mirror of
https://github.com/garrytan/gbrain.git
synced 2026-07-30 11:22:34 +00:00
948ccc7b4f
Two backlog fixes: 1. Takeover of #1476 (closes #1474): 'bench' was missing from CLI_ONLY and runBenchPublish was imported nowhere, so the documented 'gbrain bench publish' hit 'Unknown command'. Adds the 'bench' token to master's current CLI_ONLY set (the original PR rewrote the line from a stale v0.41.14 snapshot, deleting ~15 newer commands), routes bench through a no-DB bypass (bench publish is pure file I/O), and adds 'bench' to CLI_ONLY_SELF_HELP so --help reaches bench-publish's own usage text. 2. Fixes #1475: 'gbrain config set eval.capture true' persisted to the DB plane but was never read — the capture gate reads the sync file-plane ctx.config. loadConfigWithEngine now sparse-merges eval.capture / eval.scrub_pii (file wins per key), connectEngine stashes the merged values on GBRAIN_EVAL_CAPTURE / GBRAIN_EVAL_SCRUB_PII (same pattern as the multimodal flags), and the gates consult the stash when the file plane is silent. Co-authored-by: Mr-B-1 <Mr-B-1@users.noreply.github.com> Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
68 lines
2.6 KiB
TypeScript
68 lines
2.6 KiB
TypeScript
// #1474: the v0.41.1 wave shipped bench-publish.ts + docs/eval-bench.md
|
|
// advertising `gbrain bench publish`, but the cli.ts dispatcher case was never
|
|
// added — the documented command hit 'Unknown command'. These tests spawn the
|
|
// real CLI (no DB needed; bench publish is pure file I/O) and fail on any
|
|
// regression of the dispatcher wiring.
|
|
import { describe, expect, test } from 'bun:test';
|
|
import { mkdtempSync, writeFileSync, existsSync, rmSync } from 'node:fs';
|
|
import { tmpdir } from 'node:os';
|
|
import { join } from 'node:path';
|
|
import { spawnSync } from 'node:child_process';
|
|
|
|
function runCli(args: string[]): { stdout: string; stderr: string; code: number } {
|
|
const result = spawnSync(process.execPath, ['run', 'src/cli.ts', 'bench', ...args], {
|
|
encoding: 'utf8',
|
|
cwd: process.cwd(),
|
|
env: { ...process.env },
|
|
});
|
|
return { stdout: result.stdout ?? '', stderr: result.stderr ?? '', code: result.status ?? -1 };
|
|
}
|
|
|
|
describe('gbrain bench dispatcher (#1474)', () => {
|
|
test('bench --help reaches bench-publish help without a DB (was: Unknown command)', () => {
|
|
const { stdout, stderr, code } = runCli(['--help']);
|
|
expect(stderr).not.toContain('Unknown command');
|
|
expect(code).toBe(0);
|
|
expect(stdout).toContain('gbrain bench publish');
|
|
expect(stdout).toContain('--from');
|
|
});
|
|
|
|
test('unknown bench subcommand exits 2 with usage', () => {
|
|
const { stderr, code } = runCli(['bogus']);
|
|
expect(code).toBe(2);
|
|
expect(stderr).toContain('Unknown bench subcommand');
|
|
expect(stderr).toContain('bench publish');
|
|
});
|
|
|
|
test('bench publish roundtrip: captured NDJSON in, baseline file out', () => {
|
|
const tmp = mkdtempSync(join(tmpdir(), 'bench-cli-'));
|
|
try {
|
|
const row = {
|
|
tool_name: 'query',
|
|
query: 'hello world',
|
|
retrieved_slugs: ['slug-a'],
|
|
retrieved_chunk_ids: [1],
|
|
source_ids: ['default'],
|
|
expand_enabled: false,
|
|
detail: 'medium',
|
|
detail_resolved: 'medium',
|
|
vector_enabled: true,
|
|
expansion_applied: false,
|
|
latency_ms: 100,
|
|
remote: false,
|
|
job_id: null,
|
|
subagent_id: null,
|
|
};
|
|
const from = join(tmp, 'captured.ndjson');
|
|
const to = join(tmp, 'personal.baseline.ndjson');
|
|
writeFileSync(from, `${JSON.stringify(row)}\n`);
|
|
const { code, stderr } = runCli(['publish', '--from', from, '--to', to]);
|
|
expect(stderr).not.toContain('Unknown command');
|
|
expect(code).toBe(0);
|
|
expect(existsSync(to)).toBe(true);
|
|
} finally {
|
|
rmSync(tmp, { recursive: true, force: true });
|
|
}
|
|
});
|
|
});
|