]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/xhamster.py
Start new release.
[youtubedl] / youtube_dl / extractor / xhamster.py
index be3624ef2ea889835992e55949271c0028e9783b..68652a22fc7453d01802334b3af021557cc1cedd 100644 (file)
@@ -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'<div[^>]+id=["\']videoClosed["\'][^>]*>(.+?)</div>',
@@ -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'<h1[^>]*>([^<]+)</h1>',
              r'<meta[^>]+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<q>["'])(?P<thumbnail>.+?)(?P=q)''',
-             r'''<video[^>]+poster=(?P<q>["'])(?P<thumbnail>.+?)(?P=q)[^>]*>'''],
+            [r'''["']thumbUrl["']\s*:\s*(?P<q>["'])(?P<thumbnail>.+?)(?P=q)''',
+             r'''<video[^>]+"poster"=(?P<q>["'])(?P<thumbnail>.+?)(?P=q)[^>]*>'''],
             webpage, 'thumbnail', fatal=False, group='thumbnail')
 
         duration = parse_duration(self._search_regex(
@@ -167,8 +253,6 @@ class XHamsterIE(InfoExtractor):
         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)
-
         categories_html = self._search_regex(
             r'(?s)<table.+?(<span>Categories:.+?)</table>', webpage,
             'categories', default=None)
@@ -195,15 +279,16 @@ class XHamsterIE(InfoExtractor):
 
 
 class XHamsterEmbedIE(InfoExtractor):
-    _VALID_URL = r'https?://(?:www\.)?xhamster\.com/xembed\.php\?video=(?P<id>\d+)'
+    _VALID_URL = r'https?://(?:.+?\.)?xhamster\.com/xembed\.php\?video=(?P<id>\d+)'
     _TEST = {
         'url': 'http://xhamster.com/xembed.php?video=3328539',
         'info_dict': {
             'id': '3328539',
             'ext': 'mp4',
             'title': 'Pen Masturbation',
+            'timestamp': 1406581861,
             'upload_date': '20140728',
-            'uploader_id': 'anonymous',
+            'uploader': 'ManyakisArt',
             'duration': 5,
             'age_limit': 18,
         }