__init__.py

import base64
from random import random
import re
import time
import asyncio
from collections import defaultdict
from PIL import Image, ImageDraw, ImageFont, ImageColor
import html

import hoshino
from hoshino import Service, R, aiorequests
from hoshino.typing import *
from hoshino.util import FreqLimiter, concat_pic, pic2b64, silence

from .. import chara
from .. import _pcr_data
from .record import update_dic, update_record
from os.path import dirname, join, exists
from os import remove, listdir
import numpy as np
import json
from io import BytesIO
import requests
import copy
from math import log

import cv2

sv_help = '''
[怎么拆] 接防守队角色名 查询竞技场解法
[点赞] 接作业id 评价作业
[点踩] 接作业id 评价作业
'''.strip()
sv = Service('pcr-arena', help_=sv_help, bundle='pcr查询')

from . import arena

lmt = FreqLimiter(5)

aliases = ('怎么拆', '怎么解', '怎么打', '如何拆', '如何解', '如何打', 'jjc查询')
aliases_b = tuple('b' + a for a in aliases) + tuple('B' + a for a in aliases)
aliases_b = list(aliases_b)
aliases_b.append('bjjc')
aliases_b = tuple(aliases_b)

aliases_tw = tuple('台' + a for a in aliases)
aliases_tw = list(aliases_tw)
aliases_tw.append('台jjc')
aliases_tw.append('tjjc')
aliases_tw = tuple(aliases_tw)

aliases_jp = tuple('日' + a for a in aliases)
aliases_jp = list(aliases_jp)
aliases_jp.append('日jjc')
aliases_jp.append('rjjc')
aliases_jp = tuple(aliases_jp)

try:
    thumb_up_i = R.img('priconne/gadget/thumb-up-i.png').open().resize((16, 16), Image.LANCZOS)
    thumb_up_a = R.img('priconne/gadget/thumb-up-a.png').open().resize((16, 16), Image.LANCZOS)
    thumb_down_i = R.img('priconne/gadget/thumb-down-i.png').open().resize((16, 16), Image.LANCZOS)
    thumb_down_a = R.img('priconne/gadget/thumb-down-a.png').open().resize((16, 16), Image.LANCZOS)
except Exception as e:
    sv.logger.exception(e)


# 全服=1 b服=2 台服=3 日服=4
@sv.on_prefix(aliases)
async def arena_query(bot, ev):
    await _arena_query(bot, ev, region=1)


@sv.on_prefix(aliases_b)
async def arena_query_b(bot, ev):
    await _arena_query(bot, ev, region=2)


@sv.on_prefix(aliases_tw)
async def arena_query_tw(bot, ev):
    await _arena_query(bot, ev, region=3)


@sv.on_prefix(aliases_jp)
async def arena_query_jp(bot, ev):
    await _arena_query(bot, ev, region=4)


@sv.on_prefix(('testjjc'))
async def arena_query_test(bot, ev):
    await _arena_query(bot, ev, region=-20)


async def render_atk_def_teams(entries, border_pix=5):
    '''
    entries = [ {'atk': [int], 'up': int, 'down': int } ]
    '''
    n = len(entries)
    icon_size = 64
    small_icon_size = 32
    im = Image.new('RGBA', (5 * icon_size + 100, n * (icon_size + border_pix) - border_pix), (255, 255, 255, 255))
    font = ImageFont.truetype('msyh.ttc', 16)
    draw = ImageDraw.Draw(im)
    for i, e in enumerate(entries):
        if len(e) == 0:  # [] 视为空行
            continue

        y1 = i * (icon_size + border_pix)
        y2 = y1 + icon_size

        if e == "placeholder":  # 输出五个佑树
            e = {'atk': [chara.fromid(9000) for _ in range(5)], "team_type": "youshu"}

        # e此时只能是dict了
        for j, c in enumerate(e['atk']):
            x1 = j * icon_size
            x2 = x1 + icon_size
            try:
                icon = await c.render_icon(icon_size)  # 如使用旧版hoshino（不返回结果），请去掉await
                im.paste(icon, (x1, y1, x2, y2), icon)
            except:
                icon = c.render_icon(icon_size)
                im.paste(icon, (x1, y1, x2, y2), icon)

