Aether/_deprecated_py_src/api/public/models.py

"""
统一的 Models API 端点

根据请求头认证方式自动返回对应格式:
- x-api-key + anthropic-version -> Claude 格式
- x-goog-api-key (header) 或 ?key= 参数 -> Gemini 格式
- Authorization: Bearer (bearer) -> OpenAI 格式
"""

from dataclasses import dataclass
from typing import Any

from fastapi import APIRouter, Depends, Query, Request
from fastapi.responses import JSONResponse
from sqlalchemy.orm import Session

from src.api.base.adapter import ApiAdapter, ApiMode
from src.api.base.context import ApiRequestContext
from src.api.base.pipeline import get_pipeline
from src.api.base.models_service import (
    AccessRestrictions,
    ModelInfo,
    find_model_by_id,
    get_available_provider_ids,
    get_compatible_provider_formats,
    list_available_models,
)
from src.core.api_format import (
    detect_request_context,
)
from src.core.api_format.conversion import (
    format_conversion_registry,
    register_default_normalizers,
)
from src.core.logger import logger
from src.database import get_db
from src.models.database import ApiKey, User
from src.services.auth.service import AuthService
from src.services.provider.format import normalize_endpoint_signature

router = APIRouter(tags=["System Catalog"])
pipeline = get_pipeline()

# 各格式对应的 API 格式列表（包括对应的 CLI 格式）
_CLAUDE_FORMATS = ["claude:chat", "claude:cli"]
_OPENAI_FORMATS = ["openai:chat", "openai:cli", "openai:compact"]
_GEMINI_FORMATS = ["gemini:chat", "gemini:cli"]

# 所有格式（用于格式转换时的查询）
_ALL_CHAT_FORMATS = [
    *_CLAUDE_FORMATS,
    *_OPENAI_FORMATS,
    *_GEMINI_FORMATS,
]


def _detect_api_format_and_key(request: Request) -> tuple[str, str | None]:
    """
    根据请求头检测 API 格式并提取 API Key

    检测顺序:
    1. x-api-key + anthropic-version -> Claude
    2. x-goog-api-key (header) 或 ?key= -> Gemini
    3. Authorization: Bearer -> OpenAI (默认)

    Returns:
        (api_format, api_key) 元组
    """
    context = detect_request_context(request)
    return context.endpoint.key, context.credentials


def _get_formats_for_api(api_format: str) -> list[str]:
    """获取对应 API 格式的端点格式列表"""
    fam = (api_format.split(":", 1)[0] if api_format else "").strip().lower()
    if fam == "claude":
        return _CLAUDE_FORMATS
    if fam == "gemini":
        return _GEMINI_FORMATS
    return _OPENAI_FORMATS


def _is_format_conversion_enabled(db: Session) -> bool:
    """检查全局格式转换开关（从数据库配置读取，默认开启）"""
    from src.services.system.config import SystemConfigService

    return SystemConfigService.is_format_conversion_enabled(db)


def _get_convertible_formats(client_format: str) -> list[str]:
    """
    获取客户端格式可转换到的所有目标格式列表

    始终返回所有有转换器的格式（包括客户端格式本身），
    由下游 get_compatible_provider_formats 按三层开关（全局/Provider/端点）精确过滤。
    """
    client_format_norm = normalize_endpoint_signature(client_format)

    # 收集所有可转换的格式
    register_default_normalizers()
    convertible_formats: list[str] = []
    for target_format in _ALL_CHAT_FORMATS:
        target_norm = normalize_endpoint_signature(target_format)
        # 相同格式始终可用
        if target_norm == client_format_norm:
            convertible_formats.append(target_norm)
            continue

        # 检查是否有双向转换器
        if format_conversion_registry.can_convert_full(
            client_format_norm,
            target_norm,
            require_stream=False,
        ):
            convertible_formats.append(target_norm)

