]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/ceskatelevize.py
Prepare to upload.
[youtubedl] / youtube_dl / extractor / ceskatelevize.py
index 97feb6704075831fb8b5ef95a547428ecc57ec3f..b27b4e6708675027ccc0337f5abc44b4c480816b 100644 (file)
@@ -4,72 +4,94 @@ from __future__ import unicode_literals
 import re
 
 from .common import InfoExtractor
-from ..utils import (
-    compat_urllib_request,
+from ..compat import (
     compat_urllib_parse,
+    compat_urllib_parse_unquote,
     compat_urllib_parse_urlparse,
+)
+from ..utils import (
     ExtractorError,
+    float_or_none,
+    sanitized_Request,
 )
 
 
 class CeskaTelevizeIE(InfoExtractor):
-    _VALID_URL = r'https?://www\.ceskatelevize\.cz/(porady|ivysilani)/(.+/)?(?P<id>[^?#]+)'
-
-    _TESTS = [
-        {
-            'url': 'http://www.ceskatelevize.cz/ivysilani/10532695142-prvni-republika/213512120230004-spanelska-chripka',
-            'info_dict': {
-                'id': '213512120230004',
-                'ext': 'flv',
-                'title': 'První republika: Španělská chřipka',
-                'duration': 3107.4,
-            },
-            'params': {
-                'skip_download': True,  # requires rtmpdump
-            },
-            'skip': 'Works only from Czech Republic.',
+    _VALID_URL = r'https?://www\.ceskatelevize\.cz/(porady|ivysilani)/(?:[^/]+/)*(?P<id>[^/#?]+)/*(?:[#?].*)?$'
+    _TESTS = [{
+        'url': 'http://www.ceskatelevize.cz/ivysilani/ivysilani/10441294653-hyde-park-civilizace/214411058091220',
+        'info_dict': {
+            'id': '61924494876951776',
+            'ext': 'mp4',
+            'title': 'Hyde Park Civilizace',
+            'description': 'md5:fe93f6eda372d150759d11644ebbfb4a',
+            'thumbnail': 're:^https?://.*\.jpg',
+            'duration': 3350,
         },
-        {
-            'url': 'http://www.ceskatelevize.cz/ivysilani/1030584952-tsatsiki-maminka-a-policajt',
-            'info_dict': {
-                'id': '20138143440',
-                'ext': 'flv',
-                'title': 'Tsatsiki, maminka a policajt',
-                'duration': 6754.1,
-            },
-            'params': {
-                'skip_download': True,  # requires rtmpdump
-            },
-            'skip': 'Works only from Czech Republic.',
+        'params': {
+            # m3u8 download
+            'skip_download': True,
+        },
+    }, {
+        'url': 'http://www.ceskatelevize.cz/ivysilani/10532695142-prvni-republika/bonus/14716-zpevacka-z-duparny-bobina',
+        'info_dict': {
+            'id': '61924494876844374',
+            'ext': 'mp4',
+            'title': 'První republika: Zpěvačka z Dupárny Bobina',
+            'description': 'Sága mapující atmosféru první republiky od r. 1918 do r. 1945.',
+            'thumbnail': 're:^https?://.*\.jpg',
+            'duration': 88.4,
+        },
+        'params': {
+            # m3u8 download
+            'skip_download': True,
         },
-        {
-            'url': 'http://www.ceskatelevize.cz/ivysilani/10532695142-prvni-republika/bonus/14716-zpevacka-z-duparny-bobina',
+    }, {
+        # video with 18+ caution trailer
+        'url': 'http://www.ceskatelevize.cz/porady/10520528904-queer/215562210900007-bogotart/',
+        'info_dict': {
+            'id': '215562210900007-bogotart',
+            'title': 'Queer: Bogotart',
+            'description': 'Alternativní průvodce současným queer světem',
+        },
+        'playlist': [{
             'info_dict': {
-                'id': '14716',
-                'ext': 'flv',
-                'title': 'První republika: Zpěvačka z Dupárny Bobina',
-                'duration': 90,
+                'id': '61924494876844842',
+                'ext': 'mp4',
+                'title': 'Queer: Bogotart (Varování 18+)',
+                'duration': 10.2,
             },
-            'params': {
-                'skip_download': True,  # requires rtmpdump
+        }, {
+            'info_dict': {
+                'id': '61924494877068022',
+                'ext': 'mp4',
+                'title': 'Queer: Bogotart (Queer)',
+                'thumbnail': 're:^https?://.*\.jpg',
+                'duration': 1558.3,
             },
+        }],
+        'params': {
+            # m3u8 download
+            'skip_download': True,
         },
-    ]
+    }]
 
     def _real_extract(self, url):
         url = url.replace('/porady/', '/ivysilani/').replace('/video/', '')
 
         mobj = re.match(self._VALID_URL, url)
-        video_id = mobj.group('id')
+        playlist_id = mobj.group('id')
 
-        webpage = self._download_webpage(url, video_id)
+        webpage = self._download_webpage(url, playlist_id)
 
         NOT_AVAILABLE_STRING = 'This content is not available at your territory due to limited copyright.'
         if '%s</p>' % NOT_AVAILABLE_STRING in webpage:
             raise ExtractorError(NOT_AVAILABLE_STRING, expected=True)
 