        x1 = 5 * icon_size + 10
        if e["team_type"] == "normal":
            x2 = x1 + 16
            try:
                im.paste(thumb_up_a, (x1, y1 + 12, x2, y1 + 28), thumb_up_a)
            except:
                draw.text((x1, y1 + 10), f"赞", (0, 0, 0, 255), font)
            try:
                im.paste(thumb_down_a, (x1, y1 + 39, x2, y1 + 55), thumb_down_a)
            except:
                draw.text((x1, y1 + 35), f"踩", (0, 0, 0, 255), font)
            draw.text((x1 + 25, y1 + 10), f"{e['up']}", (0, 0, 0, 255), font)
            draw.text((x1 + 25, y1 + 35), f"{e['down']}", (0, 0, 0, 255), font)
        elif e["team_type"] == "approximation":
            draw.text((x1, y1 + 22), f"近似解", (0, 0, 0, 255), font)
        elif "approximation" in e["team_type"]:
            _, uid_4_1_str, uid_4_2_str = e["team_type"].split(' ')
            draw.text((x1, y1 - 3), f"近似解", (0, 0, 0, 255), font)

            chara_1 = chara.fromid(int(uid_4_1_str))
            icon_1 = await chara_1.render_icon(small_icon_size)
            im.paste(icon_1, (x1, y1 + 26), icon_1)

            draw.text((x1 + 33, y1 + 32), f"→", (0, 0, 0, 255), font)

            chara_2 = chara.fromid(int(uid_4_2_str))
            icon_2 = await chara_2.render_icon(small_icon_size)
            im.paste(icon_2, (x1 + 50, y1 + 26), icon_2)
        elif e["team_type"] == "frequency":
            draw.text((x1, y1 + 22), f"高频解", (0, 0, 0, 255), font)

    return im


async def getBox(img):
    return await getPos(img)


curpath = dirname(__file__)

dataDir = join(curpath, 'dic.npy')
if not exists(dataDir):
    update_dic()
data = np.load(dataDir, allow_pickle=True).item()
data_processed = None

best_atk_records_path = join(curpath, 'buffer/best_atk_records.json')
if not exists(best_atk_records_path):
    update_record()
with open(best_atk_records_path, "r", encoding="utf-8") as fp:
    best_atk_records = json.load(fp)


async def cut_image(image, hash_size=16):
    '''
    将图像缩小成(16+1)*16并转化成灰度图
    :param image: PIL.Image
    :return list[int]
    '''

    image1 = image.resize((hash_size + 1, hash_size), Image.ANTIALIAS).convert('L')
    pixel = list(image1.getdata())
    return pixel


async def trans_hash(lists):
    '''
    比较列表中相邻元素大小
    :param lists: list[int]
    :return list[bool]
    '''
    return [1 if lists[index - 1] > val else 0 for index, val in enumerate(lists)][1:]


async def difference_value(image_lists):
    # 获得图像差异值并获得指纹
    assert len(image_lists) == 17 * 16, "size error"
    m, n = 0, 17
    hash_list = []
    for i in range(0, 16):
        slc = slice(m, n)
        image_slc = image_lists[slc]
        hash_list.append(await trans_hash(image_slc))
        m += 17
        n += 17
    return hash_list


async def get_hash_arr(image):
    return np.array(await difference_value(await cut_image(image)))


async def calc_distance_arr(arr1, arr2):
    return sum(sum(abs(arr1 - arr2)))


async def calc_distance_img(image1, image2):
    return await calc_distance_arr(await get_hash_arr(image2) - await get_hash_arr(image1))


async def process_data():
    global data, data_processed
    data_processed = {}
    for uid in data:
        data_processed[uid] = await get_hash_arr(Image.fromarray(data[uid][25:96, 8:97, :]))


async def cutting(img, mode):
    '''
    :param img: 传入的待识别的原图片 PIL格式
    :param mode: mode=1：返回图片中最大的长方形区域，以及该区域的定位点 [PIL.Image, [x, y, w, h]] ; mode=2: 返回两个列表，列表中每个元素为正方形区域在原图的[x, y, w, h]。第一个列表为聚类结果，第二个列表为排除结果。若无正方形区域，返回[], []。
    '''
    im_grey = img.convert('L')
    totArea = (im_grey.size)[0] * (im_grey.size)[1]
    im_grey = im_grey.point(lambda x: 255 if x > 210 else 0)  # 没有用自带的二值化。考虑修改，使用更合适的函数。
    thresh = np.array(im_grey)

