"""
Track C2 — Translation cache with Redis backend.

This module provides a translation cache that:
  - Defaults to in-process LRU (same as the previous behavior)
  - Optionally uses Redis when REDIS_CACHE_ENABLED=true
  - Namespaces keys by user_id (so free and pro users don't share)
  - Includes custom_prompt_hash + glossary_hash in the key (so two
    users with different prompts/glossaries don't get cross-contamination)
  - Falls back to LRU if Redis is unavailable (transient connection
    error, etc.) — a Redis outage MUST NOT break translation jobs

Public API:
  TranslationCache — single instance per process, async-friendly.
  get_cache()       — singleton accessor.

Usage:
    cache = get_cache()
    cached = await cache.get(text, target, source, provider, user_id, custom_prompt_hash, glossary_hash)
    if cached is None:
        translated = await provider.translate(text, target, source)
        await cache.set(text, target, source, provider, user_id, custom_prompt_hash, glossary_hash, translated)
"""

from __future__ import annotations

import hashlib
import json
import os
import time
import asyncio
import threading
from collections import OrderedDict
from typing import Optional, Dict, List

from core.logging import get_logger

logger = get_logger(__name__)


# =============================================================================
# Key construction
# =============================================================================

def make_cache_key(
    text: str,
    target_language: str,
    source_language: str,
    provider: str,
    user_id: Optional[str] = None,
    custom_prompt_hash: Optional[str] = None,
    glossary_hash: Optional[str] = None,
) -> str:
    """Build a stable cache key.

    The key includes:
      - The text being translated
      - The language pair
      - The provider
      - The user (so free/pro users don't collide)
      - The custom_prompt hash (so different prompts are isolated)
      - The glossary hash (so different glossaries are isolated)
    """
    parts = [
        f"provider={provider}",
        f"src={source_language or 'auto'}",
        f"tgt={target_language}",
        f"user={user_id or 'anon'}",
        f"prompt={custom_prompt_hash or 'none'}",
        f"glossary={glossary_hash or 'none'}",
        f"text={text}",
    ]
    content = "|".join(parts)
    return hashlib.sha256(content.encode("utf-8")).hexdigest()


def hash_prompt(prompt: Optional[str]) -> Optional[str]:
    """Hash a custom prompt for use in a cache key. Returns None for None/empty."""
    if not prompt:
        return None
    return hashlib.sha256(prompt.encode("utf-8")).hexdigest()[:16]


def hash_glossary_terms(terms: Optional[List[dict]]) -> Optional[str]:
    """Hash a glossary's term list for use in a cache key.

    We only hash the source/target pairs (not metadata) so that two
    glossaries with the same content but different names still share
    cache entries.
    """
    if not terms:
        return None
    # Stable representation: list of (source, target) tuples
    normalized = sorted(
        (str(t.get("source", "")), str(t.get("target", "")))
        for t in terms
        if t.get("source") and t.get("target")
    )
    content = json.dumps(normalized, ensure_ascii=False, sort_keys=True)
    return hashlib.sha256(content.encode("utf-8")).hexdigest()[:16]


# =============================================================================
# In-process LRU fallback
# =============================================================================

class _LRUBackend:
    """Thread-safe LRU cache, used as a fallback when Redis is unavailable."""

    def __init__(self, maxsize: int = 5000):
        self._cache: "OrderedDict[str, str]" = OrderedDict()
        self._maxsize = maxsize
        self._lock = threading.RLock()
        self.hits = 0
        self.misses = 0

    def get(self, key: str) -> Optional[str]:
        with self._lock:
            if key in self._cache:
                self.hits += 1
                self._cache.move_to_end(key)
                return self._cache[key]
            self.misses += 1
            return None

    def set(self, key: str, value: str, ttl: Optional[int] = None) -> None:
        # LRU doesn't honor TTL — it's a process-lifetime cache.
        with self._lock:
            if key in self._cache:
                self._cache.move_to_end(key)
            self._cache[key] = value
            while len(self._cache) > self._maxsize:
                self._cache.popitem(last=False)

    def clear(self) -> None:
        with self._lock:
            self._cache.clear()
            self.hits = 0
            self.misses = 0

    def stats(self) -> Dict:
        with self._lock:
            total = self.hits + self.misses
            hit_rate = (self.hits / total * 100) if total > 0 else 0
            return {
                "backend": "lru",
                "size": len(self._cache),
                "maxsize": self._maxsize,
                "hits": self.hits,
                "misses": self.misses,
                "hit_rate": f"{hit_rate:.1f}%",
            }


