]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/crunchyroll.py
d/control: Build-depend on flake8 and python3-nose.
[youtubedl] / youtube_dl / extractor / crunchyroll.py
index d15fd3744d5a03e383cb36d0dd9e990d061aa6cb..85a9a577f645395d6edde83e7b72a1f4001561f8 100644 (file)
@@ -3,13 +3,15 @@ from __future__ import unicode_literals
 
 import re
 import json
 
 import re
 import json
-import base64
 import zlib
 
 from hashlib import sha1
 from math import pow, sqrt, floor
 from .common import InfoExtractor
 import zlib
 
 from hashlib import sha1
 from math import pow, sqrt, floor
 from .common import InfoExtractor
+from .vrv import VRVIE
 from ..compat import (
 from ..compat import (
+    compat_b64decode,
+    compat_etree_Element,
     compat_etree_fromstring,
     compat_urllib_parse_urlencode,
     compat_urllib_request,
     compat_etree_fromstring,
     compat_urllib_parse_urlencode,
     compat_urllib_request,
@@ -18,6 +20,8 @@ from ..compat import (
 from ..utils import (
     ExtractorError,
     bytes_to_intlist,
 from ..utils import (
     ExtractorError,
     bytes_to_intlist,
+    extract_attributes,
+    float_or_none,
     intlist_to_bytes,
     int_or_none,
     lowercase_escape,
     intlist_to_bytes,
     int_or_none,
     lowercase_escape,
@@ -26,7 +30,6 @@ from ..utils import (
     unified_strdate,
     urlencode_postdata,
     xpath_text,
     unified_strdate,
     urlencode_postdata,
     xpath_text,
-    extract_attributes,
 )
 from ..aes import (
     aes_cbc_decrypt,
 )
 from ..aes import (
     aes_cbc_decrypt,
@@ -38,8 +41,18 @@ class CrunchyrollBaseIE(InfoExtractor):
     _LOGIN_FORM = 'login_form'
     _NETRC_MACHINE = 'crunchyroll'
 
     _LOGIN_FORM = 'login_form'
     _NETRC_MACHINE = 'crunchyroll'
 
+    def _call_rpc_api(self, method, video_id, note=None, data=None):
+        data = data or {}
+        data['req'] = 'RpcApi' + method
+        data = compat_urllib_parse_urlencode(data).encode('utf-8')
+        return self._download_xml(
+            'https://www.crunchyroll.com/xml/',
+            video_id, note, fatal=False, data=data, headers={
+                'Content-Type': 'application/x-www-form-urlencoded',
+            })
+
     def _login(self):
     def _login(self):
-        (username, password) = self._get_login_info()
+        username, password = self._get_login_info()
         if username is None:
             return
 
         if username is None:
             return
 
@@ -47,7 +60,7 @@ class CrunchyrollBaseIE(InfoExtractor):
             self._LOGIN_URL, None, 'Downloading login page')
 
         def is_logged(webpage):
             self._LOGIN_URL, None, 'Downloading login page')
 
         def is_logged(webpage):
-            return '<title>Redirecting' in webpage
+            return 'href="/logout"' in webpage
 
         # Already logged in
         if is_logged(login_page):
 
         # Already logged in
         if is_logged(login_page):
@@ -90,19 +103,6 @@ class CrunchyrollBaseIE(InfoExtractor):
     def _real_initialize(self):
         self._login()
 
     def _real_initialize(self):
         self._login()
 