    # cv2.findContours. opencv3版本会返回3个值，opencv2和4只返回后两个值
    #img2, contours, hier = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
    contours, hier = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)  # 获取轮廓
    img2 = thresh

    lis = []
    icon = []  # 每个元素为：[边长，[矩阵左上点坐标，矩阵宽高]]
    # icon_area = {}  # key为边长 value为个数
    for i in range(len(contours)):
        area = cv2.contourArea(contours[i])  # 计算contour包围的像素点个数
        lis.append(area)
        if area > 500:
            if mode == 2:
                x, y, w, h = cv2.boundingRect(contours[i])  # 获取contour的aabb包围盒
                if w / h > 0.95 and w / h < 1.05:  # 近似正方形
                    are = (w + h) // 2
                    areaRatio = are * are / totArea * 100  # 获取该范围占整个输入图像的占比
                    # print(f"{areaRatio:2f}%")
                    if areaRatio >= 0.5:  # 过滤占比小于0.5%的正方形（可能为文字）
                        icon.append([are, [x, y, w, h]])
                        # icon_area[are] = icon_area.setdefault(are, 0) + 1
    # print()
    if mode == 1:
        i = lis.index(max(lis))
        x, y, w, h = cv2.boundingRect(contours[i])
        #cv2.rectangle(img2, (x, y), (x + w, y + h), (153, 153, 0), 5)
        img3 = img2[y + 2:y + h - 2, x + 2:x + w - 2]
        img4 = Image.fromarray(img3)
        return img4, [x, y, w, h]
    if mode == 2:
        if len(icon) == 0:
            return [], []

        # 对边长作简易聚类分析
        kinds = {}  # 边长: [[],[],[],...,[]]
        for i in icon:
            sidelen = i[0]
            category = -1
            for kind in kinds:
                ratio = sidelen / kind
                if ratio > 0.9 and ratio < 1.1:  # 将边长相差10%以内作为一类
                    category = kind
                    kinds[kind].append(i[1])
                    break
            if category == -1:
                kinds[sidelen] = [i[1]]

        def clusterWeight(x):
            sidelen = x[0]
            val = len(x[1])
            if val == 5:  # 该类有5个元素，优先返回。第二关键字为边长（从大到小）。
                return 5000000 + sidelen
            if val % 5 == 0:  # 该类元素个数为5的倍数，次优先返回。
                return 1000000 + sidelen
            return val * 10000 + sidelen

        kinds = sorted(kinds.items(), key=clusterWeight, reverse=True)
        kind = kinds[0]  # (边长, [[],[]])
        if len(kind[1]) % 5 == 0:  # 存在某一类，其元素个数为5的倍数，返回该类
            otherborder = []
            for x in range(1, len(kinds)):
                otherborder += kinds[x][1]
            return kind[1], otherborder
        else:
            return [x[1] for x in icon], []  # 否则返回所有找到的正方形区域，由后续程序进一步判断


async def cut(img, border):
    '''
    :param img: 待裁剪图片 PIL.Image
    :param border: 裁剪范围 [x, y, w, h]
    :return 裁剪后图像 PIL.Image
    '''
    x, y, w, h = border
    img = np.array(img)
    img = img[y + 2:y + h - 2, x + 2:x + w - 2]
    img = Image.fromarray(img)
    return img


async def getPos(img: Image):
    '''
    :param img: 待识别图片 PIL.Image
    :return 识别出的阵容的坐标, 识别出的阵容角色字符串 [[第1队第1个角色(int), 1-2, ..., 1-5], [2-1, ..., 2-5], [3-1, ..., 3-5]], str；若识别失败，返回[], ""
    '''
    img = img.convert("RGBA")
    im_grey = img

    actual_img = img
    actual_x = 0
    actual_y = 0

    nowcolor = 0
    outpImg = Image.new(mode="RGBA", size=img.size, color=ImageColor.getrgb(f'rgb({nowcolor},{nowcolor},{nowcolor})'))
    outpImgText = Image.new("RGBA", img.size, (0, 0, 0, 0))
    outpImgTextDraw = ImageDraw.Draw(outpImgText)

    cnt = 0
    while cnt <= 6:
        bo = False
        cnt += 1
        border, otherborder = await cutting(im_grey, 2)  # 获取当前图像中的正方形区域
        if len(border) == 0:  # 如果莫得 说明需要翻转图像的黑白
            bo = True
        else:
            # print(f'cnt={cnt} border={border}') # test
            # 将正方形区域按照行列分组
            def highlight(rec, color="red"):  # 识别成功red 被其它代码排除的识别项blue 超过五列未识别的green 识别不出角色的black 识别出是100031的yellow
                x, y, w, h = rec
                cropped = img.crop([x + 2, y + 2, x + w - 2, y + h - 2])
                outpImgText.paste(cropped, (actual_x + x + 2, actual_y + y + 2, actual_x + x + w - 2, actual_y + y + h - 2))
                outpImgTextDraw.rectangle((actual_x + x, actual_y + y, actual_x + x + w, actual_y + y + h), fill=None, outline=color, width=4)

            for i in otherborder:
                highlight(i, 'blue')
            if len(border) < 4:  # 由5改为4 是为了日后准备加入残缺队伍查询（通过无api版本）
                for i in border:
                    highlight(i, 'blue')
            else:
                recs = border
                recs = set([tuple(rec) for rec in recs])

                def split_last_col_recs(recs):
                    recs = sorted(recs, key=lambda x: x[0], reverse=True)
                    last_col_recs = [rec for rec in recs if abs(rec[0] - recs[0][0]) < recs[0][2] / 2]
                    last_col_recs = sorted(last_col_recs, key=lambda x: x[1])
                    return list(set(recs) - set(last_col_recs)), last_col_recs

