gitinstall 1.1.0__py3-none-any.whl

gitinstall/llm.py

@@ -0,0 +1,569 @@
+"""
+llm.py - 多 LLM 适配器
+=====================================
+
+15 级降级策略，永远不失败：
+  1. Anthropic Claude     (ANTHROPIC_API_KEY)
+  2. OpenAI GPT-4o        (OPENAI_API_KEY)
+  3. OpenRouter           (OPENROUTER_API_KEY)
+  4. Google Gemini        (GEMINI_API_KEY)
+  5. Groq Llama 3.3       (GROQ_API_KEY)
+  6. DeepSeek             (DEEPSEEK_API_KEY)
+  7. 通义千问 Qwen        (DASHSCOPE_API_KEY)
+  8. 智谱 GLM             (ZHIPU_API_KEY)
+  9. 月之暗面 Kimi         (MOONSHOT_API_KEY)
+  10. 百川智能             (BAICHUAN_API_KEY)
+  11. 零一万物 Yi          (YI_API_KEY)
+  12. 阶跃星辰 Step        (STEPFUN_API_KEY)
+  13. LM Studio           (localhost:1234，本地运行即可)
+  14. Ollama              (localhost:11434，本地运行即可)
+  15. 无 LLM 规则模式      (永远可用，无需任何配置)
+
+面向大众原则（最低配置要求）：
+  - 本地模型推荐 1.5B~3B（普通笔记本即可运行）
+  - 默认推荐：qwen2.5:1.5b（中英双语，~1GB 显存/内存）
+  - 环境变量 GITINSTALL_LLM_MODEL 可自定义模型名称
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import re
+import socket
+import sys
+import urllib.error
+import urllib.request
+
+from log import get_logger
+from i18n import t
+
+logger = get_logger(__name__)
+from abc import ABC, abstractmethod
+from typing import Optional
+
+# LLM 请求超时（秒）：超过此时间自动降级到规则模式
+# 可通过环境变量 GITINSTALL_LLM_TIMEOUT 覆盖
+LLM_TIMEOUT = int(os.getenv("GITINSTALL_LLM_TIMEOUT", "30"))
+
+
+# ─────────────────────────────────────────────
+#  抽象基类
+# ─────────────────────────────────────────────
+
+class BaseLLMProvider(ABC):
+    """所有 LLM Provider 的统一接口"""
+
+    @abstractmethod
+    def complete(self, system: str, user: str, max_tokens: int = 2048) -> str:
+        """发送对话请求，返回 AI 回复文本"""
+
+    @property
+    @abstractmethod
+    def name(self) -> str:
+        """Provider 名称，用于日志和用户提示"""
+
+    def is_available(self) -> bool:
+        """检查 Provider 是否可用（子类可覆写）"""
+        return True
+
+
+# ─────────────────────────────────────────────
+#  通用 OpenAI 兼容 Provider
+#  支持：OpenAI / OpenRouter / Groq / DeepSeek /
+#         Gemini / LM Studio / Ollama / 任何 OpenAI 兼容 API
+# ─────────────────────────────────────────────
+
+class OpenAICompatibleProvider(BaseLLMProvider):
+    """
+    适配所有 OpenAI Chat Completions 兼容接口。
+    只用 Python 标准库 urllib，无需安装任何第三方包。
+    """
+
+    def __init__(self, api_key: str, base_url: str, model: str, provider_name: str):
+        self.api_key = api_key
+        self.base_url = base_url.rstrip("/")
+        self.model = model
+        self._name = provider_name
+
+    @property
+    def name(self) -> str:
+        return self._name
+
+    def complete(self, system: str, user: str, max_tokens: int = 2048) -> str:
+        payload = json.dumps({
+            "model": self.model,
+            "messages": [
+                {"role": "system", "content": system},
+                {"role": "user", "content": user},
+            ],
+            "max_tokens": max_tokens,
+            "temperature": 0.1,   # 安装任务要确定性，不要随机性
+        }).encode("utf-8")
+
+        req = urllib.request.Request(
+            f"{self.base_url}/chat/completions",
+            data=payload,
+            headers={
+                "Content-Type": "application/json",
+                "Authorization": f"Bearer {self.api_key}",
+                "User-Agent": "gitinstall/1.0",
+            },
+            method="POST",
