]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/twitch.py
debian/watch: Add back pgpsigurlmangle.
[youtubedl] / youtube_dl / extractor / twitch.py
index 4b0ce54df4d329cf24e76621c6064c67a4befaaa..20919774d0bcf65705e1f379a433667e2a06b7e9 100644 (file)
@@ -7,13 +7,21 @@ import random
 
 from .common import InfoExtractor
 from ..compat import (
+    compat_parse_qs,
     compat_str,
-    compat_urllib_parse,
-    compat_urllib_request,
+    compat_urllib_parse_urlencode,
+    compat_urllib_parse_urlparse,
+    compat_urlparse,
 )
 from ..utils import (
     ExtractorError,
+    int_or_none,
+    js_to_json,
+    orderedSet,
+    parse_duration,
     parse_iso8601,
+    sanitized_Request,
+    urlencode_postdata,
 )
 
 
@@ -22,7 +30,8 @@ class TwitchBaseIE(InfoExtractor):
 
     _API_BASE = 'https://api.twitch.tv'
     _USHER_BASE = 'http://usher.twitch.tv'
-    _LOGIN_URL = 'https://secure.twitch.tv/user/login'
+    _LOGIN_URL = 'http://www.twitch.tv/login'
+    _NETRC_MACHINE = 'twitch'
 
     def _handle_error(self, response):
         if not isinstance(response, dict):
@@ -41,7 +50,7 @@ class TwitchBaseIE(InfoExtractor):
         for cookie in self._downloader.cookiejar:
             if cookie.name == 'api_token':
                 headers['Twitch-Api-Token'] = cookie.value
-        request = compat_urllib_request.Request(url, headers=headers)
+        request = sanitized_Request(url, headers=headers)
         response = super(TwitchBaseIE, self)._download_json(request, video_id, note)
         self._handle_error(response)
         return response
@@ -54,35 +63,48 @@ class TwitchBaseIE(InfoExtractor):
         if username is None:
             return
 
-        login_page = self._download_webpage(
+        login_page, handle = self._download_webpage_handle(
             self._LOGIN_URL, None, 'Downloading login page')
 
-        authenticity_token = self._search_regex(
-            r'<input name="authenticity_token" type="hidden" value="([^"]+)"',
-            login_page, 'authenticity token')
-
-        login_form = {
-            'utf8': '✓'.encode('utf-8'),
-            'authenticity_token': authenticity_token,
-            'redirect_on_login': '',
-            'embed_form': 'false',
-            'mp_source_action': '',
-            'follow': '',
-            'user[login]': username,
-            'user[password]': password,
-        }
+        login_form = self._hidden_inputs(login_page)
+
+        login_form.update({
+            'username': username,
+            'password': password,
+        })
+
+        redirect_url = handle.geturl()
+
+        post_url = self._search_regex(
+            r'<form[^>]+action=(["\'])(?P<url>.+?)\1', login_page,
+            'post url', default=redirect_url, group='url')
+
+        if not post_url.startswith('http'):
+            post_url = compat_urlparse.urljoin(redirect_url, post_url)
 
-        request = compat_urllib_request.Request(
-            self._LOGIN_URL, compat_urllib_parse.urlencode(login_form).encode('utf-8'))
-        request.add_header('Referer', self._LOGIN_URL)
+        request = sanitized_Request(
+            post_url, urlencode_postdata(login_form))
+        request.add_header('Referer', redirect_url)
         response = self._download_webpage(
             request, None, 'Logging in as %s' % username)
 
-        m = re.search(
-            r"id=([\"'])login_error_message\1[^>]*>(?P<msg>[^<]+)", response)
-        if m:
+        error_message = self._search_regex(
+            r'<div[^>]+class="subwindow_notice"[^>]*>([^<]+)</div>',
+            response, 'error message', default=None)
+        if error_message:
             raise ExtractorError(
-                'Unable to login: %s' % m.group('msg').strip(), expected=True)
+                'Unable to login. Twitch said: %s' % error_message, expected=True)
+
+        if '>Reset your password<' in response:
+            self.report_warning('Twitch asks you to reset your password, go to https://secure.twitch.tv/reset/submit')
+
+    def _prefer_source(self, formats):
+        try:
+            source = next(f for f in formats if f['format_id'] == 'Source')
+            source['preference'] = 10
+        except StopIteration:
+            pass  # No Source stream present
+        self._sort_formats(formats)
 
