]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/googledrive.py
Initiate new release.
[youtubedl] / youtube_dl / extractor / googledrive.py
index 766fc26d0f01145bdd2456a221940fa60ece6953..fec36cbbb7f43d2b8b37370aec270f543e8f257d 100644 (file)
@@ -6,6 +6,7 @@ from .common import InfoExtractor
 from ..utils import (
     ExtractorError,
     int_or_none,
 from ..utils import (
     ExtractorError,
     int_or_none,
+    lowercase_escape,
 )
 
 
 )
 
 
@@ -13,12 +14,12 @@ class GoogleDriveIE(InfoExtractor):
     _VALID_URL = r'https?://(?:(?:docs|drive)\.google\.com/(?:uc\?.*?id=|file/d/)|video\.google\.com/get_player\?.*?docid=)(?P<id>[a-zA-Z0-9_-]{28,})'
     _TESTS = [{
         'url': 'https://drive.google.com/file/d/0ByeS4oOUV-49Zzh4R1J6R09zazQ/edit?pli=1',
     _VALID_URL = r'https?://(?:(?:docs|drive)\.google\.com/(?:uc\?.*?id=|file/d/)|video\.google\.com/get_player\?.*?docid=)(?P<id>[a-zA-Z0-9_-]{28,})'
     _TESTS = [{
         'url': 'https://drive.google.com/file/d/0ByeS4oOUV-49Zzh4R1J6R09zazQ/edit?pli=1',
-        'md5': '881f7700aec4f538571fa1e0eed4a7b6',
+        'md5': 'd109872761f7e7ecf353fa108c0dbe1e',
         'info_dict': {
             'id': '0ByeS4oOUV-49Zzh4R1J6R09zazQ',
             'ext': 'mp4',
             'title': 'Big Buck Bunny.mp4',
         'info_dict': {
             'id': '0ByeS4oOUV-49Zzh4R1J6R09zazQ',
             'ext': 'mp4',
             'title': 'Big Buck Bunny.mp4',
-            'duration': 46,
+            'duration': 45,
         }
     }, {
         # video id is longer than 28 characters
         }
     }, {
         # video id is longer than 28 characters
@@ -55,7 +56,7 @@ class GoogleDriveIE(InfoExtractor):
     def _real_extract(self, url):
         video_id = self._match_id(url)
         webpage = self._download_webpage(
     def _real_extract(self, url):
         video_id = self._match_id(url)
         webpage = self._download_webpage(
-            'http://docs.google.com/file/d/%s' % video_id, video_id, encoding='unicode_escape')
+            'http://docs.google.com/file/d/%s' % video_id, video_id)
 
         reason = self._search_regex(r'"reason"\s*,\s*"([^"]+)', webpage, 'reason', default=None)
         if reason:
 
         reason = self._search_regex(r'"reason"\s*,\s*"([^"]+)', webpage, 'reason', default=None)
         if reason:
@@ -74,7 +75,7 @@ class GoogleDriveIE(InfoExtractor):
             resolution = fmt.split('/')[1]
             width, height = resolution.split('x')
             formats.append({
             resolution = fmt.split('/')[1]
             width, height = resolution.split('x')
             formats.append({
-                'url': fmt_url,
+                'url': lowercase_escape(fmt_url),
                 'format_id': fmt_id,
                 'resolution': resolution,
                 'width': int_or_none(width),
                 'format_id': fmt_id,
                 'resolution': resolution,
                 'width': int_or_none(width),