@renxqoo/renx-code 0.0.4 → 0.0.5

package/vendor/agent-root/src/providers/types/errors.ts

@@ -0,0 +1,323 @@
+/**
+ * Provider 错误类型定义
+ *
+ * 统一的错误类型定义，包括基础错误、可重试错误、永久性错误等
+ */
+
+// =============================================================================
+// 错误基类
+// =============================================================================
+
+export class LLMError extends Error {
+  constructor(
+    message: string,
+    public code?: string
+  ) {
+    super(message);
+    this.name = 'LLMError';
+  }
+}
+
+// =============================================================================
+// 指数退避配置
+// =============================================================================
+
+/**
+ * 指数退避配置
+ */
+export interface BackoffConfig {
+  /** 初始延迟（毫秒） */
+  initialDelayMs?: number;
+  /** 最大延迟（毫秒） */
+  maxDelayMs?: number;
+  /** 退避基数（默认 2） */
+  base?: number;
+  /** 是否启用 jitter（默认 true） */
+  jitter?: boolean;
+  /** 最大重试次数（用于计算最大退避） */
+  maxRetries?: number;
+}
+
+/**
+ * 默认退避配置
+ */
+export const DEFAULT_BACKOFF_CONFIG: Required<BackoffConfig> = {
+  initialDelayMs: 1000,
+  maxDelayMs: 60000,
+  base: 2,
+  jitter: true,
+  maxRetries: 20,
+};
+
+/**
+ * 计算带 jitter 的指数退避延迟
+ *
+ * 算法：min(maxDelay, initialDelay * (base ^ retryCount)) * random(0.5, 1.5)
+ * - 优先使用 retryAfter（服务器指定）
+ * - 支持指数增长
+ * - 添加 jitter 避免雷鸣羊群问题
+ */
+export function calculateBackoff(
+  retryCount: number,
+  retryAfterMs?: number,
+  config: BackoffConfig = {}
+): number {
+  const cfg = { ...DEFAULT_BACKOFF_CONFIG, ...config };
+
+  // 1. 如果服务器指定了 retry-after，优先使用（但不超过最大延迟）
+  if (typeof retryAfterMs === 'number' && retryAfterMs > 0) {
+    return Math.min(retryAfterMs, cfg.maxDelayMs);
+  }
+
+  // 2. 计算指数退避
+  const exponentialDelay = cfg.initialDelayMs * Math.pow(cfg.base, retryCount);
+  const cappedDelay = Math.min(exponentialDelay, cfg.maxDelayMs);
+
+  // 3. 添加 jitter（如果启用）
+  if (cfg.jitter) {
+    // 随机因子范围: [0.5, 1.5]，即 ±50%
+    const jitterFactor = 0.5 + Math.random();
+    return Math.floor(cappedDelay * jitterFactor);
+  }
+
+  return cappedDelay;
+}
+
+// =============================================================================
+// 可重试错误
+// =============================================================================
+
+export class LLMRetryableError extends LLMError {
+  /**
+   * @deprecated 请使用 calculateBackoff() 函数
+   */
+  getBackoff(retryCount: number): number {
+    return calculateBackoff(retryCount, this.retryAfter);
+  }
+
+  constructor(
+    message: string,
+    public retryAfter?: number,
+    code?: string
+  ) {
+    super(message, code);
+    this.name = 'LLMRetryableError';
+  }
+}
+
+export class LLMRateLimitError extends LLMRetryableError {
+  constructor(message: string, retryAfter?: number) {
+    super(message, retryAfter, 'RATE_LIMIT');
+    this.name = 'LLMRateLimitError';
+  }
+}
+
+// =============================================================================
+// 永久性错误
+// =============================================================================
+
+export class LLMPermanentError extends LLMError {
+  constructor(
+    message: string,
+    public statusCode?: number,
+    code?: string
+  ) {
+    super(message, code);
+    this.name = 'LLMPermanentError';
+  }
+}
+
+export class LLMAuthError extends LLMPermanentError {
+  constructor(message: string) {
+    super(message, 401, 'AUTH_FAILED');
+    this.name = 'LLMAuthError';
+  }
+}
+
+export class LLMNotFoundError extends LLMPermanentError {
+  constructor(
+    message: string,
+    public resourceType?: 'model' | 'endpoint' | 'resource'
+  ) {
+    super(message, 404, 'NOT_FOUND');
+    this.name = 'LLMNotFoundError';
+  }
+}
+
+export class LLMBadRequestError extends LLMPermanentError {
+  constructor(
+    message: string,
+    public validationErrors?: Record<string, string>
+  ) {
+    super(message, 400, 'BAD_REQUEST');
+    this.name = 'LLMBadRequestError';
+  }
+}
+
+// =============================================================================
+// 取消错误
+// =============================================================================
+
+export class LLMAbortedError extends LLMError {
+  constructor(message: string = 'Request was cancelled') {
+    super(message, 'ABORTED');
+    this.name = 'LLMAbortedError';
+  }
+}
+
