From 1ff30d7b79504035bf8f642668361ca71d905ab7 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Sergey=20M=E2=80=A4?= Date: Sat, 20 Dec 2014 18:30:56 +0600 Subject: [PATCH] [npo] Add support for streams (Closes #4276) --- youtube_dl/extractor/npo.py | 88 +++++++++++++++++++++++++++---------- 1 file changed, 65 insertions(+), 23 deletions(-) diff --git a/youtube_dl/extractor/npo.py b/youtube_dl/extractor/npo.py index ce31694a5..8da76ae45 100644 --- a/youtube_dl/extractor/npo.py +++ b/youtube_dl/extractor/npo.py @@ -9,6 +9,7 @@ from ..utils import ( qualities, strip_jsonp, url_basename, + fix_xml_ampersands, ) @@ -51,7 +52,21 @@ class NPOIE(InfoExtractor): 'upload_date': '20130225', 'duration': 3000, }, - } + }, + { + 'url': 'http://www.npo.nl/de-nieuwe-mens-deel-1/21-07-2010/WO_VPRO_043706', + 'info_dict': { + 'id': 'WO_VPRO_043706', + 'ext': 'wmv', + 'title': 'De nieuwe mens - Deel 1', + 'description': 'md5:518ae51ba1293ffb80d8d8ce90b74e4b', + 'duration': 4680, + }, + 'params': { + # mplayer mms download + 'skip_download': True, + } + }, ] def _real_extract(self, url): @@ -74,31 +89,58 @@ class NPOIE(InfoExtractor): token = self._search_regex(r'npoplayer\.token = "(.+?)"', token_page, 'token') formats = [] - quality = qualities(['adaptive', 'wmv_sb', 'h264_sb', 'wmv_bb', 'h264_bb', 'wvc1_std', 'h264_std']) - for format_id in metadata['pubopties']: - format_info = self._download_json( - 'http://ida.omroep.nl/odi/?prid=%s&puboptions=%s&adaptive=yes&token=%s' % (video_id, format_id, token), - video_id, 'Downloading %s JSON' % format_id) - if format_info.get('error_code', 0) or format_info.get('errorcode', 0): - continue - streams = format_info.get('streams') - if streams: - video_info = self._download_json( - streams[0] + '&type=json', - video_id, 'Downloading %s stream JSON' % format_id) - else: - video_info = format_info - video_url = video_info.get('url') - if not video_url: - continue - if format_id == 'adaptive': - formats.extend(self._extract_m3u8_formats(video_url, video_id)) - else: + + pubopties = metadata.get('pubopties') + if pubopties: + quality = qualities(['adaptive', 'wmv_sb', 'h264_sb', 'wmv_bb', 'h264_bb', 'wvc1_std', 'h264_std']) + for format_id in pubopties: + format_info = self._download_json( + 'http://ida.omroep.nl/odi/?prid=%s&puboptions=%s&adaptive=yes&token=%s' + % (video_id, format_id, token), + video_id, 'Downloading %s JSON' % format_id) + if format_info.get('error_code', 0) or format_info.get('errorcode', 0): + continue + streams = format_info.get('streams') + if streams: + video_info = self._download_json( + streams[0] + '&type=json', + video_id, 'Downloading %s stream JSON' % format_id) + else: + video_info = format_info + video_url = video_info.get('url') + if not video_url: + continue + if format_id == 'adaptive': + formats.extend(self._extract_m3u8_formats(video_url, video_id)) + else: + formats.append({ + 'url': video_url, + 'format_id': format_id, + 'quality': quality(format_id), + }) + + streams = metadata.get('streams') + if streams: + for i, stream in enumerate(streams): + stream_url = stream.get('url') + if not stream_url: + continue + asx = self._download_xml( + stream_url, video_id, + 'Downloading stream %d ASX playlist' % i, + transform_source=fix_xml_ampersands) + ref = asx.find('./ENTRY/Ref') + if ref is None: + continue + video_url = ref.get('href') + if not video_url: + continue formats.append({ 'url': video_url, - 'format_id': format_id, - 'quality': quality(format_id), + 'ext': stream.get('formaat', 'asf'), + 'quality': stream.get('kwaliteit'), }) + self._sort_formats(formats) return {