12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577 |
- #!/usr/bin/env python3
- # -*- coding: utf-8 -*-
- # File : cms.py
- # Author: DaShenHan&道长-----先苦后甜,任凭晚风拂柳颜------
- # Date : 2022/8/25
- import json
- # import bs4
- import requests
- import re
- import math
- import ujson
- from utils.web import *
- from utils.system import getHost
- from utils.config import playerConfig
- from utils.log import logger
- from utils.encode import base64Encode,base64Decode,fetch,post,request,getCryptoJS,getPreJs,buildUrl,getHome,atob,btoa
- from utils.encode import verifyCode,setDetail,join,urljoin2,parseText,requireCache,forceOrder,base64ToImage,encodeStr, decodeStr
- from utils.encode import md5 as mmd5
- from utils.safePython import safePython
- from utils.parser import runPy,runJScode,JsObjectWrapper,PyJsObject,PyJsString
- from utils.htmlParser import jsoup
- from urllib.parse import urljoin,quote,unquote
- from concurrent.futures import ThreadPoolExecutor # 引入线程池
- from flask import url_for,redirect,render_template_string
- from easydict import EasyDict as edict
- from controllers.service import storage_service
- def setItem(key,value):
- lsg = storage_service()
- if isinstance(key,PyJsString):
- key = parseText(str(key))
- if isinstance(value,PyJsString):
- value = parseText(str(value))
- return lsg.setItem(key,value)
- def getItem(key,value=''):
- lsg = storage_service()
- if isinstance(key,PyJsString):
- key = parseText(str(key))
- if isinstance(value,PyJsString):
- value = parseText(str(value))
- return lsg.getItem(key,value)
- def clearItem(key):
- lsg = storage_service()
- if isinstance(key,PyJsString):
- key = parseText(str(key))
- return lsg.clearItem(key)
- def encodeUrl(url):
- # return base64Encode(quote(url))
- # return base64Encode(url)
- # print(type(url))
- if isinstance(url,PyJsString):
- # obj = obj.to_dict()
- url = parseText(str(url))
- return quote(url)
- def stringify(obj):
- if isinstance(obj,PyJsObject):
- # obj = obj.to_dict()
- obj = parseText(str(obj))
- return json.dumps(obj, separators=(',', ':'), ensure_ascii=False)
- def requireObj(url):
- if isinstance(url,PyJsString):
- url = parseText(str(url))
- return requireCache(url)
- def md5(text):
- if isinstance(text,PyJsString):
- text = parseText(str(text))
- return mmd5(text)
- py_ctx = {
- 'requests':requests,'print':print,'base64Encode':base64Encode,'base64Decode':base64Decode,
- 'log':logger.info,'fetch':fetch,'post':post,'request':request,'getCryptoJS':getCryptoJS,
- 'buildUrl':buildUrl,'getHome':getHome,'setDetail':setDetail,'join':join,'urljoin2':urljoin2,
- 'PC_UA':PC_UA,'MOBILE_UA':MOBILE_UA,'UC_UA':UC_UA,'UA':UA,'IOS_UA':IOS_UA,
- 'setItem':setItem,'getItem':getItem,'clearItem':clearItem,'stringify':stringify,'encodeUrl':encodeUrl,
- 'requireObj':requireObj,'md5':md5,'atob': atob, 'btoa':btoa,'base64ToImage': base64ToImage, 'encodeStr': encodeStr,
- 'decodeStr': decodeStr
- }
- # print(getCryptoJS())
- class CMS:
- def __init__(self, rule, db=None, RuleClass=None, PlayParse=None,new_conf=None,ext=''):
- if new_conf is None:
- new_conf = {}
- self.lsg = storage_service()
- self.title = rule.get('title', '')
- self.id = rule.get('id', self.title)
- self.filter_url = rule.get('filter_url', '').replace('{{fl}}','{{fl|safe}}') # python jinjia2禁用自动编码
- cate_exclude = rule.get('cate_exclude','')
- tab_exclude = rule.get('tab_exclude','')
- self.lazy = rule.get('lazy', False)
- # self.play_disable = new_conf.get('PLAY_DISABLE',False)
- self.play_disable = self.lsg.getItem('PLAY_DISABLE',False)
- self.retry_count = new_conf.get('RETRY_CNT',3)
- # self.lazy_mode = new_conf.get('LAZYPARSE_MODE')
- self.lazy_mode = self.lsg.getItem('LAZYPARSE_MODE',2)
- self.ocr_api = new_conf.get('OCR_API')
- # self.cate_exclude = new_conf.get('CATE_EXCLUDE','')
- self.cate_exclude = self.lsg.getItem('CATE_EXCLUDE','')
- # self.tab_exclude = new_conf.get('TAB_EXCLUDE','')
- self.tab_exclude = self.lsg.getItem('TAB_EXCLUDE','')
- if cate_exclude:
- if not str(cate_exclude).startswith('|') and not str(self.cate_exclude).endswith('|'):
- self.cate_exclude = self.cate_exclude+'|'+cate_exclude
- else:
- self.cate_exclude += cate_exclude
- if tab_exclude:
- if not str(tab_exclude).startswith('|') and not str(self.tab_exclude).endswith('|'):
- self.tab_exclude = self.tab_exclude+'|'+tab_exclude
- else:
- self.tab_exclude += tab_exclude
- # print(self.cate_exclude)
- try:
- self.vod = redirect(url_for('vod')).headers['Location']
- except:
- self.vod = '/vod'
- # if not self.play_disable and self.lazy:
- if not self.play_disable:
- self.play_parse = rule.get('play_parse', False)
- try:
- play_url = getHost(self.lazy_mode)
- except:
- play_url = getHost(1,5705)
- # play_url = new_conf.get('PLAY_URL',getHost(2))
- if not play_url.startswith('http'):
- play_url = 'http://'+play_url
- # print(play_url)
- if self.play_parse:
- # self.play_url = play_url + self.vod + '?play_url='
- js0_password = self.lsg.getItem('JS0_PASSWORD')
- # print(f'js0密码:{js0_password}')
- js0_password = f'pwd={js0_password}&' if js0_password else ''
- self.play_url = f'{play_url}{self.vod}?{js0_password}rule={self.id}&ext={ext}&play_url='
- # logger.info(f'cms重定向链接:{self.play_url}')
- else:
- self.play_url = ''
- else:
- self.play_parse = False
- self.play_url = ''
- logger.info('播放免嗅地址: '+self.play_url)
- self.db = db
- self.RuleClass = RuleClass
- self.PlayParse = PlayParse
- host = rule.get('host','').rstrip('/')
- host = unquote(host)
- timeout = rule.get('timeout',5000)
- homeUrl = rule.get('homeUrl','/')
- url = rule.get('url','')
- detailUrl = rule.get('detailUrl','')
- searchUrl = rule.get('searchUrl','')
- default_headers = getHeaders(host)
- self_headers = rule.get('headers',{})
- default_headers.update(self_headers)
- headers = default_headers
- cookie = self.getCookie()
- # print(f'{self.title}cookie:{cookie}')
- self.oheaders = self_headers
- if cookie:
- headers['cookie'] = cookie
- self.oheaders['cookie'] = cookie
- limit = rule.get('limit',6)
