aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorremitamine <remitamine@gmail.com>2015-06-24 01:13:23 +0100
committerremitamine <remitamine@gmail.com>2015-07-23 11:57:08 +0100
commit984e4d487520bd2a860b31b3165416c879b28096 (patch)
tree2a1e28842bb328a7ee100c822b44c01343e75473
parent53b8247cb5b5ac4a7822c82b94ec2f1221a40625 (diff)
[googledrive] Add new extractor
-rw-r--r--youtube_dl/extractor/__init__.py1
-rw-r--r--youtube_dl/extractor/googledrive.py106
2 files changed, 107 insertions, 0 deletions
diff --git a/youtube_dl/extractor/__init__.py b/youtube_dl/extractor/__init__.py
index 3cfa804ec..6655d7eb5 100644
--- a/youtube_dl/extractor/__init__.py
+++ b/youtube_dl/extractor/__init__.py
@@ -209,6 +209,7 @@ from .globo import GloboIE
from .godtube import GodTubeIE
from .goldenmoustache import GoldenMoustacheIE
from .golem import GolemIE
+from .googledrive import GoogleDriveIE
from .googleplus import GooglePlusIE
from .googlesearch import GoogleSearchIE
from .gorillavid import GorillaVidIE
diff --git a/youtube_dl/extractor/googledrive.py b/youtube_dl/extractor/googledrive.py
new file mode 100644
index 000000000..8c611fa47
--- /dev/null
+++ b/youtube_dl/extractor/googledrive.py
@@ -0,0 +1,106 @@
+from .common import InfoExtractor
+from ..utils import RegexNotFoundError
+
+class GoogleDriveIE(InfoExtractor):
+ _VALID_URL = r'(?:https?://)?(?:video\.google\.com/get_player\?.*?docid=|(?:docs|drive)\.google\.com/(?:uc\?.*?id=|file/d/))(?P<id>.+?)(?:&|/|$)'
+ _TEST = {
+ 'url': 'https://drive.google.com/file/d/0BzpExh0WzJF0NlR5WUlxdEVsY0U/edit?pli=1',
+ 'info_dict': {
+ 'id': '0BzpExh0WzJF0NlR5WUlxdEVsY0U',
+ 'ext': 'mp4',
+ 'title': '[AHSH] Fairy Tail S2 - 01 [720p].mp4',
+ }
+ }
+ _formats = {
+ '5': {'ext': 'flv'},
+ '6': {'ext': 'flv'},
+ '13': {'ext': '3gp'},
+ '17': {'ext': '3gp'},
+ '18': {'ext': 'mp4'},
+ '22': {'ext': 'mp4'},
+ '34': {'ext': 'flv'},
+ '35': {'ext': 'flv'},
+ '36': {'ext': '3gp'},
+ '37': {'ext': 'mp4'},
+ '38': {'ext': 'mp4'},
+ '43': {'ext': 'webm'},
+ '44': {'ext': 'webm'},
+ '45': {'ext': 'webm'},
+ '46': {'ext': 'webm'},
+ '59': {'ext': 'mp4'}
+ }
+
+ def _real_extract(self, url):
+ video_id = self._match_id(url)
+ webpage = self._download_webpage(
+ 'http://docs.google.com/file/d/'+video_id, video_id, encoding='unicode_escape'
+ )
+ try:
+ title = self._html_search_regex(
+ r'"title","(?P<title>.*?)"',
+ webpage,
+ 'title',
+ group='title'
+ )
+ fmt_stream_map = self._html_search_regex(
+ r'"fmt_stream_map","(?P<fmt_stream_map>.*?)"',
+ webpage,
+ 'fmt_stream_map',
+ group='fmt_stream_map'
+ )
+ fmt_list = self._html_search_regex(
+ r'"fmt_list","(?P<fmt_list>.*?)"',
+ webpage,
+ 'fmt_list',
+ group='fmt_list'
+ )
+# timestamp = self._html_search_regex(
+# r'"timestamp","(?P<timestamp>.*?)"',
+# webpage,
+# 'timestamp',
+# group='timestamp'
+# )
+ length_seconds = self._html_search_regex(
+ r'"length_seconds","(?P<length_seconds>.*?)"',
+ webpage,
+ 'length_seconds',
+ group='length_seconds'
+ )
+ except RegexNotFoundError:
+ try:
+ reason = self._html_search_regex(
+ r'"reason","(?P<reason>.*?)"',
+ webpage,
+ 'reason',
+ group='reason'
+ )
+ self.report_warning(reason)
+ return
+ except RegexNotFoundError:
+ self.report_warning('not a video')
+ return
+
+ fmt_stream_map = fmt_stream_map.split(',')
+ fmt_list = fmt_list.split(',')
+ formats = []
+ for i in range(len(fmt_stream_map)):
+ fmt_id, fmt_url = fmt_stream_map[i].split('|')
+ resolution = fmt_list[i].split('/')[1]
+ width, height = resolution.split('x')
+ formats.append({
+ 'url': fmt_url,
+ 'format_id': fmt_id,
+ 'resolution': resolution,
+ 'width': int(width),
+ 'height': int(height),
+ 'ext': self._formats[fmt_id]['ext']
+ })
+ self._sort_formats(formats)
+
+ return {
+ 'id': video_id,
+ 'title': title,
+# 'timestamp': int(timestamp),
+ 'duration': int(length_seconds),
+ 'formats': formats
+ }