mirror of
https://github.com/bytedance/deer-flow.git
synced 2026-11-03 00:50:35 +00:00
410 lines
12 KiB
TypeScript
410 lines
12 KiB
TypeScript
import { createServer } from "node:http";
|
||
import type { AddressInfo } from "node:net";
|
||
|
||
import { expect, test } from "@playwright/test";
|
||
|
||
import { mockLangGraphAPI } from "./utils/mock-api";
|
||
|
||
const threadId = "00000000-0000-0000-0000-000000004389";
|
||
|
||
for (const { debug, content, status, label } of [
|
||
{
|
||
debug: false,
|
||
content: "denied: " + "x".repeat(20000),
|
||
status: "error",
|
||
label: "Debug disabled",
|
||
},
|
||
{
|
||
debug: true,
|
||
content: "denied: " + "x".repeat(20000),
|
||
status: "error",
|
||
label: "bounded error",
|
||
},
|
||
{
|
||
debug: true,
|
||
content: '{"run_id":9223372036854775807,"ratio":0.1234567890123456789}',
|
||
status: "success",
|
||
label: "nested numeric precision",
|
||
},
|
||
{
|
||
debug: true,
|
||
content: "9223372036854775807",
|
||
status: "success",
|
||
label: "top-level numeric precision",
|
||
},
|
||
] as const) {
|
||
test(`generic tool details: ${label}`, async ({ page }) => {
|
||
await page.addInitScript((enabled) => {
|
||
localStorage.setItem(
|
||
"deerflow.local-settings",
|
||
JSON.stringify({
|
||
tokenUsage: {
|
||
headerTotal: true,
|
||
inlineMode: enabled ? "step_debug" : "per_turn",
|
||
},
|
||
}),
|
||
);
|
||
}, debug);
|
||
mockLangGraphAPI(page, {
|
||
threads: [
|
||
{
|
||
thread_id: threadId,
|
||
title: "Tool details",
|
||
messages: [
|
||
{ type: "human", id: "human", content: "Look up the run" },
|
||
{
|
||
type: "ai",
|
||
id: "ai-tool",
|
||
content: "",
|
||
tool_calls: [
|
||
{
|
||
id: "call-4389",
|
||
name: "mcp_lookup",
|
||
args: { run_id: "run-4389" },
|
||
},
|
||
],
|
||
},
|
||
{
|
||
type: "tool",
|
||
id: "tool-result",
|
||
tool_call_id: "call-4389",
|
||
status,
|
||
content,
|
||
},
|
||
{ type: "ai", id: "answer", content: "The lookup failed." },
|
||
],
|
||
},
|
||
],
|
||
});
|
||
// 默认 mock 关闭了 Token Usage；这里启用入口，但不提供任何调用统计。
|
||
await page.route("**/api/models", (route) =>
|
||
route.fulfill({ json: { models: [], token_usage: { enabled: true } } }),
|
||
);
|
||
await page.goto(`/workspace/chats/${threadId}`);
|
||
await expect(
|
||
page.getByText("The lookup failed.", { exact: true }),
|
||
).toBeVisible();
|
||
const trigger = page.getByRole("button", {
|
||
name: "Tool details: mcp_lookup (call-4389)",
|
||
exact: true,
|
||
});
|
||
if (!debug) {
|
||
await expect(trigger).toHaveCount(0);
|
||
return;
|
||
}
|
||
await expect(trigger).toHaveAttribute("aria-expanded", "false");
|
||
await expect(
|
||
page.getByRole("region", { name: "Input", exact: true }),
|
||
).toHaveCount(0);
|
||
await trigger.click();
|
||
await expect(
|
||
page.getByRole("region", { name: "Input", exact: true }),
|
||
).toContainText("run-4389");
|
||
await expect(
|
||
page.getByRole("region", { name: "Call ID", exact: true }),
|
||
).toContainText("call-4389");
|
||
const error = page.getByRole("region", {
|
||
name: status === "error" ? "Error" : "Result",
|
||
exact: true,
|
||
});
|
||
if (status === "success") {
|
||
expect(await error.locator("pre").textContent()).toBe(content);
|
||
await expect(
|
||
error.getByText("Preview truncated", { exact: false }),
|
||
).toHaveCount(0);
|
||
await page
|
||
.context()
|
||
.grantPermissions(["clipboard-read", "clipboard-write"]);
|
||
await error
|
||
.getByRole("button", { name: "Copy to clipboard: Result", exact: true })
|
||
.click();
|
||
await expect
|
||
.poll(() => page.evaluate(() => navigator.clipboard.readText()))
|
||
.toBe(content);
|
||
return;
|
||
}
|
||
await expect(error).toContainText("denied:");
|
||
await expect(error).toContainText("Preview truncated");
|
||
expect(
|
||
(await error.locator("pre").textContent())?.length,
|
||
).toBeLessThanOrEqual(12000);
