fix: keep cancelled turns as interrupted instead of merging into the next request

- Build runtime history per turn: a cancelled or failed reply keeps its partial
  text plus an interruption marker (or a marker-only placeholder), so its user
  message is never left unanswered and merged into the next request.
- Use the same history for send, manual compaction and Main compact validation.
- OpenCode: keep a cancelled prompt in the reused native session and prefix the
  next turn with a one-time synthetic interruption notice; failed prompts are
  still removed.
- Add an opt-in DeepSeek live test covering direct model and OpenCode.
This commit is contained in:
mesalogo
2026-09-30 02:18:33 +08:00
parent a662b16f5e
commit 608a9a5fec
9 changed files with 371 additions and 25 deletions
+47
View File
@@ -4188,6 +4188,53 @@ describe("OpenCodeRuntime embedded permission mediation", () => {
await runtime.dispose();
});
it("keeps a cancelled prompt and marks the next reused-session turn as following an interruption", async () => {
const setup = runClient([]);
const controller = new AbortController();
let subscriptions = 0;
vi.mocked(setup.event.subscribe).mockImplementation(async (_input, init) => {
const first = ++subscriptions === 1;
const signal = (init as { signal: AbortSignal }).signal;
return {
stream: (async function* () {
if (first) {
await new Promise<void>((resolve) => signal.addEventListener("abort", () => resolve(), { once: true }));
return;
}
yield { type: "session.idle", properties: { sessionID: "session-1" } };
})(),
} as never;
});
const runtime = embeddedRuntime(setup.client);
const collect = async (prompt: string, signal: AbortSignal) => {
for await (const event of runtime.run({
requestId: crypto.randomUUID(), conversationId: "conversation-1", prompt, workMode: "ask",
}, signal)) void event;
};
try {
const first = collect("first question", controller.signal).catch((error: unknown) => error);
await vi.waitFor(() => expect(setup.session.promptAsync).toHaveBeenCalledTimes(1));
controller.abort(new Error("cancelled"));
await expect(first).resolves.toMatchObject({ message: "cancelled" });
expect(setup.session.deleteMessage).not.toHaveBeenCalled();
await collect("new question", new AbortController().signal);
await collect("third question", new AbortController().signal);
const parts = vi.mocked(setup.session.promptAsync).mock.calls.map(
([input]) => (input as { parts: { text?: string; synthetic?: boolean }[] }).parts,
);
expect(parts[1]).toEqual([
expect.objectContaining({ synthetic: true, text: expect.stringContaining("interrupted the previous turn") }),
{ type: "text", text: "new question" },
]);
expect(parts[2]).toEqual([{ type: "text", text: "third question" }]);
expect(setup.session.create).toHaveBeenCalledOnce();
} finally {
await runtime.dispose();
}
});
it("deletes an ephemeral OpenCode session when released", async () => {
const { client, session } = runClient([
{
+32 -3
View File
@@ -72,6 +72,8 @@ import {
} from "./opencode-subagent";
import { promptWithUntrustedConversationHistory } from "./runtime-conversation-history";
const opencodeInterruptionNotice =
"[The user interrupted the previous turn before it finished. Its request may be incomplete; answer the new message below and do not resume or re-answer the interrupted request unless asked.]";
const STARTUP_TIMEOUT_MS = 30_000;
const STARTUP_POLL_INTERVAL_MS = 100;
const STARTUP_PROBE_TIMEOUT_MS = 500;
@@ -732,7 +734,11 @@ export class OpenCodeRuntime implements AgentRuntime {
private clientInitialization?: Promise<OpencodeClient>;
private server?: OpenCodeServer;
private startingChild?: SpawnedProcess;
private readonly sessions = new Map<string, { id: string; directory: string }>();
// `interrupted` marks a reused session whose last turn the user cancelled.
private readonly sessions = new Map<
string,
{ id: string; directory: string; interrupted?: boolean }
>();
private initializationController?: AbortController;
private initializationWaiters = 0;
private readonly sessionInitializations = new Map<string, Promise<string>>();
@@ -2461,6 +2467,14 @@ export class OpenCodeRuntime implements AgentRuntime {
request,
session.created,
);
const sessionState = this.sessions.get(request.conversationId);
// The native transcript keeps a cancelled turn; tell the model it was
// interrupted so it answers the new message instead of merging both.
