diff options
Diffstat (limited to 'youtube_dl/extractor/tv5mondeplus.py')
-rw-r--r-- | youtube_dl/extractor/tv5mondeplus.py | 79 |
1 files changed, 0 insertions, 79 deletions
diff --git a/youtube_dl/extractor/tv5mondeplus.py b/youtube_dl/extractor/tv5mondeplus.py deleted file mode 100644 index 88b6baa31..000000000 --- a/youtube_dl/extractor/tv5mondeplus.py +++ /dev/null @@ -1,79 +0,0 @@ -# coding: utf-8 -from __future__ import unicode_literals - -from .common import InfoExtractor -from ..utils import ( - clean_html, - determine_ext, - extract_attributes, - get_element_by_class, - int_or_none, - parse_duration, - parse_iso8601, -) - - -class TV5MondePlusIE(InfoExtractor): - IE_DESC = 'TV5MONDE+' - _VALID_URL = r'https?://(?:www\.)?tv5mondeplus\.com/toutes-les-videos/[^/]+/(?P<id>[^/?#]+)' - _TEST = { - 'url': 'http://www.tv5mondeplus.com/toutes-les-videos/documentaire/tdah-mon-amour-tele-quebec-tdah-mon-amour-ep001-enfants', - 'md5': '12130fc199f020673138a83466542ec6', - 'info_dict': { - 'id': 'tdah-mon-amour-tele-quebec-tdah-mon-amour-ep001-enfants', - 'ext': 'mp4', - 'title': 'Tdah, mon amour - Enfants', - 'description': 'md5:230e3aca23115afcf8006d1bece6df74', - 'upload_date': '20170401', - 'timestamp': 1491022860, - } - } - _GEO_BYPASS = False - - def _real_extract(self, url): - display_id = self._match_id(url) - webpage = self._download_webpage(url, display_id) - - if ">Ce programme n'est malheureusement pas disponible pour votre zone géographique.<" in webpage: - self.raise_geo_restricted(countries=['FR']) - - series = get_element_by_class('video-detail__title', webpage) - title = episode = get_element_by_class( - 'video-detail__subtitle', webpage) or series - if series and series != title: - title = '%s - %s' % (series, title) - vpl_data = extract_attributes(self._search_regex( - r'(<[^>]+class="video_player_loader"[^>]+>)', - webpage, 'video player loader')) - - video_files = self._parse_json( - vpl_data['data-broadcast'], display_id).get('files', []) - formats = [] - for video_file in video_files: - v_url = video_file.get('url') - if not v_url: - continue - video_format = video_file.get('format') or determine_ext(v_url) - if video_format == 'm3u8': - formats.extend(self._extract_m3u8_formats( - v_url, display_id, 'mp4', 'm3u8_native', - m3u8_id='hls', fatal=False)) - else: - formats.append({ - 'url': v_url, - 'format_id': video_format, - }) - self._sort_formats(formats) - - return { - 'id': display_id, - 'display_id': display_id, - 'title': title, - 'description': clean_html(get_element_by_class('video-detail__description', webpage)), - 'thumbnail': vpl_data.get('data-image'), - 'duration': int_or_none(vpl_data.get('data-duration')) or parse_duration(self._html_search_meta('duration', webpage)), - 'timestamp': parse_iso8601(self._html_search_meta('uploadDate', webpage)), - 'formats': formats, - 'episode': episode, - 'series': series, - } |