"""群聊人设卡 —— 采集编排(Service 层)。

把 __init__.py 里的采样/复读/刷屏检测与落库编排抽到此处,使入口更薄。
"""

from __future__ import annotations

import json
import time
from collections import deque

from nonebot import get_driver

from .. import processor, repository

# 采样/复读检测的内存状态（不进库）
_recent_texts: dict[int, deque[str]] = {}  # group_id -> 最近消息文本
_last_store: dict[tuple[int, int], float] = {}  # (group_id, user_id) -> 上次落库时间戳


def _is_repeat(group_id: int, content: str) -> bool:
    """复读检测：内容在最近 N 条里出现过则丢弃"""
    queue = _recent_texts.setdefault(group_id, deque(maxlen=processor.REPEAT_WINDOW))
    if content in queue:
        return True
    queue.append(content)
    return False


def _throttled(group_id: int, user_id: int) -> bool:
    """刷屏采样：5 秒内同一个人只落库一次"""
    key = (group_id, user_id)
    now = time.time()
    if now - _last_store.get(key, 0) < processor.THROTTLE_SECONDS:
        return True
    _last_store[key] = now
    return False


async def collect(event):
    """鉴权 + 治理 + 采样 + 落库。"""
    if not await repository.is_group_enabled(event.group_id):
        return
    if not await repository.is_joined(event.user_id, event.group_id):
        return
    if event.user_id == event.self_id:
        return

    command_starts = get_driver().config.command_start
    if not processor.has_recordable_content(event):
        return
    text = processor.filter_content(
        processor.extract_text(event), command_starts=command_starts
    )
    image_hashes = processor.extract_image_hashes(event)
    content = processor.build_content(
        text or "", len(image_hashes), processor.count_faces(event)
    )
    if content is None:
        return
    context = " ".join(_recent_texts.get(event.group_id, ()))
    content = processor.desensitize(content, context)
    if _is_repeat(event.group_id, content):
        return
    if _throttled(event.group_id, event.user_id):
        return

    await repository.add_chat_log(
        user_id=event.user_id,
        group_id=event.group_id,
        content=content,
        nickname=event.sender.card or event.sender.nickname or "",
        target_user_id=processor.extract_target_user_id(event),
        reply_to_content=processor.extract_reply_to_content(event),
        image_count=len(image_hashes),
        image_hashes=json.dumps(image_hashes, ensure_ascii=False) if image_hashes else None,
    )