                _, last_col_recs = split_last_col_recs(recs)  # 先找出最右侧的一列有几行
                row_cnt = len(last_col_recs)

                arr = [[None for __ in range(5)] for _ in range(row_cnt)]
                arr_id = [[] for _ in range(row_cnt)]
                arr_id_6 = [[0 for __ in range(5)] for _ in range(row_cnt)]
                last_col_recs_xpos = [rec[1] for rec in last_col_recs]  # 以最右一列的坐标为基准

                for col_index in range(5):  # 从右往左一列一列掰，最多拿五列
                    recs, last_col_recs = split_last_col_recs(recs)
                    if len(last_col_recs) == 0:
                        break
                    for rec in last_col_recs:
                        # 看看rec能不能被识别出来
                        x, y, w, h = rec
                        cropped = img.crop([x + 2, y + 2, x + w - 2, y + h - 2])
                        uid_6, unit_id, unit_name, similarity = await getUnit(cropped)
                        # print(uid_6, unit_id, unit_name, similarity)  # 0 0 Unknown -1~-5
                        if unit_id == 0:
                            highlight(rec, "black")
                        else:
                            highlight(rec, "red" if unit_id != 1000 else "yellow")
                            most_near_row = 0
                            for row_index in range(1, len(arr)):
                                if abs(last_col_recs_xpos[row_index] - rec[1]) < abs(last_col_recs_xpos[most_near_row] - rec[1]):
                                    most_near_row = row_index
                            if arr[most_near_row][col_index] is None or abs(last_col_recs_xpos[most_near_row] - arr[most_near_row][col_index][1]) > abs(last_col_recs_xpos[most_near_row] - rec[1]):
                                arr[most_near_row][col_index] = rec
                                arr_id[most_near_row].append(unit_id)
                                arr_id_6[most_near_row][col_index] = uid_6

                for rec in recs:
                    highlight(rec, "green")

                # 创建一个 rowcnt行 5列 的画布，行间及四周留16px空隙，每行中的每列分为上下两个头像：截出来的和通过识别的id render出来的（均为64*64)。头像间隙0px。
                icon_size = 64
                compare_img = Image.new("RGBA", (icon_size * 5 + 16 * 2, icon_size * 2 * row_cnt + 16 * (row_cnt + 1)), (255, 255, 255, 255))

                for row_index in range(row_cnt):
                    none_cnt = arr[row_index].count(None)
                    if none_cnt >= 2:  # 不允许1-3个角色查询 不渲染
                        arr_id[row_index] = []
                        continue
                    for col_index in range(5):
                        if arr[row_index][4 - col_index] is None:
                            continue
                        pos_x = 16 + icon_size * col_index
                        pos_y = 16 * (row_index + 1) + icon_size * 2 * row_index
                        x, y, w, h = arr[row_index][4 - col_index]
                        cropped = img.crop([x + 2, y + 2, x + w - 2, y + h - 2]).resize((64, 64), Image.ANTIALIAS)
                        compare_img.paste(cropped, (pos_x, pos_y), cropped)  # 要不要加cropped
                        c = chara.fromid(arr_id_6[row_index][4 - col_index] // 100, arr_id_6[row_index][4 - col_index] % 100 // 10)
                        icon = await c.render_icon(icon_size)
                        compare_img.paste(icon, (pos_x, pos_y + 64), icon)

                def outp_b64(outp_img):
                    buf = BytesIO()
                    outp_img.save(buf, format='PNG')
                    base64_str = f'base64://{base64.b64encode(buf.getvalue()).decode()}'
                    return f'[CQ:image,file={base64_str}]'

                outpImg = Image.blend(outpImg, actual_img, 0.2)
                outpImg.alpha_composite(outpImgText)
                ratio = max(1, max((outpImg.size)[0], (outpImg.size)[1]) / 500)
                outpImg = outpImg.resize((int((outpImg.size)[0] / ratio), int((outpImg.size)[1] / ratio)), Image.ANTIALIAS)