 
 class TwitchItemBaseIE(TwitchBaseIE):
@@ -123,14 +145,14 @@ class TwitchItemBaseIE(TwitchBaseIE):
     def _extract_info(self, info):
         return {
             'id': info['_id'],
-            'title': info['title'],
-            'description': info['description'],
-            'duration': info['length'],
-            'thumbnail': info['preview'],
-            'uploader': info['channel']['display_name'],
-            'uploader_id': info['channel']['name'],
-            'timestamp': parse_iso8601(info['recorded_at']),
-            'view_count': info['views'],
+            'title': info.get('title') or 'Untitled Broadcast',
+            'description': info.get('description'),
+            'duration': int_or_none(info.get('length')),
+            'thumbnail': info.get('preview'),
+            'uploader': info.get('channel', {}).get('display_name'),
+            'uploader_id': info.get('channel', {}).get('name'),
+            'timestamp': parse_iso8601(info.get('recorded_at')),
+            'view_count': int_or_none(info.get('views')),
         }
 
     def _real_extract(self, url):
@@ -139,7 +161,7 @@ class TwitchItemBaseIE(TwitchBaseIE):
 
 class TwitchVideoIE(TwitchItemBaseIE):
     IE_NAME = 'twitch:video'
-    _VALID_URL = r'%s/[^/]+/b/(?P<id>[^/]+)' % TwitchBaseIE._VALID_URL_BASE
+    _VALID_URL = r'%s/[^/]+/b/(?P<id>\d+)' % TwitchBaseIE._VALID_URL_BASE
     _ITEM_TYPE = 'video'
     _ITEM_SHORTCUT = 'a'
 
@@ -150,12 +172,13 @@ class TwitchVideoIE(TwitchItemBaseIE):
             'title': 'Worlds Semifinals - Star Horn Royal Club vs. OMG',
         },
         'playlist_mincount': 12,
+        'skip': 'HTTP Error 404: Not Found',
     }
 
 
 class TwitchChapterIE(TwitchItemBaseIE):
     IE_NAME = 'twitch:chapter'
-    _VALID_URL = r'%s/[^/]+/c/(?P<id>[^/]+)' % TwitchBaseIE._VALID_URL_BASE
+    _VALID_URL = r'%s/[^/]+/c/(?P<id>\d+)' % TwitchBaseIE._VALID_URL_BASE
     _ITEM_TYPE = 'chapter'
     _ITEM_SHORTCUT = 'c'
 
@@ -166,6 +189,7 @@ class TwitchChapterIE(TwitchItemBaseIE):
             'title': 'ACRL Off Season - Sports Cars @ Nordschleife',
         },
         'playlist_mincount': 3,
+        'skip': 'HTTP Error 404: Not Found',
     }, {
         'url': 'http://www.twitch.tv/tsm_theoddone/c/2349361',
         'only_matching': True,
@@ -174,41 +198,79 @@ class TwitchChapterIE(TwitchItemBaseIE):
 
 class TwitchVodIE(TwitchItemBaseIE):
     IE_NAME = 'twitch:vod'
-    _VALID_URL = r'%s/[^/]+/v/(?P<id>[^/]+)' % TwitchBaseIE._VALID_URL_BASE
+    _VALID_URL = r'%s/[^/]+/v/(?P<id>\d+)' % TwitchBaseIE._VALID_URL_BASE
     _ITEM_TYPE = 'vod'
     _ITEM_SHORTCUT = 'v'
 
-    _TEST = {
-        'url': 'http://www.twitch.tv/ksptv/v/3622000',
+    _TESTS = [{
+        'url': 'http://www.twitch.tv/riotgames/v/6528877?t=5m10s',
         'info_dict': {
-            'id': 'v3622000',
+            'id': 'v6528877',
             'ext': 'mp4',
-            'title': '''KSPTV: Squadcast: "Everyone's on vacation so here's Dahud" Edition!''',
+            'title': 'LCK Summer Split - Week 6 Day 1',
             'thumbnail': 're:^https?://.*\.jpg$',
-            'duration': 6951,
-            'timestamp': 1419028564,
-            'upload_date': '20141219',
-            'uploader': 'KSPTV',
-            'uploader_id': 'ksptv',
+            'duration': 17208,
+            'timestamp': 1435131709,
+            'upload_date': '20150624',
+            'uploader': 'Riot Games',
+            'uploader_id': 'riotgames',
             'view_count': int,
+            'start_time': 310,
         },
         'params': {
             # m3u8 download
             'skip_download': True,
         },
-    }
+    }, {
+        # Untitled broadcast (title is None)
+        'url': 'http://www.twitch.tv/belkao_o/v/11230755',
+        'info_dict': {
+            'id': 'v11230755',
+            'ext': 'mp4',
+            'title': 'Untitled Broadcast',
+            'thumbnail': 're:^https?://.*\.jpg$',
+            'duration': 1638,
+            'timestamp': 1439746708,
+            'upload_date': '20150816',
+            'uploader': 'BelkAO_o',
+            'uploader_id': 'belkao_o',
+            'view_count': int,
+        },
+        'params': {
+            # m3u8 download
+            'skip_download': True,
+        },
+    }]
 
