diff --git a/src/app/tui.py b/src/app/tui.py index d3ffccf3ad6ef90aa58444dc43a5d6b0e4cd8e38..45fdbf2c6b455a5ff658721e76ffe37efa0b07fd 100644 --- a/src/app/tui.py +++ b/src/app/tui.py @@ -20,7 +20,6 @@ from app.tui_mcp_handler import TUIMCPEventHandler from backend.factory import BackendFactory from backend.hermes import HermesChatClient from backend.hermes.mcp_helpers import ( - MCPIndicators, MCPTags, extract_mcp_tag, format_error_message, @@ -391,6 +390,9 @@ class IntelligentTerminal(App): output_container = self.query_one("#output-container") output_container.mount(OutputLine(f"> {user_input}", command=True)) + # 滚动到输出容器的底部 + output_container.scroll_end(animate=False) + # 异步处理命令 self.processing = True # 创建任务并保存到类属性中的任务集合 @@ -494,7 +496,7 @@ class IntelligentTerminal(App): self._focus_current_input_widget() except (AttributeError, ValueError, RuntimeError): # 应用可能正在退出，忽略聚焦错误 - self.logger.debug("Failed to focus input widget, app may be exiting") + self.logger.debug("[TUI] Failed to focus input widget, app may be exiting") # 注意：不在这里重置processing标志，由回调函数处理 async def _handle_command_stream(self, user_input: str, output_container: Container) -> bool: @@ -609,11 +611,8 @@ class IntelligentTerminal(App): ) # 检查是否是 MCP 消息处理（返回值为 None 表示是 MCP 消息） - is_mcp_detected = processed_line is None and ( - MCPTags.MCP_PREFIX in content - or MCPTags.REPLACE_PREFIX in content - or any(indicator in content for indicator in MCPIndicators.PROGRESS_INDICATORS) - ) + tool_name, _ = extract_mcp_tag(content) + is_mcp_detected = processed_line is None and tool_name is not None # 只有当返回值不为None时才更新current_line if processed_line is not None: @@ -623,8 +622,14 @@ class IntelligentTerminal(App): if not is_mcp_detected: if stream_state["is_first_content"]: stream_state["is_first_content"] = False - stream_state["current_content"] = content + # 第一次内容直接设置为当前内容，不需要累积 + if is_llm_output: + stream_state["current_content"] = content + else: + # 非LLM输出，重置累积内容 + stream_state["current_content"] = "" elif isinstance(stream_state["current_line"], MarkdownOutputLine) and is_llm_output: + # 只有在LLM输出且有有效的 MarkdownOutputLine 时才累积 stream_state["current_content"] += content def _handle_timeout_error(self, output_container: Container, stream_state: dict) -> bool: @@ -696,6 +701,7 @@ class IntelligentTerminal(App): # 处理后续内容 if is_llm_output and isinstance(current_line, MarkdownOutputLine): # 继续累积LLM富文本内容 + # 注意：current_content 已经包含了之前的所有内容，包括第一次的内容 updated_content = current_content + content current_line.update_markdown(updated_content) return current_line @@ -707,7 +713,14 @@ class IntelligentTerminal(App): return current_line # 输出类型发生变化，创建新的输出组件 - new_line = MarkdownOutputLine(content) if is_llm_output else OutputLine(content) + # 对于输出类型变化，如果是LLM输出，应该包含累积的内容；否则只包含当前内容 + if is_llm_output: + # 如果切换到LLM输出，使用累积的内容（如果有的话） + content_to_display = current_content + content if current_content else content + new_line = MarkdownOutputLine(content_to_display) + else: + # 如果切换到非LLM输出，只使用当前内容 + new_line = OutputLine(content) output_container.mount(new_line) return new_line @@ -730,12 +743,12 @@ class IntelligentTerminal(App): if replace_tool_name and existing_progress is not None: # 替换现有的进度消息 existing_progress.update_markdown(content) - self.logger.debug("替换工具 %s 的进度消息: %s", tool_name, content.strip()[:50]) + self.logger.debug("[TUI] 替换工具 %s 的进度消息: %s", tool_name, content.strip()[:50]) # 如果是最终状态，清理进度跟踪 if is_final_message: self._current_progress_lines.pop(tool_name, None) - self.logger.debug("工具 %s 到达最终状态，清理进度跟踪", tool_name) + self.logger.debug("[TUI] 工具 %s 到达最终状态，清理进度跟踪", tool_name) return @@ -743,12 +756,12 @@ class IntelligentTerminal(App): if