2015-12-21 02:05:34 +00:00
|
|
|
from __future__ import unicode_literals
|
|
|
|
|
2015-06-29 06:53:21 +00:00
|
|
|
import re
|
|
|
|
|
2015-06-24 00:13:23 +00:00
|
|
|
from .common import InfoExtractor
|
2015-07-18 22:31:14 +00:00
|
|
|
from ..utils import (
|
|
|
|
ExtractorError,
|
2015-12-21 02:05:34 +00:00
|
|
|
int_or_none,
|
2017-02-05 13:41:08 +00:00
|
|
|
lowercase_escape,
|
2017-08-22 16:48:59 +00:00
|
|
|
error_to_compat_str,
|
|
|
|
update_url_query,
|
2015-07-18 22:31:14 +00:00
|
|
|
)
|
2015-06-24 00:13:23 +00:00
|
|
|
|
2015-12-21 02:05:34 +00:00
|
|
|
|
|
|
|
class GoogleDriveIE(InfoExtractor):
|
2016-03-11 18:36:39 +00:00
|
|
|
_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 = [{
|
2015-12-21 02:05:34 +00:00
|
|
|
'url': 'https://drive.google.com/file/d/0ByeS4oOUV-49Zzh4R1J6R09zazQ/edit?pli=1',
|
2017-02-05 13:41:08 +00:00
|
|
|
'md5': 'd109872761f7e7ecf353fa108c0dbe1e',
|
2015-06-29 06:53:21 +00:00
|
|
|
'info_dict': {
|
2015-12-21 02:05:34 +00:00
|
|
|
'id': '0ByeS4oOUV-49Zzh4R1J6R09zazQ',
|
2015-06-29 06:53:21 +00:00
|
|
|
'ext': 'mp4',
|
2015-12-21 02:05:34 +00:00
|
|
|
'title': 'Big Buck Bunny.mp4',
|
2017-02-05 13:41:08 +00:00
|
|
|
'duration': 45,
|
2015-06-29 06:53:21 +00:00
|
|
|
}
|
2016-03-11 18:36:39 +00:00
|
|
|
}, {
|
|
|
|
# video id is longer than 28 characters
|
|
|
|
'url': 'https://drive.google.com/file/d/1ENcQ_jeCuj7y19s66_Ou9dRP4GKGsodiDQ/edit',
|
2017-08-22 16:48:59 +00:00
|
|
|
'md5': 'c230c67252874fddd8170e3fd1a45886',
|
|
|
|
'info_dict': {
|
|
|
|
'id': '1ENcQ_jeCuj7y19s66_Ou9dRP4GKGsodiDQ',
|
|
|
|
'ext': 'mp4',
|
|
|
|
'title': 'Andreea Banica feat Smiley - Hooky Song (Official Video).mp4',
|
|
|
|
'duration': 189,
|
|
|
|
},
|
|
|
|
'only_matching': True
|
2016-03-11 18:36:39 +00:00
|
|
|
}]
|
2015-12-21 02:05:34 +00:00
|
|
|
_FORMATS_EXT = {
|
|
|
|
'5': 'flv',
|
|
|
|
'6': 'flv',
|
|
|
|
'13': '3gp',
|
|
|
|
'17': '3gp',
|
|
|
|
'18': 'mp4',
|
|
|
|
'22': 'mp4',
|
|
|
|
'34': 'flv',
|
|
|
|
'35': 'flv',
|
|
|
|
'36': '3gp',
|
|
|
|
'37': 'mp4',
|
|
|
|
'38': 'mp4',
|
|
|
|
'43': 'webm',
|
|
|
|
'44': 'webm',
|
|
|
|
'45': 'webm',
|
|
|
|
'46': 'webm',
|
|
|
|
'59': 'mp4',
|
|
|
|
}
|
2017-08-22 16:48:59 +00:00
|
|
|
_BASE_URL_CAPTIONS = 'https://drive.google.com/timedtext'
|
|
|
|
_CAPTIONS_ENTRY_TAG = {
|
|
|
|
'subtitles': 'track',
|
|
|
|
'automatic_captions': 'target',
|
|
|
|
}
|
|
|
|
_caption_formats_ext = []
|
|
|
|
_captions_by_country_xml = None
|
2015-06-29 06:53:21 +00:00
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def _extract_url(webpage):
|
|
|
|
mobj = re.search(
|
2016-03-11 18:36:39 +00:00
|
|
|
r'<iframe[^>]+src="https?://(?:video\.google\.com/get_player\?.*?docid=|(?:docs|drive)\.google\.com/file/d/)(?P<id>[a-zA-Z0-9_-]{28,})',
|
2015-06-29 06:53:21 +00:00
|
|
|
webpage)
|
|
|
|
if mobj:
|
|
|
|
return 'https://drive.google.com/file/d/%s' % mobj.group('id')
|
|
|
|
|
2017-08-22 16:48:59 +00:00
|
|
|
def _set_captions_data(self, video_id, video_subtitles_id, hl):
|
|
|
|
try:
|
|
|
|
self._captions_by_country_xml = self._download_xml(self._BASE_URL_CAPTIONS, video_id, query={
|
|
|
|
'id': video_id,