     def _real_extract(self, url):
         item_id = self._match_id(url)
+
         info = self._download_info(self._ITEM_SHORTCUT, item_id)
         access_token = self._download_json(
             '%s/api/vods/%s/access_token' % (self._API_BASE, item_id), item_id,
             'Downloading %s access token' % self._ITEM_TYPE)
+
         formats = self._extract_m3u8_formats(
-            '%s/vod/%s?nauth=%s&nauthsig=%s'
-            % (self._USHER_BASE, item_id, access_token['token'], access_token['sig']),
-            item_id, 'mp4')
+            '%s/vod/%s?%s' % (
+                self._USHER_BASE, item_id,
+                compat_urllib_parse_urlencode({
+                    'allow_source': 'true',
+                    'allow_audio_only': 'true',
+                    'allow_spectre': 'true',
+                    'player': 'twitchweb',
+                    'nauth': access_token['token'],
+                    'nauthsig': access_token['sig'],
+                })),
+            item_id, 'mp4', entry_protocol='m3u8_native')
+
+        self._prefer_source(formats)
         info['formats'] = formats
+
+        parsed_url = compat_urllib_parse_urlparse(url)
+        query = compat_parse_qs(parsed_url.query)
+        if 't' in query:
+            info['start_time'] = parse_duration(query['t'][0])
+
         return info
 
 
@@ -224,17 +286,37 @@ class TwitchPlaylistBaseIE(TwitchBaseIE):
         entries = []
         offset = 0
         limit = self._PAGE_LIMIT
+        broken_paging_detected = False
+        counter_override = None
         for counter in itertools.count(1):
             response = self._download_json(
                 self._PLAYLIST_URL % (channel_id, offset, limit),
-                channel_id, 'Downloading %s videos JSON page %d' % (self._PLAYLIST_TYPE, counter))
+                channel_id,
+                'Downloading %s videos JSON page %s'
+                % (self._PLAYLIST_TYPE, counter_override or counter))
             page_entries = self._extract_playlist_page(response)
             if not page_entries:
                 break
+            total = int_or_none(response.get('_total'))
+            # Since the beginning of March 2016 twitch's paging mechanism
+            # is completely broken on the twitch side. It simply ignores
+            # a limit and returns the whole offset number of videos.
+            # Working around by just requesting all videos at once.
+            # Upd: pagination bug was fixed by twitch on 15.03.2016.
+            if not broken_paging_detected and total and len(page_entries) > limit:
+                self.report_warning(
+                    'Twitch pagination is broken on twitch side, requesting all videos at once',
+                    channel_id)
+                broken_paging_detected = True
+                offset = total
+                counter_override = '(all at once)'
+                continue
             entries.extend(page_entries)
+            if broken_paging_detected or total and len(page_entries) >= total:
+                break
             offset += limit
         return self.playlist_result(
-            [self.url_result(entry) for entry in set(entries)],
+            [self.url_result(entry) for entry in orderedSet(entries)],
             channel_id, channel_name)
 
     def _extract_playlist_page(self, response):
@@ -276,36 +358,11 @@ class TwitchPastBroadcastsIE(TwitchPlaylistBaseIE):
     }
 
 
-class TwitchBookmarksIE(TwitchPlaylistBaseIE):
-    IE_NAME = 'twitch:bookmarks'
-    _VALID_URL = r'%s/(?P<id>[^/]+)/profile/bookmarks/?(?:\#.*)?$' % TwitchBaseIE._VALID_URL_BASE
-    _PLAYLIST_URL = '%s/api/bookmark/?user=%%s&offset=%%d&limit=%%d' % TwitchBaseIE._API_BASE
-    _PLAYLIST_TYPE = 'bookmarks'
-
-    _TEST = {
-        'url': 'http://www.twitch.tv/ognos/profile/bookmarks',
-        'info_dict': {
-            'id': 'ognos',
-            'title': 'Ognos',
-        },
-        'playlist_mincount': 3,
-    }
-
-    def _extract_playlist_page(self, response):
-        entries = []
-        for bookmark in response.get('bookmarks', []):
-            video = bookmark.get('video')
-            if not video:
-                continue
-            entries.append(video['url'])
-        return entries
-
-
 class TwitchStreamIE(TwitchBaseIE):
     IE_NAME = 'twitch:stream'
-    _VALID_URL = r'%s/(?P<id>[^/]+)/?(?:\#.*)?$' % TwitchBaseIE._VALID_URL_BASE
+    _VALID_URL = r'%s/(?P<id>[^/#?]+)/?(?:\#.*)?$' % TwitchBaseIE._VALID_URL_BASE
 
