@allbluecn/web-app 0.17.2 → 0.18.0

package/src/server/testcase/sandbox-runner.ts

@@ -0,0 +1,361 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+// AllBlue - 理想之海
+// Copyright (C) 2026 AllBlue Contributors
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <https://www.gnu.org/licenses/>.
+
+// 测试代码沙箱执行器（任务 28，规格 P5）：把 AI 生成的 vitest / Playwright spec 放进
+// 既有 sandbox provider 执行并解析 json reporter 结果，为任务 29（生成验证）与任务 31
+// （automation_code 复跑）提供统一执行底座。
+//
+// 执行流程：resolveSandboxProvider 取沙箱 → 新建一次性实例 → 宿主读取 spec 文件以
+// basename 写入沙箱 workspace（fs.write，兼容云沙箱不共享宿主路径的形态）→
+// process.exec 跑 npx vitest run / npx playwright test --reporter=json（AbortSignal
+// 超时到点 kill，进程终止由 provider 契约负责）→ 宽容解析 stdout JSON（vitest 摘要
+// numPassedTests/numFailedTests/numSkippedTests|numPendingTests + success；playwright
+// stats.expected/unexpected/skipped）→ {passed,failed,skipped,raw,...}。
+//
+// 降级与守卫（同任务 27 口径）：resolveSandboxProvider 抛错（未配置沙箱环境）时降级
+// local-process 直建（临时目录工作区）并在结果标记 degraded=true；runPlaywrightSpec
+// 需要 chromium，vercel 沙箱无浏览器二进制 → SpecRunnerUnavailableError（仅
+// local-process/daytona 可承载；E2E 验证亦仅限这两者），runVitestSpec 三种 provider
+// 均可。
+//
+// 日志治理：stdout+stderr 合并；超过 SPEC_LOG_CHAR_LIMIT 时 result.raw 截断到该上限，
+// 完整日志落 BlobStore（text/plain）返回 logBlobId；Blob 失败仅告警不反噬主链路。
+//
+// 模块形态：server/testcase/ 纯服务端模块（同 exec-pool.ts 口径），非 serverFn；
+// 测试全 mock provider 层（sandbox-runner.test.ts）不真跑测试。
+import fsPromises from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import type { ExecResult, SandboxHandle, SandboxProvider } from "@tanstack/ai-sandbox";
+import { getPorts } from "@/server/ai-gateway/runtime";
+import { localProcessSandbox } from "@/server/ai-gateway/sandbox/local-process";
+import {
+  resolveSandboxProvider,
+  type SandboxProviderKind,
+} from "@/server/ai-gateway/sandbox/provider";
+
+// ── 常量（阈值钉死，简报口径）──
+
+/** 单次 spec 执行超时默认值（到点 AbortSignal kill） */
+export const SPEC_RUN_TIMEOUT_MS = 300_000;
+/** 日志截断阈值：result.raw 上限，超过即完整日志落 Blob */
+export const SPEC_LOG_CHAR_LIMIT = 20_000;
+/** lastError 附带的原始输出尾部字符数 */
+const SPEC_LOG_TAIL_CHARS = 2_000;
+/** 宽容 JSON 提取的起点尝试上限（防超长噪声日志 O(n²) 扫描） */
+const JSON_PARSE_START_LIMIT = 50;
+/** 降级 local-process 的工作区根（实例目录在其下按实例 id 隔离） */
+const DEGRADED_SANDBOX_ROOT = path.join(os.tmpdir(), "allblue-testcase-sandbox");
+/** Blob 日志归属缺省用户（调用方未传 userId 时） */
+const DEFAULT_BLOB_USER = "testcase-runner";
+
+// ── 错误与出入参 ──
+
+/** spec 执行在当前部署形态不可用（vercel 沙箱无浏览器二进制） */
+export class SpecRunnerUnavailableError extends Error {
+  constructor(message: string) {
+    super(message);
+    this.name = "SpecRunnerUnavailableError";
+  }
+}
+
+export interface SpecRunOptions {
+  /** 单次执行超时（毫秒），到点 AbortSignal kill；缺省 SPEC_RUN_TIMEOUT_MS */
+  timeoutMs?: number;
+  /** 超长日志落 Blob 的归属用户；缺省 testcase-runner */
+  userId?: string;
+}
+
+export interface SpecRunResult {