const interruptionNotice =
!session.created && sessionState?.id === sessionId && sessionState.interrupted
? opencodeInterruptionNotice
: undefined;
if (sessionState?.id === sessionId) sessionState.interrupted = false;
const imageParts = (request.images ?? []).map((image) => ({
type: "file" as const,
mime: image.mediaType,
@@ -2515,6 +2529,15 @@ export class OpenCodeRuntime implements AgentRuntime {
.join("\n\n") || undefined,
...(toolOverrides ? { tools: toolOverrides } : {}),
parts: [
...(interruptionNotice
? [
{
type: "text" as const,
text: interruptionNotice,
synthetic: true,
},
]
: []),
{ type: "text" as const, text: promptText },
...imageParts,
],
@@ -3065,8 +3088,14 @@ export class OpenCodeRuntime implements AgentRuntime {
signal.removeEventListener("abort", abortSession);
// Keep the conversation locked until its session-wide abort settles.
await sessionAbort;
// A reused session would otherwise merge this unanswered prompt into the next turn.
if (
const currentSession = this.sessions.get(request.conversationId);
if (runFailed && signal.aborted && currentSession?.id === sessionId) {
// Keep the cancelled turn visible to the model, as it is to the user;
// the next prompt carries an interruption notice instead.
currentSession.interrupted = true;
} else if (
// A failed prompt is dropped from a reused session so it is not
// merged into the next turn.
runFailed &&
!selectedCommand &&
checklistMessageIds.size === 0 &&
+2 -5
View File
@@ -1,6 +1,7 @@
import { registerKnowledgeIpcHandlers } from './knowledge/knowledge-ipc'
import { registerModelSettingsIpcHandlers } from './model-settings-ipc'
import { knowledgeReferenceKey, toKnowledgeReference } from '../shared/knowledge-reference'
import { buildRuntimeHistory } from '../shared/runtime-history'
import { localInferenceService } from './local-inference-service'
import { inferenceActionSchema, inferenceCancelSchema } from '../shared/local-inference-contracts'
import type { NativeClientCoordinator } from './agent/native-client-coordinator'
@@ -4795,11 +4796,7 @@ export function registerIpcHandlers(
) {
throw new Error('对话 Runtime 或 Project 已更改,请刷新后重试')
}
const persistedHistory = conversation.messages
.filter(
(message) =>
message.state === 'complete' && message.content.trim()
)
const persistedHistory = buildRuntimeHistory(conversation.messages)
if (
persistedHistory.length !== request.history.length ||
persistedHistory.some(
+27
View File
@@ -23,6 +23,7 @@ import {
type ApplicationSettings,
} from "../../shared/application-settings-contracts";
import type { GlobalShortcutSettingsSnapshot } from "../../shared/shortcut";
import { cancelledResponseMarker } from "../../shared/runtime-history";
import type { TerminalSnapshot } from "../../shared/terminal-contracts";
import type { RuntimeNativeClientApi } from "./RuntimeNativeClientActions";
import { loadBrandingPreferences, saveBrandingPreferences } from "./branding";
@@ -6041,6 +6042,32 @@ describe("App", () => {
]);
});
it("sends a cancelled turn as interrupted instead of an unanswered user message", async () => {
render(<App />);
fireEvent.change(screen.getByLabelText("向 GoodBuddy 提问"), {
target: { value: "First question" },
});
fireEvent.click(await screen.findByLabelText("发送"));
await waitFor(() => expect(run).toHaveBeenCalledOnce());
const requestId = run.mock.calls[0]![0].requestId;
act(() => {
agentListener?.({ requestId, type: "text", delta: "Partial reply" });
agentListener?.({ requestId, type: "error", status: "cancelled", message: "请求已取消" });
});
fireEvent.change(screen.getByLabelText("向 GoodBuddy 提问"), {
target: { value: "New question" },
});
await waitFor(() => expect(screen.getByLabelText("发送")).toBeEnabled());
fireEvent.click(screen.getByLabelText("发送"));
await waitFor(() => expect(run).toHaveBeenCalledTimes(2));
const history = run.mock.calls[1]?.[0].history ?? [];
expect(history.slice(-2)).toEqual([
{ role: "user", content: "First question" },
{ role: "assistant", content: `Partial reply\n\n${cancelledResponseMarker}` },
]);
expect(run.mock.calls[1]?.[0].historyMessageIds).toHaveLength(history.length);
});
it("keeps a tool failure in details and hides retry after continuing", async () => {
render(<App />);
+3 -6
View File
@@ -13,6 +13,7 @@ import { PendingDocumentImports } from './PendingDocumentImports';
import { AttachmentCapabilityNotice } from './AttachmentCapabilityNotice';
import { DocumentConversationContext } from './DocumentConversationContext';