- encoding = rule.get('编码', 'utf-8')
- search_encoding = rule.get('搜索编码', '')
- self.limit = min(limit,30)
- keys = headers.keys()
- for k in headers.keys():
- if str(k).lower() == 'user-agent':
- v = headers[k]
- if v == 'MOBILE_UA':
- headers[k] = MOBILE_UA
- elif v == 'PC_UA':
- headers[k] = PC_UA
- elif v == 'UC_UA':
- headers[k] = UC_UA
- elif v == 'IOS_UA':
- headers[k] = IOS_UA
- lower_keys = list(map(lambda x:x.lower(),keys))
- if not 'user-agent' in lower_keys:
- headers['User-Agent'] = UA
- if not 'referer' in lower_keys:
- headers['Referer'] = host
- self.headers = headers
- # print(headers)
- self.host = host
- self.homeUrl = urljoin(host,homeUrl) if host and homeUrl else homeUrl or host
- if url.find('[') >-1 and url.find(']') > -1:
- u1 = url.split('[')[0]
- u2 = url.split('[')[1].split(']')[0]
- self.url = urljoin(host,u1)+'['+urljoin(host,u2)+']' if host and url else url
- else:
- self.url = urljoin(host, url) if host and url else url
- self.detailUrl = urljoin(host,detailUrl) if host and detailUrl else detailUrl
- self.searchUrl = urljoin(host,searchUrl) if host and searchUrl else searchUrl
- self.class_name = rule.get('class_name','')
- self.class_url = rule.get('class_url','')
- self.class_parse = rule.get('class_parse','')
- self.filter_name = rule.get('filter_name', '')
- self.filter_url = rule.get('filter_url', '')
- self.filter_parse = rule.get('filter_parse', '')
- self.double = rule.get('double',False)
- self.一级 = rule.get('一级','')
- self.二级 = rule.get('二级','')
- self.二级访问前 = rule.get('二级访问前','')
- self.搜索 = rule.get('搜索','')
- self.推荐 = rule.get('推荐','')
- self.图片来源 = rule.get('图片来源','')
- self.encoding = encoding
- self.search_encoding = search_encoding
- self.timeout = round(int(timeout)/1000,2)
- self.filter = rule.get('filter',[])
- self.filter_def = rule.get('filter_def',{})
- self.play_json = rule['play_json'] if 'play_json' in rule else []
- self.pagecount = rule['pagecount'] if 'pagecount' in rule else {}
- self.extend = rule.get('extend',[])
- self.d = self.getObject()
- def getName(self):
- return self.title
- def getObject(self):
- o = edict({
- 'jsp':jsoup(self.url),
- 'getParse':self.getParse,
- 'saveParse':self.saveParse,
- 'oheaders':self.oheaders,
- 'headers':self.headers, # 通用免嗅需要
- 'encoding':self.encoding,
- 'name':self.title,
- 'timeout':self.timeout,
- })
- return o
- def regexp(self,prule,text,pos=None):
- ret = re.search(prule,text).groups()
- if pos != None and isinstance(pos,int):
- return ret[pos]
- else:
- return ret
- def test(self,text,string):
- searchObj = re.search(rf'{text}', string, re.M | re.I)
- # print(searchObj)
- # global vflag
- if searchObj:
- # vflag = searchObj.group()
- pass
- return searchObj
- def blank(self):
- result = {
- 'list': []
- }
- return result
- def blank_vod(self):
- return {
- "vod_id": "id",
- "vod_name": "片名",
- "vod_pic": "",# 图片
- "type_name": "剧情",
- "vod_year": "年份",
- "vod_area": "地区",
- "vod_remarks": "更新信息",
- "vod_actor": "主演",
- "vod_director": "导演",
- "vod_content": "简介"
- }
- def jsoup(self):
- jsp = jsoup(self.url)
- pdfh = jsp.pdfh
- pdfa = jsp.pdfa
- pd = jsp.pd
- pjfh = jsp.pjfh
- pjfa = jsp.pjfa
- pj = jsp.pj
- pq = jsp.pq
- return pdfh,pdfa,pd,pq
- def getClasses(self):
- if not self.db:
- msg = '未提供数据库连接'
- print(msg)
- return []
- name = self.getName()
- # self.db.metadata.clear()
- # RuleClass = rule_classes.init(self.db)
- res = self.db.session.query(self.RuleClass).filter(self.RuleClass.name == name).first()
- # _logger.info('xxxxxx')
- if res:
- if not all([res.class_name,res.class_url]):
- return []
- cls = res.class_name.split('&')
- cls2 = res.class_url.split('&')
- classes = [{'type_name':cls[i],'type_id':cls2[i]} for i in range(len(cls))]
- # _logger.info(classes)
- logger.info(f"{self.getName()}使用缓存分类:{classes}")
- return classes
- else:
- return []
- def getCookie(self):
- name = self.getName()
- if not self.db:
- msg = f'{name}未提供数据库连接'
- print(msg)
- return False
- res = self.db.session.query(self.RuleClass).filter(self.RuleClass.name == name).first()
- if res:
- return res.cookie or None
- else:
- return None
- def saveCookie(self,cookie):
- name = self.getName()
- if not self.db:
- msg = f'{name}未提供数据库连接'
- print(msg)
- return False
- res = self.db.session.query(self.RuleClass).filter(self.RuleClass.name == name).first()
- if res:
- res.cookie = cookie
- self.db.session.add(res)
- else:
- res = self.RuleClass(name=name, cookie=cookie)
- self.db.session.add(res)
- try:
- self.db.session.commit()
- logger.info(f'{name}已保存cookie:{cookie}')
- except Exception as e:
- return f'保存cookie发生了错误:{e}'
- def saveClass(self, classes):
- if not self.db:
- msg = '未提供数据库连接'
- print(msg)
- return msg
- name = self.getName()
- class_name = '&'.join([cl['type_name'] for cl in classes])
- class_url = '&'.join([cl['type_id'] for cl in classes])
- # data = RuleClass.query.filter(RuleClass.name == '555影视').all()
- # self.db.metadata.clear()
- # RuleClass = rule_classes.init(self.db)
- res = self.db.session.query(self.RuleClass).filter(self.RuleClass.name == name).first()
- # print(res)
- if res:
- res.class_name = class_name
- res.class_url = class_url
- self.db.session.add(res)
- msg = f'{self.getName()}修改成功:{res.id}'
- else:
- res = self.RuleClass(name=name, class_name=class_name, class_url=class_url)
- self.db.session.add(res)
- res = self.db.session.query(self.RuleClass).filter(self.RuleClass.name == name).first()
- msg = f'{self.getName()}新增成功:{res.id}'
- try:
- self.db.session.commit()
- logger.info(msg)
- except Exception as e:
- return f'发生了错误:{e}'
- def getParse(self,play_url):
- if not self.db:
- msg = '未提供数据库连接'
- print(msg)
- return ''
- name = self.getName()
- # self.db.metadata.clear()
- # RuleClass = rule_classes.init(self.db)
- res = self.db.session.query(self.PlayParse).filter(self.PlayParse.play_url == play_url).first()
- # _logger.info('xxxxxx')
- if res:
- real_url = res.real_url
- logger.info(f"{name}使用缓存播放地址:{real_url}")
- return real_url