|
|
|
|
'vid': video_subtitles_id,
|
|
|
|
'hl': hl,
|
|
|
|
'v': video_id,
|
|
|
|
'type': 'list',
|
|
|
|
'tlangs': '1',
|
|
|
|
'fmts': '1',
|
|
|
|
'vssids': '1',
|
|
|
|
})
|
|
|
|
except ExtractorError as ee:
|
|
|
|
self.report_warning('unable to download video subtitles: %s' % error_to_compat_str(ee))
|
|
|
|
if self._captions_by_country_xml is not None:
|
|
|
|
caption_available_extensions = self._captions_by_country_xml.findall('format')
|
|
|
|
for caption_extension in caption_available_extensions:
|
|
|
|
if caption_extension.attrib.get('fmt_code') and not caption_extension.attrib.get('default'):
|
|
|
|
self._caption_formats_ext.append(caption_extension.attrib['fmt_code'])
|
|
|
|
|
|
|
|
def _get_captions_by_type(self, video_id, video_subtitles_id, caption_type, caption_original_lang_code=None):
|
|
|
|
if not video_subtitles_id or not caption_type:
|
|
|
|
return None
|
|
|
|
captions = {}
|
|
|
|
for caption_entry in self._captions_by_country_xml.findall(self._CAPTIONS_ENTRY_TAG[caption_type]):
|
|
|
|
caption_lang_code = caption_entry.attrib.get('lang_code')
|
|
|
|
if not caption_lang_code:
|
|
|
|
continue
|
|
|
|
caption_format_data = []
|
|
|
|
for caption_format in self._caption_formats_ext:
|
|
|
|
query = {
|
|
|
|
'vid': video_subtitles_id,
|
|
|
|
'v': video_id,
|
|
|
|
'fmt': caption_format,
|
|
|
|
'lang': caption_lang_code if caption_original_lang_code is None else caption_original_lang_code,
|
|
|
|
'type': 'track',
|
|
|
|
'name': '',
|
|
|
|
'kind': '',
|
|
|
|
}
|
|
|
|
if caption_original_lang_code is not None:
|
|
|
|
query.update({'tlang': caption_lang_code})
|
|
|
|
caption_format_data.append({
|
|
|
|
'url': update_url_query(self._BASE_URL_CAPTIONS, query),
|
|
|
|
'ext': caption_format,
|
|
|
|
})
|
|
|
|
captions[caption_lang_code] = caption_format_data
|
|
|
|
if not captions:
|
|
|
|
self.report_warning('video doesn\'t have %s' % caption_type.replace('_', ' '))
|
|
|
|
return captions
|
|
|
|
|
|
|
|
def _get_subtitles(self, video_id, video_subtitles_id, hl):
|
|
|
|
if not video_subtitles_id or not hl:
|
|
|
|
return None
|
|
|
|
if self._captions_by_country_xml is None:
|
|
|
|
self._set_captions_data(video_id, video_subtitles_id, hl)
|
|
|
|
if self._captions_by_country_xml is None:
|
|
|
|
return None
|
|
|
|
return self._get_captions_by_type(video_id, video_subtitles_id, 'subtitles')
|
|
|
|
|
|
|
|
def _get_automatic_captions(self, video_id, video_subtitles_id, hl):
|
|
|
|
if not video_subtitles_id or not hl:
|
|
|
|
return None
|
|
|
|
if self._captions_by_country_xml is None:
|
|
|
|
self._set_captions_data(video_id, video_subtitles_id, hl)
|
|
|
|
if self._captions_by_country_xml is None:
|
|
|
|
return None
|
|
|
|
self.to_screen('%s: Looking for automatic captions' % video_id)
|
|
|
|
subtitle_original_track = self._captions_by_country_xml.find('track')
|
|
|
|
if subtitle_original_track is None:
|
|
|
|
return None
|
|
|
|
subtitle_original_lang_code = subtitle_original_track.attrib.get('lang_code')
|
|
|
|
if not subtitle_original_lang_code:
|
|
|
|
return None
|
|
|
|
