]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/noco.py
Imported Upstream version 2015.02.06
[youtubedl] / youtube_dl / extractor / noco.py
index d451cd1bf8db30dd7b3315ee1edb35489ad6a8c0..251e6da07457b7e7be6b5703b5769214ae299c3d 100644 (file)
@@ -2,17 +2,28 @@
 from __future__ import unicode_literals
 
 import re
 from __future__ import unicode_literals
 
 import re
+import time
+import hashlib
 
 from .common import InfoExtractor
 
 from .common import InfoExtractor
+from ..compat import (
+    compat_str,
+    compat_urllib_parse,
+    compat_urllib_request,
+)
 from ..utils import (
 from ..utils import (
+    clean_html,
     ExtractorError,
     unified_strdate,
     ExtractorError,
     unified_strdate,
-    compat_str,
 )
 
 
 class NocoIE(InfoExtractor):
     _VALID_URL = r'http://(?:(?:www\.)?noco\.tv/emission/|player\.noco\.tv/\?idvideo=)(?P<id>\d+)'
 )
 
 
 class NocoIE(InfoExtractor):
     _VALID_URL = r'http://(?:(?:www\.)?noco\.tv/emission/|player\.noco\.tv/\?idvideo=)(?P<id>\d+)'
+    _LOGIN_URL = 'http://noco.tv/do.php'
+    _API_URL_TEMPLATE = 'https://api.noco.tv/1.1/%s?ts=%s&tk=%s'
+    _SUB_LANG_TEMPLATE = '&sub_lang=%s'
+    _NETRC_MACHINE = 'noco'
 
     _TEST = {
         'url': 'http://noco.tv/emission/11538/nolife/ami-ami-idol-hello-france/',
 
     _TEST = {
         'url': 'http://noco.tv/emission/11538/nolife/ami-ami-idol-hello-france/',
@@ -30,54 +41,111 @@ class NocoIE(InfoExtractor):
         'skip': 'Requires noco account',
     }
 
         'skip': 'Requires noco account',
     }
 
+    def _real_initialize(self):
+        self._login()
+
+    def _login(self):
+        (username, password) = self._get_login_info()
+        if username is None:
+            return
+
+        login_form = {
+            'a': 'login',
+            'cookie': '1',
+            'username': username,
+            'password': password,
+        }
+        request = compat_urllib_request.Request(self._LOGIN_URL, compat_urllib_parse.urlencode(login_form))
+        request.add_header('Content-Type', 'application/x-www-form-urlencoded; charset=UTF-8')
+
+        login = self._download_json(request, None, 'Logging in as %s' % username)
+
+        if 'erreur' in login:
+            raise ExtractorError('Unable to login: %s' % clean_html(login['erreur']), expected=True)
+
+    def _call_api(self, path, video_id, note, sub_lang=None):
+        ts = compat_str(int(time.time() * 1000))
+        tk = hashlib.md5((hashlib.md5(ts.encode('ascii')).hexdigest() + '#8S?uCraTedap6a').encode('ascii')).hexdigest()
+        url = self._API_URL_TEMPLATE % (path, ts, tk)
+        if sub_lang:
+            url += self._SUB_LANG_TEMPLATE % sub_lang
+
+        resp = self._download_json(url, video_id, note)
+
+        if isinstance(resp, dict) and resp.get('error'):
+            self._raise_error(resp['error'], resp['description'])
+
+        return resp
+
+    def _raise_error(self, error, description):
+        raise ExtractorError(
+            '%s returned error: %s - %s' % (self.IE_NAME, error, description),
+            expected=True)
+
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
         video_id = mobj.group('id')
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
         video_id = mobj.group('id')
 
-        medias = self._download_json(
-            'http://api.noco.tv/1.0/video/medias/%s' % video_id, video_id, 'Downloading video JSON')
-
-        formats = []
-
-        for fmt in medias['fr']['video_list']['default']['quality_list']:
-            format_id = fmt['quality_key']
+        medias = self._call_api(
+            'shows/%s/medias' % video_id,
+            video_id, 'Downloading video JSON')
 
