X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/33cd347759d6d999325ebf3c69b7ed5692c343b2..10aeb0702c7c06db2096fcaf319711f1f09d0508:/youtube_dl/extractor/mtv.py diff --git a/youtube_dl/extractor/mtv.py b/youtube_dl/extractor/mtv.py index 8f95657..24a79ae 100644 --- a/youtube_dl/extractor/mtv.py +++ b/youtube_dl/extractor/mtv.py @@ -26,6 +26,7 @@ class MTVIE(InfoExtractor): }, }, { + u'add_ie': ['Vevo'], u'url': u'http://www.mtv.com/videos/taylor-swift/916187/everything-has-changed-ft-ed-sheeran.jhtml', u'file': u'USCJY1331283.mp4', u'md5': u'73b4e7fcadd88929292fe52c3ced8caf', @@ -54,46 +55,57 @@ class MTVIE(InfoExtractor): def _get_thumbnail_url(self, uri, itemdoc): return 'http://mtv.mtvnimages.com/uri/' + uri - def _extract_video_url(self, metadataXml): + def _extract_video_formats(self, metadataXml): if '/error_country_block.swf' in metadataXml: raise ExtractorError(u'This video is not available from your country.', expected=True) mdoc = xml.etree.ElementTree.fromstring(metadataXml.encode('utf-8')) renditions = mdoc.findall('.//rendition') - # For now, always pick the highest quality. - rendition = renditions[-1] - - try: - _,_,ext = rendition.attrib['type'].partition('/') - format = ext + '-' + rendition.attrib['width'] + 'x' + rendition.attrib['height'] + '_' + rendition.attrib['bitrate'] - rtmp_video_url = rendition.find('./src').text - except KeyError: - raise ExtractorError('Invalid rendition field.') - video_url = self._transform_rtmp_url(rtmp_video_url) - return {'ext': ext, 'url': video_url, 'format': format} + formats = [] + for rendition in mdoc.findall('.//rendition'): + try: + _, _, ext = rendition.attrib['type'].partition('/') + rtmp_video_url = rendition.find('./src').text + formats.append({'ext': ext, + 'url': self._transform_rtmp_url(rtmp_video_url), + 'format_id': rendition.get('bitrate'), + 'width': int(rendition.get('width')), + 'height': int(rendition.get('height')), + }) + except (KeyError, TypeError): + raise ExtractorError('Invalid rendition field.') + return formats def _get_video_info(self, itemdoc): uri = itemdoc.find('guid').text video_id = self._id_from_uri(uri) self.report_extraction(video_id) mediagen_url = itemdoc.find('%s/%s' % (_media_xml_tag('group'), _media_xml_tag('content'))).attrib['url'] + # Remove the templates, like &device={device} + mediagen_url = re.sub(r'&[^=]*?={.*?}(?=(&|$))', u'', mediagen_url) if 'acceptMethods' not in mediagen_url: mediagen_url += '&acceptMethods=fms' mediagen_page = self._download_webpage(mediagen_url, video_id, u'Downloading video urls') - video_info = self._extract_video_url(mediagen_page) description_node = itemdoc.find('description') if description_node is not None: - description = description_node.text + description = description_node.text.strip() else: description = None - video_info.update({'title': itemdoc.find('title').text, - 'id': video_id, - 'thumbnail': self._get_thumbnail_url(uri, itemdoc), - 'description': description, - }) - return video_info + + info = { + 'title': itemdoc.find('title').text, + 'formats': self._extract_video_formats(mediagen_page), + 'id': video_id, + 'thumbnail': self._get_thumbnail_url(uri, itemdoc), + 'description': description, + } + + # TODO: Remove when #980 has been merged + info.update(info['formats'][-1]) + + return info def _get_videos_info(self, uri): video_id = self._id_from_uri(uri)