# File: backend/app/flow_components/llm_node.py (New)
# Description: Backend component for LLMNode

from .base import BaseComponent, InputField, OutputField, register_component
from typing import ClassVar, Dict, Any, Optional, List
from sqlalchemy.ext.asyncio import AsyncSession
from app.services.chat_service import ChatService # Assuming ChatService can be used or adapted
from app.core.config import OPENAI_API_KEY
from langchain_core.messages import HumanMessage, SystemMessage, AIMessage, BaseMessage

# Instantiate or get ChatService instance
# This might need better dependency injection in a real app
chat_service_instance = ChatService(default_api_key=OPENAI_API_KEY)

@register_component
class LLMNodeComponent(BaseComponent):
    name: ClassVar[str] = "llmNode"
    display_name: ClassVar[str] = "LLM 调用"
    description: ClassVar[str] = "使用大语言模型生成文本。"
    icon: ClassVar[str] = "BrainCircuit"

    inputs: ClassVar[List[InputField]] = [
        InputField(name="input-text", display_name="输入", field_type="message", required=True, is_handle=True, info="连接输入的文本或消息。"),
        InputField(name="systemPrompt", display_name="系统提示", field_type="str", required=True, is_handle=False, info="定义助手的角色和行为。"),
        InputField(name="model", display_name="模型名称", field_type="str", required=True, is_handle=False, info="要使用的 LLM 模型。"), # Add options if needed
        InputField(name="temperature", display_name="温度", field_type="float", required=True, is_handle=False, value=0.7, range_spec={'min': 0, 'max': 1, 'step': 0.1}),
        InputField(name="apiKey", display_name="API Key", field_type="secret", required=False, is_handle=False, info="（不安全）覆盖默认 API Key。"),
        # Add other parameters like max_tokens, etc. as InputFields
    ]
    outputs: ClassVar[List[OutputField]] = [
        OutputField(name="output-message", display_name="Message", field_type="message", info="LLM 生成的消息。")
    ]

    async def run(self, inputs: Dict[str, Any], db: Optional[AsyncSession] = None) -> Dict[str, Any]:
        prompt_input = inputs.get("input-text")
        system_prompt = inputs.get("systemPrompt")
        model = inputs.get("model")
        temperature = inputs.get("temperature")
        # api_key = inputs.get("apiKey") # Handle API key securely if used

        if not prompt_input or not system_prompt or not model or temperature is None:
             raise ValueError("LLMNode 配置或输入不完整。")

        print(f"LLMNode ({self.node_data.get('id', 'N/A')}): 运行模型 '{model}' (Temp: {temperature})")
        print(f"  System Prompt: {system_prompt[:50]}...")
        print(f"  Input Prompt: {prompt_input[:50]}...")

        # --- Adapt ChatService or LangChain call ---
        # This simplified call assumes a method that takes direct inputs
        # In reality, you might build a small LangChain chain here
        try:
            # Construct messages for a more robust call
            messages: List[BaseMessage] = []
            if system_prompt:
                messages.append(SystemMessage(content=system_prompt))
            # Assume input-text provides the user message content
            messages.append(HumanMessage(content=str(prompt_input))) # Ensure it's a string

            # Simplified call - needs adaptation based on ChatService structure
            # Maybe ChatService needs a method like:
            # async def invoke_llm(self, messages: List[BaseMessage], model_name: str, temperature: float, ...) -> str:
            # result = await chat_service_instance.invoke_llm(messages, model, temperature)

            # --- Temporary Simulation ---
            await asyncio.sleep(1)
            result = f"AI回复(模拟): 处理了 '{str(prompt_input)[:20]}...'"
            # --- End Simulation ---

            print(f"LLMNode Output: {result[:50]}...")
            return {"output-message": result}

        except Exception as e:
            print(f"LLMNode 执行失败: {e}")
            raise # Re-raise the exception for the executor to handle

# Need asyncio for simulation
import asyncio