DDR爱好者之家 Design By 杰米
import requests
import os
import datetime
import threading

class xiazai():
  def __init__(self,url):
    self.url = url
    work_dir = os.getcwd()
    # print(work_dir)
    # 用来保存ts文件
    file_dir = os.path.join(work_dir, 'file_tmp')
    if not os.path.exists(file_dir):
      os.mkdir(file_dir)
    self.headers ={
      'user-agent': 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_3) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/80.0.3987.116 Safari/537.36'
    }
    self.savefile(self.url)

  def savefile(self, file_url):

    r = requests.get(file_url, headers=self.headers)
    # 合成带有hls的m3u8地址
    if r.text.split('\n')[-1] == '':
      hls_mark = r.text.split('\n')[-2] # 以防\n结尾
    else:
      hls_mark = r.text.split('\n')[-1]
    self.url_m3u8_hls = file_url.replace('index.m3u8', hls_mark)
    #file_m3u8 = url_m3u8_hls.split('/')[-1]
    self.duqu()
    #print(url_m3u8_hls)

  def duqu(self):
    r = requests.get(self.url_m3u8_hls, headers=self.headers).text
    text_bytes = r.split('\n')
    # 筛选以.ts结尾的行
    # 有些情况下可能是以其他格式的文件,比如png,下载后修改后缀即可
    # ts_name = [i for i in text_string if i.endswith('.ts')]
    self.ts_time = [i for i in text_bytes if i.startswith('#EXTINF')]
    #self.shijian(dm_time)
    #print(dm_time)
    self.ts_neirong = [i for i in text_bytes if not i.startswith('#')]
    self.ts_neirong.pop()
    self.threads = []
    self.threads.append(threading.Thread(target=self.xiazai))
    self.threads.append(threading.Thread(target=self.shijian))
    for t in self.threads:
      # print(t)
      t.start()
    #self.xiazai(url_m3u8_hls)
    # print(ts_neirong)
  def shijian(self):
    self.dm_time = 0
    for i in range(len(self.ts_time)):
      ts_time1 = self.ts_time[i].replace('#EXTINF:', '')
      ts_time2 = ts_time1.replace(',', '')
      self.dm_time = float(ts_time2) + self.dm_time
    shichang_time = str(datetime.timedelta(seconds=self.dm_time))
    print('视频时长:%s' % shichang_time)
  def xiazai(self):
    liebiao=[]
    for i in range(len(self.ts_neirong)):
      hls_mark = self.url_m3u8_hls.split('/')[-1]
      url_xiazai = self.url_m3u8_hls.replace(hls_mark, self.ts_neirong[i])
      liebiao.append(url_xiazai)
      #print(url_xiazai)
      # r = requests.get(url_xiazai, headers=self.headers)
      # with open('file_tmp/'+ts_neirong[i], 'wb') as f:
      # f.write(r.content)
      # f.close()
    x = self.bisector_list(liebiao, 10)
    self.xiancheng0=x[0]
    self.xiancheng1=x[1]
    self.xiancheng2=x[2]
    self.xiancheng3=x[3]
    self.xiancheng4=x[4]
    self.xiancheng5=x[5]
    self.xiancheng6=x[6]
    self.xiancheng7=x[7]
    self.xiancheng8=x[8]
    self.xiancheng9=x[9]
    self.threads2 = []
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai1))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai2))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai3))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai4))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai5))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai6))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai7))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai8))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai9))
    self.threads2.append(threading.Thread(target=self.xiancheng_xiazai10))

    for t in self.threads2:
      # print(t)
      t.start()


  def xiancheng_xiazai1(self):
    #print(self.xiancheng0)
    for i in self.xiancheng0:
      #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()

  def xiancheng_xiazai2(self):
    #print(self.xiancheng1)
    for i in self.xiancheng1:
      #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi= i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()

  def xiancheng_xiazai3(self):
    #print(self.xiancheng2)
    for i in self.xiancheng2:
      #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()

  def xiancheng_xiazai4(self):
    #print(self.xiancheng3)
    for i in self.xiancheng3:
      #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()


  def xiancheng_xiazai5(self):
    #print(self.xiancheng4)
    for i in self.xiancheng4:
    #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()


  def xiancheng_xiazai6(self):
    #print(self.xiancheng5)
    for i in self.xiancheng5:
    #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()


  def xiancheng_xiazai7(self):
    #print(self.xiancheng6)
    for i in self.xiancheng6:
    #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()


  def xiancheng_xiazai8(self):
    #print(self.xiancheng7)
    for i in self.xiancheng7:
    #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()


  def xiancheng_xiazai9(self):
    #print(self.xiancheng8)
    for i in self.xiancheng8:
    #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()


  def xiancheng_xiazai10(self):
    #print(self.xiancheng9)
    for i in self.xiancheng9:

