sycommon-python-lib 0.2.5a22__tar.gz → 0.2.5a23__tar.gz

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.5a22
+Version: 0.2.5a23
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sycommon-python-lib"
-version = "0.2.5a22"
+version = "0.2.5a23"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.11"

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/src/sycommon/agent/mcp/models.py

@@ -8,6 +8,7 @@ class MCPToolStatus(BaseModel):
     """MCP 工具可用状态"""
     tool_name: str
     description: Optional[str] = None
+    display_name: Optional[str] = None
     available: bool = True
     last_check_at: Optional[str] = None
     last_error: Optional[str] = None
@@ -17,6 +18,7 @@ class MCPServerConfig(BaseModel):
     id: str
     user_id: Optional[str] = None
     name: str
+    alias: Optional[str] = None
     server_url: str
     description: Optional[str] = None
     headers: Optional[dict] = None
@@ -31,6 +33,7 @@ class MCPServerConfig(BaseModel):
 
 class MCPServerCreateRequest(BaseModel):
     name: str
+    alias: Optional[str] = None
     server_url: str
     description: Optional[str] = None
     headers: Optional[dict] = None
@@ -39,6 +42,7 @@ class MCPServerCreateRequest(BaseModel):
 
 class MCPServerUpdateRequest(BaseModel):
     name: Optional[str] = None
+    alias: Optional[str] = None
     server_url: Optional[str] = None
     description: Optional[str] = None
     headers: Optional[dict] = None

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/src/sycommon/agent/mcp/tool_loader.py

@@ -122,6 +122,7 @@ def _make_user_id_interceptor(user_id: str):
 async def load_mcp_tools(
     configs: List[MCPServerConfig],
     user_id: str = None,
+    tool_display_names: Optional[dict] = None,
     on_tool_success: Optional[Callable[[str, str], Awaitable[None]]] = None,
     on_tool_error: Optional[Callable[[str, str, str], Awaitable[None]]] = None,
     on_batch_available: Optional[Callable[[
@@ -135,6 +136,7 @@ async def load_mcp_tools(
     Args:
         configs: MCP 服务器配置列表（仅 enabled 的会被处理）
         user_id: 当前用户 ID，通过 interceptor 注入到每次 MCP 工具调用的 header 中
+        tool_display_names: 可选的工具中文显示名映射 {tool_full_name: display_name}
         on_tool_success: 单个工具调用成功回调
         on_tool_error: 单个工具调用失败回调
         on_batch_available: 服务器连接成功后批量标记工具可用回调 (config_id, tool_names)
@@ -186,6 +188,11 @@ async def load_mcp_tools(
                     lc_tool.name = f"mcp__{key}__{original_name}"
                     if lc_tool.description and not lc_tool.description.startswith("[MCP"):
                         lc_tool.description = f"[MCP:{config.name}] {lc_tool.description}"
+                    # 注入中文别名到 description，让 LLM 能识别用户说的中文名
+                    if tool_display_names:
+                        dn = tool_display_names.get(lc_tool.name)
+                        if dn:
+                            lc_tool.description = f"[别名: {dn}] {lc_tool.description}"
                     lc_tool = wrap_tool_with_error_handler(
                         lc_tool, config.id, config.name, original_name,
                         on_tool_success=on_tool_success,

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/src/sycommon/auth/wecom_ldap_service.py

@@ -293,8 +293,8 @@ class WeComLDAPService(metaclass=SingletonMeta):
 
     # ─── LDAP 查询 ───
 
-    def _load_ldap_index_sync(self) -> _LDAPIndex:
-        """同步加载 LDAP 用户索引"""
+    def _load_ldap_index_sync(self, max_retries: int = 10, retry_interval: float = 1.0) -> _LDAPIndex:
+        """同步加载 LDAP 用户索引（所有 URL 均失败则重试）"""
         import ldap3
 
         self._ensure_configs()
@@ -303,86 +303,98 @@ class WeComLDAPService(metaclass=SingletonMeta):
         bind_user = f"{config.username}@{config.default_dc}"
         raw_users = []
 
-        for url in config.urls:
-            try:
-                server = ldap3.Server(
-                    url.rstrip("/"),
-                    get_info=ldap3.ALL,
-                    connect_timeout=5,
-                )
-                conn = ldap3.Connection(
-                    server,
-                    user=bind_user,
-                    password=config.password,
-                    auto_bind=True,
-                    receive_timeout=30,
-                )
-
-                # 搜索盛业（大陆）和盛业（香港）两个 OU
-                for ou_base in [
-                    f"OU=盛业（大陆）,{config.base}",
-                    f"OU=盛业（香港）,{config.base}",
-                ]:
-                    conn.search(
-                        search_base=ou_base,
-                        search_filter="(objectClass=user)",
-                        search_scope=ldap3.SUBTREE,
-                        attributes=[
-                            "cn",
-                            "sAMAccountName",
-                            "sn",
-                            "mail",
-                            "department",
-                            "title",
-                            "telephoneNumber",
-                            "distinguishedName",
-                        ],
-                        size_limit=0,
+        for attempt in range(1, max_retries + 1):
+            raw_users = []
+            last_error = None
+            for url in config.urls:
+                try:
+                    server = ldap3.Server(
+                        url.rstrip("/"),
+                        get_info=ldap3.ALL,
+                        connect_timeout=5,
+                    )
+                    conn = ldap3.Connection(
+                        server,
+                        user=bind_user,
+                        password=config.password,
+                        auto_bind=True,
+                        receive_timeout=30,
                     )
 
