mirror of
https://github.com/alantang1977/hipy.git
synced 2024-11-25 07:53:09 +02:00
227 lines
7.9 KiB
Python
227 lines
7.9 KiB
Python
|
#!/usr/bin/env python3
|
|||
|
# -*- coding: utf-8 -*-
|
|||
|
# File : 采集分类生成器.py
|
|||
|
# Author: DaShenHan&道长-----先苦后甜,任凭晚风拂柳颜------
|
|||
|
# Date : 2024/6/21
|
|||
|
|
|||
|
import os
|
|||
|
import json
|
|||
|
import gzip
|
|||
|
import base64
|
|||
|
|
|||
|
from urllib.parse import urljoin
|
|||
|
from concurrent.futures import ThreadPoolExecutor
|
|||
|
from pprint import pprint
|
|||
|
import time
|
|||
|
|
|||
|
import requests
|
|||
|
|
|||
|
import warnings
|
|||
|
|
|||
|
# 关闭警告
|
|||
|
warnings.filterwarnings("ignore")
|
|||
|
requests.packages.urllib3.disable_warnings()
|
|||
|
|
|||
|
pool = ThreadPoolExecutor(max_workers=20) # 初始化线程池内线程数量为20
|
|||
|
|
|||
|
headers = {
|
|||
|
'User-Agent': 'Mozilla/5.0 (iPhone; CPU iPhone OS 13_2_3 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/13.0.3 Mobile/15E148 Safari/604.1',
|
|||
|
'Connection': 'close' # 设置为关闭长连接
|
|||
|
}
|
|||
|
|
|||
|
timeout = 5 # 5秒
|
|||
|
|
|||
|
use_gzip = False
|
|||
|
|
|||
|
|
|||
|
def compress_and_encode(data: str):
|
|||
|
# 压缩数据
|
|||
|
compressed_data = gzip.compress(data.encode('utf-8'))
|
|||
|
# 对压缩数据进行Base64编码
|
|||
|
encoded_data = base64.b64encode(compressed_data).decode('utf-8')
|
|||
|
return encoded_data
|
|||
|
|
|||
|
|
|||
|
def decode_and_decompress(encoded_data: str):
|
|||
|
# 解码Base64数据
|
|||
|
decoded_data = base64.b64decode(encoded_data.encode('utf-8'))
|
|||
|
# 解压缩数据
|
|||
|
decompressed_data = gzip.decompress(decoded_data).decode('utf-8')
|
|||
|
return decompressed_data
|
|||
|
|
|||
|
|
|||
|
def get_classes(rec):
|
|||
|
classes = None
|
|||
|
if rec.get('url') and str(rec['url']).startswith('http'):
|
|||
|
_class_api = rec.get('api') or '/api.php/provide/vod/'
|
|||
|
_api = urljoin(str(rec['url']).rstrip('/'), _class_api)
|
|||
|
# _api = urljoin(rec['url'], '/api.php/provide/vod/at/json')
|
|||
|
print(_api)
|
|||
|
try:
|
|||
|
r = requests.get(_api, headers=headers, timeout=timeout, verify=False)
|
|||
|
ret = r.json()
|
|||
|
if rec.get('name') == '乐视资源':
|
|||
|
print('=======乐视=========')
|
|||
|
print(ret)
|
|||
|
# print(ret)
|
|||
|
classes = ret.get('class')
|
|||
|
except Exception as e:
|
|||
|
print(f'获取资源【{rec["name"]}】({_api})分类发生错误:{e}')
|
|||
|
|
|||
|
return classes
|
|||
|
|
|||
|
|
|||
|
def convert_class(classes, name=None):
|
|||
|
"""
|
|||
|
获取的分类转静态分类格式
|
|||
|
@param classes:
|
|||
|
@return:
|
|||
|
"""
|
|||
|
if name is None:
|
|||
|
name = ''
|
|||
|
if not classes:
|
|||
|
return {
|
|||
|
"name": "",
|
|||
|
"class_name": "",
|
|||
|
"class_url": "",
|
|||
|
}
|
|||
|
class_names = []
|
|||
|
class_urls = []
|
|||
|
for cls in classes:
|
|||
|
if cls.get('type_name') and cls.get('type_id'):
|
|||
|
class_urls.append(str(cls['type_id']))
|
|||
|
class_names.append(str(cls['type_name']))
|
|||
|
global use_gzip
|
|||
|
return {
|
|||
|
"name": name,
|
|||
|
"class_name": compress_and_encode('&'.join(class_names)) if use_gzip else '&'.join(class_names),
|
|||
|
"class_url": '&'.join(class_urls),
|
|||
|
}
|
|||
|
|
|||
|
|
|||
|
def get_convert_classes(rec):
|
|||
|
classes = get_classes(rec)
|
|||
|
classes = convert_class(classes, rec.get('name'))
|
|||
|
return classes
|
|||
|
|
|||
|
|
|||
|
def check_class(api, type_name, type_id, limit_count=6):
|
|||
|
_url = f'{api}?ac=detail&pg=1&t={type_id}'
|
|||
|
try:
|
|||
|
r = requests.get(_url, headers=headers, timeout=timeout, verify=False)
|
|||
|
ret = r.json()
|
|||
|
if not ret.get("list") or len(ret["list"]) < limit_count:
|
|||
|
print(f'获取资源 {api} 分类【{type_name}】数量为:{len(ret["list"])} 小于{limit_count}视为排除')
|
|||
|
return False
|
|||
|
except Exception as e:
|
|||
|
print(f'获取资源 {_url} 分类【{type_name}】发生错误:{e}')
|
|||
|
return True
|
|||
|
|
|||
|
|
|||
|
def check_active(api):
|
|||
|
try:
|
