返回 CodeWhale
app-controls.test.mjs
根目录 / crates / tui / plugins / computer-use / tests / app-controls.test.mjs
1 import { test } from "node:test";
2 import assert from "node:assert/strict";
3 import fs from "node:fs";
4 import os from "node:os";
5 import path from "node:path";
6 import { spawn } from "node:child_process";
7 import { fileURLToPath } from "node:url";
8 import { setTimeout as delay } from "node:timers/promises";
9 import { Duplex } from "node:stream";
10 import { appRequest } from "../src/app-socket.mjs";
11
12 function controlClient(stream, { timeoutMs = 5_000, signal, diagnostics = () => "" } = {}) {
13 let seq = 0, buffer = "", failure;
14 const pending = new Map();
15 const fail = message => {
16 if (failure) return;
17 failure = new Error(`FD3 control: ${message}${diagnostics() ? `\n${diagnostics()}` : ""}`);
18 for (const { reject, timer } of pending.values()) { clearTimeout(timer); reject(failure); }
19 pending.clear();
20 stream.destroy();
21 };
22 stream.setEncoding("utf8");
23 stream.on("error", error => fail(error.message));
24 stream.on("close", () => fail("channel closed before a reply"));
25 stream.on("end", () => fail("channel ended before a reply"));
26 const abort = () => fail("test cancelled");
27 signal?.addEventListener("abort", abort, { once: true });
28 stream.once("close", () => signal?.removeEventListener("abort", abort));
29 if (signal?.aborted) abort();
30 stream.on("data", chunk => {
31 buffer += chunk;
32 if (buffer.length > 8192) return fail("reply exceeds 8192 bytes");
33 let newline;
34 while ((newline = buffer.indexOf("\n")) >= 0) {
35 const line = buffer.slice(0, newline); buffer = buffer.slice(newline + 1);
36 let state;
37 try { state = JSON.parse(line); } catch { return fail("malformed JSON reply"); }
38 if (!state || typeof state !== "object") return fail("malformed control state");
39 const waiting = pending.get(state.id);
40 if (!waiting) continue;
41 if (state.error) return fail(`${waiting.command} failed: ${state.error}`);
42 clearTimeout(waiting.timer); pending.delete(state.id); waiting.resolve(state);
43 }
44 });
45 return command => {
46 if (failure) return Promise.reject(failure);
47 const id = ++seq;
48 return new Promise((resolve, reject) => {
49 const timer = setTimeout(() => fail(`${command} timed out after ${timeoutMs}ms`), timeoutMs);
50 pending.set(id, { command, resolve, reject, timer });
51 try { stream.write(JSON.stringify({ id, command }) + "\n", error => { if (error) fail(error.message); }); }
52 catch (error) { fail(error.message); }
53 });
54 };
55 }
56
57 const root = fileURLToPath(new URL("../", import.meta.url));
58 test("human Pause and Stop cannot be bypassed; a lost owner exits and can reopen stopped", { timeout: 20_000 }, async t => {
59 const dir = fs.mkdtempSync(path.join(os.tmpdir(), "cu-control-"));
60 const log = path.join(dir, "calls.jsonl");
61 const endpoint = process.platform === "win32"
62 ? `\\\\.\\pipe\\cu-control-${process.pid}-${path.basename(dir)}`
63 : path.join(dir, "app.sock");
64 const daemonOptions = { env: { ...process.env,
65 CODEWHALE_CU_STATE_DIR: dir, CODEWHALE_CU_APP_SOCKET: endpoint, CODEWHALE_CU_APP_WARM: "off",
66 CODEWHALE_CU_TEST_BACKEND: path.join(root, "tests/fixtures/session-backend.mjs"), CU_SESSION_CALLS: log, CODEWHALE_CU_CONTROL_FD: "3" }, stdio: ["ignore", "ignore", "pipe", "overlapped"] };
67 let errors = "", daemon, exited;
68 function launch() {
69 daemon = spawn(process.execPath, [path.join(root, "app/daemon.mjs")], daemonOptions);
70 daemon.stderr.on("data", chunk => { errors += chunk; });
71 exited = new Promise(resolve => { daemon.once("exit", resolve); daemon.once("error", error => { errors += error.message; resolve(); }); });
72 }
73 launch();
74 const sockets = [];
75 const previousSocket = process.env.CODEWHALE_CU_APP_SOCKET;