-    def _download_webpage(self, url_or_request, *args, **kwargs):
-        request = (url_or_request if isinstance(url_or_request, compat_urllib_request.Request)
-                   else sanitized_Request(url_or_request))
-        # Accept-Language must be set explicitly to accept any language to avoid issues
-        # similar to https://github.com/rg3/youtube-dl/issues/6797.
-        # Along with IP address Crunchyroll uses Accept-Language to guess whether georestriction
-        # should be imposed or not (from what I can see it just takes the first language
-        # ignoring the priority and requires it to correspond the IP). By the way this causes
-        # Crunchyroll to not work in georestriction cases in some browsers that don't place
-        # the locale lang first in header. However allowing any language seems to workaround the issue.
-        request.add_header('Accept-Language', '*')
-        return super(CrunchyrollBaseIE, self)._download_webpage(request, *args, **kwargs)
-
     @staticmethod
     def _add_skip_wall(url):
         parsed_url = compat_urlparse.urlparse(url)
     @staticmethod
     def _add_skip_wall(url):
         parsed_url = compat_urlparse.urlparse(url)
@@ -111,14 +111,15 @@ class CrunchyrollBaseIE(InfoExtractor):
         # > This content may be inappropriate for some people.
         # > Are you sure you want to continue?
         # since it's not disabled by default in crunchyroll account's settings.
         # > This content may be inappropriate for some people.
         # > Are you sure you want to continue?
         # since it's not disabled by default in crunchyroll account's settings.
-        # See https://github.com/rg3/youtube-dl/issues/7202.
+        # See https://github.com/ytdl-org/youtube-dl/issues/7202.
         qs['skip_wall'] = ['1']
         return compat_urlparse.urlunparse(
             parsed_url._replace(query=compat_urllib_parse_urlencode(qs, True)))
 
 
         qs['skip_wall'] = ['1']
         return compat_urlparse.urlunparse(
             parsed_url._replace(query=compat_urllib_parse_urlencode(qs, True)))
 