                return arr_id, f'{outp_b64(outpImg)}\n{outp_b64(compare_img)}'

        try:
            im_grey, border = await cutting(im_grey, 1)  # 获取图片中最大的长方形区域
        except:
            return [], ""
        if cnt == 1 or bo:  # 如果是第一次裁剪，或者识别不到正方形区域，反色
            im_grey = im_grey.point(lambda x: 0 if x > 128 else 255)
        img = await cut(img, border)  # 将原始img也裁剪，和im_grey同步
        actual_x += border[0]
        actual_y += border[1]

        nowcolor = (nowcolor + 60) % 300
        outpImg.paste(ImageColor.getrgb(f'rgb({nowcolor},{nowcolor},{nowcolor})'), (actual_x, actual_y, actual_x + border[2], actual_y + border[3]))

        # outpImg.show()  # test

    return [], ""


async def getUnit(img2):
    img2 = img2.convert("RGB").resize((128, 128), Image.ANTIALIAS)
    img3 = np.array(img2)
    img4 = img3[25:96, 8:97, :]
    img4 = Image.fromarray(img4)
    dic = {}
    global data_processed
    if data_processed == None:
        await process_data()

    img4_arr = await get_hash_arr(img4)
    for uid in data_processed:
        dic[uid] = await calc_distance_arr(data_processed[uid], img4_arr)

    lis = list(sorted(dic.items(), key=lambda x: abs(x[1])))

    similarity = int(lis[0][1])
    if similarity > 90:  # 没一个相似的
        return 0, 0, "Unknown", 100 - similarity
    uid_6 = int(lis[0][0])
    uid = uid_6 // 100
    try:
        return uid_6, uid, chara.fromid(uid).name, 100 - similarity
    except:
        return uid_6, uid, "Unknown", 100 - similarity


async def get_pic(address):
    return await (await aiorequests.get(address, timeout=5)).content


async def _arena_query(bot, ev: CQEvent, region: int):
    uid = ev.user_id

    if not lmt.check(uid):
        await bot.finish(ev, '您查询得过于频繁，请稍等片刻', at_sender=True)

    # 处理输入数据
    defen = ""
    ret = re.match(r"\[CQ:image,file=(.*),url=(.*)\]", str(ev.message))
    if ret:
        # await bot.send(ev, "recognizing")
        image = Image.open(BytesIO(await get_pic(html.unescape(ret.group(2)))))
        boxDict, s = await getBox(image)

        if boxDict == []:
            await bot.finish(ev, "未识别到有4个及以上角色的阵容！")

        try:
            if region == -20:
                s += f'\narr_id_4={boxDict}'
            await bot.send(ev, s)
        except:
            pass

        if region == -20:
            return

        if len(boxDict) == 1:
            await __arena_query(bot, ev, region, boxDict[0])
            return

        if len(boxDict) > 3:
            await bot.finish(ev, "请截图pjjc详细对战记录（对战履历详情）（含敌我双方2或3队阵容）")

        team_has_result = 0
        lmt.start_cd(uid)

        all_query_records = [[] for _ in range(len(boxDict))]
        '''
        [
            [   
                [None, -100, None], # 通配，等待从缓存中获取配队
                [(第1队第1解),权值,render(渲染该队所需数据)]
                [(第1队第2解),权值,render]
            ],
            [
                [None, -100, None], # 通配
                [(第2队第1解),权值,render] 
            ]
        ]
        '''

        for query_index, query_team in enumerate(boxDict):
            all_query_records[query_index].append([None, -100, "placeholder"])

            if len(set(query_team)) == 1 and query_team[0] == 1000:  # 1000 1000 1000 1000 1000 pjjc情况
                continue

            # if len(query_team) == 4:
            #     boxDict[query_index].append(1000)
            #     continue

            records = await __arena_query(bot, ev, region, query_team, 1)

            if records == []:
                continue

            team_has_result += 1

            for record in records:
                record_team = tuple([chara_obj.id for chara_obj in record["atk"]])
                all_query_records[query_index].append([record_team, record["val"], record])

        if team_has_result == 0 and len(boxDict) == 3:
            await bot.finish(ev, "均未查询到解法！")

        await generateCollisionFreeTeam(bot, ev, all_query_records, team_has_result, region, boxDict)  # 最多允许补两队
    else:
        await __arena_query(bot, ev, region)