|
||
await page.screenshot({
|
||
path: test.info().outputPath("tool-details.png"),
|
||
fullPage: true,
|
||
});
|
||
await trigger.click();
|
||
await expect(error).toHaveCount(0);
|
||
});
|
||
}
|
||
|
||
for (const locale of ["en-US", "zh-CN"] as const) {
|
||
test(`tool failures and confirmed scheduler labels (${locale})`, async ({
|
||
page,
|
||
baseURL,
|
||
}) => {
|
||
const zh = locale === "zh-CN";
|
||
const runId = "00000000-0000-0000-0000-000000004390";
|
||
await page.setViewportSize({ width: 1440, height: 1000 });
|
||
await page
|
||
.context()
|
||
.addCookies([{ name: "locale", value: locale, url: baseURL! }]);
|
||
const messages: unknown[] = [
|
||
{
|
||
type: "human",
|
||
id: "human",
|
||
content: zh
|
||
? "创建定时任务，然后核对输入文件和账户。"
|
||
: "Create a scheduled task, then check the input file and account.",
|
||
},
|
||
{
|
||
type: "ai",
|
||
id: "ai-schedule",
|
||
content: "",
|
||
tool_calls: [
|
||
{ id: "schedule", name: "schedule_task", args: { action: "create" } },
|
||
],
|
||
},
|
||
];
|
||
const controller = mockLangGraphAPI(page, {
|
||
threads: [
|
||
{
|
||
thread_id: threadId,
|
||
title: zh ? "工具执行反馈" : "Tool execution feedback",
|
||
messages: [messages[0]],
|
||
},
|
||
],
|
||
});
|
||
await page.route("**/api/models", (route) =>
|
||
route.fulfill({ json: { models: [], token_usage: { enabled: true } } }),
|
||
);
|
||
let finished = false;
|
||
const run = () => ({
|
||
run_id: runId,
|
||
thread_id: threadId,
|
||
assistant_id: "lead_agent",
|
||
status: finished ? "success" : "running",
|
||
metadata: {},
|
||
kwargs: {},
|
||
created_at: "2026-10-25T00:00:00Z",
|
||
updated_at: "2026-10-25T00:00:01Z",
|
||
});
|
||
await page.route(/\/api\/langgraph\/threads\/[^/]+\/runs(\?|$)/, (route) =>
|
||
route.fulfill({ json: [run()] }),
|
||
);
|
||
await page.route(
|
||
`**/api/langgraph/threads/${threadId}/runs/${runId}`,
|
||
(route) => route.fulfill({ json: run() }),
|
||
);
|
||
// Stream actual SSE bytes so the pending assertion runs before any tool receipt.
|
||
let finishStream: (() => void) | undefined;
|
||
const server = createServer((_request, response) => {
|
||
response.writeHead(200, {
|
||
"Access-Control-Allow-Origin": "*",
|
||
"Content-Type": "text/event-stream",
|
||
});
|
||
const frame = (event: string, data: unknown) =>
|
||
`event: ${event}\ndata: ${JSON.stringify(data)}\n\n`;
|
||
response.write(frame("metadata", { run_id: runId, thread_id: threadId }));
|
||
response.write(frame("values", { messages }));
|
||
finishStream = () =>
|
||
response.end(frame("values", { messages }) + frame("end", {}));
|
||
});
|
||
await new Promise<void>((resolve) =>
|
||
server.listen(0, "127.0.0.1", resolve),
|
||
);
|
||
const { port } = server.address() as AddressInfo;
|
||
await page.route(
|
||
`**/api/langgraph/threads/${threadId}/runs/${runId}/stream*`,
|
||
(route) => route.continue({ url: `http://127.0.0.1:${port}/stream` }),
|
||
);
|
||
try {
|
||
await page.goto(`/workspace/chats/${threadId}`);
|
||
await expect(
|
||
page.getByText(zh ? "定时任务操作" : "Scheduled task operation", {
|
||
exact: true,
|
||
}),
|
||
).toBeVisible();
|
||
await expect(
|
||
page.getByText(zh ? "已创建定时任务" : "Scheduled a task", {
|
||
exact: true,
|
||
}),
|
||
).toHaveCount(0);
|
||
await expect(
|
||
page.getByRole("button", {
|
||
name: zh ? /^工具详情:/ : /^Tool details:/,
|
||
}),
|
||
).toHaveCount(0);
|
||
await page.screenshot({
|
||
path: test.info().outputPath(`scheduler-pending-${locale}.png`),
|
||
animations: "disabled",
|
||
fullPage: true,
|
||
});
|
||
messages.push({
|
||
type: "tool",
|
||
id: "schedule-result",
|
||
tool_call_id: "schedule",
|
||
content: JSON.stringify({ action: "create" }),
|
||
status: "success",
|
||
});
|
||
for (const [id, name, description, content, status, meta] of [
|
||
[
|
||
"bash-error",
|
||
"bash",
|
||