-        typ = self._html_search_regex(r'getPlaylistUrl\(\[\{"type":"(.+?)","id":".+?"\}\],', webpage, 'type')
-        episode_id = self._html_search_regex(r'getPlaylistUrl\(\[\{"type":".+?","id":"(.+?)"\}\],', webpage, 'episode_id')
+        typ = self._html_search_regex(
+            r'getPlaylistUrl\(\[\{"type":"(.+?)","id":".+?"\}\],', webpage, 'type')
+        episode_id = self._html_search_regex(
+            r'getPlaylistUrl\(\[\{"type":".+?","id":"(.+?)"\}\],', webpage, 'episode_id')
 
         data = {
             'playlist[0][type]': typ,
@@ -78,49 +100,93 @@ class CeskaTelevizeIE(InfoExtractor):
             'requestSource': 'iVysilani',
         }
 
-        req = compat_urllib_request.Request('http://www.ceskatelevize.cz/ivysilani/ajax/get-playlist-url',
-                                            data=compat_urllib_parse.urlencode(data))
+        req = sanitized_Request(
+            'http://www.ceskatelevize.cz/ivysilani/ajax/get-client-playlist',
+            data=compat_urllib_parse.urlencode(data))
 
         req.add_header('Content-type', 'application/x-www-form-urlencoded')
         req.add_header('x-addr', '127.0.0.1')
         req.add_header('X-Requested-With', 'XMLHttpRequest')
         req.add_header('Referer', url)
 
-        playlistpage = self._download_json(req, video_id)
+        playlistpage = self._download_json(req, playlist_id)
 
-        req = compat_urllib_request.Request(compat_urllib_parse.unquote(playlistpage['url']))
-        req.add_header('Referer', url)
+        playlist_url = playlistpage['url']
+        if playlist_url == 'error_region':
+            raise ExtractorError(NOT_AVAILABLE_STRING, expected=True)
 
-        playlist = self._download_xml(req, video_id)
-
-        formats = []
-        for i in playlist.find('smilRoot/body'):
-            if 'AD' not in i.attrib['id']:
-                base_url = i.attrib['base']
-                parsedurl = compat_urllib_parse_urlparse(base_url)
-                duration = i.attrib['duration']
-
-                for video in i.findall('video'):
-                    if video.attrib['label'] != 'AD':
-                        format_id = video.attrib['label']
-                        play_path = video.attrib['src']
-                        vbr = int(video.attrib['system-bitrate'])
-
-                        formats.append({
-                            'format_id': format_id,
-                            'url': base_url,
-                            'vbr': vbr,
-                            'play_path': play_path,
-                            'app': parsedurl.path[1:] + '?' + parsedurl.query,
-                            'rtmp_live': True,
-                            'ext': 'flv',
-                        })
-
-        self._sort_formats(formats)
+        req = sanitized_Request(compat_urllib_parse_unquote(playlist_url))
+        req.add_header('Referer', url)
 
+        playlist_title = self._og_search_title(webpage)
+        playlist_description = self._og_search_description(webpage)
+
+        playlist = self._download_json(req, playlist_id)['playlist']
+        playlist_len = len(playlist)
+
+        entries = []
+        for item in playlist:
+            formats = []
+            for format_id, stream_url in item['streamUrls'].items():
+                formats.extend(self._extract_m3u8_formats(
+                    stream_url, playlist_id, 'mp4', entry_protocol='m3u8_native'))
+            self._sort_formats(formats)
+
+            item_id = item.get('id') or item['assetId']
+            title = item['title']
+
+            duration = float_or_none(item.get('duration'))
+            thumbnail = item.get('previewImageUrl')
+
+            subtitles = {}
+            if item.get('type') == 'VOD':
+                subs = item.get('subtitles')
+                if subs:
+                    subtitles = self.extract_subtitles(episode_id, subs)
+
+            entries.append({
+                'id': item_id,
+                'title': playlist_title if playlist_len == 1 else '%s (%s)' % (playlist_title, title),
+                'description': playlist_description if playlist_len == 1 else None,
+                'thumbnail': thumbnail,
+                'duration': duration,
+                'formats': formats,
+                'subtitles': subtitles,
+            })
+
+        return self.playlist_result(entries, playlist_id, playlist_title, playlist_description)
+
+    def _get_subtitles(self, episode_id, subs):
+        original_subtitles = self._download_webpage(
+            subs[0]['url'], episode_id, 'Downloading subtitles')
+        srt_subs = self._fix_subtitles(original_subtitles)
         return {
-            'id': episode_id,
-            'title': self._html_search_regex(r'<title>(.+?) — iVysílání — Česká televize</title>', webpage, 'title'),
-            'duration': float(duration),
-            'formats': formats,
+            'cs': [{
+                'ext': 'srt',
+                'data': srt_subs,
+            }]
         }
+
+    @staticmethod
+    def _fix_subtitles(subtitles):
+        """ Convert millisecond-based subtitles to SRT """
+
+        def _msectotimecode(msec):
+            """ Helper utility to convert milliseconds to timecode """
+            components = []
+            for divider in [1000, 60, 60, 100]:
+                components.append(msec % divider)
+                msec //= divider
+            return '{3:02}:{2:02}:{1:02},{0:03}'.format(*components)
+
+        def _fix_subtitle(subtitle):
+            for line in subtitle.splitlines():
+                m = re.match(r'^\s*([0-9]+);\s*([0-9]+)\s+([0-9]+)\s*$', line)
+                if m:
+                    yield m.group(1)
+                    start, stop = (_msectotimecode(int(t)) for t in m.groups()[1:])
+                    yield '{0} --> {1}'.format(start, stop)
+                else:
+                    yield line
+
+        return '\r\n'.join(_fix_subtitle(subtitles))