    # 去重并保持稳定顺序
    return list(dict.fromkeys(convertible_formats)) if convertible_formats else [client_format_norm]


def _flatten_provider_formats(provider_to_formats: dict[str, set[str]]) -> list[str]:
    """合并 Provider 格式映射为唯一格式列表"""
    if not provider_to_formats:
        return []
    all_formats: set[str] = set()
    for formats in provider_to_formats.values():
        all_formats.update(formats)
    return sorted(all_formats)


def _get_family(api_format: str) -> str:
    """从 endpoint signature 提取协议族（如 'openai:chat' -> 'openai'）。"""
    return (str(api_format).split(":", 1)[0] if api_format else "").strip().lower()


def _build_empty_list_response(api_format: str) -> dict:
    """根据 API 格式构建空列表响应"""
    fam = _get_family(api_format)
    if fam == "claude":
        return {"data": [], "has_more": False, "first_id": None, "last_id": None}
    elif fam == "gemini":
        return {"models": []}
    else:
        return {"object": "list", "data": []}


def _filter_formats_by_restrictions(
    formats: list[str], restrictions: AccessRestrictions, api_format: str
) -> tuple[list[str], dict | None]:
    """
    根据访问限制过滤 API 格式

    Returns:
        (过滤后的格式列表, 空响应或None)
        如果过滤后为空，返回对应格式的空响应
    """
    if restrictions.allowed_api_formats is None:
        return formats, None
    filtered = [f for f in formats if restrictions.is_api_format_allowed(f)]
    if not filtered:
        logger.info(f"[Models] API Key 不允许访问格式 {api_format}")
        return [], _build_empty_list_response(api_format)
    return filtered, None


def _authenticate(db: Session, api_key: str | None) -> tuple[User | None, ApiKey | None]:
    """
    认证 API Key

    Returns:
        (user, api_key_record) 元组，认证失败返回 (None, None)
    """
    if not api_key:
        logger.debug("[Models] 认证失败: 未提供 API Key")
        return None, None

    result = AuthService.authenticate_api_key(db, api_key)
    if not result:
        logger.debug("[Models] 认证失败: API Key 无效")
        return None, None

    user, key_record = result
    logger.debug(f"[Models] 认证成功: {user.email} (Key: {key_record.name})")
    return result


def _build_auth_error_response(api_format: str) -> JSONResponse:
    """根据 API 格式构建认证错误响应"""
    fam = _get_family(api_format)
    if fam == "claude":
        return JSONResponse(
            status_code=401,
            content={
                "type": "error",
                "error": {
                    "type": "authentication_error",
                    "message": "Invalid API key provided",
                },
            },
        )
    elif fam == "gemini":
        return JSONResponse(
            status_code=401,
            content={
                "error": {
                    "code": 401,
                    "message": "API key not valid. Please pass a valid API key.",
                    "status": "UNAUTHENTICATED",
                }
            },
        )
    else:
        return JSONResponse(
            status_code=401,
            content={
                "error": {
                    "message": "Incorrect API key provided. You can find your API key at https://platform.openai.com/account/api-keys.",
                    "type": "invalid_request_error",
                    "param": None,
                    "code": "invalid_api_key",
                }
            },
        )


# ============================================================================
# 响应构建函数
# ============================================================================


def _build_claude_list_response(
    models: list[ModelInfo],
    before_id: str | None,
    after_id: str | None,
    limit: int,
) -> dict:
    """构建 Claude 格式的列表响应"""
    model_data_list = [
        {
            "id": m.id,
            "type": "model",
            "display_name": m.display_name,
            "created_at": m.created_at,
        }
        for m in models
    ]

    # 处理分页
    start_idx = 0
    if after_id:
        for i, m in enumerate(model_data_list):
            if m["id"] == after_id:
                start_idx = i + 1
                break

    end_idx = len(model_data_list)
    if before_id:
        for i, m in enumerate(model_data_list):
            if m["id"] == before_id:
                end_idx = i
                break

    paginated = model_data_list[start_idx:end_idx][:limit]

    first_id = paginated[0]["id"] if paginated else None
    last_id = paginated[-1]["id"] if paginated else None
    has_more = len(model_data_list[start_idx:end_idx]) > limit

    return {
        "data": paginated,
        "has_more": has_more,
        "first_id": first_id,
        "last_id": last_id,
    }


