Move MyVideo into its own file
authorPhilipp Hagemeister <phihag@phihag.de>
Sun, 23 Jun 2013 18:48:32 +0000 (20:48 +0200)
committerPhilipp Hagemeister <phihag@phihag.de>
Sun, 23 Jun 2013 18:48:32 +0000 (20:48 +0200)
youtube_dl/InfoExtractors.py
youtube_dl/extractor/myvideo.py [new file with mode: 0644]

index a3cc2e6f5d96fd82151d8dc7cd6eba3c371a03ee..5c615242738f9aadc5af81b7b95148a06d5e6773 100755 (executable)
@@ -25,6 +25,7 @@ from .extractor.dailymotion import DailymotionIE
 from .extractor.gametrailers import GametrailersIE
 from .extractor.generic import GenericIE
 from .extractor.metacafe import MetacafeIE
+from .extractor.myvideo import MyVideoIE
 from .extractor.statigram import StatigramIE
 from .extractor.photobucket import PhotobucketIE
 from .extractor.vimeo import VimeoIE
@@ -194,155 +195,6 @@ class FacebookIE(InfoExtractor):
 
 
 
-class MyVideoIE(InfoExtractor):
-    """Information Extractor for myvideo.de."""
-
-    _VALID_URL = r'(?:http://)?(?:www\.)?myvideo\.de/watch/([0-9]+)/([^?/]+).*'
-    IE_NAME = u'myvideo'
-
-    # Original Code from: https://github.com/dersphere/plugin.video.myvideo_de.git
-    # Released into the Public Domain by Tristan Fischer on 2013-05-19
-    # https://github.com/rg3/youtube-dl/pull/842
-    def __rc4crypt(self,data, key):
-        x = 0
-        box = list(range(256))
-        for i in list(range(256)):
-            x = (x + box[i] + compat_ord(key[i % len(key)])) % 256
-            box[i], box[x] = box[x], box[i]
-        x = 0
-        y = 0
-        out = ''
-        for char in data:
-            x = (x + 1) % 256
-            y = (y + box[x]) % 256
-            box[x], box[y] = box[y], box[x]
-            out += chr(compat_ord(char) ^ box[(box[x] + box[y]) % 256])
-        return out
-
-    def __md5(self,s):
-        return hashlib.md5(s).hexdigest().encode()
-
-    def _real_extract(self,url):
-        mobj = re.match(self._VALID_URL, url)
-        if mobj is None:
-            raise ExtractorError(u'invalid URL: %s' % url)
-
-        video_id = mobj.group(1)
-
-        GK = (
-          b'WXpnME1EZGhNRGhpTTJNM01XVmhOREU0WldNNVpHTTJOakpt'
-          b'TW1FMU5tVTBNR05pWkRaa05XRXhNVFJoWVRVd1ptSXhaVEV3'
-          b'TnpsbA0KTVRkbU1tSTRNdz09'
-        )
-
-        # Get video webpage
-        webpage_url = 'http://www.myvideo.de/watch/%s' % video_id
-        webpage = self._download_webpage(webpage_url, video_id)
-
-        mobj = re.search('source src=\'(.+?)[.]([^.]+)\'', webpage)
-        if mobj is not None:
-            self.report_extraction(video_id)
-            video_url = mobj.group(1) + '.flv'
-
-            video_title = self._html_search_regex('<title>([^<]+)</title>',
-                webpage, u'title')
-
-            video_ext = self._search_regex('[.](.+?)$', video_url, u'extension')
-
-            return [{
-                'id':       video_id,
-                'url':      video_url,
-                'uploader': None,
-                'upload_date':  None,
-                'title':    video_title,
-                'ext':      u'flv',
-            }]
-
-        # try encxml
-        mobj = re.search('var flashvars={(.+?)}', webpage)
-        if mobj is None:
-            raise ExtractorError(u'Unable to extract video')
-
-        params = {}
-        encxml = ''
-        sec = mobj.group(1)
-        for (a, b) in re.findall('(.+?):\'(.+?)\',?', sec):
-            if not a == '_encxml':
-                params[a] = b
-            else:
-                encxml = compat_urllib_parse.unquote(b)
-        if not params.get('domain'):
-            params['domain'] = 'www.myvideo.de'
-        xmldata_url = '%s?%s' % (encxml, compat_urllib_parse.urlencode(params))
-        if 'flash_playertype=MTV' in xmldata_url:
-            self._downloader.report_warning(u'avoiding MTV player')
-            xmldata_url = (
-                'http://www.myvideo.de/dynamic/get_player_video_xml.php'
-                '?flash_playertype=D&ID=%s&_countlimit=4&autorun=yes'
-            ) % video_id
-
-        # get enc data
-        enc_data = self._download_webpage(xmldata_url, video_id).split('=')[1]
-        enc_data_b = binascii.unhexlify(enc_data)
-        sk = self.__md5(
-            base64.b64decode(base64.b64decode(GK)) +
-            self.__md5(
-                str(video_id).encode('utf-8')
-            )
-        )
-        dec_data = self.__rc4crypt(enc_data_b, sk)
-
-        # extracting infos
-        self.report_extraction(video_id)
-
-        video_url = None
-        mobj = re.search('connectionurl=\'(.*?)\'', dec_data)
-        if mobj:
-            video_url = compat_urllib_parse.unquote(mobj.group(1))
-            if 'myvideo2flash' in video_url:
-                self._downloader.report_warning(u'forcing RTMPT ...')