 
-class CrunchyrollIE(CrunchyrollBaseIE):
-    _VALID_URL = r'https?://(?:(?P<prefix>www|m)\.)?(?P<url>crunchyroll\.(?:com|fr)/(?:media(?:-|/\?id=)|[^/]*/[^/?&]*?)(?P<video_id>[0-9]+))(?:[/?&]|$)'
+class CrunchyrollIE(CrunchyrollBaseIE, VRVIE):
+    IE_NAME = 'crunchyroll'
+    _VALID_URL = r'https?://(?:(?P<prefix>www|m)\.)?(?P<url>crunchyroll\.(?:com|fr)/(?:media(?:-|/\?id=)|(?:[^/]*/){1,2}[^/?&]*?)(?P<video_id>[0-9]+))(?:[/?&]|$)'
     _TESTS = [{
         'url': 'http://www.crunchyroll.com/wanna-be-the-strongest-in-the-world/episode-1-an-idol-wrestler-is-born-645513',
         'info_dict': {
     _TESTS = [{
         'url': 'http://www.crunchyroll.com/wanna-be-the-strongest-in-the-world/episode-1-an-idol-wrestler-is-born-645513',
         'info_dict': {
@@ -126,7 +127,7 @@ class CrunchyrollIE(CrunchyrollBaseIE):
             'ext': 'mp4',
             'title': 'Wanna be the Strongest in the World Episode 1 – An Idol-Wrestler is Born!',
             'description': 'md5:2d17137920c64f2f49981a7797d275ef',
             'ext': 'mp4',
             'title': 'Wanna be the Strongest in the World Episode 1 – An Idol-Wrestler is Born!',
             'description': 'md5:2d17137920c64f2f49981a7797d275ef',
-            'thumbnail': 'http://img1.ak.crunchyroll.com/i/spire1-tmb/20c6b5e10f1a47b10516877d3c039cae1380951166_full.jpg',
+            'thumbnail': r're:^https?://.*\.jpg$',
             'uploader': 'Yomiuri Telecasting Corporation (YTV)',
             'upload_date': '20131013',
             'url': 're:(?!.*&amp)',
             'uploader': 'Yomiuri Telecasting Corporation (YTV)',
             'upload_date': '20131013',
             'url': 're:(?!.*&amp)',
@@ -171,7 +172,7 @@ class CrunchyrollIE(CrunchyrollBaseIE):
         'info_dict': {
             'id': '727589',
             'ext': 'mp4',
         'info_dict': {
             'id': '727589',
             'ext': 'mp4',
-            'title': "KONOSUBA -God's blessing on this wonderful world! 2 Episode 1 – Give Me Deliverance from this Judicial Injustice!",
+            'title': "KONOSUBA -God's blessing on this wonderful world! 2 Episode 1 – Give Me Deliverance From This Judicial Injustice!",
             'description': 'md5:cbcf05e528124b0f3a0a419fc805ea7d',
             'thumbnail': r're:^https?://.*\.jpg$',
             'uploader': 'Kadokawa Pictures Inc.',
             'description': 'md5:cbcf05e528124b0f3a0a419fc805ea7d',
             'thumbnail': r're:^https?://.*\.jpg$',
             'uploader': 'Kadokawa Pictures Inc.',
@@ -179,7 +180,7 @@ class CrunchyrollIE(CrunchyrollBaseIE):
             'series': "KONOSUBA -God's blessing on this wonderful world!",
             'season': "KONOSUBA -God's blessing on this wonderful world! 2",
             'season_number': 2,
             'series': "KONOSUBA -God's blessing on this wonderful world!",
             'season': "KONOSUBA -God's blessing on this wonderful world! 2",
             'season_number': 2,
-            'episode': 'Give Me Deliverance from this Judicial Injustice!',
+            'episode': 'Give Me Deliverance From This Judicial Injustice!',
             'episode_number': 1,
         },
         'params': {
             'episode_number': 1,
         },
         'params': {
@@ -199,7 +200,7 @@ class CrunchyrollIE(CrunchyrollBaseIE):
         'info_dict': {
             'id': '535080',
             'ext': 'mp4',
         'info_dict': {
             'id': '535080',
             'ext': 'mp4',
-            'title': '11eyes Episode 1 – Piros éjszaka - Red Night',
+            'title': '11eyes Episode 1 – Red Night ~ Piros éjszaka',
             'description': 'Kakeru and Yuka are thrown into an alternate nightmarish world they call "Red Night".',
             'uploader': 'Marvelous AQL Inc.',
             'upload_date': '20091021',
             'description': 'Kakeru and Yuka are thrown into an alternate nightmarish world they call "Red Night".',
             'uploader': 'Marvelous AQL Inc.',
             'upload_date': '20091021',
@@ -240,6 +241,12 @@ class CrunchyrollIE(CrunchyrollBaseIE):
             # Just test metadata extraction
             'skip_download': True,
         },
             # Just test metadata extraction
             'skip_download': True,
         },
+    }, {
+        'url': 'http://www.crunchyroll.com/media-723735',
+        'only_matching': True,
+    }, {
+        'url': 'https://www.crunchyroll.com/en-gb/mob-psycho-100/episode-2-urban-legends-encountering-rumors-780921',
+        'only_matching': True,
     }]
 
     _FORMAT_IDS = {
     }]
 
     _FORMAT_IDS = {
@@ -249,9 +256,22 @@ class CrunchyrollIE(CrunchyrollBaseIE):
         '1080': ('80', '108'),
     }
 
         '1080': ('80', '108'),
     }
 
+    def _download_webpage(self, url_or_request, *args, **kwargs):
+        request = (url_or_request if isinstance(url_or_request, compat_urllib_request.Request)
+                   else sanitized_Request(url_or_request))
+        # Accept-Language must be set explicitly to accept any language to avoid issues
+        # similar to https://github.com/ytdl-org/youtube-dl/issues/6797.
+        # Along with IP address Crunchyroll uses Accept-Language to guess whether georestriction
+        # should be imposed or not (from what I can see it just takes the first language
+        # ignoring the priority and requires it to correspond the IP). By the way this causes
+        # Crunchyroll to not work in georestriction cases in some browsers that don't place
+        # the locale lang first in header. However allowing any language seems to workaround the issue.
+        request.add_header('Accept-Language', '*')
+        return super(CrunchyrollBaseIE, self)._download_webpage(request, *args, **kwargs)
+
     def _decrypt_subtitles(self, data, iv, id):
     def _decrypt_subtitles(self, data, iv, id):
-        data = bytes_to_intlist(base64.b64decode(data.encode('utf-8')))
-        iv = bytes_to_intlist(base64.b64decode(iv.encode('utf-8')))
+        data = bytes_to_intlist(compat_b64decode(data))
+        iv = bytes_to_intlist(compat_b64decode(iv))
         id = int(id)
 