+// =============================================================================
+// 工具函数
+// =============================================================================
+
+export type AbortReasonCategory = 'idle_timeout' | 'timeout' | 'abort' | 'unknown';
+
+export const PERMANENT_STREAM_ERROR_CODE_MARKERS = [
+  'invalid_request',
+  'bad_request',
+  'authentication',
+  'permission',
+  'forbidden',
+  'not_found',
+  'unsupported',
+  'context_length',
+  'content_filter',
+  'safety',
+  'invalid_parameter_error',
+] as const;
+
+export const PERMANENT_STREAM_ERROR_MESSAGE_PATTERNS: ReadonlyArray<RegExp> = [
+  /\binvalid[\s_-]?request\b/i,
+  /\bbad[\s_-]?request\b/i,
+  /\bauthentication\b/i,
+  /\bunauthorized\b/i,
+  /\bpermission\b/i,
+  /\bforbidden\b/i,
+  /\bnot[\s_-]?found\b/i,
+  /\bunsupported\b/i,
+  /\bcontext[\s_-]?length\b/i,
+  /\bcontent[\s_-]?filter\b/i,
+  /\bsafety\b/i,
+  /\binvalid[\s_-]?parameter\b/i,
+];
+
+export function abortReasonToText(reason: unknown): string {
+  if (reason instanceof Error) {
+    return `${reason.name} ${reason.message}`.trim();
+  }
+  if (typeof reason === 'string') {
+    return reason.trim();
+  }
+  return '';
+}
+
+export function isIdleTimeoutReasonText(reasonText: string): boolean {
+  const signature = reasonText.toLowerCase();
+  return (
+    signature.includes('idle timeout') ||
+    signature.includes('idletimeout') ||
+    signature.includes('idle_timeout')
+  );
+}
+
+export function isTimeoutReasonText(reasonText: string): boolean {
+  const signature = reasonText.toLowerCase();
+  return (
+    signature.includes('timeout') ||
+    signature.includes('timed out') ||
+    signature.includes('time out') ||
+    signature.includes('signal timed out')
+  );
+}
+
+export function classifyAbortReason(reason: unknown): AbortReasonCategory {
+  const reasonText = abortReasonToText(reason);
+  if (!reasonText) {
+    return 'unknown';
+  }
+
+  if (isIdleTimeoutReasonText(reasonText)) {
+    return 'idle_timeout';
+  }
+
+  if (isTimeoutReasonText(reasonText)) {
+    return 'timeout';
+  }
+
+  const signature = reasonText.toLowerCase();
+  if (signature.includes('abort') || signature.includes('cancel')) {
+    return 'abort';
+  }
+
+  return 'unknown';
+}
+
+export function isPermanentStreamChunkError(code?: string, message?: string): boolean {
+  const normalizedCode = typeof code === 'string' ? code.toLowerCase() : '';
+  if (normalizedCode) {
+    const matchedByCode = PERMANENT_STREAM_ERROR_CODE_MARKERS.some((marker) =>
+      normalizedCode.includes(marker)
+    );
+    if (matchedByCode) {
+      return true;
+    }
+  }
+
+  const normalizedMessage = typeof message === 'string' ? message : '';
+  if (!normalizedMessage) {
+    return false;
+  }
+
+  return PERMANENT_STREAM_ERROR_MESSAGE_PATTERNS.some((pattern) => pattern.test(normalizedMessage));
+}
+
+export function createErrorFromStatus(
+  status: number,
+  statusText: string,
+  errorText: string,
+  retryAfterMs?: number
+): LLMError {
+  let details = errorText;
+  try {
+    const parsed = JSON.parse(errorText);
+    details = parsed.error?.message || errorText;
+  } catch {
+    // 使用原始文本
+  }
+
+  const message = `${status} ${statusText}${details ? ` - ${details}` : ''}`;
+
+  switch (status) {
+    case 401:
+    case 403:
+      return new LLMAuthError(message);
+    case 404:
+      return new LLMNotFoundError(message, 'resource');
+    case 408:
+      return new LLMRetryableError(message, retryAfterMs, 'TIMEOUT');
+    case 429:
+      return new LLMRateLimitError(message, retryAfterMs);
+    case 400:
+      return new LLMBadRequestError(message);
+    case 501:
+      return new LLMPermanentError(message, 501, 'NOT_IMPLEMENTED');
+    case 500:
+    case 502:
+    case 503:
+    case 504:
+      return new LLMRetryableError(message, retryAfterMs, `SERVER_${status}`);
+    default:
+      return new LLMError(message, `HTTP_${status}`);
+  }
+}
+
+export function isRetryableError(error: unknown): error is LLMRetryableError {
+  return error instanceof LLMRetryableError;
+}
+
+export function isPermanentError(error: unknown): error is LLMPermanentError {
+  return error instanceof LLMPermanentError;
+}
+
+export function isAbortedError(error: unknown): error is LLMAbortedError {
+  return error instanceof LLMAbortedError;
+}

