Files
makelore/scripts/run-pi-subagent-packaged-smoke.mjs

285 lines
11 KiB
JavaScript

#!/usr/bin/env node
import { _electron as electron } from '@playwright/test';
import { mkdir, mkdtemp, rm, writeFile } from 'node:fs/promises';
import { createServer } from 'node:net';
import { tmpdir } from 'node:os';
import { dirname, join, resolve } from 'node:path';
import { pathToFileURL } from 'node:url';
import { performance } from 'node:perf_hooks';
import { defaultProductExecutable } from './lib/pi-product-artifact.mjs';
function parseArgs(argv, projectRoot = process.cwd()) {
const options = {
projectRoot: resolve(projectRoot),
runtimeRoot: undefined,
electronExecutable: undefined,
reportPath: undefined,
verifyCleanupFailure: true,
};
for (let index = 0; index < argv.length; index += 1) {
const argument = argv[index];
const value = argv[index + 1];
if (!value || value.startsWith('--')) throw new Error(`${argument} requires a value`);
if (argument === '--runtime-root') options.runtimeRoot = resolve(value);
else if (argument === '--electron-executable') options.electronExecutable = resolve(value);
else if (argument === '--report') options.reportPath = resolve(value);
else throw new Error(`Unknown argument: ${argument}`);
index += 1;
}
options.electronExecutable ??= defaultProductExecutable(options.projectRoot);
return options;
}
async function allocatePort() {
return await new Promise((resolvePort, reject) => {
const server = createServer();
server.once('error', reject);
server.listen(0, '127.0.0.1', () => {
const address = server.address();
if (!address || typeof address === 'string') {
server.close(() => reject(new Error('Failed to allocate PI proof Host API port')));
return;
}
server.close((error) => error ? reject(error) : resolvePort(address.port));
});
});
}
function assertPackagedMain(result) {
if (result?.packagedMain?.isPackaged !== true || result?.packagedMain?.appPathUsesAsar !== true) {
throw new Error(`PI proof did not execute from packaged app.asar Main: ${JSON.stringify(result?.packagedMain)}`);
}
}
function assertActivePressure(pressure) {
const expected = {
parentWorkers: 4,
childWorkers: 4,
parentProcesses: 4,
childProcesses: 4,
liveProcesses: 8,
parentProviderRequests: 4,
childProviderRequests: 4,
processBudget: 8,
childPermits: 4,
dispatches: 4,
writeLeases: 4,
};
const actual = {
parentWorkers: pressure?.parentWorkers,
childWorkers: pressure?.childWorkers,
parentProcesses: pressure?.parentProcessIds?.length,
childProcesses: pressure?.childProcessIds?.length,
liveProcesses: pressure?.liveProcessIds?.length,
parentProviderRequests: pressure?.providerRequests?.parent,
childProviderRequests: pressure?.providerRequests?.child,
processBudget: pressure?.processBudget?.active,
childPermits: pressure?.childPermits?.active,
dispatches: pressure?.dispatches?.active,
writeLeases: pressure?.writeLeases?.active,
};
if (JSON.stringify(actual) !== JSON.stringify(expected)) {
throw new Error(`PI pressure did not reach 4 parent + 4 child state: ${JSON.stringify(actual)}`);
}
}
function assertReleasedPressure(pressure) {
const counts = [
pressure?.parentWorkers,
pressure?.childWorkers,
pressure?.parentProcessIds?.length,
pressure?.childProcessIds?.length,
pressure?.liveProcessIds?.length,
pressure?.providerRequests?.parent,
pressure?.providerRequests?.child,
pressure?.processBudget?.active,
pressure?.processBudget?.waiting,
pressure?.childPermits?.active,
pressure?.childPermits?.waiting,
pressure?.dispatches?.active,
pressure?.dispatches?.parents,
pressure?.writeLeases?.active,
pressure?.writeLeases?.waiting,
];
if (counts.some((value) => value !== 0)) {
throw new Error(`PI pressure resources were not fully released: ${JSON.stringify(pressure)}`);
}
}
function assertManagedTurns(extension) {
const expectedMilestones = 'worker.queue_wait,resources.ready,worker.spawn,rpc.ready,session.open,prompt.accepted,agent.start,provider.first_event,agent.settled';
const expectedDelayMs = extension?.providerFirstEventDelayMs;
if (!Number.isFinite(expectedDelayMs) || expectedDelayMs < 1) {
throw new Error(`PI proof did not expose a controlled Provider first-event delay: ${expectedDelayMs}`);
}
for (const turn of extension?.managedTurns ?? []) {
if (turn.milestones.map(({ milestone }) => milestone).join(',') !== expectedMilestones) {
throw new Error(`PI managed turn timeline is incomplete: ${JSON.stringify(turn)}`);
}
const agentStart = turn.milestones.find(({ milestone }) => milestone === 'agent.start');
const providerFirstEvent = turn.milestones.find(({ milestone }) => milestone === 'provider.first_event');
if (agentStart?.source !== 'pi.agent_start'
|| providerFirstEvent?.source !== 'pi.assistant_message_start'
|| providerFirstEvent.at <= agentStart.at
|| providerFirstEvent.durationMs < Math.floor(expectedDelayMs * 0.75)) {
throw new Error(`PI Provider first-event milestone is not Provider-response-backed: ${JSON.stringify(turn)}`);
}
}
}
async function evaluateProof(electronApplication, action) {
return await electronApplication.evaluate(async (_electron, requestedAction) => {
const proof = globalThis.__niancodeRunPiReleaseProofE2E;
if (typeof proof !== 'function') throw new Error('Packaged Main has no PI release proof entry');
