]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/crackle.py
d/rules: Run offline tests.
[youtubedl] / youtube_dl / extractor / crackle.py
index 79238cce7a22d040e70c977219a86438bea4dfc7..49bf3a4f9ef0ecb239f9916367e2efff0af16b61 100644 (file)
 # coding: utf-8
-from __future__ import unicode_literals
+from __future__ import unicode_literals, division
+
+import hashlib
+import hmac
+import re
+import time
 
 from .common import InfoExtractor
-from ..utils import int_or_none
+from ..compat import compat_HTTPError
+from ..utils import (
+    determine_ext,
+    float_or_none,
+    int_or_none,
+    parse_age_limit,
+    parse_duration,
+    url_or_none,
+    ExtractorError
+)
 
 
 class CrackleIE(InfoExtractor):
-    _VALID_URL = r'(?:crackle:|https?://(?:www\.)?crackle\.com/(?:playlist/\d+/|(?:[^/]+/)+))(?P<id>\d+)'
-    _TEST = {
-        'url': 'http://www.crackle.com/the-art-of-more/2496419',
+    _VALID_URL = r'(?:crackle:|https?://(?:(?:www|m)\.)?(?:sony)?crackle\.com/(?:playlist/\d+/|(?:[^/]+/)+))(?P<id>\d+)'
+    _TESTS = [{
+        # geo restricted to CA
+        'url': 'https://www.crackle.com/andromeda/2502343',
         'info_dict': {
-            'id': '2496419',
+            'id': '2502343',
             'ext': 'mp4',
-            'title': 'Heavy Lies the Head',
-            'description': 'md5:bb56aa0708fe7b9a4861535f15c3abca',
+            'title': 'Under The Night',
+            'description': 'md5:d2b8ca816579ae8a7bf28bfff8cefc8a',
+            'duration': 2583,
+            'view_count': int,
+            'average_rating': 0,
+            'age_limit': 14,
+            'genre': 'Action, Sci-Fi',
+            'creator': 'Allan Kroeker',
+            'artist': 'Keith Hamilton Cobb, Kevin Sorbo, Lisa Ryder, Lexa Doig, Robert Hewitt Wolfe',
+            'release_year': 2000,
+            'series': 'Andromeda',
+            'episode': 'Under The Night',
+            'season_number': 1,
+            'episode_number': 1,
         },
         'params': {
             # m3u8 download
             'skip_download': True,
         }
-    }
-
-    # extracted from http://legacyweb-us.crackle.com/flash/QueryReferrer.ashx
-    _SUBTITLE_SERVER = 'http://web-us-az.crackle.com'
-    _UPLYNK_OWNER_ID = 'e8773f7770a44dbd886eee4fca16a66b'
-    _THUMBNAIL_TEMPLATE = 'http://images-us-am.crackle.com/%stnl_1920x1080.jpg?ts=20140107233116?c=635333335057637614'
+    }, {
+        'url': 'https://www.sonycrackle.com/andromeda/2502343',
+        'only_matching': True,
+    }]
 