+        )
+
+        try:
+            with urllib.request.urlopen(req, timeout=LLM_TIMEOUT) as resp:
+                data = json.loads(resp.read().decode("utf-8"))
+                return data["choices"][0]["message"]["content"]
+        except urllib.error.HTTPError as e:
+            body = e.read().decode("utf-8", errors="replace")
+            raise RuntimeError(f"{self._name} API 错误 {e.code}: {body[:300]}") from e
+        except (socket.timeout, TimeoutError):
+            raise RuntimeError(f"{self._name} 请求超时（{LLM_TIMEOUT}秒），自动降级到规则模式")
+        except urllib.error.URLError as e:
+            if isinstance(e.reason, (socket.timeout, TimeoutError)):
+                raise RuntimeError(f"{self._name} 请求超时（{LLM_TIMEOUT}秒），自动降级到规则模式")
+            raise RuntimeError(f"{self._name} 连接失败: {e.reason}") from e
+
+
+# ─────────────────────────────────────────────
+#  Anthropic 原生 Provider（非 OpenAI 兼容格式）
+# ─────────────────────────────────────────────
+
+class AnthropicProvider(BaseLLMProvider):
+    """Anthropic Messages API 原生接口"""
+
+    DEFAULT_MODEL = "claude-opus-4-5"
+
+    def __init__(self, api_key: str, model: Optional[str] = None):
+        self.api_key = api_key
+        self.model = model or self.DEFAULT_MODEL
+
+    @property
+    def name(self) -> str:
+        return f"Anthropic {self.model}"
+
+    def complete(self, system: str, user: str, max_tokens: int = 2048) -> str:
+        payload = json.dumps({
+            "model": self.model,
+            "max_tokens": max_tokens,
+            "system": system,
+            "messages": [{"role": "user", "content": user}],
+        }).encode("utf-8")
+
+        req = urllib.request.Request(
+            "https://api.anthropic.com/v1/messages",
+            data=payload,
+            headers={
+                "Content-Type": "application/json",
+                "x-api-key": self.api_key,
+                "anthropic-version": "2023-06-01",
+                "User-Agent": "gitinstall/1.0",
+            },
+            method="POST",
+        )
+
+        try:
+            with urllib.request.urlopen(req, timeout=LLM_TIMEOUT) as resp:
+                data = json.loads(resp.read().decode("utf-8"))
+                return data["content"][0]["text"]
+        except urllib.error.HTTPError as e:
+            body = e.read().decode("utf-8", errors="replace")
+            raise RuntimeError(f"Anthropic API 错误 {e.code}: {body[:300]}") from e
+        except (socket.timeout, TimeoutError):
+            raise RuntimeError(f"Anthropic 请求超时（{LLM_TIMEOUT}秒），自动降级到规则模式")
+        except urllib.error.URLError as e:
+            if isinstance(e.reason, (socket.timeout, TimeoutError)):
+                raise RuntimeError(f"Anthropic 请求超时（{LLM_TIMEOUT}秒），自动降级到规则模式")
+            raise RuntimeError(f"Anthropic 连接失败: {e.reason}") from e
+
+
+# ─────────────────────────────────────────────
+#  无 LLM 规则模式（永远可用）
+# ─────────────────────────────────────────────
+
+class HeuristicProvider(BaseLLMProvider):
+    """
+    无需任何 API Key 或本地模型。
+    通过正则表达式和规则库解析 README，提取安装命令。
+    
+    覆盖 90% 的主流开源项目（pip/npm/docker/brew/cargo 类型）。
+    """
+
+    @property
+    def name(self) -> str:
+        return "规则引擎（无 LLM）"
+
+    def complete(self, system: str, user: str, max_tokens: int = 2048) -> str:
+        """解析 prompt 中的 README 内容，提取安装步骤"""