- else:
- return ''
- def dealJson(self,html):
- try:
- # res = re.search('.*?\{(.*)\}',html,re.M|re.I).groups()[0]
- res = re.search('.*?\{(.*)\}',html,re.M|re.S).groups()[0]
- html = '{' + res + '}'
- return html
- except:
- return html
- def checkHtml(self,r):
- r.encoding = self.encoding
- html = r.text
- if html.find('?btwaf=') > -1:
- btwaf = re.search('btwaf(.*?)"',html,re.M|re.I).groups()[0]
- url = r.url.split('#')[0]+'?btwaf'+btwaf
- # print(f'需要过宝塔验证:{url}')
- cookies_dict = requests.utils.dict_from_cookiejar(r.cookies)
- cookie_str = ';'.join([f'{k}={cookies_dict[k]}' for k in cookies_dict])
- self.headers['cookie'] = cookie_str
- r = requests.get(url, headers=self.headers, timeout=self.timeout,verify=False)
- r.encoding = self.encoding
- html = r.text
- if html.find('?btwaf=') < 0:
- self.saveCookie(cookie_str)
- # print(html)
- return html
- def saveParse(self, play_url,real_url):
- if not self.db:
- msg = '未提供数据库连接'
- print(msg)
- return msg
- name = self.getName()
- # data = RuleClass.query.filter(RuleClass.name == '555影视').all()
- # self.db.metadata.clear()
- # RuleClass = rule_classes.init(self.db)
- res = self.db.session.query(self.PlayParse).filter(self.PlayParse.play_url == play_url).first()
- # print(res)
- if res:
- res.real_url = real_url
- self.db.session.add(res)
- msg = f'{name}服务端免嗅修改成功:{res.id}'
- else:
- res = self.PlayParse(play_url=play_url, real_url=real_url)
- self.db.session.add(res)
- res = self.db.session.query(self.PlayParse).filter(self.PlayParse.play_url == play_url).first()
- msg = f'{name}服务端免嗅新增成功:{res.id}'
- try:
- self.db.session.commit()
- logger.info(msg)
- except Exception as e:
- return f'{name}发生了错误:{e}'
- def homeContent(self,fypage=1):
- # yanaifei
- # https://yanetflix.com/vodtype/dianying.html
- t1 = time()
- result = {}
- classes = []
- video_result = self.blank()
- if self.class_url and self.class_name:
- class_names = self.class_name.split('&')
- class_urls = self.class_url.split('&')
- cnt = min(len(class_urls), len(class_names))
- for i in range(cnt):
- classes.append({
- 'type_name': class_names[i],
- 'type_id': class_urls[i]
- })
- # print(self.url)
- print(self.headers)
- has_cache = False
- # print(self.homeUrl)
- if self.homeUrl.startswith('http'):
- # print(self.class_parse)
- try:
- if self.class_parse:
- t2 = time()
- cache_classes = self.getClasses()
- logger.info(f'{self.getName()}读取缓存耗时:{get_interval(t2)}毫秒')
- if len(cache_classes) > 0:
- classes = cache_classes
- # print(cache_classes)
- has_cache = True
- # logger.info(f'是否有缓存分类:{has_cache}')
- if has_cache and not self.推荐:
- pass
- else:
- new_classes = []
- r = requests.get(self.homeUrl, headers=self.headers, timeout=self.timeout,verify=False)
- html = self.checkHtml(r)
- # print(html)
- # print(self.headers)
- if self.class_parse and not has_cache:
- p = self.class_parse.split(';')
- # print(p[0])
- # print(html)
- jsp = jsoup(self.url)
- pdfh = jsp.pdfh
- pdfa = jsp.pdfa
- pd = jsp.pd
- items = pdfa(html,p[0])
- # print(len(items))
- # print(items)
- for item in items:
- title = pdfh(item, p[1])
- # 过滤排除掉标题名称
- if self.cate_exclude and jsp.test(self.cate_exclude, title):
- continue
- url = pd(item, p[2])
- # print(url)
- tag = url
- if len(p) > 3 and p[3].strip():
- try:
- tag = self.regexp(p[3].strip(),url,0)
- except:
- logger.info(f'分类匹配错误:{title}对应的链接{url}无法匹配{p[3]}')
- continue
- new_classes.append({
- 'type_name': title,
- 'type_id': tag
- })
- if len(new_classes) > 0:
- classes.extend(new_classes)
- self.saveClass(classes)
- video_result = self.homeVideoContent(html,fypage)
- except Exception as e:
- logger.info(f'{self.getName()}主页发生错误:{e}')
- classes = list(filter(lambda x:not self.cate_exclude or not jsoup(self.url).test(self.cate_exclude, x['type_name']),classes))
- result['class'] = classes
- if self.filter:
- if isinstance(self.filter,dict):
- result['filters'] = self.filter
- else:
- result['filters'] = playerConfig['filter']
- result.update(video_result)
- # print(result)
- logger.info(f'{self.getName()}获取首页总耗时(包含读取缓存):{get_interval(t1)}毫秒')
- return result
- def homeVideoContent(self,html,fypage=1):
- p = self.推荐
- if not p:
- return self.blank()
- jsp = jsoup(self.homeUrl)
- result = {}
- videos = []
- is_js = isinstance(p, str) and str(p).strip().startswith('js:') # 是js
- if is_js:
- headers['Referer'] = getHome(self.host)
- py_ctx.update({
- 'input': self.homeUrl,
- 'HOST': self.host,
- 'TYPE': 'home', # 海阔js环境标志
- 'oheaders':self.d.oheaders,
- 'fetch_params': {'headers': self.headers, 'timeout': self.d.timeout, 'encoding': self.d.encoding},
- 'd': self.d,
- 'getParse': self.d.getParse,
- 'saveParse': self.d.saveParse,
- 'jsp': jsp,'jq':jsp,'setDetail': setDetail,
- })
- ctx = py_ctx
- jscode = getPreJs() + p.strip().replace('js:', '', 1)
- # print(jscode)
- try:
- loader, _ = runJScode(jscode, ctx=ctx)
- # print(loader.toString())
- vods = loader.eval('VODS')
- # print(vods)
- if isinstance(vods, JsObjectWrapper):
- videos = vods.to_list()
- except Exception as e:
- logger.info(f'首页推荐执行js获取列表出错:{e}')
- else:
- if p == '*' and self.一级:
- p = self.一级
- self.double = False
- logger.info(f'首页推荐继承一级: {p}')
- p = p.strip().split(';') # 解析
- if not self.double and len(p) < 5:
- return self.blank()
- if self.double and len(p) < 6:
- return self.blank()
- jsp = jsoup(self.homeUrl)
- pp = self.一级.split(';')
- def getPP(p,pn,pp,ppn):
- try:
- ps = pp[ppn] if p[pn] == '*' and len(pp) > ppn else p[pn]
- return ps
- except Exception as e:
- return ''
- p0 = getPP(p,0,pp,0)
- is_json = str(p0).startswith('json:')
- if is_json:
- html = self.dealJson(html)
- pdfh = jsp.pjfh if is_json else jsp.pdfh
- pdfa = jsp.pjfa if is_json else jsp.pdfa
- pd = jsp.pj if is_json else jsp.pd
- # print(html)
- try:
- if self.double:
- items = pdfa(html, p0.replace('json:',''))
- # print(p[0])
- # print(items)
- # print(len(items))
- p1 = getPP(p, 1, pp, 0)
- p2 = getPP(p, 2, pp, 1)
- p3 = getPP(p, 3, pp, 2)
- p4 = getPP(p, 4, pp, 3)