# =============================================================================
# Redis backend
# =============================================================================

class _RedisBackend:
    """Redis-backed cache with TTL and async support.

    Connection is lazy: we don't connect on import. If Redis is
    unreachable, every operation falls back to the LRU backend.
    """

    def __init__(
        self,
        url: str,
        namespace: str = "translate_cache",
        default_ttl: int = 86400,           # 24 hours
        lru_fallback: Optional[_LRUBackend] = None,
        socket_timeout: float = 2.0,
    ):
        self._url = url
        self._namespace = namespace
        self._default_ttl = default_ttl
        self._lru_fallback = lru_fallback
        self._socket_timeout = socket_timeout
        self._client = None
        self._lock = threading.RLock()
        self.hits = 0
        self.misses = 0
        self.errors = 0
        self._available = None  # None = unknown, True/False = probed

    def _get_client(self):
        """Lazy client init. Never raises; returns None on failure."""
        if self._client is not None:
            return self._client
        with self._lock:
            if self._client is not None:
                return self._client
            try:
                import redis
                client = redis.from_url(
                    self._url,
                    socket_timeout=self._socket_timeout,
                    socket_connect_timeout=self._socket_timeout,
                    decode_responses=True,
                )
                # Probe once
                client.ping()
                self._client = client
                self._available = True
                logger.info("redis_cache_connected", url=self._url.split("@")[-1])
                return client
            except Exception as e:
                self._available = False
                self.errors += 1
                logger.warning(
                    "redis_cache_unavailable",
                    error=str(e)[:200],
                    fallback="lru",
                )
                return None

    def _key(self, key: str) -> str:
        return f"{self._namespace}:{key}"

    def get(self, key: str) -> Optional[str]:
        client = self._get_client()
        if client is None:
            # Fall back to LRU
            if self._lru_fallback is not None:
                v = self._lru_fallback.get(key)
                if v is not None:
                    self.hits += 1
                else:
                    self.misses += 1
                return v
            return None
        try:
            v = client.get(self._key(key))
            if v is not None:
                self.hits += 1
                # Promote to LRU too (in-process L1 cache)
                if self._lru_fallback is not None:
                    self._lru_fallback.set(key, v)
                return v
            self.misses += 1
            return None
        except Exception as e:
            self.errors += 1
            self._available = False
            self._client = None  # force reconnect on next call
            logger.warning("redis_cache_get_error", error=str(e)[:200])
            if self._lru_fallback is not None:
                return self._lru_fallback.get(key)
            return None

    def set(self, key: str, value: str, ttl: Optional[int] = None) -> None:
        client = self._get_client()
        if client is not None:
            try:
                client.set(
                    self._key(key),
                    value,
                    ex=ttl if ttl is not None else self._default_ttl,
                )
                # Also write to LRU
                if self._lru_fallback is not None:
                    self._lru_fallback.set(key, value)
                return
            except Exception as e:
                self.errors += 1
                self._available = False
                self._client = None
                logger.warning("redis_cache_set_error", error=str(e)[:200])
        # Fall through to LRU
        if self._lru_fallback is not None:
            self._lru_fallback.set(key, value, ttl=ttl)

    def clear(self) -> None:
        if self._lru_fallback is not None:
            self._lru_fallback.clear()
        client = self._get_client()
        if client is not None:
            try:
                # Delete only keys in our namespace
                cursor = 0
                pattern = f"{self._namespace}:*"
                while True:
                    cursor, keys = client.scan(cursor=cursor, match=pattern, count=500)
                    if keys:
                        client.delete(*keys)
                    if cursor == 0:
                        break
            except Exception as e:
                logger.warning("redis_cache_clear_error", error=str(e)[:200])

