1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-22 06:49:31 +08:00

Merge remote-tracking branch 'pulpe/teachertube'

Conflicts:
	youtube_dl/extractor/__init__.py
This commit is contained in:
Philipp Hagemeister 2014-06-07 15:20:12 +02:00
commit 24f5251cce
3 changed files with 98 additions and 0 deletions

View File

@ -28,6 +28,7 @@ from youtube_dl.extractor import (
SoundcloudSetIE,
SoundcloudUserIE,
SoundcloudPlaylistIE,
TeacherTubeClassroomIE,
LivestreamIE,
NHLVideocenterIE,
BambuserChannelIE,
@ -360,5 +361,13 @@ class TestPlaylists(unittest.TestCase):
result['title'], 'Brace Yourself - Today\'s Weirdest News')
self.assertTrue(len(result['entries']) >= 10)
def test_TeacherTubeClassroom(self):
dl = FakeYDL()
ie = TeacherTubeClassroomIE(dl)
result = ie.extract('http://www.teachertube.com/view_classroom.php?user=rbhagwati2')
self.assertIsPlaylist(result)
self.assertEqual(result['id'], 'rbhagwati2')
self.assertTrue(len(result['entries']) >= 20)
if __name__ == '__main__':
unittest.main()

View File

@ -267,6 +267,10 @@ from .swrmediathek import SWRMediathekIE
from .syfy import SyfyIE
from .sztvhu import SztvHuIE
from .tagesschau import TagesschauIE
from .teachertube import (
TeacherTubeIE,
TeacherTubeClassroomIE,
)
from .teachingchannel import TeachingChannelIE
from .teamcoco import TeamcocoIE
from .techtalks import TechTalksIE

View File

@ -0,0 +1,85 @@
# -*- coding: utf-8 -*-
from __future__ import unicode_literals
import re
from .common import InfoExtractor
class TeacherTubeIE(InfoExtractor):
IE_NAME = 'teachertube'
IE_DESC = 'teachertube.com videos'
_VALID_URL = r'https?://(?:www\.)?teachertube\.com/viewVideo\.php\?video_id=(?P<id>\d+)'
_TESTS = [{
'url': 'http://www.teachertube.com/viewVideo.php?video_id=339997',
'md5': 'f9434ef992fd65936d72999951ee254c',
'info_dict': {
'id': '339997',
'ext': 'mp4',
'title': 'Measures of dispersion from a frequency table_x264',
'description': 'md5:a3e9853487185e9fcd7181a07164650b',
'thumbnail': 're:http://.*\.jpg',
},
}, {
'url': 'http://www.teachertube.com/viewVideo.php?video_id=340064',
'md5': '0d625ec6bc9bf50f70170942ad580676',
'info_dict': {
'id': '340064',
'ext': 'mp4',
'title': 'How to Make Paper Dolls _ Paper Art Projects',
'description': 'md5:2ca52b20cd727773d1dc418b3d6bd07b',
'thumbnail': 're:http://.*\.jpg',
},
}]
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
webpage = self._download_webpage(url, video_id)
url = self._html_search_meta('twitter:player:stream', webpage, 'twitter player')
formats = [{
'format_id': 'flv',
'url': url.replace('mp4v', 'flv').replace('.mp4', '.flv'),
'quality': 0,
'ext': 'flv',
}, {
'format_id': 'mp4',
'url': url,
'quality': 1,
'ext': 'mp4',
}]
self._sort_formats(formats)
return {
'id': video_id,
'title': self._og_search_title(webpage),
'thumbnail': self._og_search_thumbnail(webpage),
'formats': formats,
'description': self._og_search_description(webpage),
}
class TeacherTubeClassroomIE(InfoExtractor):
IE_NAME = 'teachertube:classroom'
IE_DESC = 'teachertube.com online classrooms'
_VALID_URL = r'https?://(?:www\.)?teachertube\.com/view_classroom\.php\?user=(?P<user>[0-9a-zA-Z]+)'
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
user_id = mobj.group('user')
rss = self._download_xml('http://www.teachertube.com/rssclassroom.php?mode=user&username=%s' % user_id,
user_id, 'Downloading classroom RSS')
entries = []
for url in rss.findall('.//{http://search.yahoo.com/mrss/}player'):
entries.append(self.url_result(url.attrib['url'], 'TeacherTube'))
return self.playlist_result(entries, user_id)