+  /** 通过用例数（vitest numPassedTests / playwright stats.expected） */
+  passed: number;
+  /** 失败用例数（vitest numFailedTests / playwright stats.unexpected） */
+  failed: number;
+  /** 跳过用例数（vitest numSkippedTests|numPendingTests / playwright stats.skipped） */
+  skipped: number;
+  /** stdout+stderr 合并日志（超 SPEC_LOG_CHAR_LIMIT 截断；完整版见 logBlobId） */
+  raw: string;
+  /** 退出码（超时/exec 异常无 ExecResult 时为 -1 哨兵） */
+  exitCode: number;
+  /** 实际承载执行的沙箱 provider */
+  provider: SandboxProviderKind;
+  /** true = 未配置沙箱环境，降级 local-process 执行 */
+  degraded: boolean;
+  /** true = 超时被 kill */
+  timedOut: boolean;
+  /** 失败原因（超时 / json 解析失败 / exec 异常）；解析成功（含有用例失败）时缺省 */
+  lastError?: string;
+  /** 超长完整日志的 Blob 引用（blobStore.put 产物 id） */
+  logBlobId?: string;
+}
+
+// ── 入口：runVitestSpec / runPlaywrightSpec ──
+
+/** 在沙箱内执行单个 vitest spec 文件并解析 json reporter 结果 */
+export async function runVitestSpec(
+  specAbsPath: string,
+  options: SpecRunOptions = {},
+): Promise<SpecRunResult> {
+  return runSpecCommand(`npx vitest run ${shellQuote(path.basename(specAbsPath))} --reporter=json`, {
+    specAbsPath,
+    env: {},
+    options,
+  });
+}
+
+/** 在沙箱内执行单个 Playwright spec 文件（baseUrl 经 PLAYWRIGHT_TEST_BASE_URL 注入） */
+export async function runPlaywrightSpec(
+  specAbsPath: string,
+  baseUrl: string,
+  options: SpecRunOptions = {},
+): Promise<SpecRunResult> {
+  return runSpecCommand(
+    `npx playwright test ${shellQuote(path.basename(specAbsPath))} --reporter=json`,
+    {
+      specAbsPath,
+      env: { PLAYWRIGHT_TEST_BASE_URL: baseUrl },
+      options,
+      needsChromium: true,
+    },
+  );
+}
+
+async function runSpecCommand(
+  command: string,
+  input: {
+    specAbsPath: string;
+    env: Record<string, string>;
+    options: SpecRunOptions;
+    needsChromium?: boolean;
+  },
+): Promise<SpecRunResult> {
+  const { kind, provider, degraded } = await resolveProvider();
+  // chromium 承载守卫（同任务 27）：vercel 云沙箱无浏览器二进制，fail-closed
+  if (input.needsChromium && kind === "vercel") {
+    throw new SpecRunnerUnavailableError(
+      "Playwright spec 执行在 vercel 沙箱不可用（无浏览器二进制，仅 local-process/daytona 可承载）",
+    );
+  }
+
+  const handle = await provider.create({});
+  // 宿主读取 spec → 沙箱 workspace 以 basename 写入（云沙箱不共享宿主路径）
+  try {
+    const content = await fsPromises.readFile(input.specAbsPath, "utf8");
+    await handle.fs.write(path.basename(input.specAbsPath), content);
+  } catch (err) {
+    await destroyQuietly(handle);
+    throw new Error(`spec 文件读取/写入沙箱失败：${errMessage(err)}`);
+  }
+
+  // 超时 kill：到点触发 AbortSignal，进程终止由 provider exec 契约负责
+  const timeoutMs = input.options.timeoutMs ?? SPEC_RUN_TIMEOUT_MS;
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(), timeoutMs);
+  let execResult: ExecResult | null = null;
+  let failure: { timedOut: boolean; message: string } | null = null;
+  try {
+    execResult = await handle.process.exec(command, {
+      cwd: ".",
