Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
/**
* #4318 — promptYesNo close-race regression.
*
* The old inline twins (sync-cost-gate.ts / reindex-code.ts) called
* rl.close() before resolving, and their unguarded 'close' listener
* resolved(false) synchronously during that close — a typed "y" was read as
* a decline (verified reproducible under bun). Pins the shared helper:
* answer resolves first; 'close' only declines on a true EOF.
*/
import { describe, test, expect } from 'bun:test';
import { PassThrough } from 'stream';
import { promptYesNo } from '../src/core/confirm-prompt.ts';
function mkStreams() {
return { input: new PassThrough(), output: new PassThrough() };
}
describe('promptYesNo (#4318)', () => {
test('"y" answer resolves TRUE (the close-race regression)', async () => {
const { input, output } = mkStreams();
const p = promptYesNo('Proceed? [y/N] ', { input, output });
input.write('y\n');
expect(await p).toBe(true);
});
test('"yes" (case/space-insensitive) resolves true', async () => {
const { input, output } = mkStreams();
const p = promptYesNo('Proceed? [y/N] ', { input, output });
input.write(' YES \n');
expect(await p).toBe(true);
});
test('"n" resolves false', async () => {
const { input, output } = mkStreams();
const p = promptYesNo('Proceed? [y/N] ', { input, output });
input.write('n\n');
expect(await p).toBe(false);
});
test('empty answer resolves false (the [y/N] default)', async () => {
const { input, output } = mkStreams();
const p = promptYesNo('Proceed? [y/N] ', { input, output });
input.write('\n');
expect(await p).toBe(false);
});
test('EOF without an answer (Ctrl-D) resolves false', async () => {
const { input, output } = mkStreams();
const p = promptYesNo('Proceed? [y/N] ', { input, output });
input.end(); // no line ever arrives
expect(await p).toBe(false);
});
test('answer followed by immediate EOF still resolves true', async () => {
const { input, output } = mkStreams();
const p = promptYesNo('Proceed? [y/N] ', { input, output });
input.end('y\n');
expect(await p).toBe(true);
});
});
/**
* pglite-repair.ts had its own inline promptYesNo (not this shared helper)
* with the same close-before-resolve race: `rl.close()` inside the
* rl.question callback fired the unguarded `rl.on('close', ...)` listener
* synchronously, so `resolve(false)` won before the typed "y" resolve ran —
* `gbrain pglite-repair` always treated an affirmative interactive answer as
* decline. Pins the exact call now wired at that call site (question text +
* `{ output: process.stderr }`, matching `src/commands/pglite-repair.ts`) so
* a regression to the old inline shape — or a stdout-leaking `output`
* default — is caught here rather than only in the generic cases above.
*/
describe('promptYesNo — pglite-repair call-site wiring (#4318 residual)', () => {
test('"Repair now?" [y/N] question resolves true on "y", written to stderr not stdout', async () => {
const input = new PassThrough();
const output = new PassThrough();
const written: string[] = [];
output.on('data', (chunk) => written.push(chunk.toString()));
const p = promptYesNo('Repair now? [y/N] ', { output, input });
input.write('y\n');
expect(await p).toBe(true);
// The prompt text itself must have gone to the injected `output` stream
// (which stands in for process.stderr at the real call site), not been
// silently dropped or duplicated with a second "[y/N]" suffix.
expect(written.join('')).toBe('Repair now? [y/N] ');
});
test('"Repair now?" [y/N] question resolves false on decline', async () => {
const { input, output } = mkStreams();
const p = promptYesNo('Repair now? [y/N] ', { output, input });
input.write('n\n');
expect(await p).toBe(false);
});
});