]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/tvp.py
d/p/remove-autoupdate-mechanism.patch: Update patch with metadata at the top.
[youtubedl] / youtube_dl / extractor / tvp.py
index a645800057fc6dc88850885cba7737243c95574e..3954f0b93812e5bd58d950e8401c954d0ea0a7c6 100644 (file)
+# coding: utf-8
 from __future__ import unicode_literals
 
+import re
+
 from .common import InfoExtractor
+from ..utils import (
+    determine_ext,
+    clean_html,
+    get_element_by_attribute,
+    ExtractorError,
+)
 
 
-class TvpIE(InfoExtractor):
-    IE_NAME = 'tvp.pl'
-    _VALID_URL = r'https?://www\.tvp\.pl/.*?wideo/(?P<date>\d+)/(?P<id>\d+)'
+class TVPIE(InfoExtractor):
+    IE_NAME = 'tvp'
+    IE_DESC = 'Telewizja Polska'
+    _VALID_URL = r'https?://[^/]+\.tvp\.(?:pl|info)/(?:video/(?:[^,\s]*,)*|(?:(?!\d+/)[^/]+/)*)(?P<id>\d+)'
 
-    _TEST = {
-        'url': 'http://www.tvp.pl/warszawa/magazyny/campusnews/wideo/31102013/12878238',
-        'md5': '148408967a6a468953c0a75cbdaf0d7a',
+    _TESTS = [{
+        'url': 'https://vod.tvp.pl/video/czas-honoru,i-seria-odc-13,194536',
+        'md5': '8aa518c15e5cc32dfe8db400dc921fbb',
+        'info_dict': {
+            'id': '194536',
+            'ext': 'mp4',
+            'title': 'Czas honoru, I seria – odc. 13',
+            'description': 'md5:381afa5bca72655fe94b05cfe82bf53d',
+        },
+    }, {
+        'url': 'http://www.tvp.pl/there-can-be-anything-so-i-shortened-it/17916176',
+        'md5': 'b0005b542e5b4de643a9690326ab1257',
+        'info_dict': {
+            'id': '17916176',
+            'ext': 'mp4',
+            'title': 'TVP Gorzów pokaże filmy studentów z podroży dookoła świata',
+            'description': 'TVP Gorzów pokaże filmy studentów z podroży dookoła świata',
+        },
+    }, {
+        # page id is not the same as video id(#7799)
+        'url': 'https://wiadomosci.tvp.pl/33908820/28092017-1930',
+        'md5': '84cd3c8aec4840046e5ab712416b73d0',
         'info_dict': {
-            'id': '12878238',
-            'ext': 'wmv',
-            'title': '31.10.2013 - Odcinek 2',
-            'description': '31.10.2013 - Odcinek 2',
+            'id': '33908820',
+            'ext': 'mp4',
+            'title': 'Wiadomości, 28.09.2017, 19:30',
+            'description': 'Wydanie główne codziennego serwisu informacyjnego.'
         },
-        'skip': 'Download has to use same server IP as extraction. Therefore, a good (load-balancing) DNS resolver will make the download fail.'
-    }
+    }, {
+        'url': 'http://vod.tvp.pl/seriale/obyczajowe/na-sygnale/sezon-2-27-/odc-39/17834272',
+        'only_matching': True,
+    }, {
+        'url': 'http://wiadomosci.tvp.pl/25169746/24052016-1200',
+        'only_matching': True,
+    }, {
+        'url': 'http://krakow.tvp.pl/25511623/25lecie-mck-wyjatkowe-miejsce-na-mapie-krakowa',
+        'only_matching': True,
+    }, {
+        'url': 'http://teleexpress.tvp.pl/25522307/wierni-wzieli-udzial-w-procesjach',
+        'only_matching': True,
+    }, {
+        'url': 'http://sport.tvp.pl/25522165/krychowiak-uspokaja-w-sprawie-kontuzji-dwa-tygodnie-to-maksimum',
+        'only_matching': True,
+    }, {
+        'url': 'http://www.tvp.info/25511919/trwa-rewolucja-wladza-zdecydowala-sie-na-pogwalcenie-konstytucji',
+        'only_matching': True,
+    }]
+
+    def _real_extract(self, url):
+        page_id = self._match_id(url)
+        webpage = self._download_webpage(url, page_id)
+        video_id = self._search_regex([
+            r'<iframe[^>]+src="[^"]*?object_id=(\d+)',
+            r"object_id\s*:\s*'(\d+)'",
+            r'data-video-id="(\d+)"'], webpage, 'video id', default=page_id)
+        return {
+            '_type': 'url_transparent',
+            'url': 'tvp:' + video_id,
+            'description': self._og_search_description(webpage, default=None),
+            'thumbnail': self._og_search_thumbnail(webpage),
+            'ie_key': 'TVPEmbed',
+        }
+
+
+class TVPEmbedIE(InfoExtractor):
+    IE_NAME = 'tvp:embed'
+    IE_DESC = 'Telewizja Polska'
+    _VALID_URL = r'(?:tvp:|https?://[^/]+\.tvp\.(?:pl|info)/sess/tvplayer\.php\?.*?object_id=)(?P<id>\d+)'
+
+    _TESTS = [{
+        'url': 'http://www.tvp.pl/sess/tvplayer.php?object_id=22670268',
+        'md5': '8c9cd59d16edabf39331f93bf8a766c7',
+        'info_dict': {
+            'id': '22670268',
+            'ext': 'mp4',
+            'title': 'Panorama, 07.12.2015, 15:40',
+        },
+    }, {
+        'url': 'tvp:22670268',
+        'only_matching': True,
+    }]
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
-        webpage = self._download_webpage(url, video_id)
-        json_url = 'http://www.tvp.pl/pub/stat/videofileinfo?video_id=%s' % video_id
-        params = self._download_json(
-            json_url, video_id, "Downloading video metadata")
-        video_url = params['video_url']
+
+        webpage = self._download_webpage(
+            'http://www.tvp.pl/sess/tvplayer.php?object_id=%s' % video_id, video_id)
+
+        error_massage = get_element_by_attribute('class', 'msg error', webpage)
+        if error_massage:
+            raise ExtractorError('%s said: %s' % (
+                self.IE_NAME, clean_html(error_massage)), expected=True)
+
+        title = self._search_regex(
+            r'name\s*:\s*([\'"])Title\1\s*,\s*value\s*:\s*\1(?P<title>.+?)\1',
+            webpage, 'title', group='title')
+        series_title = self._search_regex(
+            r'name\s*:\s*([\'"])SeriesTitle\1\s*,\s*value\s*:\s*\1(?P<series>.+?)\1',
+            webpage, 'series', group='series', default=None)
+        if series_title:
+            title = '%s, %s' % (series_title, title)
+
+        thumbnail = self._search_regex(
+            r"poster\s*:\s*'([^']+)'", webpage, 'thumbnail', default=None)
+
+        video_url = self._search_regex(
+            r'0:{src:([\'"])(?P<url>.*?)\1', webpage,
+            'formats', group='url', default=None)
+        if not video_url or 'material_niedostepny.mp4' in video_url:
+            video_url = self._download_json(
+                'http://www.tvp.pl/pub/stat/videofileinfo?video_id=%s' % video_id,
+                video_id)['video_url']
+
+        formats = []
+        video_url_base = self._search_regex(
+            r'(https?://.+?/video)(?:\.(?:ism|f4m|m3u8)|-\d+\.mp4)',
+            video_url, 'video base url', default=None)
+        if video_url_base:
+            # TODO: <Group> found instead of <AdaptationSet> in MPD manifest.
+            # It's not mentioned in MPEG-DASH standard. Figure that out.
+            # formats.extend(self._extract_mpd_formats(
+            #     video_url_base + '.ism/video.mpd',
+            #     video_id, mpd_id='dash', fatal=False))
+            formats.extend(self._extract_ism_formats(
+                video_url_base + '.ism/Manifest',
+                video_id, 'mss', fatal=False))
+            formats.extend(self._extract_f4m_formats(
+                video_url_base + '.ism/video.f4m',
+                video_id, f4m_id='hds', fatal=False))
+            m3u8_formats = self._extract_m3u8_formats(
+                video_url_base + '.ism/video.m3u8', video_id,
+                'mp4', 'm3u8_native', m3u8_id='hls', fatal=False)
+            self._sort_formats(m3u8_formats)
+            m3u8_formats = list(filter(
+                lambda f: f.get('vcodec') != 'none', m3u8_formats))
+            formats.extend(m3u8_formats)
+            for i, m3u8_format in enumerate(m3u8_formats, 2):
+                http_url = '%s-%d.mp4' % (video_url_base, i)
+                if self._is_valid_url(http_url, video_id):
+                    f = m3u8_format.copy()
+                    f.update({
+                        'url': http_url,
+                        'format_id': f['format_id'].replace('hls', 'http'),
+                        'protocol': 'http',
+                    })
+                    formats.append(f)
+        else:
+            formats = [{
+                'format_id': 'direct',
+                'url': video_url,
+                'ext': determine_ext(video_url, 'mp4'),
+            }]
+
+        self._sort_formats(formats)
 