         def obfuscate_key_aux(count, modulo, start):
         id = int(id)
 
         def obfuscate_key_aux(count, modulo, start):
@@ -365,15 +385,19 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
     def _get_subtitles(self, video_id, webpage):
         subtitles = {}
         for sub_id, sub_name in re.findall(r'\bssid=([0-9]+)"[^>]+?\btitle="([^"]+)', webpage):
     def _get_subtitles(self, video_id, webpage):
         subtitles = {}
         for sub_id, sub_name in re.findall(r'\bssid=([0-9]+)"[^>]+?\btitle="([^"]+)', webpage):
-            sub_page = self._download_webpage(
-                'http://www.crunchyroll.com/xml/?req=RpcApiSubtitle_GetXml&subtitle_script_id=' + sub_id,
-                video_id, note='Downloading subtitles for ' + sub_name)
-            id = self._search_regex(r'id=\'([0-9]+)', sub_page, 'subtitle_id', fatal=False)
-            iv = self._search_regex(r'<iv>([^<]+)', sub_page, 'subtitle_iv', fatal=False)
-            data = self._search_regex(r'<data>([^<]+)', sub_page, 'subtitle_data', fatal=False)
-            if not id or not iv or not data:
+            sub_doc = self._call_rpc_api(
+                'Subtitle_GetXml', video_id,
+                'Downloading subtitles for ' + sub_name, data={
+                    'subtitle_script_id': sub_id,
+                })
+            if not isinstance(sub_doc, compat_etree_Element):
+                continue
+            sid = sub_doc.get('id')
+            iv = xpath_text(sub_doc, 'iv', 'subtitle iv')
+            data = xpath_text(sub_doc, 'data', 'subtitle data')
+            if not sid or not iv or not data:
                 continue
                 continue
-            subtitle = self._decrypt_subtitles(data, iv, id).decode('utf-8')
+            subtitle = self._decrypt_subtitles(data, iv, sid).decode('utf-8')
             lang_code = self._search_regex(r'lang_code=["\']([^"\']+)', subtitle, 'subtitle_lang_code', fatal=False)
             if not lang_code:
                 continue
             lang_code = self._search_regex(r'lang_code=["\']([^"\']+)', subtitle, 'subtitle_lang_code', fatal=False)
             if not lang_code:
                 continue
@@ -390,7 +414,9 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
         else:
             webpage_url = 'http://www.' + mobj.group('url')
 
         else:
             webpage_url = 'http://www.' + mobj.group('url')
 
-        webpage = self._download_webpage(self._add_skip_wall(webpage_url), video_id, 'Downloading webpage')
+        webpage = self._download_webpage(
+            self._add_skip_wall(webpage_url), video_id,
+            headers=self.geo_verification_headers())
         note_m = self._html_search_regex(
             r'<div class="showmedia-trailer-notice">(.+?)</div>',
             webpage, 'trailer-notice', default='')
         note_m = self._html_search_regex(
             r'<div class="showmedia-trailer-notice">(.+?)</div>',
             webpage, 'trailer-notice', default='')
@@ -406,13 +432,22 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
         if 'To view this, please log in to verify you are 18 or older.' in webpage:
             self.raise_login_required()
 
         if 'To view this, please log in to verify you are 18 or older.' in webpage:
             self.raise_login_required()
 