package/vendor/agent-root/src/providers/types/index.ts

@@ -0,0 +1,72 @@
+/**
+ * Types 统一导出
+ *
+ * 统一导出所有类型定义，提供单一的导入入口
+ */
+
+// API 相关类型
+export type {
+  ToolCall,
+  Role,
+  TextContentPart,
+  ImageUrlContentPart,
+  InputAudioContentPart,
+  InputVideoContentPart,
+  FileContentPart,
+  InputContentPart,
+  MessageContent,
+  Usage,
+  StreamOptions,
+  BaseLLMMessage,
+  LLMResponseMessage,
+  LLMRequestMessage,
+  FinishReason,
+  LLMResponse,
+  StreamChunkError,
+  Chunk,
+  StreamCallback,
+  Tool,
+  LLMGenerateOptions,
+  LLMRequest,
+  AnthropicStreamEvent,
+} from './api';
+
+// 配置相关类型
+export type {
+  BaseAPIConfig,
+  BaseProviderConfig,
+  OpenAICompatibleConfig,
+  ProviderLogger,
+} from './config';
+
+// Provider 相关类型
+export { LLMProvider } from './provider';
+
+// 错误类型
+export {
+  PERMANENT_STREAM_ERROR_CODE_MARKERS,
+  PERMANENT_STREAM_ERROR_MESSAGE_PATTERNS,
+  isPermanentStreamChunkError,
+  abortReasonToText,
+  isIdleTimeoutReasonText,
+  isTimeoutReasonText,
+  classifyAbortReason,
+  LLMError,
+  LLMRetryableError,
+  LLMRateLimitError,
+  LLMPermanentError,
+  LLMAuthError,
+  LLMNotFoundError,
+  LLMBadRequestError,
+  LLMAbortedError,
+  createErrorFromStatus,
+  isRetryableError,
+  isPermanentError,
+  isAbortedError,
+  calculateBackoff,
+  DEFAULT_BACKOFF_CONFIG,
+} from './errors';
+export type { BackoffConfig } from './errors';
+
+// Registry 相关类型
+export type { ProviderType, ModelId, ModelConfig } from './registry';

