); |
| 377 | eq(backendActiveId, tabA.id, "late X activation reasserts A as backend owner"); |
| 378 | eq(controller?.state.ask?.id, "pending-tab-a", "late X completion cannot clear A's ask"); |
| 379 | |
| 380 | // useController owns periodic runtime metadata refreshes. Unmount explicitly |
| 381 | // so the suite verifies their cleanup and does not keep the discovery runner |
| 382 | // alive after all assertions have passed. |
| 383 | await act(async () => root.unmount()); |
| 384 | dom.window.close(); |
| 385 | |
| 386 | console.log(`\n${passed} passed, ${failed} failed`); |
| 387 | if (failed > 0) process.exit(1); |
| 388 |