    def stats(self) -> Dict:
        out = {
            "backend": "redis",
            "available": self._available,
            "url": self._url.split("@")[-1] if self._url else None,
            "namespace": self._namespace,
            "default_ttl": self._default_ttl,
            "hits": self.hits,
            "misses": self.misses,
            "errors": self.errors,
        }
        if self._lru_fallback is not None:
            lru_stats = self._lru_fallback.stats()
            out["lru_size"] = lru_stats["size"]
        return out


# =============================================================================
# TranslationCache — public API
# =============================================================================

class TranslationCache:
    """High-level translation cache.

    Delegates to a backend (Redis or LRU) based on configuration.
    Provides an async-friendly API (sync methods, but no blocking
    Redis I/O — Redis client is async-compatible via redis.asyncio
    if we want to add that later).
    """

    def __init__(
        self,
        backend: str = "auto",
        redis_url: Optional[str] = None,
        redis_namespace: str = "translate_cache",
        default_ttl: int = 86400,
        lru_maxsize: int = 5000,
    ):
        # Decide backend
        if backend == "auto":
            backend = "redis" if (
                os.getenv("REDIS_CACHE_ENABLED", "false").lower() == "true"
                and (redis_url or os.getenv("REDIS_URL", ""))
            ) else "lru"

        self._lru = _LRUBackend(maxsize=lru_maxsize)
        if backend == "redis" and (redis_url or os.getenv("REDIS_URL", "")):
            self._backend: object = _RedisBackend(
                url=redis_url or os.getenv("REDIS_URL", ""),
                namespace=redis_namespace,
                default_ttl=default_ttl,
                lru_fallback=self._lru,
            )
            self._backend_name = "redis"
        else:
            self._backend = self._lru
            self._backend_name = "lru"

    @property
    def backend_name(self) -> str:
        return self._backend_name

    def get(
        self,
        text: str,
        target_language: str,
        source_language: str,
        provider: str,
        user_id: Optional[str] = None,
        custom_prompt_hash: Optional[str] = None,
        glossary_hash: Optional[str] = None,
    ) -> Optional[str]:
        key = make_cache_key(
            text, target_language, source_language, provider,
            user_id, custom_prompt_hash, glossary_hash,
        )
        return self._backend.get(key)

    def set(
        self,
        text: str,
        target_language: str,
        source_language: str,
        provider: str,
        translation: str,
        user_id: Optional[str] = None,
        custom_prompt_hash: Optional[str] = None,
        glossary_hash: Optional[str] = None,
        ttl: Optional[int] = None,
    ) -> None:
        key = make_cache_key(
            text, target_language, source_language, provider,
            user_id, custom_prompt_hash, glossary_hash,
        )
        self._backend.set(key, translation, ttl=ttl)

    def clear(self) -> None:
        self._backend.clear()

    def stats(self) -> Dict:
        return self._backend.stats()


# =============================================================================
# Singleton accessor
# =============================================================================

_cache: Optional[TranslationCache] = None
_cache_lock = threading.RLock()


def get_cache() -> TranslationCache:
    """Get the process-wide translation cache (lazy init)."""
    global _cache
    if _cache is not None:
        return _cache
    with _cache_lock:
        if _cache is None:
            _cache = TranslationCache(
                backend="auto",
                redis_url=os.getenv("REDIS_URL", "") or None,
                redis_namespace=os.getenv("REDIS_CACHE_NAMESPACE", "translate_cache"),
                default_ttl=int(os.getenv("REDIS_CACHE_TTL", "86400")),
                lru_maxsize=int(os.getenv("LRU_CACHE_MAXSIZE", "5000")),
            )
            logger.info(
                "translation_cache_initialized",
                backend=_cache.backend_name,
            )
    return _cache


def reset_cache_for_tests() -> None:
    """Reset the singleton — only for tests."""
    global _cache
    with _cache_lock:
        _cache = None