 
-            file = self._download_json(
-                'http://api.noco.tv/1.0/video/file/%s/fr/%s' % (format_id.lower(), video_id),
-                video_id, 'Downloading %s video JSON' % format_id)
+        qualities = self._call_api(
+            'qualities',
+            video_id, 'Downloading qualities JSON')
 
 
-            file_url = file['file']
-            if not file_url:
-                continue
+        formats = []
 
 
-            if file_url == 'forbidden':
-                raise ExtractorError(
-                    '%s returned error: %s - %s' % (
-                        self.IE_NAME, file['popmessage']['title'], file['popmessage']['message']),
-                    expected=True)
-
-            formats.append({
-                'url': file_url,
-                'format_id': format_id,
-                'width': fmt['res_width'],
-                'height': fmt['res_lines'],
-                'abr': fmt['audiobitrate'],
-                'vbr': fmt['videobitrate'],
-                'filesize': fmt['filesize'],
-                'format_note': fmt['quality_name'],
-                'preference': fmt['priority'],
-            })
+        for lang, lang_dict in medias['fr']['video_list'].items():
+            for format_id, fmt in lang_dict['quality_list'].items():
+                format_id_extended = '%s-%s' % (lang, format_id) if lang != 'none' else format_id
+
+                video = self._call_api(
+                    'shows/%s/video/%s/fr' % (video_id, format_id.lower()),
+                    video_id, 'Downloading %s video JSON' % format_id_extended,
+                    lang if lang != 'none' else None)
+
+                file_url = video['file']
+                if not file_url:
+                    continue
+
+                if file_url in ['forbidden', 'not found']:
+                    popmessage = video['popmessage']
+                    self._raise_error(popmessage['title'], popmessage['message'])
+
+                formats.append({
+                    'url': file_url,
+                    'format_id': format_id_extended,
+                    'width': fmt['res_width'],
+                    'height': fmt['res_lines'],
+                    'abr': fmt['audiobitrate'],
+                    'vbr': fmt['videobitrate'],
+                    'filesize': fmt['filesize'],
+                    'format_note': qualities[format_id]['quality_name'],
+                    'preference': qualities[format_id]['priority'],
+                })
 
         self._sort_formats(formats)
 
 
         self._sort_formats(formats)
 
-        show = self._download_json(
-            'http://api.noco.tv/1.0/shows/show/%s' % video_id, video_id, 'Downloading show JSON')[0]
+        show = self._call_api(
+            'shows/by_id/%s' % video_id,
+            video_id, 'Downloading show JSON')[0]
 
 
-        upload_date = unified_strdate(show['indexed'])
+        upload_date = unified_strdate(show['online_date_start_utc'])
         uploader = show['partner_name']
         uploader_id = show['partner_key']
         duration = show['duration_ms'] / 1000.0
         uploader = show['partner_name']
         uploader_id = show['partner_key']
         duration = show['duration_ms'] / 1000.0
-        thumbnail = show['screenshot']
+
+        thumbnails = []
+        for thumbnail_key, thumbnail_url in show.items():
+            m = re.search(r'^screenshot_(?P<width>\d+)x(?P<height>\d+)$', thumbnail_key)
+            if not m:
+                continue
+            thumbnails.append({
+                'url': thumbnail_url,
+                'width': int(m.group('width')),
+                'height': int(m.group('height')),
+            })
 
         episode = show.get('show_TT') or show.get('show_OT')
         family = show.get('family_TT') or show.get('family_OT')
 
         episode = show.get('show_TT') or show.get('show_OT')
         family = show.get('family_TT') or show.get('family_OT')
@@ -97,10 +165,10 @@ class NocoIE(InfoExtractor):
             'id': video_id,
             'title': title,
             'description': description,
             'id': video_id,
             'title': title,
             'description': description,
-            'thumbnail': thumbnail,
+            'thumbnails': thumbnails,
             'upload_date': upload_date,
             'uploader': uploader,
             'uploader_id': uploader_id,
             'duration': duration,
             'formats': formats,
             'upload_date': upload_date,
             'uploader': uploader,
             'uploader_id': uploader_id,
             'duration': duration,
             'formats': formats,
-        }
\ No newline at end of file
+        }