mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-24 05:36:56 +01:00
parent
55eae65b39
commit
fa7a1cc5ef
@ -1,12 +1,11 @@
|
||||
# encoding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
int_or_none,
|
||||
unified_strdate,
|
||||
js_to_json,
|
||||
)
|
||||
|
||||
|
||||
@ -22,59 +21,48 @@ def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
|
||||
playerdata = self._download_webpage(
|
||||
'http://player.screenwavemedia.com/play/player.php?id=%s' % video_id,
|
||||
'http://player.screenwavemedia.com/player.php?id=%s' % video_id,
|
||||
video_id, 'Downloading player webpage')
|
||||
|
||||
vidtitle = self._search_regex(
|
||||
r'\'vidtitle\'\s*:\s*"([^"]+)"', playerdata, 'vidtitle').replace('\\/', '/')
|
||||
vidurl = self._search_regex(
|
||||
r'\'vidurl\'\s*:\s*"([^"]+)"', playerdata, 'vidurl').replace('\\/', '/')
|
||||
|
||||
videolist_url = None
|
||||
playerconfig = self._download_webpage(
|
||||
'http://player.screenwavemedia.com/player.js',
|
||||
video_id, 'Downloading playerconfig webpage')
|
||||
|
||||
mobj = re.search(r"'videoserver'\s*:\s*'(?P<videoserver>[^']+)'", playerdata)
|
||||
if mobj:
|
||||
videoserver = mobj.group('videoserver')
|
||||
mobj = re.search(r'\'vidid\'\s*:\s*"(?P<vidid>[^\']+)"', playerdata)
|
||||
vidid = mobj.group('vidid') if mobj else video_id
|
||||
videolist_url = 'http://%s/vod/smil:%s.smil/jwplayer.smil' % (videoserver, vidid)
|
||||
else:
|
||||
mobj = re.search(r"file\s*:\s*'(?P<smil>http.+?/jwplayer\.smil)'", playerdata)
|
||||
if mobj:
|
||||
videolist_url = mobj.group('smil')
|
||||
videoserver = self._search_regex(r"'videoserver'\s*:\s*'([^']+)", playerconfig, 'videoserver')
|
||||
|
||||
sources = self._parse_json(
|
||||
js_to_json(
|
||||
self._search_regex(
|
||||
r"sources\s*:\s*(\[[^\]]+?\])", playerconfig,
|
||||
'sources',
|
||||
).replace(
|
||||
"' + thisObj.options.videoserver + '",
|
||||
videoserver
|
||||
).replace(
|
||||
"' + playerVidId + '",
|
||||
video_id
|
||||
)
|
||||
),
|
||||
video_id
|
||||
)
|
||||
|
||||
if videolist_url:
|
||||
videolist = self._download_xml(videolist_url, video_id, 'Downloading videolist XML')
|
||||
formats = []
|
||||
baseurl = vidurl[:vidurl.rfind('/') + 1]
|
||||
for video in videolist.findall('.//video'):
|
||||
src = video.get('src')
|
||||
if not src:
|
||||
continue
|
||||
file_ = src.partition(':')[-1]
|
||||
width = int_or_none(video.get('width'))
|
||||
height = int_or_none(video.get('height'))
|
||||
bitrate = int_or_none(video.get('system-bitrate'), scale=1000)
|
||||
format = {
|
||||
'url': baseurl + file_,
|
||||
'format_id': src.rpartition('.')[0].rpartition('_')[-1],
|
||||
}
|
||||
if width or height:
|
||||
format.update({
|
||||
'tbr': bitrate,
|
||||
'width': width,
|
||||
for source in sources:
|
||||
if source['type'] == 'hls':
|
||||
formats.extend(self._extract_m3u8_formats(source['file'], video_id))
|
||||
else:
|
||||
format_label = source.get('label')
|
||||
height = int_or_none(self._search_regex(
|
||||
r'^(\d+)[pP]', format_label, 'height', default=None))
|
||||
formats.append({
|
||||
'url': source['file'],
|
||||
'format': format_label,
|
||||
'ext': source.get('type'),
|
||||
'height': height,
|
||||
})
|
||||
else:
|
||||
format.update({
|
||||
'abr': bitrate,
|
||||
'vcodec': 'none',
|
||||
})
|
||||
formats.append(format)
|
||||
else:
|
||||
formats = [{
|
||||
'url': vidurl,
|
||||
}]
|
||||
self._sort_formats(formats)
|
||||
|
||||
return {
|
||||
|
Loading…
Reference in New Issue
Block a user