yt-dlp/youtube_dl/extractor/googleplus.py

94 lines
3.2 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import datetime
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
)
class GooglePlusIE(InfoExtractor):
IE_DESC = 'Google Plus'
_VALID_URL = r'https://plus\.google\.com/(?:[^/]+/)*?posts/(?P<id>\w+)'
IE_NAME = 'plus.google'
_TEST = {
'url': 'https://plus.google.com/u/0/108897254135232129896/posts/ZButuJc6CtH',
'info_dict': {
'id': 'ZButuJc6CtH',
'ext': 'flv',
'upload_date': '20120613',
'uploader': '井上ヨシマサ',
'title': '嘆きの天使 降臨',
}
}
def _real_extract(self, url):
# Extract id from URL
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
# Step 1, Retrieve post webpage to extract further information
webpage = self._download_webpage(url, video_id, 'Downloading entry webpage')
self.report_extraction(video_id)
# Extract update date
upload_date = self._html_search_regex(
r'''(?x)<a.+?class="o-U-s\s[^"]+"\s+style="display:\s*none"\s*>
([0-9]{4}-[0-9]{2}-[0-9]{2})</a>''',
webpage, 'upload date', fatal=False, flags=re.VERBOSE)
if upload_date:
# Convert timestring to a format suitable for filename
upload_date = datetime.datetime.strptime(upload_date, "%Y-%m-%d")
upload_date = upload_date.strftime('%Y%m%d')
# Extract uploader
uploader = self._html_search_regex(r'rel\="author".*?>(.*?)</a>',
webpage, 'uploader', fatal=False)
# Extract title
# Get the first line for title
video_title = self._html_search_regex(r'<meta name\=\"Description\" content\=\"(.*?)[\n<"]',
webpage, 'title', default='NA')
# Step 2, Simulate clicking the image box to launch video
DOMAIN = 'https://plus.google.com/'
video_page = self._search_regex(r'<a href="((?:%s)?photos/.*?)"' % re.escape(DOMAIN),
webpage, 'video page URL')
if not video_page.startswith(DOMAIN):
video_page = DOMAIN + video_page
webpage = self._download_webpage(video_page, video_id, 'Downloading video page')
# Extract video links all sizes
pattern = r'\d+,\d+,(\d+),"(http\://redirector\.googlevideo\.com.*?)"'
mobj = re.findall(pattern, webpage)
if len(mobj) == 0:
raise ExtractorError('Unable to extract video links')
# Sort in resolution
links = sorted(mobj)
# Choose the lowest of the sort, i.e. highest resolution
video_url = links[-1]
# Only get the url. The resolution part in the tuple has no use anymore
video_url = video_url[-1]
# Treat escaped \u0026 style hex
try:
video_url = video_url.decode("unicode_escape")
except AttributeError: # Python 3
video_url = bytes(video_url, 'ascii').decode('unicode-escape')
return {
'id': video_id,
'url': video_url,
'uploader': uploader,
'upload_date': upload_date,
'title': video_title,
'ext': 'flv',
}