zh ? "核对输入文件" : "Check input file",
|
||
"File not found",
|
||
"error",
|
||
"error",
|
||
],
|
||
[
|
||
"mcp-error",
|
||
"mcp_lookup",
|
||
zh ? "查询账户" : "Look up account",
|
||
"Provider denied the request",
|
||
"success",
|
||
"error",
|
||
],
|
||
[
|
||
"read-ok",
|
||
"read_file",
|
||
zh ? "读取本地记录" : "Read local record",
|
||
'{"error":"ordinary document data"}',
|
||
"success",
|
||
"success",
|
||
],
|
||
] as const) {
|
||
messages.push(
|
||
{
|
||
type: "ai",
|
||
id: `ai-${id}`,
|
||
content: "",
|
||
tool_calls: [
|
||
{
|
||
id,
|
||
name,
|
||
args: {
|
||
description,
|
||
path: "/mnt/user-data/workspace/input.json",
|
||
},
|
||
},
|
||
],
|
||
},
|
||
{
|
||
type: "tool",
|
||
id: `result-${id}`,
|
||
tool_call_id: id,
|
||
content,
|
||
status,
|
||
additional_kwargs: { deerflow_tool_meta: { status: meta } },
|
||
},
|
||
);
|
||
}
|
||
messages.push({
|
||
type: "ai",
|
||
id: "answer",
|
||
content: zh
|
||
? "定时任务已创建，文件和账户核对未成功。"
|
||
: "The scheduled task was created; the file and account checks failed.",
|
||
});
|
||
controller.upsertThread({
|
||
thread_id: threadId,
|
||
title: zh ? "工具执行反馈" : "Tool execution feedback",
|
||
messages,
|
||
});
|
||
finished = true;
|
||
finishStream?.();
|
||
const expandSteps = () =>
|
||
page.getByRole("button", {
|
||
name: zh ? /查看其他.*步骤/ : /more steps/,
|
||
});
|
||
await expandSteps().click();
|
||
await expect(
|
||
page.getByText(zh ? "已创建定时任务" : "Scheduled a task", {
|
||
exact: true,
|
||
}),
|
||
).toBeVisible();
|
||
await expect(
|
||
page.getByText(zh ? "执行失败" : "Failed", { exact: true }),
|
||
).toHaveCount(2);
|
||
await expect(
|
||
page.getByRole("button", {
|
||
name: zh ? /^工具详情:/ : /^Tool details:/,
|
||
}),
|
||
).toHaveCount(0);
|
||
await expect(
|
||
page.getByText(zh ? "结果可能不完整" : "Result may be incomplete", {
|
||
exact: true,
|
||
}),
|
||
).toHaveCount(0);
|
||
const successfulRead = page.getByText(
|
||
zh ? "读取本地记录" : "Read local record",
|
||
{ exact: true },
|
||
);
|
||
await expect(successfulRead).toBeVisible();
|
||
await expect
|
||
.poll(() =>
|
||
successfulRead.evaluate((element) => {
|
||
let opacity = 1;
|
||
for (
|
||
let node: Element | null = element;
|
||
node;
|
||
node = node.parentElement
|
||
) {
|
||
opacity *= Number(getComputedStyle(node).opacity);
|
||
}
|
||
return opacity;
|
||
}),
|
||
)
|
||
.toBe(1);
|
||
await page.mouse.move(0, 0);
|
||
await page.screenshot({
|
||
path: test.info().outputPath(`normal-errors-${locale}.png`),
|
||
animations: "disabled",
|
||
fullPage: true,
|
||
});
|
||
await page.reload();
|
||
await expandSteps().click();
|
||
await expect(
|
||
page.getByText(zh ? "执行失败" : "Failed", { exact: true }),
|
||
).toHaveCount(2);
|
||
await page.evaluate(() =>
|
||
localStorage.setItem(
|
||
"deerflow.local-settings",
|
||
JSON.stringify({
|
||
tokenUsage: { headerTotal: true, inlineMode: "step_debug" },
|
||
}),
|
||
),
|
||
);
|
||
await page.reload();
|
||
await expandSteps().click();
|
||
await expect(
|
||
page.getByRole("button", {
|
||
name: zh ? /^工具详情:/ : /^Tool details:/,
|
||
}),
|
||
).toHaveCount(2);
|
||
await page
|
||
.getByRole("button", {
|
||
name: `${zh ? "工具详情" : "Tool details"}: mcp_lookup (mcp-error)`,
|
||
exact: true,
|
||
})
|
||
.click();
|
||
await expect(
|
||
page.getByRole("region", { name: zh ? "错误" : "Error", exact: true }),
|
||
).toContainText("Provider denied the request");
|
||
await expect(
|
||
page.getByRole("region", {
|
||
name: zh ? "调用 ID" : "Call ID",
|
||
exact: true,
|
||
}),
|
||
).toContainText("mcp-error");
|
||
await page.screenshot({
|
||
path: test.info().outputPath(`existing-debug-${locale}.png`),
|
||
animations: "disabled",
|
||
fullPage: true,
|
||
});
|
||
} finally {
|
||
server.closeAllConnections();
|
||
await new Promise<void>((resolve) => server.close(() => resolve()));
|
||
}
|
||
});
|
||
}
|