|||
|
r = requests.get(api, headers=headers, timeout=timeout, verify=False)
|
|||
|
ret = r.json()
|
|||
|
if not ret.get("class"):
|
|||
|
return False
|
|||
|
except Exception as e:
|
|||
|
print(f'检查api: {api} 存活发生错误:{e}')
|
|||
|
return False
|
|||
|
return True
|
|||
|
|
|||
|
|
|||
|
def main(fname='采集'):
|
|||
|
file_path = f'./{fname}.json'
|
|||
|
out_file_path = file_path.replace('.json', '静态.json')
|
|||
|
if not os.path.exists(file_path):
|
|||
|
exit(f'不存在采集文件路径:{file_path}')
|
|||
|
with open(file_path, encoding='utf-8') as f:
|
|||
|
data = f.read()
|
|||
|
records = json.loads(data)
|
|||
|
print(records)
|
|||
|
# for rec in records:
|
|||
|
# ret = get_convert_classes(rec)
|
|||
|
# pprint(ret)
|
|||
|
tasks = [pool.submit(get_convert_classes, rec) for rec in records] # 构造一个列表,循环向线程池内submit提交执行的方法
|
|||
|
pool.shutdown(wait=True) # 线程数等待所有线程结束,这里 卡住主线程
|
|||
|
results = [task.result() for task in tasks]
|
|||
|
print(results)
|
|||
|
new_records = []
|
|||
|
for record in records:
|
|||
|
rec_name = record["name"]
|
|||
|
if rec_name:
|
|||
|
has_name = [ret for ret in results if ret.get("name") == rec_name]
|
|||
|
if has_name:
|
|||
|
record.update(has_name[-1])
|
|||
|
new_records.append(record)
|
|||
|
pprint(new_records)
|
|||
|
print(f'转换静态数据成功记录数:{len(new_records)}')
|
|||
|
with open(out_file_path, mode='w+', encoding='utf-8') as f:
|
|||
|
f.write(json.dumps(new_records, ensure_ascii=False, indent=2))
|
|||
|
|
|||
|
|
|||
|
def main_exclude(fname='采集静态', max_workers=0):
|
|||
|
file_path = f'./{fname}.json'
|
|||
|
if not os.path.exists(file_path):
|
|||
|
exit(f'不存在采集文件路径:{file_path}')
|
|||
|
with open(file_path, encoding='utf-8') as f:
|
|||
|
data = f.read()
|
|||
|
records = json.loads(data)
|
|||
|
if len(records) < 1 or not records[0].get('class_name'):
|
|||
|
exit('输入数据有误,疑似不是静态数据')
|
|||
|
print(records)
|
|||
|
new_records = []
|
|||
|
for rec in records:
|
|||
|
new_rec = rec.copy()
|
|||
|
if rec.get('api'):
|
|||
|
api_url = urljoin(rec['url'], rec['api'])
|
|||
|
else:
|
|||
|
api_url = urljoin(rec['url'], '/api.php/provide/vod/')
|
|||
|
print(api_url)
|
|||
|
cate_excludes = []
|
|||
|
if not check_active(api_url):
|
|||
|
print(f'{rec["name"]} ({rec["url"]})视为不存活,跳过分类检测')
|
|||
|
else:
|
|||
|
class_names = decode_and_decompress(rec['class_name']).split('&')
|
|||
|
class_urls = rec['class_url'].split('&')
|
|||
|
rec_pool = ThreadPoolExecutor(max_workers=max_workers or len(class_names)) # 初始化线程池内线程数量为分类数量
|
|||
|
tasks = []
|
|||
|
for i in range(len(class_names)):
|
|||
|
type_name = class_names[i]
|
|||
|
type_id = class_urls[i]
|
|||
|
tasks.append(rec_pool.submit(check_class, api_url, type_name, type_id))
|
|||
|
rec_pool.shutdown(wait=True) # 线程数等待所有线程结束,这里 卡住主线程
|
|||
|
results = [task.result() for task in tasks]
|
|||
|
print(results)
|
|||
|
for i in range(len(class_names)):
|
|||
|
type_name = class_names[i]
|
|||
|
# type_id = class_urls[i]
|
|||
|
if not results[i]:
|
|||
|
cate_excludes.append(type_name)
|
|||
|
if len(cate_excludes) > 0:
|
|||
|
new_rec['cate_excludes'] = cate_excludes
|
|||
|
new_records.append(new_rec)
|
|||
|
|
|||
|
with open(file_path, mode='w+', encoding='utf-8') as f:
|
|||
|
f.write(json.dumps(new_records, ensure_ascii=False, indent=2))
|
|||
|
|
|||
|
|
|||
|
if __name__ == '__main__':
|
|||
|
use_gzip = True
|
|||
|
fmode = str(input('请输入处理文件方式(0:生成分类 1:添加分类过滤),留空默认为生成静态分类:\n'))
|
|||
|
ftips = '采集静态' if fmode == '1' else '采集'
|
|||
|
fname = str(input(f'请输入文件名(q结束程序),留空默认为{ftips}:\n'))
|
|||
|
t1 = time.time()
|
|||
|
if fname == 'q':
|
|||
|
exit('已主动结束脚本')
|
|||
|
if not fmode or fmode == '0':
|
|||
|
fname = fname or '采集'
|
|||
|
main(fname)
|
|||
|
elif fmode == '1':
|
|||
|
fname = fname or '采集静态'
|
|||
|
main_exclude(fname, 10)
|
|||
|
else:
|
|||
|
exit(f'未知的处理类型:{fmode}')
|
|||
|
t2 = time.time()
|
|||
|
print(f'本次程序运行耗时:{round(t2 - t1, 2)}秒')
|