297 lines
11 KiB
JavaScript
297 lines
11 KiB
JavaScript
import assert from 'node:assert/strict';
|
||
import test from 'node:test';
|
||
import { mkdtempSync, mkdirSync, writeFileSync, appendFileSync, rmSync } from 'node:fs';
|
||
import { tmpdir } from 'node:os';
|
||
import { join } from 'node:path';
|
||
|
||
import { createSessionScanner } from '../src/session-scan.mjs';
|
||
|
||
/** 造一个会话目录树。返回根目录，用完由调用方删。 */
|
||
function makeRoot() {
|
||
return mkdtempSync(join(tmpdir(), 'pi-scan-'));
|
||
}
|
||
|
||
/** 写一条会话文件。lines 是 header 之后的行（对象，会被 JSON 化）。 */
|
||
function writeSession(root, cwdSlug, fileName, header, lines = []) {
|
||
const dir = join(root, cwdSlug);
|
||
mkdirSync(dir, { recursive: true });
|
||
const file = join(dir, fileName);
|
||
const body = [JSON.stringify({ type: 'session', version: 3, ...header })]
|
||
.concat(lines.map((l) => JSON.stringify(l)))
|
||
.join('\n');
|
||
writeFileSync(file, `${body}\n`);
|
||
return file;
|
||
}
|
||
|
||
const msg = (text) => ({
|
||
type: 'message',
|
||
id: Math.random().toString(36).slice(2, 10),
|
||
message: { role: 'user', content: [{ type: 'text', text }] },
|
||
});
|
||
|
||
const info = (name) => ({ type: 'session_info', id: 'aa', parentId: 'bb', name });
|
||
|
||
test('取到 header 的 id 与 cwd，以及 session_info 的 name', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
writeSession(root, '--tmp-proj--', 'a.jsonl',
|
||
{ id: 'sess-1', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/tmp/proj' },
|
||
[msg('你好'), info('重构导入路径'), msg('继续')]);
|
||
|
||
const { scan } = createSessionScanner({ sessionsDir: root });
|
||
const out = await scan();
|
||
assert.equal(out.length, 1);
|
||
assert.equal(out[0].id, 'sess-1');
|
||
assert.equal(out[0].cwd, '/tmp/proj');
|
||
assert.equal(out[0].name, '重构导入路径');
|
||
assert.ok(out[0].modified instanceof Date);
|
||
assert.ok(out[0].path.endsWith('a.jsonl'));
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('多次改名时取最后一个 session_info（与 SDK 语义一致）', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' },
|
||
[info('旧名'), msg('干活'), info('新名')]);
|
||
const { scan } = createSessionScanner({ sessionsDir: root });
|
||
assert.equal((await scan())[0].name, '新名');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('显式清名（session_info 不带 name）→ name 为 undefined', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' },
|
||
[info('有名字'), { type: 'session_info', id: 'cc', parentId: 'dd' }]);
|
||
const { scan } = createSessionScanner({ sessionsDir: root });
|
||
assert.equal((await scan())[0].name, undefined);
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('老会话的 cwd 是空串时照实返回，不冒充', async () => {
|
||
// snapshotPiSessions 会按空 workspace 上报。拿桥自己的 cwd 顶上去会让
|
||
// 那条会话在错误的工作目录下出现在补全候选里。
|
||
const root = makeRoot();
|
||
try {
|
||
writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z' },
|
||
[info('无 cwd 的老会话')]);
|
||
const { scan } = createSessionScanner({ sessionsDir: root });
|
||
assert.equal((await scan())[0].cwd, '');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('第二次扫描不重读未变化的文件（这就是省下来的那 282MB）', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's1', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' },
|
||
[info('甲'), msg('内容')]);
|
||
writeSession(root, '--y--', 'b.jsonl',
|
||
{ id: 's2', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/y' },
|
||
[info('乙')]);
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
await sc.scan();
|
||
const after1 = sc.stats();
|
||
assert.equal(after1.fullScans, 2);
|
||
|
||
await sc.scan();
|
||
const after2 = sc.stats();
|
||
assert.equal(after2.fullScans, 2, '没变化的文件不该被重扫');
|
||
assert.equal(after2.tailScans, 0);
|
||
assert.equal(after2.tailBytes, after1.tailBytes, '一个字节都不该多读');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('文件追加后只扫尾部，且能看到新名字', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
const file = writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' },
|
||
[info('第一版')]);
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
assert.equal((await sc.scan())[0].name, '第一版');
|
||
const bytes1 = sc.stats().tailBytes;
|
||
|
||
appendFileSync(file, `${JSON.stringify(msg('新一轮'))}\n${JSON.stringify(info('第二版'))}\n`);
|
||
const out = await sc.scan();
|
||
assert.equal(out[0].name, '第二版');
|
||
assert.equal(sc.stats().tailScans, 1);
|
||
const delta = sc.stats().tailBytes - bytes1;
|
||
assert.ok(delta > 0 && delta < 400, `只该读新增的那一小段，实际 ${delta} 字节`);
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('★尾部没有 session_info 时保留旧 name，不能清空', async () => {
|
||
// 写错成 `cached.name = name` 会让每次有新消息的会话都丢掉名字，
|
||
// 而没有 name 的会话不上报（S-1）—— 于是**活跃**会话反而从补全里消失。
|
||
const root = makeRoot();
|
||
try {
|
||
const file = writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' },
|
||
[info('要保住的名字')]);
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