-                video_url = video_url.replace('rtmpe://', 'rtmpt://')
-
-        if not video_url:
-            # extract non rtmp videos
-            mobj = re.search('path=\'(http.*?)\' source=\'(.*?)\'', dec_data)
-            if mobj is None:
-                raise ExtractorError(u'unable to extract url')
-            video_url = compat_urllib_parse.unquote(mobj.group(1)) + compat_urllib_parse.unquote(mobj.group(2))
-
-        video_file = self._search_regex('source=\'(.*?)\'', dec_data, u'video file')
-        video_file = compat_urllib_parse.unquote(video_file)
-
-        if not video_file.endswith('f4m'):
-            ppath, prefix = video_file.split('.')
-            video_playpath = '%s:%s' % (prefix, ppath)
-            video_hls_playlist = ''
-        else:
-            video_playpath = ''
-            video_hls_playlist = (
-                video_filepath + video_file
-            ).replace('.f4m', '.m3u8')
-
-        video_swfobj = self._search_regex('swfobject.embedSWF\(\'(.+?)\'', webpage, u'swfobj')
-        video_swfobj = compat_urllib_parse.unquote(video_swfobj)
-
-        video_title = self._html_search_regex("<h1(?: class='globalHd')?>(.*?)</h1>",
-            webpage, u'title')
-
-        return [{
-            'id':                 video_id,
-            'url':                video_url,
-            'tc_url':             video_url,
-            'uploader':           None,
-            'upload_date':        None,
-            'title':              video_title,
-            'ext':                u'flv',
-            'play_path':          video_playpath,
-            'video_file':         video_file,
-            'video_hls_playlist': video_hls_playlist,
-            'player_url':         video_swfobj,
-        }]
-
 
 class ComedyCentralIE(InfoExtractor):
     """Information extractor for The Daily Show and Colbert Report """
diff --git a/youtube_dl/extractor/myvideo.py b/youtube_dl/extractor/myvideo.py
new file mode 100644 (file)
index 0000000..47a44e3
--- /dev/null
@@ -0,0 +1,164 @@
+import binascii
+import base64
+import hashlib
+import re
+
+from .common import InfoExtractor
+from ..utils import (
+    compat_ord,
+    compat_urllib_parse,
+
+    ExtractorError,
+)
+
+
+
+class MyVideoIE(InfoExtractor):
+    """Information Extractor for myvideo.de."""