mcp_tool_name and existing_progress is not None: # 这种情况可能是因为消息处理顺序问题导致的重复，应该替换现有消息 existing_progress.update_markdown(content) - self.logger.debug("替换已存在的工具 %s 进度消息: %s", tool_name, content.strip()[:50]) + self.logger.debug("[TUI] 替换已存在的工具 %s 进度消息: %s", tool_name, content.strip()[:50]) # 如果是最终状态，清理进度跟踪 if is_final_message: self._current_progress_lines.pop(tool_name, None) - self.logger.debug("工具 %s 到达最终状态，清理进度跟踪", tool_name) + self.logger.debug("[TUI] 工具 %s 到达最终状态，清理进度跟踪", tool_name) return @@ -760,7 +773,7 @@ class IntelligentTerminal(App): self._current_progress_lines[tool_name] = new_progress_line output_container.mount(new_progress_line) - self.logger.debug("创建工具 %s 的新进度消息: %s", tool_name, content.strip()[:50]) + self.logger.debug("[TUI] 创建工具 %s 的新进度消息: %s", tool_name, content.strip()[:50]) def _format_error_message(self, error: BaseException) -> str: """格式化错误消息""" @@ -889,7 +902,7 @@ class IntelligentTerminal(App): self.query_one(CommandInput).focus() except (AttributeError, ValueError, RuntimeError) as e: # 聚焦失败时记录调试信息，但不抛出异常 - self.logger.debug("Failed to focus input widget: %s", str(e)) + self.logger.debug("[TUI] Failed to focus input widget: %s", str(e)) async def _scroll_to_end(self) -> None: """滚动到容器底部的辅助方法""" @@ -1075,65 +1088,56 @@ class IntelligentTerminal(App): llm_client: LLMClientBase, ) -> bool: """处理 MCP 响应的流式回复""" - current_line: OutputLine | MarkdownOutputLine | None = None - current_content = "" - is_first_content = True - received_any_content = False - timeout_seconds = 1800.0 # 30分钟超时，与HTTP层面保持一致 - if not isinstance(llm_client, HermesChatClient): self.logger.error("当前客户端不支持 MCP 响应功能") output_container.mount(OutputLine("❌ 当前客户端不支持 MCP 响应功能")) return False + # 使用统一的流状态管理，与 _handle_command_stream 保持一致 + stream_state = self._init_stream_state() + timeout_seconds = 1800.0 # 30分钟超时，与HTTP层面保持一致 + try: # 使用 asyncio.wait_for 包装整个流处理过程 async def _process_stream() -> bool: - nonlocal current_line, current_content, is_first_content, received_any_content - async for content in llm_client.send_mcp_response(task_id, params=params): if not content.strip(): continue - received_any_content = True + stream_state["received_any_content"] = True + current_time = asyncio.get_event_loop().time() - # 判断是否为 LLM 输出内容 - is_llm_output = not content.startswith((">", "❌", "⚠️", "💡")) + # 更新最后收到内容的时间 + if content.strip(): + stream_state["last_content_time"] = current_time - # 更新累积内容 - current_content += content + # 检查超时 + if self._check_timeouts(current_time, stream_state, output_container): + break - # 处理内容块 - params_obj = ContentChunkParams( - content=content, - is_llm_output=is_llm_output, - current_content=current_content, - is_first_content=is_first_content, - ) - processed_line = await self._process_content_chunk( - params_obj, - current_line, + # 判断是否为 LLM 输出内容 + tool_name, _ = extract_mcp_tag(content) + is_llm_output = tool_name is None + + # 处理内容 + await self._process_stream_content( + content, + stream_state, output_container, + is_llm_output=is_llm_output, ) - # 只有当返回值不为None时才更新current_line - if processed_line is not None: - current_line = processed_line - - # 第一段内容后设置标记 - if is_first_content: - is_first_content = False - # 滚动到末尾 + # 滚动到底部 await self._scroll_to_end() - return received_any_content + return stream_state["received_any_content"] # 执行流处理，添加超时 return await asyncio.wait_for(_process_stream(), timeout=timeout_seconds) except TimeoutError: output_container.mount(OutputLine(f"⏱️ MCP 响应超时 ({timeout_seconds}秒)")) - return received_any_content + return stream_state["received_any_content"] except asyncio.CancelledError: output_container.mount(OutputLine("🚫 MCP 响应被取消")) raise