123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290 |
- # coding=utf-8
- # !/usr/bin/python
- import sys
- sys.path.append('..')
- from base.spider import Spider
- import base64
- import hashlib
- import requests
- from Crypto.Cipher import AES
- import urllib
- class Spider(Spider): # 元类 默认的元类 type
- def getName(self):
- return "厂长资源"
- def init(self, extend=""):
- print("============{0}============".format(extend))
- pass
- def homeContent(self, filter):
- result = {}
- cateManual = {
- "豆瓣电影Top250": "dbtop250",
- "最新电影": "zuixindianying",
- "电视剧": "dsj",
- "国产剧": "gcj",
- "美剧": "meijutt",
- "韩剧": "hanjutv",
- "番剧": "fanju",
- "动漫": "dm"
- }
- classes = []
- for k in cateManual:
- classes.append({
- 'type_name': k,
- 'type_id': cateManual[k]
- })
- result['class'] = classes
- return result
- def homeVideoContent(self):
- url = "https://czspp.com"
- header = {
- "Connection": "keep-alive",
- "Referer": url,
- "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/104.0.0.0 Safari/537.36"
- }
- session = self.getCookie(url,header)
- rsp = session.get(url, headers=header)
- root = self.html(self.cleanText(rsp.text))
- aList = root.xpath("//div[@class='mi_btcon']//ul/li")
- videos = []
- for a in aList:
- name = a.xpath('./a/img/@alt')[0]
- pic = a.xpath('./a/img/@data-original')[0]
- mark = a.xpath("./div[@class='hdinfo']/span/text()")[0]
- sid = a.xpath("./a/@href")[0]
- sid = self.regStr(sid, "/movie/(\\S+).html")
- videos.append({
- "vod_id": sid,
- "vod_name": name,
- "vod_pic": pic,
- "vod_remarks": mark
- })
- result = {
- 'list': videos
- }
- return result
- def getCookie(self,url):
- header = {
- "Referer": 'https://czspp.com/',
- "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/104.0.0.0 Safari/537.36"
- }
- session = requests.session()
- rsp = session.get(url)
- if '人机验证' in rsp.text:
- append = self.regStr(rsp.text, 'src=\"(/.*?)\"')
- nurl = 'https://czspp.com' + append
- nrsp = session.get(nurl, headers=header)
- key = self.regStr(nrsp.text, 'var key=\"(.*?)\"')
- avalue = self.regStr(nrsp.text, 'value=\"(.*?)\"')
- c = ''
- for i in range(0, len(avalue)):
- a = avalue[i]
- b = ord(a)
- c = c + str(b)
- value = hashlib.md5(c.encode()).hexdigest()
- session.get('https://czspp.com/a20be899_96a6_40b2_88ba_32f1f75f1552_yanzheng_ip.php?type=96c4e20a0e951f471d32dae103e83881&key={0}&value={1}'.format(key, value), headers=header)
- return session.get(url, headers=header)
- elif '检测中' in rsp.text:
- append = self.regStr(rsp.text, 'href =\"(/.*?)\"')
- session.get('https://czspp.com{0}'.format(append), headers=header)
- return session.get(url, headers=header)
- else:
- return rsp
- def categoryContent(self, tid, pg, filter, extend):
- result = {}
- url = 'https://czspp.com/{0}/page/{1}'.format(tid,pg)
- rsp = self.getCookie(url)
- root = self.html(self.cleanText(rsp.text))
- aList = root.xpath("//div[contains(@class,'bt_img mi_ne_kd mrb')]/ul/li")
- videos = []
- for a in aList:
- name = a.xpath('./a/img/@alt')[0]
- pic = a.xpath('./a/img/@data-original')[0]
- mark = a.xpath(".//div[@class='jidi']/span/text()")
- if mark ==[]:
- mark = a.xpath("./div[@class='hdinfo']/span/text()")
- mark = mark[0]
- sid = a.xpath("./a/@href")[0]
- sid = self.regStr(sid, "/movie/(\\S+).html")
- videos.append({
- "vod_id": sid,
- "vod_name": name,
- "vod_pic": pic,
- "vod_remarks": mark
- })
- result['list'] = videos
- result['page'] = pg
- result['pagecount'] = 9999
- result['limit'] = 90
- result['total'] = 999999
- return result
- def detailContent(self, array):
- tid = array[0]
- url = 'https://czspp.com/movie/{0}.html'.format(tid)
- rsp = self.getCookie(url)
- root = self.html(self.cleanText(rsp.text))
- node = root.xpath("//div[@class='dyxingq']")[0]
- pic = node.xpath(".//div[@class='dyimg fl']/img/@src")[0]
- title = node.xpath('.//h1/text()')[0]
- detail = root.xpath(".//div[@class='yp_context']//p/text()")[0]
- vod = {
- "vod_id": tid,
- "vod_name": title,
- "vod_pic": pic,
- "type_name": "",
- "vod_year": "",
- "vod_area": "",
- "vod_remarks": "",
- "vod_actor": "",
- "vod_director": "",
- "vod_content": detail
- }
- infoArray = node.xpath(".//ul[@class='moviedteail_list']/li")
- for info in infoArray:
- content = info.xpath('string(.)')
