from functools import lru_cache
from pathlib import Path
from typing import Literal

from pydantic_settings import BaseSettings, SettingsConfigDict

# Repo-root .env for native dev; inside Docker the file is absent and
# configuration comes from real environment variables (which take precedence).
_REPO_ROOT = Path(__file__).resolve().parents[2]
_ENV_FILE = _REPO_ROOT / ".env"


class Settings(BaseSettings):
    model_config = SettingsConfigDict(
        env_prefix="PABLAN_", env_file=_ENV_FILE, extra="ignore"
    )

    env: Literal["development", "production"] = "development"
    database_url: str = "postgresql+asyncpg://pablan:change-me@localhost:5432/pablan"

    # Secure=false is needed for dev over plain http on non-localhost
    # addresses (WireGuard IPs) — see .env.example.
    cookie_secure: bool = True
    auth_session_ttl_days: int = 14

    query_retention_days: int = 90

    # The shipped template catalog (repo templates/ in dev; the customer
    # stack mounts the directory and overrides this path).
    templates_dir: str = str(_REPO_ROOT / "templates")
    help_dir: str = str(_REPO_ROOT / "help")

    # The instance's own language: which blueprint variant the first-install
    # starter set uses, and the fallback when a visitor states no preference.
    # Per-user choice lives on users.locale and wins over this.
    default_locale: Literal["de", "en"] = "de"

    log_level: str = "INFO"
    # Content debug logging (prompts/responses) — NEVER in production.
    debug_log_prompts: bool = False
    llm_timeout_seconds: float = 120.0
    # How many requests Pablan lets one endpoint see at once, and how long a
    # request waits for a free slot before it is answered with "busy". Match
    # llm_max_parallel to the server's parallel slots (llama.cpp: --parallel).
    # See app/llm/gate.py.
    llm_max_parallel: int = 4
    llm_queue_wait_seconds: float = 20.0
    llm_max_queued: int = 24
    job_poll_seconds: float = 1.0

    chat_base_url: str = "http://localhost:8001/v1"
    chat_api_key: str = "none"
    chat_model: str = ""
    utility_base_url: str = "http://localhost:8001/v1"
    utility_api_key: str = "none"
    utility_model: str = ""
    embedding_base_url: str = "http://localhost:8002/v1"
    embedding_api_key: str = "none"
    embedding_model: str = ""


@lru_cache
def get_settings() -> Settings:
    return Settings()