diff --git a/youtube_dl/extractor/vgtv.py b/youtube_dl/extractor/vgtv.py index 0964b96aa..5ab6168a9 100644 --- a/youtube_dl/extractor/vgtv.py +++ b/youtube_dl/extractor/vgtv.py @@ -4,70 +4,113 @@ from __future__ import unicode_literals import re from .common import InfoExtractor +from ..utils import float_or_none -from ..utils import ( - ExtractorError -) class VGTVIE(InfoExtractor): - # Because of the #! in the URL structure we need to add ' before and after given URL. - # Or else it will cry: -bash: !/video/100495/lars-og-lars-sesong-6-episode-6-lakselus: event not found - _VALID_URL = r'http://(?:www\.)?vgtv\.no/#!/(?:.*)/(?P[0-9]+)/(?P[^?#]*)' - _TEST = { - 'url': 'http://www.vgtv.no/#!/video/84196/hevnen-er-soet-episode-10-abu', - 'md5': 'b8be7a234cebb840c0d512c78013e02f', - 'info_dict': { - 'id': '84196', - 'ext': 'mp4', - 'title': 'Hevnen er søt episode 10: Abu', - 'description': 'md5:e25e4badb5f544b04341e14abdc72234', - 'timestamp': 1404626400, - 'upload_date': '20140706' - } - } + _VALID_URL = r'http://(?:www\.)?vgtv\.no/#!/(?:.*)/(?P<id>[0-9]+)' + _TESTS = [ + { + # streamType: vod + 'url': 'http://www.vgtv.no/#!/video/84196/hevnen-er-soet-episode-10-abu', + 'md5': 'b8be7a234cebb840c0d512c78013e02f', + 'info_dict': { + 'id': '84196', + 'ext': 'mp4', + 'title': 'Hevnen er søt episode 10: Abu', + 'description': 'md5:e25e4badb5f544b04341e14abdc72234', + 'thumbnail': 're:^https?://.*\.jpg', + 'duration': 648.000, + 'timestamp': 1404626400, + 'upload_date': '20140706' + }, + }, + { + # streamType: wasLive + 'url': 'http://www.vgtv.no/#!/live/100764/opptak-vgtv-foelger-em-kvalifiseringen', + 'info_dict': { + 'id': '100764', + 'ext': 'mp4', + 'title': 'OPPTAK: VGTV følger EM-kvalifiseringen', + 'description': 'md5:3772d9c0dc2dff92a886b60039a7d4d3', + 'thumbnail': 're:^https?://.*\.jpg', + 'duration': 9056.000, + 'timestamp': 1410113864, + 'upload_date': '20140907' + }, + 'params': { + # m3u8 download + 'skip_download': True, + }, + }, + { + # streamType: live + 'url': 'http://www.vgtv.no/#!/live/100015/direkte-her-kan-du-se-laksen-live-fra-suldalslaagen', + 'info_dict': { + 'id': '100015', + 'ext': 'mp4', + 'title': 'DIREKTE: Her kan du se laksen live fra Suldalslågen!', + 'description': 'md5:9a60cc23fa349f761628924e56eeec2d', + 'thumbnail': 're:^https?://.*\.jpg', + 'duration': 0, + 'timestamp': 1407423348, + 'upload_date': '20140807' + }, + 'params': { + # m3u8 download + 'skip_download': True, + }, + }, + ] - def _real_extract(self, url): - mobj = re.match(self._VALID_URL, url) - video_id = mobj.group('id') + def _real_extract(self, url): + mobj = re.match(self._VALID_URL, url) + video_id = mobj.group('id') - # Download JSON file containing video info. - data = self._download_json('http://svp.vg.no/svp/api/v1/vgtv/assets/%s?appName=vgtv-website' % video_id, video_id, 'Downloading media JSON') + data = self._download_json( + 'http://svp.vg.no/svp/api/v1/vgtv/assets/%s?appName=vgtv-website' % video_id, + video_id, 'Downloading media JSON') - # Known streamType: vod, live, wasLive - # Will it even be possible to add support for live streams? - if data['streamType'] != 'vod': - raise ExtractorError('Stream type \'%s\' is not yet supported.' % data['streamType'], expected=True) + streams = data['streamUrls'] - # Add access token to image or it will fail. - thumbnail = data['images']['main'] + '?t[]=900x506q80' + formats = [] - formats = [] + hls_url = streams.get('hls') + if hls_url: + formats.extend(self._extract_m3u8_formats(hls_url, video_id, 'mp4')) - # Most videos are in MP4, but some are either HLS or HDS. - # Don't want to support HDS. - if data['streamUrls']['mp4'] is not None: - formats.append({ - 'url': data['streamUrls']['mp4'], - 'format_id': 'mp4', - 'ext': 'mp4' - }) - elif data['streamUrls']['hls'] is not None: - self.to_screen(u'No MP4 URL found, using m3u8. This may take some extra time.') - formats.append({ - 'url': data['streamUrls']['hls'], - 'format_id': 'm3u8', - 'ext': 'mp4' - }) - else: - raise ExtractorError('No download URL found for video: %s.' % video_id, expected=True) + hds_url = streams.get('hds') + if hds_url: + formats.extend(self._extract_f4m_formats(hds_url + '?hdcore=3.2.0&plugin=aasp-3.2.0.77.18', video_id)) - return { - 'id': video_id, - 'title': data['title'], - 'description': data['description'], - 'thumbnail': thumbnail, - 'timestamp': data['published'], - 'duration': data['duration'], - 'view_count': data['displays'], - 'formats': formats, - } \ No newline at end of file + mp4_url = streams.get('mp4') + if mp4_url: + _url = hls_url or hds_url + MP4_URL_TEMPLATE = '%s/%%s.%s' % (mp4_url.rpartition('/')[0], mp4_url.rpartition('.')[-1]) + for mp4_format in _url.split(','): + m = re.search('(?P<width>\d+)_(?P<height>\d+)_(?P<vbr>\d+)', mp4_format) + if not m: + continue + width = int(m.group('width')) + height = int(m.group('height')) + vbr = int(m.group('vbr')) + formats.append({ + 'url': MP4_URL_TEMPLATE % mp4_format, + 'format_id': 'mp4-%s' % vbr, + 'width': width, + 'height': height, + 'vbr': vbr, + 'preference': 1, + }) + self._sort_formats(formats) + + return { + 'id': video_id, + 'title': data['title'], + 'description': data['description'], + 'thumbnail': data['images']['main'] + '?t[]=900x506q80', + 'timestamp': data['published'], + 'duration': float_or_none(data['duration'], 1000), + 'view_count': data['displays'], + 'formats': formats, + } \ No newline at end of file