宜博 / lemonboss

/**
 * Unit Tests — R-19 产物索引（27-P4 §4.6）
 *
 * 锁的是第四次真 CLI 冒烟那个断点的正解。当时 org-planner-prd 的归因原话是
 * 「不是上游没产出，是它的产出**结构上到不了我手上**」—— 边只承载控制流
 * （condition / edge_kind / fail_reason），上游 brd.md 躺在节点私有目录，
 * 下游提示词里没有任何可读路径。
 *
 * 每条用例都按「产生 → 传递 → 落库 → 读回」四环走真入口：
 *   ①  版本链         同路径二次产出标 superseded_by，不删行（落库）
 *   ②  部分唯一索引    「同一键只有一个活的 + 留历史」这个选择本身
 *   ③a 沿边传递        出边声明 kind → 引擎登记实例 → 下游提示词里是绝对路径（全四环）
 *   ③b 任务级共享      acceptance 对**非相邻**节点也可见（R-9.3 planner→(coder)→validator）
 *   ④  缺输入告警      声明了却查不到实例时提示词里出现 ⚠️，不静默省略（R-19.4 读回端）
 *   ⑤  产出侧过滤      attempt-*.json 不登记（引擎自读的归因文件）；节点 fail 不登记
 *   ⑥  索引重建        删掉 graph_files 后能从 graph_run_nodes 重建
 *
 * describeInputs / indexNodeOutputs / companyRootDir **故意不导出**（graph-engine.js
 * module.exports 那段注释写明了口径：不为测试新开公开面），所以这里全部通过 advanceOne
 * 这个真入口验 —— 输入侧读 executor 收到的 ctx.task，产出侧读 graph_files 落库行。
 *
 * 隔离手法同 graph-boundary.test.js：先设 GRAPH_DB_PATH / GRAPH_RUNS_DIR 再 require，
 * 仓库 data/ 全程不被碰；executor 全用注入的假货，不起 Claude CLI。
 * （注入 executor 还顺带关掉了边界守卫 —— boundaryGuardApplicable 只在
 *   agentExecutor === defaultAgentExecutor 时成立，所以这些用例不扫宿主仓库。）
 */
import { describe, it, expect, beforeAll, afterAll } from 'vitest';
import fs from 'fs';
import os from 'os';
import path from 'path';

describe('R-19 产物索引：graph_files / graph_edge_contracts', () => {
  let wdb;
  let engine;
  let runsDir;
  const TEST_DB_PATH = '/tmp/test-graph-files-index.sqlite';

  beforeAll(() => {
    for (const f of [TEST_DB_PATH, `${TEST_DB_PATH}-shm`, `${TEST_DB_PATH}-wal`]) {
      try { fs.unlinkSync(f); } catch {}
    }
    process.env.GRAPH_DB_PATH = TEST_DB_PATH;
    runsDir = fs.mkdtempSync(path.join(os.tmpdir(), 'graph-files-runs-'));
    process.env.GRAPH_RUNS_DIR = runsDir;
    for (const m of [
      '../../services/graph-db',
      '../../services/graph-service',
      '../../services/org-graph-service',
      '../../services/work-graph-db',
      '../../services/graph-run-sse',
      '../../services/graph-engine',
    ]) {
      delete require.cache[require.resolve(m)];
    }
    wdb = require('../../services/work-graph-db');
    engine = require('../../services/graph-engine');
  });

  afterAll(() => {
    for (const f of [TEST_DB_PATH, `${TEST_DB_PATH}-shm`, `${TEST_DB_PATH}-wal`]) {
      try { fs.unlinkSync(f); } catch {}
    }
    delete process.env.GRAPH_DB_PATH;
    try { fs.rmSync(runsDir, { recursive: true, force: true }); } catch {}
    delete process.env.GRAPH_RUNS_DIR;
    for (const m of ['../../services/graph-run-sse', '../../services/graph-engine']) {
      delete require.cache[require.resolve(m)];
    }
  });

  // ── 测试脚手架 ────────────────────────────────────────────────

  function ensureAgents(ids) {
    const db = wdb.getDb();
    const ts = new Date().toISOString();
    const ins = db.prepare(`
      INSERT OR IGNORE INTO system_agents
        (id, name, scope, owner_company_id, source_kind, legacy_path, status, created_at, updated_at)
      VALUES (?, ?, 'public', NULL, 'global_agent', ?, 'active', ?, ?)
    `);
    for (const id of new Set(ids)) ins.run(id, id, `system/agents/${id}`, ts, ts);
  }

