config/settings.py

"""
Centralized configuration for Face Intel.

All settings are environment-driven via pydantic-settings so the same
code runs in dev, test, and production without code changes.

After the refactor, `settings` is still importable for backward
compatibility with provider modules that read tuning knobs, BUT every
stateful service (storage, cache, orchestrator, services) receives its
dependencies through constructor injection — never by importing this
module directly.
"""

from __future__ import annotations

from pathlib import Path
from typing import List

from pydantic import Field
from pydantic_settings import BaseSettings, SettingsConfigDict


BASE_DIR = Path(__file__).resolve().parent.parent
DATA_DIR = BASE_DIR / "data"
MODELS_DIR = DATA_DIR / "models"
GALLERY_DIR = DATA_DIR / "gallery"
UPLOADS_DIR = DATA_DIR / "uploads"
JOBS_DIR = DATA_DIR / "jobs"
UI_DIR = BASE_DIR / "ui" / "static"

# Ensure runtime dirs exist (idempotent)
for _d in (DATA_DIR, MODELS_DIR, GALLERY_DIR, UPLOADS_DIR, JOBS_DIR):
    _d.mkdir(parents=True, exist_ok=True)


class Settings(BaseSettings):
    """Environment-driven settings.  All fields map to FI_* env vars."""

    model_config = SettingsConfigDict(
        env_file=".env",
        env_prefix="FI_",
        case_sensitive=False,
        extra="ignore",
    )

    # ------------------------------------------------------------------ #
    # Core
    # ------------------------------------------------------------------ #
    app_name: str = "Face Intel"
    app_version: str = "1.0.0"
    environment: str = "development"
    host: str = "0.0.0.0"
    port: int = 8000
    debug: bool = False

    # ------------------------------------------------------------------ #
    # Provider enable flags
    # ------------------------------------------------------------------ #
    # Detection
    enable_haar: bool = True
    enable_dnn: bool = False          # requires model download
    # Recognition
    enable_insightface: bool = False  # requires onnxruntime + model
    # Reverse search
    enable_serpapi: bool = False      # requires API key
    enable_social_lookup: bool = True  # pure stdlib, always available
    # Image analysis
    enable_image_quality: bool = True
    enable_image_properties: bool = True
    # Metadata
    enable_exif: bool = True
    # Forensics
    enable_image_integrity: bool = True
    enable_duplicate_detector: bool = True
    enable_ela: bool = True
    enable_image_similarity: bool = True
    # OCR (optional — requires onnxruntime)
    enable_ocr: bool = False
    # Object detection (optional — requires onnxruntime)
    enable_yolov8: bool = False
    # QR code + barcode detection (pure OpenCV — always available)
    enable_qr_code: bool = True
    enable_barcode: bool = True
    # Scene classification (optional — requires onnxruntime)
    enable_places365: bool = False
    # NSFW detection (optional — requires onnxruntime)
    enable_nudenet: bool = False
    # AI image detection (optional — requires onnxruntime)
    enable_ai_image_detector: bool = False
    # Embeddings (optional — requires onnxruntime)
    enable_mobilenet_embed: bool = False

    # ------------------------------------------------------------------ #
    # Reverse face search (PimEyes-style indexed search)
    # ------------------------------------------------------------------ #
    enable_face_index: bool = True
    face_index_path: str = str(DATA_DIR / "face_index.db")
    face_index_top_k: int = 10
    face_index_threshold: float = 0.5
    face_index_max_faces: int = 100_000
    face_index_crawler_user_agent: str = "FaceIntel/1.0"
    face_index_crawler_seed_wikipedia: bool = False
    face_index_crawler_seed_imdb: bool = False

    # ------------------------------------------------------------------ #
    # Detection tuning
    # ------------------------------------------------------------------ #
    dnn_confidence_threshold: float = 0.7
    haar_scale_factor: float = 1.1
    haar_min_neighbors: int = 5

    # ------------------------------------------------------------------ #
    # Recognition tuning
    # ------------------------------------------------------------------ #
    insightface_model_pack: str = "buffalo_s"  # small variant for low-RAM
    recognition_match_threshold: float = 0.5

    # ------------------------------------------------------------------ #
    # Scraping (used by reverse-search providers)
    # ------------------------------------------------------------------ #
    scrape_timeout: int = 30
    scrape_max_images: int = 50
    user_agent: str = (
        "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 "
        "(KHTML, like Gecko) Chrome/121.0.0.0 Safari/537.36"
    )

    # ------------------------------------------------------------------ #
    # Reverse image search
    # ------------------------------------------------------------------ #
    reverse_search_max_results: int = 20
    serpapi_key: str = ""

    # ------------------------------------------------------------------ #
    # Orchestrator
    # ------------------------------------------------------------------ #
    orchestrator_timeout_seconds: float = 90.0
    orchestrator_max_concurrency: int = 8
    retry_max_attempts: int = 3
    retry_initial_backoff_seconds: float = 0.5
    retry_max_backoff_seconds: float = 8.0

    # ------------------------------------------------------------------ #
    # Cache
    # ------------------------------------------------------------------ #
    cache_enabled: bool = True
    cache_ttl_seconds: int = 3600
    cache_max_entries: int = 1000

    # ------------------------------------------------------------------ #
    # Health
    # ------------------------------------------------------------------ #
    health_check_interval_seconds: int = 60
    circuit_breaker_failure_threshold: int = 5
    circuit_breaker_recovery_seconds: int = 120

    # ------------------------------------------------------------------ #
    # Storage
    # ------------------------------------------------------------------ #
    db_path: str = str(DATA_DIR / "face_intel.db")
    audit_log_path: str = str(DATA_DIR / "audit.log")
    job_retention_days: int = 7

    # ------------------------------------------------------------------ #
    # API
    # ------------------------------------------------------------------ #
    rate_limit_per_minute: int = 30
    cors_origins: List[str] = Field(default_factory=lambda: ["*"])
    require_consent_header: bool = True
    consent_header_name: str = "X-Consent-Statement"
    max_image_bytes: int = 20 * 1024 * 1024  # 20 MB upload limit
    job_timeout_seconds: float = 300.0        # 5 min overall job cap
    max_request_body_bytes: int = 25 * 1024 * 1024  # 25 MB HTTP body cap

    # ------------------------------------------------------------------ #
    # Model management
    # ------------------------------------------------------------------ #
    models_dir: str = str(MODELS_DIR)
    models_auto_download: bool = True  # auto-download on first use
    onnx_intra_op_threads: int = 2     # limit CPU threads for low-RAM

    # ------------------------------------------------------------------ #
    # Logging
    # ------------------------------------------------------------------ #
    log_level: str = "INFO"
    log_json: bool = False


def make_settings(**overrides) -> Settings:
    """Factory used by the DI container to build a Settings instance.

    Tests can override individual fields without touching env vars:
        make_settings(environment="test", cache_enabled=False)
    """
    if overrides:
        # Build a fresh instance with overrides merged
        base = Settings()
        data = base.model_dump()
        data.update(overrides)
        return Settings(**data)
    return Settings()


# Default singleton — fine for read-only config consumption by providers.
# Stateful services MUST be injected, not import this directly.
settings = Settings()