-    # extracted from http://legacyweb-us.crackle.com/flash/ReferrerRedirect.ashx
     _MEDIA_FILE_SLOTS = {
-        'c544.flv': {
-            'width': 544,
-            'height': 306,
-        },
         '360p.mp4': {
             'width': 640,
             'height': 360,
         },
         '480p.mp4': {
-            'width': 852,
-            'height': 478,
+            'width': 768,
+            'height': 432,
         },
         '480p_1mbps.mp4': {
             'width': 852,
-            'height': 478,
+            'height': 480,
         },
     }
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
-        item = self._download_xml(
-            'http://legacyweb-us.crackle.com/app/revamp/vidwallcache.aspx?flags=-1&fm=%s' % video_id,
-            video_id).find('i')
-        title = item.attrib['t']
-
-        thumbnail = None
-        subtitles = {}
-        formats = self._extract_m3u8_formats(
-            'http://content.uplynk.com/ext/%s/%s.m3u8' % (self._UPLYNK_OWNER_ID, video_id),
-            video_id, 'mp4', m3u8_id='hls', fatal=None)
-        path = item.attrib.get('p')
-        if path:
-            thumbnail = self._THUMBNAIL_TEMPLATE % path
-            http_base_url = 'http://ahttp.crackle.com/' + path
-            for mfs_path, mfs_info in self._MEDIA_FILE_SLOTS.items():
-                formats.append({
-                    'url': http_base_url + mfs_path,
-                    'format_id': 'http-' + mfs_path.split('.')[0],
-                    'width': mfs_info['width'],
-                    'height': mfs_info['height'],
-                })
-            for cc in item.findall('cc'):
-                locale = cc.attrib.get('l')
-                v = cc.attrib.get('v')
-                if locale and v:
-                    if locale not in subtitles:
-                        subtitles[locale] = []
-                    subtitles[locale] = [{
-                        'url': '%s/%s%s_%s.xml' % (self._SUBTITLE_SERVER, path, locale, v),
-                        'ext': 'ttml',
-                    }]
-        self._sort_formats(formats, ('width', 'height', 'tbr', 'format_id'))
-
-        return {
-            'id': video_id,
-            'title': title,
-            'description': item.attrib.get('d'),
-            'duration': int(item.attrib.get('r'), 16) if item.attrib.get('r') else None,
-            'series': item.attrib.get('sn'),
-            'season_number': int_or_none(item.attrib.get('se')),
-            'episode_number': int_or_none(item.attrib.get('ep')),
-            'thumbnail': thumbnail,
-            'subtitles': subtitles,
-            'formats': formats,
-        }
+
+        country_code = self._downloader.params.get('geo_bypass_country', None)
+        countries = [country_code] if country_code else (
+            'US', 'AU', 'CA', 'AS', 'FM', 'GU', 'MP', 'PR', 'PW', 'MH', 'VI')
+
+        last_e = None
+
+        for country in countries:
+            try:
+                # Authorization generation algorithm is reverse engineered from:
+                # https://www.sonycrackle.com/static/js/main.ea93451f.chunk.js
+                media_detail_url = 'https://web-api-us.crackle.com/Service.svc/details/media/%s/%s?disableProtocols=true' % (video_id, country)
+                timestamp = time.strftime('%Y%m%d%H%M', time.gmtime())
+                h = hmac.new(b'IGSLUQCBDFHEOIFM', '|'.join([media_detail_url, timestamp]).encode(), hashlib.sha1).hexdigest().upper()
+                media = self._download_json(
+                    media_detail_url, video_id, 'Downloading media JSON as %s' % country,
+                    'Unable to download media JSON', headers={
+                        'Accept': 'application/json',
+                        'Authorization': '|'.join([h, timestamp, '117', '1']),
+                    })
+            except ExtractorError as e:
+                # 401 means geo restriction, trying next country
+                if isinstance(e.cause, compat_HTTPError) and e.cause.code == 401:
+                    last_e = e
+                    continue
+                raise
+
+            media_urls = media.get('MediaURLs')
+            if not media_urls or not isinstance(media_urls, list):
+                continue
+
+            title = media['Title']
+
+            formats = []
+            for e in media['MediaURLs']:
+                if e.get('UseDRM') is True:
+                    continue
+                format_url = url_or_none(e.get('Path'))
+                if not format_url:
+                    continue
+                ext = determine_ext(format_url)
+                if ext == 'm3u8':
+                    formats.extend(self._extract_m3u8_formats(
+                        format_url, video_id, 'mp4', entry_protocol='m3u8_native',
+                        m3u8_id='hls', fatal=False))
+                elif ext == 'mpd':
+                    formats.extend(self._extract_mpd_formats(
+                        format_url, video_id, mpd_id='dash', fatal=False))
+                elif format_url.endswith('.ism/Manifest'):
+                    formats.extend(self._extract_ism_formats(
+                        format_url, video_id, ism_id='mss', fatal=False))
+                else:
+                    mfs_path = e.get('Type')
+                    mfs_info = self._MEDIA_FILE_SLOTS.get(mfs_path)
+                    if not mfs_info:
+                        continue
+                    formats.append({
+                        'url': format_url,
+                        'format_id': 'http-' + mfs_path.split('.')[0],
+                        'width': mfs_info['width'],
+                        'height': mfs_info['height'],
+                    })
+            self._sort_formats(formats)
+
+            description = media.get('Description')
+            duration = int_or_none(media.get(
+                'DurationInSeconds')) or parse_duration(media.get('Duration'))
+            view_count = int_or_none(media.get('CountViews'))
+            average_rating = float_or_none(media.get('UserRating'))
+            age_limit = parse_age_limit(media.get('Rating'))
+            genre = media.get('Genre')
+            release_year = int_or_none(media.get('ReleaseYear'))
+            creator = media.get('Directors')
+            artist = media.get('Cast')
+
+            if media.get('MediaTypeDisplayValue') == 'Full Episode':
+                series = media.get('ShowName')
+                episode = title
+                season_number = int_or_none(media.get('Season'))
+                episode_number = int_or_none(media.get('Episode'))
+            else:
+                series = episode = season_number = episode_number = None
+
+            subtitles = {}
+            cc_files = media.get('ClosedCaptionFiles')
+            if isinstance(cc_files, list):
+                for cc_file in cc_files:
+                    if not isinstance(cc_file, dict):
+                        continue
+                    cc_url = url_or_none(cc_file.get('Path'))
+                    if not cc_url:
+                        continue
+                    lang = cc_file.get('Locale') or 'en'
+                    subtitles.setdefault(lang, []).append({'url': cc_url})
+
+            thumbnails = []
+            images = media.get('Images')
+            if isinstance(images, list):
+                for image_key, image_url in images.items():
+                    mobj = re.search(r'Img_(\d+)[xX](\d+)', image_key)
+                    if not mobj:
+                        continue
+                    thumbnails.append({
+                        'url': image_url,
+                        'width': int(mobj.group(1)),
+                        'height': int(mobj.group(2)),
+                    })
+
+            return {
+                'id': video_id,
+                'title': title,
+                'description': description,
+                'duration': duration,
+                'view_count': view_count,
+                'average_rating': average_rating,
+                'age_limit': age_limit,
+                'genre': genre,
+                'creator': creator,
+                'artist': artist,
+                'release_year': release_year,
+                'series': series,
+                'episode': episode,
+                'season_number': season_number,
+                'episode_number': episode_number,
+                'thumbnails': thumbnails,
+                'subtitles': subtitles,
+                'formats': formats,
+            }
+
+        raise last_e