+      env: input.env,
+      signal: controller.signal,
+    });
+  } catch (err) {
+    failure = {
+      timedOut: controller.signal.aborted,
+      message: controller.signal.aborted
+        ? `执行超时（上限 ${timeoutMs}ms），进程已经 AbortSignal kill`
+        : `沙箱命令执行失败：${errMessage(err)}`,
+    };
+  } finally {
+    clearTimeout(timer);
+  }
+  await destroyQuietly(handle);
+
+  // 日志合并与宽容解析（JSON 在 stdout 为主，stderr 合并兜底）
+  const stdout = execResult?.stdout ?? "";
+  const stderr = execResult?.stderr ?? "";
+  const fullLog = [stdout, stderr].filter((s) => s.trim().length > 0).join("\n");
+  const counts = parseReporterCounts(stdout) ?? parseReporterCounts(fullLog);
+
+  const result: SpecRunResult = {
+    passed: counts?.passed ?? 0,
+    failed: counts?.failed ?? 0,
+    skipped: counts?.skipped ?? 0,
+    raw: fullLog.slice(0, SPEC_LOG_CHAR_LIMIT),
+    exitCode: execResult?.exitCode ?? -1,
+    provider: kind,
+    degraded,
+    timedOut: failure?.timedOut ?? false,
+  };
+  if (failure) {
+    result.lastError = `${failure.message}${fullLog ? `\n---- 输出尾部 ----\n${tailOf(fullLog)}` : ""}`;
+  } else if (!counts) {
+    result.lastError = `无法解析 json reporter 输出（exitCode=${result.exitCode}）；原始尾部：\n${tailOf(fullLog)}`;
+  }
+  return finalizeLog(result, fullLog, input.options.userId);
+}
+
+// ── provider 解析与降级 ──
+
+async function resolveProvider(): Promise<{
+  kind: SandboxProviderKind;
+  provider: SandboxProvider;
+  degraded: boolean;
+}> {
+  try {
+    const resolved = await resolveSandboxProvider();
+    return { kind: resolved.kind, provider: resolved.provider, degraded: false };
+  } catch {
+    // 无沙箱环境降级：local-process 直建（临时目录工作区），结果标记 degraded
+    return {
+      kind: "local-process",
+      provider: localProcessSandbox({ root: DEGRADED_SANDBOX_ROOT }),
+      degraded: true,
+    };
+  }
+}
+
+async function destroyQuietly(handle: SandboxHandle): Promise<void> {
+  try {
+    await handle.destroy();
+  } catch (err) {
+    console.warn("[sandbox-runner] 沙箱实例清理失败（不反噬主链路）:", errMessage(err));
+  }
+}
+
+// ── json reporter 宽容解析 ──
+
+interface SpecCounts {
+  passed: number;
+  failed: number;
+  skipped: number;
+}
+
+/** 先按 vitest 摘要形状解析，再按 playwright stats 形状兜底 */
+function parseReporterCounts(output: string): SpecCounts | null {
+  const json = extractJsonObject(output);
+  if (!json) return null;
+  return parseVitestSummary(json) ?? parsePlaywrightSummary(json);
+}
+
+function parseVitestSummary(json: Record<string, unknown>): SpecCounts | null {
+  if (typeof json.numPassedTests !== "number" && typeof json.numFailedTests !== "number") {
+    return null;
+  }
+  return {
+    passed: numOf(json.numPassedTests),
+    failed: numOf(json.numFailedTests),
+    // vitest 5 摘要将 skip 折入 numPendingTests；兼容 numSkippedTests 命名（简报口径）
+    skipped: numOf(json.numSkippedTests ?? json.numPendingTests),
+  };
+}
+