assert.equal((await sc.scan())[0].name, '要保住的名字');
|
||
|
||
appendFileSync(file, `${JSON.stringify(msg('只是普通消息'))}\n`);
|
||
assert.equal((await sc.scan())[0].name, '要保住的名字');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('★巨大的 message 行不进内存也不影响解析', async () => {
|
||
// 本机实测单行最长 2.63MB。listAll 会把它整行读进来并 parse；
|
||
// 这里只要求 name 仍能取到，且扫描不抛错。
|
||
const root = makeRoot();
|
||
try {
|
||
const huge = 'x'.repeat(3 * 1024 * 1024);
|
||
writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' },
|
||
[msg(huge), info('巨行之后的名字'), msg(huge)]);
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
const out = await sc.scan();
|
||
assert.equal(out.length, 1);
|
||
assert.equal(out[0].name, '巨行之后的名字');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('首行不是 session 的文件被忽略，且不会每拍重读', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
const dir = join(root, '--x--');
|
||
mkdirSync(dir, { recursive: true });
|
||
writeFileSync(join(dir, 'junk.jsonl'), `${JSON.stringify({ type: 'message' })}\n`);
|
||
writeFileSync(join(dir, 'empty.jsonl'), '');
|
||
writeFileSync(join(dir, 'broken.jsonl'), '这不是 json\n');
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
assert.deepEqual(await sc.scan(), []);
|
||
const n = sc.stats().fullScans;
|
||
await sc.scan();
|
||
assert.equal(sc.stats().fullScans, n, '空壳条目不该被反复重读');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('★会话文件被删后缓存条目跟着走（缓存自身不是下一个泄露源）', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's1', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' }, [info('甲')]);
|
||
writeSession(root, '--x--', 'b.jsonl',
|
||
{ id: 's2', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' }, [info('乙')]);
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
await sc.scan();
|
||
assert.equal(sc.stats().tracked, 2);
|
||
|
||
rmSync(join(root, '--x--', 'a.jsonl'));
|
||
const out = await sc.scan();
|
||
assert.equal(out.length, 1);
|
||
assert.equal(sc.stats().tracked, 1, '磁盘上没了的文件必须从缓存里消失');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('文件变小（被截断/重写）时整份重扫而不是从越界偏移读', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
const file = writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' },
|
||
[msg('很多内容'.repeat(200)), info('旧')]);
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
await sc.scan();
|
||
const fulls = sc.stats().fullScans;
|
||
|
||
// 重写成更短的内容，且换了名字
|
||
writeSession(root, '--x--', 'a.jsonl',
|
||
{ id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' }, [info('新')]);
|
||
|
||
const out = await sc.scan();
|
||
assert.equal(out[0].name, '新');
|
||
assert.equal(sc.stats().fullScans, fulls + 1, '变小必须触发整份重扫');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('目录不存在 → 空列表（pi 从没跑过是正常状态）', async () => {
|
||
const sc = createSessionScanner({ sessionsDir: join(tmpdir(), `nope-${Date.now()}`) });
|
||
assert.deepEqual(await sc.scan(), []);
|
||
});
|
||
|
||
test('★读目录遇到非 ENOENT 错误必须抛出，不能返回空数组', async () => {
|
||
// 返回空数组的语义是「平台确实没有会话」，会把服务端镜像抹掉（W-3 / N-7）。
|
||
// 一次 EACCES 就能清空别人的补全候选 —— 必须让调用方看到失败并省略字段。
|
||
const root = makeRoot();
|
||
try {
|
||
// 用一个普通文件当 sessionsDir：readdir 会给 ENOTDIR，而不是 ENOENT
|
||
const notADir = join(root, 'file');
|
||
writeFileSync(notADir, 'x');
|
||
const sc = createSessionScanner({ sessionsDir: notADir });
|
||
await assert.rejects(() => sc.scan(), (e) => e?.code === 'ENOTDIR');
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('缺 sessionsDir 时当场抛错', () => {
|
||
assert.throws(() => createSessionScanner(), /sessionsDir/);
|
||
assert.throws(() => createSessionScanner({}), /sessionsDir/);
|
||
});
|
||
|
||
test('跨多个 cwd 子目录汇总', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
writeSession(root, '--a--', '1.jsonl', { id: 'a1', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/a' }, [info('甲')]);
|
||
writeSession(root, '--b--', '2.jsonl', { id: 'b1', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/b' }, [info('乙')]);
|
||
writeSession(root, '--b--', '3.jsonl', { id: 'b2', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/b' }, [info('丙')]);
|
||
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
const out = await sc.scan();
|
||
assert.equal(out.length, 3);
|
||
assert.deepEqual([...out.map((s) => s.id)].sort(), ['a1', 'b1', 'b2']);
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|
||
|
||
test('非 .jsonl 文件被跳过', async () => {
|
||
const root = makeRoot();
|
||
try {
|
||
const dir = join(root, '--x--');
|
||
mkdirSync(dir, { recursive: true });
|
||
writeFileSync(join(dir, 'notes.txt'), 'hello');
|
||
writeSession(root, '--x--', 'a.jsonl', { id: 's', timestamp: '2026-09-01T00:00:00.000Z', cwd: '/x' }, [info('甲')]);
|
||
const sc = createSessionScanner({ sessionsDir: root });
|
||
assert.equal((await sc.scan()).length, 1);
|
||
} finally {
|
||
rmSync(root, { recursive: true, force: true });
|
||
}
|
||
});
|