mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-26 18:32:39 +00:00
[mxplayer] Add new extractor
https://github.com/ytdl-org/youtube-dl/pull/27325 Authored by: codeasashu
This commit is contained in:
parent
f3b737ed19
commit
097b056c5a
@ -738,6 +738,7 @@
|
|||||||
)
|
)
|
||||||
from .muenchentv import MuenchenTVIE
|
from .muenchentv import MuenchenTVIE
|
||||||
from .mwave import MwaveIE, MwaveMeetGreetIE
|
from .mwave import MwaveIE, MwaveMeetGreetIE
|
||||||
|
from .mxplayer import MxplayerIE
|
||||||
from .mychannels import MyChannelsIE
|
from .mychannels import MyChannelsIE
|
||||||
from .myspace import MySpaceIE, MySpaceAlbumIE
|
from .myspace import MySpaceIE, MySpaceAlbumIE
|
||||||
from .myspass import MySpassIE
|
from .myspass import MySpassIE
|
||||||
|
103
yt_dlp/extractor/mxplayer.py
Normal file
103
yt_dlp/extractor/mxplayer.py
Normal file
@ -0,0 +1,103 @@
|
|||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
import re
|
||||||
|
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..utils import (
|
||||||
|
ExtractorError,
|
||||||
|
js_to_json,
|
||||||
|
url_or_none,
|
||||||
|
urljoin,
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
|
VALID_STREAMS = ('dash', )
|
||||||
|
|
||||||
|
|
||||||
|
class MxplayerIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?mxplayer\.in/movie/(?P<slug>[a-z0-9]+(?:-[a-z0-9]+)*)'
|
||||||
|
_TEST = {
|
||||||
|
'url': 'https://www.mxplayer.in/movie/watch-knock-knock-hindi-dubbed-movie-online-b9fa28df3bfb8758874735bbd7d2655a?watch=true',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'b9fa28df3bfb8758874735bbd7d2655a',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Knock Knock Movie | Watch 2015 Knock Knock Full Movie Online- MX Player',
|
||||||
|
'description': 'md5:b195ba93ff1987309cfa58e2839d2a5b'
|
||||||
|
},
|
||||||
|
'params': {
|
||||||
|
'skip_download': True,
|
||||||
|
'format': 'bestvideo'
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
def _get_best_stream_url(self, stream):
|
||||||
|
best_stream = list(filter(None, [v for k, v in stream.items()]))
|
||||||
|
return best_stream.pop(0) if len(best_stream) else None
|
||||||
|
|
||||||
|
def _get_stream_urls(self, video_dict):
|
||||||
|
stream_dict = video_dict.get('stream', {'provider': {}})
|
||||||
|
stream_provider = stream_dict.get('provider')
|
||||||
|
|
||||||
|
if not stream_dict[stream_provider]:
|
||||||
|
message = 'No stream provider found'
|
||||||
|
raise ExtractorError('%s said: %s' % (self.IE_NAME, message), expected=True)
|
||||||
|
|
||||||
|
streams = []
|
||||||
|
for stream_name, v in stream_dict[stream_provider].items():
|
||||||
|
if stream_name in VALID_STREAMS:
|
||||||
|
stream_url = self._get_best_stream_url(v)
|
||||||
|
if stream_url is None:
|
||||||
|
continue
|
||||||
|
streams.append((stream_name, stream_url))
|
||||||
|
return streams
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
mobj = re.match(self._VALID_URL, url)
|
||||||
|
video_slug = mobj.group('slug')
|
||||||
|
|
||||||
|
video_id = video_slug.split('-')[-1]
|
||||||
|
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
|
||||||
|
window_state_json = self._html_search_regex(
|
||||||
|
r'(?s)<script>window\.state\s*[:=]\s(\{.+\})\n(\w+).*(</script>).*',
|
||||||
|
webpage, 'WindowState')
|
||||||
|
|
||||||
|
source = self._parse_json(js_to_json(window_state_json), video_id)
|
||||||
|
if not source:
|
||||||
|
raise ExtractorError('Cannot find source', expected=True)
|
||||||
|
|
||||||
|
config_dict = source['config']
|
||||||
|
video_dict = source['entities'][video_id]
|
||||||
|
stream_urls = self._get_stream_urls(video_dict)
|
||||||
|
|
||||||
|
title = self._og_search_title(webpage, fatal=True, default=video_dict['title'])
|
||||||
|
|
||||||
|
formats = []
|
||||||
|
headers = {'Referer': url}
|
||||||
|
for stream_name, stream_url in stream_urls:
|
||||||
|
if stream_name == 'dash':
|
||||||
|
format_url = url_or_none(urljoin(config_dict['videoCdnBaseUrl'], stream_url))
|
||||||
|
if not format_url:
|
||||||
|
continue
|
||||||
|
formats.extend(self._extract_mpd_formats(
|
||||||
|
format_url, video_id, mpd_id='dash', headers=headers))
|
||||||
|
|
||||||
|
self._sort_formats(formats)
|
||||||
|
info = {
|
||||||
|
'id': video_id,
|
||||||
|
'title': title,
|
||||||
|
'description': video_dict.get('description'),
|
||||||
|
'formats': formats
|
||||||
|
}
|
||||||
|
|
||||||
|
if video_dict.get('imageInfo'):
|
||||||
|
info['thumbnails'] = list(map(lambda i: dict(i, **{
|
||||||
|
'url': urljoin(config_dict['imageBaseUrl'], i['url'])
|
||||||
|
}), video_dict['imageInfo']))
|
||||||
|
|
||||||
|
if video_dict.get('webUrl'):
|
||||||
|
last_part = video_dict['webUrl'].split("/")[-1]
|
||||||
|
info['display_id'] = last_part.replace(video_id, "").rstrip("-")
|
||||||
|
|
||||||
|
return info
|
Loading…
Reference in New Issue
Block a user