  /** 建图 + 建 run + 置 ready + start，并给出「按节点名拿 defNodeId / 边 id」的取值器 */
  function buildRun({ companyId, taskId, nodes, edges = [] }) {
    ensureAgents(nodes.map((n) => n.systemAgentId || n.nodeName));
    const g = wdb.createWorkGraph({ companyId, taskId, nodes, edges });
    expect(g.errors || []).toEqual([]);
    const def = wdb.readDefGraph('work', g.id);
    const byName = new Map(def.nodes.map((n) => [n.name, n.defNodeId]));
    const run = wdb.createRun({ graphKind: 'work', graphId: g.id, companyId, taskId });
    engine.markRunReady(run.id);
    engine.startRun(run.id, { mode: 'manual' });
    return {
      def,
      runId: run.id,
      defNodeId: (name) => byName.get(name),
      edgeId: (from, to) =>
        def.edges.find((e) => e.from === byName.get(from) && e.to === byName.get(to)).id,
    };
  }

  /** 节点私有产物目录 —— 与 graph-engine.nodeArtifactDir 同源（GRAPH_RUNS_DIR 覆盖分支） */
  const nodeDir = (rootRunId, nodeRunId) => path.join(runsDir, rootRunId, 'nodes', nodeRunId);
  /** 公司根 —— 与 graph-engine.companyRootDir 同源（同一个 override 分支） */
  const companyRoot = (companyId) => path.join(runsDir, '_companies', companyId);

  /**
   * 跑 n 步，收集每个节点看到的提示词。writes[nodeName] 给出这个节点要落盘的文件，
   * 落在它自己的私有目录里、outputRef 指向该目录 —— 与 defaultAgentExecutor 的形状一致。
   */
  async function advance(runId, steps, { writes = {}, status = () => 'success' } = {}) {
    const tasks = new Map();
    engine.setAgentExecutor(async (ctx) => {
      tasks.set(ctx.nodeName, ctx.task);
      const files = writes[ctx.nodeName];
      if (!files) return { status: status(ctx.nodeName) };
      const dir = nodeDir(ctx.run.root_run_id, ctx.nodeRunId);
      fs.mkdirSync(dir, { recursive: true });
      for (const [name, body] of Object.entries(files)) {
        fs.writeFileSync(path.join(dir, name), body);
      }
      return { status: status(ctx.nodeName), outputRef: dir };
    });
    try {
      for (let i = 0; i < steps; i++) await engine.advanceOne(runId);
    } finally {
      engine.setAgentExecutor(null);
    }
    return tasks;
  }

  const allFiles = (companyId) =>
    wdb.getDb()
      .prepare(`SELECT * FROM graph_files WHERE company_id = ? ORDER BY created_at, id`)
      .all(companyId);

  // ── ① 版本链 ─────────────────────────────────────────────────

  it('① 同一 (company, task, relPath) 二次产出：旧行标 superseded_by 而不是被删，findFiles 只给最新那份', () => {
    const companyId = 'co-files-ver';
    const taskId = 'task-ver';
    const rel = 'tasks/task-ver/brd.md';

    const v1 = wdb.recordFile({ companyId, taskId, relPath: rel, kind: 'brd' });
    const v2 = wdb.recordFile({ companyId, taskId, relPath: rel, kind: 'brd' });
    expect(v2).not.toBe(v1);

    const rows = allFiles(companyId);
    // 历史不被删：两行都在
    expect(rows.map((r) => r.id).sort()).toEqual([v1, v2].sort());
    // 旧行指向新行
    expect(rows.find((r) => r.id === v1).superseded_by).toBe(v2);
    // 当前版本恰好一行
    expect(rows.filter((r) => r.superseded_by === null).map((r) => r.id)).toEqual([v2]);

    // 读回：findFiles 只返回最新那份
    const found = wdb.findFiles({ companyId, taskId, kinds: ['brd'] });
    expect(found.map((f) => f.id)).toEqual([v2]);
  });