+function parsePlaywrightSummary(json: Record<string, unknown>): SpecCounts | null {
+  const stats = json.stats;
+  if (!stats || typeof stats !== "object" || Array.isArray(stats)) return null;
+  const s = stats as Record<string, unknown>;
+  if (
+    typeof s.expected !== "number" &&
+    typeof s.unexpected !== "number" &&
+    typeof s.skipped !== "number"
+  ) {
+    return null;
+  }
+  return {
+    passed: numOf(s.expected),
+    failed: numOf(s.unexpected),
+    skipped: numOf(s.skipped),
+  };
+}
+
+/**
+ * 宽容 JSON 提取：先整段 parse；失败则从每个 "{" 起点尝试（JSON 前有噪声前缀 /
+ * 后有尾注的场景），起点数受 JSON_PARSE_START_LIMIT 约束。找不到合法对象返回 null。
+ */
+function extractJsonObject(output: string): Record<string, unknown> | null {
+  const trimmed = output.trim();
+  if (trimmed) {
+    try {
+      const parsed: unknown = JSON.parse(trimmed);
+      if (isPlainObject(parsed)) return parsed;
+    } catch {
+      // 落入噪声容忍扫描
+    }
+  }
+  const lastBrace = output.lastIndexOf("}");
+  let start = output.indexOf("{");
+  for (let tries = 0; start !== -1 && tries < JSON_PARSE_START_LIMIT; tries++) {
+    if (lastBrace > start) {
+      try {
+        const parsed: unknown = JSON.parse(output.slice(start, lastBrace + 1));
+        if (isPlainObject(parsed)) return parsed;
+      } catch {
+        // 尝试下一个起点
+      }
+    }
+    start = output.indexOf("{", start + 1);
+  }
+  return null;
+}
+
+function isPlainObject(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+
+function numOf(value: unknown): number {
+  return typeof value === "number" && Number.isFinite(value) ? value : 0;
+}
+
+// ── 日志治理 ──
+
+/** 超长日志落 Blob（text/plain），失败仅告警不反噬主链路 */
+async function finalizeLog(
+  result: SpecRunResult,
+  fullLog: string,
+  userId?: string,
+): Promise<SpecRunResult> {
+  if (fullLog.length <= SPEC_LOG_CHAR_LIMIT) return result;
+  try {
+    const ref = await getPorts().blobStore.put(
+      `testcase-spec-log-${Date.now()}-${Math.random().toString(36).slice(2, 8)}.log`,
+      Buffer.from(fullLog, "utf8"),
+      "text/plain",
+      userId ?? DEFAULT_BLOB_USER,
+    );
+    result.logBlobId = ref.id;
+  } catch (err) {
+    console.warn("[sandbox-runner] 超长日志落 Blob 失败（不反噬主链路）:", errMessage(err));
+  }
+  return result;
+}
+
+function tailOf(log: string): string {
+  return log.length > SPEC_LOG_TAIL_CHARS
+    ? `…（前略，共 ${log.length} 字符）…\n${log.slice(-SPEC_LOG_TAIL_CHARS)}`
+    : log;
+}
+
+// ── 杂项 ──
+
+/** sh -c 单引号包裹（防文件名元字符注入命令） */
+function shellQuote(value: string): string {
+  return `'${value.replaceAll("'", `'\\''`)}'`;
+}
+
+function errMessage(err: unknown): string {
+  return err instanceof Error ? err.message : String(err);
+}

package/src/server/testcase/severity-map.ts

@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+// AllBlue - 理想之海
+// Copyright (C) 2026 AllBlue Contributors
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <https://www.gnu.org/licenses/>.