package/vendor/agent-root/src/providers/types/provider.ts

@@ -0,0 +1,45 @@
+/**
+ * Provider 接口类型定义
+ *
+ * Provider 相关的接口和类型定义
+ */
+
+import { BaseProviderConfig } from './config';
+import { Chunk, LLMGenerateOptions, LLMRequestMessage, LLMResponse } from './api';
+
+/**
+ * Provider 抽象基类
+ */
+export abstract class LLMProvider {
+  config: BaseProviderConfig;
+
+  protected constructor(config: BaseProviderConfig) {
+    this.config = config;
+  }
+
+  /**
+   * 从提供商生成非流式响应
+   * @param messages 对话消息列表
+   * @param options 可选参数
+   * @returns LLM 响应
+   */
+  abstract generate(
+    messages: LLMRequestMessage[],
+    options?: LLMGenerateOptions
+  ): Promise<LLMResponse>;
+
+  /**
+   * 从提供商生成流式响应
+   * @param messages 对话消息列表
+   * @param options 可选参数
+   * @returns 流式 chunk 生成器
+   */
+  abstract generateStream(
+    messages: LLMRequestMessage[],
+    options?: LLMGenerateOptions
+  ): AsyncGenerator<Chunk>;
+
+  abstract getTimeTimeout(): number;
+  abstract getLLMMaxTokens(): number;
+  abstract getMaxOutputTokens(): number;
+}

package/vendor/agent-root/src/providers/types/registry.ts

@@ -0,0 +1,88 @@
+/**
+ * Registry 相关类型定义
+ *
+ * Provider Registry 相关的类型定义
+ */
+
+/**
+ * Provider 厂商类型
+ */
+export type ProviderType =
+  | 'anthropic'
+  | 'kimi'
+  | 'deepseek'
+  | 'glm'
+  | 'minimax'
+  | 'openai'
+  | 'openrouter'
+  | 'qwen';
+
+/**
+ * 模型唯一标识
+ */
+export type ModelId =
+  // Anthropic 系列
+  | 'claude-opus-4.6'
+  // GLM 系列
+  | 'glm-4.7'
+  // MiniMax 系列
+  | 'minimax-2.5'
+  // Kimi 系列
+  | 'kimi-k2.5'
+  // DeepSeek 系列
+  // | 'deepseek-chat'
+  // GLM 5.0 系列
+  | 'glm-5'
+  // Qwen 系列
+  | 'qwen3.5-plus'
+  | 'qwen-kimi-k2.5'
+  | 'qwen-glm-5'
+  | 'qwen3.5-max'
+  | 'qwen-minimax-2.5'
+  | 'deepseek-reasoner'
+  | 'gpt-5.3'
+  | 'gpt-5.4'
+  | 'openrouter/hunter-alpha';
+
+/**
+ * 模型配置
+ */
+export interface ModelConfig {
+  /** 模型唯一标识 */
+  id: ModelId;
+  /** 所属厂商 */
+  provider: ProviderType;
+  /** 显示名称 */
+  name: string;
+  /** API 端点路径 */
+  endpointPath: string;
+  /** API Key 环境变量名 */
+  envApiKey: string;
+  /** Base URL 环境变量名 */
+  envBaseURL: string;
+  /** API 基础 URL */
+  baseURL: string;
+  /** API 模型名称 */
+  model: string;
+  /** 最大输出 token 数 */
+  max_tokens: number;
+  /** 最大上下文 token 数 */
+  LLMMAX_TOKENS: number;
+  /** 支持的特性 */
+  features: string[];
+  /** 多模态输入能力 */
+  modalities?: {
+    image?: boolean;
+    audio?: boolean;
+    video?: boolean;
+  };
+  /** API 密钥（可选） */
+  apiKey?: string;
+  /** 温度（可选） */
+  temperature?: number;
+  /** 默认工具流式输出（可选） */
+  tool_stream?: boolean;
+  thinking?: boolean;
+  timeout?: number;
+  model_reasoning_effort?: 'low' | 'medium' | 'high';
+}