+
+    _VALID_URL = r'(?:http://)?(?:www\.)?myvideo\.de/watch/([0-9]+)/([^?/]+).*'
+    IE_NAME = u'myvideo'
+
+    # Original Code from: https://github.com/dersphere/plugin.video.myvideo_de.git
+    # Released into the Public Domain by Tristan Fischer on 2013-05-19
+    # https://github.com/rg3/youtube-dl/pull/842
+    def __rc4crypt(self,data, key):
+        x = 0
+        box = list(range(256))
+        for i in list(range(256)):
+            x = (x + box[i] + compat_ord(key[i % len(key)])) % 256
+            box[i], box[x] = box[x], box[i]
+        x = 0
+        y = 0
+        out = ''
+        for char in data:
+            x = (x + 1) % 256
+            y = (y + box[x]) % 256
+            box[x], box[y] = box[y], box[x]
+            out += chr(compat_ord(char) ^ box[(box[x] + box[y]) % 256])
+        return out
+
+    def __md5(self,s):
+        return hashlib.md5(s).hexdigest().encode()
+
+    def _real_extract(self,url):
+        mobj = re.match(self._VALID_URL, url)
+        if mobj is None:
+            raise ExtractorError(u'invalid URL: %s' % url)
+
+        video_id = mobj.group(1)
+
+        GK = (
+          b'WXpnME1EZGhNRGhpTTJNM01XVmhOREU0WldNNVpHTTJOakpt'
+          b'TW1FMU5tVTBNR05pWkRaa05XRXhNVFJoWVRVd1ptSXhaVEV3'
+          b'TnpsbA0KTVRkbU1tSTRNdz09'
+        )
+
+        # Get video webpage
+        webpage_url = 'http://www.myvideo.de/watch/%s' % video_id
+        webpage = self._download_webpage(webpage_url, video_id)
+
+        mobj = re.search('source src=\'(.+?)[.]([^.]+)\'', webpage)
+        if mobj is not None:
+            self.report_extraction(video_id)
+            video_url = mobj.group(1) + '.flv'
+
+            video_title = self._html_search_regex('<title>([^<]+)</title>',
+                webpage, u'title')
+
+            video_ext = self._search_regex('[.](.+?)$', video_url, u'extension')
+
+            return [{
+                'id':       video_id,
+                'url':      video_url,
+                'uploader': None,
+                'upload_date':  None,
+                'title':    video_title,
+                'ext':      video_ext,
+            }]
+
+        # try encxml
+        mobj = re.search('var flashvars={(.+?)}', webpage)
+        if mobj is None:
+            raise ExtractorError(u'Unable to extract video')
+
+        params = {}
+        encxml = ''
+        sec = mobj.group(1)
+        for (a, b) in re.findall('(.+?):\'(.+?)\',?', sec):
+            if not a == '_encxml':
+                params[a] = b
+            else:
+                encxml = compat_urllib_parse.unquote(b)
+        if not params.get('domain'):
+            params['domain'] = 'www.myvideo.de'
+        xmldata_url = '%s?%s' % (encxml, compat_urllib_parse.urlencode(params))
+        if 'flash_playertype=MTV' in xmldata_url:
+            self._downloader.report_warning(u'avoiding MTV player')
+            xmldata_url = (
+                'http://www.myvideo.de/dynamic/get_player_video_xml.php'
+                '?flash_playertype=D&ID=%s&_countlimit=4&autorun=yes'
+            ) % video_id
+
+        # get enc data
+        enc_data = self._download_webpage(xmldata_url, video_id).split('=')[1]
+        enc_data_b = binascii.unhexlify(enc_data)
+        sk = self.__md5(
+            base64.b64decode(base64.b64decode(GK)) +
+            self.__md5(
+                str(video_id).encode('utf-8')
+            )
+        )
+        dec_data = self.__rc4crypt(enc_data_b, sk)
+
+        # extracting infos
+        self.report_extraction(video_id)
+
+        video_url = None
+        mobj = re.search('connectionurl=\'(.*?)\'', dec_data)
+        if mobj:
+            video_url = compat_urllib_parse.unquote(mobj.group(1))
+            if 'myvideo2flash' in video_url:
+                self._downloader.report_warning(u'forcing RTMPT ...')
+                video_url = video_url.replace('rtmpe://', 'rtmpt://')
+
+        if not video_url:
+            # extract non rtmp videos
+            mobj = re.search('path=\'(http.*?)\' source=\'(.*?)\'', dec_data)
+            if mobj is None:
+                raise ExtractorError(u'unable to extract url')
+            video_url = compat_urllib_parse.unquote(mobj.group(1)) + compat_urllib_parse.unquote(mobj.group(2))
+
+        video_file = self._search_regex('source=\'(.*?)\'', dec_data, u'video file')
+        video_file = compat_urllib_parse.unquote(video_file)
+
+        if not video_file.endswith('f4m'):
+            ppath, prefix = video_file.split('.')
+            video_playpath = '%s:%s' % (prefix, ppath)
+            video_hls_playlist = ''
+        else:
+            video_playpath = ''
+            video_hls_playlist = (
+                video_file
+            ).replace('.f4m', '.m3u8')
+
+        video_swfobj = self._search_regex('swfobject.embedSWF\(\'(.+?)\'', webpage, u'swfobj')
+        video_swfobj = compat_urllib_parse.unquote(video_swfobj)
+
+        video_title = self._html_search_regex("<h1(?: class='globalHd')?>(.*?)</h1>",
+            webpage, u'title')
+
+        return [{
+            'id':                 video_id,
+            'url':                video_url,
+            'tc_url':             video_url,
+            'uploader':           None,
+            'upload_date':        None,
+            'title':              video_title,
+            'ext':                u'flv',
+            'play_path':          video_playpath,
+            'video_file':         video_file,
+            'video_hls_playlist': video_hls_playlist,
+            'player_url':         video_swfobj,
+        }]
+