2 changed files with 145 additions and 0 deletions
Split View
Diff Options
@ -0,0 +1,144 @@ |
|||
# coding: utf-8 |
|||
from __future__ import unicode_literals |
|||
|
|||
from .common import InfoExtractor |
|||
from ..utils import qualities |
|||
from ..compat import compat_urllib_request |
|||
|
|||
|
|||
_QUALITY = qualities(('low', 'high')) |
|||
|
|||
|
|||
class FilmOnIE(InfoExtractor): |
|||
_VALID_URL = r'https?://(?:www\.)?filmon\.com/(?:tv|channel)/(?P<id>[a-z0-9-]+)' |
|||
_TESTS = [{ |
|||
'url': 'https://www.filmon.com/channel/filmon-sports', |
|||
'only_matching': True, |
|||
}, { |
|||
'url': 'https://www.filmon.com/tv/2894', |
|||
'only_matching': True, |
|||
}] |
|||
|
|||
def _real_extract(self, url): |
|||
channel_id = self._match_id(url) |
|||
|
|||
request = compat_urllib_request.Request('https://www.filmon.com/channel/%s' % (channel_id)) |
|||
request.add_header('X-Requested-With', 'XMLHttpRequest') |
|||
channel_info = self._download_json(request, channel_id) |
|||
now_playing = channel_info['now_playing'] |
|||
|
|||
thumbnails = [] |
|||
for thumb in now_playing.get('images', ()): |
|||
if thumb['type'] != '2': |
|||
continue |
|||
thumbnails.append({ |
|||
'url': thumb['url'], |
|||
'width': int(thumb['width']), |
|||
'height': int(thumb['height']), |
|||
}) |
|||
|
|||
formats = [] |
|||
|
|||
for stream in channel_info['streams']: |
|||
formats.append({ |
|||
'format_id': str(stream['id']), |
|||
# this is an m3u8 stream, but we are deliberately not using _extract_m3u8_formats |
|||
# because 0) it doesn't have bitrate variants anyway, and 1) the ids generated |
|||
# by that method are highly unstable (because the bitrate is variable) |
|||
'url': stream['url'], |
|||
'resolution': stream['name'], |
|||
'format_note': 'expires after %u seconds' % int(stream['watch-timeout']), |
|||
'ext': 'mp4', |
|||
'quality': _QUALITY(stream['quality']), |
|||
'preference': int(stream['watch-timeout']), |
|||
}) |
|||
self._sort_formats(formats) |
|||
|
|||
return { |
|||
'id': str(channel_info['id']), |
|||
'display_id': channel_info['alias'], |
|||
'formats': formats, |
|||
# XXX: use the channel description (channel_info['description'])? |
|||
'uploader_id': channel_info['alias'], |
|||
'uploader': channel_info['title'], # XXX: kinda stretching it... |
|||
'title': now_playing.get('programme_name') or channel_info['title'], |
|||
'description': now_playing.get('programme_description'), |
|||
'thumbnails': thumbnails, |
|||
'is_live': True, |
|||
} |
|||
|
|||
|
|||
class FilmOnVODIE(InfoExtractor): |
|||
_VALID_URL = r'https?://(?:www\.)?filmon\.com/vod/view/(?P<id>\d+)' |
|||
_TESTS = [{ |
|||
'url': 'https://www.filmon.com/vod/view/24869-0-plan-9-from-outer-space', |
|||
'info_dict': { |
|||
'id': '24869', |
|||
'ext': 'mp4', |
|||
'title': 'Plan 9 From Outer Space', |
|||
'description': 'Dead human, zombies and vampires', |
|||
}, |
|||
}, { |
|||
'url': 'https://www.filmon.com/vod/view/2825-1-popeye-series-1', |
|||
'info_dict': { |
|||
'id': '2825', |
|||
'title': 'Popeye Series 1', |
|||
}, |
|||
'playlist_count': 8, |
|||
}] |
|||
|
|||
def _real_extract(self, url): |
|||
video_id = self._match_id(url) |
|||
|
|||
result = self._download_json('https://www.filmon.com/api/vod/movie?id=%s' % (video_id), video_id) |
|||
if result['code'] != 200: |
|||
raise ExtractorError('FilmOn said: %s' % (result['reason']), expected=True) |
|||
|
|||
response = result['response'] |
|||
|
|||
if response.get('episodes'): |
|||
return { |
|||
'_type': 'playlist', |
|||
'id': video_id, |
|||
'title': response['title'], |
|||
'entries': [{ |
|||
'_type': 'url', |
|||
'url': 'https://www.filmon.com/vod/view/%s' % (ep), |
|||
} for ep in response['episodes']] |
|||
} |
|||
|
|||
formats = [] |
|||
for (id, stream) in response['streams'].items(): |
|||
formats.append({ |
|||
'format_id': id, |
|||
'url': stream['url'], |
|||
'resolution': stream['name'], |
|||
'format_note': 'expires after %u seconds' % int(stream['watch-timeout']), |
|||
'ext': 'mp4', |
|||
'quality': _QUALITY(stream['quality']), |
|||
'preference': int(stream['watch-timeout']), |
|||
}) |
|||
self._sort_formats(formats) |
|||
|
|||
poster = response['poster'] |
|||
thumbnails = [{ |
|||
'id': 'poster', |
|||
'url': poster['url'], |
|||
'width': poster['width'], |
|||
'height': poster['height'], |
|||
}] |
|||
for (id, thumb) in poster['thumbs'].items(): |
|||
thumbnails.append({ |
|||
'id': id, |
|||
'url': thumb['url'], |
|||
'width': thumb['width'], |
|||
'height': thumb['height'], |
|||
}) |
|||
|
|||
return { |
|||
'id': video_id, |
|||
'title': response['title'], |
|||
'formats': formats, |
|||
'description': response['description'], |
|||
'thumbnails': thumbnails, |
|||
} |
Write
Preview
Loading…
Cancel
Save