from __future__ import annotations

import os
from dataclasses import dataclass, field
from pathlib import Path

from dotenv import load_dotenv


BACKEND_ROOT = Path(__file__).resolve().parents[1]
DEFAULT_ENV_FILE = BACKEND_ROOT / ".env"


def _as_bool(value: str | None, default: bool) -> bool:
    if value is None:
        return default
    normalized = value.strip().lower()
    if normalized in {"1", "true", "yes", "on"}:
        return True
    if normalized in {"0", "false", "no", "off"}:
        return False
    raise ValueError(f"Invalid boolean configuration value: {value!r}")


def _as_int(name: str, value: str | None, default: int) -> int:
    if value is None:
        return default
    parsed = int(value)
    if parsed < 0:
        raise ValueError(f"{name} must be non-negative")
    return parsed


def _as_float(name: str, value: str | None, default: float) -> float:
    if value is None:
        return default
    parsed = float(value)
    if parsed <= 0:
        raise ValueError(f"{name} must be positive")
    return parsed


@dataclass(frozen=True, slots=True)
class Settings:
    """Runtime settings with the API key deliberately hidden from repr output."""

    llm_provider: str = "auto"
    openai_api_key: str | None = field(default=None, repr=False)
    openai_base_url: str | None = None
    openai_model: str = "gpt-4o-mini"
    embedding_provider: str = "tei"
    embedding_base_url: str = "http://127.0.0.1:8081"
    embedding_model: str = "BAAI/bge-m3"
    embedding_dimensions: int = 1024
    embedding_timeout_seconds: float = 30.0
    embedding_batch_size: int = 32
    openai_timeout_seconds: float = 30.0
    openai_max_retries: int = 2
    light_opt_rate_limit: int = 20
    light_opt_rate_window_seconds: float = 3600.0
    structured_output_retries: int = 1
    structured_output_mode: str = "json_schema"
    fallback_to_rules: bool = True
    intent_router_mode: str = "off"
    intent_model: str | None = None
    knowledge_admin_token: str | None = field(default=None, repr=False)
    database_url: str | None = field(default=None, repr=False)
    deep_max_questions: int = 6
    deep_min_questions: int = 2
    deep_gap_threshold: float = 5.0

    @property
    def use_openai(self) -> bool:
        if self.llm_provider == "openai":
            if not self.openai_api_key:
                raise ValueError("OPENAI_API_KEY is required when LLM provider is openai")
            return True
        if self.llm_provider == "volcengine":
            if not self.openai_api_key:
                raise ValueError(
                    "VOLCENGINE_API_KEY is required when LLM provider is volcengine"
                )
            return True
        if self.llm_provider == "rule":
            return False
        if self.llm_provider != "auto":
            raise ValueError(
                "RESUME_AGENT_LLM_PROVIDER must be auto, openai, volcengine, or rule"
            )
        return bool(self.openai_api_key)