+
+// 用例优先级 → Bug 严重度映射（规格 §10 钉死）：用例转 Bug 时由 caseSnapshot
+// 的 priority 推导 severity。纯函数，禁止引入 prisma。
+import type { TestCasePriority } from "@/server/serverFns/testcase/constants";
+
+export type BugSeverity = "critical" | "high" | "medium" | "low";
+
+const PRIORITY_TO_SEVERITY: Readonly<Record<TestCasePriority, BugSeverity>> = {
+  p0: "critical",
+  p1: "high",
+  p2: "medium",
+  p3: "low",
+};
+
+/** p0→critical、p1→high、p2→medium、p3→low */
+export function priorityToSeverity(priority: TestCasePriority): BugSeverity {
+  return PRIORITY_TO_SEVERITY[priority];
+}

package/src/server/testcase/snapshot.ts

@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+// AllBlue - 理想之海
+// Copyright (C) 2026 AllBlue Contributors
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <https://www.gnu.org/licenses/>.
+
+// 用例执行时快照（规格 §3.7）：纯函数结构拷贝，锁定 {title, precondition, steps,
+// priority, type} 五字段——TestRunCase.caseSnapshot 存此形状，执行后用例编辑不影响
+// 历史批次记录（版本锁由调用方写入 caseVersion，本模块只负责内容定格）。
+// 纯服务端逻辑模块：无副作用、不触 DB，禁止引入 prisma。
+import type {
+  TestCaseCaseSnapshot,
+  TestCaseStep,
+  TestCaseSteps,
+} from "@/server/serverFns/testcase/types";
+import type { TestCasePriority, TestCaseType } from "@/server/serverFns/testcase/constants";
+
+/** 快照源：任何含五字段的用例形态（Prisma TestCase 行 / TestCaseSummary DTO 均可） */
+export type CaseSnapshotSource = {
+  title: string;
+  precondition: string | null;
+  steps: TestCaseSteps;
+  priority: TestCasePriority;
+  type: TestCaseType;
+};
+
+/** 从用例定格五字段快照（深拷贝 steps，源对象后续变更不影响快照） */
+export function buildCaseSnapshot(testCase: CaseSnapshotSource): TestCaseCaseSnapshot {
+  return {
+    title: testCase.title,
+    precondition: testCase.precondition,
+    steps: copySteps(testCase.steps),
+    priority: testCase.priority,
+    type: testCase.type,
+  };
+}
+
+function copySteps(steps: TestCaseSteps): TestCaseSteps {
+  return { steps: steps.steps.map(copyStep) };
+}
+
+function copyStep(step: TestCaseStep): TestCaseStep {
+  return step.expected === undefined
+    ? { order: step.order, action: step.action }
+    : { order: step.order, action: step.action, expected: step.expected };
+}

package/src/server/testcase/step-events.ts

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+// AllBlue - 理想之海
+// Copyright (C) 2026 AllBlue Contributors
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <https://www.gnu.org/licenses/>.
+
+// 执行步骤事件总线（任务 26）：MCP 工具（report_step，server 请求上下文 A）与
+// SSE 路由（/api/test-runs/$runId/events，请求上下文 B）之间的进程内发布/订阅通道，
+// 照 server/notifications/event-bus.ts 先例（module-level EventEmitter，键 run:<runId>）。
+//
+// ⚠ 多实例边界：EventEmitter 只存活于本进程——serverless 多实例部署下，report_step
+// 与 SSE 订阅若落到不同实例，事件不可达（SSE 只见本实例事件）。MVP 接受（web 端
+// AI 执行以同实例短窗为主）；跨实例需 Redis pub-sub 或轮询降级，见任务报告。
+import { EventEmitter } from "node:events";
+import type { TestRunStepEvent } from "@/server/serverFns/testcase/types";
+
+const emitter = new EventEmitter();
+emitter.setMaxListeners(0);
+
+export const runStepEventBus = {
+  emit(runId: string, event: TestRunStepEvent) {
+    emitter.emit(`run:${runId}`, event);
+  },
+  subscribe(runId: string, listener: (event: TestRunStepEvent) => void) {
+    emitter.on(`run:${runId}`, listener);
+    return () => emitter.off(`run:${runId}`, listener);
+  },
+};