  // ── ② 部分唯一索引 ───────────────────────────────────────────

  it('② 部分唯一索引：两行同路径且都 superseded_by IS NULL 被拒；历史行同路径可共存', () => {
    const db = wdb.getDb();
    const companyId = 'co-files-idx';
    const rel = 'tasks/task-idx/prd.md';
    const ins = db.prepare(`
      INSERT INTO graph_files (id, company_id, task_id, rel_path, kind, created_at, superseded_by)
      VALUES (?, ?, 'task-idx', ?, 'prd', ?, ?)
    `);
    const now = new Date().toISOString();

    ins.run('gf-live-1', companyId, rel, now, null);
    // 第二个「当前版本」必须被索引挡住 —— 否则 findFiles 会同时返回两份，
    // 下游拿不准该读哪个（这条锁的是「用部分唯一索引」这个选择本身）
    expect(() => ins.run('gf-live-2', companyId, rel, now, null)).toThrow(/UNIQUE/i);
    // 而历史行（superseded_by 非空）同路径可以随便共存 —— 索引是**部分**的，不是全量的
    expect(() => ins.run('gf-old-1', companyId, rel, now, 'gf-live-1')).not.toThrow();
    expect(() => ins.run('gf-old-2', companyId, rel, now, 'gf-live-1')).not.toThrow();
    expect(allFiles(companyId).filter((r) => r.superseded_by === null)).toHaveLength(1);
  });

  // ── ③a 沿边传递（全四环）+ ⑤ attempt-*.json 不登记 ──────────

  it('③a 沿边传递：出边声明 brd → 引擎登记实例 → 下游提示词里是绝对路径；attempt-*.json 不登记', async () => {
    const companyId = 'co-files-edge';
    const taskId = 'task-edge';
    const b = buildRun({
      companyId, taskId,
      nodes: [{ nodeName: 'brd-writer' }, { nodeName: 'prd-writer' }],
      edges: [{ from: 'brd-writer', to: 'prd-writer', condition: 'success' }],
    });
    // 产生：规划时就在边上声明「这条边传一份 brd」
    wdb.declareEdgeContract({
      edgeLayer: 'work', edgeId: b.edgeId('brd-writer', 'prd-writer'), kind: 'brd',
    });

    const tasks = await advance(b.runId, 2, {
      writes: {
        'brd-writer': {
          'brd.md': '# BRD',
          // 引擎自读的归因文件，不是给下游的产物
          'attempt-1.json': JSON.stringify({ fail_reason: null }),
        },
      },
    });

    // 落库：只登记 brd.md，kind 取出边声明的 brd（不是兜底的 artifact）
    const rows = allFiles(companyId);
    expect(rows).toHaveLength(1);
    expect(rows[0].kind).toBe('brd');
    expect(path.basename(rows[0].rel_path)).toBe('brd.md');
    // attempt-*.json 一行都不该有
    expect(rows.filter((r) => /attempt-/.test(r.rel_path))).toEqual([]);

    // 传递：入边契约查得到
    const inbound = wdb.inboundContracts('work', b.def.edges, b.defNodeId('prd-writer'));
    expect(inbound.map((c) => c.kind)).toEqual(['brd']);

    // 读回：下游提示词里是能直接 open 的绝对路径
    const abs = path.join(companyRoot(companyId), rows[0].rel_path);
    expect(fs.existsSync(abs)).toBe(true);
    const prompt = tasks.get('prd-writer');
    expect(prompt).toContain('[上游输入');
    expect(prompt).toContain(`- brd：${abs}`);
    // 查得到实例就不该出现缺输入告警（段尾那句「标 ⚠️ 的表示…」是固定说明，不算告警）
    expect(prompt).not.toContain('⚠️ 上游声明会产出');
    // 上游自己那一步没有入边，不该凭空多一段
    expect(tasks.get('brd-writer')).not.toContain('[上游输入');
  });

  // ── ③b 任务级共享：非相邻可见 ────────────────────────────────

  it('③b 任务级共享：planner 产的 acceptance，隔着 coder 的 validator 也查得到（R-9.3 的形状）', async () => {
    const companyId = 'co-files-shared';
    const taskId = 'task-shared';
    const b = buildRun({
      companyId, taskId,
      nodes: [{ nodeName: 'planner' }, { nodeName: 'coder' }, { nodeName: 'validator' }],
      edges: [
        { from: 'planner', to: 'coder', condition: 'success' },
        { from: 'coder', to: 'validator', condition: 'success' },
      ],
    });
    // 刻意**不**声明任何边契约：planner→validator 根本没有边，纯边绑定表达不了这条依赖

    const tasks = await advance(b.runId, 3, {
      writes: { planner: { 'acceptance.json': JSON.stringify({ checks: ['a'] }) } },
    });

    const rows = allFiles(companyId);
    expect(rows).toHaveLength(1);
    expect(rows[0].kind).toBe('acceptance'); // 无声明时按文件名定 kind
    const abs = path.join(companyRoot(companyId), rows[0].rel_path);