76 process.env.CODEWHALE_CU_APP_SOCKET = endpoint;
77 t.after(async () => {
78 if (previousSocket === undefined) delete process.env.CODEWHALE_CU_APP_SOCKET;
79 else process.env.CODEWHALE_CU_APP_SOCKET = previousSocket;
80 sockets.forEach(socket => socket.destroy()); daemon.stdio[3].destroy();
81 let deadline;
82 const forceKill = setTimeout(() => daemon.kill("SIGKILL"), 3_000);
83 try {
84 if (daemon.exitCode === null && daemon.signalCode === null) daemon.kill("SIGTERM");
85 await Promise.race([exited, new Promise((_, reject) => { deadline = setTimeout(() => reject(new Error("Daemon did not exit after SIGKILL")), 4_000); })]);
86 } finally { clearTimeout(forceKill); clearTimeout(deadline); fs.rmSync(dir, { recursive: true, force: true }); }
87 });
88 async function until(predicate) { for(let i=0;i<250;i++) { if(predicate()) return; if (daemon.exitCode !== null || daemon.signalCode !== null) throw new Error(`Daemon exited: ${errors}`); await delay(20, undefined, { signal: t.signal }); } throw new Error(`Timed out: ${errors}`); }
89 // Named pipes on Windows have no filesystem entry; both transports publish
90 // the same run receipt only after the listener is ready.
91 await until(() => fs.existsSync(path.join(dir, "app-run.json")));
92 function request(payload, keepOpen = false) {
93 return appRequest(payload, { keepOpen, timeoutMs: 5_000, signal: t.signal }).then(result => {
94 if (!keepOpen) return result;
95 sockets.push(result.socket); return result.reply;
96 });
97 }
98 const control = controlClient(daemon.stdio[3], { signal: t.signal, diagnostics: () => errors + (fs.existsSync(log) ? fs.readFileSync(log,"utf8") : "") });
99 assert.equal((await control("status")).mode, "ready", "human-control channel responds before input starts");
100 assert.equal((await request({tool:"hello"})).app.controlOwner,true);
101 const sessionId="human-controls";
102 const { leaseToken }=await request({tool:"open_session",sessionId},true);
103 const call=(tool,args={})=>request({tool,args,sessionId,leaseToken});
104 assert.equal((await call("get_app_state",{app_ref:{name:"Practice"}})).ok,true);
105 const held=call("hold_key",{text:"must be cancelled"});
106 held.catch(() => {}); // The result is asserted below; avoid orphan rejections if control fails first.
107 await until(()=>fs.existsSync(log)&&fs.readFileSync(log,"utf8").includes("child_started"));
108 const queued=call("type",{text:"must not replay"});
109 queued.catch(() => {});
110 t.diagnostic("control while input runs: " + JSON.stringify(await control("status")));
111 const pausing = control("pause"); pausing.catch(() => {});
112 await delay(200);
113 t.diagnostic("control after pause: " + JSON.stringify(await control("status")));
114 const paused=await pausing;
115 assert.equal(paused.mode,"paused"); assert.equal(paused.cleanupPending,false);
116 assert.equal((await held).ok,false); assert.equal((await queued).ok,false);
117 assert.equal((await call("type",{text:"while paused"})).error.code,"control_paused");
118 for(const tool of ["resume","set_control_mode","control","updates"]) assert.equal((await call(tool)).error.code,"tool_not_allowed");
119 assert.equal((await control("resume")).mode,"ready");
120 assert.equal((await call("type",{text:"allowed again"})).ok,true);
121 assert.equal((await control("stop")).mode,"stopped");
122 await control("resume");
123 assert.equal((await call("type",{text:"old stopped owner"})).error.code,"control_stopped");
124 const records=fs.readFileSync(log,"utf8").trim().split("\n").map(JSON.parse);
125 assert.ok(records.some(record=>record.method==="child_released"));
126 assert.deepEqual(records.filter(record=>record.method==="type").map(record=>record.text),["allowed again"]);
127 // Owner loss must release input and retire the listener, so relaunching
128 // restores the menu controls without silently authorizing new input.