def recommend1Team(already_used_units: List[int]):
    '''
    already_used_units: [1110,1008,1011,1026,1089] 不可使用的角色id（四位）
    return : render | "placeholder"
    '''
    global best_atk_records
    for record_6 in best_atk_records:  # [111451,101261,110351,103461,103261]
        record_4 = [x // 100 for x in record_6]  # [1114,1012,1103,1034,1032]
        team_mix = already_used_units + record_4
        if len(team_mix) == len(set(team_mix)):  # 推荐配队成功
            return {"atk": [chara.fromid(uid_6 // 100, uid_6 % 100 // 10) for uid_6 in record_6], "team_type": "frequency"}
    return "placeholder"


def recommend2Teams(already_used_units: List[int]):
    '''
    return : render, render 或 "placeholder", "placeholder"
    '''
    global best_atk_records

    try_combinations = []  # [查询1解序号, 查询2解序号, 优先级]
    for record_1_index in range(len(best_atk_records) - 1):
        for record_2_index in range(record_1_index + 1, len(best_atk_records)):
            try_combinations.append([record_1_index, record_2_index, record_1_index + record_2_index])
    try_combinations = list(sorted(try_combinations, key=lambda x: x[-1]))
    for try_combination in try_combinations:
        record_6_1_index = try_combination[0]  # [111451,101261,110351,103461,103261]
        record_6_1 = best_atk_records[record_6_1_index]
        record_4_1 = [x // 100 for x in record_6_1]

        record_6_2_index = try_combination[1]
        record_6_2 = best_atk_records[record_6_2_index]
        record_4_2 = [x // 100 for x in record_6_2]

        team_mix = already_used_units + record_4_1 + record_4_2
        if len(team_mix) == len(set(team_mix)):  # 推荐配队成功
            # print(f'\n\n成功配队{already_used_units}\n{record_4_1}\n{record_4_2}')  # test
            return {"atk": [chara.fromid(uid_6 // 100, uid_6 % 100 // 10) for uid_6 in record_6_2], "team_type": "frequency"}, {"atk": [chara.fromid(uid_6 // 100, uid_6 % 100 // 10) for uid_6 in record_6_1], "team_type": "frequency"}

    return "placeholder", "placeholder"


async def generateCollisionFreeTeam(bot: HoshinoBot, ev: CQEvent, all_query_records, team_has_result, region, boxDict):
    '''
    all_query_records
    [
        [
            [None, -100, "placeholder"], # 通配，等待从缓存中获取配队
            [(1110,1008,1011,1026,1089), 2.105, render], # [(第1队第1解),权值,render(渲染该队所需数据)]
            [(1111,1008,1802,1012,1014), 1.152, render] # [(第1队第2解),权值,render]
        ],
        [
            [None, -100, "placeholder"], # 通配
            [(第2队第1解),权值,render]
        ]
    ]
    '''
    collision_free_match_cnt = 0
    outp_render = []
    collision_free_match_cnt_2 = 0  # 处理三队查询只能两队无冲的情况
    outp_render_2 = []

    if len(all_query_records) == 2:
        try_combinations = []  # [查询1解序号, 查询2解序号, 优先级]
        for query_1_index, query_1_record in enumerate(all_query_records[0]):
            for query_2_index, query_2_record in enumerate(all_query_records[1]):
                val = query_1_record[1] + query_2_record[1]
                try_combinations.append([query_1_index, query_2_index, val])
        try_combinations = sorted(try_combinations, key=lambda x: x[-1], reverse=True)

        for try_combination in try_combinations:
            record_1 = all_query_records[0][try_combination[0]]  # [(1110,1008,1011,1026,1089), 2.105, render] # 或通配
            record_2 = all_query_records[1][try_combination[1]]
            team_1 = [] if record_1[0] is None else list(record_1[0])  # (1110,1008,1011,1026,1089)
            team_2 = [] if record_2[0] is None else list(record_2[0])
            team_mix = team_1 + team_2  # list
            if len(team_mix) != len(set(team_mix)):  # 存在冲突
                continue

            succ = False
            val = try_combination[-1]
            if val < -250:
                break
            if val < -150:  # 已有0队，要补2队 # 只会出现一次
                team_recommend_1, team_recommend_2 = recommend2Teams(team_mix)
                if team_recommend_1 == "placeholder" or team_recommend_2 == "placeholder":
                    continue
                record_1[-1] = team_recommend_1
                record_2[-1] = team_recommend_2
                succ = True
            elif val < -50:  # 已有1队，要补1队
                team_recommend = recommend1Team(team_mix)
                if team_recommend == "placeholder":
                    continue
                if team_1 == []:
                    record_1[-1] = team_recommend
                if team_2 == []:
                    record_2[-1] = team_recommend
                succ = True
            else:  # 已有2队
                succ = True