- p5 = getPP(p, 5, pp, 4)
- p6 = getPP(p, 6, pp, 5)
- for item in items:
- items2 = pdfa(item,p1)
- # print(len(items2))
- for item2 in items2:
- try:
- title = pdfh(item2, p2)
- # print(title)
- try:
- img = pd(item2, p3)
- except:
- img = ''
- try:
- desc = pdfh(item2, p4)
- except:
- desc = ''
- links = [pd(item2, _p5) if not self.detailUrl else pdfh(item2, _p5) for _p5 in p5.split('+')]
- vid = '$'.join(links)
- if len(p) > 6 and p[6]:
- content = pdfh(item2, p6)
- else:
- content = ''
- if self.二级 == '*':
- vid = vid + '@@' + title + '@@' + img
- videos.append({
- "vod_id": vid,
- "vod_name": title,
- "vod_pic": img,
- "vod_remarks": desc,
- "no_use":{
- "vod_content": content,
- "type_id": 1,
- "type_name": "首页推荐",
- },
- })
- except:
- pass
- else:
- items = pdfa(html, p0.replace('json:',''))
- # print(items)
- p1 = getPP(p, 1, pp, 1)
- p2 = getPP(p, 2, pp, 2)
- p3 = getPP(p, 3, pp, 3)
- p4 = getPP(p, 4, pp, 4)
- p5 = getPP(p, 5, pp, 5)
- for item in items:
- try:
- title = pdfh(item, p1)
- try:
- img = pd(item, p2)
- except:
- img = ''
- try:
- desc = pdfh(item, p3)
- except:
- desc = ''
- # link = pd(item, p[4])
- links = [pd(item, _p5) if not self.detailUrl else pdfh(item, _p5) for _p5 in p4.split('+')]
- vid = '$'.join(links)
- if len(p) > 5 and p[5]:
- content = pdfh(item, p5)
- else:
- content = ''
- if self.二级 == '*':
- vid = vid + '@@' + title + '@@' + img
- videos.append({
- "vod_id": vid,
- "vod_name": title,
- "vod_pic": img,
- "vod_remarks": desc,
- "no_use": {
- "vod_content": content,
- "type_id": 1,
- "type_name": "首页推荐",
- },
- })
- except:
- pass
- # result['list'] = videos[min((fypage-1)*self.limit,len(videos)-1):min(fypage*self.limit,len(videos))]
- except Exception as e:
- logger.info(f'首页内容获取失败:{e}')
- return self.blank()
- if self.图片来源:
- for video in videos:
- if video.get('vod_pic','') and str(video['vod_pic']).startswith('http'):
- video['vod_pic'] = f"{video['vod_pic']}{self.图片来源}"
- result['list'] = videos
- # print(videos)
- result['no_use'] = {
- 'code': 1,
- 'msg': '数据列表',
- 'page': fypage,
- 'pagecount': math.ceil(len(videos) / self.limit),
- 'limit': self.limit,
- 'total': len(videos),
- 'now_count': len(result['list']),
- }
- # print(result)
- return result
- def categoryContent(self, fyclass, fypage, fl=None):
- """
- 一级带分类的数据返回
- :param fyclass: 分类标识
- :param fypage: 页码
- :param fl: 筛选
- :return: cms一级数据
- """
- if fl is None:
- fl = {}
- # print(f'fl:{fl}')
- if self.filter_def and isinstance(self.filter_def,dict):
- try:
- if self.filter_def.get(fyclass) and isinstance(self.filter_def[fyclass],dict):
- self_filter_def = self.filter_def[fyclass]
- filter_def = ujson.loads(ujson.dumps(self_filter_def))
- filter_def.update(fl)
- fl = filter_def
- except Exception as e:
- print(f'合并不同分类对应的默认筛选出错:{e}')
- # print(fl)
- result = {}
- # urlParams = ["", "", "", "", "", "", "", "", "", "", "", ""]
- # urlParams = [""] * 12
- # urlParams[0] = tid
- # urlParams[8] = str(pg)
- # for key in self.extend:
- # urlParams[int(key)] = self.extend[key]
- # params = '-'.join(urlParams)
- # print(params)
- # url = self.url + '/{0}.html'.format
- t1 = time()
- pg = str(fypage)
- url = self.url.replace('fyclass',fyclass)
- if fypage == 1 and self.test('[\[\]]',url):
- url = url.split('[')[1].split(']')[0]
- elif fypage > 1 and self.test('[\[\]]',url):
- url = url.split('[')[0]
- if self.filter_url:
- if not 'fyfilter' in url: # 第一种情况,默认不写fyfilter关键字,视为直接拼接在链接后面当参数
- if not url.endswith('&') and not self.filter_url.startswith('&'):
- url += '&'
- url += self.filter_url
- else: # 第二种情况直接替换关键字为待拼接的结果后面渲染,适用于 ----fypage.html的情况
- url = url.replace('fyfilter', self.filter_url)
- # print(f'url渲染:{url}')
- url = render_template_string(url,fl=fl)
- # fl_url = render_template_string(self.filter_url,fl=fl)
- # if not 'fyfilter' in url: # 第一种情况,默认不写fyfilter关键字,视为直接拼接在链接后面当参数
- # if not url.endswith('&') and not fl_url.startswith('&'):
- # url += '&'
- # url += fl_url
- # else: # 第二种情况直接替换关键字为渲染后的结果,适用于 ----fypage.html的情况
- # url = url.replace('fyfilter',fl_url)
- if url.find('fypage') > -1:
- if '(' in url and ')' in url:
- # url_rep = url[url.find('('):url.find(')')+1]
- # cnt_page = url.split('(')[1].split(')')[0].replace('fypage',pg)
- # print(url_rep)
- url_rep = re.search('.*?\((.*)\)',url,re.M|re.S).groups()[0]
- cnt_page = url_rep.replace('fypage', pg)
- # print(url_rep)
- # print(cnt_page)
- cnt_ctx = {}
- exec(f'cnt_pg={cnt_page}', cnt_ctx)
- cnt_pg = str(cnt_ctx['cnt_pg']) # 计算表达式的结果
- url = url.replace(url_rep,str(cnt_pg)).replace('(','').replace(')','')
- # print(url)
- else:
- url = url.replace('fypage',pg)
- # print(url)
- logger.info(url)
- p = self.一级
- jsp = jsoup(self.url)
- videos = []
- is_js = isinstance(p, str) and str(p).startswith('js:') # 是js
- if is_js:
- headers['Referer'] = getHome(url)
- py_ctx.update({
- 'input': url,
- 'TYPE': 'cate', # 海阔js环境标志
- 'oheaders': self.d.oheaders,
- 'fetch_params': {'headers': self.headers, 'timeout': self.d.timeout, 'encoding': self.d.encoding},
- 'd': self.d,
- 'MY_CATE':fyclass, # 分类id
- 'MY_FL':fl, # 筛选
- 'MY_PAGE':fypage, # 页数
- 'detailUrl':self.detailUrl or '', # 详情页链接
- 'getParse': self.d.getParse,
- 'saveParse': self.d.saveParse,
- 'jsp': jsp,'jq':jsp, 'setDetail': setDetail,
- })
- ctx = py_ctx
- # print(ctx)
- jscode = getPreJs() + p.replace('js:', '', 1)
- # print(jscode)
- loader, _ = runJScode(jscode, ctx=ctx)
- # print(loader.toString())
- vods = loader.eval('VODS')
- # print('vods:',vods)
- if isinstance(vods, JsObjectWrapper):
- videos = vods.to_list()
- else:
- p = p.split(';') # 解析
- # print(len(p))
- # print(p)
- if len(p) < 5:
- return self.blank()
- is_json = str(p[0]).startswith('json:')
- pdfh = jsp.pjfh if is_json else jsp.pdfh
- pdfa = jsp.pjfa if is_json else jsp.pdfa
- pd = jsp.pj if is_json else jsp.pd
- # print(pdfh(r.text,'body a.module-poster-item.module-item:eq(1)&&Text'))
- # print(pdfh(r.text,'body a.module-poster-item.module-item:eq(0)'))
- # print(pdfh(r.text,'body a.module-poster-item.module-item:first'))
