mirror of
https://github.com/l1ving/youtube-dl
synced 2024-11-30 12:55:37 +08:00
193 lines
6.9 KiB
Python
193 lines
6.9 KiB
Python
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import (
|
|
compat_urlparse,
|
|
compat_str,
|
|
)
|
|
from ..utils import (
|
|
parse_duration,
|
|
js_to_json,
|
|
parse_iso8601,
|
|
)
|
|
|
|
|
|
class ViideaIE(InfoExtractor):
|
|
_VALID_URL = r'''(?x)http://(?:www\.)?(?:
|
|
videolectures\.net|
|
|
flexilearn\.viidea\.net|
|
|
presentations\.ocwconsortium\.org|
|
|
video\.travel-zoom\.si|
|
|
video\.pomp-forum\.si|
|
|
tv\.nil\.si|
|
|
video\.hekovnik.com|
|
|
video\.szko\.si|
|
|
kpk\.viidea\.com|
|
|
inside\.viidea\.net|
|
|
video\.kiberpipa\.org|
|
|
bvvideo\.si|
|
|
kongres\.viidea\.net|
|
|
edemokracija\.viidea\.com
|
|
)(?:/lecture)?/(?P<id>[^/]+)(?:/video/(?P<part>\d+))?/*(?:[#?].*)?$'''
|
|
|
|
_TESTS = [{
|
|
'url': 'http://videolectures.net/promogram_igor_mekjavic_eng/',
|
|
'info_dict': {
|
|
'id': '20171',
|
|
'display_id': 'promogram_igor_mekjavic_eng',
|
|
'ext': 'mp4',
|
|
'title': 'Automatics, robotics and biocybernetics',
|
|
'description': 'md5:815fc1deb6b3a2bff99de2d5325be482',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
'timestamp': 1372349289,
|
|
'upload_date': '20130627',
|
|
'duration': 565,
|
|
},
|
|
'params': {
|
|
# m3u8 download
|
|
'skip_download': True,
|
|
},
|
|
}, {
|
|
# video with invalid direct format links (HTTP 403)
|
|
'url': 'http://videolectures.net/russir2010_filippova_nlp/',
|
|
'info_dict': {
|
|
'id': '14891',
|
|
'display_id': 'russir2010_filippova_nlp',
|
|
'ext': 'flv',
|
|
'title': 'NLP at Google',
|
|
'description': 'md5:fc7a6d9bf0302d7cc0e53f7ca23747b3',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
'timestamp': 1284375600,
|
|
'upload_date': '20100913',
|
|
'duration': 5352,
|
|
},
|
|
'params': {
|
|
# rtmp download
|
|
'skip_download': True,
|
|
},
|
|
}, {
|
|
# event playlist
|
|
'url': 'http://videolectures.net/deeplearning2015_montreal/',
|
|
'info_dict': {
|
|
'id': '23181',
|
|
'title': 'Deep Learning Summer School, Montreal 2015',
|
|
'description': 'md5:0533a85e4bd918df52a01f0e1ebe87b7',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
'timestamp': 1438560000,
|
|
},
|
|
'playlist_count': 30,
|
|
}, {
|
|
# multi part lecture
|
|
'url': 'http://videolectures.net/mlss09uk_bishop_ibi/',
|
|
'info_dict': {
|
|
'id': '9737',
|
|
'display_id': 'mlss09uk_bishop_ibi',
|
|
'title': 'Introduction To Bayesian Inference',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
'timestamp': 1251622800,
|
|
},
|
|
'playlist': [{
|
|
'info_dict': {
|
|
'id': '9737_part1',
|
|
'display_id': 'mlss09uk_bishop_ibi_part1',
|
|
'ext': 'wmv',
|
|
'title': 'Introduction To Bayesian Inference (Part 1)',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
'duration': 4622,
|
|
'timestamp': 1251622800,
|
|
'upload_date': '20090830',
|
|
},
|
|
}, {
|
|
'info_dict': {
|
|
'id': '9737_part2',
|
|
'display_id': 'mlss09uk_bishop_ibi_part2',
|
|
'ext': 'wmv',