def _build_openai_list_response(models: list[ModelInfo]) -> dict:
    """构建 OpenAI 格式的列表响应"""
    data = [
        {
            "id": m.id,
            "object": "model",
            "created": m.created_timestamp,
            "owned_by": m.provider_name,
        }
        for m in models
    ]
    return {"object": "list", "data": data}


def _build_gemini_list_response(
    models: list[ModelInfo],
    page_size: int,
    page_token: str | None,
) -> dict:
    """构建 Gemini 格式的列表响应"""
    # 处理分页
    start_idx = 0
    if page_token:
        try:
            start_idx = int(page_token)
        except ValueError:
            start_idx = 0

    end_idx = start_idx + page_size
    paginated_models = models[start_idx:end_idx]

    models_data = [
        {
            "name": f"models/{m.id}",
            "baseModelId": m.id,
            "version": "001",
            "displayName": m.display_name,
            "description": m.description or f"Model {m.id}",
            "inputTokenLimit": m.context_limit if m.context_limit is not None else 128000,
            "outputTokenLimit": m.output_limit if m.output_limit is not None else 8192,
            "supportedGenerationMethods": ["generateContent", "countTokens"],
            "temperature": 1.0,
            "maxTemperature": 2.0,
            "topP": 0.95,
            "topK": 64,
        }
        for m in paginated_models
    ]

    response: dict = {"models": models_data}
    if end_idx < len(models):
        response["nextPageToken"] = str(end_idx)

    return response


def _build_claude_model_response(model_info: ModelInfo) -> dict:
    """构建 Claude 格式的模型详情响应"""
    return {
        "id": model_info.id,
        "type": "model",
        "display_name": model_info.display_name,
        "created_at": model_info.created_at,
    }


def _build_openai_model_response(model_info: ModelInfo) -> dict:
    """构建 OpenAI 格式的模型详情响应"""
    return {
        "id": model_info.id,
        "object": "model",
        "created": model_info.created_timestamp,
        "owned_by": model_info.provider_name,
    }


def _build_gemini_model_response(model_info: ModelInfo) -> dict:
    """构建 Gemini 格式的模型详情响应"""
    return {
        "name": f"models/{model_info.id}",
        "baseModelId": model_info.id,
        "version": "001",
        "displayName": model_info.display_name,
        "description": model_info.description or f"Model {model_info.id}",
        "inputTokenLimit": (
            model_info.context_limit if model_info.context_limit is not None else 128000
        ),
        "outputTokenLimit": (
            model_info.output_limit if model_info.output_limit is not None else 8192
        ),
        "supportedGenerationMethods": ["generateContent", "countTokens"],
        "temperature": 1.0,
        "maxTemperature": 2.0,
        "topP": 0.95,
        "topK": 64,
    }


# ============================================================================
# 404 响应
# ============================================================================


def _build_404_response(model_id: str, api_format: str) -> JSONResponse:
    """根据 API 格式构建 404 响应"""
    fam = _get_family(api_format)
    if fam == "claude":
        return JSONResponse(
            status_code=404,
            content={
                "type": "error",
                "error": {"type": "not_found_error", "message": f"Model '{model_id}' not found"},
            },
        )
    elif fam == "gemini":
        return JSONResponse(
            status_code=404,
            content={
                "error": {
                    "code": 404,
                    "message": f"models/{model_id} is not found",
                    "status": "NOT_FOUND",
                }
            },
        )
    else:
        return JSONResponse(
            status_code=404,
            content={
                "error": {
                    "message": f"The model '{model_id}' does not exist",
                    "type": "invalid_request_error",
                    "param": "model",
                    "code": "model_not_found",
                }
            },
        )