129 daemon.stdio[3].destroy();
130 await until(() => daemon.exitCode !== null);
131 assert.equal(daemon.exitCode,0,errors);
132 assert.equal(JSON.parse(fs.readFileSync(path.join(dir,"control.json"))).mode,"stopped");
133 assert.equal(fs.existsSync(path.join(dir,"app-run.json")),false);
134 await assert.rejects(request({tool:"hello"}),error=>error.code==="app_unavailable");
135 launch();
136 await until(() => fs.existsSync(path.join(dir,"app-run.json")));
137 const reopened = controlClient(daemon.stdio[3], { signal: t.signal, diagnostics: () => errors + (fs.existsSync(log) ? fs.readFileSync(log,"utf8") : "") });
138 assert.equal((await reopened("status")).mode,"stopped");
139 assert.equal((await request({tool:"hello"})).app.controlOwner,true);
140 const fresh=await request({tool:"open_session",sessionId:"fresh"},true);
141 assert.equal((await request({tool:"get_app_state",sessionId:"fresh",leaseToken:fresh.leaseToken})).error.code,"control_stopped");
142 assert.equal((await reopened("resume")).mode,"ready");
143 assert.equal((await request({tool:"get_app_state",args:{app_ref:{name:"Reopened"}},sessionId:"fresh",leaseToken:fresh.leaseToken})).ok,true);
144 assert.equal((await call("type",{text:"old lease after reopen"})).error.code,"session_owner_required");
145 });
146
147 test("silent FD3 peer rejects every pending command and closes the channel", { timeout: 2_000 }, async () => {
148 const stream = new Duplex({ read() {}, write(_chunk, _encoding, done) { done(); } });
149 const control = controlClient(stream, { timeoutMs: 30, diagnostics: () => "fixture daemon stderr" });
150 const results = await Promise.allSettled([control("pause"), control("status")]);
151 for (const result of results) {
152 assert.equal(result.status, "rejected");
153 assert.match(result.reason.message, /pause timed out after 30ms\nfixture daemon stderr/);
154 }
155 assert.equal(stream.destroyed, true);
156 await assert.rejects(control("resume"), /timed out/);
157 });
158
159 for (const [name, respond, expected] of [
160 ["EOF", stream => stream.push(null), /channel ended/],
161 ["close", stream => stream.destroy(), /channel closed/],
162 ["error", stream => stream.destroy(new Error("broken pipe")), /broken pipe/],
163 ["invalid JSON", stream => stream.push("not json\n"), /malformed JSON/],
164 ["invalid state", stream => stream.push("null\n"), /malformed control state/],
165 ["oversized reply", stream => stream.push("x".repeat(8193)), /exceeds 8192/],
166 ["command failure", stream => stream.push('{"id":1,"error":"disk full"}\n'), /pause failed: disk full/],
167 ]) {
168 test(`FD3 ${name} rejects a waiting control command`, { timeout: 2_000 }, async () => {
169 const stream = new Duplex({ read() {}, write(_chunk, _encoding, done) { done(); queueMicrotask(() => respond(this)); } });
170 const control = controlClient(stream);
171 await assert.rejects(control("pause"), expected);
172 assert.equal(stream.destroyed, true);
173 });
174 }
175
176 test("FD3 test cancellation rejects pending commands", { timeout: 2_000 }, async () => {
177 const stream = new Duplex({ read() {}, write(_chunk, _encoding, done) { done(); } });
178 const abort = new AbortController();
179 const control = controlClient(stream, { signal: abort.signal });
180 const rejected = assert.rejects(control("pause"), /test cancelled/);
181 abort.abort(); await rejected;
182 assert.equal(stream.destroyed, true);
183 });
184
185 test("FD3 matches fragmented replies to their command IDs", { timeout: 2_000 }, async () => {
186 const requests = [];
187 const stream = new Duplex({ read() {}, write(chunk, _encoding, done) { requests.push(JSON.parse(chunk)); done(); } });
188 const control = controlClient(stream);
189 const replies = Promise.all([control("status"), control("pause")]);
190 stream.push('{"id":2,"mode":"pau');
191 stream.push('sed"}\n{"id":1,"mode":"ready"}\n');
192 assert.deepEqual(await replies, [{ id: 1, mode: "ready" }, { id: 2, mode: "paused" }]);
193 assert.deepEqual(requests.map(request => request.command), ["status", "pause"]);
194 stream.destroy();
195 });
196
196 lines Plain Text