mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-11-30 23:22:56 +01:00
133 lines
5.4 KiB
Python
133 lines
5.4 KiB
Python
from __future__ import unicode_literals
|
|
|
|
import json
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import int_or_none
|
|
|
|
|
|
class LiveLeakIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:\w+\.)?liveleak\.com/view\?(?:.*?)i=(?P<id>[\w_]+)(?:.*)'
|
|
_TESTS = [{
|
|
'url': 'http://www.liveleak.com/view?i=757_1364311680',
|
|
'md5': '50f79e05ba149149c1b4ea961223d5b3',
|
|
'info_dict': {
|
|
'id': '757_1364311680',
|
|
'ext': 'flv',
|
|
'description': 'extremely bad day for this guy..!',
|
|
'uploader': 'ljfriel2',
|
|
'title': 'Most unlucky car accident',
|
|
'thumbnail': 're:^https?://.*\.jpg$'
|
|
}
|
|
}, {
|
|
'url': 'http://www.liveleak.com/view?i=f93_1390833151',
|
|
'md5': 'b13a29626183c9d33944e6a04f41aafc',
|
|
'info_dict': {
|
|
'id': 'f93_1390833151',
|
|
'ext': 'mp4',
|
|
'description': 'German Television Channel NDR does an exclusive interview with Edward Snowden.\r\nUploaded on LiveLeak cause German Television thinks the rest of the world isn\'t intereseted in Edward Snowden.',
|
|
'uploader': 'ARD_Stinkt',
|
|
'title': 'German Television does first Edward Snowden Interview (ENGLISH)',
|
|
'thumbnail': 're:^https?://.*\.jpg$'
|
|
}
|
|
}, {
|
|
'url': 'http://www.liveleak.com/view?i=4f7_1392687779',
|
|
'md5': '42c6d97d54f1db107958760788c5f48f',
|
|
'info_dict': {
|
|
'id': '4f7_1392687779',
|
|
'ext': 'mp4',
|
|
'description': "The guy with the cigarette seems amazingly nonchalant about the whole thing... I really hope my friends' reactions would be a bit stronger.\r\n\r\nAction-go to 0:55.",
|
|
'uploader': 'CapObveus',
|
|
'title': 'Man is Fatally Struck by Reckless Car While Packing up a Moving Truck',
|
|
'age_limit': 18,
|
|
}
|
|
}, {
|
|
# Covers https://github.com/rg3/youtube-dl/pull/5983
|
|
'url': 'http://www.liveleak.com/view?i=801_1409392012',
|
|
'md5': '0b3bec2d888c20728ca2ad3642f0ef15',
|
|
'info_dict': {
|
|
'id': '801_1409392012',
|
|
'ext': 'mp4',
|
|
'description': 'Happened on 27.7.2014. \r\nAt 0:53 you can see people still swimming at near beach.',
|
|
'uploader': 'bony333',
|
|
'title': 'Crazy Hungarian tourist films close call waterspout in Croatia',
|
|
'thumbnail': 're:^https?://.*\.jpg$'
|
|
}
|
|
}]
|
|
|
|
@staticmethod
|
|
def _extract_url(webpage):
|
|
mobj = re.search(
|
|
r'<iframe[^>]+src="https?://(?:\w+\.)?liveleak\.com/ll_embed\?(?:.*?)i=(?P<id>[\w_]+)(?:.*)',
|
|
webpage)
|
|
if mobj:
|
|
return 'http://www.liveleak.com/view?i=%s' % mobj.group('id')
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
video_title = self._og_search_title(webpage).replace('LiveLeak.com -', '').strip()
|
|
video_description = self._og_search_description(webpage)
|
|
video_uploader = self._html_search_regex(
|
|
r'By:.*?(\w+)</a>', webpage, 'uploader', fatal=False)
|
|
age_limit = int_or_none(self._search_regex(
|
|
r'you confirm that you are ([0-9]+) years and over.',
|
|
webpage, 'age limit', default=None))
|
|
video_thumbnail = self._og_search_thumbnail(webpage)
|
|
|
|
sources_raw = self._search_regex(
|
|
r'(?s)sources:\s*(\[.*?\]),', webpage, 'video URLs', default=None)
|
|
if sources_raw is None:
|
|
alt_source = self._search_regex(
|
|
r'(file: ".*?"),', webpage, 'video URL', default=None)
|
|
if alt_source:
|
|
sources_raw = '[{ %s}]' % alt_source
|
|
else:
|
|
# Maybe an embed?
|
|
embed_url = self._search_regex(
|
|
r'<iframe[^>]+src="(http://www.prochan.com/embed\?[^"]+)"',
|
|
webpage, 'embed URL')
|
|
return {
|
|
'_type': 'url_transparent',
|
|
'url': embed_url,
|
|
'id': video_id,
|
|
'title': video_title,
|
|
'description': video_description,
|
|
'uploader': video_uploader,
|
|
'age_limit': age_limit,
|
|
}
|
|
|
|
sources_json = re.sub(r'\s([a-z]+):\s', r'"\1": ', sources_raw)
|
|
sources = json.loads(sources_json)
|
|
|
|
formats = [{
|
|
'format_id': '%s' % i,
|
|
'format_note': s.get('label'),
|
|
'url': s['file'],
|
|
} for i, s in enumerate(sources)]
|
|
for i, s in enumerate(sources):
|
|
# Removing '.h264_*.mp4' gives the raw video, which is essentially
|
|
# the same video without the LiveLeak logo at the top (see
|
|
# https://github.com/rg3/youtube-dl/pull/4768)
|
|
orig_url = re.sub(r'\.h264_.+?\.mp4', '', s['file'])
|
|
if s['file'] != orig_url:
|
|
formats.append({
|
|
'format_id': 'original-%s' % i,
|
|
'format_note': s.get('label'),
|
|
'url': orig_url,
|
|
'preference': 1,
|
|
})
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': video_title,
|
|
'description': video_description,
|
|
'uploader': video_uploader,
|
|
'formats': formats,
|
|
'age_limit': age_limit,
|
|
'thumbnail': video_thumbnail,
|
|
}
|