+        media = self._parse_json(self._search_regex(
+            r'vilos\.config\.media\s*=\s*({.+?});',
+            webpage, 'vilos media', default='{}'), video_id)
+        media_metadata = media.get('metadata') or {}
+
+        language = self._search_regex(
+            r'(?:vilos\.config\.player\.language|LOCALE)\s*=\s*(["\'])(?P<lang>(?:(?!\1).)+)\1',
+            webpage, 'language', default=None, group='lang')
+
         video_title = self._html_search_regex(
             r'(?s)<h1[^>]*>((?:(?!<h1).)*?<span[^>]+itemprop=["\']title["\'][^>]*>(?:(?!<h1).)+?)</h1>',
             webpage, 'video_title')
         video_title = re.sub(r' {2,}', ' ', video_title)
         video_title = self._html_search_regex(
             r'(?s)<h1[^>]*>((?:(?!<h1).)*?<span[^>]+itemprop=["\']title["\'][^>]*>(?:(?!<h1).)+?)</h1>',
             webpage, 'video_title')
         video_title = re.sub(r' {2,}', ' ', video_title)
-        video_description = self._parse_json(self._html_search_regex(
+        video_description = (self._parse_json(self._html_search_regex(
             r'<script[^>]*>\s*.+?\[media_id=%s\].+?({.+?"description"\s*:.+?})\);' % video_id,
             r'<script[^>]*>\s*.+?\[media_id=%s\].+?({.+?"description"\s*:.+?})\);' % video_id,
-            webpage, 'description', default='{}'), video_id).get('description')
+            webpage, 'description', default='{}'), video_id) or media_metadata).get('description')
         if video_description:
             video_description = lowercase_escape(video_description.replace(r'\r\n', '\n'))
         video_upload_date = self._html_search_regex(
         if video_description:
             video_description = lowercase_escape(video_description.replace(r'\r\n', '\n'))
         video_upload_date = self._html_search_regex(
@@ -425,105 +460,163 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
             [r'<a[^>]+href="/publisher/[^"]+"[^>]*>([^<]+)</a>', r'<div>\s*Publisher:\s*<span>\s*(.+?)\s*</span>\s*</div>'],
             webpage, 'video_uploader', fatal=False)
 
             [r'<a[^>]+href="/publisher/[^"]+"[^>]*>([^<]+)</a>', r'<div>\s*Publisher:\s*<span>\s*(.+?)\s*</span>\s*</div>'],
             webpage, 'video_uploader', fatal=False)
 
