- from __future__ import unicode_literals
-
- import os.path
- import re
-
- from .fragment import FragmentFD
- from .external import FFmpegFD
-
- from ..compat import compat_urlparse
- from ..utils import (
- encodeFilename,
- sanitize_open,
- )
-
-
- class HlsFD(FragmentFD):
- """ A limited implementation that does not require ffmpeg """
-
- FD_NAME = 'hlsnative'
-
- @staticmethod
- def can_download(manifest):
- UNSUPPORTED_FEATURES = (
- r'#EXT-X-KEY:METHOD=(?!NONE)', # encrypted streams [1]
- r'#EXT-X-BYTERANGE', # playlists composed of byte ranges of media files [2]
-
- # Live streams heuristic does not always work (e.g. geo restricted to Germany
- # http://hls-geo.daserste.de/i/videoportal/Film/c_620000/622873/format,716451,716457,716450,716458,716459,.mp4.csmil/index_4_av.m3u8?null=0)
- # r'#EXT-X-MEDIA-SEQUENCE:(?!0$)', # live streams [3]
-
- # This heuristic also is not correct since segments may not be appended as well.
- # Twitch vods of finished streams have EXT-X-PLAYLIST-TYPE:EVENT despite
- # no segments will definitely be appended to the end of the playlist.
- # r'#EXT-X-PLAYLIST-TYPE:EVENT', # media segments may be appended to the end of
- # # event media playlists [4]
-
- # 1. https://tools.ietf.org/html/draft-pantos-http-live-streaming-17#section-4.3.2.4
- # 2. https://tools.ietf.org/html/draft-pantos-http-live-streaming-17#section-4.3.2.2
- # 3. https://tools.ietf.org/html/draft-pantos-http-live-streaming-17#section-4.3.3.2
- # 4. https://tools.ietf.org/html/draft-pantos-http-live-streaming-17#section-4.3.3.5
- )
- return all(not re.search(feature, manifest) for feature in UNSUPPORTED_FEATURES)
-
- def real_download(self, filename, info_dict):
- man_url = info_dict['url']
- self.to_screen('[%s] Downloading m3u8 manifest' % self.FD_NAME)
- manifest = self.ydl.urlopen(man_url).read()
-
- s = manifest.decode('utf-8', 'ignore')
-
- if not self.can_download(s):
- self.report_warning(
- 'hlsnative has detected features it does not support, '
- 'extraction will be delegated to ffmpeg')
- fd = FFmpegFD(self.ydl, self.params)
- for ph in self._progress_hooks:
- fd.add_progress_hook(ph)
- return fd.real_download(filename, info_dict)
-
- fragment_urls = []
- for line in s.splitlines():
- line = line.strip()
- if line and not line.startswith('#'):
- segment_url = (
- line
- if re.match(r'^https?://', line)
- else compat_urlparse.urljoin(man_url, line))
- fragment_urls.append(segment_url)
- # We only download the first fragment during the test
- if self.params.get('test', False):
- break
-
- ctx = {
- 'filename': filename,
- 'total_frags': len(fragment_urls),
- }
-
- self._prepare_and_start_frag_download(ctx)
-
- frags_filenames = []
- for i, frag_url in enumerate(fragment_urls):
- frag_filename = '%s-Frag%d' % (ctx['tmpfilename'], i)
- success = ctx['dl'].download(frag_filename, {'url': frag_url})
- if not success:
- return False
- down, frag_sanitized = sanitize_open(frag_filename, 'rb')
- ctx['dest_stream'].write(down.read())
- down.close()
- frags_filenames.append(frag_sanitized)
-
- self._finish_frag_download(ctx)
-
- for frag_file in frags_filenames:
- os.remove(encodeFilename(frag_file))
-
- return True
|