]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - test/test_all_urls.py
Merge tag 'upstream/2015.02.28'
[youtubedl] / test / test_all_urls.py
index 6b9764c67e98ba47a63b227d824bb97b82757b53..e66264b4b16147cae6e41d329bf07dcc31ff83e4 100644 (file)
@@ -1,5 +1,7 @@
 #!/usr/bin/env python
 
+from __future__ import unicode_literals
+
 # Allow direct execution
 import os
 import sys
@@ -7,11 +9,11 @@ import unittest
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
 
-from test.helper import get_testcases
+from test.helper import gettestcases
 
 from youtube_dl.extractor import (
+    FacebookIE,
     gen_extractors,
-    JustinTVIE,
     YoutubeIE,
 )
 
@@ -28,21 +30,24 @@ class TestAllURLsMatching(unittest.TestCase):
 
     def test_youtube_playlist_matching(self):
         assertPlaylist = lambda url: self.assertMatch(url, ['youtube:playlist'])
-        assertPlaylist(u'ECUl4u3cNGP61MdtwGTqZA0MreSaDybji8')
-        assertPlaylist(u'UUBABnxM4Ar9ten8Mdjj1j0Q') #585
-        assertPlaylist(u'PL63F0C78739B09958')
-        assertPlaylist(u'https://www.youtube.com/playlist?list=UUBABnxM4Ar9ten8Mdjj1j0Q')
-        assertPlaylist(u'https://www.youtube.com/course?list=ECUl4u3cNGP61MdtwGTqZA0MreSaDybji8')
-        assertPlaylist(u'https://www.youtube.com/playlist?list=PLwP_SiAcdui0KVebT0mU9Apz359a4ubsC')
-        assertPlaylist(u'https://www.youtube.com/watch?v=AV6J6_AeFEQ&playnext=1&list=PL4023E734DA416012') #668
-        self.assertFalse('youtube:playlist' in self.matching_ies(u'PLtS2H6bU1M'))
+        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(u'PLtS2H6bU1M'))
-        self.assertFalse(YoutubeIE.suitable(u'https://www.youtube.com/watch?v=AV6J6_AeFEQ&playnext=1&list=PL4023E734DA416012')) #668
+        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'])
@@ -62,24 +67,12 @@ class TestAllURLsMatching(unittest.TestCase):
     def test_youtube_show_matching(self):
         self.assertMatch('http://www.youtube.com/show/airdisasters', ['youtube:show'])
 
-    def test_justin_tv_channelid_matching(self):
-        self.assertTrue(JustinTVIE.suitable(u"justin.tv/vanillatv"))
-        self.assertTrue(JustinTVIE.suitable(u"twitch.tv/vanillatv"))
-        self.assertTrue(JustinTVIE.suitable(u"www.justin.tv/vanillatv"))
-        self.assertTrue(JustinTVIE.suitable(u"www.twitch.tv/vanillatv"))
-        self.assertTrue(JustinTVIE.suitable(u"http://www.justin.tv/vanillatv"))
-        self.assertTrue(JustinTVIE.suitable(u"http://www.twitch.tv/vanillatv"))
-        self.assertTrue(JustinTVIE.suitable(u"http://www.justin.tv/vanillatv/"))
-        self.assertTrue(JustinTVIE.suitable(u"http://www.twitch.tv/vanillatv/"))
-
-    def test_justintv_videoid_matching(self):
-        self.assertTrue(JustinTVIE.suitable(u"http://www.twitch.tv/vanillatv/b/328087483"))
-
-    def test_justin_tv_chapterid_matching(self):
-        self.assertTrue(JustinTVIE.suitable(u"http://www.twitch.tv/tsm_theoddone/c/2349361"))
+    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'])
 
     def test_youtube_extract(self):
-        assertExtractId = lambda url, id: self.assertEqual(YoutubeIE()._extract_id(url), id)
+        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')
@@ -87,15 +80,21 @@ class TestAllURLsMatching(unittest.TestCase):
         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 get_testcases():
+        for tc in gettestcases(include_onlymatching=True):
             url = tc['url']
             for ie in ies:
-                if type(ie).__name__ in ['GenericIE', tc['name'] + 'IE']:
+                if type(ie).__name__ in ('GenericIE', tc['name'] + 'IE'):
                     self.assertTrue(ie.suitable(url), '%s should match URL %r' % (type(ie).__name__, url))
                 else:
-                    self.assertFalse(ie.suitable(url), '%s should not match URL %r' % (type(ie).__name__, url))
+                    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'])
@@ -103,12 +102,33 @@ class TestAllURLsMatching(unittest.TestCase):
         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'])
+        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'])
+
+    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__':