+        return self._extract_install_plan(user)
+
+    def _extract_install_plan(self, content: str) -> str:
+        """从 README 文本中提取安装命令"""
+        steps = []
+
+        # 提取所有代码块
+        code_blocks = re.findall(
+            r'```(?:bash|shell|sh|zsh|powershell|cmd|console|text)?\n(.*?)```',
+            content,
+            re.DOTALL | re.IGNORECASE,
+        )
+
+        # 按优先级排序的安装命令模式
+        patterns = [
+            # Git 克隆（几乎所有项目的第一步）
+            (r'git\s+clone\s+(?:--depth[= ]\S+\s+)?(?:https?://|git@)\S+', "git_clone"),
+            # Python 安装
+            (r'pip(?:3)?\s+install[^\n]+', "pip"),
+            (r'pip(?:3)?\s+install\s+-r\s+requirements[^\n]*\.txt', "pip_req"),
+            (r'conda\s+(?:install|env\s+create)[^\n]+', "conda"),
+            (r'python(?:3)?\s+setup\.py\s+install', "setup_py"),
+            (r'python(?:3)?\s+-m\s+pip\s+install[^\n]+', "pip_m"),
+            # Node.js
+            (r'npm\s+install[^\n]*', "npm"),
+            (r'pnpm\s+install[^\n]*', "pnpm"),
+            (r'yarn(?:\s+install)?[^\n]*', "yarn"),
+            # 系统包管理器
+            (r'brew\s+install[^\n]+', "brew"),
+            (r'apt(?:-get)?\s+install[^\n]+', "apt"),
+            (r'dnf\s+install[^\n]+', "dnf"),
+            (r'pacman\s+-S[^\n]+', "pacman"),
+            (r'winget\s+install[^\n]+', "winget"),
+            (r'choco\s+install[^\n]+', "choco"),
+            # 其他语言
+            (r'cargo\s+install[^\n]+', "cargo"),
+            (r'go\s+install[^\n]+', "go"),
+            # Docker
+            (r'docker\s+(?:pull|run)[^\n]+', "docker"),
+            (r'docker-compose\s+up[^\n]*', "docker_compose"),
+            # 安装脚本
+            (r'curl\s+[^\n]+\s*\|[^\n]+(?:bash|sh)', "curl_pipe"),
+            (r'bash\s+(?:install|setup)\.sh[^\n]*', "bash_script"),
+            (r'make(?:\s+install)?[^\n]*', "make"),
+        ]
+
+        seen = set()
+        for block in code_blocks:
+            for pattern, kind in patterns:
+                for match in re.finditer(pattern, block, re.IGNORECASE):
+                    cmd = match.group(0).strip()
+                    # 安全过滤
+                    if self._is_dangerous(cmd):
+                        continue
+                    if cmd not in seen:
+                        seen.add(cmd)
+                        steps.append({
+                            "command": cmd,
+                            "type": kind,
+                            "description": self._describe(kind),
+                        })
+
+        if not steps:
+            return json.dumps({
+                "mode": "heuristic",
+                "status": "insufficient_data",
+                "steps": [],
+                "message": (
+                    "规则模式未能从 README 提取到安装命令。\n"
+                    "建议：配置任意 LLM（哪怕是免费的 Groq）以获得更好效果，\n"
+                    "或手动查阅项目 README。"
+                ),
+            }, ensure_ascii=False, indent=2)
+
+        return json.dumps({
+            "mode": "heuristic",
+            "status": "ok",
+            "steps": steps,
+            "warning": "规则模式，建议人工确认步骤后再执行",
+        }, ensure_ascii=False, indent=2)
+
+    @staticmethod
+    def _is_dangerous(cmd: str) -> bool:
+        """过滤危险命令"""
+        dangerous = [
+            r'rm\s+-rf\s+/',
+            r'rm\s+-rf\s+~',
