aboutsummaryrefslogtreecommitdiffstats
path: root/youtube_dl/extractor/iqiyi.py
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor/iqiyi.py')
-rw-r--r--youtube_dl/extractor/iqiyi.py219
1 files changed, 0 insertions, 219 deletions
diff --git a/youtube_dl/extractor/iqiyi.py b/youtube_dl/extractor/iqiyi.py
deleted file mode 100644
index 6df521426..000000000
--- a/youtube_dl/extractor/iqiyi.py
+++ /dev/null
@@ -1,219 +0,0 @@
-# coding: utf-8
-from __future__ import unicode_literals
-
-import hashlib
-import itertools
-import re
-import time
-
-from .common import InfoExtractor
-from ..compat import (
- compat_str,
-)
-from ..utils import (
- clean_html,
- get_element_by_id,
- get_element_by_attribute,
- ExtractorError,
- ohdave_rsa_encrypt,
- remove_start,
-)
-
-
-def md5_text(text):
- return hashlib.md5(text.encode('utf-8')).hexdigest()
-
-
-class IqiyiIE(InfoExtractor):
- IE_NAME = 'iqiyi'
- IE_DESC = '爱奇艺'
-
- _VALID_URL = r'https?://(?:(?:[^.]+\.)?iqiyi\.com|www\.pps\.tv)/.+\.html'
-
- _NETRC_MACHINE = 'iqiyi'
-
- _TESTS = [{
- 'url': 'http://www.iqiyi.com/v_19rrojlavg.html',
- # MD5 checksum differs on my machine and Travis CI
- 'info_dict': {
- 'id': '9c1fb1b99d192b21c559e5a1a2cb3c73',
- 'ext': 'mp4',
- 'title': '美国德州空中惊现奇异云团 酷似UFO',
- }
- }, {
- 'url': 'http://www.iqiyi.com/v_19rrhnnclk.html',
- 'md5': 'b7dc800a4004b1b57749d9abae0472da',
- 'info_dict': {
- 'id': 'e3f585b550a280af23c98b6cb2be19fb',
- 'ext': 'mp4',
- # This can be either Simplified Chinese or Traditional Chinese
- 'title': r're:^(?:名侦探柯南 国语版:第752集 迫近灰原秘密的黑影 下篇|名偵探柯南 國語版:第752集 迫近灰原秘密的黑影 下篇)$',
- },
- 'skip': 'Geo-restricted to China',
- }, {
- 'url': 'http://www.iqiyi.com/w_19rt6o8t9p.html',
- 'only_matching': True,
- }, {
- 'url': 'http://www.iqiyi.com/a_19rrhbc6kt.html',
- 'only_matching': True,
- }, {
- 'url': 'http://yule.iqiyi.com/pcb.html',
- 'info_dict': {
- 'id': '4a0af228fddb55ec96398a364248ed7f',
- 'ext': 'mp4',
- 'title': '第2017-04-21期 女艺人频遭极端粉丝骚扰',
- },
- }, {
- # VIP-only video. The first 2 parts (6 minutes) are available without login
- # MD5 sums omitted as values are different on Travis CI and my machine
- 'url': 'http://www.iqiyi.com/v_19rrny4w8w.html',
- 'info_dict': {
- 'id': 'f3cf468b39dddb30d676f89a91200dc1',
- 'ext': 'mp4',
- 'title': '泰坦尼克号',
- },
- 'skip': 'Geo-restricted to China',
- }, {
- 'url': 'http://www.iqiyi.com/a_19rrhb8ce1.html',
- 'info_dict': {
- 'id': '202918101',
- 'title': '灌篮高手 国语版',
- },
- 'playlist_count': 101,
- }, {
- 'url': 'http://www.pps.tv/w_19rrbav0ph.html',
- 'only_matching': True,
- }]
-
- _FORMATS_MAP = {
- '96': 1, # 216p, 240p
- '1': 2, # 336p, 360p
- '2': 3, # 480p, 504p
- '21': 4, # 504p
- '4': 5, # 720p
- '17': 5, # 720p
- '5': 6, # 1072p, 1080p
- '18': 7, # 1080p
- }
-
- def _real_initialize(self):
- self._login()
-
- @staticmethod
- def _rsa_fun(data):
- # public key extracted from http://static.iqiyi.com/js/qiyiV2/20160129180840/jobs/i18n/i18nIndex.js
- N = 0xab86b6371b5318aaa1d3c9e612a9f1264f372323c8c0f19875b5fc3b3fd3afcc1e5bec527aa94bfa85bffc157e4245aebda05389a5357b75115ac94f074aefcd
- e = 65537
-
- return ohdave_rsa_encrypt(data, e, N)
-
- def _login(self):
- raise ExtractorError("iQiyi's non-free authentication algorithm has made login impossible", expected=True)
-