# ============================================================================
# Adapter helpers
# ============================================================================


class PublicModelsApiAdapter(ApiAdapter):
    mode = ApiMode.PUBLIC
    eager_request_body = False

    def authorize(self, context: ApiRequestContext) -> None:  # type: ignore[override]
        return None


def _handle_model_list_request(
    context: ApiRequestContext,
    *,
    before_id: str | None,
    after_id: str | None,
    limit: int,
    page_size: int,
    page_token: str | None,
) -> dict | JSONResponse:
    request = context.request
    db = context.db

    api_format, api_key = _detect_api_format_and_key(request)
    logger.info(f"[Models] GET /v1/models | format={api_format}")

    user, key_record = _authenticate(db, api_key)
    if not user:
        return _build_auth_error_response(api_format)

    restrictions = AccessRestrictions.from_api_key_and_user(key_record, user)
    global_conversion_enabled = _is_format_conversion_enabled(db)
    candidate_formats = _get_convertible_formats(api_format)
    candidate_formats, empty_response = _filter_formats_by_restrictions(
        candidate_formats, restrictions, api_format
    )
    if empty_response is not None:
        return empty_response

    provider_to_formats = get_compatible_provider_formats(
        db, api_format, candidate_formats, global_conversion_enabled
    )
    formats = _flatten_provider_formats(provider_to_formats)

    available_provider_ids = get_available_provider_ids(db, formats, provider_to_formats)
    if not available_provider_ids:
        return _build_empty_list_response(api_format)

    async def _list() -> dict:
        models = await list_available_models(
            db,
            available_provider_ids,
            formats,
            restrictions,
            provider_to_formats=provider_to_formats,
            client_format=api_format,
        )
        logger.debug(f"[Models] 返回 {len(models)} 个模型")

        if _get_family(api_format) == "claude":
            return _build_claude_list_response(models, before_id, after_id, limit)
        if _get_family(api_format) == "gemini":
            return _build_gemini_list_response(models, page_size, page_token)
        return _build_openai_list_response(models)

    return _list()


def _handle_model_detail_request(
    context: ApiRequestContext,
    *,
    model_id: str,
    force_gemini_name: bool,
) -> dict | JSONResponse:
    request = context.request
    db = context.db
    api_format, api_key = _detect_api_format_and_key(request)

    resolved_model_id = model_id
    if force_gemini_name:
        resolved_model_id = model_id[7:] if model_id.startswith("models/") else model_id
        logger.info(f"[Models] GET /v1beta/models/{resolved_model_id} | format=gemini")
    else:
        if _get_family(api_format) == "gemini" and resolved_model_id.startswith("models/"):
            resolved_model_id = resolved_model_id[7:]
        logger.info(f"[Models] GET /v1/models/{resolved_model_id} | format={api_format}")

    user, key_record = _authenticate(db, api_key)
    if not user:
        return _build_auth_error_response(api_format)

    restrictions = AccessRestrictions.from_api_key_and_user(key_record, user)
    global_conversion_enabled = _is_format_conversion_enabled(db)
    candidate_formats = _get_convertible_formats(api_format)
    candidate_formats, _ = _filter_formats_by_restrictions(
        candidate_formats, restrictions, api_format
    )
    provider_to_formats = get_compatible_provider_formats(
        db, api_format, candidate_formats, global_conversion_enabled
    )
    formats = _flatten_provider_formats(provider_to_formats)
    if not formats:
        return _build_404_response(resolved_model_id, api_format)

    available_provider_ids = get_available_provider_ids(db, formats, provider_to_formats)
    model_info = find_model_by_id(
        db,
        resolved_model_id,
        available_provider_ids,
        formats,
        restrictions,
        provider_to_formats=provider_to_formats,
    )

    if not model_info:
        return _build_404_response(resolved_model_id, api_format)

    if _get_family(api_format) == "claude":
        return _build_claude_model_response(model_info)
    if _get_family(api_format) == "gemini":
        return _build_gemini_model_response(model_info)
    return _build_openai_model_response(model_info)


@dataclass
class PublicModelsListAdapter(PublicModelsApiAdapter):
    before_id: str | None
    after_id: str | None
    limit: int
    page_size: int
    page_token: str | None

    async def handle(self, context: ApiRequestContext) -> dict | JSONResponse:  # type: ignore[override]
        result = _handle_model_list_request(
            context,
            before_id=self.before_id,
            after_id=self.after_id,
            limit=self.limit,
            page_size=self.page_size,
            page_token=self.page_token,
        )
        if hasattr(result, "__await__"):
            return await result
        return result