-        available_fmts = []
-        for a, fmt in re.findall(r'(<a[^>]+token=["\']showmedia\.([0-9]{3,4})p["\'][^>]+>)', webpage):
-            attrs = extract_attributes(a)
-            href = attrs.get('href')
-            if href and '/freetrial' in href:
-                continue
-            available_fmts.append(fmt)
-        if not available_fmts:
-            for p in (r'token=["\']showmedia\.([0-9]{3,4})p"', r'showmedia\.([0-9]{3,4})p'):
-                available_fmts = re.findall(p, webpage)
-                if available_fmts:
-                    break
-        video_encode_ids = []
         formats = []
         formats = []
-        for fmt in available_fmts:
-            stream_quality, stream_format = self._FORMAT_IDS[fmt]
-            video_format = fmt + 'p'
-            streamdata_req = sanitized_Request(
-                'http://www.crunchyroll.com/xml/?req=RpcApiVideoPlayer_GetStandardConfig&media_id=%s&video_format=%s&video_quality=%s'
-                % (video_id, stream_format, stream_quality),
-                compat_urllib_parse_urlencode({'current_page': url}).encode('utf-8'))
-            streamdata_req.add_header('Content-Type', 'application/x-www-form-urlencoded')
-            streamdata = self._download_xml(
-                streamdata_req, video_id,
-                note='Downloading media info for %s' % video_format)
-            stream_info = streamdata.find('./{default}preload/stream_info')
-            video_encode_id = xpath_text(stream_info, './video_encode_id')
-            if video_encode_id in video_encode_ids:
-                continue
-            video_encode_ids.append(video_encode_id)
-
-            video_file = xpath_text(stream_info, './file')
-            if not video_file:
-                continue
-            if video_file.startswith('http'):
-                formats.extend(self._extract_m3u8_formats(
-                    video_file, video_id, 'mp4', entry_protocol='m3u8_native',
-                    m3u8_id='hls', fatal=False))
-                continue
+        for stream in media.get('streams', []):
+            audio_lang = stream.get('audio_lang')
+            hardsub_lang = stream.get('hardsub_lang')
+            vrv_formats = self._extract_vrv_formats(
+                stream.get('url'), video_id, stream.get('format'),
+                audio_lang, hardsub_lang)
+            for f in vrv_formats:
+                if not hardsub_lang:
+                    f['preference'] = 1
+                language_preference = 0
+                if audio_lang == language:
+                    language_preference += 1
+                if hardsub_lang == language:
+                    language_preference += 1
+                if language_preference:
+                    f['language_preference'] = language_preference
+            formats.extend(vrv_formats)
+        if not formats:
+            available_fmts = []
+            for a, fmt in re.findall(r'(<a[^>]+token=["\']showmedia\.([0-9]{3,4})p["\'][^>]+>)', webpage):
+                attrs = extract_attributes(a)
+                href = attrs.get('href')
+                if href and '/freetrial' in href:
+                    continue
+                available_fmts.append(fmt)
+            if not available_fmts:
+                for p in (r'token=["\']showmedia\.([0-9]{3,4})p"', r'showmedia\.([0-9]{3,4})p'):
+                    available_fmts = re.findall(p, webpage)
+                    if available_fmts:
+                        break
+            if not available_fmts:
+                available_fmts = self._FORMAT_IDS.keys()
+            video_encode_ids = []
+
+            for fmt in available_fmts:
+                stream_quality, stream_format = self._FORMAT_IDS[fmt]
+                video_format = fmt + 'p'
+                stream_infos = []
+                streamdata = self._call_rpc_api(
+                    'VideoPlayer_GetStandardConfig', video_id,
+                    'Downloading media info for %s' % video_format, data={
+                        'media_id': video_id,
+                        'video_format': stream_format,
+                        'video_quality': stream_quality,
+                        'current_page': url,
+                    })
+                if isinstance(streamdata, compat_etree_Element):
+                    stream_info = streamdata.find('./{default}preload/stream_info')
+                    if stream_info is not None:
+                        stream_infos.append(stream_info)
+                stream_info = self._call_rpc_api(
+                    'VideoEncode_GetStreamInfo', video_id,
+                    'Downloading stream info for %s' % video_format, data={
+                        'media_id': video_id,
+                        'video_format': stream_format,
+                        'video_encode_quality': stream_quality,
+                    })
+                if isinstance(stream_info, compat_etree_Element):
+                    stream_infos.append(stream_info)
+                for stream_info in stream_infos:
+                    video_encode_id = xpath_text(stream_info, './video_encode_id')
+                    if video_encode_id in video_encode_ids:
+                        continue
+                    video_encode_ids.append(video_encode_id)
+
+                    video_file = xpath_text(stream_info, './file')
+                    if not video_file:
+                        continue
+                    if video_file.startswith('http'):
+                        formats.extend(self._extract_m3u8_formats(
+                            video_file, video_id, 'mp4', entry_protocol='m3u8_native',
+                            m3u8_id='hls', fatal=False))
+                        continue
+
+                    video_url = xpath_text(stream_info, './host')
+                    if not video_url:
+                        continue
+                    metadata = stream_info.find('./metadata')
+                    format_info = {
+                        'format': video_format,
+                        'height': int_or_none(xpath_text(metadata, './height')),
+                        'width': int_or_none(xpath_text(metadata, './width')),
+                    }
+
+                    if '.fplive.net/' in video_url:
+                        video_url = re.sub(r'^rtmpe?://', 'http://', video_url.strip())
+                        parsed_video_url = compat_urlparse.urlparse(video_url)
+                        direct_video_url = compat_urlparse.urlunparse(parsed_video_url._replace(
+                            netloc='v.lvlt.crcdn.net',
+                            path='%s/%s' % (remove_end(parsed_video_url.path, '/'), video_file.split(':')[-1])))
+                        if self._is_valid_url(direct_video_url, video_id, video_format):
+                            format_info.update({
+                                'format_id': 'http-' + video_format,
+                                'url': direct_video_url,
+                            })
+                            formats.append(format_info)
+                            continue
 