- def get_raw_data(self, tvid, video_id):
- tm = int(time.time() * 1000)
-
- key = 'd5fb4bd9d50c4be6948c97edd7254b0e'
- sc = md5_text(compat_str(tm) + key + tvid)
- params = {
- 'tvid': tvid,
- 'vid': video_id,
- 'src': '76f90cbd92f94a2e925d83e8ccd22cb7',
- 'sc': sc,
- 't': tm,
- }
-
- return self._download_json(
- 'http://cache.m.iqiyi.com/jp/tmts/%s/%s/' % (tvid, video_id),
- video_id, transform_source=lambda s: remove_start(s, 'var tvInfoJs='),
- query=params, headers=self.geo_verification_headers())
-
- def _extract_playlist(self, webpage):
- PAGE_SIZE = 50
-
- links = re.findall(
- r'<a[^>]+class="site-piclist_pic_link"[^>]+href="(http://www\.iqiyi\.com/.+\.html)"',
- webpage)
- if not links:
- return
-
- album_id = self._search_regex(
- r'albumId\s*:\s*(\d+),', webpage, 'album ID')
- album_title = self._search_regex(
- r'data-share-title="([^"]+)"', webpage, 'album title', fatal=False)
-
- entries = list(map(self.url_result, links))
-
- # Start from 2 because links in the first page are already on webpage
- for page_num in itertools.count(2):
- pagelist_page = self._download_webpage(
- 'http://cache.video.qiyi.com/jp/avlist/%s/%d/%d/' % (album_id, page_num, PAGE_SIZE),
- album_id,
- note='Download playlist page %d' % page_num,
- errnote='Failed to download playlist page %d' % page_num)
- pagelist = self._parse_json(
- remove_start(pagelist_page, 'var tvInfoJs='), album_id)
- vlist = pagelist['data']['vlist']
- for item in vlist:
- entries.append(self.url_result(item['vurl']))
- if len(vlist) < PAGE_SIZE:
- break
-
- return self.playlist_result(entries, album_id, album_title)
-
- def _real_extract(self, url):
- webpage = self._download_webpage(
- url, 'temp_id', note='download video page')
-
- # There's no simple way to determine whether an URL is a playlist or not
- # Sometimes there are playlist links in individual videos, so treat it
- # as a single video first
- tvid = self._search_regex(
- r'data-(?:player|shareplattrigger)-tvid\s*=\s*[\'"](\d+)', webpage, 'tvid', default=None)
- if tvid is None:
- playlist_result = self._extract_playlist(webpage)
- if playlist_result:
- return playlist_result
- raise ExtractorError('Can\'t find any video')
-
- video_id = self._search_regex(
- r'data-(?:player|shareplattrigger)-videoid\s*=\s*[\'"]([a-f\d]+)', webpage, 'video_id')
-
- formats = []
- for _ in range(5):
- raw_data = self.get_raw_data(tvid, video_id)
-
- if raw_data['code'] != 'A00000':
- if raw_data['code'] == 'A00111':
- self.raise_geo_restricted()
- raise ExtractorError('Unable to load data. Error code: ' + raw_data['code'])
-
- data = raw_data['data']
-
- for stream in data['vidl']:
- if 'm3utx' not in stream:
- continue
- vd = compat_str(stream['vd'])
- formats.append({
- 'url': stream['m3utx'],
- 'format_id': vd,
- 'ext': 'mp4',
- 'preference': self._FORMATS_MAP.get(vd, -1),
- 'protocol': 'm3u8_native',
- })
-
- if formats:
- break
-
- self._sleep(5, video_id)
-
- self._sort_formats(formats)
- title = (get_element_by_id('widget-videotitle', webpage)
- or clean_html(get_element_by_attribute('class', 'mod-play-tit', webpage))
- or self._html_search_regex(r'<span[^>]+data-videochanged-title="word"[^>]*>([^<]+)</span>', webpage, 'title'))
-
- return {
- 'id': video_id,
- 'title': title,
- 'formats': formats,
- }