- items = []
- try:
- r = requests.get(url, headers=self.headers, timeout=self.timeout,verify=False)
- html = self.checkHtml(r)
- print(self.headers)
- # print(html)
- if is_json:
- html = self.dealJson(html)
- html = json.loads(html)
- # else:
- # soup = bs4.BeautifulSoup(html, 'lxml')
- # html = soup.prettify()
- # print(html)
- # with open('1.html',mode='w+',encoding='utf-8') as f:
- # f.write(html)
- items = pdfa(html,p[0].replace('json:','',1))
- except:
- pass
- # print(items)
- for item in items:
- # print(item)
- try:
- title = pdfh(item, p[1])
- img = pd(item, p[2])
- desc = pdfh(item, p[3])
- links = [pd(item, p4) if not self.detailUrl else pdfh(item, p4) for p4 in p[4].split('+')]
- link = '$'.join(links)
- content = '' if len(p) < 6 else pdfh(item, p[5])
- # sid = self.regStr(sid, "/video/(\\S+).html")
- vod_id = f'{fyclass}${link}' if self.detailUrl else link # 分类,播放链接
- if self.二级 == '*':
- vod_id = vod_id+'@@'+title+'@@'+img
- videos.append({
- "vod_id": vod_id,
- "vod_name": title,
- "vod_pic": img,
- "vod_remarks": desc,
- "vod_content": content,
- })
- except Exception as e:
- print(f'发生了错误:{e}')
- pass
- if self.图片来源:
- for video in videos:
- if video.get('vod_pic','') and str(video['vod_pic']).startswith('http'):
- video['vod_pic'] = f"{video['vod_pic']}{self.图片来源}"
- print('videos:',videos)
- limit = 40
- cnt = 9999 if len(videos) > 0 else 0
- pagecount = 0
- if self.pagecount and isinstance(self.pagecount,dict) and fyclass in self.pagecount:
- print(f'fyclass:{fyclass},self.pagecount:{self.pagecount}')
- pagecount = int(self.pagecount[fyclass])
- result['list'] = videos
- result['page'] = fypage
- result['pagecount'] = pagecount or max(cnt,fypage)
- result['limit'] = limit
- result['total'] = cnt
- # print(result)
- # print(result['pagecount'])
- logger.info(f'{self.getName()}获取分类{fyclass}第{fypage}页耗时:{get_interval(t1)}毫秒,共计{round(len(str(result)) / 1000, 2)} kb')
- nodata = {
- 'list': [{'vod_name': '无数据,防无限请求', 'vod_id': 'no_data', 'vod_remarks': '不要点,会崩的',
- 'vod_pic': 'https://ghproxy.com/https://raw.githubusercontent.com/hjdhnx/dr_py/main/404.jpg'}],
- 'total': 1, 'pagecount': 1, 'page': 1, 'limit': 1
- }
- # return result
- return result if len(result['list']) > 0 else nodata
- def 二级渲染(self,parse_str:'str|dict',**kwargs):
- # *args是不定长参数 列表
- # ** args是不定长参数字典
- p = parse_str # 二级传递解析表达式 js的obj json对象
- detailUrl = kwargs.get('detailUrl','') # 不定长字典传递的二级详情页vod_id详情处理数据
- orId = kwargs.get('orId','') # 不定长字典传递的二级详情页vod_id原始数据
- url = kwargs.get('url','') # 不定长字典传递的二级详情页链接智能拼接数据
- vod = kwargs.get('vod',self.blank_vod()) # 最终要返回的二级详情页数据 默认空
- html = kwargs.get('html','') # 不定长字典传递的源码(如果不传才会在下面程序中去获取)
- show_name = kwargs.get('show_name','') # 是否显示来源(用于drpy区分)
- jsp = kwargs.get('jsp','') # jsp = jsoup(self.url) 传递的jsp解析
- fyclass = kwargs.get('fyclass','') # 二级传递的分类名称,可以得知进去的类别
- play_url = self.play_url
- vod_name = '片名'
- vod_pic = ''
- # print('二级url:',url)
- if self.二级 == '*':
- extra = orId.split('@@')
- vod_name = extra[1] if len(extra) > 1 else vod_name
- vod_pic = extra[2] if len(extra) > 2 else vod_pic
- if self.play_json:
- play_url = play_url.replace('&play_url=', '&type=json&play_url=')
- if p == '*': # 解析表达式为*默认一级直接变播放
- vod['vod_play_from'] = '道长在线'
- vod['vod_remarks'] = detailUrl
- vod['vod_actor'] = '没有二级,只有一级链接直接嗅探播放'
- # vod['vod_content'] = url if not show_name else f'({self.id}) {url}'
- vod['vod_content'] = url
- vod['vod_id'] = orId
- vod['vod_name'] = vod_name
- vod['vod_pic'] = vod_pic
- vod['vod_play_url'] = '嗅探播放$' + play_url + url.split('@@')[0]
- elif not p or (not isinstance(p, dict) and not isinstance(p, str)) or (isinstance(p, str) and not str(p).startswith('js:')):
- pass
- else:
- is_json = p.get('is_json', False) if isinstance(p, dict) else False # 二级里加is_json参数
- pdfh = jsp.pjfh if is_json else jsp.pdfh
- pdfa = jsp.pjfa if is_json else jsp.pdfa
- pd = jsp.pj if is_json else jsp.pd
- pq = jsp.pq
- vod['vod_id'] = orId
- if not html: # 没传递html参数接detailUrl下来智能获取
- r = requests.get(url, headers=self.headers, timeout=self.timeout,verify=False)
- html = self.checkHtml(r)
- if is_json:
- html = self.dealJson(html)
- html = json.loads(html)
- tt1 = time()
- if p.get('title'):
- p1 = p['title'].split(';')
- vod['vod_name'] = pdfh(html, p1[0]).replace('\n', ' ').strip()
- vod['type_name'] = pdfh(html, p1[1]).replace('\n',' ').strip() if len(p1)>1 else ''
- if p.get('desc'):
- try:
- p1 = p['desc'].split(';')
- vod['vod_remarks'] = pdfh(html, p1[0]).replace('\n', '').strip()
- vod['vod_year'] = pdfh(html, p1[1]).replace('\n', ' ').strip() if len(p1) > 1 else ''
- vod['vod_area'] = pdfh(html, p1[2]).replace('\n', ' ').strip() if len(p1) > 2 else ''
- vod['vod_actor'] = pdfh(html, p1[3]).replace('\n', ' ').strip() if len(p1) > 3 else ''
- vod['vod_director'] = pdfh(html, p1[4]).replace('\n', ' ').strip() if len(p1) > 4 else ''
- except:
- pass
- if p.get('content'):
- p1 = p['content'].split(';')
- try:
- content = '\n'.join([pdfh(html, i).replace('\n', ' ') for i in p1])
- vod['vod_content'] = content
- except:
- pass
- if p.get('img'):
- p1 = p['img']
- try:
- img = pd(html, p1)
- vod['vod_pic'] = img
- except Exception as e:
- logger.info(f'二级图片定位失败,但不影响使用{e}')
- vod_play_from = '$$$'
- playFrom = []
- init_flag = {'ctx':False}
- def js_pre():
- headers['Referer'] = getHome(url)
- py_ctx.update({
- 'input': url,
- 'html': html,
- 'TYPE': 'detail', # 海阔js环境标志
- 'MY_CATE': fyclass, # 分类id
- 'oheaders': self.d.oheaders,
- 'fetch_params': {'headers': self.headers, 'timeout': self.d.timeout, 'encoding': self.d.encoding},
- 'd': self.d,
- 'getParse': self.d.getParse,
- 'saveParse': self.d.saveParse,
- 'jsp': jsp,'jq':jsp, 'setDetail': setDetail,'play_url':play_url
- })
- init_flag['ctx'] = True
- if p.get('重定向') and str(p['重定向']).startswith('js:'):
- if not init_flag['ctx']:
- js_pre()
- ctx = py_ctx
- # print(ctx)
- rcode = p['重定向'].replace('js:', '', 1)
- jscode = getPreJs() + rcode
