"""structured output 공용 래퍼.

OpenAI 호환 /chat/completions를 쓰는 프로바이더면 base_url만 바꿔 그대로 쓴다.
"""
from typing import TypeVar

from openai import AsyncOpenAI
from pydantic import BaseModel

T = TypeVar("T", bound=BaseModel)


class StructuredLLM:
    def __init__(self, model: str, api_key: str, base_url: str | None = None):
        self.model = model
        self._client = AsyncOpenAI(api_key=api_key, base_url=base_url)

    async def ask(
        self,
        schema: type[T],
        prompt: str,
        *,
        system: str | None = None,
        temperature: float = 0.0,
    ) -> T:
        messages = [{"role": "system", "content": system}] if system else []
        messages.append({"role": "user", "content": prompt})

        msg = (await self._client.chat.completions.parse(
            model=self.model,
            messages=messages,
            response_format=schema,
            temperature=temperature,
        )).choices[0].message

        if msg.refusal:
            raise RuntimeError(f"{self.model} 거절: {msg.refusal}")
        if msg.parsed is None:
            raise RuntimeError(f"{self.model} 파싱 실패: {msg.content!r}")
        return msg.parsed