X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/00368b4c3a5d4e909e1b7ecfc4030bf28da020f3..9113dfef91df19343cf76c6274dd0a85258c1004:/youtube_dl/extractor/xhamster.py
diff --git a/youtube_dl/extractor/xhamster.py b/youtube_dl/extractor/xhamster.py
index c42b59e..68652a2 100644
--- a/youtube_dl/extractor/xhamster.py
+++ b/youtube_dl/extractor/xhamster.py
@@ -6,10 +6,12 @@ from .common import InfoExtractor
from ..compat import compat_str
from ..utils import (
clean_html,
+ determine_ext,
dict_get,
ExtractorError,
int_or_none,
parse_duration,
+ try_get,
unified_strdate,
)
@@ -32,6 +34,7 @@ class XHamsterIE(InfoExtractor):
'display_id': 'femaleagent_shy_beauty_takes_the_bait',
'ext': 'mp4',
'title': 'FemaleAgent Shy beauty takes the bait',
+ 'timestamp': 1350194821,
'upload_date': '20121014',
'uploader': 'Ruseful2011',
'duration': 893,
@@ -45,6 +48,7 @@ class XHamsterIE(InfoExtractor):
'display_id': 'britney_spears_sexy_booty',
'ext': 'mp4',
'title': 'Britney Spears Sexy Booty',
+ 'timestamp': 1379123460,
'upload_date': '20130914',
'uploader': 'jojo747400',
'duration': 200,
@@ -61,6 +65,7 @@ class XHamsterIE(InfoExtractor):
'id': '5667973',
'ext': 'mp4',
'title': '....',
+ 'timestamp': 1454948101,
'upload_date': '20160208',
'uploader': 'parejafree',
'duration': 72,
@@ -70,6 +75,10 @@ class XHamsterIE(InfoExtractor):
'params': {
'skip_download': True,
},
+ }, {
+ # mobile site
+ 'url': 'https://m.xhamster.com/videos/cute-teen-jacqueline-solo-masturbation-8559111',
+ 'only_matching': True,
}, {
'url': 'https://xhamster.com/movies/2272726/amber_slayed_by_the_knight.html',
'only_matching': True,
@@ -88,7 +97,8 @@ class XHamsterIE(InfoExtractor):
video_id = mobj.group('id') or mobj.group('id_2')
display_id = mobj.group('display_id') or mobj.group('display_id_2')
- webpage = self._download_webpage(url, video_id)
+ desktop_url = re.sub(r'^(https?://(?:.+?\.)?)m\.', r'\1', url)
+ webpage = self._download_webpage(desktop_url, video_id)
error = self._html_search_regex(
r'
]+id=["\']videoClosed["\'][^>]*>(.+?)
',
@@ -96,6 +106,83 @@ class XHamsterIE(InfoExtractor):
if error:
raise ExtractorError(error, expected=True)
+ age_limit = self._rta_search(webpage)
+
+ def get_height(s):
+ return int_or_none(self._search_regex(
+ r'^(\d+)[pP]', s, 'height', default=None))
+
+ initials = self._parse_json(
+ self._search_regex(
+ r'window\.initials\s*=\s*({.+?})\s*;\s*\n', webpage, 'initials',
+ default='{}'),
+ video_id, fatal=False)
+ if initials:
+ video = initials['videoModel']
+ title = video['title']
+ formats = []
+ for format_id, formats_dict in video['sources'].items():
+ if not isinstance(formats_dict, dict):
+ continue
+ for quality, format_item in formats_dict.items():
+ if format_id == 'download':
+ # Download link takes some time to be generated,
+ # skipping for now
+ continue
+ if not isinstance(format_item, dict):
+ continue
+ format_url = format_item.get('link')
+ filesize = int_or_none(
+ format_item.get('size'), invscale=1000000)
+ else:
+ format_url = format_item
+ filesize = None
+ if not isinstance(format_url, compat_str):
+ continue
+ formats.append({
+ 'format_id': '%s-%s' % (format_id, quality),
+ 'url': format_url,
+ 'ext': determine_ext(format_url, 'mp4'),
+ 'height': get_height(quality),
+ 'filesize': filesize,
+ })
+ self._sort_formats(formats)
+
+ categories_list = video.get('categories')
+ if isinstance(categories_list, list):
+ categories = []
+ for c in categories_list:
+ if not isinstance(c, dict):
+ continue
+ c_name = c.get('name')
+ if isinstance(c_name, compat_str):
+ categories.append(c_name)
+ else:
+ categories = None
+
+ return {
+ 'id': video_id,
+ 'display_id': display_id,
+ 'title': title,
+ 'description': video.get('description'),
+ 'timestamp': int_or_none(video.get('created')),
+ 'uploader': try_get(
+ video, lambda x: x['author']['name'], compat_str),
+ 'thumbnail': video.get('thumbURL'),
+ 'duration': int_or_none(video.get('duration')),
+ 'view_count': int_or_none(video.get('views')),
+ 'like_count': int_or_none(try_get(
+ video, lambda x: x['rating']['likes'], int)),
+ 'dislike_count': int_or_none(try_get(
+ video, lambda x: x['rating']['dislikes'], int)),
+ 'comment_count': int_or_none(video.get('views')),
+ 'age_limit': age_limit,
+ 'categories': categories,
+ 'formats': formats,
+ }
+
+ # Old layout fallback
+
title = self._html_search_regex(
[r']*>([^<]+)
',
r']+itemprop=".*?caption.*?"[^>]+content="(.+?)"',
@@ -119,8 +206,7 @@ class XHamsterIE(InfoExtractor):
formats.append({
'format_id': format_id,
'url': format_url,
- 'height': int_or_none(self._search_regex(
- r'^(\d+)[pP]', format_id, 'height', default=None))
+ 'height': get_height(format_id),
})
video_url = self._search_regex(
@@ -148,8 +234,8 @@ class XHamsterIE(InfoExtractor):
webpage, 'uploader', default='anonymous')
thumbnail = self._search_regex(
- [r'''thumb\s*:\s*(?P["'])(?P.+?)(?P=q)''',
- r'''