"""OpenAI 兼容 VLM 实现：覆盖 Ollama / GLM / MiniMax / Moonshot / OpenRouter / OpenAI。"""
from __future__ import annotations

from pathlib import Path
from typing import Any

from app.core.logger import get_logger

from .base import VLMProvider, VLMResult
from .openai_vision_client import chat_completions, safe_parse_json


logger = get_logger(__name__)


_SYSTEM_PROMPT = """你是一个截图整理助手。用户会给你一张截图（可能附带 OCR 文本）。
请用简洁的中文，按以下 JSON 结构返回分析结果，**只输出 JSON，不要解释**：

{
  "title": "一句话标题，不超过 24 个字",
  "summary": "2-3 句话总结这张截图的内容、要点或笑点",
  "category": "从给定分类列表中选一个最贴切的名字；如果都不符合就填'其他'",
  "tags": ["3-6 个能帮助检索的细分标签"],
  "todos": [
     {"title": "如果截图里出现'待看/待读/待办/想试试/记一下'的内容，抽成一条 todo", "kind": "待读|待看|待办|学习", "note": "可空"}
  ],
  "suggestion": "可选：给用户的进一步行动建议或同类资源提示，可空"
}

要求：
- 标题要可读，不要复述"这是一张..."。
- summary 不要超过 80 字。
- todos 没有可识别项时给空数组。"""


class OpenAICompatVLM(VLMProvider):
    """统一调用 /v1/chat/completions，图片以 base64 data URL 传入。"""

    name = "openai_compat"

    def __init__(
        self,
        base_url: str,
        api_key: str,
        model: str,
        timeout: float = 60.0,
    ) -> None:
        self.base_url = base_url.rstrip("/")
        self.api_key = api_key
        self.model = model
        self.timeout = timeout

    async def analyze(
        self,
        image_path: Path,
        ocr_text: str,
        categories: list[str],
        allow_upload: bool,
    ) -> VLMResult:
        """调用模型并解析结构化 JSON。"""
        prompt = (
            f"可选分类：{', '.join(categories)}\n\n"
            f"OCR 文本（可能不完整或为空）：\n{ocr_text or '（无）'}"
        )
        content = await chat_completions(
            base_url=self.base_url,
            api_key=self.api_key,
            model=self.model,
            system_prompt=_SYSTEM_PROMPT,
            user_text=prompt,
            image_path=image_path if allow_upload else None,
            allow_upload=allow_upload,
            timeout=self.timeout,
            json_mode=True,
        )
        parsed = safe_parse_json(content)
        return _to_vlm_result(parsed)


def _to_vlm_result(data: dict[str, Any]) -> VLMResult:
    """JSON -> dataclass，容错地兜住字段。"""
    todos_raw = data.get("todos") or []
    todos: list[dict[str, str]] = []
    if isinstance(todos_raw, list):
        for item in todos_raw:
            if isinstance(item, dict) and item.get("title"):
                todos.append(
                    {
                        "title": str(item.get("title", ""))[:512],
                        "kind": str(item.get("kind", "")) or "待办",
                        "note": str(item.get("note", "") or ""),
                    }
                )
            elif isinstance(item, str):
                todos.append({"title": item, "kind": "待办", "note": ""})
    tags_raw = data.get("tags") or []
    if not isinstance(tags_raw, list):
        tags_raw = []
    return VLMResult(
        title=str(data.get("title", "") or "")[:128],
        summary=str(data.get("summary", "") or ""),
        category=str(data.get("category") or "") or None,
        tags=[str(t) for t in tags_raw if t][:8],
        todos=todos,
        suggestion=str(data.get("suggestion", "") or ""),
        raw=data,
    )