    // 非相邻的 validator 拿到了路径 —— 这才是「两类数据流都要」的理由
    const prompt = tasks.get('validator');
    expect(prompt).toContain('[上游输入');
    expect(prompt).toContain(`- acceptance（任务级共享）：${abs}`);
    // 相邻的 coder 同样看得到（任务级共享不受相邻性限制）
    expect(tasks.get('coder')).toContain(abs);
    // planner 跑的时候索引里还没有实例，不该有段落
    expect(tasks.get('planner')).not.toContain('[上游输入');
  });

  // ── ④ 缺输入告警（R-19.4 读回端）────────────────────────────

  it('④ 声明了 required 输入却查不到实例：提示词里出现 ⚠️，不静默省略', async () => {
    const companyId = 'co-files-missing';
    const taskId = 'task-missing';
    const b = buildRun({
      companyId, taskId,
      nodes: [{ nodeName: 'sdd-writer' }, { nodeName: 'impl' }],
      edges: [{ from: 'sdd-writer', to: 'impl', condition: 'success' }],
    });
    wdb.declareEdgeContract({
      edgeLayer: 'work', edgeId: b.edgeId('sdd-writer', 'impl'), kind: 'sdd', required: true,
    });

    // 上游什么都不写 → 声明在、实例不在
    const tasks = await advance(b.runId, 2);
    expect(allFiles(companyId)).toEqual([]);

    const prompt = tasks.get('impl');
    expect(prompt).toContain('[上游输入');
    expect(prompt).toContain('- sdd：⚠️');
    expect(prompt).toContain('索引里查不到实例');
  });

  // ── ⑤ 节点 fail 不登记 ──────────────────────────────────────

  it('⑤ 节点 fail 时落盘的半成品不进索引（下游不该把它当输入读）', async () => {
    const companyId = 'co-files-fail';
    const taskId = 'task-fail';
    const b = buildRun({ companyId, taskId, nodes: [{ nodeName: 'flaky' }] });

    await advance(b.runId, 1, {
      writes: { flaky: { 'brd.md': '# 半成品' } },
      status: () => 'fail',
    });

    // 文件真的落盘了（不是没写），但索引里一行都没有
    const state = wdb.getRunState(b.runId);
    // 图上没有任何 fail 边，失败节点最终落 blocked（fail → blocked 是 FSM 的终态转换）
    expect(state.nodes[0].status).toBe('blocked');
    expect(fs.existsSync(path.join(state.nodes[0].outputRef, 'brd.md'))).toBe(true);
    expect(allFiles(companyId)).toEqual([]);
  });

  // ── ⑥ 索引重建 ─────────────────────────────────────────────

  it('⑥ graph_files 被清掉后能从 graph_run_nodes 重建（有损：kind 退成 artifact）', async () => {
    const companyId = 'co-files-rebuild';
    const taskId = 'task-rebuild';
    const b = buildRun({ companyId, taskId, nodes: [{ nodeName: 'brd-writer' }] });

    await advance(b.runId, 1, { writes: { 'brd-writer': { 'brd.md': '# BRD' } } });
    expect(allFiles(companyId)).toHaveLength(1);

    // 索引坏了 / 被清空
    wdb.getDb().prepare(`DELETE FROM graph_files WHERE company_id = ?`).run(companyId);
    expect(wdb.findFiles({ companyId, taskId })).toEqual([]);

    // 节点行还在 → 能重建出「这个节点产出过哪个目录」
    expect(wdb.rebuildFilesFromNodes(companyId)).toBe(1);
    const rebuilt = wdb.findFiles({ companyId, taskId });
    expect(rebuilt).toHaveLength(1);
    expect(rebuilt[0].kind).toBe('artifact'); // 重建恢复不了语义，只恢复位置
    expect(rebuilt[0].rel_path).toBe(wdb.getRunState(b.runId).nodes[0].outputRef);
  });