@dataclass
class PublicModelDetailAdapter(PublicModelsApiAdapter):
    model_id: str
    force_gemini_name: bool = False

    async def handle(self, context: ApiRequestContext) -> dict | JSONResponse:  # type: ignore[override]
        return _handle_model_detail_request(
            context,
            model_id=self.model_id,
            force_gemini_name=self.force_gemini_name,
        )


@router.get("/v1/models", response_model=None)
async def list_models(
    request: Request,
    # Claude 分页参数
    before_id: str | None = Query(None, description="返回此 ID 之前的结果 (Claude)"),
    after_id: str | None = Query(None, description="返回此 ID 之后的结果 (Claude)"),
    limit: int = Query(20, ge=1, le=1000, description="返回数量限制 (Claude)"),
    # Gemini 分页参数
    page_size: int = Query(50, alias="pageSize", ge=1, le=1000, description="每页数量 (Gemini)"),
    page_token: str | None = Query(None, alias="pageToken", description="分页 token (Gemini)"),
    db: Session = Depends(get_db),
) -> dict | JSONResponse:
    """
    列出可用模型（统一端点）

    根据请求头中的认证方式自动检测 API 格式，并返回相应格式的模型列表。
    此接口兼容 Claude、OpenAI 和 Gemini 三种 API 格式。

    **格式检测规则**
    - x-api-key + anthropic-version → Claude 格式
    - x-goog-api-key 或 ?key= → Gemini 格式
    - Authorization: Bearer → OpenAI 格式（默认）

    **查询参数**

    Claude 格式：
    - before_id: 返回此 ID 之前的结果，用于向前分页
    - after_id: 返回此 ID 之后的结果，用于向后分页
    - limit: 返回数量限制，默认 20，范围 1-1000

    Gemini 格式：
    - pageSize: 每页数量，默认 50，范围 1-1000
    - pageToken: 分页 token，用于获取下一页

    **返回字段**

    Claude 格式：
    - data: 模型列表，每个模型包含：
      - id: 模型标识符
      - type: "model"
      - display_name: 显示名称
      - created_at: 创建时间（ISO 8601 格式）
    - has_more: 是否有更多结果
    - first_id: 当前页第一个模型 ID
    - last_id: 当前页最后一个模型 ID

    OpenAI 格式：
    - object: "list"
    - data: 模型列表，每个模型包含：
      - id: 模型标识符
      - object: "model"
      - created: Unix 时间戳
      - owned_by: 提供商名称

    Gemini 格式：
    - models: 模型列表，每个模型包含：
      - name: 模型资源名称（如 models/gemini-pro）
      - baseModelId: 基础模型 ID
      - version: 版本号
      - displayName: 显示名称
      - description: 描述信息
      - inputTokenLimit: 输入 token 上限
      - outputTokenLimit: 输出 token 上限
      - supportedGenerationMethods: 支持的生成方法
      - temperature: 默认温度参数
      - maxTemperature: 最大温度参数
      - topP: Top-P 参数
      - topK: Top-K 参数
    - nextPageToken: 下一页的 token（如果有更多结果）

    **错误响应**
    401: API Key 无效或未提供（格式根据检测到的 API 格式返回）
    """
    adapter = PublicModelsListAdapter(
        before_id=before_id,
        after_id=after_id,
        limit=limit,
        page_size=page_size,
        page_token=page_token,
    )
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.PUBLIC)


@router.get("/v1/models/{model_id:path}", response_model=None)
async def retrieve_model(
    model_id: str,
    request: Request,
    db: Session = Depends(get_db),
) -> dict | JSONResponse:
    """
    获取单个模型详情（统一端点）

    根据请求头中的认证方式自动检测 API 格式，并返回相应格式的模型详情。
    此接口兼容 Claude、OpenAI 和 Gemini 三种 API 格式。

    **格式检测规则**
    - x-api-key + anthropic-version → Claude 格式
    - x-goog-api-key 或 ?key= → Gemini 格式
    - Authorization: Bearer → OpenAI 格式（默认）

