1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-14 13:25:38 +00:00
youtube-dl/test/test_all_urls.py

155 lines
7.7 KiB
Python
Raw Normal View History

#!/usr/bin/env python
from __future__ import unicode_literals
# Allow direct execution
import os
import sys
import unittest
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from test.helper import gettestcases
from youtube_dl.extractor import (
FacebookIE,
gen_extractors,
JustinTVIE,
YoutubeIE,
)
class TestAllURLsMatching(unittest.TestCase):
def setUp(self):
self.ies = gen_extractors()
def matching_ies(self, url):
return [ie.IE_NAME for ie in self.ies if ie.suitable(url) and ie.IE_NAME != 'generic']
def assertMatch(self, url, ie_list):
self.assertEqual(self.matching_ies(url), ie_list)
def test_youtube_playlist_matching(self):
assertPlaylist = lambda url: self.assertMatch(url, ['youtube:playlist'])
assertPlaylist('ECUl4u3cNGP61MdtwGTqZA0MreSaDybji8')
assertPlaylist('UUBABnxM4Ar9ten8Mdjj1j0Q') #585
assertPlaylist('PL63F0C78739B09958')
assertPlaylist('https://www.youtube.com/playlist?list=UUBABnxM4Ar9ten8Mdjj1j0Q')
assertPlaylist('https://www.youtube.com/course?list=ECUl4u3cNGP61MdtwGTqZA0MreSaDybji8')
assertPlaylist('https://www.youtube.com/playlist?list=PLwP_SiAcdui0KVebT0mU9Apz359a4ubsC')
assertPlaylist('https://www.youtube.com/watch?v=AV6J6_AeFEQ&playnext=1&list=PL4023E734DA416012') #668
self.assertFalse('youtube:playlist' in self.matching_ies('PLtS2H6bU1M'))
# Top tracks
assertPlaylist('https://www.youtube.com/playlist?list=MCUS.20142101')
def test_youtube_matching(self):
self.assertTrue(YoutubeIE.suitable('PLtS2H6bU1M'))
self.assertFalse(YoutubeIE.suitable('https://www.youtube.com/watch?v=AV6J6_AeFEQ&playnext=1&list=PL4023E734DA416012')) #668
self.assertMatch('http://youtu.be/BaW_jenozKc', ['youtube'])
self.assertMatch('http://www.youtube.com/v/BaW_jenozKc', ['youtube'])
self.assertMatch('https://youtube.googleapis.com/v/BaW_jenozKc', ['youtube'])
self.assertMatch('http://www.cleanvideosearch.com/media/action/yt/watch?videoId=8v_4O44sfjM', ['youtube'])
def test_youtube_channel_matching(self):
assertChannel = lambda url: self.assertMatch(url, ['youtube:channel'])
assertChannel('https://www.youtube.com/channel/HCtnHdj3df7iM')
assertChannel('https://www.youtube.com/channel/HCtnHdj3df7iM?feature=gb_ch_rec')
assertChannel('https://www.youtube.com/channel/HCtnHdj3df7iM/videos')
def test_youtube_user_matching(self):
self.assertMatch('www.youtube.com/NASAgovVideo/videos', ['youtube:user'])
def test_youtube_feeds(self):
self.assertMatch('https://www.youtube.com/feed/watch_later', ['youtube:watch_later'])
self.assertMatch('https://www.youtube.com/feed/subscriptions', ['youtube:subscriptions'])
self.assertMatch('https://www.youtube.com/feed/recommended', ['youtube:recommended'])
self.assertMatch('https://www.youtube.com/my_favorites', ['youtube:favorites'])
def test_youtube_show_matching(self):
self.assertMatch('http://www.youtube.com/show/airdisasters', ['youtube:show'])
def test_youtube_search_matching(self):
self.assertMatch('http://www.youtube.com/results?search_query=making+mustard', ['youtube:search_url'])
self.assertMatch('https://www.youtube.com/results?baz=bar&search_query=youtube-dl+test+video&filters=video&lclk=video', ['youtube:search_url'])
2013-06-04 11:06:49 +00:00
def test_justin_tv_channelid_matching(self):
2014-04-19 11:01:15 +00:00
self.assertTrue(JustinTVIE.suitable('justin.tv/vanillatv'))
self.assertTrue(JustinTVIE.suitable('twitch.tv/vanillatv'))
self.assertTrue(JustinTVIE.suitable('www.justin.tv/vanillatv'))
self.assertTrue(JustinTVIE.suitable('www.twitch.tv/vanillatv'))
self.assertTrue(JustinTVIE.suitable('http://www.justin.tv/vanillatv'))