  // ── ⑦ 文件名认不出 kind 时靠声明认领 ────────────────────────
  //
  // 第五次真 CLI 冒烟里这一环成立**纯属侥幸**：brd 节点恰好把文件命名成 `brd.md`。
  // agent 给产物起中文名是完全正常的行为，不该由文件名决定索引对不对。

  it('⑦ 文件名认不出 kind 时按出边声明认领（一对一无歧义），下游照样拿到路径', async () => {
    const companyId = 'co-files-claim';
    const taskId = 'task-claim';
    const b = buildRun({
      companyId, taskId,
      nodes: [{ nodeName: 'brd-writer' }, { nodeName: 'prd-writer' }],
      edges: [{ from: 'brd-writer', to: 'prd-writer', condition: 'success' }],
    });
    wdb.declareEdgeContract({
      edgeLayer: 'work', edgeId: b.edgeId('brd-writer', 'prd-writer'), kind: 'brd',
    });

    const tasks = await advance(b.runId, 2, {
      writes: {
        'brd-writer': {
          // 文件名里一个 "brd" 字都没有 —— 老逻辑会落成 kind=artifact
          '业务需求文档.md': '# 业务需求',
          // 同目录还有一个能按文件名认出来的（真跑里就是这个形状），
          // 它必须先被 contract 认领掉，否则「还剩一个认不出的文件」不成立
          'contract.json': JSON.stringify({ outputs: [] }),
          'attempt-1.json': JSON.stringify({ fail_reason: null }),
        },
      },
    });

    const rows = allFiles(companyId);
    const byKind = new Map(rows.map((r) => [r.kind, r]));
    expect([...byKind.keys()].sort()).toEqual(['brd', 'contract']);
    expect(path.basename(byKind.get('brd').rel_path)).toBe('业务需求文档.md');

    // 真正要防的是下游拿不到输入 —— 落成 artifact 时这一行会变成 ⚠️
    const prompt = tasks.get('prd-writer');
    expect(prompt).toContain(`- brd：${path.join(companyRoot(companyId), byKind.get('brd').rel_path)}`);
    expect(prompt).not.toContain('⚠️ 上游声明会产出');
  });

  it('⑦ 有歧义时不猜：两个认不出的文件配一个声明 → 都落 artifact', async () => {
    const companyId = 'co-files-ambiguous';
    const taskId = 'task-ambiguous';
    const b = buildRun({
      companyId, taskId,
      nodes: [{ nodeName: 'brd-writer' }, { nodeName: 'prd-writer' }],
      edges: [{ from: 'brd-writer', to: 'prd-writer', condition: 'success' }],
    });
    wdb.declareEdgeContract({
      edgeLayer: 'work', edgeId: b.edgeId('brd-writer', 'prd-writer'), kind: 'brd',
    });

    const tasks = await advance(b.runId, 2, {
      writes: { 'brd-writer': { '文档一.md': 'a', '文档二.md': 'b' } },
    });

    // 猜错比留 artifact 更坏：错的 kind 会让下游读到一份不是它要的文件，
    // 而 artifact 只是查不到 —— 后者会如实报 ⚠️，前者是静默的错。
    const rows = allFiles(companyId);
    expect(rows.map((r) => r.kind)).toEqual(['artifact', 'artifact']);
    expect(tasks.get('prd-writer')).toContain('- brd：⚠️');
  });

  // ── ⑧ 任务级共享按 kind 去重 ────────────────────────────────

  it('⑧ 两个节点各写一份同 kind 的任务级产物：提示词里只列最新那一份，不逐个累加', async () => {
    const companyId = 'co-files-dedupe';
    const taskId = 'task-dedupe';
    const b = buildRun({
      companyId, taskId,
      nodes: [{ nodeName: 'first' }, { nodeName: 'second' }, { nodeName: 'third' }],
      edges: [
        { from: 'first', to: 'second', condition: 'success' },
        { from: 'second', to: 'third', condition: 'success' },
      ],
    });

    // 真跑里的形状：brd 与 prd 两个节点各写了一份 contract.json，都是 kind=contract
    const tasks = await advance(b.runId, 3, {
      writes: {
        first: { 'contract.json': JSON.stringify({ v: 1 }) },
        second: { 'contract.json': JSON.stringify({ v: 2 }) },
      },
    });

    // 两份都在索引里（各自路径不同，都是当前版本）——去重只发生在拼提示词那一步
    const rows = allFiles(companyId).filter((r) => r.kind === 'contract');
    expect(rows).toHaveLength(2);

    const prompt = tasks.get('third');
    const listed = prompt.split('\n').filter((l) => l.includes('（任务级共享）'));
    expect(listed).toHaveLength(1);
    // 最新的赢（findFiles 是 created_at 升序，与沿边那一半的 byKind.set 同口径）
    expect(listed[0]).toContain(path.join(companyRoot(companyId), rows[1].rel_path));
    expect(listed[0]).not.toContain(rows[0].rel_path);
  });
});