import { maximumAttachmentsPerMessage } from '../../shared/attachment-limits';
import { buildRuntimeHistory } from '../../shared/runtime-history';
import {
ApplicationAvailability,
ApplicationCenter,
@@ -7091,9 +7092,7 @@ function App(): React.JSX.Element {
8,
);
const historySnapshot = conversationSnapshot.messages;
const retainedHistorySnapshot = historySnapshot.filter(
(message) => message.state === "complete" && message.content.trim(),
);
const retainedHistorySnapshot = buildRuntimeHistory(historySnapshot);
const projectIdSnapshot = queuedInput
? queuedInput.projectId
: activeProjectId || undefined;
@@ -7435,9 +7434,7 @@ function App(): React.JSX.Element {
notify({ tone: "error", message: t("notices.remoteConversationRefreshFailed") });
return;
}
const history = completeConversation.messages.filter(
(message) => message.state === "complete" && message.content.trim(),
);
const history = buildRuntimeHistory(completeConversation.messages);
if (history.length < 2) {
notify({
tone: "info",
+1 -11
View File
@@ -1,11 +1 @@
import type { ConversationMessage } from '../../shared/assistant-contracts'
export function isCancelledMessage(message: Pick<ConversationMessage,
'state' | 'terminalStatus' | 'status'>): boolean {
if (message.state !== 'error') return false
if (message.terminalStatus !== undefined) return message.terminalStatus === 'cancelled'
// Released snapshots lack terminalStatus. Only recognize exact old labels;
// every new terminal event supplies the structured value above.
return message.status === '请求已取消' || message.status === '已取消' ||
message.status === '远端 Runtime 请求已取消'
}
export { isCancelledMessage } from '../../shared/runtime-history'
+51
View File
@@ -0,0 +1,51 @@
import { describe, expect, it } from 'vitest'
import {
buildRuntimeHistory,
cancelledResponseMarker,
failedResponseMarker
} from './runtime-history'
const id = (n: number) => `00000000-0000-4000-8000-${String(n).padStart(12, '0')}`
describe('buildRuntimeHistory', () => {
it('keeps a cancelled turn answered by its partial reply and an interruption marker', () => {
expect(buildRuntimeHistory([
{ id: id(1), role: 'user', content: 'First question', state: 'complete' },
{ id: id(2), role: 'assistant', content: 'Partial answer', state: 'error', terminalStatus: 'cancelled' },
{ id: id(3), role: 'user', content: 'continue', state: 'complete' },
{ id: id(4), role: 'assistant', content: '', state: 'error', terminalStatus: 'cancelled' },
{ id: id(5), role: 'user', content: 'continue', state: 'complete' },
{ id: id(6), role: 'assistant', content: 'Final answer', state: 'complete' }
])).toEqual([
{ id: id(1), role: 'user', content: 'First question' },
{ id: id(2), role: 'assistant', content: `Partial answer\n\n${cancelledResponseMarker}` },
{ id: id(3), role: 'user', content: 'continue' },
{ id: id(4), role: 'assistant', content: cancelledResponseMarker },
{ id: id(5), role: 'user', content: 'continue' },
{ id: id(6), role: 'assistant', content: 'Final answer' }
])
})
it('marks failed replies and recognizes legacy cancelled labels', () => {
expect(buildRuntimeHistory([
{ id: id(1), role: 'user', content: 'A', state: 'complete' },
{ id: id(2), role: 'assistant', content: '', state: 'error', status: '请求已取消' },
{ id: id(3), role: 'user', content: 'B', state: 'complete' },
{ id: id(4), role: 'assistant', content: 'half', state: 'error', terminalStatus: 'failed' }
]).map((message) => message.content)).toEqual([
'A', cancelledResponseMarker, 'B', `half\n\n${failedResponseMarker}`
])
})
it('skips streaming, empty and orphaned terminal messages', () => {
expect(buildRuntimeHistory([
{ id: id(1), role: 'assistant', content: 'orphan', state: 'error', terminalStatus: 'cancelled' },
{ id: id(2), role: 'user', content: ' ', state: 'complete' },
{ id: id(3), role: 'user', content: 'Q', state: 'complete' },
{ id: id(4), role: 'assistant', content: 'A', state: 'complete' },
{ id: id(5), role: 'assistant', content: 'late', state: 'error', terminalStatus: 'failed' },
{ id: id(6), role: 'user', content: 'Now', state: 'complete' },
{ id: id(7), role: 'assistant', content: 'typing', state: 'streaming' }
]).map((message) => message.id)).toEqual([id(3), id(4), id(6)])
})
})
+61
View File
@@ -0,0 +1,61 @@
import type { ConversationMessage } from './assistant-contracts'
export const cancelledResponseMarker =
'[This response was interrupted by the user before it finished.]'