- # print(jscode)
- loader, _ = runJScode(jscode, ctx=ctx)
- # print(loader.toString())
- logger.info(f'开始执行二级重定向代码:{rcode}')
- html = loader.eval('html')
- if isinstance(vod, JsObjectWrapper):
- html = str(html)
- if p.get('tabs'):
- vodHeader = []
- if str(p['tabs']).startswith('js:'):
- if not init_flag['ctx']:
- js_pre()
- ctx = py_ctx
- rcode = p['tabs'].replace('js:', '', 1)
- jscode = getPreJs() + rcode
- # print(jscode)
- loader, _ = runJScode(jscode, ctx=ctx)
- # print(loader.toString())
- logger.info(f'开始执行tabs代码:{rcode}')
- vHeader = loader.eval('TABS')
- if isinstance(vod, JsObjectWrapper):
- vHeader = vHeader.to_list()
- vodHeader = vHeader
- else:
- tab_parse = p['tabs'].split(';')[0]
- # print('tab_parse:',tab_parse)
- vHeader = pdfa(html, tab_parse)
- # print(vHeader)
- print(f'二级线路定位列表数:{len((vHeader))}')
- # print(vHeader[0].outerHtml())
- # print(vHeader[0].toString())
- # from lxml import etree
- # print(str(etree.tostring(vHeader[0], pretty_print=True), 'utf-8'))
- from lxml.html import tostring as html2str
- # print(html2str(vHeader[0].root).decode('utf-8'))
- tab_text = p.get('tab_text','') or 'body&&Text'
- # print('tab_text:'+tab_text)
- if not is_json:
- for v in vHeader:
- # 过滤排除掉线路标题
- # v_title = pq(v).text()
- v_title = pdfh(v,tab_text).strip()
- # print(v_title)
- if self.tab_exclude and jsp.test(self.tab_exclude, v_title):
- continue
- vodHeader.append(v_title)
- else:
- vodHeader = vHeader
- print(f'过滤后真实线路列表数:{len((vodHeader))} {vodHeader}')
- else:
- vodHeader = ['道长在线']
- # print(vodHeader)
- # print(vod)
- new_map = {}
- for v in vodHeader:
- if not v in new_map:
- new_map[v] = 1
- else:
- new_map[v] += 1
- if new_map[v] > 1:
- v = f'{v}{new_map[v]-1}'
- playFrom.append(v)
- vod_play_from = vod_play_from.join(playFrom)
- vod_play_url = '$$$'
- vod_tab_list = []
- if p.get('lists'):
- if str(p['lists']).startswith('js:'):
- if not init_flag['ctx']:
- js_pre()
- ctx = py_ctx
- ctx['TABS'] = vodHeader # 把选集列表传过去
- rcode = p['lists'].replace('js:', '', 1)
- jscode = getPreJs() + rcode
- # print(jscode)
- loader, _ = runJScode(jscode, ctx=ctx)
- # print(loader.toString())
- logger.info(f'开始执行lists代码:{rcode}')
- vlists = loader.eval('LISTS')
- if isinstance(vod, JsObjectWrapper):
- vlists = vlists.to_list() # [['第1集$http://1.mp4','第2集$http://2.mp4'],['第3集$http://1.mp4','第4集$http://2.mp4']]
- for i in range(len(vlists)):
- try:
- vlists[i] = list(map(lambda x:'$'.join(x.split('$')[:2]),vlists[i]))
- except Exception as e:
- logger.info(f'LISTS格式化发生错误:{e}')
- vod_play_url = vod_play_url.join(list(map(lambda x:'#'.join(x),vlists)))
- else:
- list_text = p.get('list_text','') or 'body&&Text'
- list_url = p.get('list_url','') or 'a&&href'
- print('list_text:' + list_text)
- print('list_url:' + list_url)
- is_tab_js = p['tabs'].strip().startswith('js:')
- for i in range(len(vodHeader)):
- tab_name = str(vodHeader[i])
- # print(tab_name)
- tab_ext = p['tabs'].split(';')[1] if len(p['tabs'].split(';')) > 1 and not is_tab_js else ''
- p1 = p['lists'].replace('#idv', tab_name).replace('#id', str(i))
- tab_ext = tab_ext.replace('#idv', tab_name).replace('#id', str(i))
- # print(p1)
- vodList = pdfa(html, p1) # 1条线路的选集列表
- # print(vodList)
- # vodList = [pq(i).text()+'$'+pd(i,'a&&href') for i in vodList] # 拼接成 名称$链接
- # pq(i).text()
- if self.play_parse: # 自动base64编码
- vodList = [(pdfh(html, tab_ext) if tab_ext else tab_name) + '$' + play_url + encodeUrl(i) for i
- in vodList] if is_json else \
- [pdfh(i,list_text) + '$' + play_url + encodeUrl(pd(i, list_url)) for i in vodList] # 拼接成 名称$链接
- else:
- vodList = [(pdfh(html, tab_ext) if tab_ext else tab_name) + '$' + play_url + i for i in
- vodList] if is_json else \
- [pdfh(i,list_text) + '$' + play_url + pd(i, list_url) for i in vodList] # 拼接成 名称$链接
- # print(vodList)
- vodList = forceOrder(vodList,option=lambda x:x.split('$')[0])
- # print(vodList)
- vlist = '#'.join(vodList) # 拼多个选集
- # print(vlist)
- vod_tab_list.append(vlist)
- vod_play_url = vod_play_url.join(vod_tab_list)
- vod_play_url_str = vod_play_url[:min(len(vod_play_url),500)]
- print(vod_play_url_str)
- vod['vod_play_from'] = vod_play_from
- # print(vod_play_from)
- vod['vod_play_url'] = vod_play_url
- logger.info(f'{self.getName()}仅二级渲染{len(vod_play_url.split("$$$")[0].split("$"))}集耗时:{get_interval(tt1)}毫秒,共计{round(len(str(vod)) / 1000, 2)} kb')
- if show_name:
- vod['vod_content'] = f'({self.id}){vod.get("vod_content", "")}'
- return vod
- def detailOneVod(self,id,fyclass='',show_name=False):
- vod = self.blank_vod()
- orId = str(id)
- orUrl = orId
- if fyclass:
- orUrl = f'{fyclass}${orId}'
- detailUrl = orId.split('@@')[0]
- # print(detailUrl)
- if not detailUrl.startswith('http') and not '/' in detailUrl:
- url = self.detailUrl.replace('fyid', detailUrl).replace('fyclass',fyclass)
- # print(url)
- elif '/' in detailUrl:
- url = urljoin(self.homeUrl,detailUrl)
- else:
- url = detailUrl
- if self.二级访问前:
- logger.info(f'尝试在二级访问前执行代码: {self.二级访问前}')
- py_ctx.update({
- 'MY_URL': url,
- 'oheaders': self.d.oheaders,
- 'fetch_params': {'headers': self.headers, 'timeout': self.d.timeout, 'encoding': self.d.encoding},
- 'd': self.d,
- })
- ctx = py_ctx
- jscode = getPreJs() + self.二级访问前.replace('js:', '', 1)
- # print(jscode)
- loader, _ = runJScode(jscode, ctx=ctx)
- try:
- MY_URL = loader.eval('MY_URL')
- if isinstance(MY_URL, JsObjectWrapper):
- MY_URL = str(MY_URL)
- if MY_URL:
- url = MY_URL
- except Exception as e:
- logger.info(f'执行二级访问前发生错误: {e}')
- logger.info(f'进入详情页: {url}')
- try:
- p = self.二级 # 解析
- jsp = jsoup(url) if url.startswith('http') else jsoup(self.url)
- is_js = isinstance(p,str) and str(p).startswith('js:') # 是js
- if is_js:
- headers['Referer'] = getHome(url)
- play_url = self.play_url
- if self.play_json:
- play_url = play_url.replace('&play_url=', '&type=json&play_url=')
- py_ctx.update({
- 'input': url,
- 'TYPE': 'detail', # 海阔js环境标志
- # 'VID': id, # 传递的vod_id
- '二级': self.二级渲染, # 二级解析函数,可以解析dict