         return {
             'id': video_id,
-            'title': self._og_search_title(webpage),
-            'ext': 'wmv',
-            'url': video_url,
-            'description': self._og_search_description(webpage),
-            'thumbnail': self._og_search_thumbnail(webpage),
+            'title': title,
+            'thumbnail': thumbnail,
+            'formats': formats,
+        }
+
+
+class TVPSeriesIE(InfoExtractor):
+    IE_NAME = 'tvp:series'
+    _VALID_URL = r'https?://vod\.tvp\.pl/(?:[^/]+/){2}(?P<id>[^/]+)/?$'
+
+    _TESTS = [{
+        'url': 'http://vod.tvp.pl/filmy-fabularne/filmy-za-darmo/ogniem-i-mieczem',
+        'info_dict': {
+            'title': 'Ogniem i mieczem',
+            'id': '4278026',
+        },
+        'playlist_count': 4,
+    }, {
+        'url': 'http://vod.tvp.pl/audycje/podroze/boso-przez-swiat',
+        'info_dict': {
+            'title': 'Boso przez świat',
+            'id': '9329207',
+        },
+        'playlist_count': 86,
+    }]
+
+    def _real_extract(self, url):
+        display_id = self._match_id(url)
+        webpage = self._download_webpage(url, display_id, tries=5)
+
+        title = self._html_search_regex(
+            r'(?s) id=[\'"]path[\'"]>(?:.*? / ){2}(.*?)</span>', webpage, 'series')
+        playlist_id = self._search_regex(r'nodeId:\s*(\d+)', webpage, 'playlist id')
+        playlist = self._download_webpage(
+            'http://vod.tvp.pl/vod/seriesAjax?type=series&nodeId=%s&recommend'
+            'edId=0&sort=&page=0&pageSize=10000' % playlist_id, display_id, tries=5,
+            note='Downloading playlist')
+
+        videos_paths = re.findall(
+            '(?s)class="shortTitle">.*?href="(/[^"]+)', playlist)
+        entries = [
+            self.url_result('http://vod.tvp.pl%s' % v_path, ie=TVPIE.ie_key())
+            for v_path in videos_paths]
+
+        return {
+            '_type': 'playlist',
+            'id': playlist_id,
+            'display_id': display_id,
+            'title': title,
+            'entries': entries,
         }