            if succ:
                collision_free_match_cnt += 1
                outp_render += [record_1[-1], record_2[-1], []]
                if collision_free_match_cnt >= 8:
                    break

    if len(all_query_records) == 3:
        try_combinations = []  # [查询1解序号, 查询2解序号, 查询3解序号, 优先级]
        for query_1_index, query_1_record in enumerate(all_query_records[0]):
            for query_2_index, query_2_record in enumerate(all_query_records[1]):
                for query_3_index, query_3_record in enumerate(all_query_records[2]):
                    val = query_1_record[1] + query_2_record[1] + query_3_record[1]
                    try_combinations.append([query_1_index, query_2_index, query_3_index, val])
        try_combinations = sorted(try_combinations, key=lambda x: x[-1], reverse=True)

        for try_combination in try_combinations:
            record_1 = all_query_records[0][try_combination[0]]  # [(1110,1008,1011,1026,1089), 2.105, render] # 或通配
            record_2 = all_query_records[1][try_combination[1]]
            record_3 = all_query_records[2][try_combination[2]]
            team_1 = [] if record_1[0] is None else list(record_1[0])  # (1110,1008,1011,1026,1089)
            team_2 = [] if record_2[0] is None else list(record_2[0])
            team_3 = [] if record_3[0] is None else list(record_3[0])
            team_mix = team_1 + team_2 + team_3  # list
            if len(team_mix) != len(set(team_mix)):  # 存在冲突
                continue

            succ = False
            val = try_combination[-1]
            if val < -250:
                break
            if val < -150:  # 已有1队，要补2队
                team_recommend_1, team_recommend_2 = recommend2Teams(team_mix)
                if team_recommend_1 == "placeholder" or team_recommend_2 == "placeholder":
                    continue
                if team_1 != []:
                    record_2[-1] = team_recommend_1
                    record_3[-1] = team_recommend_2
                if team_2 != []:
                    record_3[-1] = team_recommend_1
                    record_1[-1] = team_recommend_2
                if team_3 != []:
                    record_1[-1] = team_recommend_1
                    record_2[-1] = team_recommend_2
                succ = True
            elif val < -50:  # 已有2队，要补1队 # 此时已有两队无冲
                team_recommend = recommend1Team(team_mix)
                if team_recommend == "placeholder":
                    collision_free_match_cnt_2 += 1
                    outp_render_2 += [record_1[-1], record_2[-1], record_3[-1], []]
                else:
                    if team_1 == []:
                        record_1[-1] = team_recommend
                    if team_2 == []:
                        record_2[-1] = team_recommend
                    if team_3 == []:
                        record_3[-1] = team_recommend
                    succ = True
            else:  # 已有3队
                succ = True

            if succ:
                collision_free_match_cnt += 1
                outp_render += [record_1[-1], record_2[-1], record_3[-1], []]
                # print(f'当前无冲配队数={collision_free_match_cnt} len(outp_render)={len(outp_render)}')  # test
                if collision_free_match_cnt >= 6:
                    break

    if collision_free_match_cnt:
        # print(f'\n\n总共无冲配队数={collision_free_match_cnt} len(outp_render)={len(outp_render)}')  # test
        teams = await render_atk_def_teams(outp_render[:-1])
        await bot.finish(ev, str(MessageSegment.image(pic2b64(teams))))
    elif collision_free_match_cnt_2:
        teams = await render_atk_def_teams(outp_render_2[:-1])
        await bot.finish(ev, str(MessageSegment.image(pic2b64(teams))))
    else:
        if len(all_query_records) == 2:  # 查两队
            if team_has_result == 0:
                await bot.finish(ev, "均未查询到解法！")
            elif team_has_result == 1:
                for index, records in enumerate(all_query_records):
                    if len(records) > 1:
                        await bot.send(ev, f"仅第{index+1}队查询到解法！")
                        await __arena_query(bot, ev, region, boxDict[index], only_use_cache=True)
            else:  # 2队有结果，但凑不满2队无冲
                await bot.send(ev, "无冲配对失败，返回单步查询结果")
                for index, records in enumerate(all_query_records):
                    await __arena_query(bot, ev, region, boxDict[index], only_use_cache=True)
        else:  # 查三队
            if team_has_result == 1:
                for index, records in enumerate(all_query_records):
                    if len(records) > 1:
                        await bot.send(ev, f"仅第{index+1}队查询到解法！")
                        await __arena_query(bot, ev, region, boxDict[index], only_use_cache=True)
            else:  # 2~3队有结果，但凑不满2队无冲
                await bot.send(ev, "无冲配对失败，返回单步查询结果")
                for index, records in enumerate(all_query_records):
                    if len(records) > 1:
                        await __arena_query(bot, ev, region, boxDict[index], only_use_cache=True)