export const failedResponseMarker =
'[This response failed before it finished.]'
type HistorySourceMessage = Pick<
ConversationMessage,
'id' | 'role' | 'content' | 'state' | 'terminalStatus' | 'status'
>
export type RuntimeHistoryMessage = {
id: string
role: ConversationMessage['role']
content: string
}
export function isCancelledMessage(
message: Pick<ConversationMessage, 'state' | 'terminalStatus' | 'status'>
): boolean {
if (message.state !== 'error') return false
if (message.terminalStatus !== undefined) return message.terminalStatus === 'cancelled'
// Released snapshots lack terminalStatus. Only recognize exact old labels;
// every new terminal event supplies the structured value above.
return message.status === '请求已取消' || message.status === '已取消' ||
message.status === '远端 Runtime 请求已取消'
}
/**
* Builds the model-visible history from stored messages. A cancelled or failed
* reply stays in place with its partial text and an explicit marker, so its
* user message is never left unanswered and merged into the next turn.
*/
export function buildRuntimeHistory(
messages: readonly HistorySourceMessage[]
): RuntimeHistoryMessage[] {
const history: RuntimeHistoryMessage[] = []
for (const message of messages) {
if (message.state === 'complete') {
if (message.content.trim()) {
history.push({ id: message.id, role: message.role, content: message.content })
}
continue
}
if (message.state !== 'error' || message.role !== 'assistant') continue
// Only a reply to a retained user turn needs a marker.
if (history.at(-1)?.role !== 'user') continue
const marker = isCancelledMessage(message)
? cancelledResponseMarker
: failedResponseMarker
history.push({
id: message.id,
role: 'assistant',
content: message.content.trim()
? `${message.content.trimEnd()}\n\n${marker}`
: marker
})
}
return history
}
+147
View File
@@ -0,0 +1,147 @@
// @vitest-environment node
import { mkdtemp, readFile, rm } from 'node:fs/promises'
import { tmpdir } from 'node:os'
import { join, resolve } from 'node:path'
import { parseEnv } from 'node:util'
import { afterAll, beforeAll, describe, expect, it } from 'vitest'
import { ModelAgentRuntime } from '../src/main/agent/model-runtime'
import { OpenCodeRuntime } from '../src/main/agent/opencode-runtime'
import type { AgentRuntime, RuntimeEvent } from '../src/main/agent/runtime'
import { buildRuntimeHistory } from '../src/shared/runtime-history'
// Opt-in: GB_CANCEL_LIVE_ENV=.env.deepseek.local npx vitest run tests/cancelled-turn-live.test.ts
const envPath = process.env.GB_CANCEL_LIVE_ENV
const opencodePath = resolve('.runtime-resources', process.arch, process.platform === 'win32' ? 'opencode.exe' : 'opencode')
const firstPrompt = 'List 40 European capital cities, one per line, each with a one-sentence description.'
const secondPrompt = 'What is 17 + 25? Reply with only the number and nothing else.'
type Env = { apiKey: string; baseUrl: string; model: string }
async function loadEnv(): Promise<Env> {
const env = parseEnv(await readFile(resolve(envPath!), 'utf8'))
return { apiKey: env.DEEPSEEK_API_KEY!, baseUrl: env.DEEPSEEK_BASE_URL!, model: env.DEEPSEEK_MODEL! }
}
/** Runs until some answer text streams, then cancels like the Stop button. */
async function runAndCancel(runtime: AgentRuntime, conversationId: string): Promise<string> {
const controller = new AbortController()
let partial = ''
try {
for await (const event of runtime.run({
requestId: crypto.randomUUID(), conversationId, workMode: 'ask', prompt: firstPrompt
}, controller.signal)) {
if (event.type === 'text') partial += event.delta
if (partial.length >= 60 && !controller.signal.aborted) controller.abort(new Error('用户取消了请求'))
}
} catch (error) {
if (!controller.signal.aborted) throw error
}
expect(controller.signal.aborted).toBe(true)
return partial
}
async function runText(
runtime: AgentRuntime,
request: Parameters<AgentRuntime['run']>[0]
): Promise<string> {
let output = ''
for await (const event of runtime.run(request, new AbortController().signal) as AsyncGenerator<RuntimeEvent>) {
if (event.type === 'text') output += event.delta
}
return output.trim()
}
function expectOnlyNewAnswer(output: string): void {
console.info(`SECOND_TURN_OUTPUT=${JSON.stringify(output)}`)
expect(output).toMatch(/\b42\b/u)
// Merging the cancelled turn would resume the capital list.