return self._get_captions_by_type(video_id, video_subtitles_id, 'automatic_captions', subtitle_original_lang_code)
|
|
|
|
|
2015-06-29 06:53:21 +00:00
|
|
|
def _real_extract(self, url):
|
|
|
|
video_id = self._match_id(url)
|
2015-12-21 02:05:34 +00:00
|
|
|
webpage = self._download_webpage(
|
2017-02-05 13:41:08 +00:00
|
|
|
'http://docs.google.com/file/d/%s' % video_id, video_id)
|
2015-06-29 06:53:21 +00:00
|
|
|
|
2015-12-21 02:05:34 +00:00
|
|
|
reason = self._search_regex(r'"reason"\s*,\s*"([^"]+)', webpage, 'reason', default=None)
|
|
|
|
if reason:
|
|
|
|
raise ExtractorError(reason)
|
2015-06-24 00:13:23 +00:00
|
|
|
|
2015-12-21 02:05:34 +00:00
|
|
|
title = self._search_regex(r'"title"\s*,\s*"([^"]+)', webpage, 'title')
|
|
|
|
duration = int_or_none(self._search_regex(
|
|
|
|
r'"length_seconds"\s*,\s*"([^"]+)', webpage, 'length seconds', default=None))
|
|
|
|
fmt_stream_map = self._search_regex(
|
|
|
|
r'"fmt_stream_map"\s*,\s*"([^"]+)', webpage, 'fmt stream map').split(',')
|
|
|
|
fmt_list = self._search_regex(r'"fmt_list"\s*,\s*"([^"]+)', webpage, 'fmt_list').split(',')
|
2015-06-24 00:13:23 +00:00
|
|
|
|
2017-06-20 15:58:33 +00:00
|
|
|
resolutions = {}
|
|
|
|
for fmt in fmt_list:
|
|
|
|
mobj = re.search(
|
|
|
|
r'^(?P<format_id>\d+)/(?P<width>\d+)[xX](?P<height>\d+)', fmt)
|
|
|
|
if mobj:
|
|
|
|
resolutions[mobj.group('format_id')] = (
|
|
|
|
int(mobj.group('width')), int(mobj.group('height')))
|
|
|
|
|
2015-06-24 00:13:23 +00:00
|
|
|
formats = []
|
2017-06-20 15:58:33 +00:00
|
|
|
for fmt_stream in fmt_stream_map:
|
|
|
|
fmt_stream_split = fmt_stream.split('|')
|
|
|
|
if len(fmt_stream_split) < 2:
|
|
|
|
continue
|
|
|
|
format_id, format_url = fmt_stream_split[:2]
|
|
|
|
f = {
|
|
|
|
'url': lowercase_escape(format_url),
|
|
|
|
'format_id': format_id,
|
|
|
|
'ext': self._FORMATS_EXT[format_id],
|
|
|
|
}
|
|
|
|
resolution = resolutions.get(format_id)
|
|
|
|
if resolution:
|
|
|
|
f.update({
|
|
|
|
'width': resolution[0],
|
2017-07-08 17:26:13 +00:00
|
|
|
'height': resolution[1],
|
2017-06-20 15:58:33 +00:00
|
|
|
})
|
|
|
|
formats.append(f)
|
2015-06-24 00:13:23 +00:00
|
|
|
self._sort_formats(formats)
|
|
|
|
|
2017-08-22 16:48:59 +00:00
|
|
|
hl = self._search_regex(
|
|
|
|
r'"hl"\s*,\s*"([^"]+)', webpage, 'hl', default=None)
|
|
|
|
video_subtitles_id = None
|
|
|
|
ttsurl = self._search_regex(
|
|
|
|
r'"ttsurl"\s*,\s*"([^"]+)', webpage, 'ttsurl', default=None)
|
|
|
|
if ttsurl:
|
|
|
|
# the video Id for subtitles will be the last value in the ttsurl query string
|
|
|
|
video_subtitles_id = ttsurl.encode('utf-8').decode('unicode_escape').split('=')[-1]
|
|
|
|
|
2015-06-24 00:13:23 +00:00
|
|
|
return {
|
|
|
|
'id': video_id,
|
|
|
|
'title': title,
|
2016-02-21 21:13:18 +00:00
|
|
|
'thumbnail': self._og_search_thumbnail(webpage, default=None),
|
2015-12-21 02:05:34 +00:00
|
|
|
'duration': duration,
|
|
|
|
'formats': formats,
|
2017-08-22 16:48:59 +00:00
|
|
|
'subtitles': self.extract_subtitles(video_id, video_subtitles_id, hl),
|
|
|
|
'automatic_captions': self.extract_automatic_captions(video_id, video_subtitles_id, hl),
|
2015-06-24 00:13:23 +00:00
|
|
|
}
|