self.assertTrue(JustinTVIE.suitable('http://www.twitch.tv/vanillatv'))
self.assertTrue(JustinTVIE.suitable('http://www.justin.tv/vanillatv/'))
self.assertTrue(JustinTVIE.suitable('http://www.twitch.tv/vanillatv/'))
2013-06-04 11:06:49 +00:00
def test_justintv_videoid_matching(self):
2014-04-19 11:01:15 +00:00
self.assertTrue(JustinTVIE.suitable('http://www.twitch.tv/vanillatv/b/328087483'))
2013-06-04 11:06:49 +00:00
def test_justin_tv_chapterid_matching(self):
2014-04-19 11:01:15 +00:00
self.assertTrue(JustinTVIE.suitable('http://www.twitch.tv/tsm_theoddone/c/2349361'))
2013-06-04 11:06:49 +00:00
def test_youtube_extract(self):
assertExtractId = lambda url, id: self.assertEqual(YoutubeIE.extract_id(url), id)
assertExtractId('http://www.youtube.com/watch?&v=BaW_jenozKc', 'BaW_jenozKc')
assertExtractId('https://www.youtube.com/watch?&v=BaW_jenozKc', 'BaW_jenozKc')
assertExtractId('https://www.youtube.com/watch?feature=player_embedded&v=BaW_jenozKc', 'BaW_jenozKc')
assertExtractId('https://www.youtube.com/watch_popup?v=BaW_jenozKc', 'BaW_jenozKc')
assertExtractId('http://www.youtube.com/watch?v=BaW_jenozKcsharePLED17F32AD9753930', 'BaW_jenozKc')
assertExtractId('BaW_jenozKc', 'BaW_jenozKc')
def test_facebook_matching(self):
self.assertTrue(FacebookIE.suitable('https://www.facebook.com/Shiniknoh#!/photo.php?v=10153317450565268'))
self.assertTrue(FacebookIE.suitable('https://www.facebook.com/cindyweather?fref=ts#!/photo.php?v=10152183998945793'))
def test_no_duplicates(self):
ies = gen_extractors()
for tc in gettestcases(include_onlymatching=True):
url = tc['url']
for ie in ies:
if type(ie).__name__ in ('GenericIE', tc['name'] + 'IE'):
self.assertTrue(ie.suitable(url), '%s should match URL %r' % (type(ie).__name__, url))
else:
2014-08-27 09:35:43 +00:00
self.assertFalse(
ie.suitable(url),
'%s should not match URL %r . That URL belongs to %s.' % (type(ie).__name__, url, tc['name']))
def test_keywords(self):
self.assertMatch(':ytsubs', ['youtube:subscriptions'])
self.assertMatch(':ytsubscriptions', ['youtube:subscriptions'])
self.assertMatch(':ythistory', ['youtube:history'])
self.assertMatch(':thedailyshow', ['ComedyCentralShows'])
self.assertMatch(':tds', ['ComedyCentralShows'])
self.assertMatch(':colbertreport', ['ComedyCentralShows'])
self.assertMatch(':cr', ['ComedyCentralShows'])
def test_vimeo_matching(self):
self.assertMatch('http://vimeo.com/channels/tributes', ['vimeo:channel'])
self.assertMatch('http://vimeo.com/channels/31259', ['vimeo:channel'])
self.assertMatch('http://vimeo.com/channels/31259/53576664', ['vimeo'])
self.assertMatch('http://vimeo.com/user7108434', ['vimeo:user'])
2014-01-07 06:13:42 +00:00
self.assertMatch('http://vimeo.com/user7108434/videos', ['vimeo:user'])
self.assertMatch('https://vimeo.com/user21297594/review/75524534/3c257a1b5d', ['vimeo:review'])
# https://github.com/rg3/youtube-dl/issues/1930
def test_soundcloud_not_matching_sets(self):
self.assertMatch('http://soundcloud.com/floex/sets/gone-ep', ['soundcloud:set'])
2014-01-28 02:37:23 +00:00
def test_tumblr(self):
self.assertMatch('http://tatianamaslanydaily.tumblr.com/post/54196191430/orphan-black-dvd-extra-behind-the-scenes', ['Tumblr'])
self.assertMatch('http://tatianamaslanydaily.tumblr.com/post/54196191430', ['Tumblr'])
def test_pbs(self):
# https://github.com/rg3/youtube-dl/issues/2350
self.assertMatch('http://video.pbs.org/viralplayer/2365173446/', ['PBS'])
self.assertMatch('http://video.pbs.org/widget/partnerplayer/980042464/', ['PBS'])
def test_yahoo_https(self):
# https://github.com/rg3/youtube-dl/issues/2701
self.assertMatch(
'https://screen.yahoo.com/smartwatches-latest-wearable-gadgets-163745379-cbs.html',
['Yahoo'])
if __name__ == '__main__':
unittest.main()