expect(output).not.toMatch(/Paris|Berlin|Madrid|Vienna|Rome|London/u)
expect(output.length).toBeLessThan(40)
}
describe.skipIf(!envPath)('cancelled turn handling with a real model', () => {
let env: Env
let workspace: string
beforeAll(async () => {
env = await loadEnv()
workspace = await mkdtemp(join(tmpdir(), 'gb-cancel-live-'))
})
afterAll(async () => {
await rm(workspace, { recursive: true, force: true })
})
it('direct model answers only the new message after a cancelled turn', async () => {
const runtime = new ModelAgentRuntime({
apiKey: env.apiKey, baseUrl: env.baseUrl, model: env.model,
protocol: 'openai-chat-completions', authentication: 'api-key'
})
try {
const conversationId = crypto.randomUUID()
const partial = await runAndCancel(runtime, conversationId)
console.info(`DIRECT_PARTIAL_CHARS=${partial.length}`)
// Same stored shape the renderer keeps after Stop.
const history = buildRuntimeHistory([
{ id: crypto.randomUUID(), role: 'user', content: firstPrompt, state: 'complete' },
{ id: crypto.randomUUID(), role: 'assistant', content: partial, state: 'error', terminalStatus: 'cancelled' }
])
expect(history).toHaveLength(2)
expectOnlyNewAnswer(await runText(runtime, {
requestId: crypto.randomUUID(), conversationId, workMode: 'ask', prompt: secondPrompt,
history: history.map(({ role, content }) => ({ role, content })),
historyMessageIds: history.map(({ id }) => id)
}))
} finally {
await runtime.dispose()
}
}, 180_000)
it('direct model answers only the new message when the cancelled turn had no output', async () => {
const runtime = new ModelAgentRuntime({
apiKey: env.apiKey, baseUrl: env.baseUrl, model: env.model,
protocol: 'openai-chat-completions', authentication: 'api-key'
})
try {
const history = buildRuntimeHistory([
{ id: crypto.randomUUID(), role: 'user', content: firstPrompt, state: 'complete' },
{ id: crypto.randomUUID(), role: 'assistant', content: '', state: 'error', terminalStatus: 'cancelled' }
])
expectOnlyNewAnswer(await runText(runtime, {
requestId: crypto.randomUUID(), conversationId: crypto.randomUUID(), workMode: 'ask', prompt: secondPrompt,
history: history.map(({ role, content }) => ({ role, content })),
historyMessageIds: history.map(({ id }) => id)
}))
} finally {
await runtime.dispose()
}
}, 180_000)
it('OpenCode reused session answers only the new message after a cancelled turn', async () => {
const runtime = new OpenCodeRuntime({
embedded: true, binaryPath: '', bundledBinaryPath: opencodePath, configPath: '',
defaultWorkspace: workspace,
modelProfile: {
id: crypto.randomUUID(), name: 'DeepSeek live', baseUrl: env.baseUrl, modelName: env.model,
apiKey: env.apiKey, protocol: 'openai-chat-completions', authentication: 'api-key'
}
})
try {
const conversationId = crypto.randomUUID()
const partial = await runAndCancel(runtime, conversationId)
console.info(`OPENCODE_PARTIAL_CHARS=${partial.length}`)
expectOnlyNewAnswer(await runText(runtime, {
requestId: crypto.randomUUID(), conversationId, workMode: 'ask', prompt: secondPrompt
}))
// The cancelled request stays in the native transcript, so the model
// still knows what was asked before.
const recall = await runText(runtime, {
requestId: crypto.randomUUID(), conversationId, workMode: 'ask',
prompt: 'In one short phrase, what did I ask for in my very first message of this conversation? Do not list anything.'
})
console.info(`OPENCODE_RECALL=${JSON.stringify(recall)}`)
expect(recall).toMatch(/capital/iu)
} finally {
await runtime.dispose()
}
}, 300_000)
})