|
|
'title': 'Introduction To Bayesian Inference (Part 2)',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
'duration': 5641,
|
|
'timestamp': 1251622800,
|
|
'upload_date': '20090830',
|
|
},
|
|
}],
|
|
'playlist_count': 2,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
lecture_slug, explicit_part_id = re.match(self._VALID_URL, url).groups()
|
|
|
|
webpage = self._download_webpage(url, lecture_slug)
|
|
|
|
cfg = self._parse_json(self._search_regex(
|
|
[r'cfg\s*:\s*({.+?})\s*,\s*[\da-zA-Z_]+\s*:\s*\(?\s*function',
|
|
r'cfg\s*:\s*({[^}]+})'],
|
|
webpage, 'cfg'), lecture_slug, js_to_json)
|
|
|
|
lecture_id = compat_str(cfg['obj_id'])
|
|
|
|
base_url = self._proto_relative_url(cfg['livepipe'], 'http:')
|
|
|
|
lecture_data = self._download_json(
|
|
'%s/site/api/lecture/%s?format=json' % (base_url, lecture_id),
|
|
lecture_id)['lecture'][0]
|
|
|
|
lecture_info = {
|
|
'id': lecture_id,
|
|
'display_id': lecture_slug,
|
|
'title': lecture_data['title'],
|
|
'timestamp': parse_iso8601(lecture_data.get('time')),
|
|
'description': lecture_data.get('description_wiki'),
|
|
'thumbnail': lecture_data.get('thumb'),
|
|
}
|
|
|
|
playlist_entries = []
|
|
lecture_type = lecture_data.get('type')
|
|
parts = [compat_str(video) for video in cfg.get('videos', [])]
|
|
if parts:
|
|
multipart = len(parts) > 1
|
|
|
|
def extract_part(part_id):
|
|
smil_url = '%s/%s/video/%s/smil.xml' % (base_url, lecture_slug, part_id)
|
|
smil = self._download_smil(smil_url, lecture_id)
|
|
info = self._parse_smil(smil, smil_url, lecture_id)
|
|
info['id'] = lecture_id if not multipart else '%s_part%s' % (lecture_id, part_id)
|
|
info['display_id'] = lecture_slug if not multipart else '%s_part%s' % (lecture_slug, part_id)
|
|
if multipart:
|
|
info['title'] += ' (Part %s)' % part_id
|
|
switch = smil.find('.//switch')
|
|
if switch is not None:
|
|
info['duration'] = parse_duration(switch.attrib.get('dur'))
|
|
item_info = lecture_info.copy()
|
|
item_info.update(info)
|
|
return item_info
|
|
|
|
if explicit_part_id or not multipart:
|
|
result = extract_part(explicit_part_id or parts[0])
|
|
else:
|
|
result = {
|
|
'_type': 'multi_video',
|
|
'entries': [extract_part(part) for part in parts],
|
|
}
|
|
result.update(lecture_info)
|
|
|
|
# Immediately return explicitly requested part or non event item
|
|
if explicit_part_id or lecture_type != 'evt':
|
|
return result
|
|
|
|
playlist_entries.append(result)
|
|
|
|
# It's probably a playlist
|
|
if not parts or lecture_type == 'evt':
|
|
playlist_webpage = self._download_webpage(
|
|
'%s/site/ajax/drilldown/?id=%s' % (base_url, lecture_id), lecture_id)
|
|
entries = [
|
|
self.url_result(compat_urlparse.urljoin(url, video_url), 'Viidea')
|
|
for _, video_url in re.findall(
|
|
r'<a[^>]+href=(["\'])(.+?)\1[^>]+id=["\']lec=\d+', playlist_webpage)]
|
|
playlist_entries.extend(entries)
|
|
|
|
playlist = self.playlist_result(playlist_entries, lecture_id)
|
|
playlist.update(lecture_info)
|
|
return playlist
|