+            r':\(\)\{',          # fork bomb
+            r'format\s+[cC]:',
+            r'mkfs\.',
+            r'dd\s+if=',
+            r'wget[^\n]+\|\s*(?:sudo\s+)?(?:bash|sh)',
+        ]
+        return any(re.search(p, cmd, re.IGNORECASE) for p in dangerous)
+
+    @staticmethod
+    def _describe(kind: str) -> str:
+        descriptions = {
+            "git_clone": "克隆代码仓库",
+            "pip": "安装 Python 包",
+            "pip_req": "安装 Python 依赖（requirements.txt）",
+            "pip_m": "安装 Python 包（pip module 方式）",
+            "conda": "安装 Conda 包",
+            "setup_py": "编译安装 Python 包",
+            "npm": "安装 Node.js 包",
+            "pnpm": "安装 Node.js 包（pnpm）",
+            "yarn": "安装 Node.js 包（yarn）",
+            "brew": "通过 Homebrew 安装（macOS）",
+            "apt": "通过 apt 安装（Debian/Ubuntu）",
+            "dnf": "通过 dnf 安装（Fedora/RHEL）",
+            "pacman": "通过 pacman 安装（Arch Linux）",
+            "winget": "通过 winget 安装（Windows）",
+            "choco": "通过 Chocolatey 安装（Windows）",
+            "cargo": "通过 Cargo 安装（Rust）",
+            "go": "通过 go install 安装",
+            "docker": "通过 Docker 运行",
+            "docker_compose": "通过 Docker Compose 启动",
+            "curl_pipe": "下载并执行安装脚本",
+            "bash_script": "执行安装脚本",
+            "make": "编译安装",
+        }
+        return descriptions.get(kind, "执行安装命令")
+
+
+# ─────────────────────────────────────────────
+#  自动检测 + 工厂函数
+# ─────────────────────────────────────────────
+
+def _is_port_open(host: str, port: int, timeout: float = 1.0) -> bool:
+    """检查本地端口是否在监听"""
+    try:
+        with socket.create_connection((host, port), timeout=timeout):
+            return True
+    except (socket.timeout, ConnectionRefusedError, OSError):
+        return False
+
+
+# 用户可通过环境变量指定本地模型，方便普通用户自定义
+# 例：export GITINSTALL_LLM_MODEL=qwen2.5:1.5b
+_DEFAULT_SMALL_MODEL = "qwen2.5:1.5b"    # ~1GB，普通笔记本可跑
+_DEFAULT_MEDIUM_MODEL = "qwen2.5:3b"    # ~2GB，推荐有独显用户
+
+# 优先推荐的小模型列表（按质量/大小权衡排序）
+# 如果 Ollama 里安装了这些模型之一，优先使用最小的
+_PREFERRED_SMALL_MODELS = [
+    "qwen2.5:1.5b",          # 1.5B，中英双语最佳（推荐大众用户）
+    "deepseek-r1:1.5b",      # 1.5B，有推理链
+    "smollm2:1.7b",          # 1.7B，英文为主
+    "qwen2.5-coder:1.5b",    # 1.5B，代码理解强（已有 base 版本）
+    "gemma3:1b",             # 1B，超小
+    "qwen2.5:3b",            # 3B，质量更好
+    "llama3.2:3b",           # 3B，Meta 官方
+    "llama3.2:1b",           # 1B，Meta 官方最小
+]
+
+
+def _get_local_model(base_url: str, endpoint: str = "/v1/models",
+                     fallback: str = _DEFAULT_SMALL_MODEL) -> str:
+    """
+    获取本地运行的模型名称。
+
+    优先级：
+      1. 环境变量 GITINSTALL_LLM_MODEL（用户显式指定）
+      2. 本地已安装的小模型（_PREFERRED_SMALL_MODELS 顺序）
+      3. 本地第一个可用模型
+      4. fallback 默认值
+    """
+    # 用户显式指定模型（最高优先级）
+    user_model = os.getenv("GITINSTALL_LLM_MODEL", "").strip()
+    if user_model:
+        return user_model
+
+    try:
+        req = urllib.request.Request(
+            f"{base_url}{endpoint}",
+            headers={"Authorization": "Bearer local"},
+        )
+        with urllib.request.urlopen(req, timeout=3) as resp:
+            data = json.loads(resp.read())
+            models = data.get("data", [])
+            if not models:
+                return fallback
+            installed_ids = [m.get("id", "") for m in models]
+            # 优先选用已知小模型
