mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-02 14:37:21 +00:00
153 lines
5.1 KiB
Python
153 lines
5.1 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import os
|
|
import re
|
|
import subprocess
|
|
import tempfile
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
check_executable,
|
|
determine_ext,
|
|
encodeArgument,
|
|
ExtractorError,
|
|
)
|
|
|
|
|
|
class OpenloadIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:openload\.(?:co|io)|oload\.tv)/(?:f|embed)/(?P<id>[a-zA-Z0-9-_]+)'
|
|
|
|
_TESTS = [{
|
|
'url': 'https://openload.co/f/kUEfGclsU9o',
|
|
'md5': 'bf1c059b004ebc7a256f89408e65c36e',
|
|
'info_dict': {
|
|
'id': 'kUEfGclsU9o',
|
|
'ext': 'mp4',
|
|
'title': 'skyrim_no-audio_1080.mp4',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
},
|
|
}, {
|
|
'url': 'https://openload.co/embed/rjC09fkPLYs',
|
|
'info_dict': {
|
|
'id': 'rjC09fkPLYs',
|
|
'ext': 'mp4',
|
|
'title': 'movie.mp4',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
'subtitles': {
|
|
'en': [{
|
|
'ext': 'vtt',
|
|
}],
|
|
},
|
|
},
|
|
'params': {
|
|
'skip_download': True, # test subtitles only
|
|
},
|
|
}, {
|
|
'url': 'https://openload.co/embed/kUEfGclsU9o/skyrim_no-audio_1080.mp4',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://openload.io/f/ZAn6oz-VZGE/',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://openload.co/f/_-ztPaZtMhM/',
|
|
'only_matching': True,
|
|
}, {
|
|
# unavailable via https://openload.co/f/Sxz5sADo82g/, different layout
|
|
# for title and ext
|
|
'url': 'https://openload.co/embed/Sxz5sADo82g/',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://oload.tv/embed/KnG-kKZdcfY/',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
_PHANTOMJS_SCRIPT = r'''
|
|
phantom.onError = function(msg, trace) {
|
|
var msgStack = ['PHANTOM ERROR: ' + msg];
|
|
if(trace && trace.length) {
|
|
msgStack.push('TRACE:');
|
|
trace.forEach(function(t) {
|
|
msgStack.push(' -> ' + (t.file || t.sourceURL) + ': ' + t.line
|
|
+ (t.function ? ' (in function ' + t.function +')' : ''));
|
|
});
|
|
}
|
|
console.error(msgStack.join('\n'));
|
|
phantom.exit(1);
|
|
};
|
|
var page = require('webpage').create();
|
|
page.settings.resourceTimeout = 10000;
|
|
page.onInitialized = function() {
|
|
page.evaluate(function() {
|
|
delete window._phantom;
|
|
delete window.callPhantom;
|
|
});
|
|
};
|
|
page.open('https://openload.co/embed/%s/', function(status) {
|
|
var info = page.evaluate(function() {
|
|
return {
|
|
decoded_id: document.getElementById('streamurl').innerHTML,
|
|
title: document.querySelector('meta[name="og:title"],'
|
|
+ 'meta[name=description]').content
|
|
};
|
|
});
|
|
console.log(info.decoded_id + ' ' + info.title);
|
|
phantom.exit();
|
|
});'''
|
|
|
|
@staticmethod
|
|
def _extract_urls(webpage):
|
|
return re.findall(
|
|
r'<iframe[^>]+src=["\']((?:https?://)?(?:openload\.(?:co|io)|oload\.tv)/embed/[a-zA-Z0-9-_]+)',
|
|
webpage)
|
|
|
|
def _real_extract(self, url):
|
|
exe = check_executable('phantomjs', ['-v'])
|
|
if not exe:
|
|
raise ExtractorError('PhantomJS executable not found in PATH, '
|
|
'download it from http://phantomjs.org',
|
|
expected=True)
|
|
|
|
video_id = self._match_id(url)
|
|
url = 'https://openload.co/embed/%s/' % video_id
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
if 'File not found' in webpage or 'deleted by the owner' in webpage:
|
|
raise ExtractorError('File not found', expected=True, video_id=video_id)
|
|
|
|
script_file = tempfile.NamedTemporaryFile(mode='w', delete=False)
|
|
|
|
# write JS script to file and close it
|
|
with script_file:
|
|
script_file.write(self._PHANTOMJS_SCRIPT % video_id)
|
|
|
|
self.to_screen('%s: Decoding video ID with PhantomJS' % video_id)
|
|
|
|
p = subprocess.Popen([exe, '--ssl-protocol=any', script_file.name],
|
|
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
|
|
output, err = p.communicate()
|
|
if p.returncode != 0:
|
|
raise ExtractorError('Decoding failed\n:'
|
|
+ encodeArgument(err))
|
|
else:
|
|
decoded_id, title = encodeArgument(output).strip().split(' ', 1)
|
|
|
|
os.remove(script_file.name)
|
|
|
|
video_url = 'https://openload.co/stream/%s?mime=true' % decoded_id
|
|
|
|
entries = self._parse_html5_media_entries(url, webpage, video_id)
|
|
entry = entries[0] if entries else {}
|
|
subtitles = entry.get('subtitles')
|
|
|
|
info_dict = {
|
|
'id': video_id,
|
|
'title': title,
|
|
'thumbnail': entry.get('thumbnail') or self._og_search_thumbnail(webpage, default=None),
|
|
'url': video_url,
|
|
# Seems all videos have extensions in their titles
|
|
'ext': determine_ext(title, 'mp4'),
|
|
'subtitles': subtitles,
|
|
}
|
|
return info_dict
|