X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/b8d8e13c1f9e4d3cdd7d41c5c9d711a36dd5f9c3..4cd837fe5fe8d00dd8edcc06fef3a4c9b63c32bc:/youtube_dl/extractor/rtlnl.py?ds=inline diff --git a/youtube_dl/extractor/rtlnl.py b/youtube_dl/extractor/rtlnl.py index 54076de..be36acc 100644 --- a/youtube_dl/extractor/rtlnl.py +++ b/youtube_dl/extractor/rtlnl.py @@ -12,10 +12,10 @@ class RtlNlIE(InfoExtractor): IE_NAME = 'rtl.nl' IE_DESC = 'rtl.nl and rtlxl.nl' _VALID_URL = r'''(?x) - https?://(?:www\.)? + https?://(?:(?:www|static)\.)? (?: rtlxl\.nl/[^\#]*\#!/[^/]+/| - rtl\.nl/system/videoplayer/(?:[^/]+/)+(?:video_)?embed\.html\b.+?\buuid= + rtl\.nl/(?:(?:system/videoplayer/(?:[^/]+/)+(?:video_)?embed\.html|embed)\b.+?\buuid=|video/) ) (?P[0-9a-f-]+)''' @@ -70,6 +70,12 @@ class RtlNlIE(InfoExtractor): }, { 'url': 'http://rtlxl.nl/?_ga=1.204735956.572365465.1466978370#!/rtl-nieuws-132237/3c487912-023b-49ac-903e-2c5d79f8410f', 'only_matching': True, + }, { + 'url': 'https://www.rtl.nl/video/c603c9c2-601d-4b5e-8175-64f1e942dc7d/', + 'only_matching': True, + }, { + 'url': 'https://static.rtl.nl/embed/?uuid=1a2970fc-5c0b-43ff-9fdc-927e39e6d1bc&autoplay=false&publicatiepunt=rtlnieuwsnl', + 'only_matching': True, }] def _real_extract(self, url): @@ -87,58 +93,11 @@ class RtlNlIE(InfoExtractor): meta = info.get('meta', {}) - # m3u8 streams are encrypted and may not be handled properly by older ffmpeg/avconv. - # To workaround this previously adaptive -> flash trick was used to obtain - # unencrypted m3u8 streams (see https://github.com/rg3/youtube-dl/issues/4118) - # and bypass georestrictions as well. - # Currently, unencrypted m3u8 playlists are (intentionally?) invalid and therefore - # unusable albeit can be fixed by simple string replacement (see - # https://github.com/rg3/youtube-dl/pull/6337) - # Since recent ffmpeg and avconv handle encrypted streams just fine encrypted - # streams are used now. videopath = material['videopath'] m3u8_url = meta.get('videohost', 'http://manifest.us.rtl.nl') + videopath formats = self._extract_m3u8_formats( m3u8_url, uuid, 'mp4', m3u8_id='hls', fatal=False) - - video_urlpart = videopath.split('/adaptive/')[1][:-5] - PG_URL_TEMPLATE = 'http://pg.us.rtl.nl/rtlxl/network/%s/progressive/%s.mp4' - - PG_FORMATS = ( - ('a2t', 512, 288), - ('a3t', 704, 400), - ('nettv', 1280, 720), - ) - - def pg_format(format_id, width, height): - return { - 'url': PG_URL_TEMPLATE % (format_id, video_urlpart), - 'format_id': 'pg-%s' % format_id, - 'protocol': 'http', - 'width': width, - 'height': height, - } - - if not formats: - formats = [pg_format(*pg_tuple) for pg_tuple in PG_FORMATS] - else: - pg_formats = [] - for format_id, width, height in PG_FORMATS: - try: - # Find hls format with the same width and height corresponding - # to progressive format and copy metadata from it. - f = next(f for f in formats if f.get('height') == height) - # hls formats may have invalid width - f['width'] = width - f_copy = f.copy() - f_copy.update(pg_format(format_id, width, height)) - pg_formats.append(f_copy) - except StopIteration: - # Missing hls format does mean that no progressive format with - # such width and height exists either. - pass - formats.extend(pg_formats) self._sort_formats(formats) thumbnails = []