+            for preferred in _PREFERRED_SMALL_MODELS:
+                for installed in installed_ids:
+                    if preferred in installed or installed.startswith(preferred.split(":")[0] + ":1"):
+                        return installed
+            # 否则返回第一个
+            return installed_ids[0]
+    except Exception:
+        pass
+    return fallback
+
+
+def create_provider(force: Optional[str] = None) -> BaseLLMProvider:
+    """
+    自动检测并创建最优可用的 LLM Provider。
+    
+    Args:
+        force: 强制指定 Provider，可选值：
+               "anthropic" | "openai" | "openrouter" | "gemini" |
+               "groq" | "deepseek" | "lmstudio" | "ollama" | "none"
+               None 表示自动检测
+               
+    Returns:
+        可用的 LLM Provider 实例，永远不会返回 None
+    """
+
+    # ── 强制指定模式 ──
+    if force == "none":
+        logger.info(t("llm.using_heuristic"))
+        return HeuristicProvider()
+
+    if force == "lmstudio":
+        model = _get_local_model("http://localhost:1234", fallback=_DEFAULT_SMALL_MODEL)
+        logger.info(t("llm.using_with_model", name="LM Studio", model=model))
+        return OpenAICompatibleProvider("lm-studio", "http://localhost:1234/v1", model, "LM Studio")
+
+    if force == "ollama":
+        model = _get_local_model("http://localhost:11434", fallback=_DEFAULT_SMALL_MODEL)
+        logger.info(t("llm.using_with_model", name="Ollama", model=model))
+        logger.info(t("llm.ollama_hint", model=model))
+        return OpenAICompatibleProvider("ollama", "http://localhost:11434/v1", model, f"Ollama ({model})")
+
+    # ── 自动检测（按质量/成本优先级排序）──
+    # 格式：(环境变量名, Provider 构造函数)
+    cloud_providers = [
+        (
+            "ANTHROPIC_API_KEY",
+            lambda k: AnthropicProvider(k),
+        ),
+        (
+            "OPENAI_API_KEY",
+            lambda k: OpenAICompatibleProvider(k, "https://api.openai.com/v1", "gpt-4o", "OpenAI GPT-4o"),
+        ),
+        (
+            "OPENROUTER_API_KEY",
+            lambda k: OpenAICompatibleProvider(k, "https://openrouter.ai/api/v1", "anthropic/claude-opus-4-5", "OpenRouter"),
+        ),
+        (
+            "GEMINI_API_KEY",
+            lambda k: OpenAICompatibleProvider(
+                k,
+                "https://generativelanguage.googleapis.com/v1beta/openai",
+                "gemini-2.0-flash",
+                "Google Gemini",
+            ),
+        ),
+        (
+            "GROQ_API_KEY",
+            lambda k: OpenAICompatibleProvider(k, "https://api.groq.com/openai/v1", "llama-3.3-70b-versatile", "Groq Llama"),
+        ),
+        (
+            "DEEPSEEK_API_KEY",
+            lambda k: OpenAICompatibleProvider(k, "https://api.deepseek.com/v1", "deepseek-chat", "DeepSeek"),
+        ),
+        # ── 中国 LLM 提供商 ──
+        (
+            "DASHSCOPE_API_KEY",
+            lambda k: OpenAICompatibleProvider(
+                k, "https://dashscope.aliyuncs.com/compatible-mode/v1",
+                "qwen-plus", "通义千问 Qwen",
+            ),
+        ),
+        (
+            "ZHIPU_API_KEY",
+            lambda k: OpenAICompatibleProvider(
+                k, "https://open.bigmodel.cn/api/paas/v4",
+                "glm-4-flash", "智谱 GLM",
+            ),
+        ),
+        (
+            "MOONSHOT_API_KEY",
+            lambda k: OpenAICompatibleProvider(