-    _TEST = {
+    _TESTS = [{
         'url': 'http://www.twitch.tv/shroomztv',
         'info_dict': {
             'id': '12772022048',
@@ -324,7 +381,10 @@ class TwitchStreamIE(TwitchBaseIE):
             # m3u8 download
             'skip_download': True,
         },
-    }
+    }, {
+        'url': 'http://www.twitch.tv/miracle_doto#profile-0',
+        'only_matching': True,
+    }]
 
     def _real_extract(self, url):
         channel_id = self._match_id(url)
@@ -339,30 +399,30 @@ class TwitchStreamIE(TwitchBaseIE):
                 'http://www.twitch.tv/%s/profile' % channel_id,
                 'TwitchProfile', channel_id)
 
+        # Channel name may be typed if different case than the original channel name
+        # (e.g. http://www.twitch.tv/TWITCHPLAYSPOKEMON) that will lead to constructing
+        # an invalid m3u8 URL. Working around by use of original channel name from stream
+        # JSON and fallback to lowercase if it's not available.
+        channel_id = stream.get('channel', {}).get('name') or channel_id.lower()
+
         access_token = self._download_json(
             '%s/api/channels/%s/access_token' % (self._API_BASE, channel_id), channel_id,
             'Downloading channel access token')
 
         query = {
             'allow_source': 'true',
+            'allow_audio_only': 'true',
             'p': random.randint(1000000, 10000000),
             'player': 'twitchweb',
             'segment_preference': '4',
-            'sig': access_token['sig'],
-            'token': access_token['token'],
+            'sig': access_token['sig'].encode('utf-8'),
+            'token': access_token['token'].encode('utf-8'),
         }
-
         formats = self._extract_m3u8_formats(
             '%s/api/channel/hls/%s.m3u8?%s'
-            % (self._USHER_BASE, channel_id, compat_urllib_parse.urlencode(query).encode('utf-8')),
+            % (self._USHER_BASE, channel_id, compat_urllib_parse_urlencode(query)),
             channel_id, 'mp4')
-
-        # prefer the 'source' stream, the others are limited to 30 fps
-        def _sort_source(f):
-            if f.get('m3u8_media') is not None and f['m3u8_media'].get('NAME') == 'Source':
-                return 1
-            return 0
-        formats = sorted(formats, key=_sort_source)
+        self._prefer_source(formats)
 
         view_count = stream.get('viewers')
         timestamp = parse_iso8601(stream.get('created_at'))
@@ -395,3 +455,45 @@ class TwitchStreamIE(TwitchBaseIE):
             'formats': formats,
             'is_live': True,
         }
+
+
+class TwitchClipsIE(InfoExtractor):
+    IE_NAME = 'twitch:clips'
+    _VALID_URL = r'https?://clips\.twitch\.tv/(?:[^/]+/)*(?P<id>[^/?#&]+)'
+
+    _TEST = {
+        'url': 'https://clips.twitch.tv/ea/AggressiveCobraPoooound',
+        'md5': '761769e1eafce0ffebfb4089cb3847cd',
+        'info_dict': {
+            'id': 'AggressiveCobraPoooound',
+            'ext': 'mp4',
+            'title': 'EA Play 2016 Live from the Novo Theatre',
+            'thumbnail': 're:^https?://.*\.jpg',
+            'creator': 'EA',
+            'uploader': 'stereotype_',
+            'uploader_id': 'stereotype_',
+        },
+    }
+
+    def _real_extract(self, url):
+        video_id = self._match_id(url)
+
+        webpage = self._download_webpage(url, video_id)
+
+        clip = self._parse_json(
+            self._search_regex(
+                r'(?s)clipInfo\s*=\s*({.+?});', webpage, 'clip info'),
+            video_id, transform_source=js_to_json)
+
+        video_url = clip['clip_video_url']
+        title = clip['channel_title']
+
+        return {
+            'id': video_id,
+            'url': video_url,
+            'title': title,
+            'thumbnail': self._og_search_thumbnail(webpage),
+            'creator': clip.get('broadcaster_display_name') or clip.get('broadcaster_login'),
+            'uploader': clip.get('curator_login'),
+            'uploader_id': clip.get('curator_display_name'),
+        }