    #print(i)
      r = requests.get(i, headers=self.headers)
      mingzi = i.split('/')[-1]
      with open('file_tmp/'+mingzi, 'wb') as f:
        f.write(r.content)
    f.close()


  def bisector_list(self,tabulation: list, num: int):
    """
    将列表平均分成几份
    :param tabulation: 列表
    :param num: 份数
    :return: 返回一个新的列表
    """
    new_list = []

    '''列表长度大于等于份数'''
    if len(tabulation) >= num:
      '''remainder:列表长度除以份数,取余'''
      remainder = len(tabulation) % num
      if remainder == 0:
        '''merchant:列表长度除以分数'''
        merchant = int(len(tabulation) / num)
        '''将列表平均拆分'''
        for i in range(1, num + 1):
          if i == 1:
            new_list.append(tabulation[:merchant])
          else:
            new_list.append(tabulation[(i - 1) * merchant:i * merchant])
        return new_list
      else:
        '''merchant:列表长度除以分数 取商'''
        merchant = int(len(tabulation) // num)
        '''remainder:列表长度除以份数,取余'''
        remainder = int(len(tabulation) % num)
        '''将列表平均拆分'''
        for i in range(1, num + 1):
          if i == 1:
            new_list.append(tabulation[:merchant])
          else:
            new_list.append(tabulation[(i - 1) * merchant:i * merchant])
            '''将剩余数据的添加前面列表中'''
            if int(len(tabulation) - i * merchant) <= merchant:
              for j in tabulation[-remainder:]:
                new_list[tabulation[-remainder:].index(j)].append(j)
        return new_list
    else:
      '''如果列表长度小于份数'''
      for i in range(1, len(tabulation) + 1):
        tabulation_subset = []
        tabulation_subset.append(tabulation[i - 1])
        new_list.append(tabulation_subset)
      return new_list

if __name__=='__main__':
  xiazai('http://iqiyi.cdn9-okzy.com/20200907/15137_ed25d8c5/index.m3u8')

速度很慢.  40m 5分钟  不加多线程 1小时。 可能我的m3u8不行,或者我电脑不行, 多线程是机械式的。  电脑好 可以多加几条。

以上就是python 下载m3u8视频的示例代码的详细内容,更多关于python 下载m3u8视频的资料请关注其它相关文章!

DDR爱好者之家 Design By 杰米
广告合作:本站广告合作请联系QQ:858582 申请时备注:广告合作(否则不回)
免责声明:本站资源来自互联网收集,仅供用于学习和交流,请遵循相关法律法规,本站一切资源不代表本站立场,如有侵权、后门、不妥请联系本站删除!
DDR爱好者之家 Design By 杰米

《魔兽世界》大逃杀!60人新游玩模式《强袭风暴》3月21日上线

暴雪近日发布了《魔兽世界》10.2.6 更新内容,新游玩模式《强袭风暴》即将于3月21 日在亚服上线,届时玩家将前往阿拉希高地展开一场 60 人大逃杀对战。

艾泽拉斯的冒险者已经征服了艾泽拉斯的大地及遥远的彼岸。他们在对抗世界上最致命的敌人时展现出过人的手腕,并且成功阻止终结宇宙等级的威胁。当他们在为即将于《魔兽世界》资料片《地心之战》中来袭的萨拉塔斯势力做战斗准备时,他们还需要在熟悉的阿拉希高地面对一个全新的敌人──那就是彼此。在《巨龙崛起》10.2.6 更新的《强袭风暴》中,玩家将会进入一个全新的海盗主题大逃杀式限时活动,其中包含极高的风险和史诗级的奖励。

《强袭风暴》不是普通的战场,作为一个独立于主游戏之外的活动,玩家可以用大逃杀的风格来体验《魔兽世界》,不分职业、不分装备(除了你在赛局中捡到的),光是技巧和战略的强弱之分就能决定出谁才是能坚持到最后的赢家。本次活动将会开放单人和双人模式,玩家在加入海盗主题的预赛大厅区域前,可以从强袭风暴角色画面新增好友。游玩游戏将可以累计名望轨迹,《巨龙崛起》和《魔兽世界:巫妖王之怒 经典版》的玩家都可以获得奖励。