mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-27 07:06:52 +01:00
96d0f8c1cb
Authored by: michal-repo
138 lines
4.9 KiB
Python
138 lines
4.9 KiB
Python
from .common import InfoExtractor
|
|
from ..utils import (
|
|
ExtractorError,
|
|
encode_base_n,
|
|
get_elements_by_class,
|
|
int_or_none,
|
|
join_nonempty,
|
|
merge_dicts,
|
|
parse_duration,
|
|
str_to_int,
|
|
url_or_none,
|
|
)
|
|
|
|
|
|
class EpornerIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?eporner\.com/(?:(?:hd-porn|embed)/|video-)(?P<id>\w+)(?:/(?P<display_id>[\w-]+))?'
|
|
_TESTS = [{
|
|
'url': 'http://www.eporner.com/hd-porn/95008/Infamous-Tiffany-Teen-Strip-Tease-Video/',
|
|
'md5': '39d486f046212d8e1b911c52ab4691f8',
|
|
'info_dict': {
|
|
'id': 'qlDUmNsj6VS',
|
|
'display_id': 'Infamous-Tiffany-Teen-Strip-Tease-Video',
|
|
'ext': 'mp4',
|
|
'title': 'Infamous Tiffany Teen Strip Tease Video',
|
|
'description': 'md5:764f39abf932daafa37485eb46efa152',
|
|
'timestamp': 1232520922,
|
|
'upload_date': '20090121',
|
|
'duration': 1838,
|
|
'view_count': int,
|
|
'age_limit': 18,
|
|
},
|
|
'params': {
|
|
'proxy': '127.0.0.1:8118'
|
|
}
|
|
}, {
|
|
# New (May 2016) URL layout
|
|
'url': 'http://www.eporner.com/hd-porn/3YRUtzMcWn0/Star-Wars-XXX-Parody/',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'http://www.eporner.com/hd-porn/3YRUtzMcWn0',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'http://www.eporner.com/embed/3YRUtzMcWn0',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.eporner.com/video-FJsA19J3Y3H/one-of-the-greats/',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
mobj = self._match_valid_url(url)
|
|
video_id = mobj.group('id')
|
|
display_id = mobj.group('display_id') or video_id
|
|
|
|
webpage, urlh = self._download_webpage_handle(url, display_id)
|
|
|
|
video_id = self._match_id(urlh.url)
|
|
|
|
hash = self._search_regex(
|
|
r'hash\s*[:=]\s*["\']([\da-f]{32})', webpage, 'hash')
|
|
|
|
title = self._og_search_title(webpage, default=None) or self._html_search_regex(
|
|
r'<title>(.+?) - EPORNER', webpage, 'title')
|
|
|
|
# Reverse engineered from vjs.js
|
|
def calc_hash(s):
|
|
return ''.join((encode_base_n(int(s[lb:lb + 8], 16), 36) for lb in range(0, 32, 8)))
|
|
|
|
video = self._download_json(
|
|
'http://www.eporner.com/xhr/video/%s' % video_id,
|
|
display_id, note='Downloading video JSON',
|
|
query={
|
|
'hash': calc_hash(hash),
|
|
'device': 'generic',
|
|
'domain': 'www.eporner.com',
|
|
'fallback': 'false',
|
|
})
|
|
|
|
if video.get('available') is False:
|
|
raise ExtractorError(
|
|
'%s said: %s' % (self.IE_NAME, video['message']), expected=True)
|
|
|
|
sources = video['sources']
|
|
|
|
formats = []
|
|
has_av1 = bool(get_elements_by_class('download-av1', webpage))
|
|
for kind, formats_dict in sources.items():
|
|
if not isinstance(formats_dict, dict):
|
|
continue
|
|
for format_id, format_dict in formats_dict.items():
|
|
if not isinstance(format_dict, dict):
|
|
continue
|
|
src = url_or_none(format_dict.get('src'))
|
|
if not src or not src.startswith('http'):
|
|
continue
|
|
if kind == 'hls':
|
|
formats.extend(self._extract_m3u8_formats(
|
|
src, display_id, 'mp4', entry_protocol='m3u8_native',
|
|
m3u8_id=kind, fatal=False))
|
|
else:
|
|
height = int_or_none(self._search_regex(
|
|
r'(\d+)[pP]', format_id, 'height', default=None))
|
|
fps = int_or_none(self._search_regex(
|
|
r'(\d+)fps', format_id, 'fps', default=None))
|
|
|
|
formats.append({
|
|
'url': src,
|
|
'format_id': format_id,
|
|
'height': height,
|
|
'fps': fps,
|
|
})
|
|
if has_av1:
|
|
formats.append({
|
|
'url': src.replace('.mp4', '-av1.mp4'),
|
|
'format_id': join_nonempty('av1', format_id),
|
|
'height': height,
|
|
'fps': fps,
|
|
'vcodec': 'av1',
|
|
})
|
|
|
|
json_ld = self._search_json_ld(webpage, display_id, default={})
|
|
|
|
duration = parse_duration(self._html_search_meta(
|
|
'duration', webpage, default=None))
|
|
view_count = str_to_int(self._search_regex(
|
|
r'id=["\']cinemaviews1["\'][^>]*>\s*([0-9,]+)',
|
|
webpage, 'view count', default=None))
|
|
|
|
return merge_dicts(json_ld, {
|
|
'id': video_id,
|
|
'display_id': display_id,
|
|
'title': title,
|
|
'duration': duration,
|
|
'view_count': view_count,
|
|
'formats': formats,
|
|
'age_limit': 18,
|
|
})
|