'description': 'Please use this to sell something. www.jonlajoie.com',
'thumbnail': 're:^http:.*\.jpg$',
},
+ 'params': {
+ 'skip_download': True,
+ },
}, {
'url': 'http://www.funnyordie.com/articles/ebf5e34fc8/10-hours-of-walking-in-nyc-as-a-man',
'only_matching': True,
formats = []
- formats.extend(self._extract_m3u8_formats(
- m3u8_url, video_id, 'mp4', 'm3u8_native', m3u8_id='hls', fatal=False))
+ m3u8_formats = self._extract_m3u8_formats(
+ m3u8_url, video_id, 'mp4', 'm3u8_native',
+ m3u8_id='hls', fatal=False)
+ source_formats = list(filter(
+ lambda f: f.get('vcodec') != 'none' and f.get('resolution') != 'multiple',
+ m3u8_formats))
- bitrates = [int(bitrate) for bitrate in re.findall(r'[,/]v(\d+)[,/]', m3u8_url)]
+ bitrates = [int(bitrate) for bitrate in re.findall(r'[,/]v(\d+)(?=[,/])', m3u8_url)]
bitrates.sort()
- for bitrate in bitrates:
- for link in links:
- formats.append({
- 'url': self._proto_relative_url('%s%d.%s' % (link[0], bitrate, link[1])),
- 'format_id': '%s-%d' % (link[1], bitrate),
- 'vbr': bitrate,
- })
+ if source_formats:
+ self._sort_formats(source_formats)
+
+ for bitrate, f in zip(bitrates, source_formats or [{}] * len(bitrates)):
+ for path, ext in links:
+ ff = f.copy()
+ if ff:
+ if ext != 'mp4':
+ ff = dict(
+ [(k, v) for k, v in ff.items()
+ if k in ('height', 'width', 'format_id')])
+ ff.update({
+ 'format_id': ff['format_id'].replace('hls', ext),
+ 'ext': ext,
+ 'protocol': 'http',
+ })
+ else:
+ ff.update({
+ 'format_id': '%s-%d' % (ext, bitrate),
+ 'vbr': bitrate,
+ })
+ ff['url'] = self._proto_relative_url(
+ '%s%d.%s' % (path, bitrate, ext))
+ formats.append(ff)
+ self._check_formats(formats, video_id)
+
+ formats.extend(m3u8_formats)
+ self._sort_formats(
+ formats, field_preference=('height', 'width', 'tbr', 'format_id'))
subtitles = {}
for src, src_lang in re.findall(r'<track kind="captions" src="([^"]+)" srclang="([^"]+)"', webpage):