    **路径参数**
    - model_id: 模型标识符（Gemini 格式支持 models/ 前缀，会自动移除）

    **返回字段**

    Claude 格式：
    - id: 模型标识符
    - type: "model"
    - display_name: 显示名称
    - created_at: 创建时间（ISO 8601 格式）

    OpenAI 格式：
    - id: 模型标识符
    - object: "model"
    - created: Unix 时间戳
    - owned_by: 提供商名称

    Gemini 格式：
    - name: 模型资源名称（如 models/gemini-pro）
    - baseModelId: 基础模型 ID
    - version: 版本号
    - displayName: 显示名称
    - description: 描述信息
    - inputTokenLimit: 输入 token 上限
    - outputTokenLimit: 输出 token 上限
    - supportedGenerationMethods: 支持的生成方法
    - temperature: 默认温度参数
    - maxTemperature: 最大温度参数
    - topP: Top-P 参数
    - topK: Top-K 参数

    **错误响应**
    401: API Key 无效或未提供
    404: 模型不存在或不可访问
    """
    adapter = PublicModelDetailAdapter(model_id=model_id)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.PUBLIC)


# Gemini 专用路径 /v1beta/models
@router.get("/v1beta/models", response_model=None)
async def list_models_gemini(
    request: Request,
    page_size: int = Query(50, alias="pageSize", ge=1, le=1000),
    page_token: str | None = Query(None, alias="pageToken"),
    db: Session = Depends(get_db),
) -> dict | JSONResponse:
    """
    列出可用模型（Gemini v1beta 专用端点）

    Gemini API 的专用模型列表端点，使用 x-goog-api-key 或 ?key= 参数进行认证。
    返回 Gemini 格式的模型列表。

    **查询参数**
    - pageSize: 每页数量，默认 50，范围 1-1000
    - pageToken: 分页 token，用于获取下一页

    **返回字段**
    - models: 模型列表，每个模型包含：
      - name: 模型资源名称（如 models/gemini-pro）
      - baseModelId: 基础模型 ID
      - version: 版本号
      - displayName: 显示名称
      - description: 描述信息
      - inputTokenLimit: 输入 token 上限
      - outputTokenLimit: 输出 token 上限
      - supportedGenerationMethods: 支持的生成方法列表
      - temperature: 默认温度参数
      - maxTemperature: 最大温度参数
      - topP: Top-P 参数
      - topK: Top-K 参数
    - nextPageToken: 下一页的 token（如果有更多结果）

    **错误响应**
    401: API Key 无效或未提供
    """
    adapter = PublicModelsListAdapter(
        before_id=None,
        after_id=None,
        limit=20,
        page_size=page_size,
        page_token=page_token,
    )
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.PUBLIC)


@router.get("/v1beta/models/{model_name:path}", response_model=None)
async def get_model_gemini(
    request: Request,
    model_name: str,
    db: Session = Depends(get_db),
) -> dict | JSONResponse:
    """
    获取单个模型详情（Gemini v1beta 专用端点）

    Gemini API 的专用模型详情端点，使用 x-goog-api-key 或 ?key= 参数进行认证。
    返回 Gemini 格式的模型详情。

    **路径参数**
    - model_name: 模型名称或资源路径（支持 models/ 前缀，会自动移除）

    **返回字段**
    - name: 模型资源名称（如 models/gemini-pro）
    - baseModelId: 基础模型 ID
    - version: 版本号
    - displayName: 显示名称
    - description: 描述信息
    - inputTokenLimit: 输入 token 上限
    - outputTokenLimit: 输出 token 上限
    - supportedGenerationMethods: 支持的生成方法列表
    - temperature: 默认温度参数
    - maxTemperature: 最大温度参数
    - topP: Top-P 参数
    - topK: Top-K 参数

    **错误响应**
    401: API Key 无效或未提供
    404: 模型不存在或不可访问
    """
    adapter = PublicModelDetailAdapter(model_id=model_name, force_gemini_name=True)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.PUBLIC)