 
-            video_url = xpath_text(stream_info, './host')
-            if not video_url:
-                continue
-            metadata = stream_info.find('./metadata')
-            format_info = {
-                'format': video_format,
-                'format_id': video_format,
-                'height': int_or_none(xpath_text(metadata, './height')),
-                'width': int_or_none(xpath_text(metadata, './width')),
-            }
-
-            if '.fplive.net/' in video_url:
-                video_url = re.sub(r'^rtmpe?://', 'http://', video_url.strip())
-                parsed_video_url = compat_urlparse.urlparse(video_url)
-                direct_video_url = compat_urlparse.urlunparse(parsed_video_url._replace(
-                    netloc='v.lvlt.crcdn.net',
-                    path='%s/%s' % (remove_end(parsed_video_url.path, '/'), video_file.split(':')[-1])))
-                if self._is_valid_url(direct_video_url, video_id, video_format):
                     format_info.update({
                     format_info.update({
-                        'url': direct_video_url,
+                        'format_id': 'rtmp-' + video_format,
+                        'url': video_url,
+                        'play_path': video_file,
+                        'ext': 'flv',
                     })
                     formats.append(format_info)
                     })
                     formats.append(format_info)
-                    continue
-
-            format_info.update({
-                'url': video_url,
-                'play_path': video_file,
-                'ext': 'flv',
-            })
-            formats.append(format_info)
-        self._sort_formats(formats)
+        self._sort_formats(formats, ('preference', 'language_preference', 'height', 'width', 'tbr', 'fps'))
 
 
-        metadata = self._download_xml(
-            'http://www.crunchyroll.com/xml', video_id,
-            note='Downloading media info', query={
-                'req': 'RpcApiVideoPlayer_GetMediaMetadata',
+        metadata = self._call_rpc_api(
+            'VideoPlayer_GetMediaMetadata', video_id,
+            note='Downloading media info', data={
                 'media_id': video_id,
             })
 
                 'media_id': video_id,
             })
 
-        subtitles = self.extract_subtitles(video_id, webpage)
+        subtitles = {}
+        for subtitle in media.get('subtitles', []):
+            subtitle_url = subtitle.get('url')
+            if not subtitle_url:
+                continue
+            subtitles.setdefault(subtitle.get('language', 'enUS'), []).append({
+                'url': subtitle_url,
+                'ext': subtitle.get('format', 'ass'),
+            })
+        if not subtitles:
+            subtitles = self.extract_subtitles(video_id, webpage)
 
         # webpage provide more accurate data than series_title from XML
         series = self._html_search_regex(
 
         # webpage provide more accurate data than series_title from XML
         series = self._html_search_regex(
-            r'id=["\']showmedia_about_episode_num[^>]+>\s*<a[^>]+>([^<]+)',
+            r'(?s)<h\d[^>]+\bid=["\']showmedia_about_episode_num[^>]+>(.+?)</h\d',
             webpage, 'series', fatal=False)
             webpage, 'series', fatal=False)
-        season = xpath_text(metadata, 'series_title')
 