def load_settings(env_file: str | Path | None = None) -> Settings:
    selected_file = Path(
        env_file or os.getenv("RESUME_AGENT_ENV_FILE", str(DEFAULT_ENV_FILE))
    )
    load_dotenv(selected_file, override=False)
    mode = os.getenv("OPENAI_STRUCTURED_OUTPUT_MODE", "json_schema").strip().lower()
    if mode not in {"json_schema", "json_object"}:
        raise ValueError(
            "OPENAI_STRUCTURED_OUTPUT_MODE must be json_schema or json_object"
        )
    provider = os.getenv("RESUME_AGENT_LLM_PROVIDER", "auto").strip().lower()
    if provider == "volcengine":
        llm_api_key = os.getenv("VOLCENGINE_API_KEY") or None
        llm_base_url = os.getenv("VOLCENGINE_BASE_URL") or None
        llm_model = os.getenv("VOLCENGINE_MODEL", "").strip()
    else:
        llm_api_key = os.getenv("OPENAI_API_KEY") or None
        llm_base_url = os.getenv("OPENAI_BASE_URL") or None
        llm_model = os.getenv("OPENAI_MODEL", "gpt-4o-mini").strip()
    intent_mode = os.getenv("RESUME_AGENT_INTENT_ROUTER_MODE", "off").strip().lower()
    if intent_mode not in {"off", "shadow", "on"}:
        raise ValueError("RESUME_AGENT_INTENT_ROUTER_MODE must be off, shadow, or on")
    settings = Settings(
        llm_provider=provider,
        openai_api_key=llm_api_key,
        openai_base_url=llm_base_url,
        openai_model=llm_model,
        embedding_provider=os.getenv("EMBEDDING_PROVIDER", "tei").strip().lower(),
        embedding_base_url=os.getenv("EMBEDDING_BASE_URL", "http://127.0.0.1:8081").rstrip("/"),
        embedding_model=os.getenv("EMBEDDING_MODEL", "BAAI/bge-m3").strip(),
        embedding_dimensions=_as_int(
            "EMBEDDING_DIMENSIONS", os.getenv("EMBEDDING_DIMENSIONS"), 1024
        ),
        embedding_timeout_seconds=_as_float(
            "EMBEDDING_TIMEOUT_SECONDS", os.getenv("EMBEDDING_TIMEOUT_SECONDS"), 30.0
        ),
        embedding_batch_size=_as_int(
            "EMBEDDING_BATCH_SIZE", os.getenv("EMBEDDING_BATCH_SIZE"), 32
        ),
        openai_timeout_seconds=_as_float(
            "OPENAI_TIMEOUT_SECONDS", os.getenv("OPENAI_TIMEOUT_SECONDS"), 30.0
        ),
        openai_max_retries=_as_int(
            "OPENAI_MAX_RETRIES", os.getenv("OPENAI_MAX_RETRIES"), 2
        ),
        light_opt_rate_limit=_as_int(
            "RESUME_AGENT_LIGHT_OPT_RATE_LIMIT",
            os.getenv("RESUME_AGENT_LIGHT_OPT_RATE_LIMIT"),
            20,
        ),
        light_opt_rate_window_seconds=_as_float(
            "RESUME_AGENT_LIGHT_OPT_RATE_WINDOW_SECONDS",
            os.getenv("RESUME_AGENT_LIGHT_OPT_RATE_WINDOW_SECONDS"),
            3600.0,
        ),
        structured_output_retries=_as_int(
            "OPENAI_STRUCTURED_OUTPUT_RETRIES",
            os.getenv("OPENAI_STRUCTURED_OUTPUT_RETRIES"),
            1,
        ),
        structured_output_mode=mode,
        fallback_to_rules=_as_bool(
            os.getenv("RESUME_AGENT_LLM_FALLBACK_TO_RULES"), True
        ),
        intent_router_mode=intent_mode,
        intent_model=os.getenv("RESUME_AGENT_INTENT_MODEL", "").strip() or None,
        knowledge_admin_token=os.getenv("KNOWLEDGE_ADMIN_TOKEN") or None,
        database_url=os.getenv("DATABASE_URL") or None,
        deep_max_questions=_as_int(
            "RESUME_AGENT_DEEP_MAX_QUESTIONS",
            os.getenv("RESUME_AGENT_DEEP_MAX_QUESTIONS"),
            6,
        ),
        deep_min_questions=_as_int(
            "RESUME_AGENT_DEEP_MIN_QUESTIONS",
            os.getenv("RESUME_AGENT_DEEP_MIN_QUESTIONS"),
            2,
        ),
        deep_gap_threshold=_as_float(
            "RESUME_AGENT_DEEP_GAP_THRESHOLD",
            os.getenv("RESUME_AGENT_DEEP_GAP_THRESHOLD"),
            5.0,
        ),
    )
    if settings.embedding_provider not in {"tei", "openai", "hash"}:
        raise ValueError("EMBEDDING_PROVIDER must be tei, openai, or hash")
    if not settings.embedding_model:
        raise ValueError("EMBEDDING_MODEL cannot be blank")
    if settings.embedding_provider == "tei" and not settings.embedding_base_url:
        raise ValueError("EMBEDDING_BASE_URL cannot be blank when EMBEDDING_PROVIDER is tei")
    if settings.embedding_dimensions < 1:
        raise ValueError("EMBEDDING_DIMENSIONS must be positive")
    if settings.embedding_batch_size < 1:
        raise ValueError("EMBEDDING_BATCH_SIZE must be positive")
    if settings.deep_max_questions < settings.deep_min_questions:
        raise ValueError(
            "RESUME_AGENT_DEEP_MAX_QUESTIONS must be at least "
            "RESUME_AGENT_DEEP_MIN_QUESTIONS"
        )
    if settings.deep_max_questions < 1:
        raise ValueError("RESUME_AGENT_DEEP_MAX_QUESTIONS must be positive")
    if not settings.openai_model:
        model_setting = "VOLCENGINE_MODEL" if provider == "volcengine" else "OPENAI_MODEL"
        raise ValueError(f"{model_setting} cannot be blank")
    return settings