- 'MY_CATE': fyclass, # 分类id
- 'oheaders': self.d.oheaders,
- 'fetch_params': {'headers': self.headers, 'timeout': self.d.timeout, 'encoding': self.d.encoding},
- 'd': self.d,
- 'getParse': self.d.getParse,
- 'saveParse': self.d.saveParse,
- 'jsp':jsp,'jq':jsp,'setDetail':setDetail,'play_url':play_url
- })
- ctx = py_ctx
- # print(ctx)
- jscode = getPreJs() + p.replace('js:','',1)
- # print(jscode)
- loader, _ = runJScode(jscode, ctx=ctx)
- # print(loader.toString())
- vod = loader.eval('VOD')
- if isinstance(vod,JsObjectWrapper):
- vod = vod.to_dict()
- if show_name:
- vod['vod_content'] = f'({self.id}){vod.get("vod_content", "")}'
- else:
- vod = self.blank_vod()
- else:
- vod = self.二级渲染(p,detailUrl=detailUrl,orId=orUrl,url=url,vod=vod,show_name=show_name,jsp=jsp,fyclass=fyclass)
- except Exception as e:
- logger.info(f'{self.getName()}获取单个详情页{detailUrl}出错{e}')
- if self.图片来源:
- if vod.get('vod_pic','') and str(vod['vod_pic']).startswith('http'):
- vod['vod_pic'] = f"{vod['vod_pic']}{self.图片来源}"
- if not vod.get('vod_id') or ('$' in orUrl and vod['vod_id']!=orUrl):
- vod['vod_id'] = orUrl
- # print(vod)
- return vod
- def detailContent(self, fypage, array,show_name=False):
- """
- cms二级数据
- :param array:
- :return:
- """
- # print('进入二级')
- t1 = time()
- array = array if len(array) <= self.limit else array[(fypage-1)*self.limit:min(self.limit*fypage,len(array))]
- thread_pool = ThreadPoolExecutor(min(self.limit,len(array))) # 定义线程池来启动多线程执行此任务
- obj_list = []
- try:
- for vod_url in array:
- print(vod_url)
- vod_class = ''
- if vod_url.find('$') > -1:
- tmp = vod_url.split('$')
- vod_class = tmp[0]
- vod_url = tmp[1]
- obj = thread_pool.submit(self.detailOneVod, vod_url,vod_class,show_name)
- obj_list.append(obj)
- thread_pool.shutdown(wait=True) # 等待所有子线程并行完毕
- vod_list = [obj.result() for obj in obj_list]
- result = {
- 'list': vod_list
- }
- logger.info(f'{self.getName()}获取详情页耗时:{get_interval(t1)}毫秒,共计{round(len(str(result)) / 1000, 2)} kb')
- except Exception as e:
- result = {
- 'list': []
- }
- logger.info(f'{self.getName()}获取详情页耗时:{get_interval(t1)}毫秒,发生错误:{e}')
- # print(result)
- return result
- def searchContent(self, key, fypage=1,show_name=False):
- if self.search_encoding:
- if str(self.search_encoding).lower() != 'utf-8':
- key = encodeStr(key,self.search_encoding)
- elif self.encoding and str(self.encoding).startswith('gb'):
- # key = quote(key.encode('utf-8').decode('utf-8').encode(self.encoding,'ignore'))
- key = encodeStr(key,self.encoding)
- # print(key)
- pg = str(fypage)
- if not self.searchUrl:
- return self.blank()
- url = self.searchUrl.replace('**', key).replace('fypage',pg)
- logger.info(f'{self.getName()}搜索链接:{url}')
- if not self.搜索:
- return self.blank()
- # p = self.一级.split(';') if self.搜索 == '*' and self.一级 else self.搜索.split(';') # 解析
- p = self.一级 if self.搜索 == '*' and self.一级 else self.搜索
- pp = self.一级.split(';')
- jsp = jsoup(url) if url.startswith('http') else jsoup(self.url)
- videos = []
- is_js = isinstance(p, str) and str(p).startswith('js:') # 是js
- def getPP(p, pn, pp, ppn):
- try:
- ps = pp[ppn] if p[pn] == '*' and len(pp) > ppn else p[pn]
- return ps
- except:
- return ''
- if is_js:
- headers['Referer'] = getHome(url)
- py_ctx.update({
- 'input': url,
- 'oheaders': self.d.oheaders,
- 'fetch_params': {'headers': self.headers, 'timeout': self.d.timeout, 'encoding': self.d.encoding},
- 'd': self.d,
- 'MY_PAGE': fypage,
- 'KEY': key, # 搜索关键字
- 'TYPE': 'search', # 海阔js环境标志
- 'detailUrl': self.detailUrl or '',
- # 详情页链接
- 'getParse': self.d.getParse,
- 'saveParse': self.d.saveParse,
- 'jsp': jsp,'jq':jsp, 'setDetail': setDetail,
- })
- ctx = py_ctx
- # print(ctx)
- jscode = getPreJs() + p.replace('js:', '', 1)
- # print(jscode)
- loader, _ = runJScode(jscode, ctx=ctx)
- # print(loader.toString())
- vods = loader.eval('VODS')
- # print(len(vods),type(vods))
- if isinstance(vods, JsObjectWrapper):
- videos = vods.to_list()
- # print(videos)
- else:
- p = p.split(';')
- if len(p) < 5:
- return self.blank()
- is_json = str(p[0]).startswith('json:')
- pdfh = jsp.pjfh if is_json else jsp.pdfh
- pdfa = jsp.pjfa if is_json else jsp.pdfa
- pd = jsp.pj if is_json else jsp.pd
- pq = jsp.pq
- try:
- req_method = url.split(';')[1].lower() if len(url.split(';'))>1 else 'get'
- if req_method == 'post':
- rurls = url.split(';')[0].split('#')
- rurl = rurls[0]
- params = rurls[1] if len(rurls)>1 else ''
- # params = quote(params)
- print(f'rurl:{rurl},params:{params}')
- new_dict = {}
- new_tmp = params.split('&')
- # print(new_tmp)
- for i in new_tmp:
- new_dict[i.split('=')[0]] = i.split('=')[1]
- # data = ujson.dumps(new_dict)
- data = new_dict
- # print(data)
- logger.info(self.headers)
- r = requests.post(rurl, headers=self.headers,data=data, timeout=self.timeout, verify=False)
- elif req_method == 'postjson':
- rurls = url.split(';')[0].split('#')
- rurl = rurls[0]
- params = rurls[1] if len(rurls) > 1 else '{}'
- headers_cp = self.headers.copy()
- headers_cp.update({'Content-Type':'application/json'})
- try:
- params = ujson.dumps(ujson.loads(params))
- except:
- params = '{}'
- # params = params.encode()
- logger.info(headers_cp)
- logger.info(params)
- r = requests.post(rurl, headers=headers_cp, data=params, timeout=self.timeout, verify=False)
- else:
- r = requests.get(url, headers=self.headers,timeout=self.timeout,verify=False)
- html = self.checkHtml(r)
- if is_json:
- html = self.dealJson(html)
- html = json.loads(html)
- # if not is_json and html.find('输入验证码') > -1:
- if not is_json and re.search('系统安全验证|输入验证码',html,re.M|re.S):
- cookie = verifyCode(url,self.headers,self.timeout,self.retry_count,self.ocr_api)
- # cookie = ''
- if not cookie:
- return {
- 'list': videos
- }
- self.saveCookie(cookie)
- self.headers['cookie'] = cookie
- r = requests.get(url, headers=self.headers, timeout=self.timeout,verify=False)
- r.encoding = self.encoding
- html = r.text
- if not show_name and not str(html).find(key) > -1:
- logger.info('搜索结果源码未包含关键字,疑似搜索失败,正为您打印结果源码')