return await proof(requestedAction);
}, action);
}
async function closeApplication(electronApplication) {
await Promise.race([
electronApplication.close().catch(() => undefined),
new Promise((resolveTimeout) => setTimeout(resolveTimeout, 5_000)),
]);
}
export async function runPackagedProductProof(options) {
const scratchRoot = await mkdtemp(join(tmpdir(), 'makelore-pi-final-product-proof-'));
const homeDir = join(scratchRoot, 'home');
const userDataDir = join(scratchRoot, 'user-data');
await Promise.all([
mkdir(join(homeDir, '.config'), { recursive: true }),
mkdir(join(homeDir, 'AppData', 'Local'), { recursive: true }),
mkdir(join(homeDir, 'AppData', 'Roaming'), { recursive: true }),
mkdir(userDataDir, { recursive: true }),
]);
const hostApiPort = await allocatePort();
let electronApplication;
let pressureActive = false;
try {
electronApplication = await electron.launch({
executablePath: options.electronExecutable,
env: {
...process.env,
HOME: homeDir,
USERPROFILE: homeDir,
APPDATA: join(homeDir, 'AppData', 'Roaming'),
LOCALAPPDATA: join(homeDir, 'AppData', 'Local'),
XDG_CONFIG_HOME: join(homeDir, '.config'),
NIANCODE_E2E: '1',
NIANCODE_E2E_SKIP_SETUP: '1',
NIANCODE_USER_DATA_DIR: userDataDir,
NIANCODE_PORT_NIANCODE_HOST_API: String(hostApiPort),
...(process.platform === 'linux' ? { ELECTRON_DISABLE_SANDBOX: '1' } : {}),
},
timeout: 90_000,
});
const page = await electronApplication.firstWindow();
await page.waitForLoadState('domcontentloaded');
const extension = await evaluateProof(electronApplication, 'extension');
assertPackagedMain(extension);
assertManagedTurns(extension.extension);
if (extension.extension?.subagentStatus !== 'complete'
|| extension.extension?.childToolNames?.join(',') !== 'find,grep,ls,read'
|| !extension.extension?.parentToolNames?.includes('subagent')
|| extension.extension?.parentProcessIds?.length < 2
|| extension.extension?.childProcessIds?.length !== 1
|| extension.extension?.providerRequests?.child !== 1
|| extension.extension?.managedTurns?.length !== 2) {
throw new Error(`Final ASAR extension/subagent proof failed: ${JSON.stringify(extension.extension)}`);
}
const pressureStart = await evaluateProof(electronApplication, 'pressure.start');
pressureActive = true;
assertPackagedMain(pressureStart);
assertActivePressure(pressureStart.pressure);
const uiStartedAt = performance.now();
await page.getByTestId('ai-module-option-programming').click();
await page.getByTestId('main-layout').waitFor({ state: 'visible', timeout: 10_000 });
const uiInteractiveMs = Math.round(performance.now() - uiStartedAt);
const pressureFinish = await evaluateProof(electronApplication, 'pressure.finish');
pressureActive = false;
assertPackagedMain(pressureFinish);
assertReleasedPressure(pressureFinish.pressure);
let failureCleanup;
if (options.verifyCleanupFailure === true) {
const failureStart = await evaluateProof(electronApplication, 'pressure.start');
pressureActive = true;
assertPackagedMain(failureStart);
assertActivePressure(failureStart.pressure);
let injectedFailure;
try {
await evaluateProof(electronApplication, 'pressure.finish.inject-failure');
} catch (error) {
injectedFailure = error;
}
if (!injectedFailure
|| !String(injectedFailure).includes('PI release pressure cleanup failed: parents.settle')) {
throw new Error(`PI pressure cleanup failure injection did not fail as expected: ${String(injectedFailure)}`);
}
const retryFinish = await evaluateProof(electronApplication, 'pressure.finish');
pressureActive = false;
assertPackagedMain(retryFinish);
assertReleasedPressure(retryFinish.pressure);
failureCleanup = {
injectedAt: 'parents.settle',
firstFinish: 'failed-as-injected',
retry: 'pass',
released: retryFinish.pressure,
};
}
const report = {
schemaVersion: 1,
generatedAt: new Date().toISOString(),
executable: options.electronExecutable,
runtimeRoot: options.runtimeRoot ?? null,
packagedMain: extension.packagedMain,
extension: extension.extension,
pressure: {
processKind: 'final-product-executable-with-ELECTRON_RUN_AS_NODE',
active: pressureStart.pressure,
ui: {
action: 'select Makelore Code and render main layout',
interactive: true,
durationMs: uiInteractiveMs,
},
released: pressureFinish.pressure,
...(failureCleanup ? { failureCleanup } : {}),
},
result: 'pass',
};
if (options.reportPath) {
await mkdir(dirname(options.reportPath), { recursive: true });
await writeFile(options.reportPath, `${JSON.stringify(report, null, 2)}\n`);
}
return report;
} finally {
if (electronApplication && pressureActive) {
await evaluateProof(electronApplication, 'pressure.finish').catch(() => undefined);
}
if (electronApplication) await closeApplication(electronApplication);
await rm(scratchRoot, { recursive: true, force: true, maxRetries: 5, retryDelay: 200 });
}
}
async function main() {
const options = parseArgs(process.argv.slice(2));
const report = await runPackagedProductProof(options);
process.stdout.write(`${JSON.stringify(report, null, 2)}\n`);
}
const isMain = process.argv[1]
&& pathToFileURL(resolve(process.argv[1])).href === import.meta.url;
if (isMain) {
main().catch((error) => {
process.stderr.write(`${error.stack ?? error.message}\n`);
process.exitCode = 1;
});
}