mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-23 19:33:59 +01:00
136 lines
4.9 KiB
Python
136 lines
4.9 KiB
Python
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
ExtractorError,
|
|
unified_strdate,
|
|
str_to_int,
|
|
int_or_none,
|
|
parse_duration,
|
|
)
|
|
|
|
|
|
class XHamsterIE(InfoExtractor):
|
|
"""Information Extractor for xHamster"""
|
|
_VALID_URL = r'(?P<proto>https?)://(?:.+?\.)?xhamster\.com/movies/(?P<id>[0-9]+)/(?P<seo>.+?)\.html(?:\?.*)?'
|
|
_TESTS = [
|
|
{
|
|
'url': 'http://xhamster.com/movies/1509445/femaleagent_shy_beauty_takes_the_bait.html',
|
|
'info_dict': {
|
|
'id': '1509445',
|
|
'ext': 'mp4',
|
|
'title': 'FemaleAgent Shy beauty takes the bait',
|
|
'upload_date': '20121014',
|
|
'uploader_id': 'Ruseful2011',
|
|
'duration': 893,
|
|
'age_limit': 18,
|
|
}
|
|
},
|
|
{
|
|
'url': 'http://xhamster.com/movies/2221348/britney_spears_sexy_booty.html?hd',
|
|
'info_dict': {
|
|
'id': '2221348',
|
|
'ext': 'mp4',
|
|
'title': 'Britney Spears Sexy Booty',
|
|
'upload_date': '20130914',
|
|
'uploader_id': 'jojo747400',
|
|
'duration': 200,
|
|
'age_limit': 18,
|
|
}
|
|
},
|
|
{
|
|
'url': 'https://xhamster.com/movies/2272726/amber_slayed_by_the_knight.html',
|
|
'only_matching': True,
|
|
},
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
def extract_video_url(webpage):
|
|
mp4 = re.search(r'<video\s+.*?file="([^"]+)".*?>', webpage)
|
|
if mp4 is None:
|
|
raise ExtractorError('Unable to extract media URL')
|
|
else:
|
|
return mp4.group(1)
|
|
|
|
def is_hd(webpage):
|
|
return '<div class=\'icon iconHD\'' in webpage
|
|
|
|
mobj = re.match(self._VALID_URL, url)
|
|
|
|
video_id = mobj.group('id')
|
|
seo = mobj.group('seo')
|
|
proto = mobj.group('proto')
|
|
mrss_url = '%s://xhamster.com/movies/%s/%s.html' % (proto, video_id, seo)
|
|
webpage = self._download_webpage(mrss_url, video_id)
|
|
|
|
title = self._html_search_regex(r'<title>(?P<title>.+?) - xHamster\.com</title>', webpage, 'title')
|
|
|
|
# Only a few videos have an description
|
|
mobj = re.search(r'<span>Description: </span>([^<]+)', webpage)
|
|
description = mobj.group(1) if mobj else None
|
|
|
|
upload_date = self._html_search_regex(r'hint=\'(\d{4}-\d{2}-\d{2}) \d{2}:\d{2}:\d{2} [A-Z]{3,4}\'',
|
|
webpage, 'upload date', fatal=False)
|
|
if upload_date:
|
|
upload_date = unified_strdate(upload_date)
|
|
|
|
uploader_id = self._html_search_regex(r'<a href=\'/user/[^>]+>(?P<uploader_id>[^<]+)',
|
|
webpage, 'uploader id', default='anonymous')
|
|
|
|
thumbnail = self._html_search_regex(r'<video\s+.*?poster="([^"]+)".*?>', webpage, 'thumbnail', fatal=False)
|
|
|
|
duration = parse_duration(self._html_search_regex(r'<span>Runtime:</span> (\d+:\d+)</div>',
|
|
webpage, 'duration', fatal=False))
|
|
|
|
view_count = self._html_search_regex(r'<span>Views:</span> ([^<]+)</div>', webpage, 'view count', fatal=False)
|
|
if view_count:
|
|
view_count = str_to_int(view_count)
|
|
|
|
mobj = re.search(r"hint='(?P<likecount>\d+) Likes / (?P<dislikecount>\d+) Dislikes'", webpage)
|
|
(like_count, dislike_count) = (mobj.group('likecount'), mobj.group('dislikecount')) if mobj else (None, None)
|
|
|
|
mobj = re.search(r'</label>Comments \((?P<commentcount>\d+)\)</div>', webpage)
|
|
comment_count = mobj.group('commentcount') if mobj else 0
|
|
|
|
age_limit = self._rta_search(webpage)
|
|
|
|
hd = is_hd(webpage)
|
|
|
|
video_url = extract_video_url(webpage)
|
|
formats = [{
|
|
'url': video_url,
|
|
'format_id': 'hd' if hd else 'sd',
|
|
'preference': 1,
|
|
}]
|
|
|
|
if not hd:
|
|
mrss_url = self._search_regex(r'<link rel="canonical" href="([^"]+)', webpage, 'mrss_url')
|
|
webpage = self._download_webpage(mrss_url + '?hd', video_id, note='Downloading HD webpage')
|
|
if is_hd(webpage):
|
|
video_url = extract_video_url(webpage)
|
|
formats.append({
|
|
'url': video_url,
|
|
'format_id': 'hd',
|
|
'preference': 2,
|
|
})
|
|
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'description': description,
|
|
'upload_date': upload_date,
|
|
'uploader_id': uploader_id,
|
|
'thumbnail': thumbnail,
|
|
'duration': duration,
|
|
'view_count': view_count,
|
|
'like_count': int_or_none(like_count),
|
|
'dislike_count': int_or_none(dislike_count),
|
|
'comment_count': int_or_none(comment_count),
|
|
'age_limit': age_limit,
|
|
'formats': formats,
|
|
}
|