+                k, "https://api.moonshot.cn/v1",
+                "moonshot-v1-8k", "月之暗面 Kimi",
+            ),
+        ),
+        (
+            "BAICHUAN_API_KEY",
+            lambda k: OpenAICompatibleProvider(
+                k, "https://api.baichuan-ai.com/v1",
+                "Baichuan4", "百川智能",
+            ),
+        ),
+        (
+            "YI_API_KEY",
+            lambda k: OpenAICompatibleProvider(
+                k, "https://api.lingyiwanwu.com/v1",
+                "yi-lightning", "零一万物 Yi",
+            ),
+        ),
+        (
+            "STEPFUN_API_KEY",
+            lambda k: OpenAICompatibleProvider(
+                k, "https://api.stepfun.com/v1",
+                "step-2-16k", "阶跃星辰 Step",
+            ),
+        ),
+    ]
+
+    # 检查环境变量，支持 force 指定特定 Provider
+    for env_var, factory in cloud_providers:
+        provider_name = env_var.replace("_API_KEY", "").lower()
+        if force and force != provider_name:
+            continue
+        key = os.getenv(env_var, "").strip()
+        if key:
+            provider = factory(key)
+            logger.info(t("llm.using_named", name=provider.name))
+            return provider
+
+    # 检查本地服务
+    if not force or force == "lmstudio":
+        if _is_port_open("localhost", 1234):
+            model = _get_local_model("http://localhost:1234", fallback="local-model")
+            logger.info(t("llm.detected_local", name="LM Studio", model=model))
+            return OpenAICompatibleProvider("lm-studio", "http://localhost:1234/v1", model, "LM Studio")
+
+    if not force or force == "ollama":
+        if _is_port_open("localhost", 11434):
+            model = _get_local_model("http://localhost:11434", fallback=_DEFAULT_SMALL_MODEL)
+            logger.info(t("llm.detected_local", name="Ollama", model=model))
+            return OpenAICompatibleProvider("ollama", "http://localhost:11434/v1", model, f"Ollama ({model})")
+
+    # 最终兜底：规则模式
+    logger.warning(t("llm.no_provider"))
+    logger.info(t("llm.hint_ollama"))
+    logger.info(t("llm.hint_groq"))
+    return HeuristicProvider()
+
+
+# ─────────────────────────────────────────────
+#  标准化 System Prompt（供各模块复用）
+# ─────────────────────────────────────────────
+
+# System Prompt 有两个版本：
+# - INSTALL_SYSTEM_PROMPT       完整版（大模型/云端 API 使用）
+# - INSTALL_SYSTEM_PROMPT_SMALL 精简版（1.5B~3B 本地小模型使用，避免 context 超限）
+
+INSTALL_SYSTEM_PROMPT = """\
+你是开源软件安装专家。根据用户提供的项目信息和环境，输出安装步骤（纯 JSON，无代码块）。
+
+JSON 格式：
+{"project_name":"名","steps":[{"id":1,"description":"说明","command":"命令"}],"launch_command":"启动命令","notes":"注意"}
+
+规则：Python项目用venv；Apple Silicon用pip install torch（无--index-url）；NVIDIA CUDA12用--index-url .../cu121；只输出JSON。
+"""
+
+# 精简版 prompt，适配 1.5B~3B 本地小模型
+# 关键：越短越好，指令越清楚越好，JSON schema 越简单越好
+INSTALL_SYSTEM_PROMPT_SMALL = """\
+Output JSON only. No explanation. Format:
+{"steps":[{"description":"step","command":"shell cmd"}],"launch_command":"start cmd"}
+Rules: use venv for Python; Apple MPS: pip install torch (no --index-url); CUDA12: add --index-url https://download.pytorch.org/whl/cu121
+"""
+
+ERROR_FIX_SYSTEM_PROMPT = """\
+你是一个开源软件安装报错修复专家。
+用户在安装 GitHub 项目时遇到了报错，请分析报错原因并提供修复方案。
+
+输出 JSON 格式：
+{
+  "root_cause": "报错根本原因（一句话）",
+  "fix_commands": ["修复命令1", "修复命令2"],
+  "explanation": "详细解释",
+  "prevention": "如何避免再次出现"
+}
+"""