- print(html)
- p0 = getPP(p,0,pp,0)
- items = pdfa(html,p0.replace('json:','',1))
- # print(len(items),items)
- videos = []
- p1 = getPP(p, 1, pp, 1)
- p2 = getPP(p, 2, pp, 2)
- p3 = getPP(p, 3, pp, 3)
- p4 = getPP(p, 4, pp, 4)
- p5 = getPP(p, 5, pp, 5)
- for item in items:
- # print(item)
- try:
- # title = pdfh(item, p[1])
- title = ''.join([pdfh(item, i) for i in p1.split('||')])
- try:
- img = pd(item, p2)
- except:
- img = ''
- try:
- desc = pdfh(item, p3)
- except:
- desc = ''
- if len(p) > 5 and p[5]:
- content = pdfh(item, p5)
- else:
- content = ''
- # link = '$'.join([pd(item, p4) for p4 in p[4].split('+')])
- links = [pd(item, _p4) if not self.detailUrl else pdfh(item, _p4) for _p4 in p4.split('+')]
- link = '$'.join(links)
- # print(content)
- # sid = self.regStr(sid, "/video/(\\S+).html")
- vod_id = link
- if self.二级 == '*':
- vod_id = vod_id + '@@' + title + '@@' + img
- videos.append({
- "vod_id": vod_id,
- "vod_name": title,
- "vod_pic": img,
- "vod_remarks": desc,
- "vod_content": content, # 无用参数
- })
- except Exception as e:
- print(f'搜索列表解析发生错误:{e}')
- pass
- # print(videos)
- except Exception as e:
- logger.info(f'搜索{self.getName()}发生错误:{e}')
- if self.图片来源:
- for video in videos:
- if video.get('vod_pic','') and str(video['vod_pic']).startswith('http'):
- video['vod_pic'] = f"{video['vod_pic']}{self.图片来源}"
- if show_name and len(videos) > 0:
- for video in videos:
- video['vod_name'] = self.id + ' '+video['vod_name']
- video['vod_rule'] = self.id
- video['vod_id'] = video['vod_id'] +'#' + self.id
- result = {
- 'list': videos
- }
- return result
- def playContent(self, play_url,jxs=None,flag=None):
- # flag参数只有类型为4的时候才有,可以忽略
- # logger.info('播放免嗅地址: ' + self.play_url)
- # 注意:全局flags里的视频没法执行免嗅代码,因为会自动拦截去调用解析: url=yoursite:5705/vod?play_url=xxxx
- if not jxs:
- jxs = []
- # print(play_url)
- if play_url.find('http') == -1: # 字符串看起来被编码的
- try:
- play_url = base64Decode(play_url) # 自动base64解码
- except:
- pass
- # print(unquote(play_url))
- play_url = unquote(play_url)
- origin_play_url = play_url
- print(origin_play_url)
- if self.lazy:
- print(f'{play_url}->开始执行免嗅代码{type(self.lazy)}->{self.lazy}')
- t1 = time()
- try:
- if type(self.lazy) == JsObjectWrapper:
- logger.info(f'lazy非纯文本免嗅失败耗时:{get_interval(t1)}毫秒,播放地址:{play_url}')
- elif str(self.lazy).startswith('py:'):
- pycode = runPy(self.lazy)
- if pycode:
- # print(pycode)
- pos = pycode.find('def lazyParse')
- if pos < 0:
- return play_url
- pyenv = safePython(self.lazy,pycode[pos:])
- lazy_url = pyenv.action_task_exec('lazyParse',[play_url,self.d])
- logger.info(f'py免嗅耗时:{get_interval(t1)}毫秒,播放地址:{lazy_url}')
- if isinstance(lazy_url,str) and lazy_url.startswith('http'):
- play_url = lazy_url
- else:
- jscode = str(self.lazy).strip().replace('js:', '', 1) if str(self.lazy).startswith('js:') else js_code
- jsp = jsoup(self.url)
- # jscode = f'var input={play_url};{jscode}'
- # print(jscode)
- headers['Referer'] = getHome(play_url)
- py_ctx.update({
- 'input': play_url,
- 'oheaders': self.d.oheaders,
- 'fetch_params':{'headers':self.headers,'timeout':self.d.timeout,'encoding':self.d.encoding},
- 'd': self.d,
- 'jxs':jxs,
- 'getParse':self.d.getParse,
- 'saveParse':self.d.saveParse,
- 'jsp': jsp,
- 'jq': jsp,
- 'pdfh': self.d.jsp.pdfh,
- 'pdfa': self.d.jsp.pdfa, 'pd': self.d.jsp.pd,'play_url':self.play_url
- })
- ctx = py_ctx
- # print(ctx)
- jscode = getPreJs() + jscode
- # print(jscode)
- loader,_ = runJScode(jscode,ctx=ctx)
- # print(loader.toString())
- play_url = loader.eval('input')
- if isinstance(play_url,JsObjectWrapper):
- play_url = play_url.to_dict()
- # print(type(play_url))
- # print(play_url)
- logger.info(f'js免嗅耗时:{get_interval(t1)}毫秒,播放地址:{play_url}')
- if not play_url and play_url!='' and play_url!={}:
- play_url = origin_play_url
- # if play_url == {}:
- # play_url = None
- except Exception as e:
- logger.info(f'免嗅耗时:{get_interval(t1)}毫秒,并发生错误:{e}')
- # return play_url
- else:
- logger.info(f'播放重定向到:{play_url}')
- # return play_url
- if self.play_json:
- # 如果传了 play_json 参数并且是个大于0的列表的话
- if isinstance(self.play_json,list) and len(self.play_json) > 0:
- # 获取播放链接
- web_url = play_url if isinstance(play_url,str) else play_url.get('url')
- for pjson in self.play_json:
- if pjson.get('re') and (pjson['re']=='*' or re.search(pjson['re'],web_url,re.S|re.M)):
- if pjson.get('json') and isinstance(pjson['json'], dict):
- if isinstance(play_url, str):
- base_json = pjson['json']
- base_json['url'] = web_url
- play_url = base_json
- elif isinstance(play_url, dict):
- base_json = pjson['json']
- play_url.update(base_json)
- # 不管有没有效,匹配到了就跑??? (当然不行了,要不然写来干嘛)
- break
- else: # 没有指定列表默认表示需要解析,解析播放 (如果不要解析,我想也是没人会去写这个参数)
- base_json = {
- 'jx':1, # 解析开
- 'parse':1, # 嗅探 关 pluto这个标识有问题 只好双1了
- }
- if isinstance(play_url,str):
- base_json['url'] = play_url
- play_url = base_json
- elif isinstance(play_url,dict):
- play_url.update(base_json)
- logger.info(f'最终返回play_url:{play_url}')
- return play_url
- if __name__ == '__main__':
- print(urljoin('https://api.web.360kan.com/v1/f',
- '//0img.hitv.com/preview/sp_images/2022/01/28/202201281528074643023.jpg'))
- # exit()
- from utils import parser
- # js_path = f'js/玩偶姐姐.js'
- # js_path = f'js/555影视.js'
- with open('../js/模板.js', encoding='utf-8') as f:
- before = f.read().split('export')[0]
- js_path = f'js/360影视.js'
- ctx, js_code = parser.runJs(js_path,before=before)
- ruleDict = ctx.rule.to_dict()
- # lazy = ctx.eval('lazy')
- # print(lazy)
- # ruleDict['id'] = rule # 把路由请求的id装到字典里,后面播放嗅探才能用
- cms = CMS(ruleDict)
- print(cms.title)
- print(cms.homeContent())
- # print(cms.categoryContent('5',1))
- # print(cms.categoryContent('latest',1))
- # print(cms.detailContent(['https://www.2345ka.com/v/45499.html']))
- # print(cms.detailContent(1,['https://cokemv.me/voddetail/40573.html']))
- # cms.categoryContent('dianying',1)
- # print(cms.detailContent(['67391']))
- # print(cms.searchContent('斗罗大陆'))
- print(cms.searchContent('独行月球'))
|