 
-        episode = xpath_text(metadata, 'episode_title')
-        episode_number = int_or_none(xpath_text(metadata, 'episode_number'))
+        season = episode = episode_number = duration = thumbnail = None
+
+        if isinstance(metadata, compat_etree_Element):
+            season = xpath_text(metadata, 'series_title')
+            episode = xpath_text(metadata, 'episode_title')
+            episode_number = int_or_none(xpath_text(metadata, 'episode_number'))
+            duration = float_or_none(media_metadata.get('duration'), 1000)
+            thumbnail = xpath_text(metadata, 'episode_image_url')
+
+        if not episode:
+            episode = media_metadata.get('title')
+        if not episode_number:
+            episode_number = int_or_none(media_metadata.get('episode_number'))
+        if not thumbnail:
+            thumbnail = media_metadata.get('thumbnail', {}).get('url')
 
         season_number = int_or_none(self._search_regex(
 
         season_number = int_or_none(self._search_regex(
-            r'(?s)<h4[^>]+id=["\']showmedia_about_episode_num[^>]+>.+?</h4>\s*<h4>\s*Season (\d+)',
+            r'(?s)<h\d[^>]+id=["\']showmedia_about_episode_num[^>]+>.+?</h\d>\s*<h4>\s*Season (\d+)',
             webpage, 'season number', default=None))
 
         return {
             'id': video_id,
             'title': video_title,
             'description': video_description,
             webpage, 'season number', default=None))
 
         return {
             'id': video_id,
             'title': video_title,
             'description': video_description,
-            'thumbnail': xpath_text(metadata, 'episode_image_url'),
+            'duration': duration,
+            'thumbnail': thumbnail,
             'uploader': video_uploader,
             'upload_date': video_upload_date,
             'series': series,
             'uploader': video_uploader,
             'upload_date': video_upload_date,
             'series': series,
@@ -538,7 +631,7 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
 
 class CrunchyrollShowPlaylistIE(CrunchyrollBaseIE):
     IE_NAME = 'crunchyroll:playlist'
 
 class CrunchyrollShowPlaylistIE(CrunchyrollBaseIE):
     IE_NAME = 'crunchyroll:playlist'
-    _VALID_URL = r'https?://(?:(?P<prefix>www|m)\.)?(?P<url>crunchyroll\.com/(?!(?:news|anime-news|library|forum|launchcalendar|lineup|store|comics|freetrial|login))(?P<id>[\w\-]+))/?(?:\?|$)'
+    _VALID_URL = r'https?://(?:(?P<prefix>www|m)\.)?(?P<url>crunchyroll\.com/(?!(?:news|anime-news|library|forum|launchcalendar|lineup|store|comics|freetrial|login|media-\d+))(?P<id>[\w\-]+))/?(?:\?|$)'
 
     _TESTS = [{
         'url': 'http://www.crunchyroll.com/a-bridge-to-the-starry-skies-hoshizora-e-kakaru-hashi',
 
     _TESTS = [{
         'url': 'http://www.crunchyroll.com/a-bridge-to-the-starry-skies-hoshizora-e-kakaru-hashi',
@@ -565,10 +658,11 @@ class CrunchyrollShowPlaylistIE(CrunchyrollBaseIE):
     def _real_extract(self, url):
         show_id = self._match_id(url)
 
     def _real_extract(self, url):
         show_id = self._match_id(url)
 
-        webpage = self._download_webpage(self._add_skip_wall(url), show_id)
-        title = self._html_search_regex(
-            r'(?s)<h1[^>]*>\s*<span itemprop="name">(.*?)</span>',
-            webpage, 'title')
+        webpage = self._download_webpage(
+            self._add_skip_wall(url), show_id,
+            headers=self.geo_verification_headers())
+        title = self._html_search_meta('name', webpage, default=None)
+
         episode_paths = re.findall(
             r'(?s)<li id="showview_videos_media_(\d+)"[^>]+>.*?<a href="([^"]+)"',
             webpage)
         episode_paths = re.findall(
             r'(?s)<li id="showview_videos_media_(\d+)"[^>]+>.*?<a href="([^"]+)"',
             webpage)