def remove_buffer(uid: str):
    curpath = dirname(__file__)
    bufferpath = join(curpath, 'buffer/buffer.json')

    buffer = {}
    with open(bufferpath, 'r', encoding="utf-8") as fp:
        buffer = json.load(fp)

    try:
        remove(join(curpath, f'buffer/{uid}.json'))
    except:
        pass

    if uid in buffer:
        buffer.pop(uid)
        with open(bufferpath, 'w', encoding="utf-8") as fp:
            json.dump(buffer, fp, ensure_ascii=False, indent=4)


async def __arena_query(bot, ev: CQEvent, region: int, defen="", raw=0, only_use_cache=False):
    uid = ev.user_id
    unknown = ""
    if defen == "":
        defen = ev.message.extract_plain_text()
        defen = re.sub(r'[?？，,_]', '', defen)
        defen, unknown = chara.roster.parse_team(defen)

    if unknown:
        _, name, score = chara.guess_id(unknown)
        if score < 70 and not defen:
            return  # 忽略无关对话
        msg = f'无法识别"{unknown}"' if score < 70 else f'无法识别"{unknown}" 您说的有{score}%可能是{name}'
        await bot.finish(ev, msg)
    if not defen:
        await bot.finish(ev, '查询请发送"b/r/tjjc+防守队伍"，无需+号', at_sender=True)
    if len(defen) > 5:
        await bot.finish(ev, '编队不能多于5名角色', at_sender=True)
    if len(defen) < 4:
        await bot.finish(ev, '编队角色过少', at_sender=True)
    if len(defen) != len(set(defen)):
        await bot.finish(ev, '编队中含重复角色', at_sender=True)
    if any(chara.is_npc(i) for i in defen):
        await bot.finish(ev, '编队中含未实装角色', at_sender=True)

    key = ''.join([str(x) for x in sorted(defen)]) + str(region)
    # 执行查询
    lmt.start_cd(uid)
    res = await arena.do_query(defen, region, -1 if only_use_cache else 1)

    defen = [chara.fromid(x).name for x in defen]
    defen = f"防守方【{' '.join(defen)}】"

    # 处理查询结果
    if res is None:
        remove_buffer(key)
        if not raw:
            await bot.finish(ev, f'{defen}\npcrdfans未返回数据')
        else:
            return []
    if not len(res):
        remove_buffer(key)
        if not raw:
            await bot.finish(ev, f'{defen}\n未查询到解法')
        else:
            return []

    if raw:
        return res

    # 发送回复
    res = res[:10]  # 限制显示数量，截断结果
    sv.logger.info('Arena generating picture...')
    teams = await render_atk_def_teams(res)
    teams = pic2b64(teams)
    teams = MessageSegment.image(teams)
    sv.logger.info('Arena picture ready!')
    msg = [defen, str(teams)]

    if region == 1:
        msg.append('※使用"b怎么拆"或"台怎么拆"可按服过滤')

    sv.logger.debug('Arena sending result...')
    await bot.send(ev, '\n'.join(msg))
    sv.logger.debug('Arena result sent!')


@sv.on_fullmatch('竞技场更新卡池')
async def _update_dic(bot, ev):
    try:
        await bot.send(ev, f'{update_dic()}')
        global data
        data = np.load(dataDir, allow_pickle=True).item()
        await process_data()
    except Exception as e:
        await bot.send(ev, f'Error: {e}')


@sv.scheduled_job('cron', hour='3', minute='21')
async def _update_dic_cron():
    try:
        update_dic()
        global data
        data = np.load(dataDir, allow_pickle=True).item()
        await process_data()
    except Exception as e:
        pass
    try:
        global best_atk_records
        update_record()
        with open(best_atk_records_path, "r", encoding="utf-8") as fp:
            best_atk_records = json.load(fp)
    except:
        pass


# @sv.on_fullmatch('恢复竞技场查询记录')
def restore_record(bot, ev):
    curpath = dirname(__file__)
    bufferpath = join(curpath, 'buffer/')
    with open(join(bufferpath, "buffer.json"), "r", encoding="utf-8") as fp:
        buffer = json.load(fp)

    for filename in listdir(bufferpath):
        if len(filename) != 26:
            continue
        filename = filename[:-5]
        if filename not in buffer:
            buffer[filename] = 1670000000

    with open(join(bufferpath, "buffer.json"), "w", encoding="utf-8") as fp:
        json.dump(buffer, fp, ensure_ascii=False, indent=4)