- if content.startswith('地区'):
- tpyeare = ''
- for inf in info:
- tn = inf.text
- tpyeare = tpyeare +'/'+'{0}'.format(tn)
- vod['vod_area'] = tpyeare.strip('/')
- if content.startswith('年份'):
- vod['vod_year'] = content.replace("年份:","")
- if content.startswith('主演'):
- tpyeact = ''
- for inf in info:
- tn = inf.text
- tpyeact = tpyeact +'/'+'{0}'.format(tn)
- vod['vod_actor'] = tpyeact.strip('/')
- if content.startswith('导演'):
- tpyedire = ''
- for inf in info:
- tn = inf.text
- tpyedire = tpyedire +'/'+'{0}'.format(tn)
- vod['vod_director'] = tpyedire .strip('/')
- vod_play_from = '$$$'
- playFrom = ['厂长']
- vod_play_from = vod_play_from.join(playFrom)
- vod_play_url = '$$$'
- playList = []
- vodList = root.xpath("//div[@class='paly_list_btn']")
- for vl in vodList:
- vodItems = []
- aList = vl.xpath('./a')
- for tA in aList:
- href = tA.xpath('./@href')[0]
- name = tA.xpath('./text()')[0].replace('\xa0','')
- tId = self.regStr(href, '/v_play/(\\S+).html')
- vodItems.append(name + "$" + tId)
- joinStr = '#'
- joinStr = joinStr.join(vodItems)
- playList.append(joinStr)
- vod_play_url = vod_play_url.join(playList)
- vod['vod_play_from'] = vod_play_from
- vod['vod_play_url'] = vod_play_url
- result = {
- 'list': [
- vod
- ]
- }
- return result
- def searchContent(self, key, quick):
- url = 'https://czspp.com/xssearch?q={0}'.format(urllib.parse.quote(key))
- rsp = self.getCookie(url)
- root = self.html(self.cleanText(rsp.text))
- vodList = root.xpath("//div[contains(@class,'mi_ne_kd')]/ul/li/a")
- videos = []
- for vod in vodList:
- name = vod.xpath('./img/@alt')[0]
- pic = vod.xpath('./img/@data-original')[0]
- href = vod.xpath('./@href')[0]
- tid = self.regStr(href, 'movie/(\\S+).html')
- res = vod.xpath('./div[@class="jidi"]/span/text()')
- if len(res) == 0:
- remark = '全1集'
- else:
- remark = vod.xpath('./div[@class="jidi"]/span/text()')[0]
- videos.append({
- "vod_id": tid,
- "vod_name": name,
- "vod_pic": pic,
- "vod_remarks": remark
- })
- result = {
- 'list': videos
- }
- return result
- config = {
- "player": {},
- "filter": {}
- }
- header = {
- "Referer": "https://czspp.com/",
- "User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/100.0.4896.75 Safari/537.36"
- }
- def parseCBC(self, enc, key, iv):
- keyBytes = key.encode("utf-8")
- ivBytes = iv.encode("utf-8")
- cipher = AES.new(keyBytes, AES.MODE_CBC, ivBytes)
- msg = cipher.decrypt(enc)
- paddingLen = msg[len(msg) - 1]
- return msg[0:-paddingLen]
- def playerContent(self, flag, id, vipFlags):
- result = {}
- url = 'https://czspp.com/v_play/{0}.html'.format(id)
- rsp = self.getCookie(url)
- pat = '\\"([^\\"]+)\\";var [\\d\\w]+=function dncry.*md5.enc.Utf8.parse\\(\\"([\\d\\w]+)\\".*md5.enc.Utf8.parse\\(([\\d]+)\\)'
- html = rsp.text
- content = self.regStr(html, pat)
- if content == '':
- str3 = url
- pars = 1
- header = {
- "User-Agent": "Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/86.0.4240.198 Safari/537.36"
- }
- else:
- key = self.regStr(html, pat, 2)
- iv = self.regStr(html, pat, 3)
- decontent = self.parseCBC(base64.b64decode(content), key, iv).decode()
- urlPat = 'video: \\{url: \\\"([^\\\"]+)\\\"'
- vttPat = 'subtitle: \\{url:\\\"([^\\\"]+\\.vtt)\\\"'
- str3 = self.regStr(decontent, urlPat)
- str4 = self.regStr(decontent, vttPat)
- self.loadVtt(str3)
- pars = 0
- header = ''
- if len(str4) > 0:
- result['subf'] = '/vtt/utf-8'
- result['subt'] = ''
- result = {
- 'parse': pars,
- 'playUrl': '',
- 'url': str3,
- 'header': header
- }
- return result
- def loadVtt(self, url):
- pass
- def isVideoFormat(self, url):
- pass
- def manualVideoCheck(self):
- pass
- def localProxy(self, param):
- action = {}
- return [200, "video/MP2T", action, ""]
|