-                    for entry in conn.entries:
-                        cn = str(entry.cn).strip() if entry.cn else ""
-                        sam = (
-                            str(entry.sAMAccountName).strip()
-                            if entry.sAMAccountName
-                            else ""
-                        )
-                        if not cn or not sam or sam.startswith("$"):
-                            continue
-
-                        raw_users.append(
-                            {
-                                "cn": cn,
-                                "sAMAccountName": sam,
-                                "sn": str(entry.sn).strip() if entry.sn else "",
-                                "mail": str(entry.mail).strip()
-                                if entry.mail
-                                else "",
-                                "department": str(entry.department).strip()
-                                if entry.department
-                                else "",
-                                "title": str(entry.title).strip()
-                                if entry.title
-                                else "",
-                                "telephoneNumber": str(
-                                    entry.telephoneNumber
-                                ).strip()
-                                if entry.telephoneNumber
-                                else "",
-                                "distinguishedName": str(
-                                    entry.distinguishedName
-                                ).strip()
-                                if entry.distinguishedName
-                                else "",
-                            }
+                    for ou_base in [
+                        f"OU=盛业（大陆）,{config.base}",
+                        f"OU=盛业（香港）,{config.base}",
+                    ]:
+                        conn.search(
+                            search_base=ou_base,
+                            search_filter="(objectClass=user)",
+                            search_scope=ldap3.SUBTREE,
+                            attributes=[
+                                "cn",
+                                "sAMAccountName",
+                                "sn",
+                                "mail",
+                                "department",
+                                "title",
+                                "telephoneNumber",
+                                "distinguishedName",
+                            ],
+                            size_limit=0,
                         )
 
-                conn.unbind()
-                break  # 第一个可用的 LDAP 服务器成功即停止
-
-            except Exception as e:
-                SYLogger.warning(f"[WeComLDAP] LDAP 连接 {url} 失败: {e}")
-                continue
+                        for entry in conn.entries:
+                            cn = str(entry.cn).strip() if entry.cn else ""
+                            sam = (
+                                str(entry.sAMAccountName).strip()
+                                if entry.sAMAccountName
+                                else ""
+                            )
+                            if not cn or not sam or sam.startswith("$"):
+                                continue
+
+                            raw_users.append(
+                                {
+                                    "cn": cn,
+                                    "sAMAccountName": sam,
+                                    "sn": str(entry.sn).strip() if entry.sn else "",
+                                    "mail": str(entry.mail).strip()
+                                    if entry.mail
+                                    else "",
+                                    "department": str(entry.department).strip()
+                                    if entry.department
+                                    else "",
+                                    "title": str(entry.title).strip()
+                                    if entry.title
+                                    else "",
+                                    "telephoneNumber": str(
+                                        entry.telephoneNumber
+                                    ).strip()
+                                    if entry.telephoneNumber
+                                    else "",
+                                    "distinguishedName": str(
+                                        entry.distinguishedName
+                                    ).strip()
+                                    if entry.distinguishedName
+                                    else "",
+                                }
+                            )
+
+                    conn.unbind()
+                    break  # 任一 URL 成功即停止
+
+                except Exception as e:
+                    last_error = e
+                    SYLogger.warning(f"[WeComLDAP] LDAP 连接 {url} 失败 (第 {attempt} 次): {e}")
+                    continue
+            else:
+                # 所有 URL 都失败
+                if attempt < max_retries:
+                    time.sleep(retry_interval)
+                    continue
+                else:
+                    SYLogger.error(f"[WeComLDAP] LDAP 全部 {max_retries} 次重试均失败，最后一次错误: {last_error}")
+
+            break  # 成功则跳出重试循环
 
         index = _LDAPIndex()
         index.build(raw_users)

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/src/sycommon/llm/llm_with_token_tracking.py

@@ -99,7 +99,7 @@ class LLMWithTokenTracking(BaseChatModel):
     def _clamp_max_tokens(self, messages, **kwargs):
         """动态调整 max_tokens，确保 input + max_output 不超过上下文窗口
 
-        每次 LLM 调用前，估算当前 input tokens，
+        每次 LLM 调用前，精确计算当前 input tokens，
         将 max_tokens = min(配置的 maxOutputTokens, context_window - input - buffer)
         """
         if not self.llmConfig or not self.llmConfig.maxTokens:
@@ -108,17 +108,66 @@ class LLMWithTokenTracking(BaseChatModel):
         if not max_output:
             return
 
-        from langchain_core.messages.utils import count_tokens_approximately
         try:
-            input_est = count_tokens_approximately(messages, chars_per_token=2.0)
+            input_est = self._count_tokens(messages)
         except Exception:
             return
 
-        buffer = 4096
+        buffer = 8192
         available = self.llmConfig.maxTokens - input_est - buffer
         dynamic_max = min(max_output, max(available, 1024))
         self.llm.max_tokens = dynamic_max
 
+    @staticmethod
+    def _count_tokens(messages) -> int:
+        """精确计算 messages 的 token 数
+
+        优先使用 tiktoken（OpenAI 兼容模型），否则回退到字符估算。
+        中文等非 ASCII 文本 chars_per_token 约 1.0-1.5，比英文 2.0-4.0 更密集。
+        """
+        total_chars = 0
+        for msg in messages:
+            content = getattr(msg, 'content', '')
+            if isinstance(content, str):
+                total_chars += len(content)
+            elif isinstance(content, list):
+                for part in content:
+                    if isinstance(part, str):
+                        total_chars += len(part)
+                    elif isinstance(part, dict):
+                        total_chars += len(str(part))
+
+        # 检查是否有高比例非 ASCII（中文/日文等）
+        sample = str(total_chars)
+        non_ascii = sum(1 for c in sample if ord(c) > 127)
+        ratio = non_ascii / max(len(sample), 1)
+
+        # 中文等语言 chars_per_token 约 1.5，英文约 4.0
+        chars_per_token = 1.5 if ratio > 0.3 else 3.5
+
+        # 尝试 tiktoken 精确计数
+        try:
+            import tiktoken
+            enc = tiktoken.get_encoding("cl100k_base")
+            token_count = 0
+            for msg in messages:
+                content = getattr(msg, 'content', '')
+                if isinstance(content, str):
+                    token_count += len(enc.encode(content))
+                elif isinstance(content, list):
+                    for part in content:
+                        if isinstance(part, str):
+                            token_count += len(enc.encode(part))
+                        elif isinstance(part, dict):
+                            token_count += len(enc.encode(str(part)))
+            # 加上消息格式开销（role、separator 等）
+            return token_count + len(messages) * 4
+        except Exception:
+            pass
+
+        # 回退：字符估算 + 每条消息 4 token 格式开销
+        return int(total_chars / chars_per_token) + len(messages) * 4
+
     def _generate(self, messages, stop=None, run_manager=None, **kwargs):
         """
         同步生成 - 核心方法

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/src/sycommon/llm/usage_token.py

@@ -174,17 +174,57 @@ class LLMWithAutoTokenUsage(BaseChatModel):
         if not max_output:
             return
 
-        from langchain_core.messages.utils import count_tokens_approximately
         try:
-            input_est = count_tokens_approximately(messages, chars_per_token=2.0)
+            input_est = self._count_tokens(messages)
         except Exception:
             return
 
-        buffer = 4096
+        buffer = 8192
         available = self.llmConfig.maxTokens - input_est - buffer
         dynamic_max = min(max_output, max(available, 1024))
         self.llm.max_tokens = dynamic_max
 
+    @staticmethod
+    def _count_tokens(messages) -> int:
+        """精确计算 messages 的 token 数"""
+        total_chars = 0
+        for msg in messages:
+            content = getattr(msg, 'content', '')
+            if isinstance(content, str):
+                total_chars += len(content)
+            elif isinstance(content, list):
+                for part in content:
+                    if isinstance(part, str):
+                        total_chars += len(part)
+                    elif isinstance(part, dict):
+                        total_chars += len(str(part))
+
+        # 尝试 tiktoken 精确计数
+        try:
+            import tiktoken
+            enc = tiktoken.get_encoding("cl100k_base")
+            token_count = 0
+            for msg in messages:
+                content = getattr(msg, 'content', '')
+                if isinstance(content, str):
+                    token_count += len(enc.encode(content))
+                elif isinstance(content, list):
+                    for part in content:
+                        if isinstance(part, str):
+                            token_count += len(enc.encode(part))
+                        elif isinstance(part, dict):
+                            token_count += len(enc.encode(str(part)))
+            return token_count + len(messages) * 4
+        except Exception:
+            pass
+
+        # 回退：字符估算（中文 1.5，英文 3.5）
+        sample = str(total_chars)
+        non_ascii = sum(1 for c in sample if ord(c) > 127)
+        ratio = non_ascii / max(len(sample), 1)
+        chars_per_token = 1.5 if ratio > 0.3 else 3.5
+        return int(total_chars / chars_per_token) + len(messages) * 4
+
     # ========== 实现BaseChatModel抽象方法 ==========
     def _generate(self, messages, stop=None, run_manager=None, **kwargs):
         self._clamp_max_tokens(messages, **kwargs)

{sycommon_python_lib-0.2.5a22 → sycommon_python_lib-0.2.5a23}/src/sycommon_python_lib.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.5a22
+Version: 0.2.5a23
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown